<%BANNER%>

Approximate Probabilistic Optimization of a Wingbox Model Using Exact-Capacity-Approximate-Response-Distribution (ECARD)

Permanent Link: http://ufdc.ufl.edu/UFE0022213/00001

Material Information

Title: Approximate Probabilistic Optimization of a Wingbox Model Using Exact-Capacity-Approximate-Response-Distribution (ECARD)
Physical Description: 1 online resource (57 p.)
Language: english
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: Mechanical and Aerospace Engineering -- Dissertations, Academic -- UF
Genre: Mechanical Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: There are two major obstacles that affect probabilistic (or reliability-based) structural optimization. First, uncertainties associated with errors in structural and aerodynamic modeling and quality of construction are not well characterized as statistical distributions and it has been shown that insufficient information may lead to large errors in probability calculations. Second, probabilistic optimization is computationally expensive due to multiple analyses, typically finite element analyses, for calculating probability as the structure is being redesigned. In order to overcome these obstacles, we propose an approximate probabilistic optimization method where the probabilistic calculation is confined to failure stress. This takes advantage of the fact that statistical characterization of failure stresses is required by Federal Aviation Administration regulations. The need for expensive stress distribution calculations are eliminated by condensing the stress distribution into a representative deterministic value, transforming a probabilistic optimization problem into a semi-deterministic optimization problem. By starting the approximate probabilistic optimization from the deterministic optimum design, a small number of iterations is expected and reliability analysis is required only once per iteration. This proposed method provides approximate sensitivity of failure probability with respect to the design variables, which is essential in risk allocation. This method is demonstrated in two examples. The first example uses a ten-bar truss which demonstrates the risk allocation between the truss elements. The second example uses a wingbox model based on a Boeing 767-400 aircraft which demonstrates the risk allocation between two different failure modes of stress and displacement.
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 (M.S.)--University of Florida, 2008.
Local: Adviser: Kim, Nam Ho.
Local: Co-adviser: Haftka, Raphael T.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022213:00001

Permanent Link: http://ufdc.ufl.edu/UFE0022213/00001

Material Information

Title: Approximate Probabilistic Optimization of a Wingbox Model Using Exact-Capacity-Approximate-Response-Distribution (ECARD)
Physical Description: 1 online resource (57 p.)
Language: english
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: Mechanical and Aerospace Engineering -- Dissertations, Academic -- UF
Genre: Mechanical Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: There are two major obstacles that affect probabilistic (or reliability-based) structural optimization. First, uncertainties associated with errors in structural and aerodynamic modeling and quality of construction are not well characterized as statistical distributions and it has been shown that insufficient information may lead to large errors in probability calculations. Second, probabilistic optimization is computationally expensive due to multiple analyses, typically finite element analyses, for calculating probability as the structure is being redesigned. In order to overcome these obstacles, we propose an approximate probabilistic optimization method where the probabilistic calculation is confined to failure stress. This takes advantage of the fact that statistical characterization of failure stresses is required by Federal Aviation Administration regulations. The need for expensive stress distribution calculations are eliminated by condensing the stress distribution into a representative deterministic value, transforming a probabilistic optimization problem into a semi-deterministic optimization problem. By starting the approximate probabilistic optimization from the deterministic optimum design, a small number of iterations is expected and reliability analysis is required only once per iteration. This proposed method provides approximate sensitivity of failure probability with respect to the design variables, which is essential in risk allocation. This method is demonstrated in two examples. The first example uses a ten-bar truss which demonstrates the risk allocation between the truss elements. The second example uses a wingbox model based on a Boeing 767-400 aircraft which demonstrates the risk allocation between two different failure modes of stress and displacement.
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 (M.S.)--University of Florida, 2008.
Local: Adviser: Kim, Nam Ho.
Local: Co-adviser: Haftka, Raphael T.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022213:00001


