<%BANNER%>

Enhanced Control Performance and Application to Fuel Cell Systems

Permanent Link: http://ufdc.ufl.edu/UFE0021944/00001

Material Information

Title: Enhanced Control Performance and Application to Fuel Cell Systems
Physical Description: 1 online resource (140 p.)
Language: english
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: cell, control, fuel, gpc, predictive, ramp, vcl
Chemical Engineering -- Dissertations, Academic -- UF
Genre: Chemical Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The inverted-pendulum virtual control lab, a simulation environment for teaching advanced concepts of process control, is designed using the LabVIEW software tool. Significant advantages of using this simulation tool for pedagogical purposes include avoiding the potential issue of schedule conflicts for securing equipment-access time in a physical laboratory and providing a learning resource that becomes accessible to students located in remote geographical places. A set of tuning relationships are proposed for standard proportional-integral controllers and proportional double-integral controllers for the purpose of tracking the slope of a ramp trajectory. Three different performance metrics are investigated to serve as the criteria for optimality, and a numerical optimization procedure is used to minimize each metric over 20,000 different plants. The proportional integral controller with tuning parameters selected to optimize value of the integral of the time-weighted absolute error is recommended for tracking the slope of a ramp trajectory. A generalized predictive control (GPC) strategy is proposed for a fuel cell system, where the controller incorporates a measured disturbance in the control design. The control objective is to maintain oxygen excess ratio at a prescribed constant value. The performance of the GPC control design is compared with that of the controllers proposed in literature for various scenarios including model uncertainty. The GPC controller has zero offset when the performance variable is measured and performs better than competing designs offered in the literature. The GPC controller is also robust with respect to model uncertainty. A battery of observers with a switching strategy is proposed for estimating the value of the performance variable when it is not measured. The GPC controller with a battery of observers has no offset demonstrating better performance than analogous designs proposed in literature. However, the control performance is not robust when the estimator battery is used and linear models used for observer design are uncertain given that the response offset is not completely eliminated.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis: Thesis (Ph.D.)--University of Florida, 2008.
Local: Adviser: Crisalle, Oscar D.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0021944:00001

Permanent Link: http://ufdc.ufl.edu/UFE0021944/00001

Material Information

Title: Enhanced Control Performance and Application to Fuel Cell Systems
Physical Description: 1 online resource (140 p.)
Language: english
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: cell, control, fuel, gpc, predictive, ramp, vcl
Chemical Engineering -- Dissertations, Academic -- UF
Genre: Chemical Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The inverted-pendulum virtual control lab, a simulation environment for teaching advanced concepts of process control, is designed using the LabVIEW software tool. Significant advantages of using this simulation tool for pedagogical purposes include avoiding the potential issue of schedule conflicts for securing equipment-access time in a physical laboratory and providing a learning resource that becomes accessible to students located in remote geographical places. A set of tuning relationships are proposed for standard proportional-integral controllers and proportional double-integral controllers for the purpose of tracking the slope of a ramp trajectory. Three different performance metrics are investigated to serve as the criteria for optimality, and a numerical optimization procedure is used to minimize each metric over 20,000 different plants. The proportional integral controller with tuning parameters selected to optimize value of the integral of the time-weighted absolute error is recommended for tracking the slope of a ramp trajectory. A generalized predictive control (GPC) strategy is proposed for a fuel cell system, where the controller incorporates a measured disturbance in the control design. The control objective is to maintain oxygen excess ratio at a prescribed constant value. The performance of the GPC control design is compared with that of the controllers proposed in literature for various scenarios including model uncertainty. The GPC controller has zero offset when the performance variable is measured and performs better than competing designs offered in the literature. The GPC controller is also robust with respect to model uncertainty. A battery of observers with a switching strategy is proposed for estimating the value of the performance variable when it is not measured. The GPC controller with a battery of observers has no offset demonstrating better performance than analogous designs proposed in literature. However, the control performance is not robust when the estimator battery is used and linear models used for observer design are uncertain given that the response offset is not completely eliminated.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis: Thesis (Ph.D.)--University of Florida, 2008.
Local: Adviser: Crisalle, Oscar D.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0021944:00001


This item has the following downloads:


Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20101111_AAAABA INGEST_TIME 2010-11-11T11:26:11Z PACKAGE UFE0021944_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 1997 DFID F20101111_AAARHN ORIGIN DEPOSITOR PATH shishodia_v_Page_026.txt GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
b8dea7f369ccf88c854ce70b818cd5d7
SHA-1
8d3eaf6459bc4f36c3b1afc6fe7eac867d2756bb
2146 F20101111_AAARIB shishodia_v_Page_042.txt
a152ecaaccd76e54ba9ed9b8b23c25b8
12e02e53730ca72c1bac76513f93398c9ee8016b
2204 F20101111_AAARHO shishodia_v_Page_027.txt
9682e4105ee8f7f96c0bcfad69ac8540
95a50a3c26b9064bc65628e27acd44a3def7cbe5
1981 F20101111_AAARGZ shishodia_v_Page_011.txt
c82355cbf17774890aeae1f15dbd36e4
5c15153f26c7e3e90134bcd01317e160b4888512
2058 F20101111_AAARIC shishodia_v_Page_043.txt
61e89388e9762e008a34626c2db52c88
5f08ceb5f828ee4386206387141fefb003897622
2171 F20101111_AAARHP shishodia_v_Page_029.txt
053bc869851dd2e6d1c8ceed229e1593
90009a15e5085a4f8cfeb08f28b1fbbce3329785
2242 F20101111_AAARID shishodia_v_Page_044.txt
1b7d08f4124bd64d292cce36ec52da8a
4bc0c7cb7354f45ed727c763aaaeef538ff8b3b2
989 F20101111_AAARHQ shishodia_v_Page_030.txt
e67951cc7a083c7d68b64e338d65fc61
3281f9597ba79e44c445203cb4cc8ac178bd19a2
2166 F20101111_AAARIE shishodia_v_Page_045.txt
eaf72506d1ff3cb4cd5a694470e2eef4
c7001b4af4e00364efeb2cbddbfb1228f2befd01
87 F20101111_AAARHR shishodia_v_Page_031.txt
ddf805f04926409226b7b6faadd00f7a
c33f37930d629693f6d59bde41c58f4ab6b48bfe
2286 F20101111_AAARIF shishodia_v_Page_046.txt
5d825c222dda55bdf1be9ce5d9f294c5
d0a07ffc6bb1afa6ab82fbbfda335fbddb4cc3a4
384 F20101111_AAARHS shishodia_v_Page_032.txt
01ba78b5b1eb1f9093cd0aa16d30f1d6
436c0fb48c2117a5fa88f587e08a23139de18f5f
2122 F20101111_AAARIG shishodia_v_Page_047.txt
49d0df3ada5cb50df4403b2cd3066f32
0baf26580e75a6d2d064e5f4291fed010e61cf8a
242 F20101111_AAARHT shishodia_v_Page_033.txt
e3ba7518fff91bdd4fa0bb990d939e23
0a73fccef98d6d28623db411f8c1be48f45cebf4
1986 F20101111_AAARIH shishodia_v_Page_048.txt
5e069ad85ec151096d219e21867e50df
84a0b002cdafd4fbb7d85c58ead8d7985526a3e2
948 F20101111_AAARHU shishodia_v_Page_035.txt
46422d0feed2b3727d8c4a35e9593336
befa276af9965a0702dc47628151b083b0626f9a
271 F20101111_AAARII shishodia_v_Page_049.txt
9f63587fdfc20fe6c5f3ba9e064d73a0
58da071406e8c18790a9ca0dfef326c2c9cb5d67
1209 F20101111_AAARHV shishodia_v_Page_036.txt
2d9d858bfc402d50fd4d50fd09fcd7ac
bead224f3d8337cdd5b957e846c88206167812a6
625 F20101111_AAARIJ shishodia_v_Page_050.txt
f66256449c3527d36101ad2210aa5444
dac2058d4fbc88e71cc17a5e7fa4c6f20c7a4bc8
412 F20101111_AAARHW shishodia_v_Page_037.txt
7ad9415bfae0d9c30af4b7f7dc6d7418
c8709a077421224ecc356dc172c4737383892586
364 F20101111_AAARIK shishodia_v_Page_051.txt
c411bba030836ae9559f9b9eaf1fb662
af51a10c58b6795b47a8d856d664433161df9df5
2092 F20101111_AAARHX shishodia_v_Page_038.txt
b52888832cf42e51a06d241beabde5bc
da4e1a86309c03e3a35e3adacc580b5531558519
556 F20101111_AAARIL shishodia_v_Page_052.txt
71c011bed8596aa278593e405b50d825
090766d50e65b784980b0afd900ae3c53cf005ab
2062 F20101111_AAARHY shishodia_v_Page_039.txt
9318ba5d73fd76bf7e613ff90a5c54b6
837ae5a3a11621bbb84c43501bbf34b4d1c7ea2d
1719 F20101111_AAARJA shishodia_v_Page_069.txt
852b1862819dfc1c21abfb925b7b93b8
8e79e826433ea7d981dbeddaef3ee901a71b2585
1318 F20101111_AAARIM shishodia_v_Page_053.txt
fb75e12cf601b371b2f2b2dbcc16411a
6715ed47e586a5c82264ec3c566b67615072f0d8
1928 F20101111_AAARHZ shishodia_v_Page_040.txt
5e2e5128469cd6edbf99120337a24779
2154c164e54bb571a8d3c118db0d540656435f6d
1901 F20101111_AAARJB shishodia_v_Page_070.txt
4330b7d26beeb1501de78cb24e7b734a
819a0669b0283fa499b01de10d9ecd565af86736
424 F20101111_AAARIN shishodia_v_Page_054.txt
c5bfc5d2429f46457f55a46e3f5957b1
0f162cb6c504b134f902257d2de4fd1bbed3737a
1835 F20101111_AAARJC shishodia_v_Page_071.txt
178689cd33d32ed7ff12b2ad8894c3e4
8304096151a2763cf617adb9e718efb30762c395
433 F20101111_AAARIO shishodia_v_Page_055.txt
5e860d4942f17da016515c669ccc3333
e1c6c01a8ed9b4ffa573cd96856659848f11c3f6
2153 F20101111_AAARJD shishodia_v_Page_072.txt
c0441d136b6cd813f90e650c9f27eaa3
7228e489d59e62c163fb8fd21c32be96b7374e49
744 F20101111_AAARIP shishodia_v_Page_057.txt
af368fd3393198c288e3755851185efa
f6812e4dd2e824d2d0723e0fdcf868411b3e4b75
1565 F20101111_AAARJE shishodia_v_Page_074.txt
186086132fcf0822d9ad0ca625a0fb6f
7e631d52aecfd31e9bd481676f512f22dbf2ab40
420 F20101111_AAARIQ shishodia_v_Page_058.txt
7abd1fa0b3eaafcc5df0d197644722d6
bb6b1a579c3a4e84e96b353f72ec8024ed5b6c55
1992 F20101111_AAARJF shishodia_v_Page_075.txt
7e01dc5d19a9308f9f44145dc2d6f9df
f179e8ef154510ef88aa10b9d96fcc80ff47ebbe
438 F20101111_AAARIR shishodia_v_Page_059.txt
234755cd4c82e2c64b6c8bf714bd1df0
03d0fee1942cbf132b8ebaa0a325c6825656b2b6
1650 F20101111_AAARJG shishodia_v_Page_076.txt
6fc2ab3b13e720e864f12fa4a25cc15f
68893e89aa3f8066b9c1d2787fe6641bb1ab1618
441 F20101111_AAARIS shishodia_v_Page_060.txt
2129b169ebf8ad0cc7c71d6a9a4844d2
91ac6098471c247715cfac6db836e0861a13384f
1918 F20101111_AAARJH shishodia_v_Page_077.txt
addfcc93b8d14ea5ce096c4d40c72a86
982e329c4d941137d8baef888de4b9c52f3999e2
2060 F20101111_AAARIT shishodia_v_Page_062.txt
26773966a6018204f9f8905d8a4b97a9
ed6088dc22919c12d17aa46430536492b6999579
1784 F20101111_AAARJI shishodia_v_Page_078.txt
ad7e8b0bda99e3e8fda2bed87cbfc9e8
4a7af6227e3ea1fd0dccb77e2e71b1c3e50453c3
2142 F20101111_AAARIU shishodia_v_Page_063.txt
73063921aeb5c8529ff4c4731fd4ff4f
7bd4efb51a55452f8a9763341e75c5f10eda4b21
1542 F20101111_AAARJJ shishodia_v_Page_079.txt
69359435c6aa3acb7c0476044cd996d8
89dd14ca0a00a9560d0e99d98cdf3a2a887c1e72
2091 F20101111_AAARIV shishodia_v_Page_064.txt
aad2cbe39a0642f47daef9c0b9869b59
08c6b87c3b9c79491a824cb697961a5911fa1389
1893 F20101111_AAARJK shishodia_v_Page_080.txt
28f967ac3a2f9f8fb146a9e1222db6c7
4c4451e1f1756a22e7c495249a6f5b63cbf63c67
1659 F20101111_AAARIW shishodia_v_Page_065.txt
44e0808401ba50fe44e029d0f129a748
8abdc7bdb56ef4971940a07705eee75ba4cc4634
1816 F20101111_AAARJL shishodia_v_Page_081.txt
0eba1d2e01f70d5b5f9af308da639b38
6ef5b9e799ecdd966ded43f96c34d2b12839673c
2051 F20101111_AAARIX shishodia_v_Page_066.txt
35261f32aa19d687c7b85a9a90340661
ef2b2f9690f7376c436775220ed2d46a4a5c0d88
2744 F20101111_AAARKA shishodia_v_Page_099.txt
73432b3db87b432d2bb63df7d606025e
d648c8c2f6762c5d2c1eeb065e8fc68030a3ce4e
2244 F20101111_AAARJM shishodia_v_Page_082.txt
318282a5c1e29eec19d05f241cfacfde
93699be31bb0a0bd2b7e03af7fa14793847185fe
1925 F20101111_AAARIY shishodia_v_Page_067.txt
464122637068e4e6d727237d001ed274
9293c13757f26f0b184f6a1e65f7ca5b28f3cf92
2178 F20101111_AAARJN shishodia_v_Page_083.txt
a8c8e0a3529eb02185445bbb5643d119
9c7189929e871e157c6c258acaad4559fa1840b6
1713 F20101111_AAARIZ shishodia_v_Page_068.txt
313fed9553b0450ae119613c2a9b1a27
7ba70c3f78efa3588ca49092e405eccaff62eb42
397 F20101111_AAARKB shishodia_v_Page_100.txt
3eed28a8c71b5fb49dae23235fcbe23a
954c823a76c7bca0a0e44e42a3bc5a37095b2e3d
2075 F20101111_AAARJO shishodia_v_Page_085.txt
3b8a4ddba73fce7a8b6879bdfd757381
f2ec74a2170358aa2f5a532ece3ad9fbfbf4bd83
524 F20101111_AAARKC shishodia_v_Page_102.txt
30524ba2998f86e1c9ad5e0804eef82d
7e0f0bbfd4ee4db36ca855b3d227cdf4a011af2a
2263 F20101111_AAARJP shishodia_v_Page_086.txt
f4a322b540c8263f65c53a8e7d230a80
40b42815aba2347c9dad362bf4c76c6df937face
356 F20101111_AAARKD shishodia_v_Page_103.txt
5e10325531cd3548b4eb7bf60c810988
4e16cc6ec72da851bbdc0259c55d9c7bb698ef19
1961 F20101111_AAARJQ shishodia_v_Page_087.txt
4e41499f8795f58e81d3c38ecf6e4274
3017c69d0cc91232fc4c7731a3ba33ff6b04d589
314 F20101111_AAARKE shishodia_v_Page_104.txt
953f75c5fa1915e313bf5d6018f163ad
725bd22da3a581da426a8a7a7e14347d523d19cb
1936 F20101111_AAARJR shishodia_v_Page_088.txt
6481b6fba8e0cdf64831b0b7cc44b49f
4d8ad3b7b010d038e08e1a56cdff2203e07c3b8f
439 F20101111_AAARKF shishodia_v_Page_105.txt
c620fc320e9b8615692f5e6834879042
96e5be779506a64b0fda513b8579e2c83744e826
F20101111_AAARJS shishodia_v_Page_091.txt
ddd861197e25e91963b439cea7644c6e
ee0c3fbe26cb589c6a5979cfd3b85856dc8d139b
306 F20101111_AAARKG shishodia_v_Page_106.txt
88da344669774b79231f394ad8037278
0697f5bf3b64de1d0aa89eef893a53a28ec66624
1837 F20101111_AAARJT shishodia_v_Page_092.txt
0c166171a112db5e4202709840407d49
db05121b1340e95656405675ac5a54e7e8e4b697
542 F20101111_AAARKH shishodia_v_Page_107.txt
df68f764ffe4c24bf2d28d5c0f03ad00
747b3254c9de12726300ab7d1351cd12c01542eb
2042 F20101111_AAARJU shishodia_v_Page_093.txt
f4054c11c009cbf1d1c2477b3ae358ed
5bdb11baecdf7705e391d3fdeb269433f092cde2
282 F20101111_AAARKI shishodia_v_Page_108.txt
79c37e68c12b868b116f86d09fd51416
4062081eca87d7fc4785163d8e08af472557eeca
1968 F20101111_AAARJV shishodia_v_Page_094.txt
f6b837503c1752a78a4baf522ef1f080
2b472c1b734bdd8cc818d774578997d6050f983c
287 F20101111_AAARKJ shishodia_v_Page_109.txt
0060fe6b79c90f29ac192a115e39ed6d
df01ab5eb47afd7487dd2e29967dd07829fa7f47
1752 F20101111_AAARJW shishodia_v_Page_095.txt
f7cbec45557ac15cb813b4770fbd897e
6df9553a6ea869e0966f330214500a2d4fbb921a
362 F20101111_AAARKK shishodia_v_Page_110.txt
b6c2765bf7315f59f7502a449ea0c11f
c97b5aa271d80823bec0d2b2add54137a14855ca
339 F20101111_AAARJX shishodia_v_Page_096.txt
71d1c15ff6b59b04704ac7e42146e19d
ed3adaa7f8b52a9ecc72c2086296a52060fb05db
786 F20101111_AAARKL shishodia_v_Page_111.txt
fcc1dc0932101121c493f06eb0d13b68
5be8326c659b3d9c0f58e3880999acb971326ca7
297 F20101111_AAARJY shishodia_v_Page_097.txt
ae5ae56072917983d74d4a541c95ec7c
ef296e0d3235f65db3d90c9c0cda38606160d047
471 F20101111_AAARLA shishodia_v_Page_127.txt
307c6df83cff3c04d2e85e10da0a4a76
796008636aa99489ba2103e76b1333a682116e54
327 F20101111_AAARKM shishodia_v_Page_113.txt
8091ac97009932fa44a32d29e8a90d6b
dea8ab83257a67fc144d9a20dd5dd87ff0a32a34
F20101111_AAARJZ shishodia_v_Page_098.txt
18d97e0130a5b21983623b9abe0c6b7e
0126549fd9198894291d141213c1a88bc004354d
475 F20101111_AAARLB shishodia_v_Page_128.txt
3dd6e980d8151a220dba8e7060a37dc5
f30e9c45892d3c5657b826ce73d562144048fc2d
329 F20101111_AAARKN shishodia_v_Page_114.txt
2b833fbc24b9bec2b8bc126bd8cd98df
95eeb1a9840650e137985db40bee486194f0f27b
410 F20101111_AAARLC shishodia_v_Page_129.txt
c1c496e7c2e64a1fc978bf6ba56617ef
d6342613a67343985849c5082d28746304958f00
385 F20101111_AAARKO shishodia_v_Page_115.txt
524244ce02ab1c0a142bd365d23ae9a7
c7472d398e0b1749fcd9991456cd24d48653c370
515 F20101111_AAARLD shishodia_v_Page_131.txt
42665e2b58537a4fc79becfc8b9c319e
40a609ce625ffdbd49f83590c12324a837e1dd69
378 F20101111_AAARKP shishodia_v_Page_116.txt
bd9d68901b434a9f6dd159253a7e26ee
afbedd76f6d6118cfbf43126d09f5df831f4365e
2023 F20101111_AAARLE shishodia_v_Page_132.txt
91fedd1ca50b068bfc071587cfdcffac
6fe4ec1bbeeb9ae96f3041beb2f72b5cee2c7cff
448 F20101111_AAARKQ shishodia_v_Page_117.txt
c7eb76ce9dcc722757b6a0a86cfb9b18
0e8e1274f35ca32e302fbef48ef570febfc48499
1570 F20101111_AAARLF shishodia_v_Page_133.txt
e1d005796951ee3c4a244c87f6c045ab
9b0158c3ab5ab547b1204498d0f61044c35e5904
423 F20101111_AAARKR shishodia_v_Page_118.txt
717244dad161dcd2cf58210692b0e480
d42e52e7994c49402838897f128525a23a747354
1693 F20101111_AAARLG shishodia_v_Page_134.txt
6265244052ad6387c33853fa13df7417
1ebd76fa84acbab2bd4a43634fd3790ca6303873
294 F20101111_AAARKS shishodia_v_Page_119.txt
28006be84efc36ae346873ea6e78ee8f
6c13de3aec64b88975b33ab385c44dd02ada6ac9
1180 F20101111_AAARLH shishodia_v_Page_135.txt
af3c0ca5153ea2925646983c9593ddf2
ea10512cd0c261d7a4896bc7682e779c22531a29
F20101111_AAARKT shishodia_v_Page_120.txt
c2f163fc61aef44380023d0f53c4d65d
1fb4a4c2578ff2e777d3b4572e2be9acf67956aa
932 F20101111_AAARLI shishodia_v_Page_136.txt
21cffc28864d1b022cd6b08ae345c754
5262d588c0677510424ab18b54080da7c76b557b
523 F20101111_AAARKU shishodia_v_Page_121.txt
2f62669601ecee392bd0e86865997608
299ad7272aab5481397e8eed928a85fe4b20b0c0
2761 F20101111_AAARLJ shishodia_v_Page_137.txt
2df8fc140a03f9c5651af246f2fc4f5b
f05d2a17da927d9bc2cd3d2fe7b12bf3969be303
699 F20101111_AAARKV shishodia_v_Page_122.txt
d34d78c252b2edde1ccfe1db6fff0113
2a05617ac675719cbc73721111a945be855cb433
2620 F20101111_AAARLK shishodia_v_Page_138.txt
367281dcf5052b66970845037972198a
2514ef69ec051835f376276a8cfe53cf3852f813
389 F20101111_AAARKW shishodia_v_Page_123.txt
249f5f494a57f09158f411c8e510b02f
d41515983ae8b07d18e11be0bb9586234b133a3d
1656 F20101111_AAARLL shishodia_v_Page_139.txt
e56f9fdb8005bd5b9ff507780ed1b6ef
aec0c60ab20aa85143b317fd944584b8dedf874e
503 F20101111_AAARKX shishodia_v_Page_124.txt
c1ef98c315deb8e4a134f646c4eafee1
f81d721a0c36efecd3b51638bb0fbdec2e41097f
3369 F20101111_AAARMA shishodia_v_Page_055thm.jpg
397a67ab0edc01bae5ab8b2fb2312f25
89ba38f6d18442f24ea827df04f02e5db4478d1b
927 F20101111_AAARLM shishodia_v_Page_140.txt
59ec83b6de912b8b480e7ec90d90c26d
d415adfec9481f1cbb3d62e66c36cc70fa68a0c9
457 F20101111_AAARKY shishodia_v_Page_125.txt
6f810ae2b25ed14789fa374d204c70f1
51452e08f50ee0849d05b735451ad58048124e54
4945 F20101111_AAARMB shishodia_v_Page_051thm.jpg
dfd86c29ed8742843b34ec16669fd17f
0c247197fb0ddc8db6da5ecd6484dda8e534e03e
2279 F20101111_AAARLN shishodia_v_Page_001thm.jpg
d4b654c642249549f835058c63155651
85e48f8403151e23d481867191800f5874a2eb41
437 F20101111_AAARKZ shishodia_v_Page_126.txt
8d7c2164e717632588848c9d4b310727
04b75b83b79b1807aca69b70894a0ea580608f4c
3518 F20101111_AAARMC shishodia_v_Page_032thm.jpg
aa048b062424053af174aa6dccad81a1
44977b3a08f3619a9dfd8e29062d904969c8b99a
1619288 F20101111_AAARLO shishodia_v.pdf
132af51f5c39573467944bbad94ca286
e8293f71f0ca4e36e7d46daa050bd969ac9938ef
6945 F20101111_AAARMD shishodia_v_Page_094thm.jpg
e0c30dcbd81654b600141e1f863b0c97
05f13a793fd44fe0df1ec9d6e7cd775dc397e018
12258 F20101111_AAARLP shishodia_v_Page_037.QC.jpg
bd726f2f80712df309907859e6171482
7f2f65a5c0bd6cc9931e5b85c953d56acac3f195
3106 F20101111_AAARME shishodia_v_Page_002.QC.jpg
7667a1b9e4881cc8427730ec0d02e63b
0436f164580e5364dc580046c987fb6c8c5e7d7a
26250 F20101111_AAARLQ shishodia_v_Page_085.QC.jpg
bec5b1c316e174655e529f52a5290b3d
fc8261e80ded60f02279b1596ccdaaa487ce8085
11166 F20101111_AAARMF shishodia_v_Page_053.QC.jpg
bcbe40c7722ff212b92e78fdceda0855
edeab21c9d2ce21dc2accecdece2bc3d39a38fd8
6736 F20101111_AAARLR shishodia_v_Page_066thm.jpg
09145cb46718a1d61969d8eedafcd986
a9d6bd4380fd170cafed462fb0084a9e4a109924
4832 F20101111_AAARMG shishodia_v_Page_036thm.jpg
ef1940adca190699d0ec149797b39b22
e7f77398b8498f59674b301083f9b2b21a87d8e3
31321 F20101111_AAARLS shishodia_v_Page_010.QC.jpg
7bfac683414ab37045feecb3a4a90b95
5d5e14aa8395bae57ce79e85d0fdb51e74b56497
4694 F20101111_AAARMH shishodia_v_Page_050thm.jpg
3000d52c97ae03cdd83ba0fb329af653
49f8de8705e3102e2bf8340d93b31f7acfe30656
6424 F20101111_AAARLT shishodia_v_Page_103.QC.jpg
27b7d23333081e18d70fb2f789500d94
dd5f7ec76c6fb20579c22bccbb58c33f2e834dc2
9209 F20101111_AAARMI shishodia_v_Page_110.QC.jpg
5cbfc92c17585f430ce250cb2640e350
9185b2153b62ec8d7230f4573c97cf2550a61947
5525 F20101111_AAARLU shishodia_v_Page_133thm.jpg
72f29e41460e47d057a1a0a2ce8cba4f
c33e6a0d0d3525fedfee8e340d071e7e4055d8e9
25292 F20101111_AAARMJ shishodia_v_Page_009.QC.jpg
1ac136b9dc78a2b5ddcf9d404e0e2ab5
b4dead1baeae93277f0540b4e11bf32cef6aa435
2703 F20101111_AAARLV shishodia_v_Page_131thm.jpg
95e76b81c5ac79df2ec4b6870044b313
0e8911088c451cad1617b24dbccd695f6b4f9641
5473 F20101111_AAARMK shishodia_v_Page_139thm.jpg
cc8507075439ed843bb9ade0428a4961
8efb98b39ff45715ae131d26ca41096129259608
25442 F20101111_AAARLW shishodia_v_Page_093.QC.jpg
7dc336c8e8d1120f8f55fd15d627df2c
cd11a7acd19bd6dc635b27175817ff718f8ce3c2
14321 F20101111_AAARML shishodia_v_Page_020.QC.jpg
924c2de27f61d2936aaa3bfb216958c7
510cf5196c650772edcb98e290b45af840e9a527
3102 F20101111_AAARLX shishodia_v_Page_113thm.jpg
ce17cd930cba4d65430e3911f9098d52
2ed7fc7387f0d55a5a06887773b98b9f5216360b
26683 F20101111_AAARMM shishodia_v_Page_029.QC.jpg
d20378ee4068359cf870ef99e2515fca
6679300fd28eebe0ebb24c46f319f3d5b941742a
2510 F20101111_AAARLY shishodia_v_Page_115thm.jpg
6db6b80ac15e647cbfe1c3bce77f8a02
a2424605137e37c4eefd93fe6cfcbaf845e74103
2041 F20101111_AAARNA shishodia_v_Page_007thm.jpg
a918b539f1515b7f3e8c76c5b58382b7
fb0c2fd54ad491eda51badde94930ebd20c847a8
20610 F20101111_AAARMN shishodia_v_Page_080.QC.jpg
77ef3a60992d64657eb0a83bae840206
4942761ce87513bb7f418db3f78ee9d4c5631b2e
22639 F20101111_AAARLZ shishodia_v_Page_072.QC.jpg
94d58c237d5b540c0f4838c16cbc340e
7f242d00d805913165efa9d254985cb0a490b945
25512 F20101111_AAARNB shishodia_v_Page_008.QC.jpg
cef2092ad520719d50ca73037a672d9c
ae9a8ee4984acbd95145ee4ae078cc353dc509a0
6948 F20101111_AAARMO shishodia_v_Page_116.QC.jpg
72aa3032572fda36e12be09437bc5f58
b34ced5c2cbedde5c21f036df4f251e923d1b35d
25271604 F20101111_AAAQKA shishodia_v_Page_053.tif
6a94aafc79b937c64874866535b908a6
bd04bc076587a7db1ba215c695b89dbb17f911ec
6457 F20101111_AAARNC shishodia_v_Page_008thm.jpg
7a00585f203c79aa2302e84f9f45fa68
f37bd7971b4163494efb532caa80dc0d1f6ecb3c
212365 F20101111_AAARMP UFE0021944_00001.xml FULL
c1122fd782d316b1f130aebd65706bbc
12504a5b36e81bbee8577f9fdf841747c13951d6
F20101111_AAAQKB shishodia_v_Page_013.tif
6b8c12f1e9f83becc2106cba659642aa
5fbf04a14c2a7a6a7a99a8eec745c0e1bb2de02f
6592 F20101111_AAARND shishodia_v_Page_009thm.jpg
567921684433c2d4734e5eb34dd44d03
930e6ff4a3f4113a87a8af251bc062d9d4a7c90b
7275 F20101111_AAARMQ shishodia_v_Page_001.QC.jpg
2a08e5f3814e145ca542d12572a65ed8
5427726e218eac5f2cdc28f9bf3cedcbfef6ecf3
1053954 F20101111_AAAQKC shishodia_v_Page_102.tif
97d5cbe5a8a89623afaf593275003744
894d19029a1bf4978d37e53fc676cd359ac785eb
7801 F20101111_AAARNE shishodia_v_Page_010thm.jpg
486667a76d69aa2628bd1e20d308cb5c
c624142f796552affd75551347cbf08d6278cf7d
1418 F20101111_AAARMR shishodia_v_Page_002thm.jpg
1f54148017f5a84ed0a6ed9684af75e6
c8b7b9cda6dc1d59a2dd343e31846063fbb87185
55788 F20101111_AAAQKD shishodia_v_Page_074.jpg
533418178e28586aa644b7571367cff2
3fab74885fd432f63c7912ddad90ef9b7d0fe0d8
6585 F20101111_AAARNF shishodia_v_Page_011thm.jpg
001a49810967a69cb4821f7776e9e168
656a941d4a4d9356e6fb6c5f7952ae2d4ff76b61
2956 F20101111_AAARMS shishodia_v_Page_003.QC.jpg
051ef91e7fb33e0e31a3c375dae52960
73a82a320b4dc0f28c3e0b43a2fbfb5022de2490
24849 F20101111_AAAQKE shishodia_v_Page_048.QC.jpg
47a8d95bbbd7cfe9e4b1096ee55a9b8e
8188bae868a3fbd0a027828a321a8407439a94c8
12158 F20101111_AAARNG shishodia_v_Page_012.QC.jpg
d326bd8c4df451cc36964a1f3b2e6617
206a9947c2c12a8cd6841c5e55c0d6e38078677a
1305 F20101111_AAARMT shishodia_v_Page_003thm.jpg
73532b64d0614193876a9064627b8fbf
294830991347d09338692591004844a945b41174
18782 F20101111_AAAQKF shishodia_v_Page_096.jp2
d6c3a11ce3e95ffe2f55dc35f036215f
55c9840ea83b3af60fbf2303c08330e919d8871b
3724 F20101111_AAARNH shishodia_v_Page_012thm.jpg
8f86c757a4f2500b72997684c0f8a6cc
26c6d0abb671b3f7d86ade64063679985d7b9f98
16286 F20101111_AAARMU shishodia_v_Page_004.QC.jpg
234268d2811b4dc0b4a7cbb35733ebdc
6df2d6d65abe56e233c66389ac4d00597bf61ec7
F20101111_AAAQKG shishodia_v_Page_084.txt
090d95bef660a15be672543014c9f0fd
e76019848e9ee114e970d55854e24a23922478da
27007 F20101111_AAARNI shishodia_v_Page_013.QC.jpg
07507a6585dea0e3c9426195c8ff8aad
6e449a9d7df2d51a998adc18eb1f305883537454
20628 F20101111_AAARMV shishodia_v_Page_005.QC.jpg
16f0616afa4986fc09514fb524b3904b
f75d234b244776502ccb940adc9b1402c8795ae9
2234 F20101111_AAAQKH shishodia_v_Page_015.txt
997e9fa72594eb74442b99310e0a7c56
d2dc8274a7b748012620a9d19c33d13888545000
7330 F20101111_AAARNJ shishodia_v_Page_013thm.jpg
f1f15265e37df131c2d403cc8ade7c83
a4dc4cca16988b5b97e040b813eca93744271366
2409 F20101111_AAAQJT shishodia_v_Page_098thm.jpg
7864723d3e8e2f95faa7b99fa8dc0ae1
26435faac2053919818ca35d8cdfa2089d6e2f93
5653 F20101111_AAARMW shishodia_v_Page_005thm.jpg
f1bd64db70d3c98969f65964c17e3482
ca0ea2dfa2e937912b252ca74e79c5429ae6bc98
44985 F20101111_AAAQKI shishodia_v_Page_078.jpg
aa6000aa82338366b014735c83af09dd
724e862f6b47c40ae4a10f37367b5f382a740e16
5024 F20101111_AAARNK shishodia_v_Page_014thm.jpg
48820e284d1d421ef8ab45a593fc84d8
2bd6d84c0810449a646c820912b170dfe7a38312
17946 F20101111_AAAQJU shishodia_v_Page_014.QC.jpg
8f47aed994d5f6a6737a071773d9c10c
72cfd9a3c0a5009b4fd36f842b415404495bb835
17764 F20101111_AAAROA shishodia_v_Page_024.QC.jpg
73e55ec87fb06509b14acffbe843f45e
e2e94cd1afa9ac3ab472357a21b30438f1020d1e
20074 F20101111_AAARMX shishodia_v_Page_006.QC.jpg
aab5fc64e53dc85eb0c2c4a7d94b2c38
4eee6a2b7c8031b93721f55aa132e1bdcd16cfdf
245 F20101111_AAAQKJ shishodia_v_Page_034.txt
741b2c1835ccc5378b9a2b7658a4d072
20f045d587a195eb586f3d0567dfe803ef224d80
26892 F20101111_AAARNL shishodia_v_Page_015.QC.jpg
5f1bce523b933f909523872fbbc865d0
9074498aa689e4b1109e84ffb122be09a9cd311b
1746 F20101111_AAAQJV shishodia_v_Page_073.txt
36df34773979cf1b6132d1fc32914244
6b7ed7546bdb047dc2aa5f964e6b90bf05aa408b
5506 F20101111_AAARMY shishodia_v_Page_006thm.jpg
7c5ef5ac5e82d97e6a8088fdae6c6f80
e2e9f03975c67af312f785714c429fcc8d689559
F20101111_AAAQKK shishodia_v_Page_056.txt
2ab2700d48ea3f9223b184c274d05f1c
a4255a7c6b00055a6aa093fc69579da32cbfee2b
7344 F20101111_AAARNM shishodia_v_Page_015thm.jpg
cd4aefd1185168d766414150547ba64e
e021f2471544f7dfa4a3a8cc724e7ee77a3d71da
6864 F20101111_AAAQJW shishodia_v_Page_016thm.jpg
6fdbe8b84b1e0e28e9e2ce7d894ad5cc
6d2b1464f5885067b0392db77f5f9b47dffab342
5602 F20101111_AAAROB shishodia_v_Page_024thm.jpg
d1f450600adba5ebff3c7f12dae68d11
0ea77d2a74a0717eb830630ed36899c7bd6377fd
6748 F20101111_AAARMZ shishodia_v_Page_007.QC.jpg
8fa59e55816a5a7b7bc0628f80da06b3
f2fae469b4b0f1f2f67389b8e5312249e0226de3
792 F20101111_AAAQKL shishodia_v_Page_112.txt
6b5061749cb158242ff9e0872fba8428
59c24562119048cf1115b12effe347364e2293d4
25806 F20101111_AAARNN shishodia_v_Page_016.QC.jpg
bc0f96b5ac085b5832914015c6106d2b
7fedf25d56f75486cd92ed844003810e61f55671
25670 F20101111_AAAQJX shishodia_v_Page_039.QC.jpg
73095555c47ab976f738d34790440834
4f904cdba88ef7d125824d2f6ab393a50ea110a6
22380 F20101111_AAAROC shishodia_v_Page_025.QC.jpg
31aa4a66ff2262ee186e17efe32d8c06
e94ae7eb5d72c9f64ccf57ecba9af1f3eb1d8efe
6942 F20101111_AAAQLA shishodia_v_Page_043thm.jpg
0e4ee360f89032fa549924a057855f0e
4e54d9679d4073173c4fda92b475d81e6ce8e959
9428 F20101111_AAAQKM shishodia_v_Page_003.jpg
9d0ebea8de6737799c6628be50560433
23b4c4c8aedd3148b4b135c018bae1988a9b7257
26409 F20101111_AAARNO shishodia_v_Page_017.QC.jpg
0d4faec1a9e9db8f05516a5c019dbd4c
1cd2306e320df58d36d26493e55572dbe2d9a5a8
F20101111_AAAQJY shishodia_v_Page_115.tif
7d58ee24bedd63247b208dbbf95b5f01
3b28ee2e3ae1b9f078539040bd11be7475a2e08d
6510 F20101111_AAAROD shishodia_v_Page_025thm.jpg
5e92859f50ecb4da2aa66ac1d3db57ca
3799d24fd64ed8db35ff94d45e2208f6a90e0f45
37779 F20101111_AAAQLB shishodia_v_Page_055.jp2
b2ebef760472bbbb31c15681195eefea
f83c180cb4525ce8a5fb3c4f6da67a46cc8c450f
2138 F20101111_AAAQKN shishodia_v_Page_090.txt
98d7afd5b8a2db9efa8a4003889c59db
7ade209a1f47b9966d363b3db8bb7545c769e07f
7230 F20101111_AAARNP shishodia_v_Page_017thm.jpg
a9ca2b2af2164c424abbecc097197f35
e84ebe616eda440e692ce3f54b8f43e75423cd05
21367 F20101111_AAAQJZ shishodia_v_Page_116.jp2
a876a7926b007dcd79c3abe11eedd098
c782f1d338faa928c093799ad900f2f771657601
25274 F20101111_AAAROE shishodia_v_Page_026.QC.jpg
305a4bd0dd21b0fd01edcc3daff0c3cc
03df93dc3be8184115400420cf4c22ab732a8da1
6379 F20101111_AAAQLC shishodia_v_Page_058.pro
fa37d75351836eba698130cc89b3d3d1
03a9e97da2c066cd744872dfe94167b02265e7fb
9394 F20101111_AAAQKO shishodia_v_Page_113.QC.jpg
4b4d71cf9160c0e478d41a6e6b4ffe6e
03065a31b2e3c46007f0048ef6f09cac4b3d9a57
16938 F20101111_AAARNQ shishodia_v_Page_018.QC.jpg
8cd755c0b2aca98094e614f735c7eb65
a08f87d3a9df6e0dc40d8abb07d6c13f1e503093
7151 F20101111_AAAROF shishodia_v_Page_026thm.jpg
1dcca07e59df1c5bf814e1404d9f5d49
8baab9bfc312c157ab0abb4c5c7c044965d5510f
3083 F20101111_AAAQLD shishodia_v_Page_008.txt
4826fd2d916eb1fdd187d279c3dbb639
bf9691a21b3eded560c2b3e03183296cbea5d916
736035 F20101111_AAAQKP shishodia_v_Page_065.jp2
555ac42dd7027e71d41c361598c9ad70
5fd330d4ba7005b958e20b80690a001edf18147f
5134 F20101111_AAARNR shishodia_v_Page_018thm.jpg
37cca6e0a43e64de54ceb885b5b34c47
4f781673a75adf0b8164f264f19b5b7721e71cb9
27863 F20101111_AAAROG shishodia_v_Page_027.QC.jpg
958a0e871fa2b4895417bc033a63e897
f6bec1ec2282d6bc0654764cf0fa89fba5427e13
21505 F20101111_AAAQLE shishodia_v_Page_116.jpg
e6fb4671fa47d901c55d01b64c58ba6d
8ab40b9607a05449af42deece4750053aaafb6c2
89158 F20101111_AAAQKQ shishodia_v_Page_009.jpg
15ba19132939a6156473e73e3b4d1f88
648ee48b1a6fbdb6608f75e318d74319f7dbeb0c
11444 F20101111_AAARNS shishodia_v_Page_019.QC.jpg
0c7ca4ad94dec06215972ec5992c8c0b
a5f147ce42ebe2dad5f129523512f67dad27547e
7448 F20101111_AAAROH shishodia_v_Page_027thm.jpg
bdfd15664c99de1957f9038a78eab929
9e0a2a84c5b207a1ddd70eb01de8c3ac398314d0
36986 F20101111_AAAQLF shishodia_v_Page_056.jp2
b70bbf0f0c2dc9e6c5e3f309baa2fce7
bbcb7561a18e7a79db1ab48203d571c564f94b53
F20101111_AAAQKR shishodia_v_Page_138.tif
6dfaddae59c2d901e3a4f7496a84d1da
9e6835c2cad1c721192784498a17c86ca9a14553
3768 F20101111_AAARNT shishodia_v_Page_019thm.jpg
cc87ed66496b982fd9d651936bd5cacb
f60ebb03f897c42e5a994d718909c14f3ab1db98
26483 F20101111_AAAROI shishodia_v_Page_028.QC.jpg
63b6e7959d85cab7e7b8d1aca404fc7d
6cb80aa635cebbd59e8bef1d2bbdf122c6d2ace5
4655 F20101111_AAAQLG shishodia_v_Page_004thm.jpg
eb9eca7cb62881f748abfd2fb0935f36
282c543e18d0e81a90ac02c97de36f9d974ecb5f
81030 F20101111_AAAQKS shishodia_v_Page_038.jpg
d436807476ed86b99dad5a3f1a9881f2
4c7ff971e5929b763079a8cb792e28bf70413556
4697 F20101111_AAARNU shishodia_v_Page_020thm.jpg
4e54a8194c4da72dfe8483a9db2a8223
712257b5b2eaa276b57a3b07cc5130138d4c543a
7071 F20101111_AAAROJ shishodia_v_Page_028thm.jpg
697e7a4749d30fe76119745de44b0621
d022a62e9097e3d4c0f9dc786e63a04d6a1770a2
483 F20101111_AAAQLH shishodia_v_Page_130.txt
86c5500c402c1b23d2035cdbf7d47ecf
c1071243c2b45ea6bf9dff587053cf44aa8b50a8
7704 F20101111_AAAQKT shishodia_v_Page_120.pro
81060ed0e9e1d241fd573c0fd3bd3617
43bc92b2e7a41e90f25e9f6f1821cccfe733e232
14800 F20101111_AAARNV shishodia_v_Page_021.QC.jpg
4398562a80e0d0fa7a5fa6253ab0af4b
9ebfb95f57da73a77e5622f8bb4b730d37a8af04
7171 F20101111_AAAROK shishodia_v_Page_029thm.jpg
013617062a32beeb2cabc7bcd5b045d8
56c211f55f9c8ce8d812ec97a991710c382d85c7
25727 F20101111_AAAQLI shishodia_v_Page_038.QC.jpg
23b5c7d02259c10ef5c40baf804193ca
ca295262843cba871ea35c189340f1cff348971f
10021 F20101111_AAAQKU shishodia_v_Page_002.jpg
98e5a5a6b112462b76b4afca796470e4
49c1821df2d35ffc552f1d23fde27de3ac96280c
4981 F20101111_AAARNW shishodia_v_Page_021thm.jpg
be08643735b712b77982671aea8dcdd7
46363351a9999dd27dcd0048aa8b599083546861
17281 F20101111_AAARPA shishodia_v_Page_040.QC.jpg
e26193b8f21a82c0cb6245116d1abe8b
cdc6289e4542c2dfe5b35ded0a0a86d7515a8a72
14001 F20101111_AAAROL shishodia_v_Page_030.QC.jpg
47dc3be4573f33e0fb523ee56fdf2d2c
3818d68be309ed7f1433997c349df5c132f64be2
6515 F20101111_AAAQLJ shishodia_v_Page_067thm.jpg
c52d75b3c0ea6c3ce40e116bbf481912
a6ca551bb93f31298d8ce504cfcd229fa35691da
34851 F20101111_AAAQKV shishodia_v_Page_074.pro
e0aa165543f4b084d73e0545c8e8df62
ace8b0d5709e0320edf7d472b0e6602dae855f7a
19929 F20101111_AAARNX shishodia_v_Page_022.QC.jpg
97f58669d49c469de9b940a529fc874a
34fd5e5138f53d729fe639301119c52be01ffcee
5253 F20101111_AAARPB shishodia_v_Page_040thm.jpg
512c7a03403697c634387c34a489d889
a6280f9de0b9d295df9870ce6d5992c0ffe79a15
3962 F20101111_AAAROM shishodia_v_Page_030thm.jpg
ab32f9e544461508658879843ce3b1ef
201c17ba119de90c3d97991d444223fe7062d9ae
F20101111_AAAQLK shishodia_v_Page_101.txt
01d85dcb51a9ce7d74f3120bb819845e
1e5562a02a5f605c4248c4214db7ef545b8f8c2c
76880 F20101111_AAAQKW shishodia_v_Page_066.jpg
5b27768cfbd8648837b9cc2e1b4dbf32
fbe666231978eaaf198bcb9cb5706176832e08ed
12916 F20101111_AAARNY shishodia_v_Page_023.QC.jpg
ef12cc1ed1f78e22c14eb0a425f83e07
dc0428f239b40f1f52cf53cfb5c47d6aabf269ce
5399 F20101111_AAARON shishodia_v_Page_031.QC.jpg
6ccc80010ec5c8a77ed200b3ea50ab4b
72ed52b6524b2d15b707c75616404a4a421e6ae3
4494 F20101111_AAARNZ shishodia_v_Page_023thm.jpg
6600a9028658a43ea937fd01747e4b63
f73bb3f738c1ac3760379c6ec9b8f3fb584e4298
F20101111_AAAQLL shishodia_v_Page_126.tif
95dcdb4c11d640b3c4d270ec2b0305cb
c9f3b1b9f7c72730d626067ed60e48ce65dc78ef
1051985 F20101111_AAAQKX shishodia_v_Page_047.jp2
2722c3de7901e136c75760250b07fe75
a7336dcd0f1fde9f7afa0dc0805705838472d3ac
15901 F20101111_AAARPC shishodia_v_Page_041.QC.jpg
7cb4a649bcb69f2f2c94a68a0c740b49
ac23ad6e600e42a285a0ddbbbcafca0375734c7e
2167 F20101111_AAAROO shishodia_v_Page_031thm.jpg
3b21a6d0c5bc1b704fd90a31f4ae5e82
673ad32ab855c1e6e930deb87f5dff4d1083be3f
F20101111_AAAQLM shishodia_v_Page_134.tif
bfca070ef47de9de84297e37be269f77
af7827257c301d25f728214d266dab226c563151
F20101111_AAAQKY shishodia_v_Page_002.tif
ba494e044b273638dae74fdb64f85f3f
3357872a2cd30289ec01f2c03dcf225b024b8821
2170 F20101111_AAAQMA shishodia_v_Page_089.txt
c6ea4012b862a62e4fae6c6c8ddb642c
12683ef7b07b78f5d89759c9dd73231e20e4eeac
5074 F20101111_AAARPD shishodia_v_Page_041thm.jpg
048d19f0d601f31fb9fec420ef0742d0
3ff6f3058c39d9b784695d9d8314f97c8b96489b
10984 F20101111_AAAROP shishodia_v_Page_032.QC.jpg
0ac39abeda75d17eb689f98ed53b6435
ae6cc0e4ca708e6df1985dd71c4147802740343a
6665 F20101111_AAAQLN shishodia_v_Page_064thm.jpg
dbf974f5627c6f5cf20ced6929dee223
ce133475e0f274b0cf10d37e952610f50bd06ca1
17661 F20101111_AAAQKZ shishodia_v_Page_073.QC.jpg
ce696481ae95bc135cc18d830036a425
f51aae307ac9cd4afba48f09e053e2404e424a84
22388 F20101111_AAAQMB shishodia_v_Page_011.QC.jpg
05fdda96714606dd1ae16d1ff2d0cc89
bf532e4accb2164174d76b8284e59bf7a050697c
27634 F20101111_AAARPE shishodia_v_Page_042.QC.jpg
c87d302bb708a06bfdc633ece687d4a6
16ba712113281a0f1daa8f125b0d839f5cbaaa20
11416 F20101111_AAAROQ shishodia_v_Page_033.QC.jpg
51fa1a8f3ccd5f9044ad8bc9d8a2607b
83225163b52cf3f1daf34bbb61fe226e8911afc7
3358 F20101111_AAAQLO shishodia_v_Page_053thm.jpg
b803205389a0f1ed26b820800f4789b8
2ec3a3716d47e7430e083fcac86ec1c5a019e179
F20101111_AAAQMC shishodia_v_Page_105.tif
857630803c39932b96fceacb7df44286
e1ef5a0483af8cd4d0d5563c280e999c447a8e26
7180 F20101111_AAARPF shishodia_v_Page_042thm.jpg
2357b327c8da50938728e41ba366f23d
a16d2665fdaf25fc2d62534d0f27c98403a3096a
3491 F20101111_AAAROR shishodia_v_Page_033thm.jpg
a26cd1f537f81010ace21a3f77401ae6
315c9c64bfa310a789e8ba6f52a913a2b7071782
F20101111_AAAQLP shishodia_v_Page_003.tif
c2fcac3b2a1f644286abf41ff7c40f10
622c9d4d845b4d55c4323b6ccd2e5323eac50b81
2052 F20101111_AAAQMD shishodia_v_Page_028.txt
f494ee7213717f72bf7097ed87b977ad
22628aebb17d5956e89b68c8e9794fa808817748
25655 F20101111_AAARPG shishodia_v_Page_043.QC.jpg
264f341f879d3d3d6061f70002721256
09400ca2382e1a56e56d7669595f5adca4fd1e05
12840 F20101111_AAAROS shishodia_v_Page_034.QC.jpg
fcbff93c2c010fc95e4f1458adf3f78c
c3ad1464d335f6d686ed0a7868dd31930d70d389
27162 F20101111_AAAQLQ shishodia_v_Page_112.jpg
43e14ee5bc9c65809f540ab7584a6ecb
3e45eb518891009c63e545cf9f1ad35f993c5036
4620 F20101111_AAAQME shishodia_v_Page_134thm.jpg
df6d78204a04b2e70ec3528d54b3c39d
dad1a54f61f195cc33197f55fb418497da518491
27387 F20101111_AAARPH shishodia_v_Page_044.QC.jpg
58331e7c5dd2d88b816f9f71a2243476
47507f5b1ccf499aa5ffa124595abf5155d93aea
3778 F20101111_AAAROT shishodia_v_Page_034thm.jpg
bd59fdbe2ba68c779be1c25820501780
33c9e04501d50b99597a324422b6012ba2e2d7bd
71412 F20101111_AAAQLR shishodia_v_Page_095.jpg
190bd605bac7641fb4fae009319db7cf
049a92e1f014acce2b838b54e71312f4205f383e
F20101111_AAAQMF shishodia_v_Page_003.pro
01b30d7d317026aeeaf02ab6388083f3
94e4630341dc6e4450781a40d8680366af66cd30
7682 F20101111_AAARPI shishodia_v_Page_044thm.jpg
10d0611796fd115da39254c62fbca14c
e033d76b8d3a2ee3df068aba3f466c70c4cf44f0
12493 F20101111_AAAROU shishodia_v_Page_035.QC.jpg
da1322677246aef2b019b7f8234a4511
919651bf7ea9b425a6eda6580a86aefe086037de
24558 F20101111_AAAQLS shishodia_v_Page_030.pro
fafdedc8569c0069903ea76e0c881fd3
ba10683585a79b06d00fa0d6a04747c29ee84d1f
F20101111_AAAQMG shishodia_v_Page_024.tif
6d0a57d9db6f6a50fc2f52acc78780f8
21768fc730b271b0afbc1e1e9794b690d4f9aa85
26714 F20101111_AAARPJ shishodia_v_Page_045.QC.jpg
cc476dbc174d2652a71342641ef7590d
8dbb543413e6d3bb15b34ee73ce52fe7610eed85
3840 F20101111_AAAROV shishodia_v_Page_035thm.jpg
757ef30270863a26bfb3d08add7be65b
cee8b8681e188f666237f2723a9a8f8b18f64640
3035 F20101111_AAAQLT shishodia_v_Page_112thm.jpg
d0eb045fae876c6da91ed066c1f08c89
4b8e3f479f3ad0a5b8fe4036a6ed39d0afdbd95f
6178 F20101111_AAAQMH shishodia_v_Page_022thm.jpg
ad57d6ac2bffd57867a05f6ccd047d4f
116e5df4b9f63084c1b964edc41c6a88186922e6
7459 F20101111_AAARPK shishodia_v_Page_045thm.jpg
4259f0bbeb4bea2318505b0f9905f6db
061f52fad02cd135dd6be5ea03a48edceac03bfa
16892 F20101111_AAAROW shishodia_v_Page_036.QC.jpg
87fdede59c266ca8f4625db1c93b8dea
08823e758f0148912cb889f5f602de8acbd50d83
F20101111_AAAQLU shishodia_v_Page_069.tif
319e234f06cacf22602cc02fc7b53cd9
1f33568d385eed0fb9d964cc316966a512317258
163858 F20101111_AAAQMI UFE0021944_00001.mets
8ed8097ee13c55a23420c9579f3b3d7a
db2bbe40753fdaaefff6aade5388dfcc58986b3a
10238 F20101111_AAARQA shishodia_v_Page_057.QC.jpg
85248dbd489fdac8f1bf1a361adb82a5
f941c916693f5c3748c9741473aa6e79f73c8146
28521 F20101111_AAARPL shishodia_v_Page_046.QC.jpg
664e1386e0e6f83bdceb507a4eec9bec
48a676eab83b6c761c8400349120c6048c7380e5
3567 F20101111_AAAROX shishodia_v_Page_037thm.jpg
3901cb300e16f4456c425a13e582a6f5
004e5f240c667443eb901f4953d8f83d7a12d3ed
2197 F20101111_AAAQLV shishodia_v_Page_061.txt
940857c40b5f5e09aa95a65da0dffd0a
5aed0e2fc250bb6054e2488e625f9231b63e8de8
3332 F20101111_AAARQB shishodia_v_Page_057thm.jpg
25efdc9d88c86c28074bdb9689f7b354
7be59cfdbb49a572df0b60fad21f9b84cbab26cf
7637 F20101111_AAARPM shishodia_v_Page_046thm.jpg
f82a899b7332bd5419644bb753435da8
3349cb3ef37dea72c34923469392b445ce22fd91
7037 F20101111_AAAROY shishodia_v_Page_038thm.jpg
9e6cbcf39f52de0c8db7e6c2b134c900
a7a05da2198f6bbd2ab9b147dfa722837db5a0b1
25648 F20101111_AAAQLW shishodia_v_Page_082.QC.jpg
ba27bbb5ca454ba549224e6c1fbdc92d
3a576b942efd756656c8bc2065c73906c9c38229
10060 F20101111_AAARQC shishodia_v_Page_058.QC.jpg
486c83c1e599a92300561ddc93bbc1b7
cfcd3cf8bd034b545d1e4b0c02bebb45232097af
26608 F20101111_AAARPN shishodia_v_Page_047.QC.jpg
7c8d3d40589dd4dac92e8735b4b48e71
f0c008df3d76ece3a047c6d312470f3d39b80055
7199 F20101111_AAAROZ shishodia_v_Page_039thm.jpg
d04d7fb55481dec00c956f0f1ac40708
a3e3edacc28816c5efb92e410e3e7d2fad0db539
552365 F20101111_AAAQLX shishodia_v_Page_030.jp2
62c14a6e3c26aef6781b795cd2053bd3
283d54b06c95ba2dd1bb51983df3e36c1574a856
32049 F20101111_AAAQNA shishodia_v_Page_019.jpg
bd6220966f9e862b7043b3cf7d5c2066
5ca26ca78a851e42ae707960a4f37894612f6f39
24425 F20101111_AAAQML shishodia_v_Page_001.jpg
e03cf0a5dd527c7f8b8e0700f6b97add
43d5a8010c159b71a348c1df95dfd145767a2c69
7240 F20101111_AAARPO shishodia_v_Page_047thm.jpg
b21c2fdc660f5d81c9b4fcdc0eee01c0
3bef1322fb73dc1adf6f32916ac8b142d04345a5
3370 F20101111_AAAQLY shishodia_v_Page_054thm.jpg
054f3470fd687125d1f8972041fee208
2c37b0d743d469766445d60d9b8e86606a25d780
51753 F20101111_AAAQMM shishodia_v_Page_004.jpg
e881bacc095ac28ede4b2bb6e194044a
09572d9d26163f5d5502c035b2a655da0e3f749f
3314 F20101111_AAARQD shishodia_v_Page_058thm.jpg
809e69f0bf58ff77b1647ebba707a31d
1c4fd3f5caa56130e5e00cc5b4a3dcdd18b9eab3
6934 F20101111_AAARPP shishodia_v_Page_048thm.jpg
138011d260f02de8d5489d6937fb1a9a
2123795f405bee781c3dde55313006f219e3a733
78 F20101111_AAAQLZ shishodia_v_Page_003.txt
09695e5f20129a197cf49f3f7e0367a6
559f4561a5685d0ef6b1f1b71ca7806cf3f885a9
45919 F20101111_AAAQNB shishodia_v_Page_020.jpg
718d1adee47d1a1e156efd78097d4cd5
589024562df03b8cbc15b4847910403db6989eba
81067 F20101111_AAAQMN shishodia_v_Page_005.jpg
7a91156041a4ca41678acad2acdd80af
7fa52fff91daa63ab0d06f7088c5fb57fcffb816
11507 F20101111_AAARQE shishodia_v_Page_059.QC.jpg
9b01454eca1d71e244c8d0ebcf35f3e3
37820d0ace1281de1d511690dae4969241331eb1
7692 F20101111_AAARPQ shishodia_v_Page_049.QC.jpg
4f7d235e701646dcae91db846ab50798
b8674bca8e264aa6bba19f4f3fdc724ce38df636
47437 F20101111_AAAQNC shishodia_v_Page_021.jpg
1a1e810f7c4a29fa590c1caeabba9604
6f66e6c8b2df4b46709c548fc79a68b4f1523c87
81339 F20101111_AAAQMO shishodia_v_Page_006.jpg
537f23b0aca324ff9c718735d042a8a8
52c7587bbb57077957c169a48e6bd52cede89700
3648 F20101111_AAARQF shishodia_v_Page_059thm.jpg
6a66a3a4dff2ff6d856a427f8345b20e
d8e05780befdbcf813a5f888a83e404fd2dd88f0
2670 F20101111_AAARPR shishodia_v_Page_049thm.jpg
7f2e8b7bc6e16b859fe12b49737ec0b5
45866d88cac798aba50429666897b5e3927f7ee2
60304 F20101111_AAAQND shishodia_v_Page_022.jpg
ae1279f7382ee03ee0156a76ab618f7a
4f3246294110e47704d1f1ddf36d3b19c4fd8cb9
21611 F20101111_AAAQMP shishodia_v_Page_007.jpg
505ee7ca5e3f1ebf2c7b0e8f2f2af4ce
56056c0098a5b84accc075711198a6fa817697d9
14695 F20101111_AAARQG shishodia_v_Page_060.QC.jpg
b31fc648a47a65420e95a62f59cb6738
6ca0ad102f0c9df007ff0976a0553848a555b3b9
14318 F20101111_AAARPS shishodia_v_Page_050.QC.jpg
84fb21758943446be8ace3071ea2cd38
26b9f2474a194d86447b2472153ce66f9b8ffafe
41872 F20101111_AAAQNE shishodia_v_Page_023.jpg
2820408d8461ef2162e204cbb2d5235b
c016359f7fad8e53f433424b6892496e5235c6da
86508 F20101111_AAAQMQ shishodia_v_Page_008.jpg
dc477491c4b0fe51a7d626e7fef0f9dc
3a8f48dc12fea94ffbbf0c7db5bed0e8971ca489
4269 F20101111_AAARQH shishodia_v_Page_060thm.jpg
1cff5337cbb40981cfcd70e95e48f225
2998dd2f5f30c882e92cc798d3780c8633b94def
15253 F20101111_AAARPT shishodia_v_Page_051.QC.jpg
c1e9011dc5e45802d8705c80ec53c0fe
4fecafd27062176b5477a474cb13a1e94f28997d
56617 F20101111_AAAQNF shishodia_v_Page_024.jpg
a5ce8a1372fcbe1cc05891e5363e50bc
3595f7be600ec13ea271b76688918b1b01748209
112951 F20101111_AAAQMR shishodia_v_Page_010.jpg
db7ee8c97f3338cbca2163095c07b1da
2ab167abc02b5a63b8913b0164cd078235598772
26831 F20101111_AAARQI shishodia_v_Page_061.QC.jpg
4eb273be22571efce5f967f610662930
f42ca55f8e8f5a0ff1e0b7e71d8c96da35532279
8715 F20101111_AAARPU shishodia_v_Page_052.QC.jpg
db90b392340d7f6f07503b15fef0311d
28aa604bdfc0bc85995385831d975c778013fc5c
68610 F20101111_AAAQNG shishodia_v_Page_025.jpg
9a46d67d0ab8f527d08dce2f9e2a45ed
67c470b08fa3698783a8717844efd704cd88a3c4
75629 F20101111_AAAQMS shishodia_v_Page_011.jpg
1981c7868d119801771c52c9f7802589
ffe0205eedc0ee60bc62b3550903b8c4ad16e5a6
7219 F20101111_AAARQJ shishodia_v_Page_061thm.jpg
cc2dd6ffe93a8104fd0f6ffbfe15e404
1a96ad6351c4429d4a07db8574beee74764d6d05
2767 F20101111_AAARPV shishodia_v_Page_052thm.jpg
2398bf13e7d331ca8efd9cee25a7950d
078122c1f813dd56ce409698ddc49930f02fe715
79565 F20101111_AAAQNH shishodia_v_Page_026.jpg
261770e6cb564fc371f1655a3f26eed2
4d5bfd4bc6b194105b649b756945223c9ea42f83
39560 F20101111_AAAQMT shishodia_v_Page_012.jpg
8b603e40f50cf9bda4634dff36051805
be32b169bac3aea0d550d9129d2cc3d5e08f6fad
23654 F20101111_AAARQK shishodia_v_Page_062.QC.jpg
7d0aa89309e6f1f97fd297a8ecaf0a05
edb8091b73d8107dd5ec2c9cfe74138f96fb1d4a
10493 F20101111_AAARPW shishodia_v_Page_054.QC.jpg
64481aa149a9f9540cb7fca07d4d4723
9fa61949c069777b71adc754c4ab4c2c6b627c3c
87490 F20101111_AAAQNI shishodia_v_Page_027.jpg
758c441992c06c040fa6e9290ee28c1c
3de892da93477ed132339e734dd11a410f2e5c8d
84619 F20101111_AAAQMU shishodia_v_Page_013.jpg
dc474e790453126a8f2f32676042a64f
cd43a746271998afc532a7eef68f278474f5bdd0
6390 F20101111_AAARRA shishodia_v_Page_071thm.jpg
b42043544491b613a4f4142986625790
b67f7f2efc6f6e60ca8d48225652d828c10eb1d1
6904 F20101111_AAARQL shishodia_v_Page_062thm.jpg
85a18e996d26fda5fb606717698ffccb
227d20b9e2e817c545ccf18b091be4e3fced7ded
10447 F20101111_AAARPX shishodia_v_Page_055.QC.jpg
2b4697aadc4d39fb75138877bb3e3c99
2c8cf1824d20c4a2922c30ee2bdd207350610d8e
84198 F20101111_AAAQNJ shishodia_v_Page_028.jpg
669be1ef4773b1ca2619f6a62907fe8b
c4c750f5bed9096ed2edb143019e26ec657ce0bf
55965 F20101111_AAAQMV shishodia_v_Page_014.jpg
02558b4b147470f41d4252ab998e0b37
d491b8f38daa55d736713af82759917c71269877
6377 F20101111_AAARRB shishodia_v_Page_072thm.jpg
5defa4159f94317ed2f4632d0b642122
a65d7617b2ce13d94f6033e03d0787698711a267
26695 F20101111_AAARQM shishodia_v_Page_063.QC.jpg
a99825efd30715e1f1afa49a98aee327
b557f656b879c4cdeb0ad3098c006d2debe21b09
10440 F20101111_AAARPY shishodia_v_Page_056.QC.jpg
dcde048a172dc52ba158dd6b1970acf0
9996b43a1edfc3c1c635053ee808a05d4ac4b545
84565 F20101111_AAAQNK shishodia_v_Page_029.jpg
c9fc6a5df305cc60e5cd3b353ccfc485
55c78db2a28171bc4fba2063986fcff35b1d3f3b
85837 F20101111_AAAQMW shishodia_v_Page_015.jpg
00da6b7552287eea61110367b49210cb
9b1ee44d9fff1ac5b4e587f55652b80f152a1622
5558 F20101111_AAARRC shishodia_v_Page_073thm.jpg
25f124a01d31093eda51aa81c13addbd
1f7b56545c22a6d6b6253594a0da8b619cabee72
7210 F20101111_AAARQN shishodia_v_Page_063thm.jpg
06d60ff82a643a8406224cc24b6a955d
eacf5d2013e2dcd137cc159ee38aa8a08791d547
3393 F20101111_AAARPZ shishodia_v_Page_056thm.jpg
4ab4577952d436c52fa3b172de3eabfc
438147ca1447bc19fa219ebbff03989fbc9cba60
89050 F20101111_AAAQOA shishodia_v_Page_046.jpg
874b2624f1d709cf2e4950a9d635acf5
c73af1b7944df365faccce1685519ae70accd33a
43312 F20101111_AAAQNL shishodia_v_Page_030.jpg
abf95991d5146ed553b140033e3a2695
10f62864c8701d1e45d8a7f8bf119d851b401e68
81915 F20101111_AAAQMX shishodia_v_Page_016.jpg
acced44c7f9c6c599405b88cd4771c4b
a23b92e8f880bb2511c71f71f14e0d518dd0a50a
17878 F20101111_AAARRD shishodia_v_Page_074.QC.jpg
837861255f5d9e1fe3306065bd268340
114a6d0bfda30354242130f7ffc34f4282002376
23017 F20101111_AAARQO shishodia_v_Page_064.QC.jpg
d76bb68d08e61216d8c38836457c8a00
b093e942a60dbfa621f625c304e5a6f5a00322f6
82423 F20101111_AAAQOB shishodia_v_Page_047.jpg
36e40fae42c85d499455dc87e961e9c9
b5995cbc7ca71e80c4276ce4d688eb272d0bf5a0
16268 F20101111_AAAQNM shishodia_v_Page_031.jpg
18d5fd9dfe4db342834e05a196445265
38558fe458be922d7bd18e06a0b3d6cde725f88b
84338 F20101111_AAAQMY shishodia_v_Page_017.jpg
f894904e4f3987e37c0a72785c95abf7
0d60fd47da9879d7c5a7e941e1b3d4f2eaa54c79
17319 F20101111_AAARQP shishodia_v_Page_065.QC.jpg
085a37d4ecec214b6c455430f6f82155
f764420fdf227c0196a72b111411cd7982ddb42f
36325 F20101111_AAAQNN shishodia_v_Page_032.jpg
f58e9ece4644ff8ebcbdc2c98bce7aa3
307e52e6cc299229c1bf2f1bad6377889d52c00b
51120 F20101111_AAAQMZ shishodia_v_Page_018.jpg
e2a7e7c44c31fc2c3437d2ee12578cba
19044a82f7121758e70cf54091bcb312284f9622
5630 F20101111_AAARRE shishodia_v_Page_074thm.jpg
165189633e464d6dbf4e2c349539b864
0a4be05c01c53122be8995152641eae57510e6a5
5605 F20101111_AAARQQ shishodia_v_Page_065thm.jpg
955f152839fc193bd796317ce5fc92c6
1603f1976a36c4d4beedcc2d01988cafede79522
77585 F20101111_AAAQOC shishodia_v_Page_048.jpg
590e3a355ebc7accf9b303d3593b7e16
2a197f407113a64490b12cf22481a74b6be965fb
35845 F20101111_AAAQNO shishodia_v_Page_033.jpg
290106b4357227eb3a888996fe61fe5b
85e0164c9936d4467da73e5332db1402f847f26f
25006 F20101111_AAARRF shishodia_v_Page_075.QC.jpg
f56c62c201bcbfdbf83ce2d8d5d8a916
c8e07e3875bc5f5c332ec569304434bf97e9294f
25287 F20101111_AAARQR shishodia_v_Page_066.QC.jpg
b9a94c2e9541b0c4ce018fb62609056c
a20b80300d4b467f4a7c60c8b9e32fc73545840d
22798 F20101111_AAAQOD shishodia_v_Page_049.jpg
b787174679419e79b88a8fa92f00c7e2
0f6c4ec1e79e7e35fb6b5a07359caed20542480f
38525 F20101111_AAAQNP shishodia_v_Page_034.jpg
2fe22dcfd7727d873f6ad48880ba8759
b9fae046d950cb29f9f2a59b38147c75960ca006
6973 F20101111_AAARRG shishodia_v_Page_075thm.jpg
4b1630bd791d770c64e427b9f14391dd
b3aec5913e6d4718615a9daa23c74ffaafa4d479
22271 F20101111_AAARQS shishodia_v_Page_067.QC.jpg
0d694668f3f384f6927ba78ef1d95610
8d5a1c7c0c9354b814f372a05880981d993cefb8
53362 F20101111_AAAQOE shishodia_v_Page_050.jpg
36908ce598d58f63610e7ac74d8f6fea
535ad9c469357570cfd56cc4e1bea1778d2d9dd1
39839 F20101111_AAAQNQ shishodia_v_Page_035.jpg
5d23dd6451a16e17e085bca006af2a50
b4a14cf6cbcc1b67faa5ab0bab0fc66c0dec4039
F20101111_AAARRH shishodia_v_Page_076.QC.jpg
de1491a28fe1e421ac0ec7f140256a63
f58fb1e8166ba40eef707849ad999676906237bd
17896 F20101111_AAARQT shishodia_v_Page_068.QC.jpg
baae5e8a36b42e1ae527659c95dace50
4ca422edb88740a25895ffd891fb03c174b57e20
57561 F20101111_AAAQOF shishodia_v_Page_051.jpg
784c106e2a952ab9b18cf6f01c66374b
e3a369ba58a23ad55f5dea651d158c67dd5eedbf
57358 F20101111_AAAQNR shishodia_v_Page_036.jpg
bc82557950780504545f767f3fded38b
2e31312e3585d3d998f9efc9e974c37f10eea247
5467 F20101111_AAARRI shishodia_v_Page_076thm.jpg
936136d5a8017eed679007ee40edac4a
945acaf16aa6b379bdeadf659f938a5607d42a58
5614 F20101111_AAARQU shishodia_v_Page_068thm.jpg
6875be6b47340cb6c40b37e983bc9d6e
ab6a1561a62ade13d766b06b5ceff01586f4306b
28465 F20101111_AAAQOG shishodia_v_Page_052.jpg
8af07d4e41686ccd1d2d6ccbbda29ee1
4ef93cd86a0a578d38cde9f1e85e0e40fc88311b
40048 F20101111_AAAQNS shishodia_v_Page_037.jpg
948f93a8bad906a1619c294589003435
097fdc7ee76de2da25888d2d639ff0af0dcba0c7
12259 F20101111_AAARRJ shishodia_v_Page_077.QC.jpg
76c9bc8410135e1cfebaec1ea00691f1
fcd0b1980f8cdca24b87db361124ac9d3b198e9a
18618 F20101111_AAARQV shishodia_v_Page_069.QC.jpg
724d3a08ad47f8192590e60fef695bb8
a614f80103fed8a86e75c52ecada6e957ec92e4a
37578 F20101111_AAAQOH shishodia_v_Page_053.jpg
8326e25245b14cae4e4e257f7d080200
a3e641dfe98ffc7c133f1a6786cb09bb2ff60e16
82859 F20101111_AAAQNT shishodia_v_Page_039.jpg
f0366a4b56da953e2171038f46d2891f
b7d1de966eb4aeaa545418d14ae8b630a337c295
4163 F20101111_AAARRK shishodia_v_Page_077thm.jpg
a5e5ab8d64e9cc335c8774e826c95811
0d1e7a3e8aa51f38d5ee9a443bec9f1247d6a698
5447 F20101111_AAARQW shishodia_v_Page_069thm.jpg
c07980c2b5e33a7b11e9fe3c7cca4564
e19f29d68bda80f8747e89381c7e501ab6d1396f
30693 F20101111_AAAQOI shishodia_v_Page_054.jpg
e149f18ef005de67a1e1f054b29dbee4
234ff1892b9c2f7abe4de5033b127a36ce4cd029
51534 F20101111_AAAQNU shishodia_v_Page_040.jpg
4ecaabb985c9f4a0451a6fa6fa5f0099
186d937973bb87074c54e14ec6128efe2680e8d9
24956 F20101111_AAARSA shishodia_v_Page_087.QC.jpg
e39d39ade2e627289bc497d22ae93a73
12177c2e74e10b8d6955553ef5a406f46785196c
14465 F20101111_AAARRL shishodia_v_Page_078.QC.jpg
c37dcfce5ce8824661e16ac000da8030
adb6dced8013e64d7239e5668f68f8a489ec73df
21337 F20101111_AAARQX shishodia_v_Page_070.QC.jpg
91a99e05f7a1881229c76dfbc7a4e6f9
212312b88ff6b7ae56f33cb47b5fb02bf99c16a5
30580 F20101111_AAAQOJ shishodia_v_Page_055.jpg
83807325527913327369f313c1d6ee93
785c639a968386749e4b9863130b7ddc9c062cee
50173 F20101111_AAAQNV shishodia_v_Page_041.jpg
587b4d3851fa5b0be6a941e7493326d9
aea47d76c738e0e7d0f92a0cb57f20377c6e432e
6848 F20101111_AAARSB shishodia_v_Page_087thm.jpg
42b4e2d37e278197efca224b002ee0a6
240a454252591387a83119de31dfc9895cd0a08a
4729 F20101111_AAARRM shishodia_v_Page_078thm.jpg
1c17e1c6c4624259f9b56874b55405ec
c2705e16132d7f1b1589bb6a77f74cb5ea9b44bf
6489 F20101111_AAARQY shishodia_v_Page_070thm.jpg
304491fca0c4131fbbf5679f640fbf52
4b76d453a42c7bdf2911fde3ad2c0e128dbe407e
30304 F20101111_AAAQOK shishodia_v_Page_056.jpg
a336767240a95789da97b54b153b0ee8
670db53f78c94840739eab706bc0ae83f27785e1
84882 F20101111_AAAQNW shishodia_v_Page_042.jpg
677b9a146adc8b2bcdb6c6b840a22d47
2a1729acbdc11829eed25e08c526dd93e6674af9
23802 F20101111_AAARSC shishodia_v_Page_088.QC.jpg
3c514b479268788fbedfe56de395659b
fb9daa5ea7f59bc8b32476dd79f4ff1bd8d765ac
13788 F20101111_AAARRN shishodia_v_Page_079.QC.jpg
76f69abbde25c70de3b546424922019d
eab9206d248af578ccd89a285b97cc6ca4b087b7
20612 F20101111_AAARQZ shishodia_v_Page_071.QC.jpg
803a048c9ca938c158c68ad663ee6e3b
912afde040d1704357501b51de10b6df15badb11
30223 F20101111_AAAQOL shishodia_v_Page_057.jpg
dcdf99c7d373996afad8f21f4d5a2f23
783687f77226979ad8efda4d2b62bd298b175a6b
78856 F20101111_AAAQNX shishodia_v_Page_043.jpg
8eace395d55f2bfa688e8052d7fce141
d5c63012d44c7be6ff62651f1043c09f9a8757b7
57890 F20101111_AAAQPA shishodia_v_Page_073.jpg
45f5e9ea5c7f1242b4a219ee75cbcedb
3f23fcf9bcb4378e4981ceab0d98bf412bcc203e
6578 F20101111_AAARSD shishodia_v_Page_088thm.jpg
e2e54725518a6b0943951cd59293bc70
a12996e5f2686c78875ada7ed5491c32aebe2127
5132 F20101111_AAARRO shishodia_v_Page_079thm.jpg
9981dba1f9b905a9284add7a2d46c3ae
81525f940008463197ea80df942cf8c2cb6c05a2
87576 F20101111_AAAQNY shishodia_v_Page_044.jpg
85120b41b70a72d48a75e647209acf0c
71f1686353912aa2d04e17f57670b429476a5eb3
75730 F20101111_AAAQPB shishodia_v_Page_075.jpg
1f25bd44c6f68d35e9155ceebab583e8
899de4f3ba09aed2d49fed4fb6a78843e62c105d
28888 F20101111_AAAQOM shishodia_v_Page_058.jpg
69b586f9b10449cc4dd86f0d4dc75e78
0a7b364e3f60b10c0a42ec7bbb1b4401c1fb6086
27195 F20101111_AAARSE shishodia_v_Page_089.QC.jpg
ba231d2da9301e8c00acbddb05e17eb3
4c01c291573f33231409abd6394670e8a3a39c97
6103 F20101111_AAARRP shishodia_v_Page_080thm.jpg
047e74d979356ff0eba62fdf399c3021
2b6811530d3ade1682d2fed92fa44d7cc1fca2c2
84231 F20101111_AAAQNZ shishodia_v_Page_045.jpg
f3007b9fc5bcdb54c8406f1b1e0868fe
f3376448e9aed3cabd1dc33f4556fdd12f679303
54781 F20101111_AAAQPC shishodia_v_Page_076.jpg
611ecc4111c1ae2801a28ee2248f823a
f05f4a1134a58cfa7ed4275c4af338688be16e0e
35253 F20101111_AAAQON shishodia_v_Page_059.jpg
b54269f2d3f498b0460930f70c829757
f3dd807b9a7272eba0e1094519a205276ed9eb05
21250 F20101111_AAARRQ shishodia_v_Page_081.QC.jpg
c8b69bc5f29ac1b2089af1de4b044d28
268f174eaceff91e7f3e4e9a4273b5bc09cd8d64
55581 F20101111_AAAQOO shishodia_v_Page_060.jpg
bea9577cb661574276823059cb1d6433
f5d9e4fe0db8aa85ef1e72c95bc33f0ff0428927
7400 F20101111_AAARSF shishodia_v_Page_089thm.jpg
b82981aa65d38d2008ad8db191eaefbd
5136db487b88e7153c700d715858c07eea32a3ac
6350 F20101111_AAARRR shishodia_v_Page_081thm.jpg
292ba069d14fe63d18be31266b798456
28e9eb40162195e1314d5792469867177e312e0d
39423 F20101111_AAAQPD shishodia_v_Page_077.jpg
2e4faafe435fe11966dd913e91105bde
084c85c6cc442aa45a7a103eaffeead793eb7284
83732 F20101111_AAAQOP shishodia_v_Page_061.jpg
5c305fbe40ccfe32c8735ee3a1974ada
52e06dad5d1cfdf0e5a98db85826155897c23824
26835 F20101111_AAARSG shishodia_v_Page_090.QC.jpg
eb67a942372ab71a707873797e25ff52
108534abcb1a39f39c87bad92135d036b830da6d
7193 F20101111_AAARRS shishodia_v_Page_082thm.jpg
59afc19f835dcbb2d08bba4580c10381
3edadbcefaf16ed68f127fae55402fcd5a680016
50265 F20101111_AAAQPE shishodia_v_Page_079.jpg
fc82f645115b84f7b29db630347c014e
2ae4f7fc90c454a67dfb036194c46b3399fdfbee
76325 F20101111_AAAQOQ shishodia_v_Page_062.jpg
13dc10ba7ddc98215114248e64059984
bda38f1cdb9b186b05632fb52508e2a3660b4b99
7340 F20101111_AAARSH shishodia_v_Page_090thm.jpg
dce2cac8da71465e3d67704907782af5
5cb9f5dede00d4a71184ebbd4dbc8f0ef6e65f5e
27676 F20101111_AAARRT shishodia_v_Page_083.QC.jpg
cef472d5021a421e2879d3c5b4a9a2f6
7c377bd1dbc8c2e649dfffb1c8badd6bb3aab1b5
64522 F20101111_AAAQPF shishodia_v_Page_080.jpg
6077041d18e5106634e5e48c4dd6cafe
51aa5af195bfb5d8331b45450da97cc217ce049b
84188 F20101111_AAAQOR shishodia_v_Page_063.jpg
9a4b245735f41faa7a860ae74c2be50a
8626ac6642f1397edfef5830ba33a6b9d88a3db4
28167 F20101111_AAARSI shishodia_v_Page_091.QC.jpg
4a9e13bdc35497bc6cb5b93683a4fb2f
195cd5f060bfe3af5c10299584b96bfd5af876dd
7581 F20101111_AAARRU shishodia_v_Page_083thm.jpg
7c9b6d678c12ab075b5ff680078b1e7e
37a07f64999c2f46d94f4f82badf4e110054fd85
70669 F20101111_AAAQPG shishodia_v_Page_081.jpg
c4f106072e032f306abd1969f5ab0bd3
9c8164687801517cddc527a032da23e88614dee9
76232 F20101111_AAAQOS shishodia_v_Page_064.jpg
7f3f4fdd2a5c76e7b57dc6aa5e87bffe
e17c1e41f6d422743c666c96a56edfa37f4d50d7
7447 F20101111_AAARSJ shishodia_v_Page_091thm.jpg
63a91f506aa81b6fc2d7d9ac65d00248
8d9c82e33c6ad6c156f332c50b5a6e229fca519d
25834 F20101111_AAARRV shishodia_v_Page_084.QC.jpg
875e9020449bd6c4c6db93ad970e9de3
b47630e88c4c700d6d02f2617847054c07f9df08
86557 F20101111_AAAQPH shishodia_v_Page_082.jpg
d9e41fd562df76555b457b96ec7324e9
6d6cdbf5b14a5ebead4efffb3dd1fb4f727d0547
54146 F20101111_AAAQOT shishodia_v_Page_065.jpg
0eb8c8d48b519a246627c145c435b359
81848e9e67f92ce20cc6c1e10c3f1c170793d230
24080 F20101111_AAARSK shishodia_v_Page_092.QC.jpg
9b07d45627ef7f73f39449001ca9c256
d22709c36af54d89f2abc9c68580a5ce6c93ee9d
7107 F20101111_AAARRW shishodia_v_Page_084thm.jpg
8d07621e9d4deba95d5df0f21976226c
f9efdfe818a01a7aad6270540b4808f652b025f0
85916 F20101111_AAAQPI shishodia_v_Page_083.jpg
615fe722c2ed7793d84f943de24ee968
7f1effbca92c9192614ad29ca69e5b73a51edc8c
70184 F20101111_AAAQOU shishodia_v_Page_067.jpg
194efbdbd14a5a2cfe3904da3b83ce1b
ea7030db9a60cd0ba7d01e7e531a01e98eaa992e
3317 F20101111_AAARTA shishodia_v_Page_101thm.jpg
840209c4312886a72f71667777f2572c
1289b74d45ec6085f54f7aa7e779c46df94dad89
6483 F20101111_AAARSL shishodia_v_Page_092thm.jpg
d3a257d5a4b7c1775a1dcd557fea43a0
c220175193efed96bd991f21606f4ffeed58846e
7423 F20101111_AAARRX shishodia_v_Page_085thm.jpg
5328f88c1c9ab181a998c142c774333d
800168b868c25323b06b34346b6f5be176cd0195
83073 F20101111_AAAQPJ shishodia_v_Page_084.jpg
87d9bf5c2a19bb45d0b2844282ff8fa1
356f95465fd8295fd12d83831c57c90228bde07c
54914 F20101111_AAAQOV shishodia_v_Page_068.jpg
e125a27b09ae32d03c4bd5fd6b5d1f7a
007e54a0d4fd0b0fc59f593441a678462c455b4b
6907 F20101111_AAARTB shishodia_v_Page_102.QC.jpg
39af7e5d1c54f7963ec78346843605e4
bc05c8ff7f2ccf5df2eae358ab9a43eb52561991
7102 F20101111_AAARSM shishodia_v_Page_093thm.jpg
3b9736f40dd08fe172608f5d420f244b
aaa87d8118d3f15e721619976a45f92d96fc3815
28179 F20101111_AAARRY shishodia_v_Page_086.QC.jpg
3376c0b1ef8502067984481656dec3c2
f0c84585f33ce9e6d84c05d6eb55479e4046bfed
81520 F20101111_AAAQPK shishodia_v_Page_085.jpg
d723527987d500b2d25a1a4d14d5967c
9c6a6e45f57dba0830526623897b6528a4b7b3b5
57705 F20101111_AAAQOW shishodia_v_Page_069.jpg
17ca9e98553b70be4476b348a43f5fdd
583eceddaaccbda67c14881cba89b88c1a92559d
2546 F20101111_AAARTC shishodia_v_Page_102thm.jpg
efef0e7f315e477c722b596a2bbfed57
803556ba33b0c22f01c789b64ecc0744724ba613
23803 F20101111_AAARSN shishodia_v_Page_094.QC.jpg
0fd70045a7c312a53ce5643092522495
14b2d71ae596850e01026c34e37a5a3d5c9dc4a3
7811 F20101111_AAARRZ shishodia_v_Page_086thm.jpg
d1968153509c835ecf7764cb581c8602
f6331f21f54fafc9bd2f326ff1a59b04dea36715
21044 F20101111_AAAQQA shishodia_v_Page_102.jpg
c23c1a2a87690ebcb3e411006362bd91
54dfebbdb6392c733c7719556c1ee5dabcd2588d
90556 F20101111_AAAQPL shishodia_v_Page_086.jpg
0296be5e157c1df14c870c2df79ed926
3d93e0bbf81a6e71fff24c12555dd018f4f78cea
67849 F20101111_AAAQOX shishodia_v_Page_070.jpg
add25c9db9d4c77da2f90febac6c45d2
54cd3a1352b0fc9572f4333551b5c0a1a364e855
2283 F20101111_AAARTD shishodia_v_Page_103thm.jpg
c9ea4b31b4b76051fcad485fb70446f2
01f7b5f1f5f786739be9aba9b779fdc491f16552
22405 F20101111_AAARSO shishodia_v_Page_095.QC.jpg
ed6c8c242f4200b598e25f30725cf9f0
726cd4878aa384271ee073ee5d61a26e6f20e84a
18827 F20101111_AAAQQB shishodia_v_Page_103.jpg
c54739a8172575c03ff36dd76c40a93a
0c2c6e103492e5ac7b73cb6e0f057f6ce21ea9f3
78766 F20101111_AAAQPM shishodia_v_Page_087.jpg
ec28c465752ab2b431be8bfe10942404
3d5f42353f143680afe09b0cb2bf43f0923edbf0
63741 F20101111_AAAQOY shishodia_v_Page_071.jpg
5e79cfccb86c8ebd44ee33914bb43ff7
2f28824518eec2096f4023f452e281d3da22a357
6062 F20101111_AAARTE shishodia_v_Page_104.QC.jpg
3b3e342821150192e1a2de47eaedc2d4
ad22a8c3d36d3193082ec18f583814e90170d695
6337 F20101111_AAARSP shishodia_v_Page_095thm.jpg
01e65b038f156e9964da6dde5a852d90
3028694d0a489dabef5e0978fe16e53f9ea06344
17568 F20101111_AAAQQC shishodia_v_Page_104.jpg
b05c1521cc727ddb88bcc1d4289ac02d
502ff707f644fe801ec4e8ccd290890eb689cace
75854 F20101111_AAAQPN shishodia_v_Page_088.jpg
23bf23c3223ca99f2da5da41d01122e1
b6dedb4c078d3cef40f3124b055bd967c0a279a3
71606 F20101111_AAAQOZ shishodia_v_Page_072.jpg
7d7b99a9d6295969e46ce8c2a88d738d
6fad94de6bfa66e39ae0961aeb126b1406479607
2087 F20101111_AAARTF shishodia_v_Page_104thm.jpg
2fc96e082a3fcca4cf2218ea0ecf1a7e
cdecec13e38db7463e06bce2e1e47954d942c89c
7200 F20101111_AAARSQ shishodia_v_Page_096.QC.jpg
c5844e051685ec96f939d176fcd2fdbf
e70992c7077fdbdeb99aded6da885eda87bab4be
21524 F20101111_AAAQQD shishodia_v_Page_105.jpg
439fa6cf930c57ba34923b508e20843e
3ee938e1ae7b7238eef1620be7df3cd800817cdf
86288 F20101111_AAAQPO shishodia_v_Page_089.jpg
142a8749e9e08cf40387aa4d8506b45c
653cd305756aaa4600fd18f9ad529351af953a8e
2679 F20101111_AAARSR shishodia_v_Page_096thm.jpg
463dee96cbc7fee9587b64f1a2b26b62
4efbcba334b5c77e757bba672c5b12be41c85808
84896 F20101111_AAAQPP shishodia_v_Page_090.jpg
ae46a1e11020b6420dae345122508db2
b4c66d78761c7f5a12d7d424588be94dd0b00769
6930 F20101111_AAARTG shishodia_v_Page_105.QC.jpg
9409e2d8b0ae007c50bbfe202557b767
cb8e44edadb11ebb084bbe48cf746fb7336b3aa9
6416 F20101111_AAARSS shishodia_v_Page_097.QC.jpg
fbaff413eee364e3734d7052986a46ef
bca45f989672f6e57208d157f262869d67ef69cf
21587 F20101111_AAAQQE shishodia_v_Page_106.jpg
ffea03237f58e7835796b74933f37c99
7155086fe9e8a7e359d76966ee71f900f1dd9e09
88986 F20101111_AAAQPQ shishodia_v_Page_091.jpg
4160cb948ba77e59fc122e4b5a323e17
ded60f3a3f2b38ad744d24725dde7dbf442b1bcd
2410 F20101111_AAARTH shishodia_v_Page_105thm.jpg
d848d9592d187c2eac1ff7370da7bc0b
e781439b0f98be716b087a395b0bb50e23dccfa0
2185 F20101111_AAARST shishodia_v_Page_097thm.jpg
a50a1bc8a579575fa35ce4ab0ca5769d
dc0f44fa559df0ec62fc1afea798c46eb21564fd
20831 F20101111_AAAQQF shishodia_v_Page_107.jpg
b8a1ed312c67c9c77c5948d59c9df02d
3b59d5823711c136b9f6ca3ce90cde32fba7d44b
73992 F20101111_AAAQPR shishodia_v_Page_092.jpg
e126f7654d8490d5417860ca572bf99c
789738f3cc7248caac1260f6846335f015a03102
6954 F20101111_AAARTI shishodia_v_Page_106.QC.jpg
3a615aa9da37539370f892d184e25ee3
786a5f5a13248707115ae4ff5be3fb8f453f0b0e
7215 F20101111_AAARSU shishodia_v_Page_098.QC.jpg
36f8a4fb68500a493c6183048bea7cd9
54aa76003db791435251c8cdfd307cec041aa810
17368 F20101111_AAAQQG shishodia_v_Page_108.jpg
7d0d5aa08e8987257f4420e2b934dffb
47f672dc60fef2b78fbea7432323672ed37f53cc
82286 F20101111_AAAQPS shishodia_v_Page_093.jpg
9079b3f985b2e7d24bcb6586758038c0
2a79e71b31db86f70c53e3e1b20b2414b7a78c37
2492 F20101111_AAARTJ shishodia_v_Page_106thm.jpg
216c34ca00452dfe93a1ca4b46e27560
8305cddfd75519f5b9cd46f9a2c68e60ef2ed70e
16475 F20101111_AAARSV shishodia_v_Page_099.QC.jpg
3f359db0cf1ca0a5825922d866d128d6
ffa4d711eb5be2d73b6a9f09d6eaed1723de92e9
17679 F20101111_AAAQQH shishodia_v_Page_109.jpg
0354dfe6f7d7fa45fc3c2872b3884557
24aebc4e3d9c4ac345abd577e513de46cc81b8a4
77603 F20101111_AAAQPT shishodia_v_Page_094.jpg
e72e538438b37612189cf32c64ea3e0e
1e9c55b5e3145909a75cd56b73c3aaf84a975522
6590 F20101111_AAARTK shishodia_v_Page_107.QC.jpg
a7ec471770f9b4937ff7964376e575a1
de3d5c483b3d25fcec0090c95c8e1ad5c3ba166f
4635 F20101111_AAARSW shishodia_v_Page_099thm.jpg
67fb2b44d7311226f9da0130ae858fad
412b46325ac81f956245099f95314a3759fc32f4
27762 F20101111_AAAQQI shishodia_v_Page_110.jpg
ca86b831ddba488f3f3a4203260dcaac
618e1508ff664c714c36bedf05c7df71e1572a7a
21958 F20101111_AAAQPU shishodia_v_Page_096.jpg
1cf84672136a6f7eb6f17ccae219563a
e41b67747b923eacefb82e7c1181b70c46ba5df2
7201 F20101111_AAARUA shishodia_v_Page_118.QC.jpg
c4ba5e3a7dc9b533224f59d5dd2a56fb
afaa7f978f70ad9adbc29a261ad1675252a47b09
2379 F20101111_AAARTL shishodia_v_Page_107thm.jpg
eb8394feb2b2f55358d44df84463d60b
a2d956a46308302ada0ec2f7e2e3b6f1e27ed739
6313 F20101111_AAARSX shishodia_v_Page_100.QC.jpg
0a96ab90ce9067b37fe208b0c1b03e91
dc9c4e61afb979fbdb7ec9bcfb5a45061ae57030
28532 F20101111_AAAQQJ shishodia_v_Page_111.jpg
68ea1f4bdcc4d081896377fa34c531d5
a46f998a05a5c3fdd1bea87a20f1dd620efda335
18399 F20101111_AAAQPV shishodia_v_Page_097.jpg
25c96bd40eb0dbde2670deda4f7cde12
6b0a54fe39818aa9c16673fe7eb8f703dae0bcef
2433 F20101111_AAARUB shishodia_v_Page_118thm.jpg
998f10db82005da36d2da9b0cb6e2117
f0cd0ba4249a928d037b1a2b4cdbb4f426e0f4d1
5895 F20101111_AAARTM shishodia_v_Page_108.QC.jpg
653343ade2fb9019a7296a3e0bdff8bb
c871155741b21162fdfa9768677ee4017336cf9f
2231 F20101111_AAARSY shishodia_v_Page_100thm.jpg
54514ff9faa2bf0f03a393a10c610fec
8752e74a6d92fc7756100f474cb477ed437e1b69
26789 F20101111_AAAQQK shishodia_v_Page_113.jpg
7070729a2c5681c4179e4511c5219898
bc2b533a7c6f307a040874e7e9ac0f64c270fc0c
20966 F20101111_AAAQPW shishodia_v_Page_098.jpg
a7af790b5adcb26799d1bb5e0d93d2e2
d2b87670a7591b3faf9d6e89e60691b5f2ca3f87
F20101111_AAARUC shishodia_v_Page_119.QC.jpg
775bebeafabe63bf2da761312475dcf7
33e3f3957ee20cd506926b9339cf45e86fc89d8e
F20101111_AAARTN shishodia_v_Page_108thm.jpg
70bcab1e34d6af4f90fc4b4a1afedde4
549d21d770666b19e484499e288504503aa4eab3
8809 F20101111_AAARSZ shishodia_v_Page_101.QC.jpg
af49772da76bc8c687e42f2372ea210f
91e6c89df9a16862d73b22e216f95c8af5d69c7a
21633 F20101111_AAAQQL shishodia_v_Page_114.jpg
9bc7b7ef5965c6484201cf93cfd1d869
a291e19a50fa24cc25bf8f98acf0afd76ebf8f70
56206 F20101111_AAAQPX shishodia_v_Page_099.jpg
69655fd64f554b2220a95217d80d7e22
56f33aa12c9e88fea397a9d40898972b3cd99bbe
23431 F20101111_AAAQRA shishodia_v_Page_130.jpg
119a338b53a0d4b929876ce36b3deabf
d78e344be2d521ceede96e600dc6361d634865b3
2448 F20101111_AAARUD shishodia_v_Page_119thm.jpg
9a113ce8f8da71fee626326cd86abb40
42b1484a45969619af5cb9a8b811180599312c58
5932 F20101111_AAARTO shishodia_v_Page_109.QC.jpg
e07c4070c86388ac03c7af22bb65f741
fa57c7d31115755592dc6adaf47eb17204f48669
21901 F20101111_AAAQQM shishodia_v_Page_115.jpg
1cd538a4232dc058a074d0f0d5bd4e2a
fb73bc4ed032d3f9fc639a1a4d3d9ce61ad47953
18439 F20101111_AAAQPY shishodia_v_Page_100.jpg
17bffb9175741e41fb97a9a688f1fef1
50bce4dd56b2d5075ff933242c37e71784c05ee4
23946 F20101111_AAAQRB shishodia_v_Page_131.jpg
3824c2e7cf339d1b4bbc8e3df3e12906
d49e5bba57d3112bab77e33248b67bc080880414
8145 F20101111_AAARUE shishodia_v_Page_120.QC.jpg
c504f96ad7c3511f7e9096c9a35517ee
dd10c5384613778eacdabb6f6489aac73cf52949
2280 F20101111_AAARTP shishodia_v_Page_109thm.jpg
ef1cbf2ccc534431cce056e51b227bf0
8ecea3351da3bf9d45cf491b339347eece1ca257
21991 F20101111_AAAQQN shishodia_v_Page_117.jpg
5d4b36ade7a3589a9c81086fe3342408
678ebf678054d62c56f5a778e14589769fae7e74
25694 F20101111_AAAQPZ shishodia_v_Page_101.jpg
5d67d5ca8d850b16dae535fc9f48dc57
5fa5a6a3a1296187346055026f538e52bfb6d22a
79254 F20101111_AAAQRC shishodia_v_Page_132.jpg
5d28a48ca0bd3d38855680529b55b9ac
4c2205b857a4228681b78208f3b7e5a42fd018f2
2599 F20101111_AAARUF shishodia_v_Page_120thm.jpg
2a84a4855076e85e39135dd2ab2ffbf7
6a143e4186ed6f18c7a5becb4cf1505fde670a06
3051 F20101111_AAARTQ shishodia_v_Page_110thm.jpg
99d3f5c239357a09e08d3d8d0d4507f6
3e3e66f243ead4c0ea0dd7a5bbbeefff718c3bcc
22256 F20101111_AAAQQO shishodia_v_Page_118.jpg
8df46cd13dd7fc6a29ddb85ff90af243
6dd7e002a2bf1d06eb1687d29f8dd8f07c9816af
63131 F20101111_AAAQRD shishodia_v_Page_133.jpg
c707a54d3c167a91a93fe54aecc7e001
4e0b9e3ff51860dea47d0004235471d7f06d6fbb
8618 F20101111_AAARUG shishodia_v_Page_121.QC.jpg
31a5b4f9cfb720aecbf37143cb9b0d8b
b02eec7011fe6afc4fe1a2289706fbe787d59bb1
9906 F20101111_AAARTR shishodia_v_Page_111.QC.jpg
bbabbda06b68a33f62ec096441a6a85a
0d010b088f839a5735085e7c16afc44661520b6e
22340 F20101111_AAAQQP shishodia_v_Page_119.jpg
88e8ca9cf7b4ebb779bf9a9cc52c129e
72abcf4ee1fd77491ffdebded6a0c4f66070ea25
44790 F20101111_AAAQRE shishodia_v_Page_134.jpg
0d0d9b661dcdcfbcf90f3266f8ab87d6
3ff73eee67af15b1ebbcf02943c5c66292909e5c
3323 F20101111_AAARTS shishodia_v_Page_111thm.jpg
94701110bc873f06900370df383a96fb
1b233befef07267677f8fff1397dec318bf68cf8
23618 F20101111_AAAQQQ shishodia_v_Page_120.jpg
d156286a382baa8c8642031debeab495
e38d8b6574441a17c1af6c76338e47be23f34cea
2748 F20101111_AAARUH shishodia_v_Page_121thm.jpg
eae02008241ba41c8c6b71e48af6e352
6cefc9c9aa522cbe594eb2e68d4baccfca5ad2ec
9385 F20101111_AAARTT shishodia_v_Page_112.QC.jpg
4c66f80b64f36ab2939bfa0d556b2627
a9fffb441b4c94ecfc902fc2f7daeb1e79cb1662
26130 F20101111_AAAQQR shishodia_v_Page_121.jpg
fb5fd102918ce8f603af9ec148b0069f
1f0a272510a8b82ff1f41cd5d0b60ddb81281a2c
26040 F20101111_AAAQRF shishodia_v_Page_135.jpg
895ac962ed7778baa4bac648fffd0bf6
3971d55db47bbd1a84166401119024ae54c1c1d5
8964 F20101111_AAARUI shishodia_v_Page_122.QC.jpg
c831b1ef76e227f56fedb7bdf3dc2bdc
d8f5c0aad7c1872b44e07e67144d35dd72d16924
7031 F20101111_AAARTU shishodia_v_Page_114.QC.jpg
dce3d1defb71d1dd0e9a506077dbb1e5
7b6f02b225ada7cb89d1c0a6bb5dcd319e533d20
26115 F20101111_AAAQQS shishodia_v_Page_122.jpg
9a36af676ad8af2e6a2fe52572261a32
f947f46092bb878bab61be288c944644ad022575
34322 F20101111_AAAQRG shishodia_v_Page_136.jpg
5beb559a03581427986091576a57b851
47f72c3dc62a94ae2393339065da3067e4c4184f
2751 F20101111_AAARUJ shishodia_v_Page_122thm.jpg
54714a1f648e353bb38ec94ee3b9c34b
e03ca36e0ba9880366a4cc052661494567d5e25a
2434 F20101111_AAARTV shishodia_v_Page_114thm.jpg
c87e5a6d820cee19508c2fdebe01b661
068e7a8ce480288dbe5e51a7a760cc38c28f624c
23253 F20101111_AAAQQT shishodia_v_Page_123.jpg
2b2ea9f79fcc93fa3fbe59b1cd39f46b
aa53143f4cc69f347d60e7fa6fdb6a75f39f15b9
102833 F20101111_AAAQRH shishodia_v_Page_137.jpg
2212dd5c0c5f25411aa0c6646d76a6d1
41caba8c048ec6f60a992adc08f63bf654c77c1a
7572 F20101111_AAARUK shishodia_v_Page_123.QC.jpg
4f63c2fb5353bfeef515f1229f378920
80b4e587fae492493beb37e0848dd7dfe09955ad
7025 F20101111_AAARTW shishodia_v_Page_115.QC.jpg
e86ff3280242e8d7055aaa467b46e9e6
58f1eb04b6bc3b438cbe0c767b24561de596ebe7
24847 F20101111_AAAQQU shishodia_v_Page_124.jpg
c86b8f9779914c05053a32077e4f9698
f0a484a111b70b749bdf47c92b2d26d213be51d4
100216 F20101111_AAAQRI shishodia_v_Page_138.jpg
c81392ad43c12ce3de530ec5e67bf4e6
35c0c78dce867a88899a69fffcf3f7ec86bee60a
7913 F20101111_AAARVA shishodia_v_Page_131.QC.jpg
295751124d79eb31cefd3bbe96938c19
4da1011153ba422827a11096d93c35617457df24
2521 F20101111_AAARUL shishodia_v_Page_123thm.jpg
9d5a752951272cbca117602729e7241f
a9ee1afba1f2530d171bba5388d76cd5aaa8440d
2430 F20101111_AAARTX shishodia_v_Page_116thm.jpg
97ca8936684d72525eeec426d1caeec6
067c8b48d0cb731d0333666861c00abf56322889
24755 F20101111_AAAQQV shishodia_v_Page_125.jpg
d587af09912e5c0a40c2218dd5a5086d
7533ccd89f407ef0a6c3ab0b99d0df51302c920a
66051 F20101111_AAAQRJ shishodia_v_Page_139.jpg
6608d8d550774869be4d1095baf32f77
fc95ae8c8c4d6fc57f73c89eef7a450035455d0e
25178 F20101111_AAARVB shishodia_v_Page_132.QC.jpg
7916f1165796906e032cf8b242bbbe13
1cff846ca1112f2040ab860d6f9dc8e555bef2b9
8217 F20101111_AAARUM shishodia_v_Page_124.QC.jpg
ff2a7a5a1bccfed6f11dbf5e3c56a251
c90022b21dc61fa3ebeebd1e10fe64c17e9bc35d
7012 F20101111_AAARTY shishodia_v_Page_117.QC.jpg
a96f76654d41a1566ee4a4d6614afac9
7322f2b138806d20814f4f877fbccfb031e290b8
23218 F20101111_AAAQQW shishodia_v_Page_126.jpg
e22ce071c35e837dd139d2d023309de7
b1f5da50c4ae02de26da1008027bd04e375ad073
42606 F20101111_AAAQRK shishodia_v_Page_140.jpg
23aba2b14816631bf63031be18cc20da
5c3b8c988c0af70a84a28d2452057f3a3905eb19
6963 F20101111_AAARVC shishodia_v_Page_132thm.jpg
8ded2c2ef98e0d812d323fc212d7b8b8
8e6f87ac6657141c50d91aecccae3fcb067322f4
2777 F20101111_AAARUN shishodia_v_Page_124thm.jpg
69cbaf883c8d391dc20c2ad3918173d2
bd1dc448dbbd32f6b4b70990edc61b848af62f44
2426 F20101111_AAARTZ shishodia_v_Page_117thm.jpg
bbbd0dd15ccba089f00de9dd4074bdd8
fc05bdfc7fdd153a611015193e5c86c915bed56c
24192 F20101111_AAAQQX shishodia_v_Page_127.jpg
a5675588ece44907e9267134727aded7
5d94632ac78c8859e4c2cdc4b43b7bed9cd2da44
1051957 F20101111_AAAQSA shishodia_v_Page_016.jp2
af7a8a9276476c5206df6671b24c349c
31ecaa32e93148ff2380b6fab87a1ab2659df578
237734 F20101111_AAAQRL shishodia_v_Page_001.jp2
ad8d42dca8bf92a1de64e41d936ed73e
0b5d0cba621da98fdc2dc158e8970b4ca09eec18
19408 F20101111_AAARVD shishodia_v_Page_133.QC.jpg
fddba37348a94b73e11ad98c9064948e
678dbde0afeb7c525673ab62436ce30e30402fd6
8188 F20101111_AAARUO shishodia_v_Page_125.QC.jpg
604931fc895c7f7d308ed0d5e096e510
8ec7f0a27b8ca75227b048927309a8056c9c4028
24449 F20101111_AAAQQY shishodia_v_Page_128.jpg
5d13198372221ff85b93f45d35b86744
7c16745cf8cd8c8686e03c0d8c35f19d4c112ef4
1051973 F20101111_AAAQSB shishodia_v_Page_017.jp2
da833ff785bbff11b9dc99f6ec43fe9c
2db741435b5c9e9f1308c74afcdbd71060ca08f5
27050 F20101111_AAAQRM shishodia_v_Page_002.jp2
53879e77047169cae29e3f6facda986b
8f6b939b293805d16597a492bd98d8a44a3d617c
13931 F20101111_AAARVE shishodia_v_Page_134.QC.jpg
8f8039a0001aee1173a59aa6bf6e3c5f
0e79a94b39f1ceb44e2b6dfbfa4a9acc3eea318d
2773 F20101111_AAARUP shishodia_v_Page_125thm.jpg
0165b9e42486371d35be0221e331a89d
5f956adecd3fc926f7dc36b8245b50fb3ac66643
22892 F20101111_AAAQQZ shishodia_v_Page_129.jpg
0b8c4b4c761164091ecae4c3164ea9a1
a584b57a320218c65f922b6ba3dd641ffd5e3cfb
653282 F20101111_AAAQSC shishodia_v_Page_018.jp2
98729d7e92d1021efa79244fce7fc111
2ba163b6a9ad03b2a74875684ce9afca779dd96d
16165 F20101111_AAAQRN shishodia_v_Page_003.jp2
fc8c11d9f1a537db215a1416c50ab8b4
9eaf001dd5735a07d1e906fe6e089fabb4132d78
8649 F20101111_AAARVF shishodia_v_Page_135.QC.jpg
671e89b81206c78d3b5b042d5f01ebfb
b8a4f695ad70463976107764a95ec54db24680f0
7427 F20101111_AAARUQ shishodia_v_Page_126.QC.jpg
ae8ed86e9e6ff04c8acbd8c25b1a8d17
6a11d22e3936fcb87b5d9119276aaa3d3832c2b8
363381 F20101111_AAAQSD shishodia_v_Page_019.jp2
1434c1ae420c384daf63eda9f104c1cb
6969ca07bb41ce2d71bae121352972ffe19569f7
669728 F20101111_AAAQRO shishodia_v_Page_004.jp2
9744d869fa87e46444cd18d5ac1ae767
81403b17f6d7c3d354116391ec1e8a8d6e380564
2974 F20101111_AAARVG shishodia_v_Page_135thm.jpg
589eeb4cead6c6f4f2e907e72aea52fc
be83bb2a16edeefa85687d5d92c6007d332225d0
F20101111_AAARUR shishodia_v_Page_126thm.jpg
737095d1b52e7131f764fddc6ea4446b
c377b44a76907643a750fcf43e404a9f47a64e9e
578850 F20101111_AAAQSE shishodia_v_Page_020.jp2
3f975ae68c4db93eb320a1cce29ea6c3
53607a0294ed50b2b04cc5dfcea87e4483f6939a
1051965 F20101111_AAAQRP shishodia_v_Page_005.jp2
021c1745ee1fb0f42872c671da036427
68febfaefbc4a2cc5658d4338680506e25ac88c6
11233 F20101111_AAARVH shishodia_v_Page_136.QC.jpg
50c29e8895e271638cb509584048327a
fa77efbe9dbda37bf15ef4e6da0e061c0554bbc2
F20101111_AAARUS shishodia_v_Page_127.QC.jpg
a191ef37684170f5a76c9ae69582f923
9b2284a8b981f8103f3103c014c03271bb472428
585817 F20101111_AAAQSF shishodia_v_Page_021.jp2
2d0a571d1b2d1c6551b8c03903059145
8b9a1f391d74e418e8eedabdf24e33ace22eb8c3
1051983 F20101111_AAAQRQ shishodia_v_Page_006.jp2
03ffe8bc982cf2d86cb6539208b3e01d
6d019ef3acdc34e63cf55f7d1729380e59d9338e
2603 F20101111_AAARUT shishodia_v_Page_127thm.jpg
e30ae6460be3e402e71018901cf087f7
0bb0abf871e6af2fb221df7566077b71a759eddc
393971 F20101111_AAAQRR shishodia_v_Page_007.jp2
031b12e8efd241cb21a1bec6d977fbf8
459d3ce85f404084d61544a732a5a467056b1076
3649 F20101111_AAARVI shishodia_v_Page_136thm.jpg
3d1d1da6ecaa5c6f8b2035d598c23862
ccde0ace3a4147f843356f0f48be1819ffc4b3bc
8201 F20101111_AAARUU shishodia_v_Page_128.QC.jpg
7a749b329b08a7d12157bf27a92624be
c17bce8ef6820a1e25ef1c58e5cebb8142410b64
846161 F20101111_AAAQSG shishodia_v_Page_022.jp2
c566fc2a3bb3700f634547705284b3fc
47722e31ed60a8416b8fc67c399c6da180c599d7
1051981 F20101111_AAAQRS shishodia_v_Page_008.jp2
74600a9d2b113624cba7357ab4772343
68e738f2f6012a77624f63fa2e5ced385aff2a48
28025 F20101111_AAARVJ shishodia_v_Page_137.QC.jpg
5f7a0f9a6f5bf72857ac9431c111855f
11732efd175d9e7e2ab86ab8cf2d8d34dacbb8b3
2659 F20101111_AAARUV shishodia_v_Page_128thm.jpg
a1448d9405796ec45713e1e033bda41d
cdb96a826950303c5c84b6db7ee66b03397dbdd2
504776 F20101111_AAAQSH shishodia_v_Page_023.jp2
248cef2d22b8ff719a47cff5636d1dd5
0fee52db7f3c4cf74dea1b7a4f2dfb699945e45c
1051974 F20101111_AAAQRT shishodia_v_Page_009.jp2
50230e282212ca6f0916d4de3de5336a
addaf48b055d7795e0e1a8308659af5a1a785253
7536 F20101111_AAARVK shishodia_v_Page_137thm.jpg
1e2a9b8ee98c3b9f9bcd37625fd68e85
91f591252a6e643e85eec5902fe3ac3f41239bc7
8060 F20101111_AAARUW shishodia_v_Page_129.QC.jpg
3730115dc71553e6b6f51afdb4c69322
472e1891d99698287ea4740d492e69e2b32f3a71
724034 F20101111_AAAQSI shishodia_v_Page_024.jp2
21410eb4f17c2c4980543bee977c5d91
4ab3fdc9c61d963170dcca8dd7d3974b2d3917a7
1051953 F20101111_AAAQRU shishodia_v_Page_010.jp2
37c2ca93d9e02433f9e02d38c94b94e0
741785a2d2a583d6124ecbfff93ea19912b66dfd
28734 F20101111_AAARVL shishodia_v_Page_138.QC.jpg
fa4c8d4e4ddbf16e341580d47a301c5f
0b99778bcb89e0ad91233027b9bf2905afde247d
2513 F20101111_AAARUX shishodia_v_Page_129thm.jpg
bbc49e08144c0d8b1400f82969525ef3
3b73f700b9e4687d6e16ecc82cead07bb6ad3b6d
920913 F20101111_AAAQSJ shishodia_v_Page_025.jp2
7c501b35cc2e061a72da49a1a33059f5
fb337bab0fc0731e8bc8d481ed1018e1675f603a
1004621 F20101111_AAAQRV shishodia_v_Page_011.jp2
13362de90343841c37570ae4b8be59d5
6c90bd26030cec4a132d33fdc7336e95350e7e84
7877 F20101111_AAARVM shishodia_v_Page_138thm.jpg
30d062431dace99b1dd88d6a8aebb238
7f7a40dc2097cbf9d420fd5c4bbacfb9c03f0bbd
8215 F20101111_AAARUY shishodia_v_Page_130.QC.jpg
48db67b2cfa6f59584138c7e3baed438
ddbbc484b468523f958c45574a7f5514f17d9ccd
1051975 F20101111_AAAQSK shishodia_v_Page_026.jp2
6ce391c81ceb1201c32c2e799c962994
83ab28e1fb449ed54f778f1897f8daae778fac3c
494967 F20101111_AAAQRW shishodia_v_Page_012.jp2
d3bcb41f26cb844294f0791bb2ad416c
a030a219674b7b63c348820b907f72a2598a0016
19129 F20101111_AAARVN shishodia_v_Page_139.QC.jpg
8b2f371cdaa795ff593ba7bd90491332
57d2aa48e8078184673bfc1ed53934f9834ac72e
2548 F20101111_AAARUZ shishodia_v_Page_130thm.jpg
30686295d82c0d83da948351408d4416
f2c03207baa2aa1ba6f2a19aa2443c3b74932965
F20101111_AAAQTA shishodia_v_Page_043.jp2
0a77f14f12fba6f726de4bb2d39fbca5
2ac94048e9faeedb21cd05487177738f316f589a
1051922 F20101111_AAAQSL shishodia_v_Page_027.jp2
f0829bc43a4bceceacc9ea44fb8256c2
6fa460eeef3428a601debba256cc40552704544a
1051948 F20101111_AAAQRX shishodia_v_Page_013.jp2
b50930a880c3e072e3e80eac39cc5ea8
ba74453e710f82d4901399518b6f6b2ae29413b2
13425 F20101111_AAARVO shishodia_v_Page_140.QC.jpg
15bbe87f5bd120b630bab064856888a9
8088792df67cca1541788b7e7bab35dd092b45be
F20101111_AAAQTB shishodia_v_Page_044.jp2
c924ad509260d2c3b2a99928b2438903
b5ff9fa8e17c15b9ce2a38de0766eb356a02b1fc
1051949 F20101111_AAAQSM shishodia_v_Page_028.jp2
d86ea316f09c7ec8c8889b1ffdd84fc3
0c3f04900eb543e4d499290ff399d72b510b6913
744717 F20101111_AAAQRY shishodia_v_Page_014.jp2
8e914d1c9c9e88b6db03a3a1d4f5fc93
af94f87707f088139479aba1ae8f3fd60a50d651
3842 F20101111_AAARVP shishodia_v_Page_140thm.jpg
7e8dc3266d07ae18c13eadc9d01eb6d7
3fe755d1f6c6baa6a93028f250b2f44300bf6a64
1051979 F20101111_AAAQTC shishodia_v_Page_045.jp2
6c74deee880ca3d05ae13b2dada744c6
39fe5c008199f210b125b4d48009ec772200de93
1051970 F20101111_AAAQSN shishodia_v_Page_029.jp2
e1416c5dfddfbb4460dd6b23ef9866ea
fd80f09f654ecbacea381726f2dafd0abb962279
1051971 F20101111_AAAQRZ shishodia_v_Page_015.jp2
bbb6f96b6dbef1ab42d3893d929639af
954d0903da82c57a236be0de25c9b94f48541a37
1051961 F20101111_AAAQTD shishodia_v_Page_046.jp2
c104fb4b4c3007cc532dde180a2a38eb
6c1272af6163c512571a5e2665b195786da643f2
13656 F20101111_AAAQSO shishodia_v_Page_031.jp2
b8a3a8dc23066c5149ae1f180e9bcebe
a889fe29f4e556bc2463424f71d34ca29f77bf58
F20101111_AAAQTE shishodia_v_Page_048.jp2
0297f7e7ef80c35e56317e40b0edc1a9
bd36ba9abb1c119fa4fedad86b8de60201cfdc57
792368 F20101111_AAAQSP shishodia_v_Page_032.jp2
e60007f6bf1b7f6ade0c078658621b87
95cfc76ff32081c69eea0673c3aa982f00306f65
196053 F20101111_AAAQTF shishodia_v_Page_049.jp2
93dc4a57d9f0d21c5b5432ba09ff3f29
9f77824a94749d1631130ff782564bdd53ec9f3f
819080 F20101111_AAAQSQ shishodia_v_Page_033.jp2
818bfb6cb10efb4b9286a755697a1b39
4d625f84aab684e122cc2bced944d74f36e6b459
1051930 F20101111_AAAQTG shishodia_v_Page_050.jp2
54c03401a449cd0349130478d619512f
cc4cba505c5dbe38f780a0e1db678ad1f76e22cb
815955 F20101111_AAAQSR shishodia_v_Page_034.jp2
8a5c1501e503ab3c4b0bb6de76878c32
f3569ed5dc6ded3b90aca765f66b71a7380208fb
473621 F20101111_AAAQSS shishodia_v_Page_035.jp2
06d55db51180aa976872a14ec36acb41
959f58c27c8178743ab026418642aed50855ee2b
1051868 F20101111_AAAQTH shishodia_v_Page_051.jp2
95c77b773ecb9fbc1594d4f0d99880a5
4d637386cbd7bcf09a290f745899cbd6f04bdfb1
1051926 F20101111_AAAQST shishodia_v_Page_036.jp2
31b057671486e94ccbff90c3ae63e394
baea059bfc24459a87bc46b49eeebc79c9df43d3
300653 F20101111_AAAQTI shishodia_v_Page_052.jp2
6c89f5ffa33b296011c704a484fdae54
9264b737a159d9feddf1cf6bb07a6eea28658840
594141 F20101111_AAAQSU shishodia_v_Page_037.jp2
f1b7683d8f2ef581979048de25ff3274
a5f725e3f8789c70f5ce5e17af2a851df0bd9c85
F20101111_AAAQSV shishodia_v_Page_038.jp2
4774af66a966ce801ebce3b4c02a029e
56ade47708272b146e674162225ecdee4f663e3a
408845 F20101111_AAAQTJ shishodia_v_Page_053.jp2
c75efc285e8ba68dcfc24a0e67f97049
19cc21a9e51f8730029866edb93b73cdc5ba9a97
1051984 F20101111_AAAQSW shishodia_v_Page_039.jp2
5616fbee9ffc08954bb613a7bec3ebbd
dc4ad0101e77a07ea5141dfa576707e3c8dd1531
37809 F20101111_AAAQTK shishodia_v_Page_054.jp2
52ce5dff9fc1af75164a6b86bb65a8b2
68f5d7d571cd37012d9343fbb25aeaa83e9e7809
688936 F20101111_AAAQSX shishodia_v_Page_040.jp2
a9313a810f5e5efd93597ba33e6d0945
49d67f50b2bee4a0a2ed0ea544fe615ca5447a73
759656 F20101111_AAAQUA shishodia_v_Page_073.jp2
33e68ece915413dd15111b112654060d
648ee31965b236d2af0b4fb1b2cbd0e9772ea160
37711 F20101111_AAAQTL shishodia_v_Page_057.jp2
8fbe6d3f6107c77ae88ef1106a82b6ba
61900bbb32c4f4e35ab08373e5b9f7a369be11c2
618461 F20101111_AAAQSY shishodia_v_Page_041.jp2
77264a3a709e370e6f960ac0c5201d8d
fc4c81c9db0c053c731b180c45b078ab0d438696
735332 F20101111_AAAQUB shishodia_v_Page_074.jp2
3854dd56532fbd54da76876f9031bac2
d5b7f5ff2499a9232f44be35c99d376bd389df54
34002 F20101111_AAAQTM shishodia_v_Page_058.jp2
a765c15a46620e7fda7d64c1001d6265
578ee53df68f22623ea0f13aa1851403e1f24ac6
1051907 F20101111_AAAQSZ shishodia_v_Page_042.jp2
f1b89fcaeb4f99b1da8f97722cc9e34e
4173b0ecbaf54ae41f588982cad45bf7f6365519
1027008 F20101111_AAAQUC shishodia_v_Page_075.jp2
15f90d00e54dd5760a8fd9dbf032f0bc
e71e0cd279e425cce906da14119c92bb5b17cf8b
412570 F20101111_AAAQTN shishodia_v_Page_059.jp2
9d2ceb01ebd4535fc8af865df525d52c
218b3651f2ec18a545e6e516bdb27d9e770a2569
699681 F20101111_AAAQUD shishodia_v_Page_076.jp2
a854fa75992b718ee155bbdb0b5ba5a7
6cb120c7a4efe613d3458e52c3b6c093fdf1210f
F20101111_AAAQTO shishodia_v_Page_060.jp2
10e58e372c3e84c02a716d33ba4fd34f
3025dcf547a416f47c4059f63701695d4dade079
476449 F20101111_AAAQUE shishodia_v_Page_077.jp2
1654c7d5951d48f4e1870c66d4892473
feac3b6a3386b13e11ded67ed5b5ed138f4eca55
F20101111_AAAQTP shishodia_v_Page_061.jp2
3a36589dc7a04664edfadabc30ee394b
16dd7dc9173eee7a597ae97947760737bfe543be
576340 F20101111_AAAQUF shishodia_v_Page_078.jp2
6840f07ad2dc4b8388a441df20732564
cfb03dfe78bb11cddb9d086f826832ffea4146c1
1048579 F20101111_AAAQTQ shishodia_v_Page_062.jp2
bd67059d901c7bc3ed5fcf5b22e28e26
be26defe9ccf364d16e33be3bf71d56a2df9d752
F20101111_AAARAA shishodia_v_Page_097.tif
5837ee23b4c594f30c3a7f0aa042d6c8
279d95bcd0f3c8a78061b79c2787540a34887913
615131 F20101111_AAAQUG shishodia_v_Page_079.jp2
4930ebe0d2f47420da3eb4cfbf0f90a5
615c34d31ab246ece4c80b4db86c6730df25bddb
1051986 F20101111_AAAQTR shishodia_v_Page_063.jp2
d745f6c47e29c1cd71c04eaeacaded1d
a373a1473db526ba5172181a9113f301cd2c79f8
F20101111_AAARAB shishodia_v_Page_098.tif
45c27271f99cdd3b37097b81eeddf686
fcce994eead2c4dfccd330b3791d59b9b0e551a0
898310 F20101111_AAAQUH shishodia_v_Page_080.jp2
c79cd4696631ec29cf7ae6b165163602
0d734eaa096eea7e33f25d298ff98252c45e4f21
1009506 F20101111_AAAQTS shishodia_v_Page_064.jp2
961cf7dba0a54dbf40508280336bfa5c
f522e58f41361b42ac573a098b0fe49a10a29e29
F20101111_AAARAC shishodia_v_Page_099.tif
03654a8d09f15df0171d521b67feef2a
dd8919cc4d205d55d186d1842e9da34cca756973
1037623 F20101111_AAAQTT shishodia_v_Page_066.jp2
68100dc39acf5abdd460508ddb16a1b1
a280e0bcc6ffba4b2c6937931670d8037cd1d470
F20101111_AAARAD shishodia_v_Page_100.tif
2c4df79cf85d8ca200f8efb8a799d916
97933627937ef6a54820368b183c0dcafdd6526d
985937 F20101111_AAAQUI shishodia_v_Page_081.jp2
37abea2e3af119ef96a580cc0b368737
40351d3329ea94eb945f539643e7569e0dac4d09
983770 F20101111_AAAQTU shishodia_v_Page_067.jp2
6209474ac357310fd09170a66939cc9e
7f5e80bb4bc549830f2333ba9cf0b35c9a1a7fe4
F20101111_AAARAE shishodia_v_Page_101.tif
945805c5e710bbb4593bb52763c81c0e
ceb4aeb1f678f14710e66089a036249f320331e9
F20101111_AAAQUJ shishodia_v_Page_082.jp2
82ee39ceb505d600f71f135de86336a9
c6e11e4f21d58f5bc224d3c17663692e5093ca76
729000 F20101111_AAAQTV shishodia_v_Page_068.jp2
9f215d72d95c38458ebc051543455b44
433bb2d6b450dc32e03a1e2da7bf90267fb3d30a
F20101111_AAARAF shishodia_v_Page_103.tif
f1a7c6c9283d91d52118a6959589f614
34fe87944aa8071e4bd6a4332005a0f3b47b916b
F20101111_AAAQUK shishodia_v_Page_083.jp2
2f0ebabf16cd8bad87ad3b4dfa270691
3427fc9aaedc4752f8fecf4bed036253c74be411
771577 F20101111_AAAQTW shishodia_v_Page_069.jp2
f0a82436ff7e300800f8250de6f7cd7a
7556e47151afca0b5a115c6e344f1f185c772fa1
F20101111_AAARAG shishodia_v_Page_104.tif
163fc3fb43e587081a7e5d70400a3532
495b0ad2682ed31b649fdee5ed9647becb98b656
15810 F20101111_AAAQVA shishodia_v_Page_100.jp2
4d1c44a64631d7ad63d80332a0e2592b
75d072e40f1e0834054d37e374741ca23bb0174d
1051964 F20101111_AAAQUL shishodia_v_Page_084.jp2
88fd92056333109151ee8923692fc661
5de42b81bf7229377f217ab095ac30f58858d85f
917610 F20101111_AAAQTX shishodia_v_Page_070.jp2
612f46f76061357b2dc252f652717d67
9f79420db263e41ecae5eb290cdbee5a998e08a0
22651 F20101111_AAAQVB shishodia_v_Page_101.jp2
ab59b3b4d7254021dc20f28fe8009680
84893aad2184227b8804181b9d815eb911f9aa1d
F20101111_AAAQUM shishodia_v_Page_085.jp2
b0e8a6e5f007e0e2d3a530532ab69d3f
189bb48241aecb4f9871a3ee7d0746956b81da91
867465 F20101111_AAAQTY shishodia_v_Page_071.jp2
3f84c713b537c417c5038732514a7841
b5eb61170738362a6fd3a393d6d14aa6569830cd
F20101111_AAARAH shishodia_v_Page_106.tif
6f53050f38bdfef64e82ec49d342dd44
0cc510eb94a8219696c8a4aa1f592c667b4ad4da
17257 F20101111_AAAQVC shishodia_v_Page_102.jp2
916fbf772b0361de64d6c57ddeb1fb2a
407d2e11b1b517930876d03ac123c463cdaf655a
1051927 F20101111_AAAQUN shishodia_v_Page_086.jp2
8368c94f3e21f2d6742ec9067570e497
d025ab83f602ad21f06fa67e8a90090df3a1acd7
1006671 F20101111_AAAQTZ shishodia_v_Page_072.jp2
e0804f78290c7d13a9275338937fa732
72981c0214b1c38c3d31b668863f14dbe0fd5995
F20101111_AAARAI shishodia_v_Page_107.tif
9c60f3670d3b701ddf1a3cbe3521b104
7c73700a09554b204321f98b22d068d4515715bb
16931 F20101111_AAAQVD shishodia_v_Page_103.jp2
a8108419611a08d59906274e77e1cfb6
65339a19af29ef0d1deaff499c6c013f922ca7ff
F20101111_AAAQUO shishodia_v_Page_087.jp2
71380e1aa250ba82323dfde65239cf4f
cf3cf39866f431f8df54d30e64432170bce7bf5e
F20101111_AAARAJ shishodia_v_Page_108.tif
7b224cae659e7f8335d910301b1ccda4
4822cf75e086ed9dd7fe1256759b52b4510b5a44
14457 F20101111_AAAQVE shishodia_v_Page_104.jp2
97a212dc38ff4b2eb58935e2e012d296
c735adf0e0f7f61b0d77d28886f8ec5a801a9ed7
1050201 F20101111_AAAQUP shishodia_v_Page_088.jp2
800560ca44471d29339db8b14469b2bb
84b17da0380b26c4916f390cb85b9a3d303b34cb
F20101111_AAARAK shishodia_v_Page_109.tif
bec03a7c176e11a333a6c7b91272107e
260e2af1dd272218f6ca061390458579d6673455
21388 F20101111_AAAQVF shishodia_v_Page_105.jp2
588ca31ef3b5c63d9858869f4c44ecec
d22cd055b313a576a078022f0feb45e9a47e2158
1051968 F20101111_AAAQUQ shishodia_v_Page_089.jp2
7e580abcfb7ef4d4949fc9f35b476bdf
f2e81c19d1b3e67c0187a6015fd3857081d28541
F20101111_AAARAL shishodia_v_Page_110.tif
d4fdecdaa3e0fe84bd6c2966f4b41d7e
2cf9cfddb3cac36419ca3b0d75a9cc2e1f9ea180
21904 F20101111_AAAQVG shishodia_v_Page_106.jp2
279781d463ce513a91099063d444013e
1940be489eea7fb1581e91eb26306c9574bc2a2a
F20101111_AAAQUR shishodia_v_Page_090.jp2
9a095793aca758456f94ab7bf7cf1a9c
534a8437f83af8f97b17e3a8a85dd50db4d8c753
F20101111_AAARBA shishodia_v_Page_127.tif
f65aacbb2eb2511bfea066092755202a
f98fd760a57cf8746c456ced8164e9288b458832
F20101111_AAARAM shishodia_v_Page_111.tif
84bc7b2452b6d12641f73f2cf16d37ab
5024ba44049c5c5d66b74198d2e81658439be879
224014 F20101111_AAAQVH shishodia_v_Page_107.jp2
95753109c088a711f2759404b5e12b99
1792873708ff24a970df6239a5a0dd9c49650009
F20101111_AAAQUS shishodia_v_Page_091.jp2
a1521d9ccb0b0d180f5c6ee82285507f
fb2359835ec1dc2b87c4180984b76cfd5f8c94fc
F20101111_AAARBB shishodia_v_Page_128.tif
75d45feeacf0386e5459999ef49bb578
e0f57f4fa30f8f6d203f0f2561b25e72ded155f8
F20101111_AAARAN shishodia_v_Page_112.tif
be886b4dc963201f880df15b1b3fc244
99fb1e8b38ad0d01a2e32a77e3cbe6800ad2de1b
14963 F20101111_AAAQVI shishodia_v_Page_108.jp2
a3dfb48477016136c3699c80acc6d98a
abca1ad1097e63e29dcf9b79c7c9b98027b5cd31
1020509 F20101111_AAAQUT shishodia_v_Page_092.jp2
aea1b6badbfd60964e777ad7e927f50d
3df74ad086c0098bc49e4c053cc15948abef78ec
F20101111_AAARBC shishodia_v_Page_129.tif
6e6a7e7bc421be9356e3143d7450a20e
d7216545e866f103c6051e1b50e2ba5e45c0d7d2
F20101111_AAARAO shishodia_v_Page_113.tif
375eec64f9aadaae733d55868f3410ea
5050d345a4398b8d2cccce079d71149b91591e75
F20101111_AAAQUU shishodia_v_Page_093.jp2
6831945da1ea07540e621cfb28779b4f
ab48893e2f330cca394fe7ec75b0e181aa89eeac
F20101111_AAARBD shishodia_v_Page_130.tif
7e05eba1615344b8cc9f9bae1798fed0
b4092acc4261be13a9b75b0643398d26b253e02b
F20101111_AAARAP shishodia_v_Page_114.tif
7713345439c71d7b7a34d6b16297bc87
f6c1bb1cc7abd3808bb288581a9b637503eda33d
16855 F20101111_AAAQVJ shishodia_v_Page_109.jp2
78eb3fa313eeab10b423b08d53af956c
e756834318cdc01101bf41f8f37bd3207a74782b
F20101111_AAAQUV shishodia_v_Page_094.jp2
49138e89e05e0b01c1c8dc90ce4c5ba0
b25c4417602ea9a75a145c1727c9f56ab618dd2f
F20101111_AAARBE shishodia_v_Page_131.tif
5e7c2a04e397a50947cb55358b203f27
ff5ae0410f714e95ac6ac4597da857649938a10d
F20101111_AAARAQ shishodia_v_Page_116.tif
7ac79ccba7c92cee69981e6a4977c9ca
904918608014287b45d3d1bd54400fae14bce684
30954 F20101111_AAAQVK shishodia_v_Page_110.jp2
4df16fadfab88903dee9661163fba22e
7466aebd9c08832e3db6a2d4c8df6a3b7271ad92
964457 F20101111_AAAQUW shishodia_v_Page_095.jp2
a7cef431bcf4d939c17ecc45d759f183
075f1722c6ce08d4a3c71c7ec5915560c224ca04
F20101111_AAARBF shishodia_v_Page_132.tif
d3fe1f52bf95b5f16b0e52ec4721130d
81daf8f0b43d1b692cd0340081a50054863c62f2
F20101111_AAARAR shishodia_v_Page_117.tif
9a8cd7994f20b30751556ec506d94108
daa1672af3c8644c6d2947e23ca25843743095ce
32965 F20101111_AAAQVL shishodia_v_Page_111.jp2
f44092892176ad0ee23f8792d4fc350a
d4b55bfc51576a678c64f463b2e33f31a27f57e4
15899 F20101111_AAAQUX shishodia_v_Page_097.jp2
3bfb8016de37644d3dab707008d85866
20b98363594669dac9cd452c114716f52fdc0316
F20101111_AAARBG shishodia_v_Page_133.tif
4be5d7fa0996dc51f7d10b281d190bcf
f7fa4975f4cee1e89031cf44aa1421b22086ea36
26713 F20101111_AAAQWA shishodia_v_Page_127.jp2
30feebe0d36509e386b322942a48c081
fcdd296d79fe72d8a828c52f0f2aa2d7b573196d
F20101111_AAARAS shishodia_v_Page_118.tif
a45ac964de6bcc160e654c946571075b
19e184f4c09089cac97516774891cb511327ed09
29823 F20101111_AAAQVM shishodia_v_Page_112.jp2
04f182b6213f7e0d999002a6a5c37779
ec5e1bd2d30ff682bc6cac7c9a31bdc1398acb76
18076 F20101111_AAAQUY shishodia_v_Page_098.jp2
6e0908304afa3b0e12d00ba77aff5def
b295eee78c5dcad6df41c268e6b704c6449145bb
F20101111_AAARBH shishodia_v_Page_135.tif
f657a170db61b5c890d2a8e9af4a1174
d2cb96e45469e01049d7b1b77a880440adc779b0
27073 F20101111_AAAQWB shishodia_v_Page_128.jp2
d87eb549684ccc2247c3817f47b25d02
ea9a78046b5c1aa4b0281c757a7d4ea37e6d55fb
29872 F20101111_AAAQVN shishodia_v_Page_113.jp2
8e32e7d0d39771a56f8b7eb6249fbbb9
d1ac4c735ec359b26e8757352ed084f130ac3912
650170 F20101111_AAAQUZ shishodia_v_Page_099.jp2
2e65ad0db8a22b300b77d60b58ab0b48
973ee42f880c94887e20b6d5686d67b619646356
F20101111_AAARBI shishodia_v_Page_136.tif
62018da8aa68d35be46545ef6ecb2b72
08df26cfcd2cbe9c14d58574008d8703fddaf7e8
24384 F20101111_AAAQWC shishodia_v_Page_129.jp2
1cc391a63cb97f0fba15cebd718747ba
a9319e2e47ca5e5250590193aa389d37516f8321
F20101111_AAARAT shishodia_v_Page_119.tif
6aa1202936894540fc3c37ef1fb0077c
aa139552d85406bb87b3e43e4e57f06b55fbc97e
21720 F20101111_AAAQVO shishodia_v_Page_114.jp2
b6b64236113f8f61443455d8a751c68f
6962bec25c388a7773850a9d357dd8a2ff4e2fdd
F20101111_AAARBJ shishodia_v_Page_137.tif
5107d1bbd3c242f320c5aa1e85f942a4
0d86d59dad990396df47a8bdde9dffdf6c4c7e6d
26826 F20101111_AAAQWD shishodia_v_Page_130.jp2
7cf9b09500865232acfa873169d270f0
90e48e1262b2b65f4b78b10c0a81d415d76a6d44
F20101111_AAARAU shishodia_v_Page_120.tif
ade7e5446ed2692090963a6a9a7c7a98
ac9905adc40e0eb18bd1fdcec74ba0be11303ce1
22513 F20101111_AAAQVP shishodia_v_Page_115.jp2
c15b7f394465dd9081aae0feafd44756
f90f405d6fc74c3b90074c18eb6a00a408c57941
F20101111_AAARBK shishodia_v_Page_139.tif
5c0e4d89bbc50a7bda1701cc0cfb1293
04562a6461e184d6ecc046b578f7a5cf7c541ecf
26931 F20101111_AAAQWE shishodia_v_Page_131.jp2
86d36e88535dcb98ce12fdfd5e05a0ab
08eaa6928907ecf60003c59d1d107e8a02f3f0af
F20101111_AAARAV shishodia_v_Page_121.tif
5e224bfde5126d86b3c166e378a7a2ba
acefe2a35af928414b9820a27fc455fd0b2ddb18
22122 F20101111_AAAQVQ shishodia_v_Page_117.jp2
4740310d9cb104aafef01ee45e578166
3b2a0d78cb6d8dd09ae6178fca62aa7f4a2437fc
F20101111_AAARBL shishodia_v_Page_140.tif
06993c1c56ceac7a16791422084c9cc8
6786687a6018cd9f3fe6d8da90093dccc8ff3a98
F20101111_AAAQWF shishodia_v_Page_132.jp2
8aaca821d52703e466895345351f8766
ea2c861a7fd8ef513bae5362f4e4ea32a04f5fbb
F20101111_AAARAW shishodia_v_Page_122.tif
6d2a623a881fb712d0efc72793c39d24
a3a80e44349ab39d20d37e18e45ff1b4116e3c90
22683 F20101111_AAAQVR shishodia_v_Page_118.jp2
9a05768da8b2c207c56f5f0bc79269ff
16fb2d2c299fc3d677a06d30cc75db123a500e43
51945 F20101111_AAARCA shishodia_v_Page_016.pro
f2fb10f5bcf420d6b16b665ebe3ae7a2
56c09ef4b2b77438e98fb41fd8accd2fc2c10d79
7908 F20101111_AAARBM shishodia_v_Page_001.pro
7dc5c32659cf99638fb2adc6f08f48af
55a67706ecceb42c9c9e2ac12435add4bd9309a3
850969 F20101111_AAAQWG shishodia_v_Page_133.jp2
77c78700de989f98a32f33cdf8dc34ff
07cb217436abf751e5e259075db58e373f0590fd
F20101111_AAARAX shishodia_v_Page_123.tif
5d9ccefe40f873586b857dfe11a63dd8
e98e605ec538c504ab5d4947f9b8e4c44189eed4
22620 F20101111_AAAQVS shishodia_v_Page_119.jp2
d08fdfef2086ecb67f8661399663961d
d470702eb4008e00c46e6b9ef4f50bc1dfecad16
54251 F20101111_AAARCB shishodia_v_Page_017.pro
a6b7574b76f5af9465886b02283ec912
701a3a3dbba9c0aafd4d55476b5cde78714c4bf8
890 F20101111_AAARBN shishodia_v_Page_002.pro
233bb67ae5c94bb5affb5209dac6b588
89f38aa421fb425539951746410a5a024341e45f
549062 F20101111_AAAQWH shishodia_v_Page_134.jp2
bb344cc8a66f1ac900abd6816042ba33
aa19ba3aea735b4a7da5d6818b086aa742ab1478
F20101111_AAARAY shishodia_v_Page_124.tif
64ae618c3b0149adee45e46ef9e036c7
53f17da16f0f585773dd75da766c70416e16d744
26823 F20101111_AAAQVT shishodia_v_Page_120.jp2
7056e692462be1612cc9928beccffe39
3287c2be46ffefa3630099bbd6bd9ac1a153943e
34516 F20101111_AAARCC shishodia_v_Page_018.pro
6dd8031822748e577f56790328a1fb8f
9e61f662951cc680831d87ae42837f345e42018b
29268 F20101111_AAARBO shishodia_v_Page_004.pro
0b1f43dcc8a63b4b30f5a4ac698c7fad
d0a5c9d48261dd36951d209e42501709b46df8ca
282286 F20101111_AAAQWI shishodia_v_Page_135.jp2
2101284b5478edddb8f8452af6257cc3
f4ff87f4487c26503be087685eb079bd1a164a7a
F20101111_AAARAZ shishodia_v_Page_125.tif
1c7f5fe389450edafd34adad06d1e43e
7801acc029b67c77e9867b2d02f5560451a97905
30105 F20101111_AAAQVU shishodia_v_Page_121.jp2
590778ae06267ffbe249800d8b8b706a
00e8e9cd4a1eb2ea5e6f5bef601134aa049576ec
17898 F20101111_AAARCD shishodia_v_Page_019.pro
5f4435dace8375650090616b694c9d0c
87423cfbf26e81acd0ce61378fe28e6eafae7549
89256 F20101111_AAARBP shishodia_v_Page_005.pro
b699bc112044fa77516ecd93d7f957e4
12fc11d061df2619f1ad5085a774c76b5abbf47e
394021 F20101111_AAAQWJ shishodia_v_Page_136.jp2
0f5bed853b2fb4e28025c68e6e05942a
68c85de8f3f9783d64860e07d7a429aa6bc98a62
30137 F20101111_AAAQVV shishodia_v_Page_122.jp2
2bcb90cdcb9ae6cf6ad61a6a616bd429
fef976d292a9846e8f372e0b794cb078640cf136
F20101111_AAARCE shishodia_v_Page_020.pro
ad407009416b40bc4e8cc9092cc60a12
0a915541865585c4f0501adee5e26473e3681820
70778 F20101111_AAARBQ shishodia_v_Page_006.pro
fdc884c69f45e5ee5c966f8d822798cf
215b535952899c72f8a46a63065cb9dd04545bbc
23565 F20101111_AAAQVW shishodia_v_Page_123.jp2
59647e5bdbb92d07c5616fdfcb89fb97
2ab59f5d43cf8e49a425ca4fa99919d89472256d
33393 F20101111_AAARCF shishodia_v_Page_021.pro
a5d7a5c5b5d8575f123901c592f2a0d5
cfb0e6ac74591f99007ae1e8d15e028e4daf82cf
10029 F20101111_AAARBR shishodia_v_Page_007.pro
1ecfd64ffbaf7493595fc328819a0b22
bd4a06a3a64ee22677abcfdb4c1ee4a8bee4fe4c
F20101111_AAAQWK shishodia_v_Page_137.jp2
657616652aadb97daeed63573391cbd4
54c40d1f069329fbd165467b766612a37aa0fb4b
27493 F20101111_AAAQVX shishodia_v_Page_124.jp2
48b4a610aca468285f2a94fc9d4cc68a
9d5c6a5570e9cee105c89e83dac6b9cfbab9b46d
40378 F20101111_AAARCG shishodia_v_Page_022.pro
5f559649efb36ea4b47940768dffb91a
7c8776f9239ec056573756fc4847ea29a5fed97a
F20101111_AAAQXA shishodia_v_Page_016.tif
8d86ac7b8ef72431526c84e5e4fd3322
509ec395f82ba1bdb453a37d3465e5694084011c
76600 F20101111_AAARBS shishodia_v_Page_008.pro
2c98ee8b8cc8783808b943ab2691d3e3
340b5eb0ee2ea154b3282490269cd2eed60068fa
F20101111_AAAQWL shishodia_v_Page_138.jp2
01e63a3ba03a3696255e0d7e13f669bc
e8116d718bcd163a2b4e3996bdaf79deb49ea0df
27512 F20101111_AAAQVY shishodia_v_Page_125.jp2
70d0f57a4f61cd0c7edb62476895856d
0c4a6559c526d3ec866e35a82e94daab7bfb9f83
26158 F20101111_AAARCH shishodia_v_Page_023.pro
c8429c269543b4936957f69996780e8d
61c64dc8ed87604f48b82967f337655e1bc90b55
F20101111_AAAQXB shishodia_v_Page_017.tif
f948e8e163f190f24471110610dc7513
1f314c8d5833925c446e1ca7895a524194fc9999
73622 F20101111_AAARBT shishodia_v_Page_009.pro
182a04b86e14d305f086c435075d0803
e61bb03a65caaf0978956e4a8b2acd525bed8704
890193 F20101111_AAAQWM shishodia_v_Page_139.jp2
7fdd3217604204856a85da4b75b7e936
cc3a279edef8e36fa7b79e95ce638db91d2a5281
24151 F20101111_AAAQVZ shishodia_v_Page_126.jp2
510ca424a2dc02baee83d11dc6fe8ed8
751e939fa4c01d4e562278008396b8230fc6a1db
35623 F20101111_AAARCI shishodia_v_Page_024.pro
9cff74ac895a99fe8e5574a9722e961d
6834e3e23ac57307b2a54150efffdea0dc6d047d
F20101111_AAAQXC shishodia_v_Page_018.tif
a516d9cd0e7c4d8eb6ec1664d88c0e77
291ed82685326e36967e0fdd01b87dafe6aa4e5e
517666 F20101111_AAAQWN shishodia_v_Page_140.jp2
1a1cfd54cdcf4f25f4fbcfc95e32a881
2187516f1696aaec4054c856df365ac4042069f9
42448 F20101111_AAARCJ shishodia_v_Page_025.pro
a17e4bc815e26595c93b8623454d54a1
7b6a1bff57071edc957f12c76a91db22b6f257b8
F20101111_AAAQXD shishodia_v_Page_019.tif
851a9c7e5ff72264cf39bf489038195e
47a82c05f1e758dfc12358f3405ea00d61c1d66c
81073 F20101111_AAARBU shishodia_v_Page_010.pro
9de5c79b260097e2cac434bea0b9eadd
d9416fc40915c28ded5d8e0cd082ac99e6e4116e
F20101111_AAAQWO shishodia_v_Page_001.tif
758ea172c1dd9a3dea0921c2089a20c2
66add61cfc99140dae23ac761b04cf92c0e4d9c7
50511 F20101111_AAARCK shishodia_v_Page_026.pro
b6b00e8d30c339ba01a8c19ce5302c32
33d57848bc73a7ae5471c0065467e64a37736873
F20101111_AAAQXE shishodia_v_Page_020.tif
48806b91a80b240a6c8b4df49b95d4dd
0dd71c19bc8a56df4c3262ccaeec6dee01c23781
44374 F20101111_AAARBV shishodia_v_Page_011.pro
de07567a05b92ff04d671d4f5ee136f2
386fc6da99f8c9c5e3ae52d5ae835f18cd257e3a
F20101111_AAAQWP shishodia_v_Page_004.tif
fa192e8f692c1ded7f7c03935a83ae7e
9c04475a7d7c63c20ec06bcd2cc2c30e02b0e507
56175 F20101111_AAARCL shishodia_v_Page_027.pro
cb7a3c9525f4bb9d9fb5bceba180f8b1
234042939590c613ba8f9d89b9e48f8b00b7885a
F20101111_AAAQXF shishodia_v_Page_021.tif
45758d9a6f6c1a399a654ebd9fa8a4a2
81700600f7c1683d67a5e9b5327330d9cd5cf007
21860 F20101111_AAARBW shishodia_v_Page_012.pro
f9aea1e319a957384248d6c6e60fbe53
dff5c4c7a08cc42d9fc7f603206116038597d765
F20101111_AAAQWQ shishodia_v_Page_005.tif
83794dbf79d505d614c9df2b7304b463
2181675ee90c7103f35b26bdd2cd2a053c9e5579
52054 F20101111_AAARCM shishodia_v_Page_028.pro
abe21fc05425333c7a889818a998dbb2
e0f3ff1e054e565e00506f59a58a0c73d4837abf
F20101111_AAAQXG shishodia_v_Page_022.tif
67c9608a23aab4887472cf534462b1c3
c0accd2c450065c6aa7554c6c97255ff3b7031bf
53555 F20101111_AAARBX shishodia_v_Page_013.pro
f5c778a2f391645f8da1e5f2ae443858
b7a336ff5738793a4ba96efa79c32718f0b468f1
F20101111_AAAQWR shishodia_v_Page_006.tif
2f8832cf7a5fbc9cd2401403e2de319c
4c747738e12d6b4764495ea77db820abd810e9ab
50096 F20101111_AAARDA shishodia_v_Page_043.pro
d1d84bf6cf32f760ca06863681ae09c5
f4a25094e64f0a39b9ca1292ab1d9ea37092fecd
54301 F20101111_AAARCN shishodia_v_Page_029.pro
f5b1b4d3a3e89220647da266eaca4c14
69c6b97d55777a38d27cf1b02c9a991b0b1ebd64
F20101111_AAAQXH shishodia_v_Page_023.tif
80b40e07f2e9cb32ac3fbf6c1f31f8a2
1147ecf5a38a69e8a855c1ea4e379783bb9716a0
33298 F20101111_AAARBY shishodia_v_Page_014.pro
a0b89d044e4ef05779011b1b15b93128
b98d78ec56df7bc84fbaecdadd46c051c2da2ba7
F20101111_AAAQWS shishodia_v_Page_007.tif
e31b2f459bf9ebb618900f622a1882d0
26333e37921041419af8f020565deae31bbc89da
57072 F20101111_AAARDB shishodia_v_Page_044.pro
0d1ae1cd44269757e0975957049c463e
a1a74a1ed66952e0986986035276e3f3c7766545
1575 F20101111_AAARCO shishodia_v_Page_031.pro
48e0469908f5f3cfe5a34d02f45abb26
718d3de12c8f2fb80b3831caaa8f7526cdec4db7
F20101111_AAAQXI shishodia_v_Page_025.tif
45f674c1581b5d4e197acbb22df01f0b
5d840bb1a17383621c5b700f88a7942e2c6db4fa
54305 F20101111_AAARBZ shishodia_v_Page_015.pro
e67605c2aaf0b13088c6cc96561aba26
db69af0f6a890cfeddee83cd9f250eccf1a6bcce
F20101111_AAAQWT shishodia_v_Page_008.tif
435716119708636b829e4003a814c6f2
660109cd7472f78786bf14025a011c3f935dbb67
55137 F20101111_AAARDC shishodia_v_Page_045.pro
330ed64ae20458ad31c0ffd05a1f05ed
fcdcbb766bdcbce8e850ab9d87b85a83a12666c7
7874 F20101111_AAARCP shishodia_v_Page_032.pro
e1cd14b439d794cbe3180dd145b9a46c
e26cb2dba42b6542113fe39d00af5ee402ea06e8
F20101111_AAAQXJ shishodia_v_Page_026.tif
b7b6a6c6dbefa6047115cd5b106bea8d
a6b6a0a46fdc0f85d04675c0335321ea354c6296
F20101111_AAAQWU shishodia_v_Page_009.tif
f114827556ecf1f8ecabc4d9ae5a2a4e
a65529d3bc473f7f1e38609df48b6dd468e7a987
58616 F20101111_AAARDD shishodia_v_Page_046.pro
ff8c38b7ee99e29df4932f87193abeeb
62a299b53b233ed0872c0a9a3c3436ea79e92a55
4309 F20101111_AAARCQ shishodia_v_Page_033.pro
c7cb1f42ad0069b1e28c34aa41d07f6d
16242dc2034c0d15bfaf241514b1f13917e1b2de
F20101111_AAAQXK shishodia_v_Page_027.tif
3a157d855baaacca475b2550b9d736e1
ecafc324d9672dd5b0ebfbd818bc776cdab662fb
F20101111_AAAQWV shishodia_v_Page_010.tif
a88353d9ae144b897705316c663cc10e
5d44277bca74dfc9cc0b717872afb233ea971018
53849 F20101111_AAARDE shishodia_v_Page_047.pro
884c00b38655e044c885055e0a559207
debdf04be34f69f7057d7bd9667980bf532f5878
4448 F20101111_AAARCR shishodia_v_Page_034.pro
6695497bdfd530eca3760d555ca27644
09c3520d51f37ff6653a6e1b4fb815f5b2cffe50
F20101111_AAAQWW shishodia_v_Page_011.tif
0f2f09f99600b951e170d3da767b84d2
e45d8223efc1feacb7eda86c7cafbe3708a50ea0
49249 F20101111_AAARDF shishodia_v_Page_048.pro
0b821ccfa6c799448a67f66a7c6ffbd1
44f9abfaf04a16ca01bad79c10bc9a579016c43d
F20101111_AAAQYA shishodia_v_Page_043.tif
5bb0fbc6dc55290fbe3355910d56c82e
244cbfe4750dc4705a31a242efb2d2c53192291f
20110 F20101111_AAARCS shishodia_v_Page_035.pro
304fa6df610fecd3b27caa77be56e866
956fd7e2180651ddda858fb2c65646de42522344
F20101111_AAAQXL shishodia_v_Page_028.tif
6d946df8994ca73ee1bd51b2aaf53a26
e978ae8fdf9c057c57054fdea9a8f363e889fa3f
F20101111_AAAQWX shishodia_v_Page_012.tif
57ac3b954f4acbac886a817c328d20c5
ec47c00fc2ad898261c1867abd0f36d5bfa459f6
4975 F20101111_AAARDG shishodia_v_Page_049.pro
cb2ec173380ce8d2f40fb78e2cc4e8fe
00190173fbddde4417ac522100ba4d6b1197c354
F20101111_AAAQYB shishodia_v_Page_044.tif
d0c5122fd75bede64cdf51109506ff1b
abae779dedb85f629ac49686ef6764bcb6cc70dd
23949 F20101111_AAARCT shishodia_v_Page_036.pro
cd5e3a60d01fc19e35fcce06c2356b2b
4bf6bb0511e0ef93bbf4252e669f5ae9070a5eec
F20101111_AAAQXM shishodia_v_Page_029.tif
609071cbd43a9c672be528962a5aa964
fc8f107f66fc884ce6041093e39d145b5c5b2c54
F20101111_AAAQWY shishodia_v_Page_014.tif
36d8b4f8a29e643dd7cbebf5010894e1
a46b587e91e388a08e0ca8707f9ca869c90302e8
11356 F20101111_AAARDH shishodia_v_Page_050.pro
1b60e884a2d47b0f97834c1cf5de145b
ecae6615c3b97823fc7cc2db77b4d7c133214cbc
F20101111_AAAQYC shishodia_v_Page_045.tif
65da6749891394342fb920fec4450614
e7da3f5538fd8df1de0adf86012b4baebed530b0
8962 F20101111_AAARCU shishodia_v_Page_037.pro
6c76ac509bb18a3bf0bc4dd1379fb364
23e3f71d5765e700b6ce1bae1015acbe2b0a0093
F20101111_AAAQXN shishodia_v_Page_030.tif
f0d1876366cf93cb6f41a0b9fc28f2c1
04ce94e84440c47648418203e2e22776ea6de1b9
F20101111_AAAQWZ shishodia_v_Page_015.tif
b412aa4aaeb1f055c79a95a8061c769c
4a6e99ce353d9487266e5414f8474cd54d6a17f8
6773 F20101111_AAARDI shishodia_v_Page_051.pro
d033f68122c1550c4506d7f69a5e333c
07a0d9762a3849dd5a046e71ff14e930beca140d
F20101111_AAAQYD shishodia_v_Page_046.tif
baf27925955f614489a4057ec30932f7
71ff5ffd4fec62c8cde052ae82da17937a106731
F20101111_AAAQXO shishodia_v_Page_031.tif
a0ca3b48e18d2af62f7d7d6fc3f8719c
1b7e9f569073922f742f4c5fc334f90200dacc49
12162 F20101111_AAARDJ shishodia_v_Page_052.pro
6728372d43eabba10936ddbaa2cc04d4
c8bb87ebeefb627c45790d09eab1cd7a07537314
F20101111_AAAQYE shishodia_v_Page_047.tif
603e9c82d3397bd76e462d7b240966c2
3be208b689b467b152fd56499c93a1479a72eeed
51140 F20101111_AAARCV shishodia_v_Page_038.pro
1ff9acc0095ce2ddaaca04bf610b7fce
a23809208cb9e8afa8fe8a1c1e49d7a26c80e147
F20101111_AAAQXP shishodia_v_Page_032.tif
e39001c4029be33b7403c8690bc8a02a
e7d044f3c4deeda563cca5e9fe17a14d54785b2b
20464 F20101111_AAARDK shishodia_v_Page_053.pro
e4a489dbd3ff5dec242500acf65fc15d
34bbc83c52a2af3085e9a9319e8c6199c3696608
F20101111_AAAQYF shishodia_v_Page_048.tif
fa3eff2185efcf07c9500c87f42872c8
da2d7f7acda1f6a1760ff5c650775a0a0b885929
52180 F20101111_AAARCW shishodia_v_Page_039.pro
58a430c2ea1ab4220a63f0d786a9c0da
d8c92adc655a366d4ce070b7aefef4736d44a58a
F20101111_AAAQXQ shishodia_v_Page_033.tif
e0b53f712eee0416fd0bd157156caf90
52a4f7e5dd044b6ab36d779734e52ccc095501c6
6641 F20101111_AAARDL shishodia_v_Page_054.pro
9482eae64554f0bdd4108b12b9f0d515
507a47f6728ea748f8c0bdfb68c694b1812619d7
F20101111_AAAQYG shishodia_v_Page_049.tif
f5dbb9be32e15988a27665bd9a9a7798
60d4404abd1703d4ab5526237856e59df230822d
34259 F20101111_AAARCX shishodia_v_Page_040.pro
08caf8180250af77135d0369c78cce3f
3a2b2a0a5cb57ba3b7d50f10552c5e148a5dbca2
F20101111_AAAQXR shishodia_v_Page_034.tif
52ebb4cb915efd635033063812b59068
cd9dc91055787db07666c71e9a29f486dfd5eb2d
43079 F20101111_AAAREA shishodia_v_Page_070.pro
70a777c00d9dd2b78d6ffe869c19c662
7acd8d22d7cb68c52f47b0a60708b35a221456fb
6743 F20101111_AAARDM shishodia_v_Page_055.pro
cce86572bc6ed355e5c3ec414eb03f53
56a2acf138c258afc040584bc31c1d0925aa46f7
31956 F20101111_AAARCY shishodia_v_Page_041.pro
bc985f587310080679a1d5be0893f4c5
a32f020d72658555299190922bd53472f43391c0
F20101111_AAAQXS shishodia_v_Page_035.tif
229f33b1352171d23104dc5b8de07b4b
4dbb99223af1da63a39b5bc2b63dd536b62201a0
41508 F20101111_AAAREB shishodia_v_Page_071.pro
0ee33d5bbaf0aced4ce0d9506623df1b
dbfd88a33556c2082a2e220c8ed099b31e5b9a77
F20101111_AAARDN shishodia_v_Page_056.pro
c05c6cfee921b66be48e19e8a87fdb74
6cff9cce4e1900141dfb06c5b5835d4419399872
F20101111_AAAQYH shishodia_v_Page_050.tif
ecec6bf2acfb5d1c3b5c1182c8a6783d
a9013a54bc0c789b8d1f155012d623a18c983cc7
54493 F20101111_AAARCZ shishodia_v_Page_042.pro
b1750133756fa581ed2c9855856e4d73
b517b05b1e72cd1d955425a704f1cf5e94eab075
F20101111_AAAQXT shishodia_v_Page_036.tif
35e8034a44d3e14e324a2b297eb039db
e1cb8eec1626feebd54b56050d86c4788e8be53c
46518 F20101111_AAAREC shishodia_v_Page_072.pro
20889770862959b449fc33a6f4c79418
bd06a6f2667d5ed1f9ad011dcb02c3906fc09bdd
10562 F20101111_AAARDO shishodia_v_Page_057.pro
442f2173f520ae270b08daedf16859eb
2899fa09af8570f9ab3a6c618231a6610af5ffa8
F20101111_AAAQYI shishodia_v_Page_051.tif
1f09b6b11bcbfa0b1eb74aae7ded9228
46d8c4677d767b4af0581fccee4d0e0cd8efc861
F20101111_AAAQXU shishodia_v_Page_037.tif
5dc52d5630c61c90838008fd0fe0f452
78498847d8f5a49f0dc95773162cd1dcddf944c6
36774 F20101111_AAARED shishodia_v_Page_073.pro
f2779e99bdb6e4a63d1386008087486c
b2fc408c023daf70e5386ff9f08647d3eb8b0827
8151 F20101111_AAARDP shishodia_v_Page_059.pro
991e57727e90b28fa88866734275ef36
9d1f5d1adf7c0ce9f9fcf51b20061e8dc735ba24
F20101111_AAAQYJ shishodia_v_Page_052.tif
35e5a63a9f3c277e632fdcfe6227e19f
37184aa70f9331a02b56eb220962af427541a7c7
F20101111_AAAQXV shishodia_v_Page_038.tif
30207cbe42657f42b166624c43e10476
f544fc59423b0723aa56864490d652696203fa7b
47906 F20101111_AAAREE shishodia_v_Page_075.pro
c724616946b792050de235f31d2415c7
cec5c95bf65f4d19b69f4608a30f6cbbfd9471ce
7323 F20101111_AAARDQ shishodia_v_Page_060.pro
895ad539df892b5ab38727c704f44fda
9d58d097e6b8d2b4296bb1a97703839dd1d6c50f
F20101111_AAAQYK shishodia_v_Page_054.tif
3f3a4cfff78ddeccd5d6a4e4335eaf3a
6491f6713aebcbbb4f9b3b5e9a1fc560d5f1dbbe
F20101111_AAAQXW shishodia_v_Page_039.tif
d64e7ac878449cd78927ef75cd704b8f
24b86b13bc956cb811dc1e62a3bd553c8e01f2be
32906 F20101111_AAAREF shishodia_v_Page_076.pro
0e4bd16c7ef84981562fa8e2ae57338a
0ddc5477764a9bcd2a9afa566fd65a2d655a4a47
52741 F20101111_AAARDR shishodia_v_Page_061.pro
16bc43428f17bc923e4bb9381d063240
5f285a80c3d385068abb61301bbacac84d01fe9f
F20101111_AAAQYL shishodia_v_Page_055.tif
278bc0a3faf43ee6a4f3087afb971375
e10b05fb21708e0638fdc3067704d4e87b1e3df3
F20101111_AAAQXX shishodia_v_Page_040.tif
732340997df02cedf1107409e00b67ff
c25963f9b5d15cb7a43ddf464fe138762d9d6241
26834 F20101111_AAAREG shishodia_v_Page_077.pro
0c30b94e5b9967242e18d2813fdf1884
d538cbdfdc85fb30340ab544a357156632f713ef
F20101111_AAAQZA shishodia_v_Page_071.tif
3b8f69c8e2e362dd16fdfa7ffa042883
6e062efa0a4443d4aac0ba298283f45fb60021a1
48652 F20101111_AAARDS shishodia_v_Page_062.pro
bd362a739e369610cfc69641c48a9ae6
0ce15ebc445835599b41ba10852cf87a81115c90
F20101111_AAAQXY shishodia_v_Page_041.tif
768249a9d71a7d6fc0faa6da1b9f7361
fbcfd1b9cbdf8c3ef34026560107b77051956642
30208 F20101111_AAAREH shishodia_v_Page_078.pro
cb00266e989e77df96501ee1ce652a20
9971dd90f88a10e1ece8d7fd9721a114220cc942
F20101111_AAAQZB shishodia_v_Page_072.tif
3446a77628c444cd607bd1aec8564a49
e8ce6d3576fae71369e634805746fdcfad17ff37
54387 F20101111_AAARDT shishodia_v_Page_063.pro
37fb633c83010530eeeb912d982e912a
97f88852f7cf728f5f290020010989447ed4ce31
F20101111_AAAQYM shishodia_v_Page_056.tif
dbf883da2b10538675bad9a1655a8e87
0385e5711eb012d4e237f0b85ec4eb62a57cd58e
F20101111_AAAQXZ shishodia_v_Page_042.tif
1a7a9b7462f8aedc2aff166e4a7bbf6b
a729804ce74ad731912563c548864a5e3983636a
28639 F20101111_AAAREI shishodia_v_Page_079.pro
c7e5526d1eb6e6dc293a6c4895e6af6b
b13a69a2d255e142d6bafbf305295357f2b93a56
F20101111_AAAQZC shishodia_v_Page_073.tif
e13d2941cd616d3b7a7080dd0c1d8898
3337bf259447129449d17b974b92fc8780c0fa9d
46764 F20101111_AAARDU shishodia_v_Page_064.pro
124d5f04220f0371e8ea85d95ba6d4d8
50668ec6adf862878047e31e80514e764d47733a
F20101111_AAAQYN shishodia_v_Page_057.tif
bdce3016a313f0c0e953ef82d11a14b5
9eb6ab78880f561b9e64620cdf8ac3a1155ac0e6
43357 F20101111_AAAREJ shishodia_v_Page_080.pro
8ba3ae442d26c004757471890db7e89e
342e7a2994b938031df0171545e2dddd38ee1299
F20101111_AAAQZD shishodia_v_Page_074.tif
6663826e6d0e04be44d772e91eddf0ad
34542d7a85d4156ff16a198b062a51677ece58d3
35150 F20101111_AAARDV shishodia_v_Page_065.pro
a1c6f27b1cc7e5b321044ceccb688438
fcf8ccff1cb0f76a19474b255eae43b6b0741304
F20101111_AAAQYO shishodia_v_Page_058.tif
44f2c57502375f89773daffc05c49913
3c512813f42e5264c580a3efa3c9832aa284e8ac
44805 F20101111_AAAREK shishodia_v_Page_081.pro
f4f614efc20f44b42fb532851714720a
62e0f0bbeafa0d54dc4f4945bfb6432b6ade3d82
F20101111_AAAQZE shishodia_v_Page_075.tif
dfc8b9cebeb0cf1ffc3a2412c8ce242c
ad761ffbea97d5d65cf6c839b0ef7de6b21c0afc
F20101111_AAAQYP shishodia_v_Page_059.tif
12f1ce569a51c2601c96e96aa12cc882
f938f50faaaf47727b1f6736a853d71b6520bfe7
54194 F20101111_AAAREL shishodia_v_Page_082.pro
0ce47c7d1681e3509c5b516425144fe3
fb8da34fd1d015f7fb6de6322d221c791787dcc3
F20101111_AAAQZF shishodia_v_Page_076.tif
ca05291ed8a90ee01bfb9469c9fd8a86
53d55842034a0e7c59d7128c065c36210f43c2e7
48309 F20101111_AAARDW shishodia_v_Page_066.pro
fb5a9adc604f9ee8262802b0725708b5
8da2f04601d587710363733f1007bfba359149c5
F20101111_AAAQYQ shishodia_v_Page_060.tif
92c81cabf8b46e8f377240fcefc4d233
07600161ad66f2de60f32f04b46f11c40a28d93c
5085 F20101111_AAARFA shishodia_v_Page_097.pro
7d52001fea9a19dde1c365fc1c2bac50
70731c59908461ce999044feb0eedc597063f444
55461 F20101111_AAAREM shishodia_v_Page_083.pro
cbad82a6ddb04ea0185e50c620d0098c
c3eb1bf769f4c745cc5657d6507869414ac8bf1b
F20101111_AAAQZG shishodia_v_Page_077.tif
a17f4fca86a8c141dc97816492a12711
ac1536c6a1de55af165137c877f3828e1946b820
44804 F20101111_AAARDX shishodia_v_Page_067.pro
2475695a7e62a8e04d0045a009503c2d
c6baef0757ae7275a1e722a161fe10e0d88467bd
F20101111_AAAQYR shishodia_v_Page_061.tif
ea25a72437cce87eeb542ca0ee8f4c3d
538fe83238609e3b45b5a52a8e5c413003981672
6317 F20101111_AAARFB shishodia_v_Page_098.pro
db4b5e785c3ed1388dbfc98d18aa40a8
67ab985eea79c356958f15c619c3430a6b08e08e
52706 F20101111_AAAREN shishodia_v_Page_084.pro
2d0d90f9c573c3d564cf66864280fbd9
9cd47f404008c19497a75f9d11190bf34e9a6126
F20101111_AAAQZH shishodia_v_Page_078.tif
1c638a76eee18d55242dd5a79e53cbbc
cbab69e0785b832e9b9defe2ebfeb07fc8fe6aa5
34786 F20101111_AAARDY shishodia_v_Page_068.pro
2304f1332aa09e3a396d2447da91b361
f40dea1f4531e016a7e533f34eaaa8f4b6534be9
F20101111_AAAQYS shishodia_v_Page_062.tif
55ab45c6342095a90a939e0557d5615d
ff2047a045cba01883f7c113baa15012fd1aeaef
49440 F20101111_AAARFC shishodia_v_Page_099.pro
6679c709e540fd6f1601cd72c2a34f86
ef1bb0cfa9d99b10f5b78bf6494b9886a30ba17f
51679 F20101111_AAAREO shishodia_v_Page_085.pro
78048afa68fac2c6ed18b454cc40bf73
0aa9b405b76350813c551f817ccebb5e56924441
F20101111_AAAQZI shishodia_v_Page_079.tif
659a6a15771d43ca19200a736d701b6d
a0ecda58bfc3692087267957ad4368ca52711640
35578 F20101111_AAARDZ shishodia_v_Page_069.pro
1de3772611cd3e9bf9660aff145c2b52
cd5e2dd51cb2abd75b1bf3b4d5b1032ba72a1215
F20101111_AAAQYT shishodia_v_Page_063.tif
1945b24a089beec4dae5e9b0c888d935
e01ae79fe25ae0d79dd820f96c503137a8b8f4df
5755 F20101111_AAARFD shishodia_v_Page_100.pro
f2a18f82b18ef94e7ea9fd4dc4e3cdf8
a13e7b9cb262033b09ea7caf46e3c5f8ab4b9a4e
57635 F20101111_AAAREP shishodia_v_Page_086.pro
f45663c73aafb209259e8b7737f50773
6932e9f11e44c55d3119181d41bb702254318dae
F20101111_AAAQZJ shishodia_v_Page_080.tif
f404ab9df33bf9d3cbeabe3c503c55f5
7d8da95d8cc6d9eaa81b6657ccec755d2c0fb036
F20101111_AAAQYU shishodia_v_Page_064.tif
c68444cece7c8ab655fcba3790a8e99a
83b3bb6bd369d58cea6970a3c66e03434f654557
49558 F20101111_AAAREQ shishodia_v_Page_087.pro
7cd636bf542e70c919b5dc02e35aa840
1615ed18fb17911a80125afd442a00cd33705b3a
F20101111_AAAQZK shishodia_v_Page_081.tif
cc63ae5a9a7d1752ceec4c0451b709bc
4cf8419d33533ad6e875b897bb8beb0ec2fb9855
F20101111_AAAQYV shishodia_v_Page_065.tif
e8a610dcc92572c043371392946eefbe
7e9bae96def1dbf370408aa4c47b968d41736e58
5829 F20101111_AAARFE shishodia_v_Page_101.pro
d3da8b903080d6d20b38e33a010a4bf1
c9f4bc001b68df1b9e51881c45349b6f0e2a0ad4
47244 F20101111_AAARER shishodia_v_Page_088.pro
a244b040ccca2a303b8618ae3c7dd7fe
43c687388f71eda12fe189c0b7298be6ecc80248
F20101111_AAAQZL shishodia_v_Page_082.tif
d11bef85ae978ff706016b7e32821202
4ba7acaea4984b957ed0be02adfe962f112ec6b1
F20101111_AAAQYW shishodia_v_Page_066.tif
59294bb0dcd21a67f41c928bf178156c
bf8cfb3090450228feaa96807efa67e06750395f
8848 F20101111_AAARFF shishodia_v_Page_102.pro
2b7a5418f0a36f640a43e4ec516a1e3c
883289a72cf38a22ea36d60c8b093c03245eb8e6
55167 F20101111_AAARES shishodia_v_Page_089.pro
13500ebabeee829e32cf534bf5dd6385
9472962fcd67840a6522f66e102e60a1265998e9
F20101111_AAAQZM shishodia_v_Page_083.tif
653f16416324dd6aaec917d66d286d69
ad492795df55c252758e7bd0a382053a48d306c2
F20101111_AAAQYX shishodia_v_Page_067.tif
11b124403eec948b4d30b2adc9a5001b
c0c6e8c416821d52fa6f2b765cd00ad6f28f5aed
5837 F20101111_AAARFG shishodia_v_Page_103.pro
85091155c78a605a00a7f9677f81d659
8df34e89f3833c4fe71c06a27e85426cf94f0a84
53367 F20101111_AAARET shishodia_v_Page_090.pro
fb5c01abe58eed00b47a8551fc1d444c
968e6242cc80fa512b47bf8be20c2104f6c10b84
F20101111_AAAQYY shishodia_v_Page_068.tif
77b367c33738f5c0c2edf3bc963d52e2
8215f9652f81d3bb2c5ab9e28da68d513c459f66
4055 F20101111_AAARFH shishodia_v_Page_104.pro
a51bf7cf568528860d035548859081fe
6d70a413b0144cbc93056e15c4f37ae916b15ce2
55341 F20101111_AAAREU shishodia_v_Page_091.pro
1d95c38b119b7b35cee9796f37eb9bc2
bd74a043ec2ac64a5df2cfa461821d87010a4898
F20101111_AAAQZN shishodia_v_Page_084.tif
06d205b0473c4d987063371c8d907dd0
303f40a78112b33f4c3002792a866030a2f8a7b4
F20101111_AAAQYZ shishodia_v_Page_070.tif
8ad5cdd065c52ac21b96f2e4e7080749
d10f9df8f20c3c47983b12bf6efc4b4ad77bf509
6601 F20101111_AAARFI shishodia_v_Page_105.pro
33a246b20e277064c012d95b590bc9bc
1c84a18c5698b1ffe677186ee3b404cee17f8d88
45452 F20101111_AAAREV shishodia_v_Page_092.pro
e1f7327e13960c266b056bfb374174fa
0c688c475c44f866790dbaf2a9acd955e8a448b0
F20101111_AAAQZO shishodia_v_Page_085.tif
c4906cae0ebf040747eb086790bf9cf3
36cfba6f046bc1db9d9b27318b2e9c152e521e6c
5670 F20101111_AAARFJ shishodia_v_Page_106.pro
564111db97a7b40597783c5e41a96a68
81a1fd5b8bd953946e98d7616d6fdfad510e9df7
51870 F20101111_AAAREW shishodia_v_Page_093.pro
5cfb423d4b2a37f6dc365714b035eb27
40e7f7ddaa5ba98a6064f54d2390dd3ab033173e
F20101111_AAAQZP shishodia_v_Page_086.tif
85f4744fe51a0af456236d21f0e60bae
db28d2ce735c62dbc818e1d9ad8a6adfebf3fe3c
6018 F20101111_AAARFK shishodia_v_Page_107.pro
ef4f56e45c62cc8dcc7c40c119ba30c4
9e51ea0f6b70a44d323aeffe9ce1135c50d9a8fe
F20101111_AAAQZQ shishodia_v_Page_087.tif
3f7060ebdf618eab3f81f8e53cf54cdf
cb3f69dcca0e6c8aa554a0ed63999c3c9af14922
4202 F20101111_AAARFL shishodia_v_Page_108.pro
0bc6f0a675f27879939c06f170832b3b
95894428f308a18dcd49af591bb6be6805da7a2f
49866 F20101111_AAAREX shishodia_v_Page_094.pro
b8fed08acfbb12df6de26b9705001926
4fef29d286bb78c6a0d8c118622562f3d07576a7
F20101111_AAAQZR shishodia_v_Page_088.tif
d5e6ee6bdaa3861be425f91d1bc34937
308574c1b649fc2dcd372ce614531cd217cd0832
7946 F20101111_AAARGA shishodia_v_Page_124.pro
b5b7fe6f7fa2ea677b4f882b069ce5fd
5ed3241bfe13d575e998b8aa1a4434316a9f20ab
3876 F20101111_AAARFM shishodia_v_Page_109.pro
a327e894cf19c01083ecf5c8a74ff894
ce95322379565fe6e6961ff8c628f166cdd92fa5
42953 F20101111_AAAREY shishodia_v_Page_095.pro
69a4eed9fb4a8a04ee00320156718d5d
832d0ffdde7c5bfa1a865782f1cb6db4816eb8eb
F20101111_AAAQZS shishodia_v_Page_089.tif
c5a76b518a484a45b7a5f8cb08498aec
75dc1ae494557d1a0c69fd2271e12ef2d2960d5f
7238 F20101111_AAARGB shishodia_v_Page_125.pro
d6f3bcb640a47418b6e16212967d4bc5
a0acebe20add83d7f842af8f6f5605835f7c69e4
6630 F20101111_AAARFN shishodia_v_Page_110.pro
3b5b967887dbb409dbb37c5107823e7f
2029bb6d9b8b52a7d496ee8250f76f43b5380eec
4790 F20101111_AAAREZ shishodia_v_Page_096.pro
837ede64fc03c2a38f1d205af63a2616
5b079a56dab5fc22b23943a40ba01d2e45ddce83
F20101111_AAAQZT shishodia_v_Page_090.tif
50d8a41cc761b811835138d59448ac9c
228e9f72e6dca6d1b6913f04e62e8e67b6185f61
6742 F20101111_AAARGC shishodia_v_Page_126.pro
3895c0cae9dd5ad291274a734347518d
eb8ffc027a82042742e72c5824f7615260203cc2
10364 F20101111_AAARFO shishodia_v_Page_111.pro
1ff764cf4bafb908a57d92676d7eb713
899a4a183f9f78b4e6932993ee4575c980ca17e4
F20101111_AAAQZU shishodia_v_Page_091.tif
6eb9fc9675d0d0edb5e8258ac334f9bb
daf7967e500de55123900b9066bbd2c5d27d287b
7383 F20101111_AAARGD shishodia_v_Page_127.pro
3cb4de6755655a3229ce0e0a1c01f44e
6e6ace51c843c63112161e14eabe0d377f8a41d9
11352 F20101111_AAARFP shishodia_v_Page_112.pro
2aab75fc0050a4cde9f7d694e83f80c2
ec3a589d9f01175f265bde0489699b83e0579b73
F20101111_AAAQZV shishodia_v_Page_092.tif
d528965e518de38e5cc8d24e53e5e20a
ff5394c322054c87e9368c927682cb6c9c098209
7553 F20101111_AAARGE shishodia_v_Page_128.pro
05fcd795fb40863ad170e0eebb0203e2
dc6104c53b0627b1e31e9f33e216fd1afc0dcaa9
6240 F20101111_AAARFQ shishodia_v_Page_113.pro
bc7ca866a910b9f07579e0f4401b732a
8712aa69c4be1b673e360f76440f4aa5125b2f17
F20101111_AAAQZW shishodia_v_Page_093.tif
826f2ccceac77b2efde4f480cb1dd4c3
232cc8b5f4016822ae405bf26a48e0c34b838296
6057 F20101111_AAARGF shishodia_v_Page_129.pro
b9c7612ec4bfa3a1faf1016d709de061
8ce5fbcbaee5e55f32496c6cceeb3939ae20aaef
6991 F20101111_AAARFR shishodia_v_Page_114.pro
2a9cd5909ab6f0e9220ec32166d841c3
27c879416685c785afebaff43efeae06f49cd57f
F20101111_AAAQZX shishodia_v_Page_094.tif
7d0194e743cee142b44c9c75183152de
6e26b4395fde0bd0b0fdb7ae4842d4f05f1cef9c
7360 F20101111_AAARGG shishodia_v_Page_130.pro
af0bebf6f385fa273ba601ae161a11b2
c21b848183a47dfb48463ed8a9ebfd249ba88589
5667 F20101111_AAARFS shishodia_v_Page_115.pro
d79b3b2b9521497df444c1a8c5b80c54
e673c0042e452e1a4b806cde43ecfb7ad6921010
F20101111_AAAQZY shishodia_v_Page_095.tif
896009cd959c4021c40f69912a979aba
bd9de7b2f19ce5f39ba95ea4f4c22e773c6b26da
8232 F20101111_AAARGH shishodia_v_Page_131.pro
136e98cb8c63a774749f0688f7acfae0
7af3e8144a2d6786a8cdaeb97a7bece234252816
5466 F20101111_AAARFT shishodia_v_Page_116.pro
28d2f302d3641c8341160fde1880d3ac
176fc06edf4ca3f84c815d50c86e6c9c7c547ce6
F20101111_AAAQZZ shishodia_v_Page_096.tif
33a63ff13ce373ef6d1300dbc866dccd
7b9d6d621a948b7b1dc2004a75291765fcc2d240
48835 F20101111_AAARGI shishodia_v_Page_132.pro
df24b8dd34a0955a63ee4bc2d93d511a
bb54f79ee00cac5ae909898d904f308866a873ec
6895 F20101111_AAARFU shishodia_v_Page_117.pro
fbc46863f67fab08de0c690ec71dde15
3a706a6687d217c7bbd3ae2666b1e5c7118b80b0
38158 F20101111_AAARGJ shishodia_v_Page_133.pro
866243c96feed47d7356f96243d18c0d
ec9111547a69e0044207fd0a3412f24918d8c872
6514 F20101111_AAARFV shishodia_v_Page_118.pro
671c8ccbbf65e7b65602e2d9f04cb693
bad014b41a4e1ea89b7b974caa747e8a8d9101fe
26886 F20101111_AAARGK shishodia_v_Page_134.pro
0e86663a0a80fa0d574eaea6aa3a56e7
52a7e3b0cd6a15424d823907137c7d5466a0a616
4928 F20101111_AAARFW shishodia_v_Page_119.pro
a2a5199b406fa1c89fb5308868a0dcbb
a249ce3b06b3254f97f11bbc00d2daf8e76be92b
16238 F20101111_AAARGL shishodia_v_Page_135.pro
20aae293be7b1d77d663766ea4f0020b
def225323999096825e000086eca101116f7cfab
9190 F20101111_AAARFX shishodia_v_Page_121.pro
aa135e8aca14fe2cfb740b5486db4c1d
7850171628200bc1368267d2fe7d46c7edd4bc0c
866 F20101111_AAARHA shishodia_v_Page_012.txt
27a3f225efea53626a2c996d6f6c116d
87b3aa01dce1446943791e00be332cf3ca033e23
15869 F20101111_AAARGM shishodia_v_Page_136.pro
de95f8df68effd0d99aab50142a314c3
b2eeba3b477af17ab700fd6b7a2760c82d523cba
F20101111_AAARHB shishodia_v_Page_013.txt
d94741546d88c7e9f750f89241cf12e7
0342d9111a86ec13345dd33369628588d3ecde4b
66514 F20101111_AAARGN shishodia_v_Page_137.pro
9eb32dfe356727523d4a4963d62c0845
18c73093b29170536876bf28108fc2dbea065249
10894 F20101111_AAARFY shishodia_v_Page_122.pro
9452eccc4ff7dffb26cc53d0d912060d
ae998cf0c89d3de6df80e8f42d1a92438c7e1b3e
1324 F20101111_AAARHC shishodia_v_Page_014.txt
1efe226c58f5dc3659fa4d23fc187f57
a4c14b326a4e94b4e7a23821a4457eba34459cce
63532 F20101111_AAARGO shishodia_v_Page_138.pro
c595df1c544d1cff6058b08485b6bc1d
b770691c57a5cba21b05acabbf55f99d1f54e315
6136 F20101111_AAARFZ shishodia_v_Page_123.pro
da52d2ae67a599cdfe30f8d43ed3d353
2c7a46fa8578c4431beed6d0a2cc1f7336f2b7e2
F20101111_AAARHD shishodia_v_Page_016.txt
cc4bb08e35c739680349e45ac4c09a94
d2e1a5d4183349df6251d3276917b62f4994788d
39522 F20101111_AAARGP shishodia_v_Page_139.pro
5c461392bd6d41eaf1b018f601f89d6d
a69487960c84d440535187ef4b53f913541ecca5
2207 F20101111_AAARHE shishodia_v_Page_017.txt
f8dfda0969b9f4fea93789eefe6b9829
06cb868565ca30c142c37fc8d8ac4317db0665e1
22422 F20101111_AAARGQ shishodia_v_Page_140.pro
6db41da4e1fd5b6b4f375d1fc89c7057
e857f5633627c05b03dae4a73bcaf8ac3b4c9870
1963 F20101111_AAARHF shishodia_v_Page_018.txt
cb3b6aa79a8f90fc697e1031ede1e21e
3816b9f5842f4d7e347df8f8a7c5d2ee4522d59f
F20101111_AAARGR shishodia_v_Page_001.txt
e56914d32ce103bbc9c9b531cd7de69f
00a313fb68b4a9dd34499923504b6eb1a269b74c
1276 F20101111_AAARHG shishodia_v_Page_019.txt
72caffcb2bf89cc9afd1c565a7e9e0fb
30ed24672bf80effa8de3ebc9161abe63874f7d0
92 F20101111_AAARGS shishodia_v_Page_002.txt
4bfc0f03625ceedcb90a47bc44e8731e
df4464025e7b55ef0416bfad2024d8c1c4e85b1b
1763 F20101111_AAARHH shishodia_v_Page_020.txt
dfbc0bc74de56ca7f03102b43f924121
87812e180fd8303359460b3ecefd6b0c099f59c7
1210 F20101111_AAARGT shishodia_v_Page_004.txt
fdd4df61331fe77da4adbd96a182a05f
9138907919c79dbb56f255546dbea88c693d5eee
2129 F20101111_AAARHI shishodia_v_Page_021.txt
442c63dfc14b7c2568bd72c1467e4fea
15e5735b66a9fbe6fec6249e23393703bc5c172d
3661 F20101111_AAARGU shishodia_v_Page_005.txt
2b5c150eb6ca2c78adb98d28fdd8189c
e5de9737a327c54f0a4b7c684d4285818baaf899
2036 F20101111_AAARHJ shishodia_v_Page_022.txt
ecd08f7df4e26cf71c927d6a801d3c61
e72389f5f234ac23aa0326f7d7d083c4f76b8f12
2981 F20101111_AAARGV shishodia_v_Page_006.txt
9a07622cd40ed706169bab8a96a3c831
615fbcabf5cc4a9ad1a3d9d2e2932f79d411ce7d
1789 F20101111_AAARHK shishodia_v_Page_023.txt
3e42fd9df581a1f1577fa0847d81e7b7
cafa4fdb31144f789819eca0b8a3a768dfcdc6a2
442 F20101111_AAARGW shishodia_v_Page_007.txt
b5a3a54c96bce2a16f775015adda3559
d4bb72797fd8b59de183e5d39cc1af55d3ba7600
1850 F20101111_AAARHL shishodia_v_Page_024.txt
37aa73c1297dd998b5467fef2d235190
e267a22d5d0c21bb9fbcfc043aca43c538d469a4
2913 F20101111_AAARGX shishodia_v_Page_009.txt
e7a4bab65bf37ba4cb7191a28edc4cc0
bc73c0c724fd0c293e9871eac8d4e72c0a974a9e
1894 F20101111_AAARHM shishodia_v_Page_025.txt
2be49866294041416490635beee4f90b
b582b3f2040a1c730edcf0622e88b6a39928f26b
3271 F20101111_AAARGY shishodia_v_Page_010.txt
9991fdc9587cc8b59d0b78d558fff5df
233ae2e0316a72fd94b4457d53324cd4d919fc4d
2139 F20101111_AAARIA shishodia_v_Page_041.txt
9466b9041fec8479fdbe1d548ba268a5
0766fac53018250042e3f1f4e577ad5ced31f2a7