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 E20101106_AAAAGG INGEST_TIME 2010-11-07T02:17:36Z PACKAGE UFE0022213_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 26717 DFID F20101106_AABXEN ORIGIN DEPOSITOR PATH pippy_r_Page_39.QC.jpg GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
e6324f203c4627b168293a887a38a448
SHA-1
5b0afc71082b6537cc776d6fee26421195b15835
6902 F20101106_AABWXP pippy_r_Page_13thm.jpg
94945a952ba0224762743c3f6438c832
e99dda9528a3f9714c8b6f3b536d1df0d59a31f5
33846 F20101106_AABXJK pippy_r_Page_43.pro
cead23f96f4da5a3a85c72609c37c32e
7474c816b1d6066b18ca770a960a6269ea020420
43302 F20101106_AABXEO pippy_r_Page_31.jp2
0ac383547f8a905b832985120bd6ce44
a963167ef3bc62c9c50f48551ad6a9711add72a0
27981 F20101106_AABWXQ pippy_r_Page_22.jpg
f144f94cba585c04089162f9000d5b73
4982592091cb5a63f290a2f15240694427626ffd
4716 F20101106_AABXJL pippy_r_Page_47.pro
2b8646ef129a30edfefc84a65a26b5aa
fe3373905264e1707ce1cf227ea6fcb0b3ec3689
109406 F20101106_AABXEP pippy_r_Page_41.jp2
c6d2fbedf3c8a63d682875a0bf1b7aba
08859352ad38fad7b5e186766432a431ef399b44
5612 F20101106_AABWXR pippy_r_Page_20thm.jpg
c4834f1ff84ff7ca558e2f1c5c48993f
032c9197eab863774c385771b17123b7a045de0b
13105 F20101106_AABXJM pippy_r_Page_48.pro
7acf87a0ccf151f70bcaf3033cf01f11
ca580bef1a5fcfde5280973cd09ac80e555a4677
665627 F20101106_AABXEQ pippy_r_Page_51.jp2
78b938f48261b3fa465d1a0bffe54f8b
f0a47882232da88416540c2f99f6f49ed98d4893
1053954 F20101106_AABWXS pippy_r_Page_55.tif
aa580ccee0178157142528bde55b5c53
e0422cd11896e3e1d6f4d38fa900e2fa2bae3f14
19889 F20101106_AABXJN pippy_r_Page_49.pro
bbf547b9bd2c4b6b881af4712c04baef
ce21d076a13a8e4d081f693f22d84312ece42e65
2031 F20101106_AABXER pippy_r_Page_01thm.jpg
e5a78c1555e17887df79e39a776b9da5
8b06d6d92e6a6e4e809b0abb1f0b5fcfb6a57269
6630 F20101106_AABWXT pippy_r_Page_38thm.jpg
381b548ddc8f52c32bb68219bb86eebd
c0de815c4eca7507517660dbad1e4b1107aceb13
9169 F20101106_AABXJO pippy_r_Page_54.pro
c101f352264a323087d662a65067e1c8
5dcbd32bc35a07dda9ebc439dfb0a4d2eadfc0fa
1734 F20101106_AABWXU pippy_r_Page_43.txt
d02e15ac11268a7ab424de9ba5c14c12
7688afacf3371175de09d1b5ad4b188eb0c9eaac
90 F20101106_AABXJP pippy_r_Page_02.txt
1bece15ded308e4eb053c476b5ebb251
e75bf9ecdf103935cc35ce33aa7a08b116c72cb3
6457 F20101106_AABXES pippy_r_Page_43thm.jpg
aa063329b1de257753bf0d0b4431ad31
0f48a31762cd4089f5a467cb21bc31d29355d2de
F20101106_AABWXV pippy_r_Page_39.tif
e28f49eee2358f06392970d100a19325
6ed65db70867fc6807b741f4513662dd43d77b10
764 F20101106_AABXJQ pippy_r_Page_04.txt
280d7be788e2395206d5a5a7f2520cf9
bd6d3b80d1231d81d6c15dd6a86d3902918ddd57
368476 F20101106_AABXET pippy_r_Page_46.jp2
e68bec6e10a4ebfae9f02b0e06c50565
600d3558ae4336dbe0c87c6b5af999cdbaf6c039
1003 F20101106_AABWXW pippy_r_Page_30.txt
b3ffb23fc4d9375221222d82b733328e
f29a934854fe97e27dd203783d0eb7991c7352ce
3442 F20101106_AABXJR pippy_r_Page_05.txt
a6f622e6f1e88da4fd308f6893a7990b
a059cfe43e1259919ef04c5833aba4dbb98be167
32799 F20101106_AABXEU pippy_r_Page_36.QC.jpg
177a1a81410312372ce8963436f78d23
6612ca2d1298c167d23a258881424f8baa11d2f9
4404 F20101106_AABWXX pippy_r_Page_31thm.jpg
af87c86f721c6a735305757a0dead32c
cde66ed5283ce39994b19b239dbabb92e6b8e864
429 F20101106_AABXJS pippy_r_Page_06.txt
0b1bbd06730b8888a778c99d5ed41231
ebcf652c6291aed72f7ee8e99c510022c240b415
6935 F20101106_AABXEV pippy_r_Page_19thm.jpg
b12f45c65d628f2e44ebe127432ca85f
edf30e1de6b9be9e832ba95e4255ee99df0a82c4
F20101106_AABWXY pippy_r_Page_43.tif
3da409b89280d22f38e5f36ca344b975
98b136c1a65618f9ed80d4312eb0dc9e2dd19fed
404 F20101106_AABXJT pippy_r_Page_10.txt
54f52a55f628c16287a91bed1dcc8283
6d032313528f16b11a4f970c2cfd6cf5ca20ddc3
76943 F20101106_AABXEW pippy_r_Page_34.jpg
d93b4c2d5e688b4fb1ff136f8b2c857a
86b54719b01db610d801a34e154b852bc1765b50
34789 F20101106_AABWXZ pippy_r_Page_14.QC.jpg
b160dffbd8d10d37db6a260c6505cced
9e82d248b6e15f7ed7a4afc5229329df7fbed2ed
2229 F20101106_AABXJU pippy_r_Page_11.txt
92ffc10c67db775fb31f9210ac7b2bae
537ee064ce0372fb7629a330a0dc3ce3b7acef21
48280 F20101106_AABXEX pippy_r_Page_09.pro
96585d42fc6bd6bf0fbe39c4a5ae105c
93c54154db57256d2dbeb480cddce72956cbccd9
1767 F20101106_AABXJV pippy_r_Page_13.txt
5c2a81b1e03a58a7f1b086f61407c888
200d9b9001679c14c60e4b5997dad1590111f403
2028 F20101106_AABXJW pippy_r_Page_14.txt
ca47511e69c97e2cfed17e231da1039f
00d5355139f2b64c752bac200d9be7260ec46548
11782 F20101106_AABXCA pippy_r_Page_51.QC.jpg
988822a077587f7ab06162684e3141ae
7df041eda720939ee4cbb143d7336950d0a3ec70
52325 F20101106_AABXEY pippy_r_Page_15.pro
9e5eca571c79c7b85335e7c0e5eb34c1
626bf89165d241ebf25f505073c55713439a5c26
1561 F20101106_AABXJX pippy_r_Page_20.txt
d1414b175125d10472b0d4086425cd32
d6d830a1c11465599cc879eb8ce7fcd9ad749aba
1642 F20101106_AABXCB pippy_r_Page_08.txt
d6693726170b5727cb214dcbd3c356a7
0c4574bd454db4bfa67b80a1d5fa99d73506c446
1874 F20101106_AABXEZ pippy_r_Page_17.txt
897b1e63cccaa7a33b44aa37589729d4
2b02eddad2a496b2054148fa2c530859a3d7aa39
711 F20101106_AABXJY pippy_r_Page_21.txt
a29f896231dbed9bad1f3146ccdee858
a438ae321dce969b129a7e11cbec06396825392b
604269 F20101106_AABXCC pippy_r_Page_45.jp2
f1b87e2d390934fd1ce8b1cfc1cd0a80
15e487b53077bbda979f19338482bb835836fcda
38823 F20101106_AABXHA pippy_r_Page_53.jpg
cb6b9d0e1da38a0b389023bdb4ca69c8
bc62cf339fda96f63613c47444e64323b9d1d9c8
676 F20101106_AABXJZ pippy_r_Page_22.txt
85f1b75796a8bf3400365702de915301
4058a9b4cd1e29842993a8bd930b0a0d99d0fb8c
16847 F20101106_AABXCD pippy_r_Page_44.QC.jpg
1cc565242c2552e5c77cf7661b8c6a73
51cf8fbaba4e520f5c7e2890086a7e8fcca8a161
26386 F20101106_AABXHB pippy_r_Page_01.jp2
0204f5d831a685b0c240a147366483fe
083f72de5af19bac9e2f2a8b8ef707fd4b90a774
25271604 F20101106_AABXCE pippy_r_Page_47.tif
7efb7f2efd3a72ed6ec18def2a1229a6
19d8e9bf7c16d4711053d52f5a7199d87cb94b4e
42397 F20101106_AABXHC pippy_r_Page_04.jp2
db04d7a95dd34d59375ed7a50c8009b2
4b72ac3a606bea2e9d3734081067735cbef6af52
F20101106_AABXCF pippy_r_Page_49.tif
07f266087f044c94f1589fe23c41b2a0
378a913361e606bca3a4bd1477893f41193f23e3
160974 F20101106_AABXHD pippy_r_Page_06.jp2
c63ad0706faa6126788f4ba2630b0d87
3fab4bb4eb0c57ad39b9afe852584664e60c4cec
447 F20101106_AABXCG pippy_r_Page_46.txt
55bd71d6578e848ce6a87d100edeaa05
a3a9d7838df4fda220c5ef7510acb66222fdaf0c
1051966 F20101106_AABXHE pippy_r_Page_08.jp2
375ea5d05a96ba895f8ff6b62c17efa9
65865d24ee7768e6834f663b9287cf3b83772ce4
50713 F20101106_AABXCH pippy_r_Page_27.pro
d488509ba28df5eaa3861a5f5dd73054
add869d6f54a2511bd89317a224aeedbc943b021
25356 F20101106_AABXHF pippy_r_Page_10.jp2
b6b202b3345197f605e94234d592b685
c786b93cf1539e6b11cb0e9ac99dde1055c05952
4841 F20101106_AABXCI pippy_r_Page_07thm.jpg
82e8f006489d7f3312ed6f28cde81e3e
e6bfed48f79bf1ed4c9906da3d3148be1d5d4917
116275 F20101106_AABXHG pippy_r_Page_11.jp2
045920cc26863c381c949743f048a3e8
3db579e5957a8edef03610f7b85e21d9a7faa92b
7124 F20101106_AABXCJ pippy_r_Page_06.pro
60e626aa3cabb8ee7bb3ed3031aa94b9
64555d189a7dddaca69304d33868e7cbf81a39a6
85246 F20101106_AABXHH pippy_r_Page_13.jp2
722d8964b1a0d1fa81147de2d6d2fd83
217751ea850df30fc6a7ba52ba658c44f91518b5
885 F20101106_AABXCK pippy_r_Page_02.pro
1eecd9fcf0debddf71e78149f7495037
4d5116bfc48166b7a103d8705bf68df486623148
110403 F20101106_AABXHI pippy_r_Page_15.jp2
691bba7d20d7cd09efd3e80d0d96fab2
bc8a5beec2d9204e2204c4c93412cd1c0104820f
1521 F20101106_AABXCL pippy_r_Page_18.txt
ce1f2efc4ace46d306fc7fcb5a3891ad
2bd65089f3133f28a2ed3701ea115ed73dfa40c2
94387 F20101106_AABXHJ pippy_r_Page_16.jp2
62e3bfe46279d6c5b32663761e7d6788
492b09b879a50183084438b373ac28c66361b55b
5286 F20101106_AABXCM pippy_r_Page_29thm.jpg
215d79a3196a37c62a28334a469eb779
57a0eb8dfa5661bfdc6f34130ad903f3c73a1381
67456 F20101106_AABXHK pippy_r_Page_20.jp2
abb073cf5f62af92409652ec3a3ed086
4c47b605ae1f918c8f3fbcd4630ea8fdc6b814dd
85347 F20101106_AABXCN pippy_r_Page_12.jpg
4a90c3f6382a425f2f55895f37dabf05
0ec5db0c7cb47079371cdbc196f93d016dda1eed
100310 F20101106_AABXHL pippy_r_Page_25.jp2
8c69ff76403c77cee94f149af04c1b26
338299654f272706b326c64ed0ba154af8e75ed4
F20101106_AABXCO pippy_r_Page_15.tif
d7ba62783533f600a9e079daadcfdbcf
45382a85c0d63ec415b885f12bc0289890aa276e
106047 F20101106_AABXHM pippy_r_Page_27.jp2
00f2672ab15259007285de3159844c7d
2ddbb9d9eb791722dc1dd7e65d0fbbc4edd6affd
849 F20101106_AABXCP pippy_r_Page_50.txt
ad15ff1856d070636b623f110df45a4a
aa0debc4f5b4e66f5581dd790bd1820b32f9c59b
107450 F20101106_AABXHN pippy_r_Page_28.jp2
5a0f2372474d220b00b99b6d9253e273
4a40c32ec99bf8fa34dd05e1f6b9ee7f76adc31f
29467 F20101106_AABXCQ pippy_r_Page_01.jpg
19921ef8302d5d53003b3fff0a8d5154
e620f05c82bb8708dc8d2ebcd1f7009ee65b32dd
61655 F20101106_AABXCR pippy_r_Page_08.jpg
9d547194a3d6148b54901adca1752d36
d5ff169ad03d5eef93a7a9d0ef98343c1f5d553c
44424 F20101106_AABXHO pippy_r_Page_30.jp2
e8916c60dd2c340b85ca7190e225d1fb
c1bc9c3edd670308a5e8e275be0e100c9b804265
5956 F20101106_AABXCS pippy_r_Page_03.jpg
d1db15deb25b290e5117dcd40e37999b
95b160b399f67adaf473b82520fc02ecd56f054b
62714 F20101106_AABXHP pippy_r_Page_33.jp2
4374013a019ed5e612ee071da0c318a6
cb3310dbb8f7479c8e68969e2589bf0684a17034
32604 F20101106_AABXCT pippy_r_Page_32.jp2
8c95d44fe5b49dbbecd4104d0bac7b25
3abaf50545ea80b2a9bb8cb6beb8d1fa165490a2
108595 F20101106_AABXHQ pippy_r_Page_37.jp2
f47639033d115e2e3b82bbb5f7b2cbf6
f2c23393386b4684dbea3a970b32dda8d1948cd0
F20101106_AABXCU pippy_r_Page_38.tif
e2af19fdcb98091d5527d1b011fdecc2
704f4c697d3819431e6942ac7b53505c22e9be8d
354030 F20101106_AABXHR pippy_r_Page_50.jp2
f63e7c1cbb3fa0b97b02c8b52d6431e3
e2555ca5d65743992470e331446c1e6fd715e489
7578 F20101106_AABXCV pippy_r_Page_24thm.jpg
70f657a423ac35de0d0a55d692f5baa0
3ff34cb48d6f5c88039e6832ec1db9d99eb73a79
41220 F20101106_AABXHS pippy_r_Page_53.jp2
5a1a6be40db5cb15533f071bbf424180
2afdc186dc2a3fa0c0a1de849460589a8cdc3a44
20053 F20101106_AABXHT pippy_r_Page_54.jp2
db3ac962f9375263eae3d859ecf55ce5
20d72fd993b3abfc3dd80d1c4dbbbb87ab563736
39129 F20101106_AABXCW pippy_r_Page_51.jpg
1c99066f5fe0c0a22a31de5195d49b95
3c9013c32640e6bf5006517fda4b3b00c30beda2
101158 F20101106_AABXHU pippy_r_Page_56.jp2
9e4e5b34fd732ded8e5a287c3edd43ef
97f7e0d4abc5ab267384c8ecbd1f9490a832cd42
F20101106_AABXCX pippy_r_Page_01.tif
0a203487fb34967ff197917dcf9595dd
2685889c484799ba47787129817a79991361d093
F20101106_AABXHV pippy_r_Page_02.tif
245a7903c407450523711467c2298e9f
3c5f73d717167826fb343b873a81c78d927bf4be
85444 F20101106_AABXAA pippy_r_Page_05.pro
e6429b3fe8ef1ace0fb3e36533254c75
adf54d825a0cbe509e63431005cdd19dacccc6df
83048 F20101106_AABXCY pippy_r_Page_19.jpg
324727b28d1d03dd5d65c1776a876663
333fadcd4e2ce89234ef51ea0d1cbcb52c60b170
F20101106_AABXHW pippy_r_Page_03.tif
07d1d7763b10f0c70e44aecf97ed1724
990da276582cf017db8572f6322b2d1d5119e7c6
101898 F20101106_AABWYA pippy_r_Page_40.jpg
bd6ed49fee90da2cba3adad1074e2395
f93917393d6cb891cb729325f34d4e34c85173ec
99566 F20101106_AABXCZ pippy_r_Page_23.jp2
ff30f6333c2b87db28e496b93594478c
b93dbe6da6772513d53d59f5af4e2870e47dd927
F20101106_AABXHX pippy_r_Page_04.tif
aa48e050afcc79756a503635ef46fef8
c36e5990ed31885a2e0ea1af47c94217621e18e9
1860 F20101106_AABWYB pippy_r_Page_52.txt
3f75dab3bd3d4cfb86afc6461e24a5a5
55502b6e5c483f9702e68130aa613e73cfe5b459
100998 F20101106_AABXAB pippy_r_Page_09.jpg
fc99f8b2cbfa305d9edf3fbc55cdd104
ce69098a2d8b567f2908dd8113541b005c8fe3d3
F20101106_AABXHY pippy_r_Page_05.tif
a45b94fe5f5ad72f1b5061900894d4b2
5f96246c2a34adb546568bbdeec21956886ec727
95594 F20101106_AABWYC pippy_r_Page_17.jpg
b595d4b4cb7754089638961851742f27
7231a35ad5ce424aa28c7548c60dc356e2289c7c
106815 F20101106_AABXAC pippy_r_Page_26.jp2
26e40c218fbfe619e6ef4478a6ff4b12
8a53d1f56d961315326f42881e9622e5385b412b
99060 F20101106_AABXFA pippy_r_Page_28.jpg
c08a00abe61453fa30e30a76eda834d9
f9ed71d4752d14776c753863458554058669cadc
F20101106_AABXHZ pippy_r_Page_07.tif
40df9493a31f416e066fb2e89e2827f1
a652e0c778ccce3fb1246569b7cd60d98459a98d
36975 F20101106_AABWYD pippy_r_Page_47.jpg
baeeed61aa2543067c6f3df40397008a
b5878146042bf453dccc1b34e98d22628a4195a8
39903 F20101106_AABXAD pippy_r_Page_57.jp2
6685028096e80dfc17144fd93aa8c0d6
5157818fdcd6f23237b51bdefdaa7b70269b9f42
38160 F20101106_AABXFB pippy_r_Page_57.jpg
33c44c6c1785d515131f0a499ce4dfd4
9bab5dca47d8b1b454715fe897ea74636f018629
F20101106_AABWYE pippy_r_Page_53.tif
500d998dd4d3869c5bac20da129e5e16
39cf114ede67cc3e418b90103d362f6b4d4020f2
7039 F20101106_AABXAE pippy_r_Page_39thm.jpg
423219f3b5c8bf2f20eb56587fd36fa6
d527326a10478da3bdff0c5f35c17d48e18ec5f4
23779 F20101106_AABXFC pippy_r_Page_10.jpg
6dfc489c3dc63554bf66e90e1b1cb59a
6843b94f8c6fb0c8ba4786fa16300383518f1b90
10240 F20101106_AABWYF pippy_r_Page_22.QC.jpg
20b3632ceda4dcc2ac56dc86520150ac
3f649563be505e4c27c617655bd044b2f8d92fad
4605 F20101106_AABXAF pippy_r_Page_33thm.jpg
5ab2e2864da9193c35bd49236038dd6b
aabbdc3dadc4ed0f2c5a4e092b60987576041327
1981 F20101106_AABXKA pippy_r_Page_24.txt
47ed6949e2cc0af2e58a5f346ed86b96
5dc020a750c497de54fa804930fef83718c2f17b
323711 F20101106_AABXFD pippy_r_Page_35.jp2
1ccf9571ce5aea4c7a7124c6a6659719
ad2216e45aea678e78486eb815d260a6a1183856
2328 F20101106_AABWYG pippy_r_Page_54thm.jpg
9ec1f8593362d34735b876978b5b3552
ef3cf1437d5f15f4aeb22520a3596d76c1b706c6
5466 F20101106_AABXAG pippy_r_Page_49thm.jpg
13b235b297d115bf5e3015dc3a06e725
71e75e43a64000d8f8de19a9760fcb97b9cd7b61
2097 F20101106_AABXKB pippy_r_Page_27.txt
e5b5811007092ed3a5a53ad058f63a6b
12eb40a6d0d54e3608c26d9d99ef0a18a8ec6622
8343 F20101106_AABXFE pippy_r_Page_25thm.jpg
8f39f3724967f6368aee90cfbd46137b
882deef7d8fbe0f0c56acb318f24e581bfb9c52b
43666 F20101106_AABWYH pippy_r_Page_50.jpg
3872fb15a17170c646bc66375a7631c7
aef93ba43e79e92f827577812ddaebbe97ec7277
706931 F20101106_AABXAH pippy_r_Page_47.jp2
e8eb605da45881631e569879b8722e81
a626b334bee7f319c227f5da8a3b747e9313eb4a
1265 F20101106_AABXKC pippy_r_Page_29.txt
91086d26fb112efebbffd08e355b95d1
f3fd9b3137bd76f9380ecca64406019d0820cd92
31216 F20101106_AABXFF pippy_r_Page_09.QC.jpg
f5a9ccc6807053db88d021ff3f42f186
53acd6b1dad51de699b761e678ca4025a5085d11
70305 F20101106_AABWYI pippy_r_Page_29.jp2
790b98232d0251a13460358f991d21cd
a3b3554a346c4e00f308ed5dec2bc30262f6d08e
88269 F20101106_AABXAI pippy_r_Page_52.jp2
d9668a744b9baa4f11170c0600ce8480
5502aafe108a9164d3e796c15415539eb36539bc
771 F20101106_AABXKD pippy_r_Page_32.txt
b470a42229014427c2b8e13fa64485fe
9d3305d2a10c71e7d4fb897a36cb45ab9d57520b
7759 F20101106_AABXFG pippy_r_Page_10.QC.jpg
ef8589cbd552a2220d5eb033440d89ba
2c339368d9c66d3121ffe09c404b9f60518fc260
2101 F20101106_AABWYJ pippy_r_Page_15.txt
b572caa2137c64805338bb03dedc6adf
ee45cb834887da0a412983710092ccac7391a613
7001 F20101106_AABXAJ pippy_r_Page_12thm.jpg
3bce311db48a1313b25bee56081de2f9
1830bc9fef86f97cd6101c413445649cbdc4d60f
117 F20101106_AABXKE pippy_r_Page_35.txt
9a3401750ab46932b1fda71f013020b0
b850481a50c0e07423ac51d11600d1ff1ae61ed1
4381 F20101106_AABXFH pippy_r_Page_45thm.jpg
7f959dce1b534cf0a3a4ea2be6c9ee7f
a9d1385199d09ad5ddf4c193ece2f86af3e25f4e
527 F20101106_AABWYK pippy_r_Page_02thm.jpg
956c3a4cb0f77e2b917d2999d6f0269b
136ba4ffe27e24022ca3d761773b0603af874201
1617 F20101106_AABXAK pippy_r_Page_12.txt
91b5e600c258d6aae4894ee08dcadd3d
f1923392d8b7130522bcb4e257b5c4ee8b2f4bdc
2076 F20101106_AABXKF pippy_r_Page_37.txt
198297dbdc8afa2f032282bd9fd38bdd
323e3db58b64e5a6c7027d8c9c0d4e7ef73613ea
1051981 F20101106_AABXFI pippy_r_Page_07.jp2
9ac86c61d8e3db27d91e316ac77a72f3
f71c1db85ca12cb86b0e263d571ac1ac618c9899
1950 F20101106_AABWYL pippy_r_Page_56.txt
9410f9fca4f8b596d9da200a405f947d
ae82947e4897e4e3b01610fc9f0181b8541973f4
2053 F20101106_AABXAL pippy_r_Page_25.txt
31a9681c6ccd67a45b74d93330ab71eb
b392e8e7849f27c52559e5cebf6648d005097510
2119 F20101106_AABXKG pippy_r_Page_41.txt
8611c6c854facfb0903375f68305821f
107451a3e9a666010d1800af2eaeabcce83433ed
6013 F20101106_AABXFJ pippy_r_Page_18thm.jpg
b216f074bfb435fe6c31e2ad45952636
1a9632dd573cbd1a112ce9df56c31ef032bd71cc
3699 F20101106_AABWYM pippy_r_Page_51thm.jpg
11ec76d99a9e33d6d110b5ad4505959d
12c47b073ed8abfe6a592987b5cf58e4204f6dcf
2048 F20101106_AABXAM pippy_r_Page_40.txt
cb7a841038d283cca0f439ed615f5771
3a7bdcc0d2cf8c8cf803912c95487066aa2eb9e6
172 F20101106_AABXKH pippy_r_Page_44.txt
c6b5c4398a3f66a4fcf33402e23923a0
83f8afa9afe3831587bc6798d6540418030ef7be
7725 F20101106_AABXFK pippy_r_Page_17thm.jpg
882df49245a067c74ec1ec34a12f041b
da2a0e278aeae26cf3a4c9a4feba3b7b4f567581
2096 F20101106_AABXAN pippy_r_Page_09.txt
66b384e58bba381a15703c167ca27bff
006e0f41d0272efa15c8a0ce81b26435f0714960
270 F20101106_AABXKI pippy_r_Page_45.txt
a9501067ee60e540d3f307ae9dccc888
b6db1ede266db3851a32cd7931e91900a7c9f15a
962449 F20101106_AABXFL pippy_r.pdf
18fc5449e4f66b79cb2c9db714c65940
81ee91b6913aba43239de34954bcc41967f31c98
F20101106_AABWYN pippy_r_Page_21.tif
29c2a1b6bec8bdcf5f04814b6007b140
1ff395e209a63915a73349308845a54966272eba
23929 F20101106_AABXAO pippy_r_Page_43.QC.jpg
b56b130b906403431ce2ed1847be31fb
a51ab094f703d12b2f3c23882f2380baa8424c58
1013 F20101106_AABXKJ pippy_r_Page_49.txt
b366e3b6d07ddea445b5f9fad7b0ab2c
8da22463f89cdb9f6a349c060b9a007fbe564ded
1921 F20101106_AABXFM pippy_r_Page_39.txt
f1439b9b96e40a06920e7db1876e2cc0
2262573120ba69c4902bd6e9009147b7b2eaac6d
2139 F20101106_AABWYO pippy_r_Page_26.txt
b73adebff274b7f95042afa7446e7168
cdbf337a9ef63a24e1ccb21cb3481997e01950dc
8105 F20101106_AABXAP pippy_r_Page_46.QC.jpg
11ae3bb3764f798c5f086403e245d920
4fd2370d84882f1cd262260140028c4f810ce191
1159 F20101106_AABXFN pippy_r_Page_31.txt
a8ef94e3516e3023a57d9fd53c99ef96
48db716a6870023b35dfeff13b14c5bece8dd09f
1423 F20101106_AABWYP pippy_r_Page_42.txt
1443b348478016b6ddbcf0b36386c7e5
130127f8e43350549e5c25313915a1d277d6f7f4
F20101106_AABXAQ pippy_r_Page_37.tif
3e527837db7625dc2da41851027b7283
2ebbe08373e8a25645f9737c7055ef6432eb8e96
556 F20101106_AABXKK pippy_r_Page_54.txt
7401c60b3b4a8f3023d4309b1d5d3d97
99d6ef6da356d8f23e11993471a58d3c1a08e227
1460 F20101106_AABXFO pippy_r_Page_53.txt
bcadea6aa4108056d05ff5e8b7ea29e3
67ded280f111e9158ac2708f92df9098ae989219
42330 F20101106_AABWYQ pippy_r_Page_39.pro
fd17af1b1f1c9363adff3169b9485207
f2f493f679ea34011085be2057f1528f992ce431
33890 F20101106_AABXAR pippy_r_Page_32.jpg
2fb87dc6829f4aa3fe31325a348bf019
4f44ba286e690e68b5f5c203f1e5ed1c56048aae
2480 F20101106_AABXKL pippy_r_Page_55.txt
0217c1c330032f45feccb88190a0758c
4c6b83aae204cb823cb011a662d4401239ae770f
F20101106_AABXFP pippy_r_Page_32.tif
a2a9330e92ed7d1f38d685ac0362fe2b
b3d231956ced27527b5ae19429db439354764556
108080 F20101106_AABWYR pippy_r_Page_14.jp2
a735c8e68c9af1abe82b65735aa5334c
32139d60f8cd0fa296fac8e2b5ba210c5d60e69d
3450 F20101106_AABXAS pippy_r_Page_04thm.jpg
911b2503bda531abbeafa5f25ceb129e
0abe504dc0d8d754f2c347c745a484a52bf38471
15477 F20101106_AABXKM pippy_r_Page_50.QC.jpg
aec9fac2f497a12806fd3408a338350a
e3e60b6c615bf803831a86c53c0de2439429c865
3702 F20101106_AABXFQ pippy_r_Page_35thm.jpg
14e0f19e93030c5cb5ba505c841bb6bf
0a9d1fa67e527949ec41d26d33347a5c5033072d
27646 F20101106_AABWYS pippy_r_Page_12.QC.jpg
677a387c59d643f2ab48a6f43df8a332
611194ceeb18b1f23ab3a0f07816dc6ceefee6f8
F20101106_AABXAT pippy_r_Page_56.tif
da857b1805380ce06d91f1495deed4fa
5269941a9fea4d4eb5006a8e73459c090ac07ce9
23610 F20101106_AABXKN pippy_r_Page_42.QC.jpg
f85106e6014c87faf7410694edaaad72
4b70745e53712bb76dbdcfcba49b5bc882954617
11356 F20101106_AABXFR pippy_r_Page_47.QC.jpg
7f5f195e930985f21956820372ff7682
e2f2f30080e28237b5c44c57aa155d8380d5e442
33323 F20101106_AABWYT pippy_r_Page_18.pro
1b5f02ab7e55b78b90a03c9d1c944f12
3dfc4451efc2eb6ebe5851209b22c9571d0af264
30135 F20101106_AABXKO pippy_r_Page_16.QC.jpg
4868896dd75df2c64ec56056eccbade9
ea13694c69e5617b1c9159784fc5d25555c830be
41503 F20101106_AABXFS pippy_r_Page_07.pro
6f08bf4c0d094f67b3b96e01e588f6d4
96dbe8eb9a8c17298cdc1352553fba1823720cc1
98609 F20101106_AABWYU pippy_r_Page_36.jpg
9d2ed2daa6478d82340c3025c2d9ec4e
760627438e1fe39c7e1f2df40c9d992047c6b08c
18241 F20101106_AABXAU pippy_r_Page_04.pro
6092d140c263e6bfc5d436f3123c14e5
20db029eece6c9c8521d67c0c80f6d558e00ee2e
5588 F20101106_AABXKP pippy_r_Page_42thm.jpg
5742872fbc2c09ef7614ebb30ab7ef84
c7c2e7733f2519d912551cd82bce19e84a902105
29468 F20101106_AABXFT pippy_r_Page_35.jpg
c4504e0914aadbcf7fedbbf5fb683cb8
9f5bb77e5db0e90ecc60106a8d7bfeda776c94e7
F20101106_AABWYV pippy_r_Page_28.tif
0ac21c1d960b111d86322220f4f9ea71
22d2469fe0420aa5771230b22a163e4e924c5fac
71884 F20101106_AABXAV pippy_r_Page_34.jp2
a6a86ce86d93d8f62d1319a9e454f453
b9179306af5aef8fbb861a9baec0bc0406b8571a
8435 F20101106_AABXKQ pippy_r_Page_26thm.jpg
a7b47e5c795a32be26c429fdf8dca2da
ec77a65c3e0241343a95392ab2b4f0e98ee37a55
67818 F20101106_AABXFU pippy_r_Page_18.jp2
df5408db53fdb87f3d1fff0adf3a1aa2
7330f7130a8c1e20bd34e85a9eb7636c93af44a8
553058 F20101106_AABWYW pippy_r_Page_48.jp2
d9fcbfa80c75783a7e14e8c6ba313c7b
dccda30c4187868af24576fdc3269c99d25f9035
29044 F20101106_AABXAW pippy_r_Page_33.pro
df85b527863eefa5c1fd4d4e676ad0e1
fadd17f3e62fd314d5f76c8123feb0a9c3771873
21840 F20101106_AABXKR pippy_r_Page_29.QC.jpg
9d65fdd7dc40b438e67c814987cf2dc5
131ebf9d765489e479e381e6e4aba5080f49ace9
3522 F20101106_AABXFV pippy_r_Page_46.pro
92e2fc141f719e45a6306e3f47e152ba
77c65079097bcd364b3d0eb262597092d8d83df3
4943 F20101106_AABWYX pippy_r_Page_44thm.jpg
52503f6dec3c56c2566e8a7918ba6d4f
32a32e40628c8d5ff131e4a689cb1b57cb5ee90f
21409 F20101106_AABXAX pippy_r_Page_34.QC.jpg
657e04b4a4acc7d4f85d46f79d0db3f7
6c26b8e733873c820733055dc54eea174fd376ed
32535 F20101106_AABXKS pippy_r_Page_27.QC.jpg
cfe1e065da9d830a60e7ca0055ff9d10
345ed566e00e0f48ed3a42e948710bd30c69c743
F20101106_AABXFW pippy_r_Page_35.tif
cc7b112cc46f2ed4f28272fc18ce8b4c
95a523694ff95d52a0ef2c646a4dc232eeda2970
38629 F20101106_AABWYY pippy_r_Page_21.jpg
a156a3c5433646e4dc474a332c184317
5d3d3becd274f3d27aed249c7445ff8b8e86029d
45450 F20101106_AABXAY pippy_r_Page_31.jpg
d26c51d785f0849f0a98988288158cad
2c469527a865adda75ac3f42388f242dfc0eab44
13296 F20101106_AABXKT pippy_r_Page_53.QC.jpg
d9e4ea597a215889cc4715cb9b892ccd
75771d7110238d134f691d7b33c5d588a6481636
62597 F20101106_AABXFX pippy_r_Page_20.jpg
15fe9cd7ad738101cfa35acdf334f3c0
6b496bac429563f2ade79339f7f9328dae54b341
1949 F20101106_AABWYZ pippy_r_Page_51.pro
b53e0fd2f4d0b23aba9804ffa98d561a
21b1ccae33a597c257e694addbe02f447225b95a
7672 F20101106_AABXAZ pippy_r_Page_09thm.jpg
09f09f982769150ee1117d1e4853e4fe
968fd559d3868c24028b2c05cb11ff9a94ae27e4
7625 F20101106_AABXKU pippy_r_Page_23thm.jpg
29a6a2e0404409b9fd3708746914baf5
d7863a4a20cd16a8d586d1b135354ed7388f8fa4
F20101106_AABXFY pippy_r_Page_09.tif
26e75f78c0e1eac741a8cc6cd9ed8eda
f33211d4de98876e8e1f42ee3a6943a48e1945d8
3601 F20101106_AABXKV pippy_r_Page_06.QC.jpg
aae9ca619de5d87f2a4c2a37d5dee542
bf848cff059c47abddad8b456cb9a0a87c4c9087
4176 F20101106_AABXKW pippy_r_Page_21thm.jpg
1f361af67d7a64062b0769a54f028619
f11c91b73f178b08304f0b4cd2965ab168af7955
648010 F20101106_AABXDA pippy_r_Page_44.jp2
8c6b99babe83555d89533aad41f56f90
1d09b2269585bd79c31176237f481c6f86ad9535
3975 F20101106_AABXFZ pippy_r_Page_50thm.jpg
10541dcbf8b301bf91af75760658674d
6dac0db3cc1b2d5957217fc2d558b76959a42c37
8689 F20101106_AABXKX pippy_r_Page_11thm.jpg
7daf6750acad1a0d9ca918aee9431c20
e080c74f5ec2868ee180821e8ed69e1a0b57c72d
72780 F20101106_AABXDB pippy_r_Page_42.jp2
e53a0129a5cf533513de5ed12434f65f
13c6821fdb8102e146600a27f2d10dca46842603
15347 F20101106_AABXKY pippy_r_Page_45.QC.jpg
f3070bd8b61573c8a7716df299f0e282
df28bf1cc482043702663b097788a1d291cd66e2
35473 F20101106_AABXDC pippy_r_Page_11.QC.jpg
db2e24cba9731717381ab64b0b5dbefd
9ad6b0eb1183a201aab3fd29fa1b1eb1a9d6c296
F20101106_AABXIA pippy_r_Page_08.tif
432814ff177d5491cfd888bb94252379
f248dcd0b70bdc0956a171d1023d37785d43ea0e
87527 F20101106_AABXKZ UFE0022213_00001.xml FULL
3fb9c5f92bd73fa4fbdc68cf1078e0b3