ENHANCED CONTROL PERFORMANCE AND
APPLICATION TO FUEL CELL SYSTEMS



















By

VIKRAM SHISHODIA


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2008


































2008 Vikram Shishodia



































To Carmen









ACKNOWLEDGMENTS

I would like to express my deepest gratitude to my advisor Dr. O. Crisalle for his support

and guidance without which this work would not have been possible. I thank the members of my

supervisory committee, Dr. H. Latchman, Dr. G. Hoflund, Dr. W. Lear, and Dr. S. Svoronos, for

their guidance and serving on my supervisory committee.

I thank my colleagues in the research group who provided insightful conversations on my

research topics, besides being great friends. I would especially like to thank Christopher Peek

for providing the sample code for ramp tracking which expedited the progress on the problem

significantly. I also thank him for all the insightful discussions. I thank my parents for their

love, support and encouragement that they have given me throughout my life and during the

completion of this work.

I would like to express my deepest gratitude to my spiritual teacher Gurumayi

Chidvilasananda who has been there for me during every step of my life. Finally, I wish to thank

my wife and kids, who have been very supportive, loving and understanding during this journey.









TABLE OF CONTENTS

page

A CK N O W LED G M EN T S ................................................................. ........... ............. .....

L IST O F T A B L E S ...................................................................................................... . 7

L IST O F FIG U R E S ............................................................................... 8

A B S T R A C T ........................................... ................................................................. 1 1

CHAPTER

1 IN T R O D U C T IO N ......................................................................................................13

2 VIRTUAL CONTROL LABORATORY ............................................................... 15

2.1 Introduction ................................................................ ..... ..... ......... 15
2 .2 O bjectiv e ................................................................... ............ 17
2 .3 Inverted P endulum ................................................................17
2.4 C control D design ..................................................................................... ........ 20
2.5 Realization of an Inverted-Pendulum VCL ...................................... ........... ....25
2 .6 C o n c lu sio n s ............................................................................................................. 2 9

3 PI AND PI2 CONTROLLER TUNING FOR TRACKING THE SLOPE OF A RAMP .......38

3.1 Introduction and B background ................................................................................. ..38
3.2 Problem Statem ent and A pproach......................................... .......................... 40
3.3 Results and Discussion............................................. 43
3.3.1 Tuning Parameters of Controllers ........................... ... 43
3.3.2 Comparison of the Performance of the PI and PI2 Controllers ....................44
3.3.3 Com prison of M etrics..................... ........ .. ...................... ............... 46
3.3.4 Comparison of PI (ITAE) Controller with Literature Precedents .....................47
3.3.5 Local Minima versus Global Minima ........................................................48
3.4 Conclusions ................... ................................... 48

4 GENERALIZED PREDICTIVE CONTROL FOR FUEL CELLS ....................................61

4 .1 Intro du action .......................................................................................................6 1
4.2 Fuel C ell System B ackground...........................................................................61
4.3 O objectives of the R research .......................... ............................. .. ........... .. 64
4.4 Fuel Cell M odel ................................................................................64
4.5 Literature Precedents fo Fuel Cell Control Designs .......................................... 66
4.5.2 F eedforw ard Strategy ........................................... ................... .. ................67
4.5.2.1 Static feedforw ard controller ............... ........... ....................67
4.5.2.2 D ynam ic feedforw ard controller........................................................68









4.5.3 Combination of Static Feedforward with Optimal Feedback Controllers ........69
4.5.3.1 Case where the performance variable is measurable..........................69
4.5.3.2 Case where the performance variable is not measurable....................71
4.6 G eneralized Predictive Control ......................................................... ............... 72
4 .7 B battery of O b servers .......................................................................... .....................79
4.8 Sim ulation Studies and R esults......................................................................... ...... 81
4.8.1 Generalized Predictive Control Results ............................................82
4.8.1.1 Case where the performance variable is measured.............................82
4.8.1.2 Performance variable not measured........................................83
4.8.2 The GPC Approach Evaluated for Robustness...........................................85
4.8.2.1 Case where the performance variable is measured...........................86
4.8.2.2 Case where the performance variable is not measured.....................86
4.8.3 Comparison of the GPC Strategy with Prior Control Designs.....................88
4.8.3.1 Case where all states are measured-sFF with LQR feedback control 88
4.8.3.2 Case where all states are not measured-observer design....................89
4.8.3.3 Comparison of controller performance with respect to robustness ....90
4.8.4 Feedforw ard Control D esigns ........................................ ....................... 92
4.8.4.1 Case of original m odel......................................... ......................... 92
4.8.4.2 Case of model uncertainty ................. ........................ ............... 94
4.9 Conclusions ........... ............ ........................................... ............... .. 95

5 CONCLUSIONS AND PROPOSITIONS FOR FUTURE WORK................................132

5 .1 C o n c lu sio n s ............. ........... ..... .................. ................. ................ 13 2
5.2 F future W ork ............. ........................................... ............................. 133

APPENDIX

A OFFSET BETWEEN AUXILLIARY AND ORIGINAL RAMP................... ............ 134

B OBSERVER DESIGN USING TRANSFER FUNCTION...............................................136

L IST O F R E F E R E N C E S .................................................................................. ..................... 137

B IO G R A PH IC A L SK E T C H ......................................................................... ... ..................... 140









LIST OF TABLES


Table page

3-1 The PI2 controllers optimized tuning parameters linear least square fit equations ..........52

3-2 The PI controllers optimized tuning parameters linear least square fit equations. ............53

3-3 Values of the plant parameters used compare the performance of the controllers. ...........53









LIST OF FIGURES


Figure page

2-1 Inverted pendulum ................................... ..... ............. .......... ....... 31

2-2a Front panel of the VCL where all states are measured. ............ ........... .................32

2-2b Front panel of the VCL showing observer ............. .......... ..................................33

2-3 Interaction panel of inverted pendulum VCL. ................ ..... .......... ...............34

2-4 Controller tab of the navigation panel. ................................................... ............. .....35

2-5 A analysis tab of the navigation panel .......................................................... ......... .....36

2-6 Simulation tab of the navigation panel. .................. ..................................................... 37

3-1 Ramp r and auxiliary ramp ra with constant slope, a. ................................................. 49

3-2 Closed loop transfer function representation of plant and controller. .......................... 49

3-3 The PI2 controllers optimal tuning parameters, using ITAE (A, B), IAE (C, D), and
ISE (E, F) as the optim izing m etric. ............................................................................50

3-4 The PI controllers optimal tuning parameters, using ITAE (A, B), IAE (C, D), and
ISE (E, F) as the optim izing m etric. ............................................................................51

3-5 The PI2 and PI controllers' ramp tracking and slope tracking performance using the
optim al ITAE control param eters. ............................................ ............................. 54

3-6 The PI2 and PI controllers' ramp tracking and slope tracking performance using the
optim al IAE control param eters ................. ........................................... ............... 55

3-7 The PI2 and PI controllers' ramp tracking and slope tracking performance using the
optim al ISE control param eters. ............................................................. .....................56

3-8 The ITAE, IAE, and ISE metrics comparison for three plants, using the PI controllers
A ), B ), an d C ) ............................................................................. 57

3-9 The ITAE, IAE, and ISE metrics comparison for three plants, using the PI2
controllers A ), B ), and C). ..................... .... ............ ................. .... ....... 58

3-10 The PI controllers tuned using the ITAE metric compared with Belanger and Luyben
and Peek's controllers for three plants A), B), and C)...................................................59

3-11 Contour plots for the PI controller tuned using the ITAE metric for the three plants
A ), B ), an d C ) ............................................................................. 6 0









4-1 Schem atic of fuel cell system ................................................ ............................... 96

4-2a Fuel cell system showing input u, disturbance w, and outputs z1 z2, yz Y2, s3...................97

4-2b Fuel cell system showing sFF with feedback controller ...................................98

4-3 Matrices defining the LTI model for the fuel cell model excluding sFF .........................99

4-4 Matrices defining the LTI model for the fuel cell including sFF. .....................................99

4-5 The sFF control configurations for fuel cell system. ................ .............................. 100

4-6 The dFF controller: (a)Schematic diagram, and (b)transfer function representation. .....101

4-7 The sFF schematic with feedback controller. ...................................... ............... 102

4-8 The GPC design in feedback block diagram................................... .............103

4-9 Disturbance profile used for simulation purposes. .................................. .................104

4-10 The GPC control strategy implementation on the nonlinear fuel cell model in the
case when the controlled variable is measured. ........................................ ............... 105

4-11 The GPC feedback with four observers control scheme implementation on the
nonlinear fuel cell m odel. ...................... .. ...................... .. .... .... ............... 106

4-12 The Norm of errors from the battery of observers.............. .... ............... 107

4-13 The switching pattern of the battery of observers........... ................... ... .................108

4-14 Final voltage to the com pressor. .............................................. ............................ 109

4-15 Observer 1, error between measured and estimated values..........................................110

4-16 Observer 2, error between measured and estimated values .................. .................111

4-17 Observer 3, error between measured and estimated values ................... ....................112

4-18 Observer 4, error between measured and estimated values ................... ....................113

4-19 The GPC feedback with three observers control scheme implementation on the
nonlinear fuel cell m odel. .......... .. ........ ......................................... .............. 114

4-20 The GPC feedback with two observers control scheme implementation on the
nonlinear fuel cell m odel. .......... ...... ........ ... .. ... ... ... .... .............. 115

4-21 The GPC feedback with one observer control scheme implementation on the
nonlinear fuel cell m odel. .......... ...... ........ ... .. ...... ... .... .............. 116









4-22 The GPC control strategy implementation on the nonlinear fuel cell model with a
parameter changed from the value used for control design. ......... ..............................117

4-23 The GPC controller with the LQG observer control strategy implementation on the
altered nonlinear fuel cell m odel......... .................................................. ............... 118

4-24 The GPC controller with the 4 observers control strategy implementation on the
altered nonlinear fuel cell m odel......... .................................................. ............... 119

4-25 Comparison of the GPC control strategy with the sFF controller combined with LQR
feedback strategy on the unaltered nonlinear fuel cell model when the performance
variable is m easurable.......... ................................................................... ....... .......... 120

4-26 The sFF with the LQG observer and LQR feedback, compared to GPC with the LQG
Observer control strategy implementation on the unaltered nonlinear fuel cell model
when the performance variable is not measurable ..............................................121

4-27 The sFF with the LQG observer and LQR feedback, compared to GPC with the 4
observers control strategy implementation on the unaltered nonlinear fuel cell model
when the performance variable is not measurable ..............................................122

4-28 The sFF with the LQR feedback, compared to GPC, when performance variable is
measurable on the altered nonlinear fuel cell model. ..................................... ........... 123

4-29 The sFF with the LQG observer and the LQR feedback compared to the GPC with
the LQG observer control strategy on the altered nonlinear fuel cell model................... 124

4-30 The sFF with the LQG observer and the LQR feedback compared to the GPC with
the 4 observers control strategy on the altered nonlinear fuel cell model. ....................125

4-31 The sFF and dFF strategies and the GPC control strategy, performance compared
when applied on the unaltered nonlinear fuel cell model. .............................................126

4-32 The sFF and dFF strategies and the GPC control strategy with the LQG observer,
performance compared when applied on the unaltered nonlinear fuel cell model ........127

4-33 The performances of the sFF and dFF strategies and the GPC with the 4 observers
control strategy compared when applied on the unaltered nonlinear fuel cell model. ....128

4-34 The sFF and dFF strategies and the GPC control strategy, performance compared
when applied on the altered nonlinear fuel cell model. ................................................129

4-35 The sFF and dFF strategies and the GPC control strategy with the LQG observer,
performance compared when applied on the altered nonlinear fuel cell model. .............130

4-36 The performance of sFF and dFF strategies and the GPC with 4 observers control
strategy compared when applied on the altered nonlinear fuel cell model................131









Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

ENHANCED CONTROL PERFORMANCE AND
APPLICATION TO FUEL CELL SYSTEMS

By

Vikram Shishodia

May 2008

Chair: Oscar D. Crisalle
Major: Chemical Engineering

The inverted-pendulum virtual control lab, a simulation environment for teaching

advanced concepts of process control, is designed using the LabVIEW software tool. Significant

advantages of using this simulation tool for pedagogical purposes include avoiding the potential

issue of schedule conflicts for securing equipment-access time in a physical laboratory and

providing a learning resource that becomes accessible to students located in remote geographical

places.

A set of tuning relationships are proposed for standard proportional-integral controllers and

proportional double-integral controllers for the purpose of tracking the slope of a ramp trajectory.

Three different performance metrics are investigated to serve as the criteria for optimality, and a

numerical optimization procedure is used to minimize each metric over 20,000 different plants.

The proportional integral controller with tuning parameters selected to optimize value of the

integral of the time-weighted absolute error is recommended for tracking the slope of a ramp

traj ectory.

A generalized predictive control (GPC) strategy is proposed for a fuel cell system, where

the controller incorporates a measured disturbance in the control design. The control objective is

to maintain oxygen excess ratio at a prescribed constant value. The performance of the GPC









control design is compared with that of the controllers proposed in literature for various

scenarios including model uncertainty. The GPC controller has zero offset when the

performance variable is measured and performs better than competing designs offered in the

literature. The GPC controller is also robust with respect to model uncertainty. A battery of

observers with a switching strategy is proposed for estimating the value of the performance

variable when it is not measured. The GPC controller with a battery of observers has no offset

demonstrating better performance than analogous designs proposed in literature. However, the

control performance is not robust when the estimator battery is used and linear models used for

observer design are uncertain given that the response offset is not completely eliminated.









CHAPTER 1
INTRODUCTION

Issues relevant and critical to process control are discussed in this study. Chapter 2

investigates the design and implementation of a virtual control lab (VCL) for "The Inverted

Pendulum" problem. The LabVIEW software is used as the platform for simulating the Inverted

Pendulum model. The objective is to have a visual computer based application by virtue of

which advanced control concepts can be shared and taught to the audience who are primarily

students studying control theory and its applications. The VCL is designed in a manner such that

various scenarios for the implementation of the controllers can be achieved. The user is given

the choice of operating the application with the system in open-loop or closed-loop

configuration. The controller can be tuned manually by the user or use the tuned control

parameters computed by the specific control algorithm. The user is allowed to alter the values of

the poles for the closed loop system and see its visual impact by simulation performed by the

VCL. The VCL provides the opportunity to be operated in the scenarios when all the controlled

variables are measurable and also when all of them are not measurable. In the case when the

performance variables are not measurable an observer is incorporated in the control design to

estimate their value. The impact of all the changes performed in the VCL are displayed visually

by the animation of the inverted pendulum system. This key feature of the VCL allows the user

to see the visual impact of changing different components of control system and hence

facilitating the process of learning.

Chapter 3 discusses the problem of controller design tuning for tracking the slope of a

ramp. The control objective is to place the output of the system in a linear zone parallel to a

ramp trajectory. A first order system with time delay is considered for this study. Two kinds of

controllers are used for this study namely, proportional-integral and proportional double-integral.









Both the controllers serve the purpose of positioning the system in the desired linear zone which

is parallel to a given ramp profile. There is an offset with respect to the ramp observed when

only one integrator is used. Zero offset with the ramp is observed when two integrators are used

in the controller. In both cases the control objective is met which is to track the slope of the

ramp trajectory.. Three different metric are employed to evaluate the performance of the

controllers. The MATLAB platform in conjunction with SIMULINK module is used for

acquiring the optimized controller parameters.

Chapter 4 discusses a generalized predictive control (GPC) strategy proposed for a fuel cell

system. The control objective is to regulate the value of the performance variable i.e., the

oxygen excess ratio at a desired value. The performance of the GPC control design is compared

with that of controllers proposed in prior literature. Various scenarios are considered, including

the cases of model uncertainty and unmeasured performance variable. The GPC controller

exhibits zero offset in all cases when the performance variable is measured, and also ensure zero

or negligible offset when the performance variable is estimated via a battery of estimators.









CHAPTER 2
VIRTUAL CONTROL LABORATORY

2.1 Introduction

There is a need for the development of internet-based non-conventional pedagogical tools

for delivering knowledge to students on various topics of study. The drive stems from the

various advantages that these environments offer. First, these applications do not depend on the

availability of a physical setup or facility to run experiments [1, 2]. They are also not limited in

terms of the number of users who can access the application at any given time, as long as

appropriate adjustments are done in the server side of application. There is also no adverse safety

issue or concern of damaging expensive equipment when the product is not used correctly. Less

training is required for the user to be able to run the tool. Compared to a traditional physical

laboratory setup, in these virtual environments, there is more of an opportunity to be able to

realize a physical system and introduce more advanced topics and see their effects on the system.

A software application that simulates the behavior of a physical system, provides animation to

depict how the system behaves, and provides an interface so that the user can observe changes

made on the system performance, is highly beneficial from a learning and educational

standpoint. The application is referred to as a virtual laboratories since the nature of the "Lab" or

the application is "virtual" as it is a software emulator of the physical plant and can be

potentially used to remotely control actual physical equipment via web and networking [1, 3-6].

From the perspective of enhancing the learning experience, the virtual control lab (VCL)

supports learning by all three modes, namely active, flexible, and discovery learning. In active

learning, tools and material are made available to students so that they can use these resources to

actively learn and reinforce the theoretical concepts. Traditionally, physical laboratories,

equipment and experimental apparatus are provided to students to reinforce and test the









understanding of the student's comprehension of the theory. With limitation of available

resources and costs associated with the overall management of logistics, at times this can be a

challenging task, which can prove to be not only quite expensive, but also involve safety issues.

In those circumstances the VCL can be an excellent solution. It doesn't necessarily need to be a

complete replacement, but it can be used in conjunction with existing physical setups to promote

active learning.

Flexible learning provides opportunities to learn class material when the students or

instructor might be having challenges in terms of establishing meeting times, scheduling or

location [7]. For instance, if an individual is a part-time student with the obligations of a full-

time student, that person might have challenges meeting lab times scheduled during regular work

hours. The VCL is a most valuable tool to accommodate those circumstances. From the comfort

of home and in a more appropriate time, that individual can complete the exercises/material if the

VCL is utilized. The VCL is flexible with respect to the schedule and logistics limitations of an

individual.

The VCL also supports learning via discovery mode. In this scenario, an environment is

provided where the student has minimal supervision or instruction [8]. The student is

encouraged to learn by making changes and observing the impact of these changes. There are

some significant challenges in implementing such a setup in the alternative scenario of a physical

laboratory. Due to safety and cost considerations, facilities that promote discovery learning are

few. A VCL designed for the purpose, again, proves to be an excellent resource to implement

safe and cost-effective learning via discovery.

There are various implementations of virtual and remote labs reported in the literature.

There are several World Wide Web based labs which foster learning by different modes [3, 9-









11]. Most of these virtual labs, however, have a few shortcomings in terms of their usage. Many

do not provide a sufficiently high level of interactivity with the user. Significant modifications

need to be made to the program to implement any changes. Another disadvantage that most of

the current virtual labs have is that they are developed on a proprietary software platform. At

times significant familiarity with that software is needed to be able to utilize the application.

2.2 Objective

The intent is to build a VCL module that treats some advanced-level control concepts and

serves as a pedagogical tool that overcomes shortcomings that existing virtual labs pose from a

learning perspective and user interface. The infrastructure created by Peek et al. is used for

implementing an Inverted Pendulum VCL [12]. The intent is to develop an animated control

module that reinforces advanced control concepts with a friendly user interface. Some examples

of key control concepts illustrated in the VCL are linear state-space modeling, controllability,

pole placement and observability analysis. Sections 2.3 and 2.4 discuss the Inverted Pendulum

system, its dynamics and the associated control concepts. Section 2.5 describes the

implementation of the Inverted Pendulum system as a VCL using the LabVIEW software and its

animation features [13]. Finally, conclusions from this effort are summarized in Section 2.6.

2.3 Inverted Pendulum

The inverted pendulum considered consists of a spherical bob attached to a cart by a rod.

A schematic diagram is given in Figure 2.1. The mass of the rod is assumed to be negligible.

The rod is mounted by a hinge at the center of the cart. The input of the system is a horizontal

force applied to the cart. The cart is free to move only along one coordinate which is the

horizontal z-axis. The pendulum is free to rotate 360 degrees with respect to the cart in the x-z

plane where the x-axis is vertical. It is assumed that there is no friction between the pendulum

and the cart at the hinge.









The goal is to keep the pendulum in an upright position by manipulating the value of the

applied force. The system is inherently nonlinear. To apply linear control theory, the dynamics

must be linearized, and represented as a standard state-space realization. Consequently, the

system is linearized for small values of the angle that the pendulum makes with the vertical.

The nonlinear equations describing the dynamics of the system are

-M I s- +02/sin -gsin0cos0) (2-1)
-+sin2 0
m
1 f m+M
and 0= 1 _cos 02/cosOsinmO+ gsino (2-2)
M sin2 0 m m
m

where z = dz/dt (2-3 a)

and = dO/dt (2-3b)

where Mis the mass of the cart, m is the mass of the pendulum bob, is the length of the

pendulum rod, g is the acceleration due to gravity, z is the horizontal position of the cart, 0 is the

angle that the pendulum makes with the vertical, andfis the force (control input) acting on the

cart [14].

A linear state-space system is derived from Eqs. 2-1 2-3 by linearizing about an

operating point ( z, z, 0, ), where

z = 0 (2-4)

S= 0 (2-5)

= 0 (2-6)

0 =0 (2-7)

f =0 (2-8)









The deviation variables for the linear state-space model are

X1= ---Z--
x, = -

x2 = z-

x = 0 -


x4 =0-0




After linearization about the point

(z, z, ,, f) = (0, ,0, 0,0)

the resulting standard linear state-space model

= Ax + bu

is given by the equation


0 1 0

0 0 -mg
= M
0 0 0

0 0 (m+M)g
Ml

where the elements of the state vector x are distance (xl

and angular velocity (x4 = ), and where


0

-mg
M
0

(m + M)g
Ml


0 0

0 1
x+ M u (2-16)
1 0

0 -1

Ml, velocity ( 3 ),
z), velocity (x2 z ), angle (x3 0),


(2-17)


(2-9)

(2-10)

(2-11)


(2-12)

(2-13)


(2-14)


(2-15)










0
1
b= M (2-18)
0
-1
MI

The control is u, which is the force acting on the cart. The standard output model

y=Cx (2-19)

relates the output y to matrix C and state vector x, where the output matrix C is the identity

matrix

1000
0100
C = (2-20a)
0 10
0 001

in the case where all four states are measured. For the case where only one state is measurable,

the output matrix C adopts a row-vector form of all zeros, except for one entry that is unity at the

location corresponding to the measured state. For the particular case where the state xj, namely

the distance of the cart from its original horizontal position, is the only measured state, the output

matrix C adopts the form

C=[l 0 0 0] (2-20b)

2.4 Control Design

When the system is in the unforced configuration, a stability check done by calculating the

eigenvalues of matrix A reveals that there is one eigenvalue that lies in the open right half plane,

implying that the system in its unforced state is unstable.









This is a regulation problem, as the objective is to make the states evolve towards zero

value. For the implementation of the controller, a test for controllability needs to be performed

to verify that the system is indeed controllable. The requirement for controllability is that

det(Q,) 0 (2-21)

where Qc= b Ab A2b A'b] (2-22)

Using the definitions for A given in Eq. 2-5 and for b given in Eq. 2-6, it follows that


1 mg
M M212
mg
0 0
Ab= A2b= Ml ,A3b= -( M) (2-23)
-1 (m +M) g
0
MI Mz/22
0 (m + M)g 0
M2/2


01 0mg
M M2/2
1 0mg

0 1 (m + M)g
Hence, Q, M -1 M21 MM)g (2-24)
0 0
MI Mz/22
-1 (m + M)g
0 0
MI Mz/22

Obviously matrix Qc in Eq. 2-11 is of full rank, thus

det(Q,) 0 (2-25)

which implies that the system Eq. 2-4 is indeed controllable. The analysis of controllability

presented here is found in standard references [32-34].

Two scenarios are considered:

1. All states are measured.

2. Some states are not measured.









In the case of the first scenario in which all states are measured, a full-state feedback

approach is used in the form of the proportional state feedback control law

u = -Fx (2-26)

where F is a proportional gain used to address the regulation problem.

To determine an appropriate value for matrix F, first substitute the value ofu given by Eq.

(2-26) into the state space equation Eq. (2-15) leading to

x = Ax + b(-Fx) (2-27)

The standard solution to Eq. 2-14 is given by the Variation of Parameters formula as

x= x0e(A-BF)t (2-28)

where x0 is the vector of the initial value of the state vector x [36]. Ackerman's pole placement

algorithm is employed for computing the value of the matrix F that places the poles of the A-BF

system in the desired location [36].

In the case of the second scenario, in which all states of the system are not measurabed, a

Luenberger Observer is incorporated in the controller to estimate the value of the states. Before

carrying out an observer design, a check is performed to verify if the system is observable when

only one state is measured. The first state, the distance of the cart from the original position, is

the only state that is assumed to be measurable. For the system to be observable, the condition

det(Q) 0 (2-29)

should be satisfied, where the observability matrix Qo is given by

C

Q = CA (2-30)
CA2
MCA

Matrix A is defined by Eq. (2-17) and matrix C by Eq. 2-20b. The expressions









CA=[0 1 0 0] (2-31)


CA2 = 0 0-mg 0 (2-32)


and CA3 = O 0 0 0 mg (2-33)
M

can be used to readily build the observability matrix Qo described by Eq. 2-17, yielding


1 0 0 0

0 1 0 0
Q, = (2-34)
mg
0 0 --m 0
M
0 0 0 mg
M

Since matrix Qo is diagonal, its determinant is simply the product of the diagonal terms.

Hence,

2 2
det(Qo)= mg 0 (2-35)


Given that the determinant is nonzero, it follows that Qo is of full rank, and therefore the

system is observable. The analysis of observability presented here is also found in standard

references [34-36].

The standard equations for observer design

x = Ax + bu (2-36)


= Ai +bu + L(y -) (2-37)

produce estimated states i and estimated outputs y = Ci as a function of the measured system

output y = Cx, and the Luenberger gain L, in Eq. 2-37. The control input









u = -Fi (2-38)

is used to place the poles of Eq. 2-36 at the desired locations. The error

= x i (2-39)

is defined as the difference between the actual values of state vector x and estimated values of

state vector i. Hence, the derivative of the error

E=x-x (2-40)

is computed by differentiating Eq. 2-39. Substituting Eq. 2-36 and Eq. 2-37 in Eq. 2-40 yields

S= (A LC)E (2-41)

Invoking now the Variation of Parameters formula the solution to the differential equation

Eq. 2-41 is

= Ee (A-LC)t (2-42)

where e0 = x(0) i(0) is the initial value of the error E, i(0) is an initial guess of the value of

the estimated state vector, and x(0) is the initial value of the state vector.

The poles of matrix A-LC should lie on the open half plane for the value of error E to

evolve to zero. The poles are placed at the desired location by an appropriate choice of L, which

for low-order systems as the one considered here can be easily done via Ackerman's Pole

Placement algorithm.

Linear quadratic regulator control design. A Linear quadratic control regulator (LQR)

control design is implemented in the VCL. The LQR control law

u = -Kx (2-43)

is obtained by minimizing the cost function


J(u)= (xQ x + uTRu + 2xTNu)dt (2-44)
0









with respect to u. The weighting matrix Q must be symmetric positive semi-definite, and R

symmetric positive definite. The weighting function N is specified to be zero.

For a linear state space system

x = Ax + Bu (2-45)

the solution to the minimization of the cost function results in the steady-state Riccatti equation

[36]

ATS + SA (SB + N)R1 (BTS + NT) + Q = 0 (2-46)

The acceptable solution to Eq. 2-46 is a positive definite matrix S which is then used to specify

K from the expression

K= R-(BTS+NT) (2-47)

Since in this case N = 0, therefore

K= R1BTS (2-48)

2.5 Realization of an Inverted-Pendulum VCL

The LabVIEW software and a VCL infrastructure proposed by Peek et al., is used for the

implementation of the Inverted Pendulum VCL [12, 13]. Previous software-based control-tools

for the inverted-pendulum system reported in the literature have significant value, but the VCL

developed in this study has a number of additional desirable pedagogical features [14]. Initially,

stand-alone VIs and subVIs are generated using LabVIEW software for different components of

the design before integrating them as a part of a monolithic VCL. There are several reasons why

National Instruments' LabVIEW software is used for constructing the VCL. The ease of

structuring and maintaining a VCL is significantly high in this software. The LabVIEW

software has built-in features for deploying applications on the web. The software also has

toolkits specifically designed for control engineering. Implementation of a VCL using









LabVIEW does not rely on support from other software packages, as would be the case if some

higher-level language is used to implement the same features found in VCL.

Figure 2-2 shows the front panel of VCL as it appears to a student user. The key elements

are the Animation and Interaction Panels, respectively, located on the top and bottom-left areas

of the front panel. These two are very critical components of the VCL as the user makes most

modifications in the plant and controller setup in the Interaction panel and instantaneously

observes an animated result describing the plant and states in the Animation Panel. The

Animation Panel has a two-dimensional graphic representation of an inverted pendulum. When

the VCL is operated, the animated cart responds to the control input by moving to the left or

right and causing a pendulum swing. The third panel is the Navigation Panel on the bottom-right

area of the front panel. The Navigation Panel has five tabs (Information, Plant, Controller,

Analysis and Simulation), which provide various pieces of information about the VCL. More

information is given about these three panels in the ensuing subsections.

Animation, interaction and navigation panel. The animation panel plays the role of

providing a visual representation of the plant, namely an inverted pendulum. Any changes that

are made to the inverted pendulum mounted on the cart are visually depicted in the Animation

Panel.

The user has the ability to make changes to the plant and controller in the Interaction

Panel. The user can adjust plant parameters, initial condition of the states of the inverted

pendulum and assign the different values to control parameters to the controller of choice. The

user also has the ability to run the plant in Manual or in Auto mode. Figure 2-3 depicts some of

the various modes that the user can configure parameters in the Interaction Panel.









The user has to specify the initial states of the pendulum (position, velocity, angle of the

pendulum with the vertical and the angular velocity of the pendulum). The Animation Panel

constructs the visual representation of the inverted pendulum based on the information that the

user provides. The user has the flexibility of running the VCL in the following two scenarios:

(1) all states are measured, or (2) only one state is measured. Based on the choice of the user, an

implementation of the corresponding controller is given. When "Manual F" Control is in the

"Off' position, the user is allowed to choose the poles for the closed loop matrix A-BF and the

value of matrix F is calculated from Ackerman's pole placement algorithm. The user can

immediately see the impact of poles chosen on the stability of the inverted pendulum in the

Navigation Panel under the Analysis tab. When the "Manual F" Control is in the"On" position,

the user has the ability to choose the values of the elements of matrix F. When the controller is

operated in "Luenberger Observer" mode, as shown in Figure 2-, the user has to provide the

desired poles for matrix A-LC. The only state that can be measured in this mode is the position

of the cart. When the controller is in "Off' mode, i.e., the system is in open loop configuration

with no feedback, the Analysis tab of the Navigation Panel shows that the pendulum is in an

unstable configuration, which is ascertained by the fact that there is one eigenvalue of the system

in the open right half plane. The Interaction and Animation panels provide a suite of options and

visual representation for the user.

The Navigation panel is located on the bottom-right area of the front panel of the VCL.

The Navigation panel has five tabs entitled: (1) Help, (2) Plant, (3) Controller, (4) Analysis, and

(5) Simulation. These tabs provide pertinent and critical information about the VCL to the user.

The Help tab, when clicked on, provides general information about the operation of the VCL.

The user can access the Help tab without having to leave the VCL. The Help tab displays an









embedded PDF file. The Plant and Controller tabs are also embedded PDF files which provide

information about the dynamics of the plant (inverted pendulum) and the controller (proportional

state feedback and Luenberger Observer). The nonlinear equations and linear state space system

for the inverted pendulum are explained in the Plant tab of the Navigation Panel. Figure 2-3

shows the Plant tab of the Navigation Panel. The Controller tab provides information about pole

placement and various other aspects of control design for inverted pendulum VCL, as shown in

Figure 2-4.

The Analysis tab, shown in Figure 2-5, has information about the tools and graphs that are

used in control theory. This tab has information about the transfer function, location of poles and

zeros in complex plane and Bode plot (frequency response). The Simulation tab, shown in

Figure 2-6, shows plots of the results of numerical simulations describing the states (position,

velocity, angle and angular velocity) and inputs as a function of time. Since this is a regulation

problem, when stable choices of eigenvalues are given, using the linear state space as the

dynamic model, all states converge to the value zero, regardless of the choice of initial state

vector. The Simulation tab provides the real time curves of all the states as a function of time.

The Runge-Kutta integration algorithm is employed to compute the numerical response of the

plant to the controlling input. When the controller is toggled between "On" and "Off" modes

(closed-loop and open-loop behavior, respectively) in the Interaction panel by the user, the

impact of that change on the value of states is depicted immediately in the Simulation tab.

The LQR control strategy, as shown in Figure 2-, employed in the VCL gives the user the

opportunity to implement different control choices, such as varying the weighting on different

elements of the cost function and displaying the corresponding LQR gain.









2.6 Conclusions

A VCL for the control of an Inverted Pendulum is described. The Inverted Pendulum is a

classic example of illustrating state-space model representations and demonstrating the classical

control concepts of controllability and observability. The animation features of the VCL provide

a visual description of how an inverted pendulum responds as a function of the input force

applied. The user is given the opportunity to run the VCL in different modes, such as in open-

loop and closed-loop configurations of the system.

The VCL can be utilized as a tool for enhancing learning. The three most widely

recognized learning modes (active learning, flexible learning, and learning via discovery) can be

easily executed using this VCL module. The module can be used in conjunction with a process

control lecture for demonstrating various concepts. The animation capabilities allows the user to

see the impact of every change that is made to the control configuration. The Analysis tab also

demonstrates that the open-loop configuration (unforced system) is unstable, as one of the

eigenvalues is in open right half plane. The user is given the choice of choosing the poles for the

system and noticing its impact on the plant. The user has the choice of running the VCL in two

modes: (1) all states measured or (2) only one state is measured and implementation of a

Luenberger Observer. This utilization of the VCL supports active learning of the control

material. The VCL can also be presented before or after a conventional lecture. In the first case,

the users' motivation to learn about theory presented in class would be enhanced as they have the

opportunity to develop and experience with the VCL. In the later case, after the lecture is

delivered, an interaction with the VCL would serve as an excellent tool for reinforcing the

concepts taught in class. There is significant value in learning an abstract concept taught in

lecture and being able to relate to the concept by virtue of seeing the animation and graphs from









simulation. In an ideal scenario, the VCL should be used in all three modes (before, during, and

after lectures).

The tool can also be used to support group activities in a class for learning control theory

and doing control homework [15]. The modular feature of the VCL can be taken advantage of

by implementing sequential learning. In this mode, different versions of same VCL are

progressively given to the user as the class progresses. Each successive version makes more

features of the controller available to the user, hence helping students appreciate and learn faster

the concepts as they are progressively taught in the class.