8505c4250de48b1be77ddba47392cdc0bfc874da
1418 F20101106_AABXDD pippy_r_Page_33.txt
3de32fda532c7b47b7fb3b1841d393a9
99ef45c663abc83ff3adfb71933df53dc89d698c
F20101106_AABXIB pippy_r_Page_10.tif
ed797f95eee4f82b8105725de55d244a
349555fd8f80d60c23f210eb9791e442560f6810
8568 F20101106_AABXDE pippy_r_Page_15thm.jpg
060f7dff73f4c40368d8449e6e1b0695
87c7b004723e0fa1b20ea83cb80b4f03e2402dc1
F20101106_AABXIC pippy_r_Page_11.tif
e9b71004ef54c8df4d6587522bdf222f
a649bf4eecf650c1e170c94b7cd7c40b66dd3875
4450 F20101106_AABXDF pippy_r_Page_08thm.jpg
ca6cbe4e0f06b6b1b84b2cd9288d205d
b262b6ac877ceac644ec01f47bcca22933f5bcb2
F20101106_AABXID pippy_r_Page_12.tif
1703edf0b8a38a65c1a2762c4b9a24f0
ab478fc72f86f50fbf3b31f677ec9a77a845a1c8
29371 F20101106_AABXDG pippy_r_Page_22.jp2
96f0cee1e2a70bd7b9f9743f98973e32
de97f02b589a11c7f6e52cb3359d4dda6c1a27b2
11747 F20101106_AABWWK pippy_r_Page_35.QC.jpg
63457933962b30abbf403695a07ca4e6
36890641e784f9285ad0c64d1d03c3cb47973d43
F20101106_AABXIE pippy_r_Page_13.tif
0f6c277304123555542039b3a6c55261
bbb6d3ad8a71fe62f2028975287b3bc1f5603a2e
F20101106_AABXDH pippy_r_Page_24.tif
ef5dd61340bef163cb2c96af09008f84
c9c5ac7e98e8b900a7fa153159b94aa55c7cf3ba
F20101106_AABXIF pippy_r_Page_14.tif
cb92e59fe01dece9c413300c74c50084
f8467e08b5b4144d29ae5a71bcab73e4dfdaa064
47055 F20101106_AABXDI pippy_r_Page_56.pro
d983ae36e5e5d732e7f87f0ed992a55b
44fffc6250d1f204063262585c2cd13d6a4927b7
1793 F20101106_AABWWL pippy_r_Page_03.pro
d56428ee0c943c933dff60868db00b1a
df64b73c9c6512179b8a673fca14d88b3fd84f41
F20101106_AABXIG pippy_r_Page_16.tif
bf31d97e97ace7222993d96834c15a6c
d48f668147191ec539f83878d5eb3bf0a6bd492d
8350 F20101106_AABXDJ pippy_r_Page_40thm.jpg
db57a82a88e58d406926f5b81340f9e3
b69b873c934900c15798eacc4cf03557310d6703
122855 F20101106_AABWWM pippy_r_Page_55.jpg
25fe85315e6fee5e9ee658d88d8ec45b
14fd1f4957c3764fd154417bdd84fc2857261097
F20101106_AABXIH pippy_r_Page_18.tif
857b2ab4c024960ba444b92524fde30d
c1e451c26567bac6dd539c27b4c437c38fcf5030
1683 F20101106_AABXDK pippy_r_Page_07.txt
7412f454eb4aaa10e3465e927311e17c
f741006b24317469b5312f0e765118c0168d340a
268 F20101106_AABWWN pippy_r_Page_47.txt
c697005222374046d37eae69bd318254
fcda800714181c1ea5207e95fed9eff1a6b6e181
F20101106_AABXII pippy_r_Page_20.tif
dc885a784406b90b537d2774057358ea
83f661eeef66a766e6e72c39e62626453adf9c46
1838 F20101106_AABXDL pippy_r_Page_03.QC.jpg
de41406ae931b5b4e1b8b8c101f141e9
269928c585b1a8e15672b42c1a89b5a46ee7ace3
111 F20101106_AABWWO pippy_r_Page_03.txt
c706a8fe80e6cb67144ff4a74b6f2a44
c9d50b84447997924bf2f395fa5a0d3481da08c5
F20101106_AABXIJ pippy_r_Page_23.tif
da3200776f309d188813b65239712a9e
4ef122955517bdb16ee294b0379f7b325421e093
1964 F20101106_AABXDM pippy_r_Page_36.txt
6404ec8b69c2ee72e6e69beb5903b6c9
686012fe011519896708c5f8bd3697e821021a89
3513 F20101106_AABWWP pippy_r_Page_22thm.jpg
da32dc237fd04ef6054795fca51c0b30
1af7b440997101d83907bef84356f1097ef7d71a
F20101106_AABXIK pippy_r_Page_36.tif
1f915c884ce71d82a551df332e89fccc
78bb6f15df68ee93d096210d53a3b657b25f43ff
41490 F20101106_AABXDN pippy_r_Page_52.pro
6af8ef498f4832f6b1dd1898c34e3cea
eef534dcf9900a25deccc385060b85b258fa15a5
106218 F20101106_AABWWQ pippy_r_Page_40.jp2
ff4d3331b2909810a3ac7131533b7677
b5445a55fdb7db92a1c3bd8b66594b8e8a5e9b65
F20101106_AABXIL pippy_r_Page_41.tif
742625bd6bee9827eb285ed9cac491be
15daacfd7a209bf9eb191b04cc1016f0d85b8a36
F20101106_AABXDO pippy_r_Page_17.tif
489bc9bd2959f335c2dcfd6353104a10
e2bc22e922e758470cd0438a0c1f3f1aed22bdb4
19961 F20101106_AABWWR pippy_r_Page_54.jpg
55401cc4b7af8659161e9932f60dbded
3370cc2376ecf47768acd1db79c32f93f0552dd5
F20101106_AABXIM pippy_r_Page_42.tif
e3f2d61c8720df4105184e1632e61d30
98a5d49f7e64b9e27d8f82e44027f628274a6d91
32844 F20101106_AABXDP pippy_r_Page_41.QC.jpg
03129b0fa0d1d9e72dd03bcd04828739
ed56c2ecc709f30bd8722d42caf8f8293506921c
F20101106_AABXIN pippy_r_Page_48.tif
c84f447c0ad9248c76621ca15a90bce1
a795998e86c15d92b274ab02fae412565e993ba6
F20101106_AABXDQ pippy_r_Page_45.tif
7dc9f6bc71b3558c7b19e9c44d64e015
f105c0c23cfe3a08c52516dbbb508f7e64258d3b
41470 F20101106_AABWWS pippy_r_Page_04.jpg
2a63c6d2fb2dfb1394fe1295d6791b8c
a5761f6fadab1f8a44d3221a9812957d28864710
F20101106_AABXIO pippy_r_Page_50.tif
b6ecb2a38c31ed2ef4c02bb7c00afe23
94b2063dc82d740e8063beac7083e5f1f45680ef
3973 F20101106_AABXDR pippy_r_Page_45.pro
115b8e68d573df41aaf694766f2ec476
cf4d27e8b98e7c5a30bce805e2146aa091de0262
68771 F20101106_AABWWT pippy_r_Page_42.jpg
25b2f52f03609e631e99bfb257b2fdee
fb597cfcf5a8b277294495059ac0a43d504a0f08
F20101106_AABXIP pippy_r_Page_51.tif
00cda7097528b857df12803fa22a670a
ac9db62c24be9625986810fae97370536e276bfd
74214 F20101106_AABXDS pippy_r_Page_43.jp2
f60ca6363d6e78f04bf9442547cd3c38
c7f1a3721c0b4d68c846ca6caab66c3c3876be01
F20101106_AABWWU pippy_r_Page_29.tif
923e3fb1fa01d4b9495934d22eeb6082
341f83c59ca2ade768596e0543b09fd64403e645
F20101106_AABXIQ pippy_r_Page_54.tif
5d50fcfcfb8a8f398da54941cb3f379f
2b0977d0a6c8467f0bfd96e22f2feb795d6c7ba7
F20101106_AABXDT pippy_r_Page_31.tif
f58c7d68ba95ec3a1e2972d39b352ff7
2c9aaf2a048fc293c8def9bb00c236b7e5aefae3
F20101106_AABWWV pippy_r_Page_40.tif
39f96e30936faa78ae4ba46d1fa7599b
8f768694620d23c49f2aec9c8c287ed9b2f602f3
F20101106_AABXIR pippy_r_Page_57.tif
95d7864200889f9c122deef392655f73
623f082c71ca50a20b656b06e296a55b6456d9ac
23843 F20101106_AABXDU pippy_r_Page_53.pro
c0432d44ce8713eea09ffd16d246d8fc
90987b9c23f429fff7996231af9134e82539e3b8
34554 F20101106_AABWWW pippy_r_Page_15.QC.jpg
29542ffaec2dbbc88916ea4ebbf2441a
4a44089339ae9059c33711df53e0bc1a3dcf73b5
10158 F20101106_AABXIS pippy_r_Page_10.pro
e3f421fb706df1548c535c191b710684
95e11abe6cba13c54afda5786d17eeef9848332c
F20101106_AABXDV pippy_r_Page_25.tif
19818b4121c59bdb17862401784bd2bd
ccd830119aa1f17d8aa154c44431947c3165ce36
47121 F20101106_AABWWX pippy_r_Page_17.pro
2cba67bb8e88b9f06aaa3b3184f2c3f0
8fe273d394c94dd71302643b91beb571587b41bf
54842 F20101106_AABXIT pippy_r_Page_11.pro
181c049ba527a53d51fb59e706ef0048
c2f681c876d210babdf1e87674ab1d9bc72c2ffe
689 F20101106_AABXDW pippy_r_Page_57.txt
4a55c7f4de7d18c00057177131b535b6
1a527e0e15f96870065bd5d93b24f1a74a1b225c
100576 F20101106_AABWWY pippy_r_Page_56.jpg
51a5ac6a5ea834cfc6cb243ef0914808
d4a01f3e15f8a4ef21bec6fd2536dbf984ddf5b5
40440 F20101106_AABXIU pippy_r_Page_12.pro
a4fb7e47c1025c4af8783339e5cd9535
51e7c356114adff62fbb573be585cd53eb286095
84017 F20101106_AABWWZ pippy_r_Page_39.jp2
88a74d31c21f7eceedc2a6f14c6e8bd0
65f928a092fc7dd9f2c3998c7049843203f16fa0
50602 F20101106_AABXIV pippy_r_Page_14.pro
ab2fcc314656e7c2430371278dbd94cb
2fa564a856cfdc75434caa6111777f5346c4b571
38801 F20101106_AABXDX pippy_r_Page_13.pro
2a118fc9b0f6857b337f6f63516ebbe0
97fbc187b2a3ff964e9b0c96acd4a2947090f7b6
42927 F20101106_AABXIW pippy_r_Page_19.pro
0988311a9f84c13fc4f5ef9eb4588435
271573b23d2640dc335962133ca34d294519912f
715 F20101106_AABWZA pippy_r_Page_03thm.jpg
3f6c4f631f9c2ae865cc236ef8bd1e63
0f66d506c6ab771fe37f4ecb666851ad591aeffe
103808 F20101106_AABXBA pippy_r_Page_36.jp2
e068edf0562515231fe4bd8de55b140e
d08fe8b1f30e43049f48c265fc26a3fd9a283a48
97033 F20101106_AABXDY pippy_r_Page_24.jp2
c7b6eff5d8460498818a32d0abfae0f8
397bd2aed57b3bc7189739a3266fd0f786b830c8
33680 F20101106_AABXIX pippy_r_Page_20.pro
dd53d62eae8bdaff2a048d1d6c5fb24f
ee605090af50d2de1fc02cd2f567c3e8709bb603
75522 F20101106_AABWZB pippy_r_Page_38.jp2
6dbad03f0b9d2ff77900b27b75a624ea
dad108b4d2fee8ac40194174f35638ca30b0f643
14443 F20101106_AABXBB pippy_r_Page_30.QC.jpg
97e5797d8472e03afb89dc67b4c754fe
7de71496222494acea817a5dd6318a71d8cff034
F20101106_AABXDZ pippy_r_Page_27.tif
f03e164d5adc779299a55e160cc38e36
797fac939efb08963b2a9e1c29a3954cfe99dc28
10713 F20101106_AABXIY pippy_r_Page_22.pro
ac89feb4b117ffb0fdfb67c5112f4f76
d22e693e5093eb644afa39d0534f020dfae9b93f
1455 F20101106_AABWZC pippy_r_Page_34.txt
c255e049bc36bca1b46de44fb11d58be
f22ac66eaa3e7e25badb251ef7be5a59738dde5e
1968 F20101106_AABXBC pippy_r_Page_19.txt
b33c10351d438919ca87a6d21f96752e
ef61107162512f5742f124eabaf7a305911f3c35
46599 F20101106_AABXIZ pippy_r_Page_24.pro
742496aa11a781863312b66a80f56878
362aff4e99ed22d6ab0860679ae0b6cfc3b389af
36368 F20101106_AABWZD pippy_r_Page_21.jp2
292b84259d04fab562b44ac5361b317e
c47ba75854eadd59856ab3624b8f4f3abe918d59
60665 F20101106_AABXBD pippy_r_Page_55.pro
e9ac88037dee142773ccf4fb266d4c00
585335cea44935f56a544b8e40760a91f47ee0fb
F20101106_AABXGA pippy_r_Page_52.tif
318a9d0bc75f9da5b9861beea8dc24b6
7dafe370a16b370161fb87e07e41121d90bb6324
F20101106_AABWZE pippy_r_Page_19.tif
1764999d61e37e0b2fdc5ee7b7387ba7
a8d5f38345b6ba1bf2fb79bc4fdef7ef40a0dfff
8222 F20101106_AABXBE pippy_r_Page_37thm.jpg
cb03feb8c6bbdf935ff869c2028d25c2
bfbcf91f829934cbcef3fb9f1e5ab86291992bdb
104859 F20101106_AABXGB pippy_r_Page_09.jp2
2046cb81cc97f179516090f43b3898bc
fdd235cb0b85c4f891e8caee88dc6bdd09c4045c
F20101106_AABWZF pippy_r_Page_44.tif
1021b41c66ed844272910f4cb1f93e41
f2aa252eb118faf3f5ecc1bd39f56d4dfd18f196
90393 F20101106_AABXBF pippy_r_Page_12.jp2
d15384082c994a5f7370aef7b5d5ee51
d38d03c1325e5a3bf91b04c268abe6fbfb4d2704
34139 F20101106_AABXGC pippy_r_Page_55.QC.jpg
34bc5a88d19baa27828973019200c23f
2f5fcbd29be6f9610b59796e94f693395517ce86
3069 F20101106_AABWZG pippy_r_Page_57thm.jpg
128bdcddc0f1fd493f780f1d45ad25e9
0875004c7fcd0ff70aec8c7f4d819d0be323b37f
1898 F20101106_AABXBG pippy_r_Page_16.txt
a41f369ec8f335f96055a3fd2e3e219b
8ecdda3b8a3a22a03e6e1b68947691b67bdfca0c
13029 F20101106_AABXLA pippy_r_Page_04.QC.jpg
2cdc4b3843457305c292df6998d03328
39d6919486b33ff1ac2731dca52021cfa2cf81f4
1180 F20101106_AABXGD pippy_r_Page_02.QC.jpg
f9c2b2aa1c8c8e4a9973f3c58220a364
f646995fbbe09077cb0dd8f8f3f938b4f835a5c2
13174 F20101106_AABWZH pippy_r_Page_21.QC.jpg
005e3d9a99a42b5b72b8f588c5368e62
6455ac40c5420d01c7607b8760b6e1d8ab758846
27546 F20101106_AABXBH pippy_r_Page_05.QC.jpg
bb4aa5c056974e293620d6187454fcad
35ea2b69d3d5664c6943358c06d6a9f2109eae24
7107 F20101106_AABXLB pippy_r_Page_05thm.jpg
0e9baccc8b3bf2eee52764de4f3bbf29
f54248c73a5ac41a023e68edbc37aeab41bfe2eb
12314 F20101106_AABXGE pippy_r_Page_21.pro
d10d19af20c0aa2e5de2895d980ea067
fe457de05f1c9181706d9d2185b88f4ae2aea160
8583 F20101106_AABWZI pippy_r_Page_14thm.jpg
2c32f0b64f6d0d11c831ddef81112a56
b13320bce67b32a904d534471db3fe9aa150d9a8
101844 F20101106_AABXBI pippy_r_Page_37.jpg
874f8657b253887a27f117dcdb59b89e
0840cac3ff340097b899cc15d6ed1c941db1a96e
18160 F20101106_AABXLC pippy_r_Page_08.QC.jpg
13380d269fba51541048558714e177de
caf909908e25f63f6df4b8a0b25aaaa846f99508
110369 F20101106_AABXGF pippy_r_Page_11.jpg
d53431add8b5b4266aebc47cbab9af2a
1bbc9b493b4777ec22ed0fe45f3a957c4cc23634
3943 F20101106_AABWZJ pippy_r_Page_02.jpg
e9825c15c82b71ed522147310fb9f212
8eefd6040e4166f966ccbb5aeef30bf6eed733f3
44985 F20101106_AABXBJ pippy_r_Page_16.pro
c2cf08afaa5ba703dd2c777435066131
3d4fd517d7842d9c50a22baf3131590b30758210
19823 F20101106_AABXLD pippy_r_Page_18.QC.jpg
eebc67e4bc8ebf88d9914c97a437b806
e28a9f2ca6285dceaf8232bf1fbc8ee05f0c92b8
43793 F20101106_AABXGG pippy_r_Page_30.jpg
22165b95bafa612fd02fac6e011d5d49
e86667d621818a10e5904939f35460e52bbf185c
F20101106_AABWZK pippy_r_Page_46.tif
63df27e9f63289ddc90578329618494d
9d189df765d67d03838e1f5e5a048991ec838464
22006 F20101106_AABXBK pippy_r_Page_49.QC.jpg
685928bc946868aada1bfdccbc065d8b
0124a4a0bec05540e52e38c8f850d9109f131706
30378 F20101106_AABXLE pippy_r_Page_23.QC.jpg
3f8e7debb4ff93c5431465d75185a363
14981d36c6b6bb75c0b62ee0d9c3b58bf3d14193
1051952 F20101106_AABXGH pippy_r_Page_05.jp2
64c4c1636f37adc02bdaa78a84d84e9f
bb20aaf08621e8665afd6cd6d8bfb381cab8ee6f
1889 F20101106_AABWZL pippy_r_Page_38.txt
2a426de15178909913c59d2a0b70e61c
620e33ec6a6ea1626935a4e29e1645e2bfecc553
F20101106_AABXBL pippy_r_Page_30.tif
8e608c005f1bf6735795a8b96e26897a
e33eb118b8298c09f1d07c236d806e7150e3eff4
31428 F20101106_AABXLF pippy_r_Page_25.QC.jpg
8fefd3642a3df3235d0fb1d4e1516103
f068a5124c33c7106b35d83edfc9b6b0ec091c11
5028 F20101106_AABXGI pippy_r_Page_02.jp2
b8e3ed322e9d3ef5b463425ab1017b49
816998cc808866620961869ba00e84b18a524d8c
33661 F20101106_AABWZM pippy_r_Page_42.pro
dd4eaea96c280caedb42a387f70031ca
5d6a51690d826f3f0dec0f80644b702652591b49
34483 F20101106_AABXBM pippy_r_Page_40.QC.jpg
02c65d93f46ec33c62d33f10a448a680
cd0bf16492b5bba793b0cd99748b34b2c649d0b6
8071 F20101106_AABXLG pippy_r_Page_27thm.jpg
d76c9169a197b7a18fc81ccca4e34d68
c0215f572b50ff2ee34d2d2033c60ea441ed3fcb
68021 F20101106_AABXGJ UFE0022213_00001.mets
f78405e6ea0eef4fa76d4f4dc6ad1154
9dc00265917fbf9bc21dbdfcc64c02fe98c97a95
9010 F20101106_AABWZN pippy_r_Page_55thm.jpg
451916c1caa4b344918895f255584591
432e07fed8fb339bf3ee46567a3f7daf81f6c568
536237 F20101106_AABXBN pippy_r_Page_49.jp2
9157a609768291db3f7a65ff53d58664
41c7f70fc724b68aca11d4d572189f310d5f04d5
3803 F20101106_AABXLH pippy_r_Page_30thm.jpg
581f9c2c55523587c27fbfc7384d2f58
fec4bb01779d4e315f5e10991144ec5736291278
F20101106_AABXBO pippy_r_Page_33.tif
9c98f4ea18c73d187296d58da045d16f
2effed7e875a15a39721e0057447ece077a21b41
10896 F20101106_AABXLI pippy_r_Page_32.QC.jpg
3f0cf3ecb321411383e87346136777b6
ea7667a3f3b0a0fcaaff0afd2d9781c10252f875
78899 F20101106_AABXBP pippy_r_Page_13.jpg
b7c580999da8478632870ffa317d0b4b
bd8ce9288b0b71dedde1c32deeecd4610947a59c
2631 F20101106_AABXLJ pippy_r_Page_32thm.jpg
9dde45238abd752d110f5e081d9d993c
f36e2ce445cb5af1ee0722f37076146b5683b45b
64913 F20101106_AABXGM pippy_r_Page_07.jpg
53b29dca29763d0d158b71bc76da2318
127905f99df4ac456bc352cdff87d3279bc77de3
19731 F20101106_AABWZO pippy_r_Page_20.QC.jpg
2d98e4dc73cb539eabfff695b24d4ad3
2baf27d673de00b222c7543e98fb4661dd3b43ae
677 F20101106_AABXBQ pippy_r_Page_48.txt
7cbd4c5a66809a05dbbad5f4d7c7a692
e2f3e95ef6d183280c49643509c3eeb62f5c8254
19365 F20101106_AABXLK pippy_r_Page_33.QC.jpg
83a37675cf7de8a76da003a3bbdaaec2
107a7de9e79a7418955ac6fa40970eff079d6bf5
64052 F20101106_AABXGN pippy_r_Page_18.jpg
f088ee06a88ba85800b924c034d56de3
32e32c50789cc4cb6fa64b7d88fe8cfd42f25b96
126933 F20101106_AABWZP pippy_r_Page_55.jp2
9c9095bdc5cb3ec6f058efc9aa4ba672
26df48fbb5e083b8807f029e468a8a21fba0bd07
21053 F20101106_AABXBR pippy_r_Page_31.pro
55f84230344de6c4ef9be1cc92b046e7
0dcf1b0aea3a02275ed0d1644bada1ea3580b1ab
8469 F20101106_AABXLL pippy_r_Page_36thm.jpg
67a14f4e837bc6af21e9df697677f165
f6a61ccef62bc0c800ccd4f0e63e3ccaf4f22d99
95016 F20101106_AABXGO pippy_r_Page_23.jpg
bd068580371359075df7f9208977bb69
4813e567d72f02faeeeaa9a278202474c7150954
37778 F20101106_AABWZQ pippy_r_Page_38.pro
2d7932061c4fb621038a72bc5bc6814b
7c76f3c71d3a2eb8f83251e2ffcd15f0fd3c5cf5
4782 F20101106_AABXBS pippy_r_Page_34thm.jpg
89b4429a470818d96187d68a5ba73ed6
a0ab52008fc08cd6cb940e3dec3ec1b34c97dacb
F20101106_AABXLM pippy_r_Page_37.QC.jpg
679920a15fc3e44821f16e8ffd2624de
5fb88cd5c7d9d1f6b6d5f1c293d55b550e92f548
92008 F20101106_AABXGP pippy_r_Page_24.jpg
84db115c7ea8b0390d89b0bcb792040a
10a9b30ce2d9b7aad8f2af63305949b50c816980
20486 F20101106_AABWZR pippy_r_Page_48.QC.jpg
190e06ec275e0cfa37bca7fa943a1797
8d37a2e63b9d2eacc8b2d267ca86799fab4de891
7190 F20101106_AABXBT pippy_r_Page_03.jp2
d3121255f3fdc9272c4d9a600d9ada3b
909dbc63897aa0249ea0cac00719f8de7a9dbf47
23179 F20101106_AABXLN pippy_r_Page_38.QC.jpg
a72e03efa022e37c2bfa06c716742c38
edef365f1e726b1ca5fa928b17a9c33d50ce310a
103385 F20101106_AABXGQ pippy_r_Page_26.jpg
88632085e031a7bdc14ccfdc614e4e67
572977fa6622d3f553b6b981c4e79307fd74214b
88096 F20101106_AABWZS pippy_r_Page_19.jp2
551fd6d536fed3650f0cf5b3177de1da
c41c6240f13196bed43ac35b07111845b96d3fbb
15423 F20101106_AABXBU pippy_r_Page_31.QC.jpg
1f58f38d9517a3844647e5524b05c188
0430620dea7e03e9725640b5b476e940e61b8a87
8487 F20101106_AABXLO pippy_r_Page_41thm.jpg
e79795e7847b979edb54caed9a5ec74a
c4930edab406a584bd5f079dac348267412b8975
100352 F20101106_AABXGR pippy_r_Page_27.jpg
f1c1722264a33942dc35ebc55d24a5be
afac7905e2494ffe1da43d0d43a1066614963fc4
84116 F20101106_AABWZT pippy_r_Page_52.jpg
53d7cddea7488c66d3f5cfa238d64771
463bb62b4b742a1d2be9a94e6aa6176fa18ab336
3020 F20101106_AABXLP pippy_r_Page_46thm.jpg
1de06f8a44f390fbe36e0bf7bde8086c
7e483544cd77667e514f25a75e4a827648a5ee7d
66890 F20101106_AABXGS pippy_r_Page_29.jpg
0a3b6b82d14b58fcea7cd512b7e3b5b0
ca23fad49a18733f9cb91ffa2cc36e0361c08ab0
50947 F20101106_AABWZU pippy_r_Page_45.jpg
49541f3505746523c8e9ac7113da0dde
8a0ed6fcedab5b76d34cbdc869000b66ddf1d2d6
3613 F20101106_AABXLQ pippy_r_Page_47thm.jpg
86fcabeb5efc4afe5515b938863230a8
6a2b036486173733ffc780ba833fd7a099855d14
69199 F20101106_AABXGT pippy_r_Page_38.jpg
75656b7edeedeccbf365bb94e267c48e
f7b17ca744c1ba0c842298eef3e95212e74c3478
24113 F20101106_AABWZV pippy_r_Page_46.jpg
2337d03994c77800be5be94b95c149a1
b72c84a9e24d53d12f28f71242c0cb09a1f8a8b0
102079 F20101106_AABXBV pippy_r_Page_14.jpg
fba0d1e1bea8067ad6be7b6fc219c91e
3f6b37e3c544851f240ba28e94825d275e72dbe0
5455 F20101106_AABXLR pippy_r_Page_48thm.jpg
4c60f75791667e46fd7efad59eae35cb
cf880932e5785075f1af76053fce7371d65083ff
80582 F20101106_AABXGU pippy_r_Page_39.jpg
d12a96cae958b6685ea09dda37f6b152
2430e1d2cd91cafca95e80d6520e029843fc4c6d
8390 F20101106_AABWZW pippy_r_Page_28thm.jpg
e2fc300da8182caa9179b89d89db44d9
138ba3d211f2295efcff242086d15d161608a0e9
40135 F20101106_AABXBW pippy_r_Page_08.pro
a90766ff9ee1bdbbfc56a5cd81fe136b
62881b75d0e988b32e651ca758153030f68bf0e7
23372 F20101106_AABXLS pippy_r_Page_52.QC.jpg
8661791f3a9d83d6853c56957a7cc1a3
4729a283edc123ff6cbbddddfa152afc20388a51
103328 F20101106_AABXGV pippy_r_Page_41.jpg
8ea431792477bbdc438170b6577f0b72
5552d0e31dc33af28f4fd3f9dc51f20bfe3356d7
12301 F20101106_AABWZX pippy_r_Page_57.QC.jpg
0c00a6ab49fb1f06a98a20afc497ef07
38e64f967bd3800ae166137018f49b3f04537af9
52250 F20101106_AABXBX pippy_r_Page_41.pro
8d8322b71b263f6d902c21c3b691aa9e
2aad611e2ff7ee21cdfef463c4c012ebde4a58b5
5752 F20101106_AABXLT pippy_r_Page_52thm.jpg
6d2a227c75fc93bc8c4097c34cf80072
c9a64eaba8566e996b13d1afde77a8e4c51a7041
82375 F20101106_AABXGW pippy_r_Page_43.jpg
343d8cd134bdfcff3783f077870dee69
ed49f4ada43a6af6ae2161cf172d2a7265d892d4
2069 F20101106_AABWXA pippy_r_Page_28.txt
2dad1e9e6fb251bae5cbbad4942b5d5a
1ab47b6f6713c24aa1862e79ae673a107d00d60a
2141 F20101106_AABWZY pippy_r_Page_10thm.jpg
ab860703bbb2ef34a7fe4807f08b83d7
8be79fa4651e927bfe737265d6c2d9df762a7539
64151 F20101106_AABXBY pippy_r_Page_33.jpg
3caa9b94fd45a6caf9ac8ec1dccaae79
35d8c3748ee387d1477a88b86fc74ddf3e34821b
4256 F20101106_AABXLU pippy_r_Page_53thm.jpg
c22531f6e47b2687f15bd7625d398637
6317c8450346bf965e0c76d4618344764fbc1d0f
56988 F20101106_AABXGX pippy_r_Page_44.jpg
32fcd7fb4a2da01ded3c6d66ff57db1b
e9edcfcc013b01462cf074f01b3bef2feb8e1882
32997 F20101106_AABWXB pippy_r_Page_28.QC.jpg
ad0b817a9625784d3996a23be7409df7
a0633d4c49a9fc2f949875203f27b3aa43b70d64
1126 F20101106_AABWZZ pippy_r_Page_06thm.jpg
fcbf8135c6c33188303fd3e343115885
5247e29a5a3d14e917119f8942cee110d780cb90
8799 F20101106_AABXBZ pippy_r_Page_01.pro
840368baf08543e0b5e4cbe46801806a
9532af14c144510e7cc33e5e95d0e5a771dcb171
7262 F20101106_AABXLV pippy_r_Page_54.QC.jpg
9f9cb648c71a6b0a1885d95e0f51685d
91dca8e5b2daef2e89dc768c5a42b6e17533de85
61107 F20101106_AABXGY pippy_r_Page_48.jpg
66754cbcf67f6b89477a6f75c204fbcc
686516a6e84ba7da95ed3c86f097ff19a4523ace
115375 F20101106_AABWXC pippy_r_Page_05.jpg
fdb2b141a770ddff39b0d497f2ee02a9
08f2af6307c5a56f36f24449ef45c231f13d8862
28752 F20101106_AABXLW pippy_r_Page_56.QC.jpg
a435c4df6797f61bbd79296f97b91a89
05cb71764f7b9318ac7eaa0d2c23b59eb533d660
26196 F20101106_AABXEA pippy_r_Page_19.QC.jpg
1ecdc1bc8f5ce281ec6fec508b02fcdf
62e9cbc8c703700715a6b263e8d337942404fb9c
64553 F20101106_AABXGZ pippy_r_Page_49.jpg
0260aebb40b6e734d032cc3cd7ea4196
ba1dba220b866855cf942b67850a36f42347e60d
30219 F20101106_AABWXD pippy_r_Page_24.QC.jpg
3902fe91ca6f817104016feb71ad23af
8986dbabfaef18c3ad6678fdf24a6d0e5be05634
7179 F20101106_AABXLX pippy_r_Page_56thm.jpg
99555ccd4e9c02b92f88b58b8f01f91b
9247efd75cb5b64c34de618c1dce2246805b5b1f
16406 F20101106_AABXEB pippy_r_Page_57.pro
4e217206c36c50cf68657ab3e64227a6
49e54f5e08500f589291b025338dbcb938bf3656
88495 F20101106_AABWXE pippy_r_Page_16.jpg
412d7018707fb75a01dc5e12d9792860
b39f510d7e6e44a21e5c322c4b6d7eb0dfdcfa33
93319 F20101106_AABXEC pippy_r_Page_25.jpg
6010c32d474aa046f1c8302694c7cb28
1e291ebd9c46ccbb1c0ca54c6af7da39dc913d37
F20101106_AABWXF pippy_r_Page_06.tif
bb67daeb9776c64f2d730bd3bf886d90
03806635249b83652df89808c1faa0b82bd4e830
12496 F20101106_AABXED pippy_r_Page_06.jpg
d5a2225138802625d057ff96a5248f74
21cca3b439b15ac36ec7b8f755558f22f90008fb
F20101106_AABWXG pippy_r_Page_22.tif
8ccd8f0b513f660e5b22f1d551cc539f
b6497ae8d9f52a2b843d3d52dbfeee855389d8d9
49611 F20101106_AABXJA pippy_r_Page_25.pro
748d2f4c5747c1a460f5e6e45a159877
2eac2a481b35c908523b68c9280bf4eba57f62f2
102197 F20101106_AABXEE pippy_r_Page_15.jpg
4a8577f0356c00e53b46389624e45d3f
ad4ea9e9be28efcb6f23175ab94fb0dbe5dbbd25
32404 F20101106_AABWXH pippy_r_Page_17.QC.jpg
0b63c339df6d6dd575d96b2825eafb5b
33147783667f7b5e116868bfc09813e94f638836
51796 F20101106_AABXJB pippy_r_Page_26.pro
38cc5f647ce8998a9e281fe139a1481a
3c9c1a0fbcf00fbfbd0efc62881a9426670d3bf4
35022 F20101106_AABXEF pippy_r_Page_26.QC.jpg
ca1bf224344500df329c3beab340bd47
0034844ebd9ae869dc87f415eb168eed02fdfe87
F20101106_AABWXI pippy_r_Page_26.tif
7d2aa2911710d18f6887f239b42100c2
af7d67b276bb64c32e882042463ee95afcf7ab84
50064 F20101106_AABXJC pippy_r_Page_28.pro
64ed1ac231a62a9acf1368ee3de781cb
47e32c6d891aaaeba0f75e9bd1091e3fcad80cf5
7532 F20101106_AABXEG pippy_r_Page_16thm.jpg
637094879de525192babf0b600003c38
4920f9299e641bb0a0ad63d928c41df9e047adc5
135 F20101106_AABWXJ pippy_r_Page_51.txt
d6f1a4284456bca413997d02d2ee99f1
260f6208b6d0b37b01ca03776705b3737d345eb8
31822 F20101106_AABXJD pippy_r_Page_29.pro
933f0ef58e74c1037966aa35ab39921a
8ba1f35d4a11ed3202abed8a3aa8418b25675fa2
506 F20101106_AABXEH pippy_r_Page_01.txt
89c81b21efe14f17fbfe6dfebfcbdd9e
782f8f78cd12eba1c87f83a48e3173801486b311
F20101106_AABWXK pippy_r_Page_34.tif
5349ce1ab3b8c21d3ad5066c6ac38895
e3221ea35070177d38eb0d866d59509f14492d86
15549 F20101106_AABXJE pippy_r_Page_32.pro
826f5820a5b02a71ac496af3b10364c5
6ee5620c33a8d70cda6d2141b3cf7823622a70f4
3740 F20101106_AABXEI pippy_r_Page_44.pro
706c004877aef0ca0918c8d5a8713b9c
26516e7e53811f8a721bf4653598d687d769ef9b
25643 F20101106_AABWXL pippy_r_Page_13.QC.jpg
151de41f2343671cda82ba0f0c41719a
89ce74e4031372786bc2cb19e3e39701f3e63b91
32996 F20101106_AABXJF pippy_r_Page_34.pro
b8e234b15c1370baa1a65b8f80287196
0feef45b5b067ba430967bff58fcc4b2d21112b7
46087 F20101106_AABXEJ pippy_r_Page_23.pro
9175c0769b0a18492f6a780d6fe0b903
398d50b06d502ef983032811879edacd1c17c431
1689 F20101106_AABXJG pippy_r_Page_35.pro
657e74de090a01f8fbef69a728e002c4
0d18597bfb1e266bcf975e32fa20b4f56dcc8fe1
21657 F20101106_AABXEK pippy_r_Page_30.pro
4a786390f2d4c8e8059e3be577f5e751
c38db9f5fe8c926a1f42ce71d7b3023d9930258c
100672 F20101106_AABWXM pippy_r_Page_17.jp2
db51caa13545103327d7169d75aedca4
f9a4281d142034a2dffdfe818a039427fe3b3be3
46954 F20101106_AABXJH pippy_r_Page_36.pro
9ffdd71635a3b257f7dd51aa53bbd25e
3b2547d1d0cefd8eee3a639d0193a7ac1cd5a615
1995 F20101106_AABXEL pippy_r_Page_23.txt
6fce91af669dea99a1b1cbb94cf3e01b
86d5455ee7f5692d4b70b17f2948dd88ac44d666
18544 F20101106_AABWXN pippy_r_Page_07.QC.jpg
fc80e4688385088dcf8c0c6d9316ff14
8dd895dd39378786bbbff7257bb6559c24d2c900
50898 F20101106_AABXJI pippy_r_Page_37.pro
31b65edff5ef7dfcd770875e4ba72bbe
bcda0126b4a0740b081657b0bc05475e4b7c8db0
9110 F20101106_AABXEM pippy_r_Page_01.QC.jpg
371dbf4aa197766e83d43ba42377755a
f46d67a1ca7ae6ac5b2b449d1353ba316a5ef64c
16552 F20101106_AABWXO pippy_r_Page_50.pro
715ae9caf38b7dbbfb5866a2c497f699
04bc0bb45afa5daacdbf04cfd5961cd4c01178fd
50329 F20101106_AABXJJ pippy_r_Page_40.pro
0b7fa1b0aebd8ccf90176635e7730692
43b28d3395c54c8e5cd0ddfe17db210af9599fb8







APPROXIMATE PROBABILISTIC OPTIMIZATION OF A WINGBOX MODEL USING
EXACT-CAPACITY-APPROXIMATE-RESPONSE-DISTRIBUTION (ECARD)



















By

RICHARD J. PIPPY


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE

UNIVERSITY OF FLORIDA

2008

































2008 Richard J. Pippy



























To my parents, Richard and Evelyn, and to my brother Chris









ACKNOWLEDGMENTS

I would like to express my sincere gratitude to my advisor and the chair of my thesis

committee, Dr. Nam-Ho Kim, and to the co-chair, Dr. Rafael T. Haftka, for their guidance,

enthusiasm and continual support throughout my study. I am also grateful to committee member

Dr. Peter G. Ifju for his advice and patience in reviewing this thesis. Special thanks go to Mulu

Haile for his help in deriving the wingbox load conditions. I would like to thank my colleagues

in the Structural and Multidisciplinary Optimization Lab at the University of Florida, in

particular, Sunil Kumar, Dr. Erdem Acar, Palani Ramu, Haoyu Wang and Saad M. Mukras for

their help and encouragement.









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

LIST OF FIGURES ......................... ........ .. .... .... ............... 8

A B S T R A C T ............ ................... ............ ................................ ................ .. 9

CHAPTER

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

2 EXACT-CAPACITY-RESPONSE-DISTRIBUTION (ECARD) THEORY ......................... 13

In tro d u ctio n ................... ...................1...................3..........
Characteristic Response ................................................... ............... ...... ..... 13
C o rre ctio n F a cto r ....................................................................................................16
U sin g F O R M ........................................................................................................1 8
U sin g M C S ......................................................................................................1 8
Approximate Probabilistic Optimization............................................... 19

3 ANALYTICAL EXAMPLE: APPLICATION OF ECARD TO A TEN-BAR TRUSS........23

In tro d u ctio n ................... ...................2...................3..........
P ro b lem D e scrip tio n ............................................................................................................... 2 3
Deterministic Optimization ............................. ............... 24
Probability of Failure Calculation Using MCS .................. ...............24
Probabilistic Optimization ..... ............................................................ ..... .... .........27
Approximate Probabilistic Optimization Using ECARD ....................................................28

4 PRACTICAL EXAMPLE: APPLICATION OF ECARD TO A WINGBOX MODEL .......36

In tro d u c tio n ....................................................................................................................... 3 6
P ro b lem D e scrip tio n ............................................................................................................... 3 6
G e o m e try ................................................................................................................... 3 6
Loading Calculations ................................. .........................................37
Deterministic Optimization .......................................................39
Probability of Failure Calculation U sing M CS ............................... ............ ..................... 40
Approximate Probabilistic Optimization Using ECARD ....................................................41

5 SUMMARY AND CONCLUSIONS .................................................52

APPENDIX CALCULATION OF MEMBER FORCES OF THE TEN BAR TRUSS .......53









L IST O F R E FE R EN C E S ....................................................... ................................... 55

B IO G R A PH IC A L SK E T C H .............................................................................. .....................57




















































6









LIST OF TABLES


Table page

3-1 Param eters for the ten-bar truss problem ................................................ ........ ....... 30

3-2 Results of deterministic optimization of the ten-bar truss problem................................30

3-3 Probabilistic distribution types, parameters of errors and variabilities in the ten-bar
tru ss p ro b lem ....................................................................... .. 3 1

3-4 Probabilities of failure of the deterministic optimum areas .......................................31

3-5 Results of the probabilistic optimization of the ten-bar truss ................. ............ .......32

3-6 Results of the ECARD optimization....................................................... ...................33

3-7 Results of the ECARD optimization....................................................... ..................34

4-1 Material Properties of 7150-T77 Aluminum ...............................................................43

4-2 Results of deterministic optimization of the wingbox model ................. ............ .......43

4-3 V ariability for W ing M odel ....................................................................... ..................43

4-4 Probabilities of failure of the deterministic optimum design .........................................43

4-5 E C A R D optim ization results .................................................................. .....................43









LIST OF FIGURES

Figure page

2-2 Calculation of the probability of failure at new design.................. .................................21

2-3 Calculation of the probability of failure at new design.................. .................................22

3-1 Geometry and loadings of the ten-bar truss ....................................................................35

4-1 B oeing 767 w ing dim ensions............................................................................. ............44

4-2 B oeing 767 internal schem atic................................................ ............................... 45

4-3 A N SY S m odel of the w ingbox .................................................................................46

4-5 Sw eep of the quarter-chord ............................................................................... ........ 47

4-6 Relationship of local lift distribution and taper ratio .....................................................48

4-7 Elliptical lift distribution from the root to the tip of the wingbox model .........................48

4-8 Equilibrium of forces on the wingbox model ........................................ .....................49

4-9 Pressure distribution from root to wingtip of the model..............................................49

4-10 Force distribution from root to wingtip of the model .......... ............ ............... 50









Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science

APPROXIMATE PROBABILISTIC OPTIMIZATION OF A WINGBOX MODEL USING
EXACT-CAPACITY-RESPONSE-DISTRIBUTION (ECARD)

By

Richard J. Pippy

May 2008

Chair: Nam-Ho Kim
Co-chair: Rafael T. Haftka
Major: Mechanical Engineering

There are two major obstacles that affect probabilistic (or reliability-based) structural

optimization. First, uncertainties associated with errors in structural and aerodynamic modeling

and quality of construction are not well characterized as statistical distributions and it has been

shown that insufficient information may lead to large errors in probability calculations. Second,

probabilistic optimization is computationally expensive due to multiple analyses, typically finite

element analyses, for calculating probability as the structure is being redesigned. In order to

overcome these obstacles, we propose an approximate probabilistic optimization method where

the probabilistic calculation is confined to failure stress. This takes advantage of the fact that

statistical characterization of failure stresses is required by Federal Aviation Administration

regulations. The need for expensive stress distribution calculations are eliminated by condensing

the stress distribution into a representative deterministic value, transforming a probabilistic

optimization problem into a semi-deterministic optimization problem. By starting the

approximate probabilistic optimization from the deterministic optimum design, a small number

of iterations is expected and reliability analysis is required only once per iteration. This proposed

method provides approximate sensitivity of failure probability with respect to the design









variables, which is essential in risk allocation. This method is demonstrated in two examples.