To validate the benefits and effectiveness, it is proposed that the VCL should be used as a

pilot teaching tool in process control classes. The feedback obtained from the students would be

highly beneficial in optimizing and incorporating features that could enhance the learning

experience.













x
z


Figure 2-1. Inverted pendulum.


()


()

















I- I-


in iMenaM iUain

Controller. Of/On
LuMnbergerGin
Ir T J7. Tim
r- ijrmrrr


I-
~I-


A
fl-;..


Figure 2-2a. Front panel of the VCL where all states are measured.


I'....' I


~--tt


;,i o. na il L ,3 LLL I A
.jn ]l L lij. Ti
I,, ,, ,, ,
II i~
II-rr,























c..br.iie. I


r r .


Figure 2-2b. Front panel of the VCL showing observer.


I LpC'rT |


.11 III


I I









































r7 r0
S0 17 0
0-o .-j0 0
...... .. ........... ........ ..............


Figure 2-3. Interaction panel of inverted pendulum VCL.













Help Plant Controller Analysis Simulation


TransFer Function


Static Stik Feedhnck Controller and Liibenhr -

Static state 'Le,;LjL. controller is used for Inverted Mo.'[ei IIllI
the pendulum in vertical upright position.

The Linear State Space Model is obtained by linearization of the
equations that described ti dIn.lmicni (f the inverted pendulum:


y= ,+ .L.Sinf-g,, r,,in





Rm J

Th" Lincar State Space Model is:


f i 1 1 r
( >I I l
< I >


Pole-Zero Map Plot
1-
0.75--
0.5-
g 0.25--
0-
E
-0.25--
-0.5--
-0.75--
-1 -0.
-1 -0,8


-till lii] II


-0.6 -0.4 -0.2 0
Time


Figure 2-4. Controller tab of the navigation panel.
















Bode Magnitude
0-
-25 -
2 -50--
( -75 -
-o
S-100-
S-125 -
-150 -
-175-
10Ou


Bode Phase
50

C


-50

-100
1 rn


]-


III ____ I ____ ] ____ I ____ I ____


lm 10m 100m 1


Frequency (rad/s)


Step Response Graph
2-1 I


U

-2
U-- 4--



-4-

. - -


0 10 -k 25
10 100 1k 0 2.5 5


Pole-Zero Map Plot


'Ft __


- JU

-200 1
lOOu im


I I I I
10m 100m 1 10
Frequency (rad/s)


ID


100 1k <


Transfer Function


0.0076 s2 + 3.3952E-18 s 0.075
s4+2., s3 + 1.6653 s2 + 0.575 s- 0.075


0.
0.
0.
0.


-0.
-0.
-0.
-0,


7.5 10 12.5 15 17.5 20
Time (s)


22.5 2
22.5 2


6
4
2- -
0- --- --- -- -.... I---- --------


4-
6-


1-


-0.8 -0.6


Input:

Output:


u -I

Uiil:!a *


Figure 2-5. ls tb of te n p l


Figure 2-5. Analysis tab of the navigation panel.


-02
-0.2 0


-












Help I Plant Controller I Analysis I Simulation I


\. /


1.5
1.5


2 2,5 3 3,5 44.1
Time


11:11:1 : AI


Figure 2-6. Simulation tab of the navigation panel.


-i..i-, 51


1 1: 1 I 1:11:1
,J ,i: _:, ZO:_ :1 :
,JTh- r .*,c _,)
________ ______ _


E w pi:'l

Run









CHAPTER 3
PI AND PI2 CONTROLLER TUNING FOR TRACKING THE SLOPE OF A RAMP

3.1 Introduction and Background

In certain applications tracking the slope of the ramp is more critical than tracking the

ramp itself. It is not unusual to encounter applications where the set point is the slope of a ramp

trajectory. For instance, while growing thin films on a substrate, it is desired that the

temperature of the substrate in the reactor increases at a steady rate (i.e., following a trajectory

with a specified slope). In these kinds of applications, it becomes crucial to adopt the

appropriate choice of controller type with effective values for the tuning parameters and an

appropriate metric to ensure adequate performance. Simplicity of tuning relationships plays a

critical role in the implementation of a controller. Most successful tuning relationships have

been developed via simulation [16, 17]. The cost function normally used involves the feedback

error, which is the difference between the set point and output of the plant.

A proportional-only controller leads to a steady state offset with respect to step changes in

set point. An integrator needs to be incorporated in the controller to eliminate the offset.

Similarly, for a ramp set point, a proportional-integral (PI) controller is not sufficient to remove

the steady state offset. In this case a controller with two integrators, that is a proportional

double-integral (PI2) controller, is needed to remove the offset [18]. However, a PI controller is

sufficient to track the slope of the ramp trajectory. Belanger and Luyben proposed a

proportional-integral-double integral controller relating the tuning parameters to the ultimate

gain and ultimate period of the plant [19]. Alvarez-Ramirez et al. extended the work of Belanger

and Luyben and coined the acronym PI2 [20]. Peek examined three different versions of PI2

controllers for tracking ramp set point [21].









This study investigates the problem of controller tuning for tracking the slope of a ramp

signal as depicted in Figure 3-1. The intent is to design a controller that leads the output

trajectory to follow a line parallel to the ramp. It is also critical to minimize transients. The

control performance is deemed as poor when the system experiences large deviations during the

transients.

To investigate the problem, a first order system with time delay is adopted as the plant

model. Since the set point is a ramp, integral action needs to be incorporated in the control

design. Two kinds of controllers are used for this study, namely a proportional-integral (PI)

controller and proportional double-integral (PI2) controller. Both controllers serve the purpose of

placing the system output on the desired line, parallel to the ramp set point. However, there is an

offset with respect to the original ramp trajectory when PI controllers are used because only one

integrator is included in the control scheme.

To identify the appropriate indicator to measure the performance of the controller with

respect to the system and objective in question, three different metrics are used: integral of the

time-weighted absolute error (ITAE), integral of the absolute error (IAE), and integral of the

square of error (ISE). Each of the controllers is tuned to minimize the metric value for a given

set of plant parameters. The Simplex optimization routine is used to acquire tuning parameters

via the minimization of the metric adopted. The MATLAB platform in conjunction with the

SIMULINK module is used for conducting the simulations. A set of optimal tuning relationships

for controllers to track the slope of a ramp for 20,000 different plants are presented. The

performance of all the controllers is evaluated and results are compared with the prior work of

Belanger and Luyben. The performance of the controllers is also compared with that of those

proposed by Peek.










3.2 Problem Statement and Approach

The objective is to use PI and PI2 controller to make a first-order plant with time delay

follow the slope of a ramp. The plant is represented by the transfer function


G,(s)= K e (3-1)
TS+l

where K is the gain, r is the time constant, and 0 is the time delay of the plant. The input to the

plant is denoted as u and the output asy. The transfer function representation of the PI controller




Ge(s)= K 1+- (3-2)
i's

and the transfer function for the PI2 controller investigated here is


G,(s)= KK 1+- (3-3)
7-s

where K, is the proportional gain, and r, is the integral-action time constant. The plant and the

controller are configured in the closed-loop arrangement shown in Figure 3-2. The ramp

function serving as the set point in Figure 3-2 is given by

r(t) = at (3-4)

where t is the time and the constant slope a is taken as a = 1.

Three error metrics are used for tuning the controllers, namely, the integral of the time-

weighted absolute error (ITAE), the integral of the absolute error (IAE), and the integral of the

square of error (ISE), respectively defined by the integral equations

ITAE = lim t(3-5)
ITAE = lim \t r(t) -y(t) dt (3-5)
tf -> 0
0











IAE = lim f r(t)- y(t) dt (3-6)
tf ->CO
0

t!
ISE = lim f(r(t)- y(t))2 dt (3-7)
tf ->C 0
0

where y(t) is the plant output and tf is the extent of time over which the metric is computed.


When the closed loop uses a PI controller, the output of the system exhibits a steady state

offset es with respect to the original ramp characterized through the analytical expression


e = r (3-8)
KK,


the derivation of which is given in the APPENDIX using Final Value Theorem. A modified or

auxiliary ramp ra (t) is defined via the relationship


ra (t) = r(t) e, (3-9)


Figure 3-1 shows the ramp and the auxiliary ramp trajectories.

The corresponding auxiliary error metrics for PI controller, modified from Eq. 3-4, Eq. 3-5

and Eq. 3-6 are given by

tf
ITAE= lim it r (t) y(t) dt (3-10)
tf -*co 0


tf
AE = lim f (t) y(t) dt (3-11)
tf -,co 0
0

tI
ISE lim (r(t) -y(t))2dt (3-12)
tf -"*o 0


where the ITAE, is the auxiliary integral of the time-weighted absolute error, the AE, is the


auxiliary integral of the absolute error, and the ISE, is the integral of the square of error. It is to









be noted that in the case of PI2 controllers, the value of e, is zero i.e., r, (t) = r(t). Hence, the

ITAE, IAE and ISE error metrics for PI2 retain the original form of Eq. 3-5, Eq. 3-6, and Eq. 3-7,

respectively, even when all computations are carried out using their auxiliary counterparts.

The ITAE expression (Eq. 3-5) has been a popular choice for control parameter

optimization, as it assigns less weight to errors occurring in the initial times and more weight to

error at longer times. Error is defined as the closed-loop feedback difference between the

auxiliary set point and the output. This is traditionally a useful measure to adopt, as for a step

response it is inevitable that there is a relatively large error during initial times, which needs to

be given less significance compared to the error that is encountered at later times. The ITAE

may not be the best metric for the problem in question, however. The desired trajectory of the

output is the one which tracks the slope of the ramp without abrupt deviations in trajectory. The

ITAE is forgiving of aggressive output values at initial times as it gives less significance to error

at early times. Sometimes, it is desired to adopt a metric which gives an equal importance to

errors occurring at initial times as well. From that perspective, the two other metrics are also

used in this work for optimization purposes, namely, the IAE (Eq. 3-6) and the ISE (Eq. 3-7).

A routine in MATLAB is written to simulate the plant output for a given controller with its

parameters specified [22]. The simplex optimization routine is employed for tuning the control

parameters. For a fixed value of plant parameters (gain, time constant and time delay), the

parameters of the controller (proportional gain and integral action) are altered to minimize the

(ITAE, IAE or ISE) value of the cost function. For a PI controller, where there is offset with

respect to the ramp, an auxiliary ramp parallel to original one is constructed. The steady state

offset is calculated analytically using Eq. 3-8.









The three error metrics (ITAE, IAE and ISE) are minimized using the new ramp. These

metrics are defined over an infinitely long time; however, for practical purposes the final time is

chosen to be finite and defined by the formula

tf =15 max(r,0) (3-13)

The rationale behind this choice is that by this extent of final time, any reasonably performing

controller should make the value of error significantly small. The optimized tuning parameters

are non-dimensionalized by combining with plant parameters, and plots of optimized tuning

parameters were constructed. Time responses are constructed with the optimized values to verify

the responses of the process with the controlling action incorporated.

Peek analyzes the performance of three different configurations of PI2 controllers [21].

That study concludes that there is no significant difference in the performance of the three

configurations. The transfer function for PI2 controller given in Eq. 3-2 is used in this study as it

is the easiest to tune because it involves only two parameters, namely the proportional gain K,

and the integral-action time constant r, .

3.3 Results and Discussion

3.3.1 Tuning Parameters of Controllers

Tuning parameters are calculated for both controllers PI and PI2 using the ITAE, IAE, and

ISE as the optimizing metric. The optimized controller parameters are nondimesionalised using

the plant gain and time constant [17]. The resulting plots of KKc versus 0/r, and r /r versus 0/r

are shown in Figures 3-3 and 3-4 for the PI2 and PI controllers. The tuning parameters obtained

using ITAE, IAE and ISE as the metric are shown in the first, second, and third row of each

figure, respectively. The plant parameters K, 0, and r are selected such that K and r range from

0.1 to 50. Twenty logarithmically equally-spaced points are considered for both K and r in their









specified ranges. After each time constant r is defined, the values of the delay parameter 0 is set

by defining the ratio of 0/r to range from 0.1 to 100, with 50 logarithmically equally-spaced

points inside the range. For a fixed value of 0/r ratio, the value of 0 is computed from the value

of r and of the fixed 0/r ratio. Hence, tuning parameters were obtained for 20,000 different

plants for each controller and for each metric.

The graphs in Figures 3-3 and 3-4 show that, in general, as the 0 r ratio increases, the value

of the optimal KKc product and of the Tr/r ratio decrease. The value of the KKc product

represents the proportional control action on the closed loop system, and it is expected to vary

inversely to the 0 r ratio. In other words, the control action will be higher for smaller values of

/ r ratio, and smaller as 0/r increases. This is qualitatively reflected in Figures 3-3 and 3-4 for all

three optimizing metrics and for the two controllers considered. On the same vein, the Tr/- ratio

is indicative of integral action for the system and it is expected to behave analogously to the

proportional control action. The integral action should be more aggressive for smaller values of

/ r ratio compared to higher values of the ratio. This is, indeed, observed from Figures 3-3 and

3-4.

Least-square fits for the optimized control parameters are given in Tables 3-1 and 3-2 for

the PI2 and PI controller, respectively. The least square fit relates the optimal KKc product with

the 0 r ratios and the optimal r/-r ratio with the 0/r ratio. If the KKc versus / r curve and/or Tr/r

versus 0/r curve is significantly nonlinear, a break point at a certain value of 0/r ratio is identified

and two least-square fits are presented for the same curve, one above and one below the 0 r ratio

breakpoint value. The results are for 0/r values ranging between 10-1 and 102 only.

3.3.2 Comparison of the Performance of the PI and PI2 Controllers

The performance of the PI and PI2 controllers is characterized for three different plants

(plant parameters given in Table 3-3) using the optimal parameters prescribed by each of the









three optimizing metrics. The time response curves for each plant are constructed, for both the

controllers and the tuning parameters prescribed by the respective optimizing metric, to assess

the time-domain performance of each tuning prescription. The proportional gain of the three

plants was taken to be the same value, namely K = 1.0. Three different values of the 0 r ratio are

taken from the domain of the values for 20,000 different plants. The ratio values of 0.1, 3.0 and

100 (minimum, middle and maximum of the range considered) are selected. Several

combinations of 0/r can satisfy each value of the ratio. The value of r is selected such that it

covers the domain of the different values of r selected for all the plants in this study. The values

of 0.1, 1.9 and 50 are selected for r. From these values of r the value of 0 is computed for each

value of the ratio.

Figures 3-5, 3-6 and 3-7, illustrate the time responses of the three plants for the ITAE, IAE

and ISE metrics, respectively, using PI2 and PI controllers. Each figure demonstrates the

performance of PI2 versus PI controller for the three plants. When a PI2 controller is used, the

output follows the original ramp, whereas with a PI controller an offset is introduced in relation

to original ramp and the output follows the auxiliary ramp parallel to the original ramp. The

original and auxiliary ramp are plotted in each time response curve as well. With increasing

values of the 0 r ratio, the offset between the original and auxiliary ramp increases. This is

expected as the offset is directly proportional to r, and inversely proportional to the product of

KKc as shown by Eq. 3-8. With increasing value of 0/r, r, increases and the product ofKKc

decreases, hence the offset increases. It is also observed that it takes longer for steady state to be

reached with increasing values of the 0/r ratio.

The output of each plant tracks the slope of the original ramp for each controller and every

optimizing metric. As discussed earlier, there is offset with respect to the original ramp when PI









controller is used and there is no offset when PI2 controller is used. Regardless of the fact of

whether offset is introduced or not, as long as the output is parallel to the original ramp, the

control objective is satisfied. The better performing controller is the one that reaches faster the

original or auxiliary ramp, depending on the controller adopted, and with minimal transient

values. It is observed that for all the three plants and optimizing metrics considered, PI

controller performs better than PI2 controller. Output of each of the plant, on using PI controller,

tracks the auxiliary ramp much faster compared to the output when PI2 controller is used. It is

observed that the plant output is more oscillatory during transient time for ISE prescribed tuning

compared to those obtained from the ITAE and IAE criteria. This is an expected result as in the

ISE metric the square of the error is used. Even though both PI2 and PI controllers exhibit

oscillatory behavior, the phenomenon is more prominent in the case of PI2 controller. From

these observations, it is concluded that the PI controllers are better performing than PI2,

regardless of the optimizing metric adopted to tune, for the three plants considered, and provided

that the unavoidable resulting offset is acceptable to the user.

3.3.3 Comparison of Metrics

After it is established that PI is a better performing controller, the next step is to identify

the optimizing metric with which the PI controller gives the best performance. Time responses

for the same three plants are constructed using the optimizing metrics. Figures 3-8 and 3-9

shows the time responses for the three metrics. For the sake of comparison, even though it is

established that the PI is a better performing controller, time response curves are generated for

the PI2 controller as well. Figure 3-8 shows the PI controller time responses and Figure 3-9

shows the PI controller time responses, for the three plants. Note that the offset introduced while

using PI controller, is a function of the tuning parameters Kc and rt. Since the values of these

parameters are different for each metric, the output using a PI controller follows a different









auxiliary ramp, depending on which optimizing metric was used. All the auxiliary ramps,

however, are parallel to the original ramp.

It is observed that ISE is the worst optimizing metric, as the output of the plant is most

oscillatory and has larger deviation from the auxiliary ramp as shown in Figure 3-8 and 3-9.

Also, it takes longer in the case of the ISE to reach steady state. The other two metrics, ITAE

and IAE, are quite close in their performance. The time response curves suggest that for tuning

purposes the ITAE is a better metric for the PI controller and the IAE is better metric for the PI2

controller.

For the PI controller, the IAE demonstrates more oscillatory output compared to the ITAE

metric. Also, the output of the plants reaches steady state sooner when the ITAE is used as the

optimizing metric compared to when the IAE is adopted.

3.3.4 Comparison of PI (ITAE) Controller with Literature Precedents

After determining that the PI controller using the ITAE as the optimizing metric exhibits

highly desirable performance, the next step is to compare its performance with controllers

proposed in the literature. Peek suggests a PI2 controller with ITAE as the optimizing metric for

tracking a ramp set point [21]. Though the objective is slightly different than the one in this

study, the controller recommended by Peek does satisfy the control objective of this work [21].

Belanger and Luyben recommend a double integrator controller for tracking a ramp, which

satisfies the control objective of this study as well [19].

Figure 3-10 shows the performance of the three different controllers for the three plants

considered. It is observed that the PI controllers tuned using the ITAE metric gives the best

results followed by the controllers recommended by Peek, and then by those proposed by

Belanger and Luyben. As the 0/r ratio increases for the plants, it becomes increasingly obvious

that the PI controller gives the best results. For the plant with the ratio 0/r = 100, the highest









value, the PI controller makes the output reach steady state much faster and with least transients

compared to the other two controllers.

3.3.5 Local Minima versus Global Minima

Figure 3-11 shows the contour plots for the ITAE metric using PI controller for the three

plants. It is observed from Figure 3-11 (A) that there is more than one minima for the ITAE

metric and different values of control parameters Kc and r, for each minima. The goal is to

obtain global minima for the optimizing metric and the corresponding optimal control

parameters. If local minima is reached as opposed to global minima, that would potentially lead

to scatter in the optimal control parameters curves as shown in Figures 3-3 and 3-4. Hence, the

optimization work described here is conducted with care to avoid local minima results. The

measure adopted consists of utilizing different initial guesses for each optimization routine

execution, leading to the identification of different local minima, when they exist. The smallest

of such minima is then accepted as the best approximation to the global minimum. Although this

approach is neither rigorous nor exhaustive, it provides excellent practical results in the context

of this study.

3.4 Conclusions

Optimal tuning relationships for the PI and PI2 controllers using the optimizing metrics

ITAE, IAE and ISE are presented for 20,000 first-order plants with time delay. The plants

considered have the 0/r ratio value varying from 10-1 to 102. The validation and comparison of

the controllers performance is done by their deployment on three different plants.

On the basis of results obtained, the PI controllers using ITAE as the optimizing metric are

the best performing controllers for the purpose of tracking the slope of a ramp trajectory. They

perform better than controllers proposed in prior literature.

















r



1

I*- auxiliary ramp r.


Figure 3-1. Ramp r and auxiliary ramp r, with constant slope, a.


Figure 3-2. Closed loop transfer function representation of plant and controller.





















0-1 10 101 102
V0/

Ir:-- -----


0/r


101 102


0



101 10 101 1C
O/T


B
10 0


.............. ...................
+t

...,..-...-.-:-.i. +.9......-


O/r


D 100

o.. I _


- .. .
!. 9. "


0/r


100


" 10


10-2
101


4.-






101 102
O/r


Figure 3-3. The PI2 controllers optimal tuning parameters, using ITAE (A, B), IAE (C, D), and
ISE (E, F) as the optimizing metric.


10-1






















O/


0/r


0
1 0




10-1 10 101 1C
O/T


100 101
0/r


Figure 3-4. The PI controllers optimal tuning parameters, using ITAE (A, B), IAE (C, D), and
ISE (E, F) as the optimizing metric.


,-1


10-1


O/


ii


10-1


0/r









Table 3-1. The PI2
Criterion
ITAE


IAE


ISE


controllers optimized tuning parameters linear least square fit equations.
Optimal PI2 Parameters
1.26(O/r)-076 /r < 2.5
K 0.74(0/r)-004 /r >2.5
0{.48(/r)-050 0/T < 2.5
0 .74(0/r)-092 O/r >2.5


S1.26(0/r)-076
KK o0.78(0/r)-006

0.48(0/r)-050
T/7I 0.69(0/r)-089


KKc


,.-


1.58(/r)-076
0.94(0/r) 0o8
0.45(O/r)-052
0.65(0/r)0 88


<2.5
> 2.5
<2.5
> 2.5

<2.5
> 2.5
<2.5
> 2.5








Table 3-2. The PI controllers optimized tuning parameters linear least square fit equations.

Criterion Optimal PI Parameters
ITAE K 0.73(/r)-073 O/r 2.5
c 0.41(0/r)-006 O/r >2.5
0.82(/r) 013 0/r < 2.5
1.58(0/r) 088 O/T > 2.5

IAE 1.07(0/r)-00 O/r < 2.5
c 0.57(0/r)-008 O/r > 2.5
0.73(0/r) 026 O/r 2.5
1.42(O/r)088 /r > 2.5

ISE KK 1.40(/r)080 O/r < 2.5
C 0.74(/r) 008 O/r > 2.5
0 .75(0/r)036 O/r < 2.5
1.42(0/r)-088 0/r > 2.5



Table 3-3. Values of the plant parameters used compare the performance of the controllers.
Parameters Ratio
Plant K r 0 0/r
1 1 0.1 0.01 0.1
2 1 1.9 5.6 3.0
3 1 50 50000 100






































0 18 37
t
x 104
I '


56 75


0 1.9 3.8
t


7.5
x 104


Figure 3-5. The PI2 and PI controllers' ramp tracking and slope tracking performance using the
optimal ITAE control parameters.


56


3 37


18


0



7.5
C






































0 18 37
t
x 104
I 1


56 75


0 1.9 3.8
t


7.5
x 104


Figure 3-6. The PI2 and PI controllers' ramp tracking and slope tracking performance using the
optimal IAE control parameters.


56


3 37


18


0



7.5
C






































0 18 37
t
x 104
I 1


56 75


0 1.9 3.8
t


7.5
x 104


Figure 3-7. The PI2 and PI controllers' ramp tracking and slope tracking performance using the
optimal ISE control parameters.


56


3 37


18


0



7.5
C































37L


0 18 37 56 75
t
x 104
7.5
C ITAE
5 -6 IAE
5.6
ISE
. .3.8 R am p ...
S K=
1.9 = 50
7 = 5000
0
0 1.9 3.8 5.6 7.5
t x 104


Figure 3-8. The ITAE, IAE, and ISE metrics comparison for three plants, using the PI
controllers A), B), and C).


-ITAE .
--IAE ..
--- ISE
.. Ramp

K=
S=1.9
/ 0 = 5.6






































0 18 37
t
x 104
I 1


56 75


0 1.9 3.8
t


7.5
x 104


Figure 3-9. The ITAE, IAE, and ISE metrics comparison for three plants, using the PI2
controllers A), B), and C).


56


3 37


18


0



7.5
C

























0.4 0.8
t


x 104
7.5 F-


5.6 .- Pec
.. Rai
S3.8 Au:


1.9 ,

L./
0
0 1.9


3.8 5.6
t


Figure 3-10. The PI controllers tuned using the ITAE metric compared with Belanger and
Luyben and Peek's controllers for three plants A), B), and C).


1.2 1.5


7.5
x 104

















3.5





3
8


0.4031 --

0.403









0.4029
0.2908


0.3175





S0.3174





0.3174
5.0454


10
1/t
i


0.2908
1


5.0505
l/1


0.2908


5.0556
x 104


Figure 3-11. Contour plots for the PI controller tuned using the ITAE metric for the three plants
A), B), and C).









CHAPTER 4
GENERALIZED PREDICTIVE CONTROL FOR FUEL CELLS

4.1 Introduction

A generalized predictive control (GPC) strategy is designed and implemented for a

polymer electrolyte membrane (PEM) fuel cell system. It is vital for efficient performance of the

fuel cell to ensure the robust and precise control of the performance variable which is oxygen

excess ratio. A review of prior control strategies developed for this fuel cell system, is presented

and compared to the new GPC scheme proposed. The performance of the different controllers is

evaluated in the case where the model available for design suffers from uncertainty.

4.2 Fuel Cell System Background

Fuel cell are electrochemical devices that directly convert the chemical energy of gaseous

reactants to electrical energy. They are widely considered as an alternative to fossil fuels which

are limited in supply. For a typical fuel cell, water and heat are byproducts generated as a result

of operation. The reactants needed are hydrogen and oxygen. Both of these reactants are widely

available, and a proliferation of applications based on these fuels would tremendously reduce our

dependence on fossil fuels. This is an additional motivation for developing and engineering fuel

cell system as it is friendly to the environment. Fuel-cell based automobiles have no harmful

emissions, such as CO2 which combustion-engine automobiles contribute significantly to the

environment [23]. Fuel cells are an efficient and clean source of energy production.

William R. Grove discovered the principle of operation of fuel cells in 1839 [24]. From a

classical standpoint, a fuel cell is comprised of two electrodes with an electrolyte located

between them. The electrolyte has the special property that it allows only protons (positively

charged hydrogen atoms) to pass through it. In contrast, the membrane does not allow electrons

to pass through. Hydrogen gas passes over the anode electrode, and with the assistance of a









catalyst, breaks down into protons and electrons. The protons selectively pass through the

membrane to reach opposing cathode electrode. The membrane is an electronic insulator. The

membrane is comprised of fluorocarbon chain to which the sulfonic acid groups are attached.

On hydration of the membrane the hydrogen ions become mobile. The electrons flow through an

external circuit, creating a current flow. An oxygen flow passes over the cathode and combines

with the protons and electrons to generate water. The reaction at the anode is

2H2 -4H+ + 4e (4-1)

and at the cathode is

02 + 4H+ + 4e- 2H20 (4-2)

hence, the overall reaction is

2H2 + 02- 2H20 (4-3)

Several kinds of fuel cell designs have been developed and are currently being studied [25-

30]. A schematic diagram of an automotive fuel cell system including the structural

relationships among the input, outputs, and disturbance signals is given in Figure 4-1. A

compressor and pressurized hydrogen tank are used to provide the reactants oxygen and

hydrogen, respectively [31]. The compressor plays a crucial role as it ensures that the desired air

flow rate reaches the cathode based on power demands. The supply and return manifold models

based on thermodynamic consideration provide information about desired air flow rate needed.

A nonlinear curve fitting method is used to describe the compressor behavior [31]. The net

power delivered by the fuel cell system is the difference between the power generated and power

consumed to run the compressor motor to deliver a particular air flow rate. An excess amount of

air flow provided to the cathode is referred to as the oxygen excess ratio Al2 which is defined as

the ratio of the rate of oxygen supplied to rate of oxygen consumed. With increase in oxygen









excess ratio, resulting in high oxygen partial pressure, there is increase in power delivered.

However, it is at the cost of increased power consumption by the compressor to deliver a higher

air flow rate. Beyond a particular value of the oxygen excess ratio there is loss in net power as a

result of increased power consumption by the compressor. It has been shown in literature that

having oxygen excess ratio in the vicinity of 2, the fuel cell system delivers highest net power.

A humidifier, in fuel cell system, is used to add water to the reactants to avoid dehydration

of the membrane. A water separator is used to extract water from the air, leaving the fuel cell

stack, which is recycled back to the humidifier via the water tank. The voltage generated by the

fuel cell needs to be conditioned before it is fed to the traction motor. Appropriate usage of an

external battery with the fuel cell power supply helps in minimizing the transient responses and

delivering better system efficiency.

Polymer electrolyte membrane (PEM) fuel cells, also referred to as proton exchange

membrane fuel cells, are recognized particularly promising for utilization in automobiles as a

substitute for the internal combustion engine. This is because of the fact that the PEM fuel cells

have high power density, long cell life, low corrosion, and use a solid electrolyte. For effective

utilization of fuel cell technology, it is vital that that the fuel cell system is accurately

understood, monitored and that its process variables be held under tight control for various

operating conditions.

From a control engineering perspective the fuel cell can be divided into four subsystems,

namely (1) supply of the reactants air and hydrogen, (2) humidification of the reactants and of

the membrane, (3) heat management, and (4) power management. The model used in this study

assumes a perfect humidifier and coolers for the reactants and the membrane. The model

assumes a perfect power management system which controls the power drawn from the fuel cell









stack. A fast proportional controller is implemented on the hydrogen flow that tracks cathode

pressure [31]. This reduces the control problem to regulating the air supply, as the oxygen level

varies on the cathode side, due to varying power demands.

4.3 Objectives of the Research

The objectives of this study are the following:

1. Propose a systematic design solution for treating the of fuel-cell control problem of
regulating the oxygen excess ratio (performance variable) at a value of 2 by
synthesizing a GPC strategy.

2. Develop a systematic GPC design procedure for a fuel cell model in the scenario
where all states are measurable.

3. Develop a systematic GPC design procedure for the fuel cell model when selected
states are not measurable by incorporating an observer in the controller.

4. Evaluate the robustness of the GPC controller with respect to model uncertainty.

5. Rederive and correct, as needed, linear models obtained from nonlinear
formulations proposed in the prior literature for the fuel-cell model used for this
study [31].

6. Retune and redesign, as needed, all model-based controllers proposed in the prior
literature [31] to take into account the corrected linear models [31].

7. Compare by means of simulations the performance of the different controllers
proposed in prior literature [31] with the GPC strategy.

4.4 Fuel Cell Model

The fuel cell model proposed by Pukrushpan et al. is the basis for this study [31]. The state

equation for the model of a fuel cell system is of the form


x = f(x,u,w)

z = g(x,u, w)

y = h(x, u, w)


(4-4)

(4-5)

(4-6)


where x represents the states of the fuel cell system, u is the input, w is the disturbance, and z and

y are the outputs. The state vector is given by the expression









x=[ 0m2 mH2 2 0c Psm sm mw,n rm ]T (4-7)

where the vector elements are the mass of oxygen m02 in cathode volume, the mass of hydrogen

mH2 inside the anode volume, the mass of nitrogen m,2 in cathode volume, the rotational speed

of the compressor o)c the pressure of the supply manifold ps,, the mass of air in the supply

manifold mm, the mass of water at anode m ,a, and the pressure of the return manifold prm.

The outputs used as the performance variables are organized in the vector

z = [eP, A02 ] (4-8)

where e = P"" pdel is the difference between the desired power P"' and the actual power

delivered pd", and A02 is the oxygen excess ratio. The control objective in this study is confined

to regulating the value of A2 at the desired value Aet = 2 [31]. Three additional measured

outputs are organized in the vector

y=[WcP psm Vst ] (4-9)

where the vector elements are the mass flow rate of the air from the compressor Wc, the

pressure of the supply manifold ps and the voltage of the stack v,,.

The control input is given by

U = Vcm (4-10)

where vcm is the voltage signal sent to the compressor which in turn delivers a corresponding

flow rate of air to the cathode. Finally, the disturbance

w= It (4-11)

is the stack current I,t.









The nonlinear model described in Eqs. 4-4 4-6 and shown in Figure 4-2a is linearized

about a nominal operating point at which the fuel cell model generates a net power of

P"d =40kW and sustains an oxygen excess ratio o2 = 2. These conditions are realized when


I,, = 191 A and v, = 164 V. The corresponding values of the states are

x =[ moi2 H2 W cp m sn m iwan Prm (4-
[2.0e 3 5.6e 4 1.3e 2 8.2e3 2.2e5 3.8e 2 1.le -3 1.8e5]

The physical units of the states x are

[kgs kgs kgs rad/s pascal kgs kgs pascal] (4-11b)

respectively. Hence the nominal operating point is given by the input u = 164V, disturbance

w = 191A, performance variable z2 = 2, and vector x as described by Eq. 4-1 la.

Two linearization cases considered in this study are the following:

1. The direct linearization of the fuel cell model Eqs. 4-4 4-6 and shown in Figure 4-
2a [31].

2. The linearization of a combined system consisting of a static feedforward
controller, connected to the fuel cell model Eqs. 4-4 4-6 shown in Figure 4-2b.

The static feedforward control law used in case 2 is described in Section 4.5.2.1. The state,

input, output, and feedthrough matrices of the resulting linear time-invariant models for cases 1

and 2 are shown in Figures 4-3 and 4-4, respectively. Through personal communication with the

members of the research group at University of Michigan we learned that the matrices reported

in [31] are affected by errors, hence these matrices are re-derived and found to differ slightly

from the ones given by Pukrushpan et al. [31]

4.5 Literature Precedents fo Fuel Cell Control Designs

Pukrushpan et al. investigate the design and application of two main control strategies

along with 2 major variations of each of these strategies, as indicated in the following list [31]:









1. Feedforward strategy

a. Static feedforward (sFF)
b. Dynamic feedforward (dFF)

2. Combination of static feedforward strategy with optimal feedback control

a. Linear quadratic regulator (LQR)
b. Linear quadratic gaussian observer (LQG) in combination with LQR
feedback.

A succinct discussion of these strategies is given in the ensuing sections

4.5.2 Feedforward Strategy

The two feedforward control strategies proposed in the literature are a static feedforward

(sFF) and a dynamic feedforward (dFF) scheme. The sFF controller is derived from simulations

and using the results of substantial experimental work. The dFF controller is based on a linear

model, and thus its performance on the nonlinear model is dependent on the non linear model's

proximity to the nominal operating point. The models used for control design are discussed in

greater detail in the next subsections.

4.5.2.1 Static feedforward controller

Pukrushpan et al. propose the sFF control law [31]

u = 123w +378 (4-12)

where the stack current w is the measured disturbance impacting the fuel cell model. The control

input u is the voltage vcm applied to the compressor. A schematic of the sFF controller is shown

in Figure 4-5. The derivation of Eq. 4-12 consists of first seeking a function relating the

disturbance I,, to the required air mass flow rate Wcp, in such a fashion that the flow rate,

achieved by invoking thermodynamics principles, negates the effect of the disturbance on the

performance variable A0 A resulting static function is obtained by means of simulations and

experimental work which co-relates the control input vcm to the required air mass flow rate Wp









for a particular value of disturbance I,, maintaining the desired value of the performance

variable A0 This static function is implemented as a static sFF controller in the form of Eq. 4-

12.

4.5.2.2 Dynamic feedforward controller

Pukrushpan et al. propose the dynamic feedforward (dFF) control law

du = Kudw (4-13)

where 9u = u u and 6w = w w are the deviation values from the nominal control input u

and the nominal disturbance w respectively, and Kuw is the transfer function

K ideal
Kw = (4-14)
(1+ )(1+-)(1+-)
a1 a2 a3

where s is the Laplace variable, a1, a2, and a3 are filter constants, and K'de"l = -G ,,G2

where G2,, and G,2, are transfer functions describing the map

Z, = Gwt + G2,,ut (4-15)

where z2 = z2 Z2 is the deviation of the performance variable (oxygen excess ratio) from its

nominal value z2 [31]. The schematic and transfer-function representations of the dynamic

feedforward controller dFF are given in Figure 4-6.

The derivation of the dFF control transfer function begins by considering the transfer

function

&Z2 = GwaW + Gz,, K w, (4-16)

obtained by substituting Eq. 4-13 into Eq. 4-15. The expression 4-16 should become identically

zero for an effective dFF controller K,,. The expression









K,:al =-G1 ,G (4-17)

is obtained from Eq. 4-16 after equating 2 = 0. However, G is not a proper function. Low

pass filters are added to implement a causal controller, resulting in

1I
K G z2uG2w (4-18)
(1+ -)(1+- )(1+-)
al a2 a3

The authors of this derivation propose the filter-constant values a, = 80, a2 = 120, and

a3 =120 [31].

4.5.3 Combination of Static Feedforward with Optimal Feedback Controllers

Feedforward controllers lack robustness to disturbance and model variations. To address

this issue, feedback controllers are added in conjunction with feedforward controllers.

The following two scenarios are considered for feedback control design, in conjunction with

static feedforward control [32-34]:

a. The performance variable is measurable.

b. The performance variable is not measurable, leading to the introduction of a state
observer to estimate its value.

The model used for designing the optimal feedback controller is defined in such a fashion

that it includes the static feedforward control relationship. A schematic of the resulting

feedforward and feedback control is shown in Figure 4-7.

4.5.3.1 Case where the performance variable is measurable

Pukrushpan et al. propose the linear quadratic regulator (LQR) feedback control law:

u= up -KPx -KIq (4-19)









where up is a pre-compensator, iu = u u and & = x x are the deviation values of the

feedback control input u and the state vector x from their nominal points u and x,

respectively, and q is the integral defined through the differential equation
,Yeq (4-20)
S= y,1" y, (4-20)

where y1"e = Wc" is the analytically calculated value of the air mass flow rate needed for a

particular value of the disturbance It to attain the desired oxygen excess ratio, and y, = Wcp is

the measured value of the air mass flow rate. In the control law given by Eq. 4-19, gain matrices

K, and KI refer to the optimal gains resulting from minimizing the quadratic cost function


J= j(x'C'T C& +qTQq+ uT'R5u)dt (4-21)
0

where, Q,, Q1, and R are weighting matrices for the performance variable, state q and control

input u, respectively. The term C, refers to the second row of the C matrix, given in Figure 4-4

describing the linear state space model. Note that the second row is associated with the

performance variable Ai The pre-compensator up in the feedback control law Eq. 4-19 ,used

to take into account the disturbance effect on the performance variable, is given by

up =[ (A B,K,)- B, [D2" -C2 (A B,Kp)1 B ] (4-22)

where A, B,,B and D Z2 are elements of the linear state space model given in Figure 4-4. In

particular Bw and B, are the first and second column of the B matrix, respectively and D 2W is

the second element of the first column of the D matrix. A schematic diagram of the sFF scheme

supplemented with the feedback controller is shown in Figure 4-7.









The optimal linear quadratic regulator 4-19 is implemented to prescribe part of the

feedback control input. The additional state q, as described in Eq. 4-20, is introduced to

minimize offset. Using Qz = 10000 and Q, = 0.001 in Eq. 4-21 and minimizing the cost function

4-21 the optimal values of controller gains obtained are

K =[-23.03 -1.4e-17 -25.46 5.2e-5 2.0e-6 2.64 1.9e-17 -1.le-6] (4-23)

and K, = -0.001 (4-24)

Note that 4-23 and 4-24 differ from the values reported by Pukrushpan et al. [31]. This

discrepancy is a consequence of the difference, reported in Section 4.4, with the state matrices

used by Pukrushpan et al..

4.5.3.2 Case where the performance variable is not measurable

Optimal observer. When selected states are not measurable, Pukrushpan et al. propose

the following modification of the control law 4-19:

u = up -K,& -Kq (4-25)

where 5x = x x is the deviation of an estimated state vector k from its nominal value x The

estimated state is computed from the Kalman based observer

S= A + Bu + L(y- -) (4-26)

5 = C + Du (4-27)

where L is the optimal observer gain calculated using the linear quadratic gaussian (LQG)

method, and 5 is the estimated values of the measured outputs [31]. Three measurable outputs,

namely the compressor mass flow rate W the pressure of the supply manifold ps, and the

stack voltage v,,, are used as inputs to the observer to estimate the value of the states [31]. The

optimal gain of the Kalman observer is









7.43e-015 -12.22e-009 417.04e-015
-18.54e-015 31.69e-009 2.18e+000
-2.54e-015 25.06e-009 -349.02e-015
578.06e-012 4.22e-003 -49.54e-009
L = (4-28)
332.44e- 009 460.51e- 003 4.60e- 006 (4
-78.31e-015 124.80e-009 -143.89e-015
270.13e-018 -9.95e-009 1.85e+000
-486.77e -009 527.22e- 003 5.97e -006

Note that 4-28 differs from the value reported by Pukrushpan et al. [31]. This is a consequence

of using different state matrices than what were used by Pukrushpan et al. as explained in

Section 4.4. In addition, we also made an approximation to the noise variance matrices used for

LQG design, since only partial information is available in [31], hence adding another source of

discrepancy.

4.6 Generalized Predictive Control

The generalized predictive control (GPC) law is given by the discrete-time law [35, 36]

Au(t) = k T (m(t)- yOL(t)) (4-29)

where k is the GPC control gain vector, m(t) = A"' is the set point, yOL (t) is the vector of the

constant forcing values of the performance variable, u is the prescribed control input, and the

symbol A represents the difference operator. The vectors m(t), yOL (t),and the gain k have R

elements each, representing their respective values upto the prediction horizon R. The process

involved for obtaining the GPC control law Eq. 4-29 is discussed in the following sections.

Generalized predictive control design. The first step leading to the derivation ofEq. 4-

29 is to acquire a DARMA deterministicc autoregressive moving average) model from

continuous state space model [35-38]. To accomplish this, a discrete version of the linear state

space model is obtained. The sampling period Tis chosen such that about 10-20 samples are









taken during the transient time. The z-transform is performed on the discrete model and the z

variable is replaced with the q forward-shift operator. The numerator and the denominator are

divided with the highest power of q to obtain a denominator polynomial A and a numerator

polynomial B as functions of the backward shift operator q 1. Polynomial B is defined such that

it does not include the unit time delay introduced by sample and hold operation. Finally, the

DARMA model for the fuel cell model is obtained in the form

A(q-')y(t) = B(q-')u(t 1) (4-30)

A(q- ')= 1+aq-' +.. +a,q-' +.. +a, q- (4-31)

B(q') = b, +bq-' +..+bq- +..+b qb (4-32)

and where A and B are polynomials in the backward shift operator q'1. The factors a, and b,

are the coefficients of the powers q-' in polynomials A and B, respectively. The symbol y

represents the output (performance variable), i.e. the oxygen excess ratio in the case of the fuel

cell, and u is the control input, i.e., the voltage supplied to the compressor motor Vcm to deliver

the desired air mass flow rate [29, 30].

Second, a predictor for the performance variable is designed using the Diophantine

equations

1= EAA(q )+q F,(q ') (4-33)

where E, = 1+eq 1 +..+ejq J +..+e, q (4-34)


S= f + fq +..+fjq +..+fq "n (4-35)

i = 1,2,..,R (4-36)









and where R is the prediction horizon, A = 1 q and E and F are polynomials in powers of the

backward shift operator q'1. The factors e, andf are the coefficients of the powers q-1 in

polynomials E, and F,, respectively [30, 32]. Next, multiplying both sides of the DARMA

model (Eq. 4-30) by the polynomial q'E,A and invoking Eqs. 4-33-4-36 yields


y(t + i) = E, (q-1)B(q-1)Au(t + i- 1) + F (q-')y(t) (4-37)

A
where y(t + i) is the predicted value of the output at the instant t + i. Next, the constant forcing

value of the performance variable is calculated. The product of polynomials E, and B can be

decomposed in the form

E, (q )B(q ) = G,(q )+ q- 'F,(q ) (4-38)

where G, and F, are operator polynomials given by

F, (q-) = Yo +Y1ql +..+yJqJ +..+nb + l + (4-39)

G,(q-)= go +glql1 +..+g,q +..+g lq '+ (4-40)

and where y, and g, are the coefficients of powers of q-1 in the respective polynomials.

Substituting Eq. 4-38 into Eq. 4-37 yields


y(t +i)= G,(q)Au(t + i- 1)+,(q)Au(t- 1)+ F(q )y(t) (4-41)

A constant forcing Au(t + i -1) = 0 for i = 1,2,.., produces a constant-forcing output in Eq. 4-41

of the form

yOL (t + ) = ,(q1)Au(t- 1) + F,(q1)y(t) (4-42)

where OL (t + i) represents the constant-forcing response of y at instants t + i, i= 1,2,..,.









Third, an objective/cost-function J is established and minimized with respect to the

control-input increment Au to obtain the GPC control law. The appropriate cost function

adopted is defined by

J(Au)= (m(t) y(t))' a(m(t)- y(t)) + AuTAu (4-43)

where o is a weighting matrix, m and y are vectors of future set point and predicted future

output respectively, u is the vector of control inputs, and h is the weighting matrix for Au. The

vectors m and y have R elements each, representing their respective future values upto the

prediction horizon R. Vectors u and Au has L elements, where L is the control horizon defined

as the instant where the control design specifies that

u(t+i)=u(t+L -), i=L,L +,...,R -1 (4-43a)

In the case of the fuel cell, the performance variable y is composed of a contribution from

the disturbance variable d (namely, the stack current Ist), and a contribution from the manipulated

variable u (namely, the voltage to the compressor v,, ) leading to its definition


y = d+ = yOL + GdAd+ yL +G Ai/ (4-44)

where Yd is the component of the performance variable contributed by disturbance d. Note

that the notation for the disturbance is changed from w Sections 4.4 to 4.6 to din Sections 4.7

which involve the GPC strategy. Signal y, is the component of the performance variable y

contributed by manipulated variable u. The terms yOL and yOL represent the constant-forcing

values of Yd and y, respectively. The terms Gd and G, are the associated dynamic matrix

polynomials for Yd and y, respectively. The termGdAd is set as GdAd = 0, as Ad = 0 since d

is a disturbance and its unknown future values are assumed to be equal to the current value, i.e.,

d(t,) is assumed to be constant and equal to d(t), i = 1,2,... This reduces Eq. 4-44









y = y+y = yOL +yOL +G At, (4-45)

Hence, since y = yL when Au = 0, it follows that

yO = YdL y (4-46)

Substituting the value ofy described by Eq. 4-45 in the cost function Eq. 4-43 yields

J(Au)= (m(t) yOL(t)- GAu)TO (m(t) yOL(t)- GAu) + AuTAu (4-47)

which when minimized with respect to Au yields

Au = (GuoTGu) + I)-(GTuT)X (4-48)

where X = m(t) -yOL(t) The control law is extracted from Eq. 4-48, takes the form

Au(t)= kTX (4-49)

where kT is the first row of the matrix (G TGo) + I) -(GuoT), and u(t) = u(t -1)+kTX by

the GPC algorithm. Eq. 4-49 can be easily rewritten via a simple substitution of factor X to

reduce to Eq. 4-29. This completes the derivation of the GPC control law Eq. 4-29.

For stability analysis and simulation purpose it is important to develop a closed-loop

transfer function for the GPC loop. The GPC control law described by Eq. 4-49 in summation

form is given by

R
Au(t) = k,[m, -yr o ] (4-50)


where the subscript i denotes the values of the respective variables at the time instance t+i. Then

using Eq. 4-42

,(t + i) = Fd, (q-1)Ad(t -1)+ F, (q-1)yd() (4-51)

S(t + i) = (q-)Au(t 1) + F, (q-')y, (t) (4-52)










Substituting the values of y, and y, given by the Eqs. 4-51 and 4-52 into the GPC control


law given by Eq. 4-50 and rearranging terms yields

R R R
(I+q kjF,,)Au(t) = k q'm, A(q-l Y krd,,)d(t)
=1 R1 ,1 R (4-53)

-(_kFd, (q- 1)yd(t)-(_ k q-F,, )y (t)
z=1 z1i

R
Let R, =(l+q k, F,,)A (4-54)
i=1

R
Rd =A(q k d,,) (4-55)
i=1


T = kq' (4-56)
i=1

R
Sd = IkFd,,(q-1) (4-57)
1=1

R
and S= = kF,,(q1) (4-58)
i=1

Inserting in Eqs. 4-54 4-58 into Eq. 4-53 yields

Ruu(t) = Tm(t) Rdd(t) Sdyd (t) Su yu (t) (4-59)


Since y, (t) = y(t) yd (t) (4-60)


Eq. 4-59 can be rewritten as

Ruu(t) = Tm(t) Rdd(t) Sdy ( (t) (t) yd (t)) (4-61)

which after rearranging terms results in

Ru(t) = Tm(t) Rdd(t) (Sd S. )yd (t) Sy(t) (4-62)


Also, since Yd -1 d d(t) (4-63)
Ad








Eq. 4-61 can be expressed in the form


B,
R u(t) = Tm(t)- Rdd(t)- (Sd S )q 1 d d(t) Sy(t) (4-64)
Ad

which after a trivial series of algebraic operations can be written as

AdRu(t) = AdTm(t)-(AdRd +(Sd S )q'Bd)d(t)- AdSy(t) (4-65)

Now, let R = AdR (4-66)

T= AdTm(t) (4-67)

S = AdRd + (Sd S,)q-Bd (4-68)

and S, = AdSu (4-69)

Substituting Eqs. 4-66 4-69 into Eq. 4-65 yields

Ru(t) = Tm(t) Sdd(t) S,y(t) (4-70)

which finally leads to the closed-loop established by the GPC control strategy

Ru(t) = Tm(t) S(t) (4-71)

where S= (Sd Sy) (4-72)


and y(t) ) (4-73)
a y(t))

Figure 4-8 shows a closed loop schematic of the GPC controller for the augmented model which

incorporates the disturbance.

To obtain the closed-loop transfer functions relating the input u(t) and the outputy(t) to the

set point m(t) and the disturbance d(t), first consider the open-loop relationships


yd(t) = Bd d(t) (4-74)
Ad








B
and y (t) = q-1 u(t) (4-75)
A

Let q'Bd = Bd and q'B = B,. Substituting Eqs. 4-74 and 4-75 into Eq. 4-59 and rearranging

terms yields

u(t) r A.R T + (t)- Ad) ARd + SdB-d d(t) (4-76)
A, R, + S,) Ad A, R, + S, )B

which is the closed loop transfer function from the set point m(t) and the disturbance d(t) to the

control input u(t). Inserting into Eq. 4-65 the expression


y(t)= KBd d(t) + \ u2(t)A (4-77)

obtained from Eqs. 4-45, 4-74, and 4-75 yields


y(t)= A SBT m(t ) + B(A R B(AdR d(t) (4-78)
AR+SB Ad(A R, +SB)

which is the closed loop transfer function from the set point m(t) and the disturbance d(t) to the

output y(t).

4.7 Battery of Observers

In this scheme, a parallel battery of four Kalman based observers designed using four

different nominal points, respectively, are deployed for estimating the value of the controlled

variable where the equations for each observer i is given by

Ix = A I + Bu+L (y- ) (4-79)

y, = C,,, + D,u (4-80)









where i, and y, are the estimated state vector and estimated output vectors from observer i,

respectively. The state matrices A,, B C, and D, correspond to the nominal point for which the

observer i is designed.

The observer which delivers the best performance is used for providing its estimated value

of the controlled variable to the GPC controller. The observers are based on linear models

derived from the nonlinear model at different nominal values of the disturbance. The observer is

defined by Eqs. 4-79 4-80. The observer gain L is computed from LQG principles as defined

in previous section for each of the observers.

The determination of the best performing observer is done by comparing the norm of the

error in the measured outputs produced by each observer. The error-norm N, for observer i is

defined by

N, = e', +e,, +e3,, (4-81)

where e, = W (4-82)

e2, = Psm sm, (4-83)

e3, = vt ~t, (4-84)

are the errors defined by the difference between the measured value of the outputs (Wc psm, and

v, ) from the nonlinear model and their estimated values (Wp, ,, sm and t,,,) from the

observer i, respectively. The observer that delivers the least value of the norm is selected as the

best current observer and is implemented to deliver its estimated value of the controlled variable

to the controller. The nominal value of the disturbances for which the observers are designed are

w=100, 191, 125, 22 A. The values are chosen to account for the range of disturbance values

that the fuel cell model experiences.









4.8 Simulation Studies and Results

This section presents simulations results obtained from various control strategies

implemented on the nonlinear fuel cell model. The MATLAB and SIMULINK software

computational tools are used for simulation purposes and for assessing the controllers'

performance [22]. The nonlinear fuel cell model created by Pukrushpan et al., in the

SIMULINK environment is used for simulations [31].

The performance of the generalized predictive control strategy is presented first. The

scenario where the performance variable is measured is discussed, followed by the case when it

is not measured. In the latter case an observer is incorporated in the control design to estimate

the unmeasured value.

The robustness of the GPC control design is assessed by examining its performance on an

altered fuel cell model obtained by modifying a parameter of the original nonlinear model.

Finally, a comparison of the GPC controller's performance with that of controllers proposed in

literature is conducted.

For all the simulation studies carried out in this work, the disturbance d ( i.e., the stack

current It ) profile shown in Figure 4-9 is used, which is identical to the one used by Pukrushpan

et al [31]. The y-axis denotes the value of the disturbance and the x-axis is time. The trajectory

of the disturbance is implemented as step changes. In Figures 4-10 4-19, showing the

simulation results for various control configurations, the y-axis denotes the value of the

performance variable oxygen excess ratio Al2 and the x-axis is depicts the time t. The dotted

black line indicates the value of the desired set point i.e., m(t) = 2 for the performance variable.









4.8.1 Generalized Predictive Control Results

As discussed in Section 4.4, the GPC approach is implemented on the fuel cell model to

attain the desired objective of regulating the performance variable, i.e., keeping the oxygen

excess ratio at a value of 2. The following scenarios are considered for evaluating the

performance of the GPC strategy:

1. The controlled variable is measured.

2. The controlled variable is not measured. A battery of observers based on Kalman
filtering is included to estimate the value of performance variable.

3. The design model is uncertain. The robustness of the GPC controllers under model
uncertainty for cases (1) and (2) above is evaluated by examining the performance
on a modified fuel cell model obtained by varying a parameter of the original
model.

4. The control is benchmarked against literature precedents. The performance of the
GPC controller is compared to control designs proposed for the fuel cell model in
prior literature.

4.8.1.1 Case where the performance variable is measured

The performance of the GPC controller when the controlled variable is measured is

discussed in this section. Since in this case the controlled variable (oxygen excess ratio) is

measured, its value is directly available to the GPC controller. The GPC algorithm described in

Section 4.6 dynamically prescribes the value of the control input (the voltage to the compressor

motor vt) which in turn delivers the appropriate compressor mass flow rate (Wcp) to maintain the

value of the controlled variable at the desired set-point value of 2.

Figure 4-10 shows the results of a simulation study where the GPC control design is

implemented on the nonlinear fuel cell model. The x-axis in Figure 4-10 denotes the time and

the y-axis depicts the value of the controlled variable, i.e., the oxygen excess ratio measured

from the nonlinear fuel cell. The initial value of states, disturbance and control input for the

model are chosen as discussed in Section 4.4. The dotted line indicates the set point of the









controlled variable, which in this study is constant and equal to 2. The black solid line depicts

the measured value of the controlled variable under conditions where the fuel cell is subjected to

the disturbance profile (values of stack current It) shown in Figure 4-9 and the input is

prescribed by the GPC algorithm.

Figure 4-10 shows that the GPC controller successfully returns the controlled variable at

the desired value without steady-state offset. Only small deviations from the set point are

observed during transients. The spikes in the signal plot reflect instances where the disturbance

suddenly changes values in a stepwise fashion, as documented in Figure 4-9 and hence they are

unavoidable. A value of the controlled variable below 2 indicates that a positive step change in

disturbance took place, with the effect of depleting the oxygen concentration in the cathode and

consequently reducing the oxygen excess ratio before the control action could remedy the

problem. Analogously, a transient value of the controlled variable lying above 2 reflects an

opposite scenario of an occurrence of a negative step change in disturbance. The asymptotic

offset-free behavior observed is expected as the GPC approach incorporates an integrator as an

essential component of its structure.

4.8.1.2 Performance variable not measured

Battery of four observers. Figure 4-11 shows the time response of the performance

variable when a battery of four observers are included in the controller to account for the

situation where the controlled variable is not measured. The observer design steps are discussed

in Section 4.7. In Figure 4-11, the solid line indicates the performance variable trajectory as a

function of time. The dotted line is the desired set point value of 2 for the performance variable.

The measured outputs from the fuel cell model are (1) the compressor mass flow rate Wp,

(2) the stack voltage vt ,and (3) the supply manifold pressurepsm. These measurements are

provided to the observers to estimate the value of the performance variable. The observers









incorporated in the control strategy provide estimates for the performance variable and the norm

error value that they generate. According to the switching algorithm the best performing

observer, that generates the least value of the norm-error value, feeds its estimated value of

oxygen excess ratio to the GPC controller. The GPC strategy attempts to regulate the value of

the estimated performance variable at 2. Excellent results are obtained by using a battery of

observers which is evident by zero-offset/near zero-offset of the performance variable value from

the setpoint.

The norm-error values of the four observers are shown in Figure 4-12. The switching

pattern indicating the observer used for estimating the value of oygen excess ratio is shown in

Figure 4-13. The value of the control input, the voltage prescribe to the compressor motor, is

shown in Figure 4-14. The deviation from the estimated and measured three outputs for the

observers 1, 2, 3, and 4 are shown in Figures 4-15, 4-16, 4-17, and 4-18, respectively.

Battery of three observers. An identical strategy described above is used for observer

design except that a battery of three observers instead of four is implemented. Observer 3 is

excluded from the battery of observers. Figure 4-19 shows the time response of the performance

variable. The solid line indicates the performance variable trajectory as a function of time. The

dotted line is the desired set point value of 2 for the performance variable. A deterioration in

control performance is observed compared to the case when four observers are used. This result

is expected as there are fewer observes available that are designed in the vicinity of the different

operating points of the nonlinear fuel cell model. As a result, the estimated value of the

performance variable provided by the observers do not coincide with the actual value resulting in

some deviation from the set point because the observer design is based on a linear model. The









GPC strategy in turn attempts to regulate the erroneous estimated value of the performance

variable and consequently a certain degree of deterioration in controller performance is expected.

Battery of two observers. Figure 4-20 shows the time response of the performance

variable when only two observers are employed in battery of observers. The solid line indicates

the performance variable trajectory as a function of time. The dotted line is the desired set point

value of 2 for the performance variable. Observers 1 and 3 are excluded from the battery of

observers. A further deterioration in controller performance is noted compared to the case where

three observers are employed. This is an expected result as discussed earlier.

Employing only one observer. Figure 4-21 shows the time response of the performance

variable when only observer 2 is used to estimate the value of the performance variable. The

solid line indicates the performance variable trajectory as a function of time and the dotted line is

the set point value of 2. This scenario exhibits the worst controller performance compared to the

cases where batteries of observers having 2, 3, or 4 observers are used. The GPC controller with

one observer supplement fails to deliver offset free regulation. This is an expected result, as

discussed earlier.

4.8.2 The GPC Approach Evaluated for Robustness

For robustness considerations, the GPC strategy was evaluated for model uncertainty. The

GPC controller designed for the original fuel cell model is used on a modified model which is

acquired by changing a parameter of the original model. More specifically, the return manifold

throttle area changed from 0.0020 m2 to 0.0023 m2. The two scenarios considered for robustness

analysis are

1. The performance variable is measured.

2. The performance variable is not measured. An LQG observer is included in the
control strategy to estimate the value of performance variable.









4.8.2.1 Case where the performance variable is measured

Figure 4-22 shows the time response of the performance variable from the modified

nonlinear fuel cell model when the GPC control law is designed for the original model. The

solid black line indicates the time response of the performance variable from the modified

nonlinear fuel cell model. The intention is to examine the performance of the GPC controller in

a scenario of model uncertainty. In this case the performance variable is measured and is fed as

a direct input to the GPC controller. The simulation conditions are identical to the scenario

discussed in Section 4.7.1.1, except that the modified non linear model is used.

As observed from Figure 4-22, the GPC controller displays robustness to model

uncertainty because it ensures effective rejection of the effect of different values of disturbance,

and produces offset-free steady-state responses. The robustness of the GPC algorithm to model

uncertainty is a key advantage of the controller. Even though the GPC strategy is designed for

the original model, the feedback control of the performance variable produced by the modified

model gives the controller the opportunity to make the necessary changes to minimize deviations

from the set point. There is zero offset as the integrator in the GPC controller adjusts its output

appropriately to eliminate the error of the performance variable with respect to the set point.

4.8.2.2 Case where the performance variable is not measured

Figure 4-23 shows the time response of the performance variable from the modified

nonlinear fuel cell model when it cannot be directly measured. An LQG observer is incorporated

in the control design to estimate the value of the performance variable by utilizing the values of

the measured outputs (1) compressor mass flow rate Wp, (2) stack voltage vt, and (3) pressure of

supply manifold ps. The simulation conditions are identical to those given in Section 4.7.1.2,

except that the altered model is used. The GPC controller and LQG observer designed for the

original fuel cell model are used. The objective is to study the controller performance for model









uncertainty when the performance variable is not directly measurable. The solid line is the time

response of the performance variable from the modified model.

Figure 4-23 shows that when the LQG observer in conjunction with the GPC controller is

applied to the modified model, significant degradation in performance is observed compared to

the case when the performance variable is directly measured. This can be attributed to the fact

that the GPC controller is attempting to regulate at the desired set point value of 2 the value of

performance variable estimated by the LQG observer. However, the LQG observer is optimally

designed for the original model at a nominal operating point which is different from the actual

operating point and hence delivers an erroneous estimated value of the performance variable. In

the case when the performance variable is directly measured, previously discussed in Section

4.7.1.1 the GPC controller successfully makes adjustments to minimize a correct value of the

error. In the current case when the LQG observer is included, the GPC algorithm makes an

effort to minimize an incorrect error. The LQG observer is optimally designed for the original

linear model at the nominal point. In the current case not only is the nonlinear model used but

the problem is amplified further by the fact that there is a deviation from the original nonlinear

model by virtue of the uncertainty in one of its parameters.

Figure 4-24 shows the time response of the performance variable from the modified

nonlinear fuel cell model when a battery of 4 observers, as described in Section 4.7, are

employed to estimate the value of the performance variable. The solid line is the time response

of the performance variable from the modified model. Due to the reasons, as discussed above,

no significant improvement in controller performance is observed.









4.8.3 Comparison of the GPC Strategy with Prior Control Designs

In this section comparison of the GPC controller's performance with controllers proposed

in prior literature for fuel cell is conducted. Pukrushpant et al. propose controllers for the

following scenarios [31]:

1. All states are measured, leading to the implementation of the sFF controller with an
LQR strategy for feedback.

2. All states are not measured, leading to the implementation of an LQG observer to
estimate the value of the performance variable in addition to the sFF controller with
the LQR strategy for feedback.

4.8.3.1 Case where all states are measured-sFF with LQR feedback control

Figure 4-25 shows the time response of the performance variable for the GPC strategy and

the sFF with LQR feedback control strategy. The approach of sFF with LQR feedback control

strategy is discussed in Section 4.5.3.1. In Figure 4-25, the solid line is a plot of the performance

variable as a function of time in the case where the GPC controller is used. The dashed line

represents the performance variable trajectory when the sFF with the LQR feedback strategy is

applied. The dotted line is for the set point, fixed at value 2.

It is noted from the figure that the GPC controller delivers better performance as the sFF

with the LQR feedback control strategy is not able to eliminate the steady-state offset. The GPC

algorithm is able to eliminate offset as discussed in Section 4.7.1.1. The LQR controller

prescribes an optimal control input based on the linear model at the nominal point. However, the

controller's performance is being examined then applied on a nonlinear model. Consequently, as

there is deviation of the disturbance from the nominal operating point the performance of the

LQR controller degrades.









4.8.3.2 Case where all states are not measured-observer design

Figure 4-26 shows the time response of the performance variable when an LQG observer is

incorporated in the controller in the scenario where the performance variable is not measured.

The dashed line indicates the time response of the performance variable when the static

controller sFF with the LQG observer and the LQR feedback controller is applied. The

controller strategy is discussed in Section 4.5.3.2. The solid line is the value of the performance

variable when the GPC feedback with an observer is applied as the control strategy. It is noted

from the time response curves that there is deterioration in both controllers performance

compared to the case reported in Figure 4-25 where the performance variable is measured.

Note, however, that the LQG observer with the GPC feedback controller exhibits better

performance compared to the sFF with the LQG observer and the LQR feedback controller. The

LQG observer component is identical for both the controllers, consequently the strategy for

estimating the value of the performance variable is alike. The GPC algorithm is able to regulate

the estimated value of the performance variable in a superior fashion. The sFF with LQR

feedback controller lacks the same level of dynamic ability, and is not able to deliver the same

level of regulatory performance that the GPC approach realizes. The LQR controller computes

an optimal gain for the linear model at the nominal point. The implementation of the LQR

strategy at non-nominal point i.e., on the nonlinear model degrades its performance.

Figure 4-27 shows the time response of the performance variable when a battery of 4

observers, as described in Section 4.7, are employed to estimate the value of the performance

variable. The solid line is the time response of the performance variable. Due to the reasons, as

discussed earlier, excellent results are obtained by using a battery of observers which is evident

by zero-offset/near zero-offset of the performance variable value from the setpoint.









4.8.3.3 Comparison of controller performance with respect to robustness

The controllers discussed in Sections 4.7.3.1 and 4.7.3.2 (the sFF with LQR feedback and

the sFF with LQG observer and LQR feedback), are evaluated and compared with the GPC

controller for their robustness to model uncertainty. An approach identical to that of Section

4.7.2 is adopted. The modified model is obtained from the original nonlinear model by altering

the value of the return manifold throttle area from 0.0020 m2 to 0.0023 m2. The two scenarios

considered are:

1. All states measured i.e., performance variable is measured.

2. All states not measured, i.e., an observer is included to estimate the value of
performance variable.

All states measurable, sFF with LQR feedback. The performance of the GPC controller

and the sFF with LQR feedback controller is compared in the scenario of model uncertainty.

The robustness of the controllers is examined by implementing them on models for which they

were not originally designed. In the current case the performance variable is measured.

The control strategies for the GPC and the sFF with LQR feedback controllers described in

Section 4.6 and Section 4.5.3.1, respectively, are implemented on the modified nonlinear fuel

cell model. The difference in the simulation scenario compared to the one presented in Section

4.7.3.1 is that the model used to describe the nonlinear fuel cell dynamics has a different return

manifold throttle area. Figure 4-28 shows the time responses of the two control strategies. The

dashed line indicates the performance variable response to the sFF with LQR feedback control

strategy. The solid line is the performance variable response to the GPC control strategy.

The GPC controller delivers better performance by eliminating offset as discussed in

Section 4.7.2.1. The sFF with LQR feedback does not deliver the same level of performance as

an offset with respect to the set point is observed. The relatively poor performance of the sFF









with LQR feedback strategy is expected as the feedforward and optimal gain are designed for a

linear nominal point of the unmodified plant model.

Performance variable not measured, sFF with LQG observer and LQR feedback.

Figure 4-29 shows the time response of the performance variable for the two control strategies

(the LQG observer with GPC feedback and the sFF with LQG observer and LQR feedback) in

the case where the performance variable is not directly measured and an LQG observer is

incorporated in the control design to estimate its value. The LQG observer is identical to the sFF

with LQR feedback and the GPC control strategies. The dashed line indicates the time response

of the performance variable when the sFF with LQG observer and LQR feedback controller is

applied. The solid line is the performance variable time-response when the GPC approach with

LQG observer is implemented. The simulation conditions are identical to the one discussed in

Section 4.7.3.2, except that the modified model as discussed in Section 4.7.2 is used.

In the case when the performance variable is not measured and the GPC strategy with LQG

observer is implemented on the modified model, degradation in controller performance is

observed as compared to when the performance variable is measurable as discussed in Section

4.7.2.2. However, the GPC with an LQG observer control design delivers better performance

(smaller offset) than that of the sFF with LQG observer and LQR feedback control design. This

can be attributed to the fact that the erroneous estimated value of the performance variable is fed

as an input for the LQR feedback controller which itself is being implemented on a model it was

not designed for. Consequently, a relatively higher degradation in controller performance is seen

on incorporation of the observer. The GPC design performs comparatively better though it also

fails to deliver offset free behavior. As mentioned before, the observer is identical for the two

controllers. In the case of the sFF with LQR feedback strategy, the situation is compounded









further by the fact that the LQR gain is not only being calculated at the non-nominal point

(nonlinear model) but on a modified nonlinear model.

Figure 4-30 shows the time response of the performance variable from the modified

nonlinear fuel cell model when a battery of 4 observers, as described in Section 4.7, are

employed to estimate the value of the performance variable. The solid line is the time response

of the performance variable from the modified model. Due to the reasons, as discussed above,

no significant improvement in controller performance is observed.

4.8.4 Feedforward Control Designs

Two exclusively feedforward control designs, namely (1) static feedforward (sFF) and (2)

dynamic feedforward (dFF) without feedback components are proposed by Pukruspan et. al.

[31]. The two controllers' performance is compared with the GPC strategy for the following two

cases:

1. Application on the original model

2. Application on the modified model to assess and compare robustness to model
uncertainty.

4.8.4.1 Case of original model

Figures 4-31 and 4-32 compares the performance of the sFF and dFF controllers with the

GPC algorithm on the unaltered original nonlinear fuel cell model for the following two

scenarios: (1) the performance variable is measured, and (2) the case when it is not. In the

scenario where the performance variable is not measurable, an LQG observer is included in the

GPC control design.

The sFF and dFF controllers are discussed in Sections 4.5.2.1 and 4.5.2.2, respectively.

The GPC approach and the GPC with the LQG observer strategies are implemented as described

in Sections 4.6.1 and 4.5.3.2, respectively, and the simulations are conducted as discussed in









Section 4.7.1. The solid and dashed black lines represent the time responses of the performance

variable in the cases when the sFF and dFF controller are applied, respectively. The red and blue

lines indicate the time responses of the performance variable when the GPC strategy is adopted

for the scenarios where the performance variable is measured and not measured, respectively. In

the case when the performance variable is not measured, an LQG observer is included in the

control design to estimate its value.

In the case when the performance variable is measurable, the GPC controller delivers the

best result by eliminating offset with respect to the set point as discussed in Section 4.7.1.1. In

the case when the performance variable is not measurable and an LQG observer is incorporated

in the control design there is a relative degradation in control performance, compared to the case

when it is measured, as discussed in Section 4.7.1.2. The performance of the dFF controller is

the worst observed, as expected because the controller is designed for the linear model of a fuel

cell at the nominal point. As the deviation from the nominal point increases, the controller

dynamics of the dFF design do not correspond to the model it was designed for. The sFF

strategy performs better compared to dFF approach as the deviation from nominal point

increases. This behavior is expected as the sFF algorithm is derived from simulations and

experimental results based on the nonlinear model. The sFF, dFF and GPC with observer

strategies are not able to eliminate offset.

Figure 4-33 shows the time response of the performance variable when a battery of 4

observers, as described in Section 4.7, are employed to estimate the value of the performance

variable. The estimated value is fed to the GPC controller. The solid line is the time response of

the performance variable. Due to the reasons, as discussed earlier, excellent results are obtained









by using a battery of observers which is evident by zero-offset/near zero-offset of the

performance variable value from the setpoint.

4.8.4.2 Case of model uncertainty

The robustness of the sFF and dFF controllers to model uncertainty is assessed by

implementing them on a modified model derived by modifying a parameter in the original

model, as explained in Section 4.7.2. The performance of the sFF and dFF controller is

compared to that of the GPC approach. The two cases for the GPC strategy considered are: (1)

the performance variable is measured, and (2) the performance variable is not measured in which

case an LQG observer is included in the control design to estimate its value, as explained in

Section 4.5.3.2.

The sFF, dFF, GPC (with and without observer) control strategies designed for the original

model are applied to the modified nonlinear model. The simulations are performed in an

identical manner to that explained in Section 4.7.4.1, except that the modified nonlinear model

was adopted.

Figures 4-34 and 4-35 shows the results of the simulations. The solid lines are the traces

for the performance variable in the case when GPC approach is employed, respectively

implemented for the scenarios (1) and (2) described above. The solid and dashed lines represent

the time responses of the performance variable for the sFF and dFF control strategies,

respectively.

Figure 4-34 shows that in the case where the performance variable is measured the GPC

controller delivers the best result by eliminating offset of the performance variable with respect

to the set point, as discussed in Section 4.7.2.1. However, due to reasons discussed in Section

4.7.2.2, in the case when the performance variable is not measured, there is a relative degradation

in the GPC controller performance. The performance of the sFF and dFF controllers degrades as









well when they are implemented on the modified fuel cell model. The dFF controller performs

the worst due to reasons discussed in Section 4.7.4.1. Additionally, in the case of dFF controller

the problem is further compounded by the fact that, besides being applied at a non-nominal

operating point (adopting nonlinear model), a modified model is used.

Figure 4-36 shows the time response of the performance variable from the modified

nonlinear fuel cell model when a battery of 4 observers, as described in Section 4.7, are

employed to estimate the value of the performance variable. The estimated value of the

performance variable is fed to the GPC controller. The solid line is the time response of the

performance variable from the modified model. Due to the reasons, as discussed earlier, no

significant improvement in controller performance is observed.

4.9 Conclusions

An elegant solution is proposed for the problem of fuel cell control by implementing the

GPC scheme. The GPC approach employs the augmented model which incorporates the

measured disturbance in its algorithm. The GPC controller demonstrates better performance

compared to the control strategies proposed in literature for various scenarios. The GPC strategy

is the best performing controller regulating the value of the performance variable at the desired

set point of 2 with zero offset in all cases when it is measurable. The GPC controller

demonstrates the highest level of robustness towards the issue of model uncertainty by exhibiting

zero offset of the performance variable with respect to the set point when applied to a modified

fuel cell model.


















H2 Tank





Vcm
Motor
Compressor PS

WP-

Humidifier


Power
battery


Vst










2


Fuel cell stack


IWater Separator
Water Tank


Figure 4-1. Schematic of fuel cell system.
























Fuel Cell System


Z1 = et



Z2 = 202






Y1 = Wp




Y2 Psm




Y3 = Vst


Figure 4-2a. Fuel cell system showing input u, disturbance w, and outputs zl z2, yi, Y2, y3.


2W = Ist










U = Vcm
^ >


















W 'Is -I


Fuel Cell System


static







S4Feedback


Figure 4-2b. Fuel cell system showing sFF with feedback controller.


--- -- -- --- -- -- -- --- -- -- --- -- -- -- --- -- -- --- -- -- -- --- -- -- --- -- -- -- --- -- -- --


Z




























-12.62 0 -10.95 0 8.4e-7 6e-18 0 2.4e-7
0 -315.8 0 0 1.004e-6 0 -35.34 0
-37.57 0 -46.31 -8.6e-24 2.76e-6 2e-17 0 1.58e-6
0 0 0 -17.19 0.2032 0 0 0
2.6e8 0 2.97e8 379.4 -38.7 1.06e7 0 0
33.28 0 38.03 4.834e-5 -4.8e-6 0 0 0
0 -295.6 0 0 9.33e-7 0 -63.61 0
4.045e8 0 4.621e8 0 0 0 0 -51.22
A


-3.16e-5
-3.98e-6
0
0
0
0
-5.245e-5
0


0
0
0
405.1
0
0
0
0


4.94e6 1.967e6 -1.089e5 2.066 0 0 0 0 180.2 -165.7
-1273 0 -1454 -4.3e-22 1.388e-4 9.94e-16 0 0 -0.01049 0
0 0 0 4.834e-5 -1.16e-6 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0
2.59e4 1.03e4 -569.9 0 0 0 0 0 -0.2965 0
C D



Figure 4-3. Matrices defining the LTI model for the fuel cell model excluding sFF.


-12.62 0 -10.95 0 8.4e-7 6e-18 0 2.4e-7
0 -315.8 0 0 1.004e-6 0 -35.34 0
-37.57 0 -46.31 -8.6e-24 2.76e-6 2e-17 0 1.58e-6
0 0 0 -36.8 0.2032 0 0 0
2.6e8 0 2.97e8 379.4 -38.7 1.06e7 0 0
33.28 0 38.03 4.834e-5 -4.8e-6 0 0 0
0 -295.6 0 0 9.33e-7 0 -63.61 0
4.045e8 0 4.621e8 0 0 0 0 -51.22
A


4.94e6 1.967e6 -1.089e5 4.131 0 0 0 0
-1273 0 -1454 -4.3e-22 1.388e-4 9.94e-16 0 0
0 0 0 4.834e-5 -1.16e-6 0 0 0
0 0 0 0 1 0 0 0
2.59e4 1.03e4 -569.9 0 0 0 0 0


-3.16e-5 0
-3.98e-6 0
0 0
716.4 1065
0 0
0 0
-5.245e-5 0
0 0

B


-112.9 -435.7
-0.01049 0
0 0
0 0
-0.2965 0

D


Figure 4-4. Matrices defining the LTI model for the fuel cell including sFF.















U) ,


U = Vcm
static -->


Z1 =ep



Z2 = AZ0,


Y, = Wp
y1 p




Y2 Psm



73 = st
>-


Figure 4-5. The sFF control configurations for fuel cell system.


































100


Fuel Cell System





























U = Vcm
-- dynamic =


Z1= epFn











Y, = W
---->
Z2 sm2











2= Psm




3 V=st


Figure 4-6. The dFF controller: (a)Schematic diagram, and (b)transfer function representation.













101


w = Ist


Fuel Cell System




















w Ist z


Fuel Cell System


sFF
\ -/



-----------------------------------------------------------------


Feedback <


Figure 4-7. The sFF schematic with feedback controller.


--- -- -- --- -- -- -- --- -- -- --- -- -- -- --- -- -- --- -- -- -- --- -- -- --- -- -- -- --- -- -- --


z












d(t)


u(Q | -) B+ ^ + YO( yJ
Fu T 1/ R y) ya(t)




Figure 4-8. The GPC design in feedback block diagram.


Figure 4-8. The GPC design in feedback block diagram.


Id(t)



















285

E
i--

I 220
0



155





901
0 5 10 15 20 25 30
Time (sec)


Figure 4-9. Disturbance profile used for simulation purposes.






















F^ 2

1.8

O
1.6

1.4

1.2

1 I I I--
0 5 10 15 20 25 30
Time (sec)

Figure 4-10. The GPC control strategy implementation on the nonlinear fuel cell
model in the case when the controlled variable is measured.













.. setpoint
-- GPC with 4 observers



. .


15
Time (sec)


Figure 4-11. The GPC feedback with four observers control scheme implementation
on the nonlinear fuel cell model.


1.8

1.6


1.4

1.2













x 104

.. Observer 1
-- Observer 2
Observer 3
Observer 4


-14


I '


A


/ -' N


15
Time (sec)


Figure 4-12. The Norm of errors from the battery of observers.






























107


- -. .















Obs


-e
SObs 3



j Obs 2



Obs 1 -




0 5 10 15 20 25 30
Time (sec)


Figure 4-13. The switching pattern of the battery of observers.
















200

S180-

E 160

0 140

> 120

100

801
0 5 10 15 20 25 30
Time (sec)


Figure 4-14. Final voltage to the compressor.










0.04


0.02


0


-0.02


-0.04


10 20 30
Time (sec)


x 104


10 20
Time (sec)


Time (sec)


Figure 4-15. Observer 1, error between measured and estimated values.


zero error
-- Observer 1







b- -n--- -----











0.01
zero error
0.005 Observer 2

0

S-0.005

-0.01


-0.015
0 10 20 30
Time (sec)

15000 .
zero error
-- Observer 2
G 10000


S5000





-5000
0 10 20 30
Time (sec)

2



0

o -1

-2 ....... zero error
S Observer 2
-3

-4
0 10 20 30
Time (sec)

Figure 4-16. Observer 2, error between measured and estimated values.











0.02
Szero error
-- Observer 3
S0.01


0


-0.01


-0.02
0 10 20 30
Time (sec)
x 104
5
....... zero error
S-- Observer 3










-10
0 10 20 30
Time (sec)




CA 0 .




1-0Observer 3
-10


-15


-20
0 10 20 30
Time (sec)

Figure 4-17. Observer 3, error between measured and estimated values.
-105



-20-----------



Figure 4-17. Observer 3, error between measured and estimated values.

























10 20
Time (sec)


10 20
Time (sec)


10 "
Time (sec)


Figure 4-18. Observer 4, error between measured and estimated values.


0.01


0


-0.01


-0.02


-0.03


x 104


0


"-tv


zero error
Observer 4














.. setpoint
-- GPC with 3 observers


. . . . . . . .


15
Time (sec)


Figure 4-19. The GPC feedback with three observers control scheme implementation
on the nonlinear fuel cell model.


1.8

1.6

1.4

1.2





















S 2

< 1.8

1.6

1.4

1.2

1 I I
0 5 10 15 20 25 30
Time (sec)

Figure 4-20. The GPC feedback with two observers control scheme implementation
on the nonlinear fuel cell model.





















C [
L 2

1.8

1.6

1.4

1.2

1
0 5 10 15 20 25 30
Time (sec)

Figure 4-21. The GPC feedback with one observer control scheme implementation on
the nonlinear fuel cell model.





















FL 2

1.8

O
1.6

1.4

1.2

1 I I I--
0 5 10 15 20 25 30
Time (sec)

Figure 4-22. The GPC control strategy implementation on the nonlinear fuel cell
model with a parameter changed from the value used for control design.






















S 2-

1.8
O
1.6

1.4

1.2

1I I I I I
0 5 10 15 20 25 30
Time (sec)

Figure 4-23. The GPC controller with the LQG observer control strategy
implementation on the altered nonlinear fuel cell model.

































5 10 15 20 25 30
Time (sec)


Figure 4-24. The GPC controller with the 4 observers control
on the altered nonlinear fuel cell model.


strategy implementation





















2 C 2


0 V
1.8

1.6

1.4

1.2


0 5 10 15 20 25 30
Time (sec)
Figure 4-25. Comparison of the GPC control strategy with the sFF controller
combined with LQR feedback strategy on the unaltered nonlinear fuel cell
model when the performance variable is measurable.



















2 N-.K i --- --I --il -~--

1.8

O
1.6 -

1.4

1.2


0 5 10 15 20 25 30
Time (sec)
Figure 4-26. The sFF with the LQG observer and LQR feedback, compared to GPC
with the LQG Observer control strategy implementation on the unaltered
nonlinear fuel cell model when the performance variable is not measurable.













setpoint
2.6 GPC with 4 observers
sFF with LQG observer and LQR
2.4

2.2






1.6

1.4

1.2

I I I I
0 5 10 15 20 25 30
Time (sec)
Figure 4-27. The sFF with the LQG observer and LQR feedback, compared to GPC
with the 4 observers control strategy implementation on the unaltered
nonlinear fuel cell model when the performance variable is not measurable.






















0 2

S1.8 -

1.6

1.4

1.2


0 5 10 15 20 25 30
Time (sec)

Figure 4-28. The sFF with the LQR feedback, compared to GPC, when performance
variable is measurable on the altered nonlinear fuel cell model.




















S 2 .. ... ..

1.8

1.6

1.4

1.2

1 I I I I I
0 5 10 15 20 25 30
Time (sec)
Figure 4-29. The sFF with the LQG observer and the LQR feedback compared to the
GPC with the LQG observer control strategy on the altered nonlinear fuel
cell model.





















S 2-

S1.8 -

O
1.6

1.4

1.2

1 I---
0 5 10 15 20 25 30
Time (sec)
Figure 4-30. The sFF with the LQG observer and the LQR feedback compared to the
GPC with the 4 observers control strategy on the altered nonlinear fuel cell
model.






















C 2

1.8 -

1.6

1.4

1.2

1 I I
0 5 10 15 20 25 30
Time (sec)

Figure 4-31. The sFF and dFF strategies and the GPC control strategy, performance
compared when applied on the unaltered nonlinear fuel cell model.
























1.8
O
1.6

1.4

1.2

1I I I I I
0 5 10 15 20 25 30
Time (sec)


Figure 4-32. The sFF and dFF strategies and the GPC control strategy with the LQG
observer, performance compared when applied on the unaltered nonlinear
fuel cell model.




















2


1.8

1.6

1.4

1.2

1I I I I I
0 5 10 15 20 25 30
Time (sec)

Figure 4-33. The performances of the sFF and dFF strategies and the GPC with the 4
observers control strategy compared when applied on the unaltered
nonlinear fuel cell model.



























1.8 -
2




1.6

1.4

1.2


0 5 10 15 20 25 30
Time (sec)

Figure 4-34. The sFF and dFF strategies and the GPC control strategy, performance
compared when applied on the altered nonlinear fuel cell model.

























1.8

O
1.6

1.4

1.2

1 I I I I
0 5 10 15 20 25 30
Time (sec)
Figure 4-35. The sFF and dFF strategies and the GPC control strategy with the LQG
observer, performance compared when applied on the altered nonlinear fuel
cell model.



















2 .. .


1.8
O
1.6

1.4

1.2

1 I I I I I
0 5 10 15 20 25 30
Time (sec)
Figure 4-36. The performance of sFF and dFF strategies and the GPC with 4 observers
control strategy compared when applied on the altered nonlinear fuel cell
model.









CHAPTER 5
CONCLUSIONS AND PROPOSITIONS FOR FUTURE WORK

5.1 Conclusions

A Virtual Control Lab (VCL) with an inverted pendulum that can be utilized as a tool for

enhanced learning is described. The VCL can be used in conjunction with a process control

lecture for demonstrating various advanced concepts or can be used by students located in

geographically remote places. The animation module of the VCL allows the user to visually

observe the impact of the control design. The VCL can potentially minimize the problem of

scheduling laboratory time for physical equipment.

The problem of tracking the slope of ramp is examined. On the basis of results obtained, a

PI controller using the ITAE as the optimizing metric demonstrates the best results. The

controller performance is better than that of controllers proposed in literature. Tuning

relationships for PI and PI2 controllers, using three different optimizing metrics and 20,000

different plants is presented.

An elegant solution is proposed for the problem of fuel cell control by implementing a

GPC strategy which incorporates disturbance measurements to produce a manipulated variable.

The GPC design demonstrates better performance compared to control strategies available in

prior literature. The GPC controller results in zero offset in the performance variable in the

nonlinear model when the performance variable is measured. The GPC controller is also the best

performing controller when evaluated for model uncertainty. The controller exhibits zero offset,

showing its strength from a viewpoint of robustness, when employed on a modified model. In

the case when the performance variable is not measurable, a battery of observers is implemented

to estimate the value of the performance variable deliver the best result. However, the

robustness problems still exist.









5.2 Future Work

To validate the benefits and effectiveness, it is proposed that the VCL is tested as a pilot

teaching tool in process control classes taught at both undergraduate and graduate levels. The

feedback obtained from the instructors and students will be highly beneficial in improving and

incorporating features that could enhance the learning experience. It would be desirable to add

to the VCL a few more advanced control strategies such as the GPC approach.

For future work regarding the ramp tracking problem it is proposed to validate the

simulation results for physical setups that can take advantage of the tuning relationships

presented. It is also suggested to conduct a more comprehensive study to minimize the scatter in

the optimizing metric error for low 0/r values of the ratio, that is, for systems with little dead

time.

For future work on the fuel cell control problem it is proposed to evaluate the design of

observers which take model uncertainty into account. A battery of observers can be designed at

each operating point, with each observer corresponding to a linearized model obtained from

different values of parameters. Then a bumpless switching strategy could be designed to select

an appropriate state estimate to feed to the controller. The experimental implementation of the

control scheme in a physical fuel cell may be of significant value to confirm or refine the results

discussed in this dissertation.









APPENDIX A
OFFSET BETWEEN AUXILLIARY AND ORIGINAL RAMP

The final value theorem is used to calculate the shift between the original ramp and the

modified ramp. The transfer function for the process with time delay is given by

K
G = e (A-l)


and the transfer function for PI controller is given by


G =K, 1+- (A-2)


and the equation for ramp in the time domain is

r(t) = at (A-3)

Performing the Laplace transform of Eq. A-3, with the initial condition r(t = 0) = 0, yields


r(s) = (A-4)
s-

Using Eqs. A-i and A-2, it is possible to derive the following standard closed loop

relationship between the output y and the ramp r :

GGc
y(s) = r(s) (A-5)
I+GGC

Now consider the error

e(t) = r(t)- y(t) (A-6)

defined as the difference between the set point and output. Applying final value theorem

lime(t) = lims(r(s)- y(s)) (A-7)
t- i E s

and inserting Eq. A-4 and A-5 in Eq. A-7 yields











lime(t)= lims a
t-> s->o S2


GpGc a
1+GpG s2


(A-8)


a 1
or lime(t) = lim- (
t)co so s 1 + GpG,


Inserting Eq. A-i and Eq. A-2 in Eq. A-9 results in


lime(t) = lim (
t s s-KO K 1
1T+l e 1 Tj


Simplifying terms and applying the limits on the right hand side of Eq. A-10 yields

1
lime(t) = a (1
t)o 1
KKC
TI


Hence, lime(t) = (
(t- KKC


which is an analytical expression for the offset between the auxiliary and the original ramp.


A-9)


A-10)


A-11)


A-12)









APPENDIX B
OBSERVER DESIGN USING TRANSFER FUNCTION

The observer equations are given by

S= A + Bu + L(y C)

= C + Du

Eq. B-1 can be re-written as

= (A LC) + Bu + Ly

Performing Laplace Transform

sX(s) i(t = 0) = (A LC)X(s) + BU(s) + LY(s)

This results

(s A + LC)X(s) = BU(s) + LY(s) + (t = 0)

On further manipulation

i(s) = MBu(s) +MLy(s) + M(t = 0)

where M = (sI A +LC)1

Now if

S= C_

Z(s) = C.fBU(s) + C.LY(s) + CAf/(t = 0)

Z(s) = GMU(s) + GCY(s) + GjCoX(0)

where G,, = CB, G, = C,ML, and Ga O = CM.


(B-l)

(B-2)


(B-3)


(B-4)


(B-5)


(B-6)


(B-7)

(B-8)

(B-9)









LIST OF REFERENCES


1. T. F. Junge and C. Schmid, Web-based remote experimentation using a laboratory-scale
optical tracker, Proceedings of the American Control Conference, Chicago Illinois, June
2000.

2. D. Gillet, C. Salzmann, R. Longchamp, and D. Bonvin, Telepresence: an opportunity to
develop practical experimentation in automatic control education, European Controls
Conference, Brussels Belgium, July 1997.

3. D. Gillet, F. Geoffroy, K. Zeramdini, A. V. Nguyen, Y. Rekik and Y. Piguet, The
cockpit: An effective metaphor for Web-based experimentation in engineering education,
International Journal of Engineering Education, 19(3), August 2002, pp. 389-397.

4. D. Gillet, G. Fakas, eMersion: a new paradigm for Web-based training in engineering
education, International Conference on Engineering Education, Oslo, Norway, 2001,
84B-10.

5. A. Bhandari, M. H. Shor, Access to an instructional control laboratory experiment
through the World Wide Web, Proceedings of the American Control Conference,
Philadelphia, June 1998.

6. Ch. Schmid, T.I. Eikass, B. Foss and D. Gillet, A remote laboratory experimentation
network, 1st IFAC Conference on Telematics Applications in Automation and Robotics,
Weingarten, Germany, July 24-26, 2001.

7. H. Latchman, C. Salzmann, S. Thotapilly and H. Bouzekri, Hybrid asynchronous and
synchronous learning networks in distance education, International Conference on
Engineering Education, Rio de Janeiro, Brazil, 1998.

8. B. Armstrong and R. Perez, controls laboratory program with an accent on discovery
learning, IEEE Control Systems Magazine, February 2001, pp. 14-20.

9. J. W. Overstreet and A. Tzes, Internet-based client/server virtual instruments designs for
real time remote-access control engineering laboratory, Proceedings of the American
Control Conference, San Diego, 1999.

10. C. Schmid, The virtual control lab VCLab for education on the Web, Proceedings of the
American Control Conference, Philadelphia, June 1998.

11. O. D. Crisalle and H. A. Latchman, Virtual control laboratory for multidisciplinary
engineering education, NSF Award No. DUE-9352523, proposal funded by the National
Science Foundation under the Instrumentation and Laboratory Improvement Program,
(1993).

12. C. S. Peek; O. D. Crisalle; S. Depraz and D. Gillet, The virtual control laboratory
paradigm: Architectural design requirements and realization through a DC-motor
example, International Journal of Engineering Education, 21(6), 2005, pp. 1134-1147









13. National Instruments Corp., 11500 N. Mopac Expressway, Austin, TX 78759-3504,
USA, www.ni.com.

14. M. W. Dunnigan, Enhancing state-space control teaching with a computer-based
assignment, IEEE Transactions on Education,44(2), May 2001, pp. 129-136.

15. P. C. Wankat and F. S. Oreovicz, Teaching Engineering, McGraw-Hill, (1993).

16. A. M. Lopez; J. A. Miller, C. L. Smith and P. W. Murrill, Tuning controllers with error-
integral criteria, Instrumentation Technology, 14(11), November 1967, pp. 57-62.

17. C. Smith and A. Corripio, Principles and Practice of Automation Control. J. Wiley.
(1997).

18. W. Brogan, Modern Control Theory, 4th edn, Prentice-Hall, (1991).

19. P. W. Belanger and W. L. Luyben, Design of low-frequency compensator for
improvement of plantwide regulatory performance, Ind. Eng. Chem. Res., 36, 1997, pp.
5339-5347.

20. R. Monroy-Loperena, I. Cervantes, A. Morales and J. Alvarez-Ramirez, Robustness and
parametrization of the proportional plus double-integral compensator, Ind. Eng. Chem.
Res. 38, 1999, pp. 2013-2020.

21. C. S. Peek, Development of virtual control laboratories for control engineering education,
double integral controllers for ramp tracking and predictive control for robust control and
luenberger observers, PhD Candidacy Exam, Dept. of Chemical Engineering, College of
Engineering, University of Florida; 2006.

22. Mathworks Inc. MATLAB. Version 7.2.0.232, R2006a (2006).

23. S. Davis, Transportation Energy Data Book, 20th edn, U. S. Department of Energy,
(2000).

24. W. Grove, A small voltaic battery of great energy, Philosophical Magazine, 15, 1839, pp.
287-293.

25. J. Amphlett, R. Baumert, R. Mann, B. Peppley, P. Roberage and A. Rodrigues,
Parametric modelling of the performance of a 5-kW proton-exchange membrane fuel cell
stack, Journal of Power Sources, 49, 1994, pp. 349-356.

26. J. Amphlett, R. Baumert, R. Mann, B. Peppley and P. Roberage, Performance modeling
of the Ballard Mark IV solid polymer electrolyte fuel cell, Journal of Electrochemical
Society, 142(1), 1995, pp. 9-15.

27. D. Bemardi, Water-balance calculations for solid-polymer-electrolyte fuel cells, Journal
ofElectrochemical Society, 137(11), 1990, pp. 3344-3350.

28. D. Bernardi and M. Verbrugge, A mathematical model of the solid-polymer-electrolyte
fuel cell, Journal ofElectrochemical Society, 139(9), 1992, pp. 2477-2491.









29. J. Baschuk and X. Li, Modelling of polymer electrolyte membrane fuel cells with
variable degrees of water flooding, Journal of Power Sources, 86, 2000, pp. 186-191.

30. J. Eborn, L. Pedersen, C. Haugstetter and S. Ghosh, System level dynamic modeling of
fuel cell power plants, Proceedings of the American National Conference, 2003, pp.
2024-2029.

31. J. T. Pukrushpan, A. G. Stefanopoulou and H. Peng, Control of fuel cell power systems:
principles, modeling, analysis and feedback design, Advances in Industrial Control.
Springer-Verlag London Limited, (2004).

32. D. Seborg, T. Edgar and D. Mellichamp, Process Dynamics and Control, 2nd edn, Wiley,
(2004).

33. K. Ogata, Modern Control Engineering, 4th edn, Prentice Hall, (2001).

34. R. Stefani, B. Shahian, C. Savant and G. Hostetter, Design of Feedback Control Systems,
4th edn, Oxford University Press, (2002).

35. O. D. Crisalle, D. E. Seborg and D. A. Mellichamp, Theoretical analysis of long-range
predictive controllers, Proceedings of the American Control Conference, Pittsburgh,
Pennsylvania, 1989.

36. D. W. Clarke, C. Mohtadi and P. S. Tufts, Generalized predictive control part I. The
basic algorithm, Automatica, 23(2), 1987, pp. 137-148.

37. D. W. Clarke, C. Mohtadi and P. S. Tufts, Generalized predictive control part II.
Extensions and interpretations, Automatica, 23(2), 1987, pp. 149-160.

38. D. W. Clarke and C. Mohtadi, Properties of generalized predictive control, Automatica,
25(6), 1989, pp. 859-875.









BIOGRAPHICAL SKETCH

Vikram Shishodia was born in Delhi, India. He graduated with a B.Tech. in chemical

engineering from the Indian Institute of Technology Delhi in 1992. Mr. Vikram Shishodia

joined the graduate program at the University of Florida in 1993. He graduated with an M.S. in

materials science and engineering in 1996. After developing a successful career as a process

engineer with Intel Corporation for nine years he returned to academia to pursue higher degrees

in chemical engineering at the University of Florida. From 2005 to 2008 he was a graduate

student and simultaneously worked as Assistant Director for the Division of Student Affairs at

the College of Engineering of University of Florida. He graduated with an M.S. in chemical

engineering in 2003 and a Ph.D. in chemical engineering from the University of Florida in 2008.





PAGE 1

1 ENHANCED CONTROL PERFORMANCE AND APPLICATION TO FUEL CELL SYSTEMS By VIKRAM SHISHODIA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2008

PAGE 2

2 2008 Vikram Shishodia

PAGE 3

3 To Carmen

PAGE 4

4 ACKNOWLEDGMENTS I would like to express my deep est gratitude to m y advisor Dr O. Crisalle for his support and guidance without which this work would not ha ve been possible. I thank the members of my supervisory committee, Dr. H. Latchman, Dr. G. Hoflund, Dr. W. Lear, and Dr. S. Svoronos, for their guidance and serving on my supervisory committee. I thank my colleagues in the research group who provided insightful conversations on my research topics, besides being gr eat friends. I would especially like to thank Christopher Peek for providing the sample code for ramp track ing which expedited the progress on the problem significantly. I also thank him fo r all the insightful discussions. I thank my parents for their love, support and encouragement that they ha ve given me throughout my life and during the completion of this work. I would like to express my deepest gratit ude to my spiritual teacher Gurumayi Chidvilasananda who has been there for me during ev ery step of my life. Finally, I wish to thank my wife and kids, who have been very supportiv e, loving and understandi ng during this journey.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4LIST OF TABLES................................................................................................................. ..........7LIST OF FIGURES.........................................................................................................................8ABSTRACT...................................................................................................................................11CHAPTER 1 INTRODUCTION..................................................................................................................132 VIRTUAL CONTROL LABORATORY............................................................................... 152.1Introduction................................................................................................................... 152.2Objective.......................................................................................................................172.3Inverted Pendulum........................................................................................................172.4Control Design..............................................................................................................202.5Realization of an Inverted-Pendulum VCL.................................................................. 252.6Conclusions...................................................................................................................293 PI AND PI2 CONTROLLER TUNING FOR TRAC KING THE SLOPE OF A RAMP....... 383.1Introduction and Background........................................................................................383.2Problem Statement and Approach................................................................................. 403.3Results and Discussion.................................................................................................. 433.3.1Tuning Parameters of Controllers..................................................................... 433.3.2Comparison of the Performance of the PI and PI2 Controllers......................... 443.3.3Comparison of Metrics...................................................................................... 463.3.4Comparison of PI (ITAE) Controller with Literature Precedents..................... 473.3.5Local Minima versus Global Minima...............................................................483.4Conclusions...................................................................................................................484 GENERALIZED PREDICTIVE CONTROL FOR FUEL CELLS ....................................... 614.1Introduction................................................................................................................... 614.2Fuel Cell System Background....................................................................................... 614.3Objectives of the Research............................................................................................644.4Fuel Cell Model............................................................................................................644.5Literature Precedents fo Fuel Cell Control Designs..................................................... 664.5.2Feedforward Strategy........................................................................................674.5.2.1Static feedforward controller..............................................................674.5.2.2Dynamic feedforward controller......................................................... 68

PAGE 6

6 4.5.3Combination of Static Feedforward with Optimal Feedback Controllers........ 694.5.3.1Case where the performance variable is measurable.......................... 694.5.3.2Case where the performance variable is not measurable.................... 714.6Generalized Predictive Control..................................................................................... 724.7Battery of Observers.....................................................................................................794.8Simulation Studies and Results..................................................................................... 814.8.1Generalized Predictive Control Results............................................................ 824.8.1.1Case where the performance variable is measured............................. 824.8.1.2Performance variable not measured.................................................... 834.8.2The GPC Approach Evaluated for Robustness................................................. 854.8.2.1Case where the performance variable is measured............................. 864.8.2.2Case where the performance variable is not measured....................... 864.8.3Comparison of the GPC Strategy with Prior Control Designs.......................... 884.8.3.1Case where all states are measured-sFF with LQR feedback control 884.8.3.2Case where all states are not measured-observer design.................... 894.8.3.3Comparison of controller performance with respect to robustness....904.8.4Feedforward Control Designs...........................................................................924.8.4.1Case of original model........................................................................ 924.8.4.2Case of model uncertainty.................................................................. 944.9Conclusions...................................................................................................................955 CONCLUSIONS AND PROPOSITI ONS FOR FUTURE WORK ..................................... 1325.1Conclusions.................................................................................................................1325.2Future Work................................................................................................................133APPENDIX A OFFSET BETWEEN AUXILLIAR Y AND ORIGINAL RAMP ........................................ 134B OBSERVER DESIGN USI NG TRANSFER FUNCTION ..................................................136LIST OF REFERENCES.............................................................................................................137BIOGRAPHICAL SKETCH.......................................................................................................140

PAGE 7

7 LIST OF TABLES Table page 3-1 The PI2 controllers optimized tuning parameters linear least square fit equations............ 523-2 The PI controllers optimized tuning para meters linear least square fit equations............. 533-3 Values of the plant parameters used co mpare the performance of the controllers............ 53

PAGE 8

8 LIST OF FIGURES Figure page 2-1 Inverted pendulum.......................................................................................................... ...312-2a Front panel of the VCL wh ere all states are measured...................................................... 322-2b Front panel of the VCL showing observer......................................................................... 332-3 Interaction panel of inverted pendulum VCL....................................................................342-4 Controller tab of the navigation panel...............................................................................352-5 Analysis tab of the navigation panel..................................................................................362-6 Simulation tab of the navigation panel.............................................................................. 373-1 Ramp r and auxiliary ramp ra with constant slope, ........................................................493-2 Closed loop transfer function repr esentation of plant and controller................................493-3 The PI2 controllers optimal tuning parameters using ITAE (A, B), IAE (C, D), and ISE (E, F) as the optimizing metric................................................................................... 503-4 The PI controllers optimal tuning parame ters, using ITAE (A, B), IAE (C, D), and ISE (E, F) as the optimizing metric................................................................................... 513-5 The PI2 and PI controllers ramp tracking a nd slope tracking performance using the optimal ITAE control parameters...................................................................................... 543-6 The PI2 and PI controllers ramp tracking a nd slope tracking performance using the optimal IAE control parameters......................................................................................... 553-7 The PI2 and PI controllers ramp tracking a nd slope tracking performance using the optimal ISE control parameters......................................................................................... 563-8 The ITAE, IAE, and ISE metrics comparison for three plants, usi ng the PI controllers A), B), and C).....................................................................................................................573-9 The ITAE, IAE, and ISE metrics co mparison for three plants, using the PI2 controllers A), B), and C).................................................................................................. 583-10 The PI controllers tuned using the ITAE metric compared with Belanger and Luyben and Peeks controllers for three plants A), B), and C)....................................................... 593-11 Contour plots for the PI c ontroller tuned using the ITAE metric for the three plants A), B), and C).....................................................................................................................60

PAGE 9

9 4-1 Schematic of fuel cell system............................................................................................ 964-2a Fuel cell system showing input u, disturbance w and outputs z1, z2, y1, y2, y3...................974-2b Fuel cell system showing sFF with feedback controller.................................................... 984-3 Matrices defining the LTI model fo r the fuel cell model excluding sFF........................... 994-4 Matrices defining the LTI model for the fuel cell including sFF...................................... 994-5 The sFF control configura tions for fuel cell system........................................................1004-6 The dFF controller: (a)Schematic diagra m, and (b)transfer function representation...... 1014-7 The sFF schematic with feedback controller................................................................... 1024-8 The GPC design in feedback block diagram....................................................................1034-9 Disturbance profile used for simulation purposes........................................................... 1044-10 The GPC control strategy implementati on on the nonlinear fuel cell model in the case when the controlled va riable is measured................................................................ 1054-11 The GPC feedback with four observe rs control scheme implementation on the nonlinear fuel cell model................................................................................................. 1064-12 The Norm of errors from the battery of observers........................................................... 1074-13 The switching pattern of the battery of observers............................................................ 1084-14 Final voltage to the compressor....................................................................................... 1094-15 Observer 1, error between m easured and estimated values............................................. 1104-16 Observer 2, error between m easured and estimated values............................................. 1114-17 Observer 3, error between m easured and estimated values............................................. 1124-18 Observer 4, error between m easured and estimated values............................................. 1134-19 The GPC feedback with three observers control scheme implementation on the nonlinear fuel cell model................................................................................................. 1144-20 The GPC feedback with two observers control scheme implementation on the nonlinear fuel cell model................................................................................................. 1154-21 The GPC feedback with one observe r control scheme implementation on the nonlinear fuel cell model................................................................................................. 116

PAGE 10

10 4-22 The GPC control strategy implementati on on the nonlinear fuel ce ll model with a parameter changed from the value used for control design............................................. 1174-23 The GPC controller with the LQG obser ver control strategy implementation on the altered nonlinear fuel cell model...................................................................................... 1184-24 The GPC controller with the 4 observers control strategy implementation on the altered nonlinear fuel cell model...................................................................................... 1194-25 Comparison of the GPC control strategy w ith the sFF controller combined with LQR feedback strategy on the unaltered nonlinea r fuel cell model when the performance variable is measurable...................................................................................................... 1204-26 The sFF with the LQG observer and LQR feedback, compared to GPC with the LQG Observer control strategy implementation on the unaltered nonlin ear fuel cell model when the performance variab le is not measurable........................................................... 1214-27 The sFF with the LQG observer and LQR feedback, compared to GPC with the 4 observers control strategy implementation on the unaltered nonlin ear fuel cell model when the performance variab le is not measurable........................................................... 1224-28 The sFF with the LQR feedback, compared to GPC, when performance variable is measurable on the altered nonlinear fuel cell model....................................................... 1234-29 The sFF with the LQG observer and the LQR feedback compared to the GPC with the LQG observer control strategy on th e altered nonlinear fuel cell model................... 1244-30 The sFF with the LQG observer and the LQR feedback compared to the GPC with the 4 observers control strategy on the altered nonlinear fuel cell model....................... 1254-31 The sFF and dFF strategies and the GPC control strategy, performance compared when applied on the unaltered nonlinear fuel cell model................................................ 1264-32 The sFF and dFF strategies and the GPC control strategy with the LQG observer, performance compared when applied on the unaltered nonlinear fuel cell model.......... 1274-33 The performances of the sFF and dFF st rategies and the GPC with the 4 observers control strategy compared when applied on the unaltered nonlinear fuel cell model..... 1284-34 The sFF and dFF strategies and the GPC control strategy, performance compared when applied on the altered nonlinear fuel cell model.................................................... 1294-35 The sFF and dFF strategies and the GPC control strategy with the LQG observer, performance compared when applied on the altered nonlinear fuel cell model.............. 1304-36 The performance of sFF and dFF strategi es and the GPC with 4 observers control strategy compared when applied on th e altered nonlinear fuel cell model...................... 131

PAGE 11

11 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ENHANCED CONTROL PERFORMANCE AND APPLICATION TO FUEL CELL SYSTEMS By Vikram Shishodia May 2008 Chair: Oscar D. Crisalle Major: Chemical Engineering The inverted-pendulum virtual control la b, a simulation environment for teaching advanced concepts of process control, is desi gned using the LabVIEW soft ware tool. Significant advantages of using this simulation tool for pe dagogical purposes include avoiding the potential issue of schedule conflicts for securing equi pment-access time in a physical laboratory and providing a learning resource that becomes accessible to students lo cated in remote geographical places. A set of tuning relationships are proposed for standard proportional-in tegral controllers and proportional double-integral controllers for the purpos e of tracking the slope of a ramp trajectory. Three different performance metrics are investigated to serve as the criteria for optimality, and a numerical optimization procedure is used to minimize each metric over 20,000 different plants. The proportional integral controlle r with tuning parameters select ed to optimize value of the integral of the time-weighted absolute error is recommended for tracking the slope of a ramp trajectory. A generalized predictive contro l (GPC) strategy is proposed for a fuel cell system, where the controller incorporates a measured disturbanc e in the control design. The control objective is to maintain oxygen excess ratio at a prescribed constant value. The performance of the GPC

PAGE 12

12 control design is compared with that of the controllers propo sed in literature for various scenarios including model uncertainty. The GPC controller has zero offset when the performance variable is measured and performs better than competing designs offered in the literature. The GPC controller is also robust w ith respect to model uncertainty. A battery of observers with a switching stra tegy is proposed for estimating the value of the performance variable when it is not measure d. The GPC controller with a ba ttery of observers has no offset demonstrating better performance than analogous de signs proposed in literature. However, the control performance is not robust when the estimat or battery is used and linear models used for observer design are uncertain give n that the response offset is not completely eliminated.

PAGE 13

13 CHAPTER 1 INTRODUCTION Issues relevant and critical to process c ontrol are discussed in this study. Chapter 2 investigates the design and im plementation of a virtual control lab (VCL) for The Inverted Pendulum problem. The LabVIEW software is used as the platform for simulating the Inverted Pendulum model. The objective is to have a vi sual computer based application by virtue of which advanced control concepts can be shared and taught to the audience who are primarily students studying control th eory and its applic ations. The VCL is designed in a manner such that various scenarios for the implemen tation of the controllers can be achieved. The user is given the choice of operating the application with the system in open-loop or closed-loop configuration. The controller can be tuned manually by the us er or use the tuned control parameters computed by the specific control algorithm. The user is allowed to alter the values of the poles for the closed loop system and see it s visual impact by simulation performed by the VCL. The VCL provides the opportunity to be ope rated in the scenarios when all the controlled variables are measurable and also when all of them are not measurable. In the case when the performance variables are not measurable an observer is incorporated in the control design to estimate their value. The impact of all the cha nges performed in the VCL are displayed visually by the animation of the inverted pendulum system. This key feature of the VCL allows the user to see the visual impact of changing differe nt components of control system and hence facilitating the process of learning. Chapter 3 discusses the problem of controll er design tuning for tr acking the slope of a ramp. The control objective is to place the outpu t of the system in a linear zone parallel to a ramp trajectory. A first order system with time de lay is considered for this study. Two kinds of controllers are used for this study namely, proportio nal-integral and propor tional double-integral.

PAGE 14

14 Both the controllers serve the purpose of positioning the system in the desired linear zone which is parallel to a given ramp prof ile. There is an offset with respect to the ramp observed when only one integrator is used. Zero offset with th e ramp is observed when two integrators are used in the controller. In both cases the control objective is met which is to track the slope of the ramp trajectory.. Three different metric are employed to evaluate the performance of the controllers. The MATLAB pl atform in conjunction with SIMULINK module is used for acquiring the optimized controller parameters. Chapter 4 discusses a generalized predictive co ntrol (GPC) strategy prop osed for a fuel cell system. The control objective is to regulate the value of the pe rformance variable i.e., the oxygen excess ratio at a desired value. The perf ormance of the GPC control design is compared with that of controllers proposed in prior literature. Various scenarios are considered, including the cases of model uncertainty and unmeasured performance variable. The GPC controller exhibits zero offset in all cases when the perfor mance variable is measured, and also ensure zero or negligible offset when the performance vari able is estimated via a battery of estimators.