The first example uses a ten-bar truss which demonstrates the risk allocation between the truss

elements. The second example uses a wingbox model based on a Boeing 767-400 aircraft which

demonstrates the risk allocation between two different failure modes of stress and displacement.









CHAPTER 1
INTRODUCTION

There are two major barriers in front of probabilistic (or reliability-based) structural

optimization. First, uncertainties associated with material properties, operating conditions,

mathematical models, and manufacturing variability are not well characterized as statistical

distributions, and insufficient information may lead to large errors in probability calculations

(e.g., Ben-Haim and Elishakoff [1], Neal, et al. [2]). Due to this fact, many engineers are

reluctant to pursue probabilistic design. The second barrier to the application of probabilistic

structural optimization is computational expense. Probabilistic structural optimization is

expensive because repeated stress calculations (typically FEA) are required for updating

probability calculation as the structure is being re-designed. Targeting these two main barriers,

we propose an approximate probabilistic optimization method that dispenses with expensive

probabilistic stress calculations. In the proposed method, the probabilistic calculation is confined

only to failure stress, which is often well characterized.

Traditionally, reliability-based design optimization (RBDO) is performed based on a

double-loop optimization scheme, where the outer loop is used for design optimization while the

inner loop performs a sub-optimization for reliability analysis, using methods such as First-Order

Reliability Method (FORM). Since this traditional approach is computationally expensive, even

prohibitive for problems that require complex finite element analysis (FEA), alternative methods

have been proposed by many researchers (e.g., Lee and Kwak [3], Kiureghian et al. [4], Tu et al.

[5], Lee et al. [6], Qu and Haftka [7], Du and Chen [8] and Ba-abbad et al. [9]). These methods

replace the probabilistic optimization with sequential deterministic optimization (often using

inverse reliability measures) to reduce the computational expense. However, these approaches do

not necessarily converge to the optimum design, and they do not easily lend themselves to









allocating risk between failure modes in a structure where many components can fail [10]. We

note, however, that most of the computational expense is associated with repeated stress

calculation.

So we propose an approximate probabilistic design approach that reduces the number of

expensive stress calculations. That is, we approximate the probabilistic optimization that

separates the uncertainties which can be evaluated inexpensively and those whose effects are

expensive to evaluate. We boil down the stress distribution to a single characteristic stress by

utilizing the inverse cumulative distribution of the failure stress, and we propose an inexpensive

approximation of that characteristic stress. This proposed method will also improve upon a

deterministic design by reallocating the safety margins between different components or failure

modes. We call the proposed approximate probabilistic optimization approach Exact-Capacity-

Approximate-Response-Distribution or ECARD. The purpose of this thesis is to further advance

the version of ECARD which was originally developed by Dr. Erdem Acar and Dr. Rafael T.

Haftka [11]. There is now an improved version of ECARD which was developed by Sunil

Kumar, Dr. Rafael T. Haftka, and Dr. Nam-Ho Kim.

The remainder of the thesis is organized as follows. Chapter 2 discusses the theory behind

the ECARD method. The application of the method to a ten-bar truss problem is presented in

Chapter 3 and a wingbox design in Chapter 4. Finally, the concluding remarks are listed Chapter

5.









CHAPTER 2
EXACT-CAPACITY-RESPONSE-DISTRIBUTION (ECARD) THEORY

Introduction

In this chapter, the approximate probabilistic optimization method, named ECARD, will be

discussed.

Characteristic Response

In probabilistic optimization, the system constraint is often given in terms of failure

probability of a performance function. We consider a specific form of performance function,

given as

g(x;u) = c(x;u) r(x;u) (2-1)

where c(x; u) and r(x; u) are capacity and response, respectively. Both the capacity and

response are random because they are functions of input random variables x and depend on

deterministic design variables The system is considered to be failed when the response

exceeds the capacity; i.e., g(x; u) < 0. We assume that the probabilistic distribution of c(x; u) is

well known, while that of r(x; u) requires a large number of analyses. For example, when c(x; u)

is failure stress and r(x; u) is the maximum stress of an aircraft structure, the probabilistic

distribution of c(x; u) is already characterized by Federal Aviation Administration requirement.

However, the probabilistic distribution of r(x; u) requires repeated computational simulations,

such as finite element analysis.

Since the performance function depends on two random variables, c(x; u) and r(x; u), the

safety of the system can be estimated using a probability of failure, defined as


Pf = Pr[g(x;u) <0] = _Fc( )fR()d (2-2)
00









In the above equation, Fc() is the cumulative distribution function (CDF) of capacity, and

fR ({) is the probability density function (PDF) of response. The above integral can be evaluated

using analytical integration, Monte Carlo simulation (MCS), or first-/second-order reliability

method (FORM/SORM), among others. Smarslok et al. [18] presented a separable MCS, which

is much more accurate than the traditional MCS when the performance can be separable as in Eq.

2-1.

It is clear from Eq. 2-2 that accurate estimation of probability of failure requires accurate

assessment of the probability distributions of both the response and capacity. When the capacity

is the failure stress, the FAA requires aircraft builders to perform characterization tests in order

to construct a statistical model, and then to select the allowable failure stress (A-basis or B-basis

value) based on this model. Hence, the CDF of capacity is often reasonably well characterized.

On the other hand, the PDF of the response is poorly known, because it depends on the accuracy

of various factors, such as material properties, operating conditions, mathematical models, and

manufacturing variability. The key idea of this research is to express Pf in Eq. 2-2 using a

characteristic value of the response, and approximate the change of Pfin terms of the change of

this characteristic value.

The calculation of Pf in Eq. 2-2 can be simplified by using the information ofFc('). From

the Intermediate Value Theorem [19], there exist r* such that Eq. 2-2 can be re-written as

Pf = Fc (r*) fR ()d = Fc (r) (2-3)

In the above equation, the second equality is obtained from the fact that the integral offR() is

one. Equation 2-3 states that the effect of (the poorly characterized) probability distribution of

the response can be boiled down to a single characteristic response, r*. When the probability of









failure is given, the characteristic response can be calculated using the inverse transformation of

Fc(r*), as

r* = Fc(Pf) (2-4)

When design variables are changed during optimization, it is possible that the distributions

of both capacity and response may be changed. For the simplicity of presentation, we consider

the case that the distribution of the capacity remains unchanged. We assume that the design

change only affects the mean value of the response; i.e., the standard deviation remains constant.

This assumption is reasonable if the design perturbation is small. In such a case, redesign

changes the mean value of response from = r(px;u) to F(1 + A), where r(ux;u) is the value

of response evaluated at the mean value of the input random variables. The variable A represents

the relative change in response according to design change. At this point, A is unknown. Figure

2-1 illustrates the change in response distribution, along with the distribution of capacity.

In this research, we start the probabilistic design from a known deterministic optimum.

This is an important aspect of the approximate probabilistic optimization. Since the deterministic

optimization uses safety margins to consider the effect of uncertainties, the deterministic

optimum design is close to the probabilistic optimum design. This will satisfy the above

assumption of small design change. The goal of proposed probabilistic design is then to improve

upon the deterministic design by reallocating the safety margins between different components

or failure modes.

First, the change in design variables will change the mean of response from r to r(1 + A),

while maintaining the same standard deviation. The change in the response distribution will then

change the probability of failure according to Eq. 2-2. From Eq. 2-4, the characteristic response

will also be changed from r* to r*(l+A*), where A* is the relative change in characteristic









response. Unfortunately, this process requires calculation of the probability of failure at the new

design. The novel idea of the proposed approach is to reverse this process by approximating the

relation between A and A* so that the characteristic response at the new design can be calculated

without performing reliability analysis. For the moment, let us assume that A* can be calculated

from given A. Then, the probability of failure at the new design can be calculated from

pew = Fc(r*(1l+ A*)) (2-5)

The above probability of failure will be exact if A* is the correct relative change in

characteristic response. When A* is an approximated one, the probability of failure in the above

equation is approximate, and we will denote it P pPP"r. The procedure illustrated in Figure 2-2

does not require expensive reliability analysis. It is enough to analytically evaluate the value of

the CDF at the perturbed characteristic response.

Correction Factor

The key idea of the proposed approximate probability distribution is that the new

characteristic response can be approximated without recourse to the expensive reliability

analysis. The simplest approximation, used in this research, is that the relative change in the

characteristic response, A* is proportional to the relative change in response, A as

A* = kA (2-6)


where k is a proportional constant that depends on how the redesign affects the stress

distribution. In fact, it is the sensitivity of the characteristic response change with respect to the

response change. We call it a correction factor. The above assumption in linearity is reasonable

when A is relatively small.









Probabilistic optimization can be viewed as risk allocation between different failure modes

or different structural members. This allocation requires the sensitivity of failure probability with

respect to design variables. In the proposed approximate probabilistic optimization, this

sensitivity information is presented in the correction factor.

We will demonstrate that a linear relationship between A and A* works well given the

assumption of translating the stress distribution, especially when the design change is relatively

small. We consider a lognormally distributed capacity with mean value of [tc = 100 and

coefficient of variation of 8%, and normally distributed response with coefficient of variation of

20%. From Eq. 2-2, the mean value of the response is chosen to be [LR = 42.49 so that the

probability of failure becomes Pf = 10-7. For a given small value of A, a new probability of

failure Pfew is calculated from Eq. 2-2 with the mean of the response being IR(1+A). The

relative change in characteristic response A* is then obtained from Eq. 2-4 with pnew. Figure 2-3

shows the relation between A and A*. We can see that the linearity assumption is quite accurate

over the range of -10% < A < 10%. The slope will be the correction factor k. Figure 2-4 shows

the effect of the A approximation on the probability of failure. In practice, the correction factor

can be calculated using a finite difference method, which requires at least two reliability

calculations. We will describe the procedure using a forward finite difference method, but other

method can also be used in a similar way. Let Ao = 0.0 corresponds to the current design, and Ap

= 0.05 to the perturbed design. The correction factor can be calculated either using FORM or

MSC. We will explain both cases.









Using FORM

First, the probability of failure, Pf, at the current design is calculated from First-Order

Reliability Method (FORM) with the performance function in Eq. 2-1. If the response is

perturbed by Ap, Eq. 2-1 becomes

g,(x;u) = c(x;u) r(x;u)(1 + Ap) (2-7)

Using the above equation, reliability analysis is performed to calculated the perturbed

probability of failure, Pf It is noted that it is unnecessary to change design variable, u, because

we directly perturb the output response. Thus, the computational cost of reliability analysis using

Eq. 2-7 can be reduced significantly. Next, the characteristic responses are calculated from Eq. 2-

4, as

r* = Fc'(Pf)
,= (2-8)
r = Fc'(P) = r(1+ A*)

By comparing two terms in the above equation, the relative change in the characteristic

response can be obtained as


A* = 1 (2-9)
r

Then, the correction factor can be obtained from


k=- (2-10)


Using MCS

When MCS is employed, we generate N samples of response ri = r(xi;u),i = 1,...,N, at the

current design. In view of Eq. 2-2, the probabilities of failure at the current and perturbed design

can be calculated from









N
Pf FC (ri) (2-11)
i1

N
P = Fc (r( + A)) (2-12)
i=1

The remaining procedure is identical to that of FORM. Even if Eq. 2-11 and Eq. 2-12 are

two different MCS, they can be combined into one because the same sample, r,, will be used.



Approximate Probabilistic Optimization

The proposed approximate probabilistic optimization is composed of two stages: (1)

correction factor k and initial probability of failure Pf are calculated from reliability analysis,

and (2) a deterministic optimization problem is solved using the approximate probability of

failure from Eq. 2-5. The first stage is computationally expensive, while the second stage is

nothing but a semi-deterministic optimization. We will explain the approximate probabilistic

optimization procedure, as follows.

1. Perform deterministic optimization with safety margin. The probabilistic design starts
from the deterministic optimum design; i.e., initial design uo = Udet and cost
function Wo = Wet. Calculate the initial probability of failure, Pf = Pdet atuo0.

2. At the current design, uo, calculate deterministic value of response, ro = r(jx;uo),
using the mean value of input random variables.

3. Calculate the characteristic response, ro using the inverse CDF of the P P, and the mean
and c.o.v of the response.

4. Calculate the correction factor, k, using the procedure in the previous section.

5. Obtain optimum design uopt and optimum objective function Wopt by solving the
following optimization problem:

min W(,x,u)
U (2-13)
s.t. Papprox < pet









Where

r =r(lx;u) (2-14)


A =r 1 (2-15)
ro

A* =kA (2-16)

r* = Fc'(Pf) (2-17)

Pfapprox = Fc, (1 + A* ) (2-18)

6. Calculate the actual probability of failure, Pf, at Uopt.

7. Check convergence: Papprox PJf < e. If it converged, stop the process. Otherwise, set
uo = Uopt and go to Step 2 and continue.


The above semi-deterministic optimization process uses exact distribution of the capacity

(Fc) and an approximate distribution of response (MCS or FORM). Due to this aspect, we call it

Exact-Capacity-Approximate-Response Distribution (ECARD) method. The accuracy of

ECARD to locate the true optimum depends on the magnitudes of errors involved in the

approximations. As shown in Figure 2-3 and Figure 2-4, the approximation is accurate if changes

of the response due to redesign are small. In addition, the accuracy in estimating the correction

factor k affects the convergence rate of the proposed method. The result may be somewhat sub-

optimal because of the convergence condition and the approximate nature of the sensitivity of

probability of failure. This issue will be discussed in detail in the following section.











Changes in stress PDF


0.35 -

0.3 -

0.25 -


S015


a a(1+A)









0' C '
*


0 a


0.1


0.05

0


Initial Stress PDF
-New Stress PDF
PDF of Failure Stress
- Characterstic Stress
---- New Characterstic Stress


0 2 4 6 8 10 12 14 16 18
Stress


Figure 2-1. Distributions of response before and after redesign.








Fc (r)




p new -

Pf





r r*(+ A*)

Figure 2-2. Calculation of the probability of failure at new design.


























A*


'Ii'


-0 06


-0 08 -------


-0 1 -0 08 -0 06 -0 04 -0 02 0 002 004 006 008 01

A


Figure 2-3. Calculation of the probability of failure at new design


106



10-






f 10-7


-0 1 -0 08 -0 06 -004 -002 0 002 004 006 008

A


Figure 2-4. Calculation of the probability of failure at new design


y~-









approx
i


I I I I I









CHAPTER 3
ANALYTICAL EXAMPLE: APPLICATION OF ECARD TO A TEN-BAR TRUSS

Introduction

The first demonstration example is a ten-bar truss problem as shown in Figure 3-1. This

example will demonstrate risk allocation between the different truss members. First, a brief

description of the problem will be given. Then, a deterministic optimization of the problem will

be presented followed by the probability of failure calculation using Monte Carlo simulations.

Finally, the probabilistic and ECARD optimizations will be performed, and the efficiency and

accuracy of the ECARD method will be discussed.

Problem Description

The problem description for the ten-bar truss example was taken from Haftka and Gurdal

[20] (page 237). The truss structure is under two loads, P; and P2. The design objective is to

minimize the total weight of the truss, W, by varying the cross-sectional areas, A1, of the

members while satisfying minimum gage constraints and allowable stress constraints. Input data

for the truss is listed in Table 3-1. Member 9 was assigned a higher failure stress value in order

to make the fully stressed design non-optimal. Aircraft design often uses a knockdown factor,

Kdc, in order to conservatively estimate failure stress using A-basis or B-Basis methods. The A-

basis (or B-basis) failure stress is the value of a failure stress exceeded by 99% (or 90% for B-

basis) of the population with 95% confidence. In the conservative estimation, the allowable

stress of a member is related to the mean value of the failure stress through the following

equation:

"allowable = KdcP/ (3-1)

In the deterministic design process, the knockdown factor is a way of considering the

uncertainty in the failure stress.









Deterministic Optimization

Using the safety factor and knockdown factor, the deterministic optimization problem can

be formulated as

10
min W = pLA,
A, i=1
N, (SFPI, SFP2, A)
s.t. = ri ( allowable ),, (3-2)

0.1 < A,

where L,, N, and A, are, respectively, the length, member force, and cross-sectional area of

element i. A is the vector of cross-sectional areas, ci and (Gallowable)i are the stress and allowable

stress of an element, respectively. For this example, Gi, corresponds to the response while

(Gallowable)i to the capacity, and the loads are multiplied by a safety factor in order to consider

various uncertainties involved in the truss parameters, applied load, and computational errors.

The analytical solution for the member forces are given in the Appendix. The above optimization

problem is solved using the "fmincon" function in MATLAB. The problem converged in 7

iterations with 97 function evaluations. Table 3-2 lists the results of the deterministic

optimization. Note that elements 2, 5, and 6 cross-sectional areas reach minimum gage while

element 5 is a zero force member. At optimum design, stresses in all members except for

Member 5 and 9 are at the allowable stress.

Probability of Failure Calculation Using MCS

In this section, the probability of failure of the truss at the deterministic optimum design is

evaluated using Monte Carlo Simulation (MCS). There are two purposes in calculating the

probability of failure. First, it can evaluate the level of safety of the deterministic optimum

design. The effects of the knockdown factor and the factor of safety are evaluated in terms of the

probability of failure. Second, it can be used for the design criterion in the probabilistic









optimization. In the calculation of the probability of failure, the factor of safety and the

knockdown factor will not be considered. Instead, uncertainties related to errors and variability

in material properties, manufacturing tolerances, and applied loads will be considered in

calculating the probability of failure.