PAGE 15

15 CHAPTER 2 VIRTUAL CONTROL LABORATORY 2.1 Introduction There is a need for the developm ent of in ternet-based non-conventional pedagogical tools for delivering knowledge to students on various topics of study. The drive stems from the various advantages that these environments offer. First, these applications do not depend on the availability of a physical setup or facility to run experiments [1, 2]. They are also not limited in terms of the number of users who can access the application at any given time, as long as appropriate adjustments are done in the server side of application. There is also no adverse safety issue or concern of damaging expensive equipmen t when the product is not used correctly. Less training is required for the user to be able to run the tool. Compared to a traditional physical laboratory setup, in these virtual environments, there is more of an opportunity to be able to realize a physical system and introduce more advanced topics a nd see their effects on the system. A software application that simulates the behavior of a physical system, provides animation to depict how the system behaves, and provides an interface so that the us er can observe changes made on the system performance, is highly beneficial from a lear ning and educational standpoint. The application is referred to as a virt ual laboratories since the nature of the Lab or the application is virtual as it is a softwa re emulator of the physical plant and can be potentially used to remotely control actual phy sical equipment via web and networking [1, 3-6]. From the perspective of enhancing the learni ng experience, the virtual control lab (VCL) supports learning by all three modes, namely activ e, flexible, and discovery learning. In active learning, tools and material are made available to students so that they can use these resources to actively learn and reinforce th e theoretical concepts. Trad itionally, physical laboratories, equipment and experimental apparatus are provid ed to students to reinforce and test the

PAGE 16

16 understanding of the students comprehension of the theory. With limitation of available resources and costs associated with the overall management of logistics, at times this can be a challenging task, which can prove to be not only quite expensive, but also involve safety issues. In those circumstances the VCL can be an excellent solution. It doesnt n ecessarily need to be a complete replacement, but it can be used in conj unction with existing physical setups to promote active learning. Flexible learning provides opportunities to l earn class material when the students or instructor might be having challenges in terms of establishing meeting times, scheduling or location [7]. For instance, if an individual is a part-time student with the obligations of a fulltime student, that person might have challenges meeting lab times scheduled during regular work hours. The VCL is a most valuable tool to acco mmodate those circumstances. From the comfort of home and in a more appropriate time, that indi vidual can complete the exercises/material if the VCL is utilized. The VCL is flexible with respec t to the schedule and logistics limitations of an individual. The VCL also supports learning vi a discovery mode. In this scenario, an environment is provided where the student has minimal supervis ion or instruction [8]. The student is encouraged to learn by making changes and observ ing the impact of these changes. There are some significant challenges in implementing such a setup in the alternative scenario of a physical laboratory. Due to safety and cost consideratio ns, facilities that promote discovery learning are few. A VCL designed for the purpose, again, prove s to be an excellent resource to implement safe and cost-effective learning via discovery. There are various implementations of virtual an d remote labs reported in the literature. There are several World Wide Web based labs which foster learning by different modes [3, 9-

PAGE 17

17 11]. Most of these virtual labs, however, have a few shortcomings in terms of their usage. Many do not provide a sufficiently high level of interact ivity with the user. Significant modifications need to be made to the program to implement a ny changes. Another disa dvantage that most of the current virtual labs have is that they are developed on a proprietary software platform. At times significant familiarity with th at software is needed to be able to utilize the application. 2.2 Objective The intent is to build a VCL m odule that treat s some advanced-level control concepts and serves as a pedagogical tool that overcomes shor tcomings that existing virtual labs pose from a learning perspective and user interface. The infrastructure created by Peek et al. is used for implementing an Inverted Pendulum VCL [12]. The inte nt is to develop an animated control module that reinforces advanced control concepts with a friendly user interface. Some examples of key control concepts illustrated in the VC L are linear state-space modeling, controllability, pole placement and observability analysis. Sec tions 2.3 and 2.4 discuss the Inverted Pendulum system, its dynamics and the associated c ontrol concepts. Sec tion 2.5 describes the implementation of the Inverted Pendulum system as a VCL using the LabVIEW software and its animation features [13]. Finally, conclusions fr om this effort are summarized in Section 2.6. 2.3 Inverted Pendulum The inverted pendulum considered consists of a spherical bob attached to a cart by a rod. A schematic diagram is given in Figure 2.1. The mass of the rod is assumed to be negligible. The rod is mounted by a hinge at the center of the cart. The input of the system is a horizontal force applied to the cart. The cart is free to move only along one coor dinate which is the horizontal z -axis. The pendulum is free to rotate 360 de grees with respect to the cart in the x-z plane where the x -axis is vertical. It is assumed that there is no friction between the pendulum and the cart at the hinge.

PAGE 18

18 The goal is to keep the pendulum in an uprig ht position by manipulating the value of the applied force. The system is inherently nonlinea r. To apply linear c ontrol theory, the dynamics must be linearized, and represented as a standa rd state-space realizat ion. Consequently, the system is linearized for small values of the a ngle that the pendulum ma kes with the vertical. The nonlinear equations describing the dynamics of the system are cossinsin sin 1 z2 2g l m f m M (2-1) and sin sincos cos sin 12 2g m Mm l m f m M l (2-2) where dz/dt z (2-3a) and /dt d (2-3b) where M is the mass of the cart, m is the mass of the pendulum bob, l is the length of the pendulum rod, g is the acceleration due to gravity, z is the horizontal position of the cart, is the angle that the pendulum make s with the vertical, and f is the force (contro l input) acting on the cart [14]. A linear state-space system is derived from Eqs. 2-1 2-3 by linearizing about an operating point ( ),,,, fzz where 0 z (2-4) 0 z (2-5) 0 (2-6) 0 (2-7) 0 f (2-8)

PAGE 19

19 The deviation variables for the linear state-spa ce model are zzx 1 (2-9) zzx 2 (2-10) 3x (2-11) 4x (2-12) ffu (2-13) After linearization about the point )0,0,0,0,0(),,,,( fzz (2-14) the resulting standard linear state-space model u bAxx (2-15) is given by the equation u Ml M Ml M)g(m M mg 1 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 x x (2-16) where the elements of the state vector x are distance (x1 = z ), velocity (x2 = z ), angle ( x3 = ), and angular velocity ( x4 = ), and where 0 )( 0 0 1 0 0 0 0 0 0 0 0 1 0 Ml gMm M mg A (2-17)

PAGE 20

20 Ml M 1 0 1 0 b (2-18) The control is u, which is the force acting on the cart. The standard output model xCy (2-19) relates the output y to matrix C and state vector x, where the output matrix C is the identity matrix 1000 0100 0010 0001C (2-20a) in the case where all four states are measured. For the case where only one state is measurable, the output matrix C adopts a row-vector form of all zeros, except for one entry that is unity at the location corresponding to the measured state. For the particular case where the state x1, namely the distance of the cart from its original horizontal position, is the only meas ured state, the output matrix C adopts the form 0001 C (2-20b) 2.4 Control Design When the system is in the unforced configura tion, a stability check done by calculating the eigenvalues of matrix A reveals that there is one eigenvalue th at lies in the open right half plane, implying that the system in its unforced state is unstable.

PAGE 21

21 This is a regulation problem, as the objective is to make the states evolve towards zero value. For the implementation of the controller, a test for controllabilit y needs to be performed to verify that the system is indeed controllab le. The requirement for controllability is that 0Qc ) det( (2-21) where bAbAAbbQ3 2 c (2-22) Using the definitions for A given in Eq. 2-5 and for b given in Eq. 2-6, it follows that 0 1 0 1 Ml M Ab 22 2 2)( 0 0 lM gMm lM mg bA 0 )( 022 22 3lM gMm lM mg bA (2-23) Hence, 0 )( 0 1 )( 0 1 0 0 0 1 0 1 022 22 2 22lM gMm Ml lM gMm Ml lM mg M lM mg McQ (2-24) Obviously matrix cQ in Eq. 2-11 is of full rank, thus 0 )det( cQ (2-25) which implies that the system Eq. 2-4 is indeed controllable. The analysis of controllability presented here is found in standard references [32-34]. Two scenarios are considered: 1. All states are measured. 2. Some states are not measured.

PAGE 22

22 In the case of the first scenario in which all states are measured, a full-state feedback approach is used in the form of th e proportional state feedback control law Fx u (2-26) where F is a proportional gain used to address the regulation problem. To determine an appropriate value for matrix F first substitute the value of u given by Eq. (2-26) into the state space e quation Eq. (2-15) leading to ) ( FxbAxx (2-27) The standard solution to Eq. 2-14 is given by the Variation of Parameters formula as te)( 0 BFAxx (2-28) where 0x is the vector of the initial value of the state vector x [36]. Ackermans pole placement algorithm is employed for computing the value of the matrix F that places the poles of the A-BF system in the desired location [36]. In the case of the second scenario, in which a ll states of the system are not measurabed, a Luenberger Observer is incorporated in the contro ller to estimate the value of the states. Before carrying out an observer design, a check is performed to verify if the system is observable when only one state is measured. The first state, the di stance of the cart from the original position, is the only state that is assumed to be measurable. For the system to be observable, the condition 0 )det( oQ (2-29) should be satisfied, where the observability matrix oQ is given by 3 2CA CA CA C Qo (2-30) Matrix A is defined by Eq. (2-17) and matrix C by Eq. 2-20b. The expressions

PAGE 23

23 0010 CA (2-31) 0 002M mg CA (2-32) and M mg 0003CA (2-33) can be used to readily build the observability matrix oQ described by Eq. 2-17, yielding M mg M mgo000 0 00 0010 0001 Q (2-34) Since matrix oQ is diagonal, its determinant is simply the product of the diagonal terms. Hence, 0 det2 22 M gmoQ (2-35) Given that the determinant is nonzero, it follows that oQ is of full rank, and therefore the system is observable. The analysis of observability presented here is also found in standard references [34-36]. The standard equations for observer design u bAxx (2-36) ) ( yyLbxAx u (2-37) produce estimated states x and estimated outputs xCy as a function of the measured system output Cxy, and the Luenberger gain L in Eq. 2-37. The control input

PAGE 24

24 xF u (2-38) is used to place the poles of Eq. 2-36 at the desired locations. The error xx (2-39) is defined as the difference between the actual values of state vector x and estimated values of state vector x Hence, the derivative of the error xx (2-40) is computed by differentiating Eq. 2-39. Substitu ting Eq. 2-36 and Eq. 2-37 in Eq. 2-40 yields LC) (A (2-41) Invoking now the Variation of Parameters form ula the solution to th e differential equation Eq. 2-41 is te)( 0 LCA (2-42) where ) 0( )0(0xx is the initial value of the error ) 0( x is an initial guess of the value of the estimated state vector, and )0( x is the initial value of the state vector The poles of matrix A-LC should lie on the open half pl ane for the value of error to evolve to zero. The poles are placed at the desired location by an appropriate choice of L, which for low-order systems as the one considered here can be easily done via Ackermans Pole Placement algorithm. Linear quadratic regulator control design. A Linear quadratic control regulator (LQR) control design is implemented in the VCL. The LQR control law Kx u (2-43) is obtained by minimizing the cost function dtuRuu uJT T T)2 ()(0 Nx xQx (2-44)

PAGE 25

25 with respect to u. The weighting matrix Q must be symmetric positive semi-definite, and R symmetric positive definite. The weighting function Nis specified to be zero. For a linear state space system u BAxx (2-45) the solution to the minimization of the cost f unction results in the steady-state Riccatti equation [36] 0Q)NS(BN)R(SBSASAT T1 T (2-46) The acceptable solution to Eq. 2-46 is a positive definite matrix S which is then used to specify K from the expression )NS(BRKT T1 (2-47) Since in this case 0 N, therefore SBRKT1 (2-48) 2.5 Realization of an Inverted-Pendulum VCL The LabVIEW software and a VCL infrastructure proposed by Peek et al., is used for the implementation of the Inverted Pe ndulum VCL [12, 13]. Previous software-based control-tools for the inverted-pendulum system reported in the literature have significant value, but the VCL developed in this study has a number of additional desirable pedagogical featur es [14]. Initially, stand-alone VIs and subVIs are generated using LabVIEW software for different components of the design before integrating them as a part of a monolithic VCL. There are several reasons why National Instruments LabVIEW software is us ed for constructing the VCL. The ease of structuring and maintaining a VCL is significantly high in this software. The LabVIEW software has built-in features for deploying applications on the web. The software also has toolkits specifically designed for control e ngineering. Implementation of a VCL using

PAGE 26

26 LabVIEW does not rely on support from other soft ware packages, as would be the case if some higher-level language is used to implem ent the same features found in VCL. Figure 2-2 shows the front panel of VCL as it ap pears to a student user. The key elements are the Animation and Interaction Panels, respec tively, located on the top and bottom-left areas of the front panel. These two are very critical components of the VCL as the user makes most modifications in the plant and controller set up in the Interaction pa nel and instantaneously observes an animated result describing the plan t and states in the Animation Panel. The Animation Panel has a two-dimensional graphic re presentation of an inve rted pendulum. When the VCL is operated, the animated cart responds to the control input by moving to the left or right and causing a pendulum swing. The third panel is the Navigation Panel on the bottom-right area of the front panel. The Navigation Panel ha s five tabs (Information, Plant, Controller, Analysis and Simulation), which provide various pieces of information about the VCL. More information is given about these thre e panels in the ensuing subsections. Animation, interaction and navigation panel. The animation panel plays the role of providing a visual representation of the plant, namely an inverted pendulum. Any changes that are made to the inverted pendulum mounted on th e cart are visually depicted in the Animation Panel. The user has the ability to make changes to the plant and controller in the Interaction Panel. The user can adjust plant parameters, in itial condition of the states of the inverted pendulum and assign the different values to control parameters to the controller of choice. The user also has the ability to run the plant in Manu al or in Auto mode. Figure 2-3 depicts some of the various modes that the user can confi gure parameters in the Interaction Panel.

PAGE 27

27 The user has to specify the initial states of the pendulum (position, velocity, angle of the pendulum with the vertical and the angular velo city of the pendulum). The Animation Panel constructs the visual representa tion of the inverted pendulum ba sed on the information that the user provides. The user has th e flexibility of running the VCL in the following two scenarios: (1) all states are measured, or (2) only one state is measured. Based on the choice of the user, an implementation of the correspondi ng controller is given. When Manual F Control is in the Off position, the user is allowed to c hoose the poles for the closed loop matrix A-BF and the value of matrix F is calculated from Ackermans pole placement algorithm. The user can immediately see the impact of poles chosen on the stability of the inverted pendulum in the Navigation Panel under the Analysis tab. When the Manual F Control is in theOn position, the user has the ability to choose the values of the elements of matrix F. When the controller is operated in Luenberger Observer mode, as show n in Figure 2-, the user has to provide the desired poles for matrix A-LC. The only state that can be measured in this mode is the position of the cart. When the cont roller is in Off mode, i.e., the system is in open loop configuration with no feedback, the Analysis tab of the Navigation Panel show s that the pendulum is in an unstable configuration, which is as certained by the fact that there is one eigenvalue of the system in the open right half plane. Th e Interaction and Animation panels provide a suite of options and visual representation for the user. The Navigation panel is located on the bottom-ri ght area of the front panel of the VCL. The Navigation panel has five tabs entitled: (1) Help, (2) Plant, (3) Controller, (4) Analysis, and (5) Simulation. These tabs provide pertinent and critical information about the VCL to the user. The Help tab, when clicked on, provides genera l information about the op eration of the VCL. The user can access the Help tab without having to leave the VCL. The Help tab displays an

PAGE 28

28 embedded PDF file. The Plant and Controller ta bs are also embedded PDF files which provide information about the dynamics of the plant (inve rted pendulum) and the controller (proportional state feedback and Luenberger Observer). The nonlinear equations and linear state space system for the inverted pendulum are explained in the Plant tab of the Navigation Panel. Figure 2-3 shows the Plant tab of the Navigation Panel. Th e Controller tab provides information about pole placement and various other aspects of control de sign for inverted pendulum VCL, as shown in Figure 2-4. The Analysis tab, shown in Figure 2-5, has information about the tools and graphs that are used in control theory. This tab has information about the transfer function, location of poles and zeros in complex plane and Bode plot (freque ncy response). The Simulation tab, shown in Figure 2-6, shows plots of the re sults of numerical simulations describing the states (position, velocity, angle and angular veloci ty) and inputs as a function of tim e. Since this is a regulation problem, when stable choices of eigenvalues ar e given, using the lin ear state space as the dynamic model, all states converge to the value zero, regardless of the choice of initial state vector. The Simulation tab provides the real time curves of all the states as a function of time. The Runge-Kutta integration algor ithm is employed to compute the numerical response of the plant to the controlling input. When the controller is toggled between On and Off modes (closed-loop and open-loop behavior, respectivel y) in the Interaction panel by the user, the impact of that change on the value of states is depicted immediately in the Simulation tab. The LQR control strategy, as shown in Figure 2-, employed in the VCL gives the user the opportunity to implement different control choices, such as va rying the weighting on different elements of the cost function and displaying the corresponding LQR gain.

PAGE 29

29 2.6 Conclusions A VCL for the control of an Inverted Pendulum is described. The Inverted Pendulum is a classic example of illustrating state-space model representations and demonstrating the classical control concepts of controllabili ty and observability. The anima tion features of the VCL provide a visual description of how an inverted pendul um responds as a function of the input force applied. The user is given the opportunity to ru n the VCL in different modes, such as in openloop and closed-loop configur ations of the system. The VCL can be utilized as a tool for e nhancing learning. The three most widely recognized learning modes (active le arning, flexible learning, and le arning via discovery) can be easily executed using this VCL module. The mo dule can be used in conjunction with a process control lecture for demonstrating various concepts. The animation capabilities allows the user to see the impact of every change that is made to the control configuration. The Analysis tab also demonstrates that the open-loop configuration (u nforced system) is unstable, as one of the eigenvalues is in open right half plane. The user is given the c hoice of choosing the poles for the system and noticing its impact on the plant. Th e user has the choice of running the VCL in two modes: (1) all states measured or (2) only one state is meas ured and implementation of a Luenberger Observer. This utilization of th e VCL supports active learning of the control material. The VCL can also be presented before or after a conventional lectur e. In the first case, the users motivation to learn about theory presente d in class would be enhanced as they have the opportunity to develop and experi ence with the VCL. In the later case, af ter the lecture is delivered, an interaction with the VCL would se rve as an excellent tool for reinforcing the concepts taught in class. There is significant value in learning an abst ract concept taught in lecture and being able to relate to the concept by virtue of seeing the animation and graphs from

PAGE 30

30 simulation. In an ideal scenario the VCL should be used in all three modes (before, during, and after lectures). The tool can also be used to support group activ ities in a class for l earning control theory and doing control homeworks [15]. The modular f eature of the VCL can be taken advantage of by implementing sequential learni ng. In this mode, different versions of same VCL are progressively given to the user as the class progresses. Each successive version makes more features of the controller available to the user, hence helping students appr eciate and learn faster the concepts as they are progr essively taught in the class. To validate the benefits and effectiveness, it is proposed that the VCL should be used as a pilot teaching tool in process cont rol classes. The feedback obtai ned from the students would be highly beneficial in optimizing and incorpora ting features that could enhance the learning experience.

PAGE 31

31 mg l f z Mg x Figure 2-1. Inverted pendulum.

PAGE 32

32 Figure 2-2a. Front panel of the VC L where all states are measured.

PAGE 33

33 Figure 2-2b. Front panel of the VCL showing observer.

PAGE 34

34 Figure 2-3. Interaction pane l of inverted pendulum VCL.

PAGE 35

35 Figure 2-4. Controller ta b of the navigation panel.

PAGE 36

36 Figure 2-5. Analysis ta b of the navigation panel.

PAGE 37

37 Figure 2-6. Simulation tab of the navigation panel.

PAGE 38

38 CHAPTER 3 PI AND PI2 CONTROLLER TUNING FOR TRAC KING THE SLOPE OF A RAMP 3.1 Introduction and Background In certain applications tracking the slope of the ramp is more critical than tracking the ramp itself. It is not unusual to encounter applications where the set point is the slope of a ramp trajectory. For instance, while growing thin films on a substrate, it is desired that the temperature of the substrate in the reactor increases at a steady rate (i.e., following a trajectory with a specified slope). In these kinds of applications, it becomes crucial to adopt the appropriate choice of controller type with effective values fo r the tuning parameters and an appropriate metric to ensure adequate performa nce. Simplicity of tuni ng relationships plays a critical role in the implementation of a controlle r. Most successful t uning relationships have been developed via simulation [16, 17]. The cost function normally used involves the feedback error, which is the difference between the set point and out put of the plant. A proportional-only controller lead s to a steady state offset with respect to step changes in set point. An integrator needs to be incorporated in the cont roller to eliminate the offset. Similarly, for a ramp set point, a proportional-inte gral (PI) controller is not sufficient to remove the steady state offset. In this case a contro ller with two integrators, that is a proportional double-integral (PI2) controller, is needed to remove the o ffset [18]. However, a PI controller is sufficient to track the slope of the ramp trajectory. Belanger and Luyben proposed a proportional-integral-double integral controller relating the tuning parameters to the ultimate gain and ultimate period of th e plant [19]. Alvarez-Ramirez et al. extended the work of Belanger and Luyben and coined the acronym PI2 [20]. Peek examined three different versions of PI2 controllers for tracking ramp set point [21].

PAGE 39

39 This study investigates the problem of controller tuning for tracking the slope of a ramp signal as depicted in Figure 31. The intent is to design a controller that leads the output trajectory to follow a line parallel to the ramp. It is also critical to minimize transients. The control performance is deemed as poor when the system experiences large deviations during the transients. To investigate the problem, a first order syst em with time delay is adopted as the plant model. Since the set point is a ramp, integral action needs to be incorporated in the control design. Two kinds of controllers are used for this study, namely a proportional-integral (PI) controller and proportion al double-integral (PI2) controller. Both contro llers serve the purpose of placing the system output on the desired line, parallel to the ramp set point. However, there is an offset with respect to the original ramp trajectory when PI controllers are used because only one integrator is included in the control scheme. To identify the appropriate indicator to meas ure the performance of the controller with respect to the system and objective in question, th ree different metrics are used: integral of the time-weighted absolute error (ITAE), integral of the absolute error (IAE), and integral of the square of error (ISE). Each of the controllers is tuned to minimize the metric value for a given set of plant parameters. The Simplex optimizati on routine is used to acq uire tuning parameters via the minimization of the metric adopted. The MATLAB platform in conjunction with the SIMULINK module is used for conducting the simula tions. A set of optimal tuning relationships for controllers to track the slope of a ramp for 20,000 different plants are presented. The performance of all the controllers is evaluated an d results are compared with the prior work of Belanger and Luyben. The performance of the cont rollers is also compared with that of those proposed by Peek.

PAGE 40

40 3.2 Problem Statement and Approach The objective is to use PI and PI2 controller to make a firs t-order plant with time delay follow the slope of a ramp. The plant is represented by the transfer function s pe s K sG 1 ) ( (3-1) where K is the gain, is the time constant, and is the time delay of the plant. The input to the plant is denoted as u and the output as y. The transfer function representation of the PI controller is s KsGi c c1 1)( (3-2) and the transfer function for the PI2 controller investigated here is 21 1)( s KsGi c c (3-3) where cK is the proportional gain, andi is the integral-action time constant. The plant and the controller are configured in the closed-loop arrangement shown in Figure 3-2. The ramp function serving as the set poi nt in Figure 3-2 is given by ttr )( (3-4) where t is the time and the constant slope is taken as 1 Three error metrics are used for tuning the c ontrollers, namely, the integral of the timeweighted absolute error (ITAE), the integral of the absolute error ( IAE ), and the integral of the square of error (ISE ), respectively defined by the integral equations f ft tdttytrt ITAE0)()( lim (3-5)

PAGE 41

41 f ft tdttytr IAE0)()( lim (3-6) f ft tdttytr ISE0 2)()( lim (3-7) where ) (ty is the plant output and ft is the extent of time over which the metric is computed. When the closed loop uses a PI controller, the output of the system exhibits a steady state offset ssewith respect to the original ramp charact erized through the analytical expression c i ssKK e (3-8) the derivation of which is given in the APPEND IX using Final Value Theorem. A modified or auxiliary ramp ) (tra is defined via the relationship ss aetrtr )() ( (3-9) Figure 3-1 shows the ramp and th e auxiliary ramp trajectories. The corresponding auxiliary error metrics for PI controller, modi fied from Eq. 3-4, Eq. 3-5 and Eq. 3-6 are given by f ft a t adttytrt ITAE0)()( lim (3-10) f ft a t adttytr IAE0)()( lim (3-11) f ft a t adttytr ISE0 2)()( lim (3-12) where the aITAE is the auxiliary integral of the time-weighted absolute error, theaIAE is the auxiliary integral of the absolute error, and the aISE is the integral of the square of error. It is to

PAGE 42

42 be noted that in the case of PI2 controllers, the value of sse is zero i.e., ) ()(trtra Hence, the ITAE, IAE and ISE error metrics for PI2 retain the original form of Eq. 3-5, Eq. 3-6, and Eq. 3-7, respectively, even when all computations are carried out using their auxiliary counterparts. The ITAE expression (Eq. 3-5) has been a popular choice for control parameter optimization, as it assigns less weight to errors occurring in the initial times and more weight to error at longer times. Error is defined as the closed-loop feedback difference between the auxiliary set point and th e output. This is tradit ionally a useful measure to adopt, as for a step response it is inevitable that there is a relativel y large error during initia l times, which needs to be given less significance compared to the error that is encountered at later times. The ITAE may not be the best metric for the problem in que stion, however. The desired trajectory of the output is the one which tracks the slope of the ramp without abrupt deviations in trajectory. The ITAE is forgiving of aggressive output values at initial times as it gives less significance to error at early times. Sometimes, it is desired to adopt a metric which gives an equal importance to errors occurring at initial times as well. From that perspective, the two other metrics are also used in this work for optimization purposes, na mely, the IAE (Eq. 3-6) and the ISE (Eq. 3-7). A routine in MATLAB is written to simulate the plant output fo r a given controller with its parameters specified [22]. The simplex optimi zation routine is employed for tuning the control parameters. For a fixed value of plant parame ters (gain, time constant and time delay), the parameters of the controller (pr oportional gain and integral action) are altered to minimize the (ITAE, IAE or ISE) value of the co st function. For a PI controlle r, where there is offset with respect to the ramp, an auxiliary ramp parallel to original one is constr ucted. The steady state offset is calculated analytically using Eq. 3-8.

PAGE 43

43 The three error metrics (ITAE, IAE and ISE) are minimized using the new ramp. These metrics are defined over an infinitely long time; however, for practical purposes the final time is chosen to be finite and defined by the formula ) ,max(15 ft (3-13) The rationale behind this choice is that by this extent of fina l time, any reasonably performing controller should make the value of error significantly small. The optimized tuning parameters are non-dimensionalized by combining with plan t parameters, and plots of optimized tuning parameters were constructed. Time responses are constructed with the optimized values to verify the responses of the process with the controlling action incorporated. Peek analyzes the performance of th ree different configurations of PI2 controllers [21]. That study concludes that there is no significant difference in the performance of the three configurations. The transfer function for PI2 controller given in Eq. 3-2 is used in this study as it is the easiest to tune becaus e it involves only two parameters, namely the proportional gain cK and the integral-action time constant i 3.3 Results and Discussion 3.3.1 Tuning Parameters of Controllers Tuning parameters are calculated for both controllers PI and PI2 using the ITAE, IAE, and ISE as the optimizing metric. The optimized c ontroller parameters are nondimesionalised using the plant gain and time constant [17]. The resulting plots of KKc versus / and / I versus / are shown in Figures 3-3 and 3-4 for the PI2 and PI controllers. The tuning parameters obtained using ITAE, IAE and ISE as the metric are shown in the first, second, and third row of each figure, respectively. The plant parameters K, and are selected such that K and range from 0.1 to 50. Twenty logarithmically equally-spaced points are considered for both K and in their

PAGE 44

44 specified ranges. After each time constant is defined, the values of the delay parameter is set by defining the ratio of / to range from 0.1 to 100, with 50 logarithmically equally-spaced points inside the range. For a fixed value of / ratio, the value of is computed from the value of and of the fixed / ratio. Hence, tuning parameters were obtained for 20,000 different plants for each controller and for each metric. The graphs in Figures 3-3 and 3-4 show that, in general, as the / ratio increases, the value of the optimal KKc product and of the / i ratio decrease. The value of the KKc product represents the proportional control action on the closed loop sy stem, and it is expected to vary inversely to the / ratio. In other words, the control ac tion will be higher for smaller values of / ratio, and smaller as / increases. This is qualitatively reflected in Figures 3-3 and 3-4 for all three optimizing metrics and for the two controllers considered. On the same vein, the / i ratio is indicative of integral action for the system and it is expected to behave analogously to the proportional control action. The in tegral action should be more aggr essive for smaller values of / ratio compared to higher values of the ratio. This is, ind eed, observed from Figures 3-3 and 3-4. Least-square fits for the optimized control pa rameters are given in Tables 3-1 and 3-2 for the PI2 and PI controller, respectively. The least square fit relates the optimal KKc product with the / ratios and the optimal / i ratio with the / ratio. If the KKc versus / curve and/or / i versus / curve is significantly nonlinear, a br eak point at a certain value of / ratio is identified and two least-square fits are presented for the same curve, one above and one below the / ratio breakpoint value. The results are for / values ranging between 10-1 and 102 only. 3.3.2 Comparison of the Performance of the PI and PI2 Controllers The performance of the PI and PI2 controllers is characterize d for three different plants (plant parameters given in Table 3-3) using th e optimal parameters prescribed by each of the

PAGE 45

45 three optimizing metrics. The time response curves for each plant are constructed, for both the controllers and the tuning parameters prescribed by the respective optimizing metric, to assess the time-domain performance of each tuning pres cription. The proportiona l gain of the three plants was taken to be the same value, namely K = 1.0. Three different values of the / ratio are taken from the domain of the values for 20,000 differe nt plants. The ratio values of 0.1, 3.0 and 100 (minimum, middle and maximum of the ra nge considered) are selected. Several combinations of / can satisfy each value of the ratio. The value of is selected such that it covers the domain of the different values of selected for all the plants in this study. The values of 0.1, 1.9 and 50 are selected for From these values of the value of is computed for each value of the ratio. Figures 3-5, 3-6 and 3-7, illustrate the time re sponses of the three plants for the ITAE, IAE and ISE metrics, respectively, using PI2 and PI controllers. Each figure demonstrates the performance of PI2 versus PI controller for the three plants. When a PI2 controller is used, the output follows the original ramp, whereas with a PI controller an offset is introduced in relation to original ramp and the output follows the auxiliary ramp parallel to the original ramp. The original and auxiliary ramp are plotted in each time response curve as well. With increasing values of the / ratio, the offset between the original a nd auxiliary ramp increases. This is expected as the offset is directly proportional to i and inversely proporti onal to the product of KKc as shown by Eq. 3-8. With increasing value of / i increases and the product of KKc decreases, hence the offset increases. It is also ob served that it takes longer for steady state to be reached with increasing values of the / ratio. The output of each plant tracks the slope of the original ramp for each controller and every optimizing metric. As discussed earlier, there is offset with respect to the original ramp when PI

PAGE 46

46 controller is used and there is no offset when PI2 controller is used. Re gardless of the fact of whether offset is introduced or not, as long as the output is para llel to the original ramp, the control objective is satisfied. The better performing controller is the one that reaches faster the original or auxiliary ramp, depending on the controller adopted, and with minimal transient values. It is observed that for all the thr ee plants and optimizing metrics considered, PI controller performs better than PI2 controller. Output of each of the plant, on using PI controller, tracks the auxiliary ramp much fast er compared to the output when PI2 controller is used. It is observed that the plant output is more oscillato ry during transient time for ISE prescribed tuning compared to those obtained from the ITAE and IAE cr iteria. This is an expected result as in the ISE metric the square of the er ror is used. Even though both PI2 and PI controllers exhibit oscillatory behavior, the phenomenon is more prominent in the case of PI2 controller. From these observations, it is concluded that the PI controllers are better performing than PI2, regardless of the optimizing metric adopted to tu ne, for the three plants considered, and provided that the unavoidable resulting offs et is acceptable to the user. 3.3.3 Comparison of Metrics After it is established that PI is a better performing controller, the next step is to identify the optimizing metric with which the PI controlle r gives the best performance. Time responses for the same three plants are constructed using the optimizing metrics. Figures 3-8 and 3-9 shows the time responses for the three metrics. For the sake of comparison, even though it is established that the PI is a be tter performing controller, time response curves are generated for the PI2 controller as well. Figure 3-8 shows the PI controller time responses and Figure 3-9 shows the PI controller time respons es, for the three plants. Note th at the offset introduced while using PI controller, is a func tion of the tuning parameters Kc and I Since the values of these parameters are different for each metric, the ou tput using a PI controller follows a different

PAGE 47

47 auxiliary ramp, depending on which optimizing me tric was used. All the auxiliary ramps, however, are parallel to the original ramp. It is observed that ISE is the worst optimizi ng metric, as the output of the plant is most oscillatory and has larger deviation from the auxiliary ramp as shown in Figure 3-8 and 3-9. Also, it takes longer in the case of the ISE to reach steady state. The other two metrics, ITAE and IAE, are quite close in their performance. The time response curves suggest that for tuning purposes the ITAE is a better metric for the PI co ntroller and the IAE is better metric for the PI2 controller. For the PI controller, the IAE demonstrates mo re oscillatory output co mpared to the ITAE metric. Also, the output of the plants reaches steady state sooner when the ITAE is used as the optimizing metric compared to when the IAE is adopted. 3.3.4 Comparison of PI (ITAE) Co ntroller with Literature Precedents After determining that the PI controller using the ITAE as the optimizing metric exhibits highly desirable performance, the next step is to compare its performance with controllers proposed in the literatu re. Peek suggests a PI2 controller with ITAE as the optimizing metric for tracking a ramp set point [21]. Though the objective is slightly di fferent than the one in this study, the controller recommended by Peek does satisfy the control ob jective of this work [21]. Belanger and Luyben recommend a double integrator controller for tracking a ramp, which satisfies the control objective of this study as well [19]. Figure 3-10 shows the performance of the three different controllers for the three plants considered. It is observed that the PI contro llers tuned using the ITAE metric gives the best results followed by the controllers recomme nded by Peek, and then by those proposed by Belanger and Luyben. As the / ratio increases for the plants it becomes increasingly obvious that the PI controller gives the best results. For the plant with the ratio / = 100, the highest

PAGE 48

48 value, the PI controller makes the output reach st eady state much faster and with least transients compared to the other two controllers. 3.3.5 Local Minima versus Global Minima Figure 3-11 shows the contour plots for the ITAE metric using PI controller for the three plants. It is observed from Figure 3-11 (A) that there is more than one minima for the ITAE metric and different values of control parameters Kc and i for each minima. The goal is to obtain global minima for the optimizing me tric and the corresponding optimal control parameters. If local minima is reached as oppos ed to global minima, that would potentially lead to scatter in the optimal control parameters curves as shown in Figures 3-3 and 3-4. Hence, the optimization work described here is conducted wi th care to avoid local minima results. The measure adopted consists of utilizing different initial guesses for each optimization routine execution, leading to the identification of different local minima, when they exist. The smallest of such minima is then accepted as the best ap proximation to the global minimum. Although this approach is neither rigorous nor exhaustive, it provides excellent practical results in the context of this study. 3.4 Conclusions Optimal tuning relationships for the PI and PI2 controllers using th e optimizing metrics ITAE, IAE and ISE are presented for 20,000 first-order plants with time delay. The plants considered have the / ratio value varying from 10-1 to 102. The validation and comparison of the controllers performance is done by thei r deployment on three different plants. On the basis of results obtained, the PI contro llers using ITAE as the optimizing metric are the best performing controllers for the purpose of tracking the slope of a ramp trajectory. They perform better than controllers proposed in prio r literature.

PAGE 49

49 Figure 3-1. Ramp r and auxiliary ramp ra with constant slope, Figure 3-2. Closed loop transfer functi on representation of plant and controller.

PAGE 50

50 Figure 3-3. The PI2 controllers optimal tuning parameters using ITAE (A, B), IAE (C, D), and ISE (E, F) as the optimizing metric.

PAGE 51

51 Figure 3-4. The PI controllers optimal tuning parameters, using ITAE (A, B), IAE (C, D), and ISE (E, F) as the optimizing metric.

PAGE 52

52 Table 3-1. The PI2 controllers optimized tuning paramete rs linear least square fit equations. Criterion Optimal PI2 Parameters ITAE 5.2 74.0 5.2 48.0 5.2 74.0 5.2 26.1 92.0 50.0 04.0 76.0 i cKK IAE 5.2 69.0 5.2 48.0 5.2 78.0 5.2 26.1 89.0 50.0 06.0 76.0 i cKK ISE 5.2 65.0 5.2 45.0 5.2 94.0 5.2 58.1 88.0 52.0 08.0 76.0 i cKK

PAGE 53

53 Table 3-2. The PI controllers optimized tuni ng parameters linear least square fit equations. Criterion Optimal PI Parameters ITAE 5.2 58.1 5.2 82.0 5.2 41.0 5.2 73.0 88.0 13.0 06.0 73.0 i cKK IAE 5.2 42.1 5.2 73.0 5.2 57.0 5.2 07.1 88.0 26.0 08.0 80.0 i cKK ISE 5.2 42.1 5.2 75.0 5.2 74.0 5.2 40.1 88.0 36.0 08.0 80.0 i cKK Table 3-3. Values of the plant parameters us ed compare the performance of the controllers. Parameters Ratio Plant K / 1 1 0.1 0.01 0.1 2 1 1.9 5.6 3.0 3 1 50 50000 100

PAGE 54

54 0 0.4 0.8 1.2 1.5 0 0.4 0.8 1.2 1.5 ty A K = 1 = 0.1 = 0.01 PI2 Ramp PI Aux 0 18 37 56 75 0 18 37 56 75 ty B K = 1 = 1.9 = 5.6 PI2 Ramp PI Aux 0 1.9 3.8 5.6 7.5 x 10 4 0 1.9 3.8 5.6 7.5 x 10 4 ty C K = 1 = 50 = 5000 PI2 Ramp PI Aux Figure 3-5. The PI2 and PI controllers ramp tracking a nd slope tracking performance using the optimal ITAE control parameters.

PAGE 55

55 0 0.4 0.8 1.2 1.5 0 0.4 0.8 1.2 1.5 ty A K = 1 = 0.1 = 0.01 PI2 Ramp PI Aux 0 18 37 56 75 0 18 37 56 75 ty B K = 1 = 1.9 = 5.6 PI2 Ramp PI Aux 0 1.9 3.8 5.6 7.5 x 10 4 0 1.9 3.8 5.6 7.5 x 10 4 ty C K = 1 = 50 = 5000 PI2 Ramp PI Aux Figure 3-6. The PI2 and PI controllers ramp tracking a nd slope tracking performance using the optimal IAE control parameters.

PAGE 56

56 0 0.4 0.8 1.2 1.5 0 0.4 0.8 1.2 1.5 ty A K = 1 = 0.1 = 0.01 PI2 Ramp PI Aux 0 18 37 56 75 0 18 37 56 75 ty B K = 1 = 1.9 = 5.6 PI2 Ramp PI Aux 0 1.9 3.8 5.6 7.5 x 10 4 0 1.9 3.8 5.6 7.5 x 10 4 ty C K = 1 = 50 = 5000 PI2 Ramp PI Aux Figure 3-7. The PI2 and PI controllers ramp tracking a nd slope tracking performance using the optimal ISE control parameters.

PAGE 57

57 0 0.4 0.8 1.2 1.5 0 0.4 0.8 1.2 1.5 ty A K = 1 = 0.1 = 0.01 ITAE IAE ISE Ramp 0 18 37 56 75 0 18 37 56 75 ty B K = 1 = 1.9 = 5.6 ITAE IAE ISE Ramp 0 1.9 3.8 5.6 7.5 x 10 4 0 1.9 3.8 5.6 7.5 x 10 4 ty C K = 1 = 50 = 5000 ITAE IAE ISE Ramp Figure 3-8. The ITAE, IAE, and ISE metric s comparison for three plants, using the PI controllers A), B), and C).

PAGE 58

58 0 0.4 0.8 1.2 1.5 0 0.4 0.8 1.2 1.5 ty A K = 1 = 0.1 = 0.01 ITAE IAE ISE Ramp 0 18 37 56 75 0 18 37 56 75 ty B K = 1 = 1.9 = 5.6 ITAE IAE ISE Ramp 0 1.9 3.8 5.6 7.5 x 10 4 0 1.9 3.8 5.6 7.5 x 10 4 ty C K = 1 = 50 = 5000 ITAE IAE ISE Ramp Figure 3-9. The ITAE, IAE, and ISE metric s comparison for three plants, using the PI2 controllers A), B), and C).

PAGE 59

59 0 0.4 0.8 1.2 1.5 0 0.4 0.8 1.2 1.5 ty A K = 1 = 0.1 = 0.01 BL PI Peek Ramp Aux 0 18 37 56 75 0 18 37 56 75 ty B K = 1 = 1.9 = 5.6 BL PI Peek Ramp Aux 0 1.9 3.8 5.6 7.5 x 10 4 0 1.9 3.8 5.6 7.5 x 10 4 ty C K = 1 = 50 = 5000 BL PI Peek Ramp Aux Figure 3-10. The PI controllers tuned using th e ITAE metric compared with Belanger and Luyben and Peeks controllers for three plants A), B), and C).

PAGE 60

60 1/iKc A K = 1 = 0.1 = 0.01 8 10 12 3 3.5 4 1/iKcK = 1 = 1.9 = 5.6 0.2908 0.2908 0.2908 0.4029 0.403 0.4031 1/iKcK = 1 = 50 = 5000 5.0454 5.0505 5.0556 x 10 0.3174 0.3174 0.3175 Figure 3-11. Contour plots for the PI controller tuned using the ITAE metric for the three plants A), B), and C).

PAGE 61

61 CHAPTER 4 GENERALIZED PREDICTIVE CONTROL FOR FUEL CELLS 4.1 Introduction A generalized predictive cont rol (GPC) strategy is designed and implemented for a polymer electrolyte membrane (PEM) fuel cell system It is vital for efficient performance of the fuel cell to ensure the robust and precise control of the performance variable which is oxygen excess ratio. A review of prior control strategies developed for this fuel cell system, is presented and compared to the new GPC scheme proposed. The performance of the di fferent controllers is evaluated in the case where the model avai lable for design suffers from uncertainty. 4.2 Fuel Cell System Background Fuel cell are electrochemical devices that dire ctly convert the chemical energy of gaseous reactants to electrical energy. Th ey are widely considered as an alternative to fossil fuels which are limited in supply. For a typical fuel cell, water and heat are byproducts generated as a result of operation. The reactants need ed are hydrogen and oxygen. Both of these reactants are widely available, and a proliferation of applications based on these fuels would tremendously reduce our dependence on fossil fuels. This is an additiona l motivation for developing and engineering fuel cell system as it is friendly to the environmen t. Fuel-cell based automobiles have no harmful emissions, such as CO2 which combustion-engine automobiles contribute significantly to the environment [23]. Fuel cells are an effi cient and clean source of energy production. William R. Grove discovered the principle of op eration of fuel cells in 1839 [24]. From a classical standpoint, a fuel cell is comprised of two electrodes with an electrolyte located between them. The electrolyte has the special property that it allows only protons (positively charged hydrogen atoms) to pass through it. In contrast, the membrane does not allow electrons to pass through. Hydrogen gas passes over the an ode electrode, and with the assistance of a

PAGE 62

62 catalyst, breaks down into protons and electr ons. The protons sele ctively pass through the membrane to reach opposing cathode electrode. Th e membrane is an electronic insulator. The membrane is comprised of fluorocarbon chain to which the sulfonic acid groups are attached. On hydration of the membrane the hydrogen ions b ecome mobile. The electrons flow through an external circuit, creating a curr ent flow. An oxygen flow passe s over the cathode and combines with the protons and electrons to generate water. The reaction at the anode is 2H2 4H+ + 4e(4-1) and at the cathode is O2 + 4H+ + 4e2H2O (4-2) hence, the overall reaction is 2H2 + O2 2H2O (4-3) Several kinds of fuel cell designs have been developed and are curren tly being studied [2530]. A schematic diagram of an automotive fuel cell system including the structural relationships among the input, outputs, and distur bance signals is give n in Figure 4-1. A compressor and pressurized hydrogen tank are used to provide the reactants oxygen and hydrogen, respectively [31]. The compressor plays a crucial role as it ensure s that the desired air flow rate reaches the cathode based on power dema nds. The supply and return manifold models based on thermodynamic consideration provide inform ation about desired air flow rate needed. A nonlinear curve fitting method is used to desc ribe the compressor behavior [31]. The net power delivered by the fuel cell system is the difference between the power generated and power consumed to run the compressor motor to deliver a particular air flow rate An excess amount of air flow provided to the cathode is referred to as the oxygen excess ratio 2O which is defined as the ratio of the rate of oxygen supplied to rate of oxygen consumed. With increase in oxygen

PAGE 63

63 excess ratio, resulting in high oxygen partial pressure, there is increase in power delivered. However, it is at the cost of increased power consumption by the compressor to deliver a higher air flow rate. Beyond a particular value of the oxygen excess ratio there is loss in net power as a result of increased power consumption by the comp ressor. It has been shown in literature that having oxygen excess ratio in the vi cinity of 2, the fuel cell sy stem delivers highest net power. A humidifier, in fuel cell system, is used to add water to the reactan ts to avoid dehydration of the membrane. A water separator is used to extract water from the ai r, leaving the fuel cell stack, which is recycled back to the humidifier via the water tank. The voltage generated by the fuel cell needs to be conditioned before it is fed to the traction motor. Appropriate usage of an external battery with th e fuel cell power supply helps in mi nimizing the transient responses and delivering better system efficiency. Polymer electrolyte membrane (PEM) fuel ce lls, also referred to as proton exchange membrane fuel cells, are recognized particularly promising for ut ilization in automobiles as a substitute for the internal combustion engine. This is because of the fact that the PEM fuel cells have high power density, long cell life, low corros ion, and use a solid electrolyte. For effective utilization of fuel cell technology, it is vital that that the fuel cell system is accurately understood, monitored and that its process variab les be held under tight control for various operating conditions. From a control engineering perspective the fuel cell can be divided into four subsystems, namely (1) supply of the reactants air and hydroge n, (2) humidification of the reactants and of the membrane, (3) heat management, and (4) power management. The model used in this study assumes a perfect humidifier and coolers for the reactants and the membrane. The model assumes a perfect power management system whic h controls the power drawn from the fuel cell

PAGE 64

64 stack. A fast proportional controller is impl emented on the hydrogen flow that tracks cathode pressure [31]. This reduces the control problem to regulating the air suppl y, as the oxygen level varies on the cathode side, due to varying power demands. 4.3 Objectives of the Research The objectives of this study are the following: 1. Propose a systematic design solution for trea ting the of fuel-cell control problem of regulating the oxygen excess ratio (perform ance variable) at a value of 2 by synthesizing a GPC strategy. 2. Develop a systematic GPC design procedure for a fuel cell model in the scenario where all states are measurable. 3. Develop a systematic GPC design procedure for the fuel cell model when selected states are not measurable by incorpora ting an observer in the controller. 4. Evaluate the robustness of the GPC contro ller with respect to model uncertainty. 5. Rederive and correct, as needed, lin ear models obtained from nonlinear formulations proposed in the prior literature for the fuel -cell model used for this study [31]. 6. Retune and redesign, as needed, all mode l-based controllers pr oposed in the prior literature [31] to take into account the corrected linear models [31]. 7. Compare by means of simulations the perf ormance of the different controllers proposed in prior literature [31] with the GPC strategy. 4.4 Fuel Cell Model The fuel cell model proposed by Pukrushpan et al. is the basis for this study [31]. The state equation for the model of a fuel cell system is of the form ),,( wuxfx (4-4) ),,( wuxgz (4-5) ),,( wuxhy (4-6) where x represents the states of the fuel cell system, u is the input, w is the disturbance, and z and y are the outputs. The state vector is given by the expression

PAGE 65

65 T rm anwsm sm cp N H Opmmp mmm ] [,2 2 2 x (4-7) where the vector elements are the mass of oxygen 2Om in cathode volume, the mass of hydrogen 2Hm inside the anode volume, the mass of nitrogen 2Nm in cathode volume, the rotational speed of the compressor cp the pressure of the supply manifoldsmp, the mass of air in the supply manifold smm, the mass of water at anode anwm,, and the pressure of the return manifold rmp. The outputs used as the performance va riables are organized in the vector T O Pnete ][2 z (4-8) where del set PPPenetis the difference between the desired power set P and the actual power delivereddel P and 2O is the oxygen excess ratio. The contro l objective in this study is confined to regulating the value of 2O at the desired value 22set O [31]. Three additional measured outputs are organized in the vector T stsm cpvpW ] [ y (4-9) where the vector elements are the mass fl ow rate of the air from the compressor cpW, the pressure of the supply manifoldsmp, and the voltage of the stack stv. The control input is given by cmvu (4-10) where cmvis the voltage signal sent to the compressor which in turn delivers a corresponding flow rate of air to the cathode. Finally, the disturbance stIw (4-11) is the stack current Ist.

PAGE 66

66 The nonlinear model described in Eqs. 4-4 4-6 and shown in Figur e 4-2a is linearized about a nominal operating point at which th e fuel cell model generates a net power of del P =40kW and sustains an oxygen excess ratio 22O. These conditions are realized when 191stIA and 164cmvV. The corresponding values of the states are T T rm anwsm sm cp N H Oeeeeeeee pmmp mmm]58.131.128.352.232.823.146.530.2[ ] [,2 2 2 x (4-11a) The physical units of the states x are [kgs kgs kgs rad/s pasc al kgs kgs pascal] (4-11b) respectively. Hence the nominal ope rating point is gi ven by the input 164 uV, disturbance 191 wA, performance variable 22 z, and vector x as described by Eq. 4-11a. Two linearization cases considered in this study are the following: 1. The direct linearization of the fuel cell m odel Eqs. 4-4 4-6 and shown in Figure 42a [31]. 2. The linearization of a combined system consisting of a static feedforward controller, connected to th e fuel cell model Eqs. 4-4 4-6 shown in Figure 4-2b. The static feedforward control law used in case 2 is described in Section 4.5.2.1. The state, input, output, and feedthrough matrices of the re sulting linear time-invariant models for cases 1 and 2 are shown in Figures 4-3 and 4-4, respec tively. Through personal communication with the members of the research group at University of Michigan we lear ned that the matrices reported in [31] are affected by errors, hence these matr ices are re-derived and found to differ slightly from the ones given by Pukrushpan et al. [31] 4.5 Literature Precedents fo Fuel Cell Control Designs Pukrushpan et al. investigate the design and applicati on of two main control strategies along with 2 major variations of each of these strategies, as indi cated in the following list [31]:

PAGE 67

67 1. Feedforward strategy a. Static feedforward (sFF) b. Dynamic feedforward (dFF) 2. Combination of static feedforward st rategy with optimal feedback control a. Linear quadratic regulator (LQR) b. Linear quadratic gaussian observe r (LQG) in combination with LQR feedback. A succinct discussion of these strategies is given in the ensuing sections 4.5.2 Feedforward Strategy The two feedforward control strategies proposed in the literature are a static feedforward (sFF) and a dynamic feedforward (dFF) scheme. Th e sFF controller is de rived from simulations and using the results of substant ial experimental work. The dFF controller is based on a linear model, and thus its performance on the nonlinea r model is dependent on the non linear models proximity to the nominal operating point. The models used for control design are discussed in greater detail in th e next subsections. 4.5.2.1 Static feedforward controller Pukrushpan et al. propose the sFF control law [31] 378123 wu (4-12) where the stack current w is the measured disturbance impacting the fuel cell model. The control input u is the voltage cmv applied to the compressor. A schematic of the sFF controller is shown in Figure 4-5. The derivation of Eq. 4-12 cons ists of first seeking a function relating the disturbance stIto the required air mass flow rate cpW, in such a fashion that the flow rate, achieved by invoking thermodynamics principles, negates the effect of the disturbance on the performance variable2O A resulting static function is obt ained by means of simulations and experimental work which co -relates the control input cmv to the required air mass flow ratecpW

PAGE 68

68 for a particular value of disturbance stI, maintaining the desired value of the performance variable 2O This static function is implemented as a static sFF controller in the form of Eq. 412. 4.5.2.2 Dynamic feedforward controller Pukrushpan et al. propose the dynamic feedforward (dFF) control law wKuuw (4-13) where uuu and www are the deviation values fr om the nominal control input u and the nominal disturbance w, respectively, and uwK is the transfer function )1)(1)(1(3 2 1sss K Kideal uw uw (4-14) where s is the Laplace variable, 1 2 and 3 are filter constants, and wzuz ideal uwGGK2 1 2 where uzG2 and wzG2are transfer functions describing the map uGwGzuz wz 2 22 (4-15) where 222zzz is the deviation of the performance variable (oxygen excess ratio) from its nominal value 2z [31]. The schematic and transfer-f unction representations of the dynamic feedforward controller dFF are given in Figure 4-6. The derivation of the dFF control transfer function begins by considering the transfer function wKGwGzuwuz wz 2 22 (4-16) obtained by substituting Eq. 4-13 into Eq. 4-15. The expression 4-16 should become identically zero for an effective dFF controller uwK. The expression

PAGE 69

69 wzuz ideal uwGGK2 1 2 (4-17) is obtained from Eq. 4-16 after equating 02 z However,1 2wzGis not a proper function. Low pass filters are added to implement a causal controller, resulting in wzuz uwGG sss K2 1 2 3 2 1)1)(1)(1( 1 (4-18) The authors of this derivation pr opose the filter-constant values 801 1202 and 1203 [31]. 4.5.3 Combination of Static Feedforwar d with Optimal Feedback Controllers Feedforward controllers lack r obustness to disturbance and model variations. To address this issue, feedback controllers are added in conjunction with f eedforward controllers. The following two scenarios are considered for f eedback control design, in conjunction with static feedforward control [32-34]: a. The performance variable is measurable. b. The performance variable is not measurab le, leading to the introduction of a state observer to estimate its value. The model used for designing the optimal feedba ck controller is defined in such a fashion that it includes the static f eedforward control relationship. A schematic of the resulting feedforward and feedback cont rol is shown in Figure 4-7. 4.5.3.1 Case where the performance variable is measurable Pukrushpan et al. propose the linear quadr atic regulator (LQR) f eedback control law: qKxKuuI pp (4-19)

PAGE 70

70 where pu is a pre-compensator, uuu and xxx are the deviation values of the feedback control input u and the state vector x from their nominal points u and x respectively, and q is the integral defined th rough the differential equation 11yyqreq (4-20) where req cp reqWy 1is the analytically calculated value of the air mass flow rate needed for a particular value of the disturbance Ist to attain the desire d oxygen excess ratio, and cpWy1 is the measured value of the air mass flow rate. In the control law given by Eq. 4-19, gain matrices pK and IK refer to the optimal gains resulting fr om minimizing the quadratic cost function 0) (2 2dtuRuqQqxCQCxJT I T zz T z T (4-21) where, zQ, IQ, and R are weighting matrices for the performance variable, stateq and control input u, respectively. The term2zC refers to the second row of the C matrix, given in Figure 4-4 describing the linear state space model. Note that the second row is associated with the performance variable 2O The pre-compensator pu in the feedback control law Eq. 4-19 ,used to take into account the disturbance effect on the performance variable, is given by wBKBACDBKBACuwpu zwzupu zp1 1 1)( )(2 2 2 (4-22) where A, wB,uB, and wzD2 are elements of the linear state sp ace model given in Figure 4-4. In particular wB and uB are the first and second column of the B matrix, respectively and wzD2 is the second element of th e first column of the D matrix. A schematic diagram of the sFF scheme supplemented with the feedback cont roller is shown in Figure 4-7.

PAGE 71

71 The optimal linear quadratic regulator 4-19 is implemented to prescribe part of the feedback control input. The additional state q, as described in Eq. 4-20, is introduced to minimize offset. Using 10000 zQand 001 .0 iQ in Eq. 4-21 and minimizing the cost function 4-21 the optimal values of c ontroller gains obtained are 6-1.1e17-1.9e 2.64 6-2.0e 5-5.2e 25.4617-1.4e23.03-pK (4-23) and 001 .0 iK (4-24) Note that 4-23 and 4-24 differ from the values reported by Pukrushpan et al. [31]. This discrepancy is a consequence of the difference, reported in Secti on 4.4, with the state matrices used by Pukrushpan et al.. 4.5.3.2 Case where the performan ce variable is not measurable Optimal observer. When selected states are not measurable, Pukrushpan et al. propose the following modification of the control law 4-19: qKxKuuI pp (4-25) where xxx is the deviation of an estimated state vector x from its nominal value x The estimated state is computed from the Kalman based observer ) ( yyLBuxAx (4-26) DuxCy (4-27) where L is the optimal observer gain calculated using the linear quadratic gaussian (LQG) method, and y is the estimated values of the measured outputs [31]. Three measurable outputs, namely the compressor mass flow rate cpW, the pressure of the supply manifold smpand the stack voltage stv, are used as inputs to the observer to estimate the valu e of the states [31]. The optimal gain of the Kalman observer is

PAGE 72

72 006-5.97e003-527.22e 009-486.77e000+1.85e 009-9.95e018-270.13e 015-143.89e009-124.80e 015-78.31e006-4.60e003-460.51e 009-332.44e009-49.54e003-4.22e 012-578.06e 015-349.02e009-25.06e 015-2.54e000+2.18e 009-31.69e 015-18.54e015-417.04e 00912.22e015-7.43e L (4-28) Note that 4-28 differs from the value reported by Pukrushpan et al. [31]. This is a consequence of using different state matrices than what were used by Pukrushpan et al. as explained in Section 4.4. In addition, we also made an appr oximation to the noise variance matrices used for LQG design, since only partial in formation is available in [31], hence adding another source of discrepancy. 4.6 Generalized Predictive Control The generalized predictive control (GPC) law is given by the discre te-time law [35, 36] (t)ym(t)kOL T)(tu (4-29) where k is the GPC control gain vector, set O2m(t) is the set point, (t)yOL is the vector of the constant forcing values of the performance variable, u is the prescribed control input, and the symbol represents the difference operator. The vectors m(t), (t)yOL,and the gain k have R elements each, representing their respective values upto the prediction horizon R. The process involved for obtaining the GPC c ontrol law Eq. 4-29 is discusse d in the following sections. Generalized predictive control design. The first step leading to the derivation of Eq. 429 is to acquire a DARMA (deterministic au toregressive moving av erage) model from continuous state space model [35-38]. To accomplish this, a discrete versi on of the linear state space model is obtained. The sampling period T is chosen such that about 10-20 samples are

PAGE 73

73 taken during the transient time. The z-transform is performed on the discrete model and the z variable is replaced with the q forward-shift operator. The nu merator and the denominator are divided with the highest power of q to obtain a denominator polynomial A and a numerator polynomial B as functions of the backward shift operator 1 q. Polynomial B is defined such that it does not include the unit time delay introduced by sample and hold operation. Finally, the DARMA model for the fuel cell model is obtained in the form ) 1()()()(1 1 tuqBtyqA (4-30) a an n i iqaqaqaqA ....1)(1 1 1 (4-31) b bn n i iqbqbqbbqB .... )(1 10 1 (4-32) and where A and B are polynomials in the backward shift operator 1q. The factors ia and ib are the coefficients of the powers iq in polynomials A and B, respectively. The symbol y represents the output (performance variable), i.e. the oxygen excess ratio in the case of the fuel cell, and u is the control input, i.e., the voltage supplied to the compressor motor vcm to deliver the desired air mass flow rate [29, 30]. Second, a predictor for the performance va riable is designed using the Diophantine equations ) ()(11 1qFqqAEi i i (4-33) where 1 1 1 1.. ..1 i i j j iqeqeqeE (4-34) a an n j j iqfqfqffF .. ..1 10 (4-35) Ri,..,2,1 (4-36)

PAGE 74

74 and where R is the prediction horizon, 11q, and E and F are polynomials in powers of the backward shift operator 1q. The factors ej and fj are the coefficients of the powers jq in polynomials Ei and Fi respectively [30, 32]. Next, multiplying both sides of the DARMA model (Eq. 4-30) by the polynomial i iEq and invoking Eqs. 4-33-36 yields )()()1()()()(1 11tyqFituqBqEityi i (4-37) where ) (ity is the predicted value of the output at the instant it Next, the constant forcing value of the performance variable is calculated. The product of polynomials Ei and B can be decomposed in the form )()()()(1 1 1 1 qqqGqBqEi i i i (4-38) where Gi and i are operator polynomials given by 1 1 1 10 1.. .. )( b bn n j j iq q q q (4-39) 1 1 1 1 1.. .. )( i i j j o iqgqgqggqG (4-40) and where j and jgare the coefficients of powers of jq in the respective polynomials. Substituting Eq. 4-38 into Eq. 4-37 yields )()()1()()1()()(1 1 1tyqFtuqituqGityi i i (4-41) A constant forcing 0 )1(itu for ,.., 2,1 i produces a constant-forcing output in Eq. 4-41 of the form ) ()()1()()(1 1tyqFtuqityi i OL (4-42) where ) (ityOL represents the consta nt-forcing response of y at instants it, ,.., 2,1i.

PAGE 75

75 Third, an objective/cost-function J is established and minimized with respect to the control-input increment uto obtain the GPC control law. The appropriate cost function adopted is defined by u uy(t)m(t) y(t)m(t)T )()(TuJ (4-43) where is a weighting matrix, m and y are vectors of future set point and predicted future output respectively, u is the vector of control inputs, and is the weighting matrix for u. The vectors m and y have R elements each, representing their respective future values upto the prediction horizon R. Vectors u and u has L elements, where L is the control horizon defined as the instant where the c ontrol design specifies that ) 1()( Ltuitu, 1 ,...,1, RLLi (4-43a) In the case of the fuel cell, the performance variable y is composed of a contribution from the disturbance variable d (namely, the stack current Ist), and a contribution from the manipulated variable u (namely, the voltage to the compressorcmv) leading to its definition uGydGyyyyu OL u d OL dud (4-44) where dy is the component of th e performance variable y contributed by disturbance d Note that the notation for the di sturbance is changed from w Sections 4.4 to 4.6 to d in Sections 4.7 which involve the GPC strategy. Signal yu is the component of the performance variable y contributed by mani pulated variable u. The termsOL dy and OL uy represent the constant-forcing values of dy and uy respectively. The termsdG and uG are the associated dynamic matrix polynomials for dy and uy respectively. The termdGd is set as 0 dGd, as0 d since d is a disturbance and its unknown future values ar e assumed to be equal to the current value, i.e., )(itd is assumed to be constant and equal to ) (td, ,.. 2,1 i. This reduces Eq. 4-44

PAGE 76

76 uGyyyyyu OL u OL dud (4-45) Hence, since OLyy when 0 u, it follows that OL u OL d OLyyy (4-46) Substituting the value of y described by Eq. 4-45 in the cost function Eq. 4-43 yields u u u)G(t)y(m(t) u)G(t)y(m(t)T u OL T u OL uJ (4-47) which when minimized with respect to u yields )X (G I))G (G uTT u 1 u TT u (4-48) where (t)ym(t)XOL. The control law is extracted from Eq. 4-48, takes the form XkT)(tu (4-49) where T k is the first row of the matrix ) (G I))G (GTT u 1 u TT u and XkT)1()(tutu by the GPC algorithm. Eq. 4-49 can be easily rewr itten via a simple substitution of factor X to reduce to Eq. 4-29. This completes the de rivation of the GPC c ontrol law Eq. 4-29. For stability analysis and si mulation purpose it is importa nt to develop a closed-loop transfer function for the GPC loop. The GPC control law described by Eq. 4-49 in summation form is given by ] [)(,, 1OL iu OL idi R i iyymktu (4-50) where the subscript i denotes the values of the respectiv e variables at the time instance t+i. Then using Eq. 4-42 )()()1()()(1 1 ,tyqFtdqityd id id OL id (4-51) )()()1()()(1 1 ,tyqFtuqityu iu iu OL iu (4-52)

PAGE 77

77 Substituting the values of OL idy, and OL iuy, given by the Eqs. 4-51 and 4-52 into the GPC control law given by Eq. 4-50 and rearranging terms yields R i u iui R i d idi R i idi R i i i i iu R i ityqFktyqFk tdkqmqktukq1 1 1 1 1 1 1 1 1)()(()()(( )() ( )() 1( (4-53) Let ) 1(, 1 1 iu R i i ukqR (4-54) R i idi dkqR1 1) ( (4-55) R i i iqkT1 (4-56) R i idi dqFkS1 1 ,) ( (4-57) and R i iui uqFkS1 1 ,) ( (4-58) Inserting in Eqs. 4-54 4-58 into Eq. 4-53 yields ) ()()()()( tyStyStdRtTmtuRuu dd d u (4-59) Since ) ()()( tytytyd u (4-60) Eq. 4-59 can be rewritten as )) ()(()()()()( tytyStyStdRtTmtuRd u dd d u (4-61) which after rearranging terms results in ) ()()()()()( tyStySStdRtTmtuRu dud d u (4-62) Also, since )(1td A B qyd d d (4-63)

PAGE 78

78 Eq. 4-61 can be expressed in the form )()()()()()(1tyStd A B qSStdRtTmtuRu d d ud d u (4-64) which after a trivial series of algebraic operations can be written as ) ()())(()()(1tySAtdBqSSRAtTmAtuRAud d uddd d ud (4-65) Now, let udRAR ~ (4-66) )( ~ tTmATd (4-67) d uddddBqSSRAS1)( ~ (4-68) and udySAS ~ (4-69) Substituting Eqs. 4-66 4-69 into Eq. 4-65 yields ) ( ~ )( ~ )( ~ )( ~ tyStdStmTtuRy d (4-70) which finally leads to th e closed-loop established by the GPC control strategy )( ~ ~ ~ )( ~ )( ~ tyStmTtuR (4-71) where ydSSS ~ ~ ~ ~ (4-72) and )( )( )( ~ ty td ty (4-73) Figure 4-8 shows a closed loop schematic of the GPC controller for the augmented model which incorporates the disturbance. To obtain the closed-loop transf er functions relating the input u(t) and the output y(t) to the set point m(t) and the disturbance d(t) first consider the op en-loop relationships )( )(1td A B qtyd d d (4-74)

PAGE 79

79 and )( )(1tu A B qtyu u u (4-75) Let ddBBq 1 and uuBBq 1. Substituting Eqs. 4-74 and 4-75 into Eq. 4-59 and rearranging terms yields )( )( )(td BSRA BSRA A A tm BSRA TA tuuuuu dddd d u uuuu u (4-76) which is the closed loop transfer function from the set point m(t) and the disturbance d(t) to the control input u(t) Inserting into Eq. 4-65 the expression )()()(tu A B td A B tyu u d d (4-77) obtained from Eqs. 4-45, 4-74, and 4-75 yields )( ) ( ) () ( )( )(td BSRAA BSRABBSRAB tm BSRA TB tyuuuud dddduuuuud uuuu u (4-78) which is the closed loop transfer function from the set point m(t) and the disturbance d(t) to the output ) ( ty 4.7 Battery of Observers In this scheme, a parallel ba ttery of four Kalman based observers designed using four different nominal points, respectively, are deployed for estimating the value of the controlled variable where the equations for each observer i is given by ) ( i iiiiiyyLuBxAx (4-79) uDxCyiiii (4-80)

PAGE 80

80 where ix and iy are the estimated state vector and es timated output vectors from observer i respectively. The state matrices iiiCBA ,and iD correspond to the nominal point for which the observer i is designed. The observer which delivers the best performa nce is used for providing its estimated value of the controlled variable to the GPC controller. The observers are based on linear models derived from the nonlinear model at different nomina l values of the disturbance. The observer is defined by Eqs. 4-79 4-80. The observer gain L is computed from LQG principles as defined in previous section for each of the observers. The determination of the best performing observer is done by comparing the norm of the error in the measured outputs produced by each observer. The error-norm iN for observer i is defined by 2 ,3 2 ,2 2 ,1iii ieeeN (4-81) where icp cpiWWe, ,1 (4-82) ism smippe, ,2 (4-83) iststivve, ,3 (4-84) are the errors defined by the difference be tween the measured value of the outputs (cpW ,smp and stv ) from the nonlinear model and their estimated values (icpW, ,ismp, and istv, ) from the observer i respectively. The observer that delivers th e least value of the norm is selected as the best current observer and is implemented to delive r its estimated value of the controlled variable to the controller. The nominal value of the dist urbances for which the observers are designed are w=100, 191, 125, 22 A. The values are chosen to account for the range of disturbance values that the fuel cell model experiences.

PAGE 81

81 4.8 Simulation Studies and Results This section presents simulations results obtained from various control strategies implemented on the nonlinear fuel cell model. The MATLAB and SIMULINK software computational tools are used for simulation purposes and for assessing the controllers performance [22]. The nonlinear fu el cell model created by Pukrushpan et al., in the SIMULINK environment is used for simulations [31]. The performance of the generali zed predictive control strategy is presented first. The scenario where the performance variable is meas ured is discussed, followed by the case when it is not measured. In the latter case an observer is incorporated in the control design to estimate the unmeasured value. The robustness of the GPC control design is assessed by examining its performance on an altered fuel cell model obtained by modifying a parameter of the original nonlinear model. Finally, a comparison of the GPC controllers perf ormance with that of controllers proposed in literature is conducted. For all the simulation studies carried out in this work, the disturbance d ( i.e., the stack current Ist ) profile shown in Figure 4-9 is used, whic h is identical to the one used by Pukrushpan et al [31]. The y-axis denotes the value of th e disturbance and the x-axis is time. The trajectory of the disturbance is implemented as step ch anges. In Figures 4-10 4-19, showing the simulation results for various control configurations, the y-axis denotes the value of the performance variable oxygen excess ratio 2O and the x-axis is depicts the time t The dotted black line indicates the value of the desired set point i.e., 2 )( tm for the performance variable.

PAGE 82

82 4.8.1 Generalized Predictive Control Results As discussed in Section 4.4, the GPC approach is implemented on the fuel cell model to attain the desired objective of regulating the performance variable, i.e., keeping the oxygen excess ratio at a value of 2. The following scenarios are considered for evaluating the performance of the GPC strategy: 1. The controlled variable is measured. 2. The controlled variable is not measured. A battery of observers based on Kalman filtering is included to estimate the value of performance variable. 3. The design model is uncertain. The robustness of the GPC controllers under model uncertainty for cases (1) and (2) above is evaluated by examining the performance on a modified fuel cell model obtained by varying a parameter of the original model. 4. The control is benchmarked against literature precedents. The performance of the GPC controller is compared to control de signs proposed for the fuel cell model in prior literature. 4.8.1.1 Case where the performance variable is measured The performance of the GPC controller when the controlled variable is measured is discussed in this section. Sin ce in this case the controlled variable (oxygen excess ratio) is measured, its value is directly available to the GPC controller. The GPC algorithm described in Section 4.6 dynamically prescribes the value of the control input (the vol tage to the compressor motor vst) which in turn delivers the approp riate compressor mass flow rate ( Wcp) to maintain the value of the controlled variable at the desired set-point value of 2. Figure 4-10 shows the results of a simula tion study where the GPC control design is implemented on the nonlinear fuel cell model. The x-axis in Figure 4-10 denotes the time and the y-axis depicts the value of th e controlled variable, i.e., the oxygen excess ratio measured from the nonlinear fuel cell. The initial value of states, distur bance and control input for the model are chosen as discussed in Section 4.4. The dotted line indicates the set point of the

PAGE 83

83 controlled variable, which in this study is consta nt and equal to 2. The black solid line depicts the measured value of the controlled variable unde r conditions where the fu el cell is subjected to the disturbance profile (v alues of stack current Ist) shown in Figure 4-9 and the input is prescribed by the GPC algorithm. Figure 4-10 shows that the GPC controller suc cessfully returns the controlled variable at the desired value without steady-state offset. Only small deviations from the set point are observed during transients. The spikes in the sign al plot reflect instances where the disturbance suddenly changes values in a stepwise fashion, as documented in Figure 4-9 and hence they are unavoidable. A value of the controlled variable be low 2 indicates that a po sitive step change in disturbance took place, with the effect of depleting the oxygen concentration in the cathode and consequently reducing the oxygen excess ratio before the control action could remedy the problem. Analogously, a transient value of the controlled variable lying above 2 reflects an opposite scenario of an occurrence of a negative step change in disturba nce. The asymptotic offset-free behavior observed is expected as the GPC approach incorporates an integrator as an essential component of its structure. 4.8.1.2 Performance variable not measured Battery of four observers. Figure 4-11 shows the time response of the performance variable when a battery of four observers ar e included in the controller to account for the situation where the controlled variable is not measured. The observer desi gn steps are discussed in Section 4.7. In Figure 4-11, the solid line indicates the performance variable trajectory as a function of time. The dotted line is the desired set point value of 2 for the performance variable. The measured outputs from the fuel cell mo del are (1) the compressor mass flow rate Wcp, (2) the stack voltage vst ,and (3) the supply manifold pressure psm. These measurements are provided to the observers to estimate the valu e of the performance variable. The observers

PAGE 84

84 incorporated in the control strategy provide es timates for the performance variable and the norm error value that they generate. According to the switching algorithm the best performing observer, that generates the least value of the norm-error value, feeds its estimated value of oxygen excess ratio to the GPC controller. The GP C strategy attempts to regulate the value of the estimated performance variable at 2. Excel lent results are obtained by using a battery of observers which is evident by zero-offset/near zero -offset of the performance variable value from the setpoint. The norm-error values of the four observers are shown in Figure 4-12. The switching pattern indicating the observer used for estima ting the value of oygen ex cess ratio is shown in Figure 4-13. The value of the control input, th e voltage prescribe to the compressor motor, is shown in Figure 4-14. The deviation from th e estimated and measured three outputs for the observers 1, 2, 3, and 4 are shown in Figu res 4-15, 4-16, 4-17, and 4-18, respectively. Battery of three observers. An identical strategy described above is used for observer design except that a battery of three observers in stead of four is implemented. Observer 3 is excluded from the battery of observers. Figure 4-19 shows the time response of the performance variable. The solid line indicates the performance variable traject ory as a function of time. The dotted line is the desired set point value of 2 for the performance variable. A deterioration in control performance is observed compared to the case when four observers are used. This result is expected as there are fewer obs erves available that are designed in the vicinity of the different operating points of the nonlinear fuel cell model. As a result, the estimated value of the performance variable provided by the observers do not coincide with the actual value resulting in some deviation from the set point because the observer design is based on a linear model. The

PAGE 85

85 GPC strategy in turn attempts to regulate the erroneous estimated value of the performance variable and consequently a certain degree of dete rioration in controller pe rformance is expected. Battery of two observers. Figure 4-20 shows the time response of the performance variable when only two observers are employed in battery of observers. The solid line indicates the performance variable trajectory as a function of time. The dotted line is the desired set point value of 2 for the performance variable. Observ ers 1 and 3 are excluded from the battery of observers. A further deterioratio n in controller performance is no ted compared to the case where three observers are employed. This is an expected result as discussed earlier. Employing only one observer. Figure 4-21 shows the time response of the performance variable when only observer 2 is used to estimate the value of the performance variable. The solid line indicates the performance variable traj ectory as a function of ti me and the dotted line is the set point value of 2. This scenario exhibits the worst controller perf ormance compared to the cases where batteries of observers having 2, 3, or 4 observers are used. The GPC controller with one observer supplement fails to deliver offset fr ee regulation. This is an expected result, as discussed earlier. 4.8.2 The GPC Approach Evaluated for Robustness For robustness considerations, the GPC strategy was evaluated for model uncertainty. The GPC controller designed for the original fuel cell model is used on a modified model which is acquired by changing a parameter of the original mo del. More specifically, the return manifold throttle area changed from 0.0020 m2 to 0.0023 m2. The two scenarios considered for robustness analysis are 1. The performance variable is measured. 2. The performance variable is not measured An LQG observer is included in the control strategy to estimate the value of performance variable.

PAGE 86

86 4.8.2.1 Case where the performance variable is measured Figure 4-22 shows the time response of th e performance variable from the modified nonlinear fuel cell model when the GPC control law is designed for the original model. The solid black line indicates the time response of the performance variable from the modified nonlinear fuel cell model. The intention is to ex amine the performance of the GPC controller in a scenario of model uncertainty. In this case the performance vari able is measured and is fed as a direct input to the GPC controller. The simulation conditions are identical to the scenario discussed in Section 4.7.1.1, except that the modified non linear model is used. As observed from Figure 4-22, the GPC controller displays robustness to model uncertainty because it ensures effect ive rejection of the effect of different values of disturbance, and produces offset-free steady-state responses. The robustness of the GPC algorithm to model uncertainty is a key advantage of the controller. Even though the GPC strategy is designed for the original model, the feedback control of th e performance variable produced by the modified model gives the controller the opportunity to ma ke the necessary changes to minimize deviations from the set point. There is ze ro offset as the integrator in the GPC controller adjusts its output appropriately to eliminate the e rror of the performance variable with respect to the set point. 4.8.2.2 Case where the performance variable is not measured Figure 4-23 shows the time response of th e performance variable from the modified nonlinear fuel cell model when it cannot be directly measured. An LQG observer is incorporated in the control design to estimate the value of the performance variable by utilizing the values of the measured outputs (1) compressor mass flow rate Wcp, (2) stack voltage vst, and (3) pressure of supply manifold psm. The simulation conditions are identical to those given in Section 4.7.1.2, except that the altered model is used. The GP C controller and LQG observer designed for the original fuel cell model are used. The objective is to study the controller performance for model

PAGE 87

87 uncertainty when the performance variable is not directly measurable. The solid line is the time response of the performance variable from the modified model. Figure 4-23 shows that when the LQG observer in conjunction with the GPC controller is applied to the modified model, significant degr adation in performance is observed compared to the case when the performance variable is directly measured. This can be attributed to the fact that the GPC controller is attempting to regulate at the desired set point value of 2 the value of performance variable estimated by the LQG observer. However, the LQG observer is optimally designed for the original model at a nominal ope rating point which is different from the actual operating point and hence delivers an erroneous estimated value of the performance variable. In the case when the performance variable is direct ly measured, previously discussed in Section 4.7.1.1 the GPC controller successfully makes adjustments to minimize a correct value of the error. In the current case when the LQG observer is included, the GPC algorithm makes an effort to minimize an incorrect error. The LQG observer is optimally designed for the original linear model at the nominal point. In the curren t case not only is the nonlinear model used but the problem is amplified further by the fact that there is a deviation from the original nonlinear model by virtue of the uncertainty in one of its parameters. Figure 4-24 shows the time response of th e performance variable from the modified nonlinear fuel cell model when a battery of 4 ob servers, as described in Section 4.7, are employed to estimate the value of the performanc e variable. The solid line is the time response of the performance variable from the modified mo del. Due to the reasons, as discussed above, no significant improvement in controller performance is observed.

PAGE 88

88 4.8.3 Comparison of the GPC Strategy with Prior Control Designs In this section comparison of the GPC controll ers performance with controllers proposed in prior literature for fuel cell is conducted. Pukrushpant et al propose controllers for the following scenarios [31]: 1. All states are measured, leading to the implementation of the sFF controller with an LQR strategy for feedback. 2. All states are not measured, leading to the implementation of an LQG observer to estimate the value of the performance variab le in addition to the sFF controller with the LQR strategy for feedback. 4.8.3.1 Case where all states are measured-sFF with LQR feedback control Figure 4-25 shows the time response of the performance variable for the GPC strategy and the sFF with LQR feedback control strategy. Th e approach of sFF with LQR feedback control strategy is discussed in Section 4.5.3.1. In Figur e 4-25, the solid line is a plot of the performance variable as a function of time in the case wher e the GPC controller is used. The dashed line represents the performance variable trajectory wh en the sFF with the LQR feedback strategy is applied. The dotted line is for the set point, fixed at value 2. It is noted from the figure that the GPC cont roller delivers better performance as the sFF with the LQR feedback control strategy is not ab le to eliminate the steady-state offset. The GPC algorithm is able to eliminate offset as discus sed in Section 4.7.1.1. The LQR controller prescribes an optimal control input based on the li near model at the nominal point. However, the controllers performance is being examined then applied on a nonlinear model. Consequently, as there is deviation of the disturbance from th e nominal operating point the performance of the LQR controller degrades.

PAGE 89

89 4.8.3.2 Case where all states are not measured-observer design Figure 4-26 shows the time response of the performance variable when an LQG observer is incorporated in the controller in the scenario wh ere the performance variable is not measured. The dashed line indicates the time response of the performance variable when the static controller sFF with the LQG observer and the LQR feedback controller is applied. The controller strategy is disc ussed in Section 4.5.3.2. The solid line is the value of the performance variable when the GPC feedback with an observer is applied as the control strategy. It is noted from the time response curves that there is deterioration in both controllers performance compared to the case reported in Figure 4-25 where the performance variable is measured. Note, however, that the LQG observer with the GPC feedback controller exhibits better performance compared to the sFF with the LQG observer and the LQR feedback controller. The LQG observer component is identical for both th e controllers, consequently the strategy for estimating the value of the performance variable is alike. The GPC algorithm is able to regulate the estimated value of the performance variable in a superior fashion. The sFF with LQR feedback controller lacks the same level of dyna mic ability, and is not able to deliver the same level of regulatory performance that the GPC ap proach realizes. The LQR controller computes an optimal gain for the linear model at the nominal point. The implementation of the LQR strategy at non-nominal point i.e., on the nonlinear model degrades its performance. Figure 4-27 shows the time response of th e performance variable when a battery of 4 observers, as described in Section 4.7, are employed to estimate the value of the performance variable. The solid line is the time response of the performance variable. Due to the reasons, as discussed earlier, excellent results are obtained by using a battery of obs ervers which is evident by zero-offset/near zero-offset of the performance variable value from the setpoint.

PAGE 90

90 4.8.3.3 Comparison of controller performance with respect to robustness The controllers discussed in Sections 4.7.3.1 and 4.7.3.2 (the sFF with LQR feedback and the sFF with LQG observer and LQR feedback), are evaluated and compared with the GPC controller for their robustness to model uncertainty. An approach identical to that of Section 4.7.2 is adopted. The modified model is obtained from the original nonlinear model by altering the value of the return manifo ld throttle area from 0.0020 m2 to 0.0023 m2. The two scenarios considered are: 1. All states measured i.e., performance variable is measured. 2. All states not measured, i.e., an observer is included to estimate the value of performance variable. All states measurable, sFF with LQR feedback. The performance of the GPC controller and the sFF with LQR feedback controller is co mpared in the scenario of model uncertainty. The robustness of the controllers is examined by implementing them on models for which they were not originally designed. In the current case the performance variable is measured. The control strategies for the GPC and the sFF with LQR feedback controllers described in Section 4.6 and Section 4.5.3.1, respectively, are implemented on the modified nonlinear fuel cell model. The difference in the simulation scenario compared to the one presented in Section 4.7.3.1 is that the model used to describe the no nlinear fuel cell dynamics has a different return manifold throttle area. Figure 4-28 shows the ti me responses of the two control strategies. The dashed line indicates the performance variable response to the sFF with LQR feedback control strategy. The solid line is the performance va riable response to the GPC control strategy. The GPC controller delivers better performance by eliminating offset as discussed in Section 4.7.2.1. The sFF with LQR feedback does not deliver the same level of performance as an offset with respect to the set point is ob served. The relatively poo r performance of the sFF

PAGE 91

91 with LQR feedback strategy is expected as the feedforward and optimal gain are designed for a linear nominal point of the unmodified plant model. Performance variable not measured, sFF with LQG observer and LQR feedback. Figure 4-29 shows the time response of the perf ormance variable for the two control strategies (the LQG observer with GPC feedback and th e sFF with LQG observer and LQR feedback) in the case where the performance variable is no t directly measured and an LQG observer is incorporated in the control design to estimate its value. The LQG observer is identical to the sFF with LQR feedback and the GPC control strategies The dashed line indicates the time response of the performance variable when the sFF with LQG observer and LQR feedback controller is applied. The solid line is the performance vari able time-response when the GPC approach with LQG observer is implemented. The simulation co nditions are identical to the one discussed in Section 4.7.3.2, except that the modified m odel as discussed in Section 4.7.2 is used. In the case when the performance variable is not measured and the GPC strategy with LQG observer is implemented on the modified model, degradation in contro ller performance is observed as compared to when the performance va riable is measurable as discussed in Section 4.7.2.2. However, the GPC with an LQG observer control design delivers better performance (smaller offset) than that of the sFF with LQG observer and LQR feedback control design. This can be attributed to the fact th at the erroneous estimated value of the performance variable is fed as an input for the LQR feedback controller wh ich itself is being implemented on a model it was not designed for. Consequently, a relatively higher degradation in controller performance is seen on incorporation of the observer. The GPC desi gn performs comparatively better though it also fails to deliver offset free behavior. As mentio ned before, the observer is identical for the two controllers. In the case of the sFF with LQR feedback strategy, the situation is compounded

PAGE 92

92 further by the fact that the LQR gain is not only being calculated at the non-nominal point (nonlinear model) but on a modified nonlinear model. Figure 4-30 shows the time response of th e performance variable from the modified nonlinear fuel cell model when a battery of 4 ob servers, as described in Section 4.7, are employed to estimate the value of the performanc e variable. The solid line is the time response of the performance variable from the modified mo del. Due to the reasons, as discussed above, no significant improvement in controller performance is observed. 4.8.4 Feedforward Control Designs Two exclusively feedforward control designs, namely (1) static feedforward (sFF) and (2) dynamic feedforward (dFF) without feedback components are proposed by Pukruspan et. al. [31]. The two controllers performance is comp ared with the GPC strategy for the following two cases: 1. Application on the original model 2. Application on the modified model to assess and compare robustness to model uncertainty. 4.8.4.1 Case of original model Figures 4-31 and 4-32 compares the performance of the sFF and dFF controllers with the GPC algorithm on the unaltered original nonlinear fuel cell model for the following two scenarios: (1) the perfor mance variable is measured, and (2) the case when it is not. In the scenario where the performance variable is not m easurable, an LQG observer is included in the GPC control design. The sFF and dFF controllers are discussed in Se ctions 4.5.2.1 and 4.5.2.2, respectively. The GPC approach and the GPC with the LQG obse rver strategies are implemented as described in Sections 4.6.1 and 4.5.3.2, respectively, a nd the simulations are conducted as discussed in

PAGE 93

93 Section 4.7.1. The solid and dashed black line s represent the time responses of the performance variable in the cases when the sFF and dFF controll er are applied, respectiv ely. The red and blue lines indicate the time responses of the perfor mance variable when the GPC strategy is adopted for the scenarios where the performance variable is measured and not measured, respectively. In the case when the performance variable is not measured, an LQG observer is included in the control design to estimate its value. In the case when the performance variable is measurable, the GPC controller delivers the best result by eliminating offset with respect to the set point as discussed in Section 4.7.1.1. In the case when the performance variable is not m easurable and an LQG observer is incorporated in the control design there is a relative degradatio n in control performance, compared to the case when it is measured, as discussed in Section 4.7. 1.2. The performance of the dFF controller is the worst observed, as expected because the contro ller is designed for the linear model of a fuel cell at the nominal point. As the deviation fr om the nominal point increases, the controller dynamics of the dFF design do not correspond to the model it was designed for. The sFF strategy performs better compared to dFF a pproach as the deviation from nominal point increases. This behavior is expected as th e sFF algorithm is derived from simulations and experimental results based on the nonlinear model. The sFF, dFF and GPC with observer strategies are not able to eliminate offset. Figure 4-33 shows the time response of th e performance variable when a battery of 4 observers, as described in Section 4.7, are employed to estimate the value of the performance variable. The estimated value is fed to the GPC controller. The solid line is the time response of the performance variable. Due to the reasons, as discussed earlier, excellent results are obtained

PAGE 94

94 by using a battery of observers which is evid ent by zero-offset/near zero-offset of the performance variable value from the setpoint. 4.8.4.2 Case of model uncertainty The robustness of the sFF and dFF controll ers to model uncertainty is assessed by implementing them on a modified model derive d by modifying a parameter in the original model, as explained in Section 4.7.2. The performance of the sFF and dFF controller is compared to that of the GPC approach. The tw o cases for the GPC strategy considered are: (1) the performance variable is measured, and (2) the performance variable is not measured in which case an LQG observer is included in the control design to estimate its value, as explained in Section 4.5.3.2. The sFF, dFF, GPC (with and without observer) co ntrol strategies designed for the original model are applied to the modified nonlinear model. The simulations are performed in an identical manner to that explained in Section 4.7.4.1, except that the modified nonlinear model was adopted. Figures 4-34 and 4-35 shows th e results of the simulations. The solid lines are the traces for the performance variable in the case wh en GPC approach is employed, respectively implemented for the scenarios (1) and (2) describe d above. The solid and dashed lines represent the time responses of the performance variab le for the sFF and dFF control strategies, respectively. Figure 4-34 shows that in the case where th e performance variable is measured the GPC controller delivers the best result by eliminating offset of the performance variable with respect to the set point, as discussed in Section 4.7.2.1. However, due to reasons discussed in Section 4.7.2.2, in the case when the performance variable is not measured, there is a relative degradation in the GPC controller performance. The performance of the sFF an d dFF controllers degrades as

PAGE 95

95 well when they are implemented on the modified fu el cell model. The dFF controller performs the worst due to reasons discussed in Section 4.7. 4.1. Additionally, in the case of dFF controller the problem is further compounde d by the fact that, besides being applied at a non-nominal operating point (adopting nonlinear model), a modified model is used. Figure 4-36 shows the time response of th e performance variable from the modified nonlinear fuel cell model when a battery of 4 ob servers, as described in Section 4.7, are employed to estimate the value of the perfor mance variable. The estimated value of the performance variable is fed to the GPC controller. The solid line is the time response of the performance variable from the modified model. Due to the reasons, as discussed earlier, no significant improvement in contro ller performance is observed. 4.9 Conclusions An elegant solution is proposed for the problem of fuel cell control by implementing the GPC scheme. The GPC aproach employs the augmented model which incorporates the measured disturbance in its algorithm. The GPC controller demonstrates better performance compared to the control strategies proposed in literature for various scenarios. The GPC strategy is the best performing controller regulating the value of the performance variable at the desired set point of 2 with zero offset in all cases when it is measurable. The GPC controller demonstrates the highest level of robustness towa rds the issue of model uncertainty by exhibiting zero offset of the performance variable with resp ect to the set point when applied to a modified fuel cell model.

PAGE 96

96 battery Fuel cell stack Motor Compressor Humidifier Water Tank Water Separator2O cmvstvH2 Tank PowersmpcpWstI Figure 4-1. Schematic of fuel cell system.

PAGE 97

97 Fuel Cell System stIw cmvu netPez 122 Oz cpWy 1smpy 2stvy 3 Figure 4-2a. Fuel cell system showing input u disturbance w and outputs z1, z2, y1, y2, y3.

PAGE 98

98 Fuel Cell System stIw cmvu z y static Feedback + Figure 4-2b. Fuel cell system showing sFF with feedback controller.

PAGE 99

99 A -12.62 0 -10.95 0 8.4e-7 6e-18 0 2.4e-7 0 -315.8 0 0 1.004e-6 0 -35.34 0 -37.57 0 -46.31 -8.6e-24 2.76e-6 2e-17 0 1.58e-6 0 0 0 -17.19 0.2032 0 0 0 2.6e8 0 2.97e8 379.4 -38.7 1.06e7 0 0 33.28 0 38.03 4.834e-5 -4.8e-6 0 0 0 0 -295.6 0 0 9.33e-7 0 -63.61 0 4.045e8 0 4.621e8 0 0 0 0 -51.22 B -3.16e-5 0 -3.98e-6 0 0 0 0 405.1 0 0 0 0 -5.245e-5 0 0 0 C 4.94e6 1.967e6 -1.089e5 2.066 0 0 0 0 -1273 0 -1454 -4.3e-22 1.388e-4 9.94e-16 0 0 0 0 0 4.834e-5 -1.16e-6 0 0 0 0 0 0 0 1 0 0 0 2.59e4 1.03e4 -569.9 0 0 0 0 0 D 180.2 -165.7 -0.01049 0 0 0 0 0 -0.2965 0 Figure 4-3. Matrices defini ng the LTI model for the fuel cell model excluding sFF. A -12.62 0 -10.95 0 8.4e-7 6e-18 0 2.4e-7 0 -315.8 0 0 1.004e-6 0 -35.34 0 -37.57 0 -46.31 -8.6e-24 2.76e-6 2e-17 0 1.58e-6 0 0 0 -36.8 0.2032 0 0 0 2.6e8 0 2.97e8 379.4 -38.7 1.06e7 0 0 33.28 0 38.03 4.834e-5 -4.8e-6 0 0 0 0 -295.6 0 0 9.33e-7 0 -63.61 0 4.045e8 0 4.621e8 0 0 0 0 -51.22 B -3.16e-5 0 -3.98e-6 0 0 0 716.4 1065 0 0 0 0 -5.245e-5 0 0 0 C 4.94e6 1.967e6 -1.089e5 4.131 0 0 0 0 -1273 0 -1454 -4.3e-22 1.388e-4 9.94e-16 0 0 0 0 0 4.834e-5 -1.16e-6 0 0 0 0 0 0 0 1 0 0 0 2.59e4 1.03e4 -569.9 0 0 0 0 0 D -112.9 -435.7 -0.01049 0 0 0 0 0 -0.2965 0 Figure 4-4. Matrices defining the LTI model for the fuel cell including sFF.

PAGE 100

100 Fuel Cell System stIw cmvu netPez 122 Oz cpWy 1smpy 2stvy 3 static Figure 4-5. The sFF control conf igurations for fuel cell system.

PAGE 101

101 (a) Fuel Cell System stIw cmvu netPez 122 Oz cpWy 1smpy 2stvy 3 dynamic (b) + + u uwKw wzG2uzG22z Figure 4-6. The dFF controller: (a)Schematic diagram, and (b)transfer function representation.

PAGE 102

102 Fuel Cell System stIw cmvu z y sFF Feedback + Figure 4-7. The sFF schematic with feedback controller.

PAGE 103

103 )( tu )( td )( tm)( ~ ty)( tyR ~ /1S ~ ~ T ~ )( td + )( )( ty td )( tyu)( tyd + + d dA B u uA B Figure 4-8. The GPC design in feedback block diagram.

PAGE 104

104 0 5 10 15 20 25 30 90 155 220 285 350 Time (sec)Stack Current (Amp) Figure 4-9. Disturbance profile used for simulation purposes.

PAGE 105

105 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC Figure 4-10. The GPC control strategy implementation on the nonlinear fuel cell model in the case when the controlled variable is measured.

PAGE 106

106 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with 4 observers Figure 4-11. The GPC feedback with four observers control scheme implementation on the nonlinear fuel cell model.

PAGE 107

107 0 5 10 15 20 25 30 0 2 4 6 8 10 12 x 104 Time (sec)Normerror Observer 1 Observer 2 Observer 3 Observer 4 Figure 4-12. The Norm of errors from the battery of observers.

PAGE 108

108 0 5 10 15 20 25 30 Obs 1 Obs 2 Obs 3 Obs 4 Time (sec)Observer selected Figure 4-13. The switching patter n of the battery of observers.

PAGE 109

109 0 5 10 15 20 25 30 80 100 120 140 160 180 200 220 240 Time (sec)Voltage to compressor (volts) Figure 4-14. Final voltage to the compressor.

PAGE 110

110 0 10 20 30 .04 .02 0 0.02 0.04 Time (sec)Wcp error (kg/s) zero error Observer 1 0 10 20 30 0 2 4 6 8 x 10 4 Time (sec)Psm error (pascal) zero error Observer 1 0 10 20 30 0 5 10 15 20 Time (sec)Vst error (volt) zero error Observer 1 Figure 4-15. Observer 1, error between measured and estimated values.

PAGE 111

111 0 10 20 30 .015 .01 .005 0 0.005 0.01 Time (sec)Wcp error (kg/s) zero error Observer 2 0 10 20 30 0 5000 10000 15000 Time (sec)Psm error (pascal) zero error Observer 2 0 10 20 30 0 1 2 Time (sec)Vst error (volt) zero error Observer 2 Figure 4-16. Observer 2, error between measured and estimated values.

PAGE 112

112 0 10 20 30 .02 .01 0 0.01 0.02 Time (sec)Wcp error (kg/s) zero error Observer 3 0 10 20 30 0 5 x 10 4 Time (sec)Psm error (pascal) zero error Observer 3 0 10 20 30 0 5 Time (sec)Vst error (volt) zero error Observer 3 Figure 4-17. Observer 3, error between measured and estimated values.

PAGE 113

113 0 10 20 30 .03 .02 .01 0 0.01 Time (sec)Wcp error (kg/s) zero error Observer 4 0 10 20 30 0 5 x 10 4 Time (sec)Psm error (pascal) zero error Observer 4 0 10 20 30 0 10 Time (sec)Vst error (volt) zero error Observer 4 Figure 4-18. Observer 4, error between measured and estimated values.

PAGE 114

114 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with 3 observers Figure 4-19. The GPC feedback with thre e observers control scheme implementation on the nonlinear fuel cell model.

PAGE 115

115 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with 2 observers Figure 4-20. The GPC feedback with two observers control scheme implementation on the nonlinear fuel cell model.

PAGE 116

116 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with 1 observer Figure 4-21. The GPC feedback with one observer control scheme implementation on the nonlinear fuel cell model.

PAGE 117

117 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC Figure 4-22. The GPC control strategy implementation on the nonlinear fuel cell model with a parameter changed from the value used for control design.

PAGE 118

118 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with observer Figure 4-23. The GPC controller with the LQG observer control strategy implementation on the altered nonlinear fuel cell model.

PAGE 119

119 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with 4 observers Figure 4-24. The GPC controller with the 4 observers control strategy implementation on the altered nonlinear fuel cell model.

PAGE 120

120 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC sFF with LQR Figure 4-25. Comparison of the GPC c ontrol strategy with the sFF controller combined with LQR feedback strategy on the unaltered nonlinear fuel cell model when the performance variable is measurable.

PAGE 121

121 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with observer sFF with LQG observer and LQR Figure 4-26. The sFF with the LQG observer and LQR feedback, compared to GPC with the LQG Observer control stra tegy implementation on the unaltered nonlinear fuel cell model when the perf ormance variable is not measurable.

PAGE 122

122 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with 4 observers sFF with LQG observer and LQR Figure 4-27. The sFF with the LQG observer and LQR feedback, compared to GPC with the 4 observers control strategy implementation on the unaltered nonlinear fuel cell model when the perf ormance variable is not measurable.

PAGE 123

123 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC sFF with LQR Figure 4-28. The sFF with the LQR feedback, compared to GPC, when performance variable is measurable on the altered nonlinear fuel cell model.

PAGE 124

124 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with observer sFF with LQG observer and LQR Figure 4-29. The sFF with the LQG observer and the LQR feedback compared to the GPC with the LQG observer control stra tegy on the altered nonlinear fuel cell model.

PAGE 125

125 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with 4 observers sFF with LQG observer and LQR Figure 4-30. The sFF with the LQG observer and the LQR feedback compared to the GPC with the 4 observers control stra tegy on the altered nonlinear fuel cell model.

PAGE 126

126 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC dFF sFF Figure 4-31. The sFF and dFF strategies and the GPC control st rategy, performance compared when applied on the una ltered nonlinear fuel cell model.

PAGE 127

127 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with observer dFF sFF Figure 4-32. The sFF and dFF strategies and the GPC control strategy with the LQG observer, performance compared when applied on the unaltered nonlinear fuel cell model.

PAGE 128

128 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with 4 observers dFF sFF Figure 4-33. The performances of the sFF and dFF strategies and the GPC with the 4 observers control strategy compar ed when applied on the unaltered nonlinear fuel cell model.

PAGE 129

129 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC dFF sFF Figure 4-34. The sFF and dFF strategies and the GPC control st rategy, performance compared when applied on the al tered nonlinear fuel cell model.

PAGE 130

130 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with observer dFF sFF Figure 4-35. The sFF and dFF strategies and the GPC control strategy with the LQG observer, performance compared when applied on the altered nonlinear fuel cell model.

PAGE 131

131 0 5 10 15 20 25 30 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Time (sec)O2 Excess Ratio setpoint GPC with 4 observers dFF sFF Figure 4-36. The performance of sFF and dFF strategies and the GPC with 4 observers control strategy compared when applie d on the altered nonlinear fuel cell model.

PAGE 132

132 CHAPTER 5 CONCLUSIONS AND PROPOSITIONS FOR FUTURE WORK 5.1 Conclusions A Virtual Control Lab (VCL) with an inverted pe ndulum that can be utilized as a tool for enhanced learning is described. The VCL can be used in conjunction with a process control lecture for demonstrating various advanced conc epts or can be used by students located in geographically remote places. The animation modul e of the VCL allows the user to visually observe the impact of the control design. The VCL can potentially minimize the problem of scheduling laboratory time for physical equipment. The problem of tracking the slope of ramp is ex amined. On the basis of results obtained, a PI controller using the ITAE as the optimizing metric demonstrates the best results. The controller performance is better than that of controllers proposed in literature. Tuning relationships for PI and PI2 controllers, using three differe nt optimizing metrics and 20,000 different plants is presented. An elegant solution is proposed for the prob lem of fuel cell control by implementing a GPC strategy which incorporates disturbance m easurements to produce a manipulated variable. The GPC design demonstrates better performance co mpared to control strategies available in prior literature. The GPC controller results in zero offset in the performance variable in the nonlinear model when the performance variable is m easured. The GPC controller is also the best performing controller when evalua ted for model uncertainty. The c ontroller exhibits zero offset, showing its strength from a view point of robustness, when employed on a modified model. In the case when the performance variable is not m easurable, a battery of observers is implemented to estimate the value of the performance variable deliver the best result. However, the robustness problems still exist.

PAGE 133

133 5.2 Future Work To validate the benefits and effectiveness, it is proposed that the VCL is tested as a pilot teaching tool in process control classes taught at both undergraduate and graduate levels. The feedback obtained from the instructors and students will be highly beneficial in improving and incorporating features that coul d enhance the learning experience. It would be desirable to add to the VCL a few more advanced control strategies such as the GPC approach. For future work regarding the ramp tracking problem it is proposed to validate the simulation results for physical setups that can take advantage of the tuning relationships presented. It is also suggested to conduct a mo re comprehensive study to minimize the scatter in the optimizing metric error for low / values of the ratio, that is, for systems with little dead time. For future work on the fuel cell control problem it is proposed to evaluate the design of observers which take model uncertainty into accoun t. A battery of observers can be designed at each operating point, with each observer corres ponding to a linearized model obtained from different values of parameters. Then a bumpless switching strategy could be designed to select an appropriate state estimate to feed to the co ntroller. The experimental implementation of the control scheme in a physical fuel cell may be of si gnificant value to confirm or refine the results discussed in this dissertation.

PAGE 134

134 APPENDIX A OFFSET BETWEEN AUXILLIARY AND ORIGINAL RAMP The final value theorem is used to calculate the shift between the original ramp and the modified ramp. The transfer function for the process with time delay is given by s pe s K G 1 (A-1) and the transfer function for PI controller is given by s KGI cc1 1 (A-2) and the equation for ramp in the time domain is ttr ) ( (A-3) Performing the Laplace transform of Eq A-3, with the initial condition 0 )0( tr yields 2)( s sr (A-4) Using Eqs. A-1 and A-2, it is possible to de rive the following standard closed loop relationship between the output y and the ramp r : )( 1 )( sr GG GG sycp cp (A-5) Now consider the error )()()(tytrte (A-6) defined as the difference between the set point and output. Applying final value theorem )()(lim)(lim0sysrstes t (A-7) and inserting Eq. A-4 and A-5 in Eq. A-7 yields

PAGE 135

135 2 2 01 lim)(lim s GG GG s stecp cp s t (A-8) or cp s tGGs te 1 1 lim)(lim0 (A-9) Inserting Eq. A-1 and Eq. A-2 in Eq. A-9 results in s Ke s K s teI c s s t 1 1 1 1 1 lim)(lim0 (A-10) Simplifying terms and applying the limits on the right hand side of Eq. A-10 yields I c tKK te 1 1 )(lim (A-11) Hence, c I tKK te)(lim (A-12) which is an analytical expression for the offset between the auxiliary and the original ramp.

PAGE 136

136 APPENDIX B OBSERVER DESIGN USING TRANSFER FUNCTION The observer equations are given by ) ( xCyLBuxAx (B-1) DuxCy (B-2) Eq. B-1 can be re-written as LyBuxLCAx )( (B-3) Performing Laplace Transform )()()( )()0( )( sLYsBUsXLCAtxsXs (B-4) This results ) 0( )()()( ) ( txsLYsBUsXLCAsI (B-5) On further manipulation ) 0( )()()( txMsMLysMBusx (B-6) where 1) ( LCAsIM Now if xCzz (B-7) ) 0( )()()( txMCsMLYCsMBUCsZz z z (B-8) ) 0( )()()( )( XGsYGsUGsZoxz yz uz (B-9) where MBCGzuz, MLCGzyz, and MCGz xz )0(

PAGE 137

137 LIST OF REFERENCES 1. T. F. Junge and C. Schmid, Web-based remote experimentation us ing a laboratory-scale optical tracker, Proceedings of the American Control Conference Chicago Illinois, June 2000. 2. D. Gillet, C. Salzmann, R. Longchamp, a nd D. Bonvin, Telepresence: an opportunity to develop practical experimentation in automatic control education, European Controls Conference, Brussels Belgium July 1997. 3. D. Gillet, F. Geoffroy, K. Zeramdini, A. V. Nguyen, Y. Rekik and Y. Piguet, The cockpit: An effective metaphor for Web-based experimentation in engineering education, International Journal of Engineering Education, 19(3), August 2002, pp. 389-397. 4. D. Gillet, G. Fakas, eMersion: a new pa radigm for Web-based training in engineering education, International Conference on Engineering Education Oslo, Norway, 2001, 84B-10. 5. A. Bhandari, M. H. Shor, Access to an instructional control laboratory experiment through the World Wide Web, Proceedings of the American Control Conference Philadelphia, June 1998. 6. Ch. Schmid, T.I. Eikass, B. Foss and D. Gillet, A remote laboratory experimentation network, 1st IFAC Conference on Telematics A pplications in Automation and Robotics, Weingarten, Germany, July 24-26, 2001. 7. H. Latchman, C. Salzmann, S. Thotap illy and H. Bouzekri, Hybrid asynchronous and synchronous learning networks in distance education, International Conference on Engineering Education Rio de Janeiro, Brazil, 1998. 8. B. Armstrong and R. Perez, controls la boratory program with an accent on discovery learning, IEEE Control Systems Magazine February 2001, pp. 14-20. 9. J. W. Overstreet and A. Tzes, Internet-b ased client/server virtua l instruments designs for real time remote-access cont rol engineering laboratory, Proceedings of the American Control Conference San Diego, 1999. 10. C. Schmid, The virtual control lab VCLab for education on the Web, Proceedings of the American Control Conference Philadelphia, June 1998. 11. O. D. Crisalle and H. A. Latchman, Virtual control laboratory for multidisciplinary engineering education, NSF Award No. DUE-9352523, proposal funded by the National Science Foundation under the Instrumenta tion and Laboratory Improvement Program, (1993). 12. C. S. Peek; O. D. Crisalle; S. Depraz and D. Gillet, The virtual control laboratory paradigm: Architectural design requirement s and realization th rough a DC-motor example, International Journal of Engineering Education 21(6), 2005, pp. 1134-1147

PAGE 138

138 13. National Instruments Corp., 11500 N. Mopac Expressway, Austin, TX 78759-3504, USA, www.ni.com 14. M. W. Dunnigan, Enhancing state-space control teaching with a computer-based assignment, IEEE Transactions on Education,44(2), May 2001, pp. 129-136. 15. P. C. Wankat and F. S. Oreovicz, Teaching Engineering McGraw-Hill, (1993). 16. A. M. Lopez; J. A. Miller, C. L. Smith and P. W. Murrill, Tuni ng controllers with errorintegral criteria, Instrumentation Technology, 14(11), November 1967, pp. 57-62. 17. C. Smith and A. Corripio, Principles and Practice of Automation Control. J. Wiley. (1997). 18. W. Brogan, Modern Control Theory 4th edn, Prentice-Hall, (1991). 19. P. W. Belanger and W. L. Luyben Design of low-frequency compensator for improvement of plantwide regulatory performance, Ind. Eng. Chem. Res. 36, 1997, pp. 5339-5347. 20. R. Monroy-Loperena, I. Cervantes, A. Mo rales and J. Alvarez-Ramirez, Robustness and parametrization of the proportional plus double-integral compensator, Ind. Eng. Chem. Res. 38, 1999, pp. 2013-2020. 21. C. S. Peek, Development of virtual control laboratories for control engineering education, double integral controllers for ramp tracking and predictive control for robust control and luenberger observers, PhD Candidacy Exam Dept. of Chemical Engineering, College of Engineering, University of Florida; 2006. 22. Mathworks Inc. MATLAB. Version 7.2.0.232, R2006a (2006). 23. S. Davis, Transportation Energy Data Book, 20th edn, U. S. Department of Energy, (2000). 24. W. Grove, A small voltaic battery of great energy, Philosophical Magazine, 15, 1839, pp. 287-293. 25. J. Amphlett, R. Baumert, R. Mann, B. Peppley, P. Roberage and A. Rodrigues, Parametric modelling of the performance of a 5-kW proton-ex change membrane fuel cell stack, Journal of Power Sources 49, 1994, pp. 349-356. 26. J. Amphlett, R. Baumert, R. Mann, B. Peppley and P. Roberage, Performance modeling of the Ballard Mark IV solid polymer electrolyte fuel cell, Journal of Electrochemical Society 142(1), 1995, pp. 9-15. 27. D. Bernardi, Water-balance calculations for solid-polymer-electrolyte fuel cells, Journal of Electrochemical Society 137(11), 1990, pp. 3344-3350. 28. D. Bernardi and M. Verbrugge, A mathem atical model of the so lid-polymer-e lectrolyte fuel cell, Journal of Electrochemical Society 139(9), 1992, pp. 2477-2491.

PAGE 139

139 29. J. Baschuk and X. Li, Modelling of pol ymer electrolyte membrane fuel cells with variable degrees of water flooding, Journal of Power Sources 86, 2000, pp. 186-191. 30. J. Eborn, L. Pedersen, C. Haugstetter and S. Ghosh, System level dynamic modeling of fuel cell power plants, Proceedings of th e American National Conference, 2003, pp. 2024-2029. 31. J. T. Pukrushpan, A. G. Stefanopoulou and H. Peng, Control of fuel cell power systems: principles, modeling, analys is and feedback design, Advances in Industrial Control Springer-Verlag London Limited, (2004). 32. D. Seborg, T. Edgar and D. Mellichamp, Process Dynamics and Control 2nd edn, Wiley, (2004). 33. K. Ogata, Modern Control Engineering 4th edn, Prentice Hall, (2001). 34. R. Stefani, B. Shahian, C. Savant and G. Hostetter, Design of Feedba ck Control Systems 4th edn, Oxford University Press, (2002). 35. O. D. Crisalle, D. E. Seborg and D. A. Mellichamp, Theoretical analysis of long-range predictive controllers, Proceedings of the American Control Conference Pittsburgh, Pennsylvania, 1989. 36. D. W. Clarke, C. Mohtadi and P. S. Tufts, Generalized predictive control part I. The basic algorithm, Automatica 23(2), 1987, pp. 137-148. 37. D. W. Clarke, C. Mohtadi and P. S. Tu fts, Generalized predictiv e control part II. Extensions and interpretations, Automatica 23(2), 1987, pp. 149-160. 38. D. W. Clarke and C. Mohtadi, Prop erties of generalized predictive control, Automatica, 25(6), 1989, pp. 859-875.

PAGE 140

BIOGRAPHICAL SKETCH Vikram Shishodia was born in Delhi, India. He graduated with a B.Tech. in chemical engineering from the Indian Institute of Technology Delhi in 1992. Mr. Vikram Shishodia joined the graduate program at the University of Florida in 1993. He graduated with an M.S. in materials science and engineering in 1996. After developing a successful career as a process engineer with Intel Corporation for nine years he returned to academia to pursue higher degrees in chemical engineering at the University of Florida. From 2005 to 2008 he was a graduate student and simultaneously worked as Assistant Director for the Division of Student Affairs at the College of Engineering of University of Flor ida. He graduated with an M.S. in chemical engineering in 2003 and a Ph.D. in chemical engineering from the University of Florida in 2008.