There are many uncertainties involved in the design of the ten-bar truss, such as variability

from material properties, loads, manufacturing, and errors from numerical calculation and

modeling. Failure of an element is predicted to occur when the stress in an element is greater

than its failure stress. Knowing this, the performance function can be written as


g ( -f ) -0t (3-3)

where the subscript 'true' stands for the true value of the relevant quantity, which is different

from its calculated (or predicted) value due to errors. Adding these errors, the equation can be

rewritten as

g= (-e,)o--(1+e)o- (3-4)

Here, efis the error in failure prediction, af is the predicted failure stress, e, is the error in stress

calculation, and a is the calculated stress. The errors were formulated to where positive errors

correspond to a conservative design. Therefore, the error in calculated stress is positive, while

the error in predicted failure stress is negative. Even though the stress calculation is exact for the

ten-bar truss, the error, e,, was introduced to consider the analysis of a more complex structure

where the stresses are calculated from numerical methods. The calculated stress can be written in

the following form

S= FEA[ P(l +ejP)P1,(1l+ep)P2,(l +eA )A] (3-5)

where GFEA stands for calculated stresses using FEA, epi and ep2 are errors in loads P1 and P2,

and eA is the vector of errors (tolerances) corresponding to ten cross-sectional areas. By









substituting Eq. 3-5 into Eq. 3-4, the performance function can now be rearranged in separable

form (i.e., in a form that allows the use of separable MCS) for each element as

(1+e)
S= ("),r e) E(1+ ep l)P1,(1+ eP2 )P2,(1 + eA)A] c, (3-6)
(i ef)

where c, and r are respectively, the capacity and response. Beside errors, variabilities are

introduced into the performance function through random variables af, P1, P2, and A. The

probabilistic parameters of errors and variabilities and their distribution types are listed in Table

3-3. The probabilities of failure were calculated using separable MCS, which requires smaller

number of simulations to achieve the same accuracy as crude MCS [17]. After calculating the

probabilities of failure for each element, the total probability of failure of the system can be

approximated as

10
PFS = (P)i (3-7)
i=1

where PFs is the system probability of failure. Calculating the probability failure in this form is

Ditlevesen's first-order upper bound; therefore the system probability of failure is estimated

conservatively. Using separable MCS with 106 samples, the probabilities of failure for each

element and the system are listed in Table 3-4. The results show that members 2, 6, and 10's

probability of failure contributes to 80% of the system probability of failure.

In the deterministic design process, the uncertainty in the system is considered using safety

measures, such as knockdown factor and the factor of safety. However, as is clear form Table 3-

4, the effects of these safety measures are not evenly distributed between members. It appears

that members 2, 6, and 10 are very sensitive to these safety measures, while other members are

not. They are either at minimum gage or close to it, and yet, their probabilities of failure are

relatively high compared to the other members. Thus, it is possible to move some of the weight









from non-sensitive members to the sensitive ones so that the system probability of failure can be

reduced further while maintaining the total weight of the truss. Or, it is also possible to reduce

the weight of the truss while maintaining the same level of system probability of failure. The

latter possibility will be investigated in the probabilistic optimization.

Probabilistic Optimization

Starting from the deterministic design, the probabilistic optimization problem can be

formulated such that the weight of the structure is minimized, while maintaining the same level

of system probability of failure with that of the deterministic optimum design. Thus, we have

10
min W = pLiA,
A, i=1 (3-8)
s.t. PFS < PFSt

Results of the probabilistic optimization are shown in Table 3-5. A total of 105 samples are used

for MCS. The optimization converged after 59 iterations and 728 reliability assessments. The

relatively large number of reliability assessments is due to the fact that the problem has ten

design variables. At each iteration, the optimization algorithm calculates sensitivity using finite

different method. On the other hand, the proposed ECARD method perturbs the response

directly. Thus, ECARD will be efficient when the number of response is smaller than that of

design variables.

The overall optimization took about 125 hours using a Dell desktop computer. In order to

remove instability related to random samples, a set of input random variables are generated and

repeatedly used during the optimization. Overall weight is reduced by 6% (90.47 lbs) while

maintaining the same system probability of failure as that of the deterministic optimum design.

This reduction is achieved by reallocating the risk from the higher risk members (2, 6, and 10) to

the lower risk members. The probabilistic optimization slightly increased the cross-sectional









areas of members 2, 6, and 10, and decreased the cross-sectional areas of the other members.

While the remaining member's probabilities of failure increased slightly, members 2, 6, and 10

were reduced by an order of magnitude. This risk allocation can be achieved when the

sensitivities of probability of failure and weight with respect to design variables are available. In

the probabilistic optimization, these sensitivities are calculated using the finite difference

method. That explains the 728 reliability assessments during the optimization. In the following

section, the same optimization problem will be solved using the ECARD method, which requires

a smaller amount of reliability assessments, and yet the sensitivity information can be obtained in

the approximate sense.

Approximate Probabilistic Optimization Using ECARD

In the approximate probabilistic design, the same optimization problem is used except that

the approximate probability of failure is used. Thus, the optimization problem can be written as

10
min W = pLiA,
A, i= (3-9)

s.t. Paprox < Pt

where the approximate system probability of failure is the sum of each members contributions.

Since there are ten members, ten characteristic responses and correction factors are calculated

before the ECARD optimization. This calculation is equivalent to assessing the probability of

failure twice. Then, the above optimization is deterministic because the approximate probability

of failure can be evaluated without MCS. Since the approximation in the probability of failure is

not accurate, the above ECARD optimization is repeated until the convergence criterion, as

stated in Chapter 2, is satisfied.

The accuracy of the characteristic response, which depends on the number of MCS

samples, affects the number of iterations needed to reach an accurate optimum. A low number of









samples may appear to reduce computational costs, but actually it reduces the confidence in the

probability of failure calculation resulting in an increased number of iterations to reach the

accurate optimum. The number of MCS samples must be chosen accordingly for each problem.

The results of the ECARD optimization are displayed in Table 3-6. Using 105 MCS

samples, the ECARD optimization needed only four iterations and 8 reliability assessments to

reach close to the probabilistic optimum. This is a significant reduction from the 728 reliability

assessments of the probabilistic optimization. The weight difference when comparing the fourth

iteration to the third is 0.03%, while the approximate system probability of failure equals the

deterministic system probability of failure. In addition, the errors in the member approximate

probability of failure calculations are less than 2%. Since, the probability of failure for member

nine is very small, its probability of failure error is not accurate and the error is ignored. As

expected the ECARD optimization allocates the risk between members. The cross-sectional areas

of the smaller members increased while they decreased in the largest members.









Table 3-1. Parameters for the ten-bar truss problem
Parameters Values
Dimension, b 360 inches
Safety factor, SF 1.5
Load, P1 66.67 kips
Load, P2 66.67 kips
Knockdown factor, Kdc 0.87
Density, r 0.1 lb/in3
Modulus of elasticity, E 104 ksi
Allowable stress, Gallowable 25 ksi*
-2
Minimum gage 0.1 in2
*for Element 9, allowable stress is 75 ksi


Table 3-2. Results of deterministic optimization of the ten-bar truss problem


Element Adt (in)
1 7.900
2 0.100
3 8.100
4 3.900
5 0.100


6
7
8
9
10
Total


0.100
5.798
5.515
3.677
0.141


W, (lb)
284.400
3.600
291.600
140.400
3.600
3.600
295.200
280.800
187.200
7.200
1497.600


Stress (ksi)
25.0
25.0
-25.0
-25.0
0.0
25.0
25.0
-25.0
37.5
-25.0










Table 3-3. Probabilistic distribution types, parameters of errors and variabilities in the ten-bar
truss problem
Distribution
Uncertainties type Mean Scatter
Errors
e, Uniform 0 5%
epi Uniform 0 + 10%
ep2 Uniform 0 + 10%
eA (10x1 vector) Uniform 0 + 3%
ef Uniform 0 + 20%
Variability
P1, P2 Extreme type I 66.67 kips 10% c.o.v.
A (10x1 vector) Uniform A (10x1 vector) 4% bounds
25/kdc ksi or
/ Lognormal 8% c.o.v.
75/kdc ksi


Table 3-4. Probabilities of failure of the deterministic optimum areas
Element Pe
1 2.13E-03
2 1.06E-02
3 4.80E-04
4 2.19E-03
5 4.04E-04
6 1.07E-02
7 1.69E-03
8 1.89E-03
9 5.47E-13
10 1.07E-02
Total 4.08E-02









Table 3-5. Results of the probabilistic optimization of the ten-bar truss
Elements A"et (in) A, (P/ e), (Pf),
1 7.9 7.192 2.13E-03 5.88E-03
2 0.1 0.3243 1.06E-02 3.07E-03
3 8.1 7.162 4.80E-04 8.26E-03
4 3.9 3.701 2.19E-03 2.15E-03
5 0.1 0.4512 4.04E-04 3.18E-05
6 0.1 0.3337 1.07E-02 2.14E-03
7 5.798 5.1697 1.69E-03 1.02E-02
8 5.515 4.9782 1.89E-03 3.75E-03
9 3.677 3.5069 5.47E-13 4.70E-13
10 0.141 0.4325 1.07E-02 5.46E-03
Total 1497.6 lbs 1407.13 lbs 4.08E-02 4.08E-02









Table 3-6. Results of the ECARD optimization
Element Determ. Des. Iter 1 Iter 2 Iter 3 Iter 4

AREAS (in2)
1 7.9000 7.4487 7.4787 7.4841 7.4849
2 0.1000 0.1000 0.1000 0.1000 0.1000
3 8.1000 7.0752 7.0406 7.0401 7.0402
4 3.9000 3.9382 3.9666 3.9710 3.9716
5 0.1000 0.1000 0.1000 0.1000 0.1000
6 0.1000 0.1000 0.1000 0.1000 0.1000
7 5.7980 5.0457 5.0440 5.0442 5.0441
8 5.5150 5.3538 5.3873 5.3941 5.3951
9 3.6770 3.8416 3.9657 3.9873 3.9908
10 0.1410 0.1314 0.1310 0.1309 0.1309
Weight (lb) 1497.60 1407.16 1415.94 1417.71 1418.00
MEAN STRESSES (ksi)
1 16.6667 17.7656 17.7047 17.6934 17.6918
2 16.6667 14.479 14.0059 13.9276 13.9147
3 -16.6667 -18.9868 -19.0693 -19.0688 -19.0684
4 -16.6667 -16.5606 -16.4537 -16.4378 -16.4354
5 0 4.462 4.7514 4.7913 4.7977
6 16.6667 14.479 14.0059 13.9276 13.9147
7 16.6667 18.966 18.951 18.9472 18.9468
8 -16.6667 -17.3456 -17.2576 -17.2391 -17.2363
9 25 24.0093 23.2747 23.1515 23.1313
10 -16.6667 -15.5797 -15.1199 -15.0471 -15.0354









Table 3-7. Results of the ECARD optimization
Element Determ. Des. Iter 1 Iter 2 Iter 3 Iter 4

APPROXIMATE PF


1
2
3
4
5
6
7
8
9
10
SYSTEM


1
2
3
4
5
6
7
8
9
10
SYSTEM


5.65E-03
2.16E-03
7.44E-03
1.97E-03
4.04E-04
2.17E-03
1.23E-02
3.59E-03
3.09E-14
5.17E-03
4.08E-02


5.26E-03
2.11E-03
7.51E-03
1.77E-03
1.72E-03
2.09E-03
1.20E-02
3.22E-03
2.50E-15
5.21E-03
4.08E-02


2.13E-03
1.06E-02
4.80E-04
2.19E-03
4.04E-04
1.07E-02
1.69E-03
1.89E-03
5.47E-13
1.07E-02
4.08E-02


2.13E-03
1.06E-02
4.80E-04
2.19E-03
4.04E-04
1.07E-02
1.69E-03
1.89E-03
5.47E-13
1.07E-02
4.08E-02


5.21E-03
2.10E-03
7.51E-03
1.74E-03
1.86E-03
2.09E-03
1.20E-02
3.17E-03
1.67E-15
5.22E-03
4.08E-02


5.21E-03
2.13E-03
7.51E-03
1.74E-03
1.88E-03
2.11E-03
1.20E-02
3.17E-03
1.59E-15
5.27E-03
4.10E-02


5.20E-03
2.10E-03
7.50E-03
1.74E-03
1.88E-03
2.09E-03
1.20E-02
3.16E-03
6.66E-16
5.23E-03
4.08E-02


5.20E-03
2.11E-03
7.50E-03
1.74E-03
1.88E-03
2.09E-03
1.20E-02
3.16E-03
1.48E-15
5.24E-03
4.08E-02


ACTUAL PF
5.53E-03 5.26E-03
3.09E-03 2.25E-03
6.95E-03 7.51E-03
1.96E-03 1.77E-03
1.72E-03 1.86E-03
3.09E-03 2.23E-03
1.21E-02 1.20E-02
3.49E-03 3.22E-03
2.77E-14 2.44E-15
7.14E-03 5.50E-03
4.51E-02 4.16E-02
































Figure 3-1. Geometry and loadings of the ten-bar truss









CHAPTER 4
PRACTICAL EXAMPLE: APPLICATION OF ECARD TO A WINGBOX MODEL

Introduction

The second demonstration example is the wingbox model problem. This example will

demonstrate the risk allocation between different failure modes of stress and displacement while

minimizing the total weight of the wingbox. Publically available information was used to model

the wingbox as a Boeing 767 wing in a 2.5 g dive. First, a description of the problem will be

given which will cover model geometry, and loading calculations. Then, a deterministic

optimization of the problem will be presented followed by the probability of failure calculation

using Monte Carlo simulations. Finally, the ECARD optimization will be performed, and the

efficiency and accuracy of the ECARD method will be discussed.

Problem Description

Geometry

The geometry of the wingbox was modeled using the open source Boeing 767 wing

schematics shown in Figures 4-1 and 4-2. The root dimensions have a width of 170 inches with a

height of 50 inch. The tip dimensions are 20 inches wide, 10 inches tall. The skin thickness at the

root is 1.0 inches and decreases in steps linearly to 0.1 inches at the tip of the wing. The wing is

745 inches long from root to tip.

The wingbox is modeled using ANSYS Parametric Design Language (APDL) and

analyzed using ANSYS Finite Element Analysis software. Since the APDL is a parametric

language, it can generate finite element models with different design values. The APDL code

consists of a set of commands that instructs ANSYS to build the model, add the loading and

boundary conditions, and conduct the FEA analysis. The design variables of the model are the

skin thicknesses at the root and midpoint of the wingbox. The tip thickness is fixed to 0.1 in.









The thickness varies linearly between the root and midpoint, and between the midpoint and the

tip. Figure 4-3 shows the ANSYS model of the wingbox.

The model contains a total of 5,105 nodes, 5,652 elements and was meshed using SHELL

63 Elastic Shell elements. The element contains four nodes, four thicknesses, and orthotropic

material properties. It also has six degrees of freedom at each node, translations in the x, y, and z

directions and rotations about the nodal x, y, and z axes. The boundary conditions are that all

nodes at the root of the wing are constrained in all directions. Figure 4-4 shows the meshed

wingbox model.

The material properties of the model are based on 7150-T77 aluminum, which were taken

from Metallic Materials Properties Development and Standardization (MMPDS) [21]. These

material properties are listed in Table 4-1.

Loading Calculations

The stresses and deflections of the wingbox model are calculated using mechanics of

materials methods. Determination of these stresses and deflections requires simplifying

assumptions on the geometry of the structure. Without losing generality, the following

simplifying assumptions have been made:

1. The cross section of the wing is made of thin hollow rectangular box section and
the platform is trapezoidal. The wing is modeled as a cantilever beam clamped at
its root section and free at its ends.

2. The lift over the surface of the wing area is replaced by a line load (lift per unit
length) elliptically distributed over a sweep of a line whose locus of points are
one-quarter of the chord from leading edge (c/4 line), as seen in Figure 4-5.

Wing lift distribution is directly related to the wing geometry and determines such wing

performance characteristics as induced drag, structural weight, and stalling characteristics. The

distribution of the aerodynamic lift along the span of a wing is commonly regarded as elliptical









and depends (among other things) on the taper ratio A. For an elliptical lift distribution shown in

Figure 4-6, the lift, w on the total span length of both wings, L is defined by the equation of an

ellipse.

X2 W2
+ + =1 (4-1)
(L/2)2 w

Where, Wo is the maximum load per unit length at the center of the fuselage (x = 0). The value of

Wo is determined from the gross weight of the aircraft at landing. For straight and level flight, the

total aerodynamic lift is equal to the area of the ellipse and must be equal to the gross weight of

the aircraft. Hence, lift per wing is

1
Weight of Airplane x g-force = woL (4-2)
4

Now solving for wo in Equation 4-2 at 2.5 g,

W
w =3.181- g (4-3)
L

where Wg is the aircraft gross weight. The lift per unit length w at any section of the wing is,

hence, given by


2w L
W = -- X2 (4-4)
L 2

The lift distribution varies from 747.21 lbs/in at the root to 0 lbs/in at the wingtip. The plot of the

lift distribution is seen in Figure 4-7. The elliptical lift distribution can now be converted into a

pressure load, p on the wing and the force per unit length, F acting on the leading edge of the

chord as shown in Figure 4-8. Equilibrium of forces in the vertical direction requires

Pc+F-w=0
(4-5)
Pc+F = w

Equilibrium of the moments about the quarter chord requires










c ( c c \ c c
-F--+Pc =0 F-- +Pc--=0
4 2 4 4 4 (4-6)
F = Pc

Solving forp, and F from Eq. 4-5 and Eq. 4-6 yields

w
P=-
2c (4-7)

F=
2

Using these equations, the pressure at the root was calculated to be 2.198 lbs/in2 to 0 lbs/in2 at

the tip, while the force at the leading edge was calculated to be 373.61 lbs/in to 0 lbs/in at the tip.

Plot of the pressure and force can be seen in Figures 4-9 and 4-10. Half the distribution obtained

from Eq. 4-7 is applied to the top of the model while the other half is applied to the bottom of the

model. This was done in order to reduce errors in the FEA analysis of the wingbox model.

Deterministic Optimization

The deterministic optimization problem can be formulated as

# elements
min W= P pV
i=1
s.t. Ci < (cA-basis ), (4-8)
TipDisplacement < (TipDisplacement) alow

where Vy, ci and (GA-basis) are the volume, stress and A-basis allowable stress of the material,

respectively. The design variables are the skin thicknesses at the root and midpoint of the

wingbox. Maximum allowable values for the problem are the A-basis value of 7150-T77

aluminum for stress and 68.2 inches for displacement. In this example, ci and the calculated tip

displacement corresponds to the response while (oA-basis) and allowable tip displacement to the

capacity. The loads are multiplied by a safety factor of 1.5 in order to consider various

uncertainties involved in the geometric parameters, applied load, and computational errors. The









above optimization problem is solved using the "fmincon" function in MATLAB. The

MATLAB code will input the initial design variables into ANSYS, which will conduct the

analysis. ANSYS then sends the maximum von Mises element stress, tip displacement data, and

total volume of the structure back to MATLAB, which will work to find the design parameters

that minimizes the structural weight, while satisfying stress and displacement constraints. The

optimization problem converged in 7 iterations with 36 function evaluations. Table 4-2 lists the

results of the deterministic optimization. It should be noted that the maximum stress occurs at the

root of the wing model. A contour plot of the element stress can be viewed in Figure 4-11.

Probability of Failure Calculation Using MCS

In this section, the probability of failure of the wingbox at the deterministic optimum

design is evaluated using Monte Carlo Simulation (MCS). As stated in Chapter 3, the purpose of

the probability of failure is to evaluate the level of safety of the deterministic optimum design. It

will also be used for the design criterion in the probabilistic optimization. In the calculation of

the probability of failure, the factor of safety will not be considered. Instead, uncertainties related

to variability in material properties, manufacturing tolerances, and applied loads will be

considered in calculating the probability of failure.

Failure of the wingbox is predicted to occur when its maximum von Mises stress and tip

displacement is greater than its A-basis failure stress value and allowable displacement.

Knowing this, the performance function corresponding to the failure stress mode can be written

as

gl = (a'ba.,s) 'FEA =1 r1 (4-9)

where c1 and r, are the capacity and response ofgi. Similarity, the performance function

corresponding to the displacement failure mode can be written as









g2 = (TipDisplacement)alo)ble -(TipDisplacement)F c2 r2


where c2 and r2 are the capacity and response of g2.

Variabilities are introduced into the performance function through random variables oA_

basis, (Tip Displacement)allowable and a load factor. The means and standard deviations of random

variables are listed in Table 4-3. As in the ten-bar truss example, the probabilities of failure were

calculated using separable MCS. After calculating the probabilities of failure, the total

probability of failure of the system can be approximated as

PF =Pfl +Pf2 (4-11)

where PFis the system probability of failure, Pf] is the probability of failure of the stress failure

mode, and Pp is the probability of failure of the displacement failure mode. Using separable

MCS with 106 samples, the probabilities of failure for each element and the system are listed in

Table 4-4. The results show that the probability of failure of the displacement failure mode is an

about six times larger than the probability of failure of the stress failure mode. Since the stress is

a local performance, it depends on the thickness of the root. On the other hand, the displacement

is a global performance and its value depends on the thickness of entire wingbox. Thus, it is

possible to increase the thickness of the midpoint design variable, which will reduce the

probability of failure of the displacement failure mode, while simultaneously decreasing the

thickness of root design variable, which will increase the probability of failure of the stress

failure mode. This will make it possible to reduce the weight of the wing model while

maintaining the same level of system probability of failure.

Approximate Probabilistic Optimization Using ECARD

Starting from the deterministic design, the approximate probabilistic optimization problem

can be formulated such that the weight of the structure is minimized, while maintaining the same


(4-10)









system probability of failure with that of the deterministic optimum design. Instead of the actual

probability of failure, the approximate probability of failure from ECARD is used. Thus, the

optimization problem can be written as

# elements
min W= i pVi
i=1 (4-12)
S.t. P aPProx = PDapprox + f2approx < Fdet

where the approximate system probability of failure is the sum of each approximate probability

of failure mode. Before the ECARD optimization, a characteristic response and correction

factors are calculated for each failure mode. The results of the ECARD optimization are

displayed in Table 4-5. Using 106 MCS samples, the ECARD optimization needed only two

iterations and 4 reliability assessments to reach close to an accurate optimum. Overall weight is

reduced by 0.22% (42.7 lbs) while maintaining the same system probability of failure as that of

the deterministic optimum design. This reduction is achieved by reallocating the risk from the

higher risk displacement failure mode to the lower risk stress failure mode. The ECARD

optimization slightly increased the midpoint thickness and decreased the root thickness. This

resulted in a similar level of probabilities of failure between the displacement and stress failure

modes.










Table 4-1. Material Properties of 7150-T77 Aluminum
7150-T77 Aluminum
Yield Strength = 80.5 ksi
A-Basis = 74 ksi
B-Basis = 79 ksi


Table 4-2. Results of deterministic optimization of the wingbox model
Root Thickness (in) Midpoint Thickness (in) Deterministic Weight (lbs)
0.78226 0.44164 19,174.40


Table 4-3. Variability for Wing Model
Uncertainties Distribution type Mean Standard Deviation COV
Load Factor Normal 1 0.1 10%
Failure Stress Normal 80,500 psi 5072 psi 6.30%
Displacement Normal 68.2 in 3.41 in 5%


Table 4-4. Probabilities of failure of the deterministic optimum design


Pfl


PF


5.26E-06 3.17E-05 3.69E-05


Table 4-5. ECARD optimization results
Iteration 1
Prob Thickness (in) Mean Value Approx Pf Actual Pf
Root 0.7666 Stress: 50,257 psi 1.110E-05 1.141E-05
Midpoint 0.4533 Displacement: 45.249 in 2.584E-05 2.545E-05
Total Pf: 3.694E-05 3.687E-05
Prob Weight: 19132.6 lbs

Iteration 2
Prob Thickness (in) Mean Value Approx Pf Actual Pf
Root 0.7668 Stress: 50,246 psi 1.131E-05 1.130E-05
Midpoint 0.4531 Displacement: 45.255 in 2.563E-05 2.564E-05
Total Pf: 3.694E-05 3.694E-05


19131.7 lbs


Prob Weight:














flff"AvAf ;V297


SYSTEM SCHEMATIC MANUAL


ttuWM Fra Pm


UJMM b(I


M.M m


6
3?


Figure 4-1. Boeing 767 wing dimensions



































FRONT SPAR "
WING BOX STRUCTURE


Figure 4-2. Boeing 767 internal schematic


BOEING 767


TRAILING
EDGE
./


OF WING BOX


LEADING
EDGE










AREAS
TYPE NMl




I t lI I
III

I I I I



1 I r
,I i l I

I i 'r
I


I


Figure 4-3. ANSYS model of the wingbox











ELEMENTS


Figure 4-4. Meshed ANSYS model of the wingbox







... Sweep of
c oti iiiii^ ^^^^^tc/4 line,.


Ctipl
.--.span, b-

Figure 4-5. Sweep of the quarter-chord
















Kllipi c lomlin j


W lIrp.r rjIk
s= hi:


o .6
R e-lai Sriui ]LxCiMtprI

Figure 4-6. Relationship of local lift distribution and taper ratio


800

700

600
mo-
500 -

400 -

300 -

200 -

100 -

n


Eq 3: Span-wise Lift Distribution vs Distance from Root to Wingtip


0 100 200 300 400 500 600 700 800
Distance across wing (in)


Figure 4-7. Elliptical lift distribution from the root to the tip of the wingbox model














FA IW
c/4





p



c/2


chord, c


Figure 4-8. Equilibrium of forces on the wingbox model


Pressure vs Distance from Root to Wingtip
I I I I I I I


---------r--------r--------r---------------------- --- ------ ------ -










-- -- -" ---- -- --- - -- - T-- -- -- -T- -- -- -
---------L --------L--------L --------1 -- -- -- -- -







--------- -------- -----~----------- --------------- -------- --- ----





---------- -------- -------- --- - - -------- ---- ----
1111111 i




,I







i TTTT


Illll


0 100 200 300 400 500 600
Distance across wing (in)



Figure 4-9. Pressure distribution from root to wingtip of the model


700 800











Force at Leading Edge vs Distance from Root to Wingtip


400





300 ,------------, --------- ---- ---- --------



150 .
200 --------- --------- -------- --------I --- -------------





150 t--------- -------- -------- --------; -------;---------
100 --------------- ------------- --------------







50 -------- ------- -------- -------- --------- --------
0
S150 --- -I-- -- -. -,- -,- -










0 100 200 300 400 500 6
Distance across wing (in)


Figure 4-10. Force distribution from root to wingtip of the model


00 700 800




































Figure 4-11. Contour plot of maximum stress on the wingbox model









CHAPTER 5
SUMMARY AND CONCLUSIONS

An exact-capacity approximate-response-distribution (ECARD) probabilistic optimization

method that dispenses with most of the expensive structural response calculations (typically done

via finite element analysis) was proposed in this paper. ECARD was demonstrated with two

examples. First, probabilistic optimization of a ten-bar truss problem was performed, where risk

was allocated between truss members. Then, probabilistic optimization of a wingbox was

performed, where risk was allocated between the different failure modes. From the results

obtained in these two demonstration problems, we reached to the following conclusions.

1. In the ten-bar truss problem, ECARD converged to near optima that allocated risk
between failure modes much more efficiently than the deterministic optima. The
differences between the true and approximate optima were due to the errors
involved in probability of failure estimations, which led to errors in the
derivatives of probabilities of failure with respect to design variables that is
required in risk allocation problems.

2. The approximate optimum required four inexpensive ECARD iterations and five
probability of failure calculations for the ten-bar truss example to locate the
approximate optimum. In the wingbox example, two ECARD iterations were
required and probabilities of failure of the elements are calculated three times to
locate the approximate optimum. This represents substantial reduction in the
number of probability calculation require for full probabilistic optimization.









APPENDIX
CALCULATION OF MEMBER FORCES OF THE TEN BAR TRUSS

Analytical solution to ten-bar truss problem is given in Elishakoff et al. [16]. The member

forces satisfy the following equilibrium and compatibility equations. Note: Values with "*" are

incorrect in the reference. The correct expressions are:

1
N = P, -N (A-1)


1
N,2 N-,o (A-2)


1
N3 = -P 2P N8 (A-3)


1
N4 = -P2 N1 (A-4)


1 1
N5 = P2- N8 N (A-5)


1
N6 = N10 (A-6)


N7 = -2(P + P) + N8 (A-7)


N* = ba22 a12b (A-8)
a11a22 a12a21


N9 = 2P + N10 (A-9)


N0* a11b2 a21b1
N1o b (A-10)
a,1122 12a21

where









1 1 1 2^2 2^2 ^ 1^
all*= -+-+-+ +-- (A-11)
A, A3 A5 + A7


a 2* = a21* (A-12)
A5

1 1 1 1 2 i 24 ^ 1^
a22 -+--+ + +(A-13)
SA2 A4 A5 A6 Ag A](


b -- /P2 P, + 2 P2 P 2F2 (P +P2 (A- 14)
b=4 A(A-14)
A A3 A (A-15)


-[22 -P 4P2(A-15)
24 A 5 A I









LIST OF REFERENCES


[1] Ben Haim, Y., and Elishakoff, I., Convex Models of Uncertainty in Applied Mechanics,
Elsevier, Amsterdam, 1990.

[2] Neal, D.M., Matthews, W.T., and Vangel, M.G., "Uncertainties in Obtaining High
Reliability from Stress-Strength Models," Proceedings of the 9th DOD/NASA/FAA
Conference on Fibrous Composites in Structural Design, Vol. 1, Department of Defense,
Lake Tahoe, NV, 1992, pp. 503-521.

[3] Lee, T.W and Kwak, B.M., "A Reliability-based Optimal Design Using Advanced First
Order Second Moment Method," Mechanics of Structures and Machines, Vol. 15, No. 4,
1987, pp. 523-542.

[4] Kiureghian, A.D., Zhang, Y., and Li, C.C., "Inverse Reliability Problem," Journal of
Engineering Mechanics, Vol. 120, No. 5, 1994, pp. 1154-1159.

[5] Tu, J., Choi, K.K., and Park, Y.H., "A New Study on Reliability Based Design
Optimization," ASME Journal of Mechanical Design, Vol. 121, No. 4, 1999, pp. 557-
564.

[6] Lee, J.O., Yang, Y.S., Ruy, W.S., "A Comparative Study on Reliability-index and
Target-performance-based Probabilistic Structural Design Optimization," Computers and
Structures, Vol. 80, No. 3-4, 2002, pp. 257-269.

[7] Qu, X., and Haftka, R.T., "Reliability-based Design Optimization Using Probabilistic
Sufficiency Factor," Journal of Structural and Multidisciplinary Optimization, Vol. 27,
No.5, 2004, pp. 314-325.

[8] Du, X., and Chen, W., "Sequential Optimization and Reliability Assessment Method for
Efficient Probabilistic Design," ASME Journal of Mechanical Design, Vol. 126, No. 2,
2004, pp. 225-233.

[9] Ba-abbad, M.A., Nikolaidis, E., and Kapania, R.K., "New Approach for System
Reliability-Based Design Optimization," AIAA Journal, Vol. 44, No. 5, May 2006, pp.
1087-1096.

[10] Mogami, K., Nishiwaki, S., Izui, K, Yoshimura, M., and Kogiso, N., "Reliability-based
Structural Optimization of Frame Structures for Multiple Failure Criteria Using Topology
Optimization Techniques," in press, Structural and Multidisciplinary Optimization, 2006.

[11] Acar, E., Kumar S., Pippy R.J., Kim N.H., Haftka R. T., "Approximate Probabilistic
Optimization Using Exact-Capacity-Approximate-Response-Distribution (ECARD)",
submitted, 48th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and
Materials Conference, 2006.









[12] Oberkampf, W.L., DeLand, S.M., Rutherford, B.M., Diegert, K.V. and Alvin, K.F.,
"Estimation of Total Uncertainty in Modeling and Simulation", Sandia National
Laboratory Report, SAND2000-0824, Albuquerque, NM, April 2000.

[13] Oberkampf, W.L., Deland, S.M., Rutherford, B.M., Diegert, K.V., and Alvin, K.F.,
"Error and Uncertainty in Modeling and Simulation," Reliability Engineering and System
Safety, Vol. 75, 2002, pp. 333-357.

[14] Acar, E., Kale, A. and Haftka, R.T., "Effects of Error, Variability, Testing and Safety
Factors on Aircraft Safety," Proceedings of the NSF Workshop on Reliable Engineering
Computing, 2004, pp. 103-118.

[15] Acar, E., Kale, A., and Haftka, R.T., "Comparing Effectiveness of Measures that Improve
Aircraft Structural Safety," submitted, ASCE Journal of Aerospace Engineering, 2006.

[16] Haftka, R.T., and Gurdal, Z., "Elements of Structural Optimization," Kluwer Academic
Publishers, 3rd edition, 1992.

[17] Elishakoff, I., Haftka, R.T., and Fang, J., "Structural Design Under Bounded
Uncertainty-Optimization with Anti-optimization," Computers and Structures, Vol. 53,
No. 6, 1994, pp. 1401-1405.

[18] Smarslok, B.P., Haftka, R.T., and Kim, N.H., "Taking Advantage of Separable Limit
States in Sampling Procedures," AIAA Paper 2006-1632, 47th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials
Conference, Newport, RI, May 2006.

[19] Kreyzig E., Advanced Engineering Mathematics, Wiley, New York, pp. 848.

[20] Haftka, R.T., and Gurdal, Z., "Elements of Structural Optimization," Kluwer Academic
Publishers, 3rd edition, 1992.

[21] Richard C. Rice, Jana L. Jackson, John Bakuckas, and Steven Thompson, "Metallic
Material Properties Development and Standardization (MMPDS)," Federal Aviation
Administration, January 2003.









BIOGRAPHICAL SKETCH

Richard Pippy was born in Bronxville, New York, in 1972. Upon graduation from high

school, he joined the United States Marine Corps, where he served for six years. After his

honorable discharge, he then attended St. Petersburg College where he received an Associate of

Arts degree in 2001. He then joined the University of Florida, where he earned a Bachelor of

Science degree in mechanical engineering in 2005. In 2006, he returned to the University of

Florida to pursue a master's degree in mechanical engineering. Under the supervision of Dr.

Nam-Ho Kim, he earned his master's degree in 2008.





PAGE 1

1 APPROXIMATE PROBABILISTIC OPTIMIZATION OF A WINGBOX MODEL USING EXACT CAPACITY APPR OXIMATE RESPONSE DISTRIBUTION (ECARD) By RICHARD J. PIPPY A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FL ORIDA 2008

PAGE 2

2 2008 Richard J. Pippy

PAGE 3

3 To my parents, Richard and Evelyn, and to my brother Chris

PAGE 4

4 ACKNOWLEDGMENTS I would like to express my sincere gratitud e to my advisor and the chair of my thesis committee, Dr. Nam Ho Kim and to the co c hair Dr. Rafael T. Haftka for their guidance, enthusiasm and continual support throughout my study. I am also grateful to committee member Dr. Peter G. Ifju for his advice and patience in reviewing this thesis. Special thanks go to Mulu Haile for his hel p in deriving the wingbox load conditions. I would like to thank my colleagues in the Structural and Multidisciplinary Optimization Lab at the University of F lorida i n particular, Sunil Kumar, Dr. Erdem Acar, Palani Ramu, Haoyu Wang and Saad M. Mukras for their help and encouragement

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 7 LI ST OF FIGURES ................................ ................................ ................................ ......................... 8 ABSTRACT ................................ ................................ ................................ ................................ ..... 9 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 11 2 EXACT CAPACITY RESPON SE DISTRIBUTION (ECARD) THEORY ......................... 13 Introduction ................................ ................................ ................................ ............................. 13 Characteristic Response ................................ ................................ ................................ .......... 13 Correction Factor ................................ ................................ ................................ .................... 16 Using FORM ................................ ................................ ................................ ................... 18 Using MCS ................................ ................................ ................................ ...................... 18 App roximate Probabilistic Optimization ................................ ................................ ................ 19 3 ANALYTICAL EXAMPLE: APPLICATION OF ECARD TO A TEN BAR TRUSS ........ 23 Introduction ................................ ................................ ................................ ............................. 23 Problem Description ................................ ................................ ................................ ............... 23 Deterministic Optimization ................................ ................................ ................................ .... 24 Probability of Failur e Calculation Using MCS ................................ ................................ ...... 24 Probabilistic Optimization ................................ ................................ ................................ ...... 27 Approximate Probabilistic Optimization Using ECARD ................................ ....................... 28 4 PRACTICAL EXAMPLE: APPLICATION OF ECARD TO A WINGBOX MODEL ....... 36 Introduction ................................ ................................ ................................ ............................. 36 Pro blem Description ................................ ................................ ................................ ............... 36 Geometry ................................ ................................ ................................ ......................... 36 Loading Calculations ................................ ................................ ................................ ....... 37 Determin istic Optimization ................................ ................................ ................................ .... 39 Probability of Failure Calculation Using MCS ................................ ................................ ...... 40 Approximate Probabilistic Optimization Using ECARD ................................ ....................... 41 5 SUMMARY AND CONCLUSIONS ................................ ................................ ..................... 52 APPENDIX CALCULATION OF MEMBER FORCES OF THE TEN BAR TRUSS ........... 53

PAGE 6

6 LIST OF REFERENCE S ................................ ................................ ................................ ............... 55 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ......... 57

PAGE 7

7 LIST OF TABLES Table page 3 1 Parameters for the ten bar truss problem ................................ ................................ ........... 30 3 2 Results of deterministic optimization of the ten bar truss problem ................................ ... 30 3 3 Probabilistic distribution types, parameters of errors and variabilities in the ten bar truss problem ................................ ................................ ................................ ...................... 31 3 4 Probabilities of failure of the deterministic optimum a reas ................................ ............... 31 3 5 Results of the probabilistic optimization of the ten bar truss ................................ ............ 32 3 6 Results of the ECARD optimizatio n ................................ ................................ .................. 33 3 7 Results of the ECARD optimization ................................ ................................ .................. 34 4 1 Material Properties of 7150 T77 Aluminum ................................ ................................ ..... 43 4 2 Results of deterministic optimization of the wingbox model ................................ ............ 43 4 3 Variability for Wing Model ................................ ................................ ............................... 43 4 4 Probabilities of failure of the deterministic optimum design ................................ ............. 43 4 5 ECARD optimization results ................................ ................................ ............................. 43

PAGE 8

8 LIST OF FI GURES Figure page 2 2 Calculation of the probability of failure at new design. ................................ ..................... 21 2 3 Calculation of t he probability of failure at new design ................................ ...................... 22 3 1 Geometry and loadings of the ten bar truss ................................ ................................ ....... 35 4 1 Boeing 767 wing dimensions ................................ ................................ ............................. 44 4 2 Boeing 767 internal schematic ................................ ................................ ........................... 45 4 3 ANSYS model of the wingbox ................................ ................................ .......................... 46 4 5 Sweep of the quarter chord ................................ ................................ ................................ 47 4 6 Relationship of local lift distribution and taper ratio ................................ ......................... 48 4 7 Elliptical lift distribution from the root to the tip of the wingbox model .......................... 48 4 8 Equilibrium of forces on the wingbox model ................................ ................................ .... 49 4 9 Pressure distribution from root to wingtip of the model ................................ .................... 49 4 10 Force distribution from root to wingtip of the model ................................ ........................ 50

PAGE 9

9 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science APPROXIMATE PROBABILISTIC OPTIMIZATION OF A WINGBOX MOD EL USING EXACT CAPACITY RESPONSE DISTRIBUTION (ECARD) By Richard J. Pippy May 2008 Chair: Nam Ho Kim Co chair: Rafael T. Haftka Major: Mechanical Engineering There are two major obstacles that affect probabilistic (or reliability based) structural optimi zation. First, uncertainties associated with errors in structural and aerodynamic modeling and quality of construction are not well characterized as statistical distributions and it has been shown that insufficient information may lead to large errors in p robability calculations. Second, probabilistic optimization is computationally expensive due to multiple analyses, typically finite element analyses, for calculating probability as the structure is being redesigned. In order to overcome these obstacles, we propose an approximate probabilistic optimization method where the probabilistic calculation is confined to failure stress. This takes advantage of the fact that statistical characterization of failure stresses is required by Federal Aviation Administrati on regulations. The need for expensive stress distribution calculations are eliminated by condensing the stress distribution into a representative deterministic value, transforming a probabilistic optimization problem into a semi deterministic optimization problem. By starting the approximate probabilistic optimization from the deterministic optimum design, a small number of iterations is expected and reliability analysis is required only once per iteration. This proposed method provides approximate sensiti vity of failure probability with respect to the design

PAGE 10

10 variables which is essential in risk allocation. This method is demonstrated in two examples. The first example uses a ten bar truss which demonstrate s the risk allocation between the truss elements. The second example uses a wingbox model based on a Boeing 767 400 aircraft which demonstrates the risk allocation between two different failure modes of stress and displacement

PAGE 11

11 CHAPTER 1 INTRODUCTION There are two ma jor barriers in front of probabilistic (or reliability based) structural optimization First, uncertainties associated with material properties, operating conditions, mathematical models, and manufacturing variability are not well characterized as statisti cal distributions and insufficient information may lead to large errors in probabilit y calculations (e.g., Ben Haim and Elishakoff [1], Neal, et al. [2]). Due to this fact, many engineers are reluctant to pursue probabilistic design. The second barrier to the application of probabilistic structural optimization is computational expense. Probabilistic structural optimization is expensive because repeated stress calculations (typically FEA) are required for updating probability calculation as the structure i s being re designed. Targeting these two main barriers, we propose an approximate probabilistic optimization method that dispenses with expensive probabilistic stress calculations. In the proposed method, the probabilistic calculation is confined only to f ailure stress, which is often well characterized. Traditionally, reliability based design optimization (RBDO) is performed based on a double loop optimization scheme, where the outer loop is used for design optimization while the inner loop performs a sub optimization for reliability analysis, using methods such as First Order Reliability Method (FORM). Since this traditional approach is computationally expensive, even prohibitive for problems that require complex finite element analysis (FEA), alternative methods have been proposed by many researchers (e.g., Lee and Kwak [3], Kiureghian et al. [4], Tu et al. [5], Lee et al. [6], Qu and Haftka [7], Du and Chen [8] and Ba abbad et al. [9]). These methods replace the probabilistic optimization with sequential deterministic optimization (often using inverse reliability measures) to reduce the computational expense. However, these approaches do not necessarily converge to the optimum design, and they do not easily lend themselves to

PAGE 12

12 allocating risk between failur e modes in a structure where many components can fail [10]. We note, however, that most of the computational expense is associated with repeated stress calculation. So we propose an approximate probabilistic design approach that reduces the number of expe nsive stress calculations. That is, we approximate the probabilistic optimization that separates the uncertainties which can be evaluated inexpensively and those whose effects are expensive to evaluate. We boil down the stress distribution to a single char acteristic stress by utilizing the inverse cumulative distribution of the failure stress, and we propose an inexpensive approximation of that characteristic stress. This proposed method will also improve upon a deterministic design by reallocating the safe ty margins between different components or failure modes. We call the proposed approximate probabilistic optimization approach Exact Capacity Approximate R esponse D istribution or ECAR D The purpose of this thesis is to further advance the version of ECARD which was originally developed by Dr. Erdem Acar and Dr. Rafael T. Haftka [11] There is now an improved version of ECARD which was developed by Sunil Kumar, Dr. Rafael T. Haftka, and Dr. Nam Ho Kim. The remainder of the thesis is organized as follows. Cha pter 2 discusses the theory behind the ECARD method. The application of the method to a ten bar truss problem is presented in Chapter 3 and a wingbox design in Chapter 4. Finally, the concluding remarks are listed Chapter 5.

PAGE 13

13 CHAPTER 2 EXACT CAPACI TY RESPONSE DISTRIBUTION (ECARD) THEORY Introduction In this chapter, the approximate probabilistic optimization method named ECARD will be discussed. Characteristic Response In probabilistic optimization, the system constraint is often given in terms o f failure probability of a performance function. We consider a specific form of performance function, given as (2 1) where and are capacity and response, respectively. Both the capacity and response are random because they are functions of input random variables and depend on deterministic design variables The system is considered to be failed when the response exceeds the capacity ; i.e., g ( x ; u ) < 0 We assume that the probabilistic distribution of is well known, while that of requires a l arge number of analyses. For example, when c ( x ; u ) is failure stress and r ( x ; u ) is the maximum stress of an aircraft structure, the probabilistic distribution of c ( x ; u ) is already characterized by Federal Aviation Administration requirement. However, the probabilistic distribution of r ( x ; u ) requires repeated computational simulations, such as finite element analysis. Since the performance function depends on two random variables, and the safety of t he system can be estimated using a probability of failure, defined as (2 2)

PAGE 14

14 In the above equation, is the cumulative distribution function (CDF) o f capacity, and is the probability density function (PDF) of response. The above integral can be evaluated using analytical integration, Monte Carlo simulation (MCS), or first /second order reliability method (FORM/SORM), among o thers. Smarslok et al [ 18 ] presented a separable MCS, which is much more accurate than the traditional MCS when the performance can be separable as in Eq. 2 1. It is clear from Eq. 2 2 that accurate estimation of probability of failure requires accurate a ssessment of the probability distributions of both the response and capacity. When the capacity is the failure stress, the FAA requires aircraft builders to perform characterization tests in order to construct a statistical model, and then to select the al lowable failure stress (A basis or B basis value) based on this model. Hence, the CDF of capacity is often reasonably well characterized. On the other hand, the PDF of the response is poorly known, because it depends on the accuracy of various factors, suc h as material properties, operating conditions, mathematical models, and manufacturing variability. The key idea of this research is to express P f in Eq. 2 2 using a characteristic value of the response, and approximate the change of P f in terms of the ch ange of this characteristic value. The calculation of in Eq. 2 2 can be simplified by using the information of From the Intermediate Value Theorem [ 19 ], there exist such that Eq. 2 2 can be re written as (2 3) In the above equation, the second equality is obtained from the fact that the integral of f R is one. Equation 2 3 states tha t the effect of (the poorly characterized) probability distribution of the response can be boiled down to a single characteristic response, r When the probability of

PAGE 15

15 failure is given, the characteristic response can be calculated using the inverse transf ormation of F C ( r ) as (2 4) When design variables are changed during optimization, it is possible that the distributions of b oth capacity and response may be changed. For the simplicity of presentation, w e consider the case that the distribution of the capacit y remains unchanged. We assume that the design change only affects the mean value of the response; i.e., the standard dev iation remains constant. This assumption is reasonable if the design perturbation is small. In such a case, redesign changes the mean value of response from to where is the v alue the relative change in response according to design change. Figure 2 1 illustrates the change in response distribution, along with the distribution of capacity. In this research we start the probabilistic design from a known deterministic optimum. This is an important aspect of the approximate probabilistic optimization. Since the deterministic optimization uses safety margins to consider the effect of uncertainties, the deterministic optimum design is close to the probabilistic optimum design. This will satisfy the above assumption of small design change. The goal of proposed probabilistic design is then to improve upon the det erministic design by reallocating the safety margins between different components or failure modes. First, the change in design variables will change the mean of response from to while maintaining the same st andard deviation. The change in the response distribution will then change the probability of failure according to Eq. 2 2. From Eq. 2 4, the characteristic response will also be changed from r* to r* where is the relative change in characteristic

PAGE 16

16 response. Unfortunately, t his process requires calculation of the probability of failure at the new design. The novel idea of the proposed approach is to reverse this process by approximating the re lation and so that the characteristic response at the new design can be calculated without performing reliability analysis For the moment, let us assume can be calculated from given Then, the probability of failure at the new desig n can be calculated from (2 5) The above probability of failure will be exact if is the correct relative change in characteristic response. W hen is an approximated one, the probability of failure in the above equation is approximate, and we will denote it The procedure illustrated in Figure 2 2 does not require expensive reliability analysis. It is enough to analyt ically evaluate the value of the CDF at the perturbed characteristic response. Correction Factor The key idea of the proposed approximate probability distribution is that the new characteristic response can be approximated without recourse to the expensive reliability analysis. The simplest approximation, used in this research, is that the relative change in the characteristic response, is proportional to the relative change in response, as (2 6) where is a proportional c onstant that depends on how the redesign af fects the stress distribution. In fact, i t is the sensitivity of the characteristic response change with respect to the r esponse change. W e call it a correction factor The above assumption in linearity is reasona ble when is relatively small.

PAGE 17

17 Probabilistic optimization can be viewed as risk allocation between different failure modes or different structural members. This allocation requires the sensitivity of failure probability with respect to design variables. In the pr oposed approximate probabilistic optimization, this sensitivity information is presented in the correction factor. We will demonstrate that a linear relationship between works well given the assumption of translating the stress distribution espec ially when the design change is relatively small We consider a lognorma l l y distributed capacity with mean value of C = 100 and coefficient of variation of 8%, and normally distributed response with coefficient of variation of 20%. From Eq. 2 2, the mean value of the response is chosen to be R = 42.49 so that the probability of failure becomes P f = 10 7 For a given small value o a new probability of failure is calculated from Eq. 2 2 with the mean of the response being R ( ) The relative change in characteristic response is then obtained from Eq. 2 4 with Figure 2 3 shows the relation *. We can see that the linearity assumption is quite accurate over the range of %. The slope will be the correction factor k Figure 2 4 shows the effect of the approximation on the probability of failure. In practice, the correction factor can be calculated using a finite difference method, which requires at least two reliability c alculations. We will describe the procedure using a forward finite difference method, but other method can also be used in a similar way. Let o = 0.0 corresponds to the current design, and p = 0.05 to the perturbed design. The correction factor can be ca lculated either using FORM or MSC. We will explain both cases.

PAGE 18

18 Using FORM First, the probability of failure at the current design is calculated from First Order Reliability Method (FORM) with the performance function in Eq. 2 1. If the response is perturbed by p Eq. 2 1 becomes (2 7) Using the above equation, reliability analysis is performed to calculated the perturbed probability of failure, It is noted that it is unnecessary to change design variable, because we directly perturb the output response. Thus, the computational cost of reliability analysis using Eq. 2 7 can be reduced significantly. Next, the characte ristic responses are calculated from Eq. 2 4, as (2 8) By comparing two terms in the above equation, the relative change in the characteristic res ponse can be obtained as (2 9) Then, the correction factor can be obtained from (2 10) Using MCS When MCS is employed, we generate N samples of response at the current design. In view of Eq. 2 2, the probabili ties of failure at the current and perturbed design can be calculated from

PAGE 19

19 (2 11) (2 12) The remaining procedure is identical to that of FORM. Even if Eq. 2 11 and Eq. 2 12 are two different MCS, they can be combined into one because the same sample, r i will be used. Approxi mate Probabilistic Optimization The proposed approximate probabilistic optimization is composed of two stages: (1) correction factor and initial probability of failure are calculated from reliability an alysis, and (2) a deterministic optimization problem is solved using the approximate probability of failure from Eq. 2 5. The first stage is computationally expensive, while the second stage is nothing but a semi deterministic optimization. We will explain the approximate probabilistic optimization procedure, as follows. 1. Perform deterministic optimization with safety margin. The probabilistic design starts from the deterministic optimum design; i.e., initial de s ign and cost functi on Calculate the initial probability of failure at 2. At the current design calculate deterministic value of response, u sing the mean value of input random variables. 3. Calculate the characteristic response, r 0 using the inverse CDF of the P f p and the mean and c.o.v of the response. 4. Calculate the correction factor using the procedure in the p revious section. 5. Obtain optimum design and optimum objective function by solv ing the following optimization problem: (2 13)

PAGE 20

20 Where (2 14) (2 15) (2 16) (2 17) (2 18) 6. Calculate the actual probability of fail ure, at 7. Check convergence: If i t converged, stop the process. Otherwise, set and go to Step 2 and continue The above semi deterministic optimiz ation process uses exact distribution of the capacity and an approximate distribution of response (MCS or FORM). Due to this aspect, we call it Exact Capacity Approximate Response Distribution (ECARD) method. The accuracy of ECAR D to locate the true optimum depends on the magnitudes of errors involved in the approximations. As shown in Figure 2 3 and Figure 2 4 the approximation is accurate if changes of the response due to redesign are small. In addition, the accuracy in estimat ing the correction factor affects the convergence rate of the proposed method. The result may be somewhat sub optimal because of the convergence condition and the approximate nature of the sensitivity of probability of failure. T his issue will be discussed in detail in the following section.

PAGE 21

21 Figure 2 1 Distributions of response before and after redesign. Figure 2 2 Calculation of the probability of failure at new design.

PAGE 22

22 Figure 2 3 Calculation of the probability of failure at new design Figure 2 4 Calculation of the probability of failure at new design

PAGE 23

23 CHAPTER 3 ANALYTICAL EXAMPLE: APPLICATION OF ECARD TO A TEN BAR TRUSS Introduction The first demonstration example is a ten ba r truss problem as shown in Figure 3 1. This example will demonstrate risk allocation between the different truss members. First, a brief description of the problem will be given. Then, a deterministic optimization of the problem will be presented followed by the probability of failure calculation using Monte Carlo simulations. Finally, the probabilistic and ECARD optimizations will be performed and the efficiency and accuracy of the ECARD method will be discussed Problem Description The problem descrip tion for the ten bar truss example was taken from Haftka and Gurdal [ 20 ] (page 237). The truss structure is under two loads, P 1 and P 2 The design objective is to minimize the total weight of the truss, W by varying the cross sectional areas, A i of the m embers while satisfying minimum gage constraints and allowable stress constraints. Input data for the truss is listed in Table 3 1. Member 9 was assigned a higher failure stress value in order to make the fully stressed design non optimal Aircraft design often uses a knockdown factor, K dc in order to conservatively estimate failure stress using A basis or B Basis methods The A basis (or B basis) failure stress is the value of a failure stress exceeded by 99% ( or 90% for B basis) of the pop u lation with 95 % confidence. In the conservative estimation, t he allowable stress of a member is related to the mean value of the failure stress through the following equation : (3 1) In the deterministic design process, the knockdown factor is a way of considering the uncertainty in the failure stress

PAGE 24

24 Deterministic Optimization Using the safety factor and knockdown factor, t he deterministic optim ization problem can be formulated as (3 2) w here L i N i and A i are, respectively, the length, member force, and cross sectional area of element i A is the vector of cross s ectional areas, i and allowable ) i are the stress and allowable stress of an element, respectively. i corresponds to the response while allowable ) i to the capacity, and the loads are multiplied by a safety factor in order to consider various uncert ainties involved in the truss parameters applied load, and computational errors The analytical solution for the member forces are given in the Appendix. The above optimization problem is solved using the d in 7 iterations with 97 function evaluations. Table 3 2 lists the results of the deterministic optimization. Note that elements 2, 5, and 6 cross sectional areas reach minimum gage while element 5 is a zero force member. At optimum design, stresses in al l members except for Member 5 and 9 are at the allowable stress. Probability of Failure Calculation Using MCS In this section, the probability of failure of the truss at the deterministic optimum design is evaluated using Monte Carlo Simulation (MCS). Ther e are two purposes in calculating the probability of failure. First, it can evaluate the level of safety of the deterministic optimum design. The effects of the knockdown factor and the factor of safety are evaluated in terms of the probability of failure. Second, it can be used for the design criterion in the probabilistic

PAGE 25

25 optimization. In the calculation of the probability of failure, the factor of safety and the knockdown factor will not be considered. Instead, uncertainties related to errors and variabi lity in material properties, manufacturing tolerances, and applied loads will be considered in calculating the probability of failure. There are many uncertainties involved in the design of the ten bar truss, such as variability from material properties, loads, manufacturing, and errors from numerical calculation and modeling. Failure of an element is predicted to occur when the stress in an element is greater than its failure stress. Knowing this, the performance function can be written as (3 3) from its calculated (or predicted) value due to errors. Adding these errors, t he equation ca n be rewritten as (3 4) Here, e f is the error in failure prediction, f is the predicted fa ilure stress, e is the error in stress calculation and is the calculated stress The errors were formulated to where positive errors correspond to a conservative design. Therefore, the error in calculated stress is positive, while the error in predicte d failure stress is negative. Even though the stress calculation is exact for the ten bar truss the error, e was introduced to consider the analysis of a more complex structure where the stresses are calculated from numerical methods The calcula ted str ess can be written in the following form (3 5) where FEA stands for calculated stresses using FEA, e P1 and e P2 are errors in loads P 1 and P 2 and e A is the vector of err ors (tolerances) corresponding to ten cross sectional areas. By

PAGE 26

26 substituting Eq. 3 5 into Eq. 3 4, t he performance function can now be rearranged in separable form (i.e., in a form that allows the use of separable MCS) for each element as (3 6) where and are respectively, the capacity and response. Beside errors, variabilities are introduced into the performance function through random variables f P 1 P 2 and A. The probabilistic parameters of errors and variabilities and their distribution types are listed in Table 3 3. The probabilities of failure were calculated using separable MCS, which requires smaller number of simulations to achieve the same accuracy as crude MCS [ 17 ] After calculating the probabilities of failure for each element, the total probability of failure of the system can be approximated as (3 7) where is the system probability of failure. Calculating the probability failure in this form is Ditle order upper bound; therefore the system probability of failure is estimated conservatively. Using separable MCS with 10 6 samples the probabilities of failure for each element and the system are listed in Table 3 4. The results show that members probability of failure contributes to 80% of the system prob ability of failure. In the deterministic design process, the uncertainty in the system is considered using safety measures, such as knockdown factor and the factor of safety. However, as is clear form Table 3 4, the effects of these safety measures are no t evenly distributed between members. It appears that members 2, 6, and 10 are very sensitive to these safety measures, while other members are not. They are either at minimum gage or close to it, and yet, their probabilities of failure are relatively high compared to the other members. Thus, it is possible to move some of the weight

PAGE 27

27 from non sensitive members to the sensitive ones so that the system probability of failure can be reduced further while maintaining the total weight of the truss. Or, it is als o possible to reduce the weight of the truss while maintaining the same level of system probability of failure. The latter possibility will be investigated in the probabilistic optimization. Probabilistic Optimization Starting from the deterministic desig n, the probabilistic optimization problem can be formulated such that the weight of the structure is minimized, while maintaining the same level of system probability of failure with that of the deterministic optimum design. Thus, we have (3 8) Results of the probabilistic optimization are shown in Table 3 5. A total of 10 5 samples are used for MCS. The optimization converged after 59 i terations and 728 reliability assessments. The relatively large number of reliability assessments is due to the fact that the problem has ten design variables. At each iteration, the optimization algorithm calculates sensitivity using finite different meth od. On the other hand, the proposed ECARD method perturbs the response directly. Thus, ECARD will be efficient when the number of response is smaller than that of design variables. The ov erall optimization t ook about 125 hours using a Dell deskt op compute r In order to remove instability related to random samples, a set of input random variables are generated and repeatedly used during the optimization. Overall weight is reduced by 6% (90.47 lbs) while maintaining the same system probability of failure as that of the deterministic optimum design This reduction is achieved by reallocating the risk from the higher risk members (2, 6, and 10) to the lower risk members. The probabilistic optimization slightly increased the cross sectional

PAGE 28

28 areas of members 2, 6 and 10, and decreased t he cross sectional areas of the other members. While the remaining probabilities of failure increased slightly, members 2, 6 and 10 were reduced by an order of magnitude. This risk allocation can be achieved when the sens itivities of probability of failure and weight with respect to design variables are available. In the probabilistic optimization, these sensitivities are calculated using the finite difference method. That explains the 728 reliability assessments during th e optimization. In the following section, the same optimization problem will be solved using the ECARD method, which requires a smaller amount of reliability assessments, and yet the sensitivity information can be obtained in the approximate sense. Approx imate Probabilistic Optimization Using ECARD In the approximate probabilistic design, the same optimization problem is used except that the approximate probability of failure is used. Thus, the optimization problem can be written as (3 9) where the approximate system probability of failure is the sum of each members contributions. Since there are ten members, ten characteristic response s and correction factors are calculated before the ECARD optimization. This calculation is equivalent to assessing the probability of failure twice. Then, the above optimization is deterministic because the approximate probability of failure c an be evaluat ed without MCS. Since the approximation in the probability of failure is not accurate, the above ECARD optimization is repeated until the convergence criterion, as stated in Chapter 2, is satisfied. The accuracy of the characteristic response, which depen ds on the number of MCS samples, affects the number of iterations needed to reach an accurate optimum. A low number of

PAGE 29

29 samples may appear to reduce computational costs, but actually it reduces the confidence in the probability of failure calculation result ing in an increased number of iterations to reach the accurate optimum. The number of MCS samples must be chosen accordingly for each problem. The results of the ECARD optimization are displayed in Table 3 6. Using 10 5 MCS samples, the ECARD optimization n eeded only four iterations and 8 reliability assessments to reach close to the probabilistic optimum This is a significant reduction from the 728 reliability assessments of the probabilistic optimization The weight difference when comparing the fourth it eration to the third is 0.03%, while the approximate system probability of failure equals the deterministic system probability of failure. In addition, the errors in the member approximate probability of failure calculations are less than 2%. Since, the pr obability of failure for member nine is very small, its probability of failure error is not accurate and the error is ignored As expected the ECARD optimization allocates the risk between members. T he cross sectional areas of the smaller members increased while they decreased in the largest members.

PAGE 30

30 Table 3 1. Parameters for the ten bar truss problem Parameters Values Dimension, b 360 inches Safety factor, S F 1.5 Load, P 1 66.67 kips Load, P 2 66.67 kips Knockdown factor, K dc 0.87 Density, r 0.1 lb/ in 3 Modulus of elasticity, E 10 4 ksi Allowable stress, allowable 25 ksi* Minimum gage 0.1 in 2 *for Element 9, allowable stress is 75 ksi Table 3 2. Results of deterministic optimization of the ten bar truss problem Element A i det (in) W i (lb) Str ess (ksi) 1 7.900 284.400 25.0 2 0.100 3.600 25.0 3 8.100 291.600 25.0 4 3.900 140.400 25.0 5 0.100 3.600 0.0 6 0.100 3.600 25.0 7 5.798 295.200 25.0 8 5.515 280.800 25.0 9 3.677 187.200 37.5 10 0.141 7.200 25.0 Total --1497.600 --

PAGE 31

31 Table 3 3. Proba bilistic distribution types, parameters of errors and variabilities in the ten bar truss problem Uncertainties Distribution type Mean Scatter Errors e Uniform 0 5% e P1 Uniform 0 10% e P2 Uniform 0 10% e A (10 1 vector) Uniform 0 3% e f Uniform 0 20% Variability P 1 P 2 Extreme type I 66.67 k ips 10% c.o.v. A (10 1 vector) Uniform A (10 1 vector) 4% bounds Lognormal 25 /k dc ksi or 8% c.o.v. 75 /k dc ksi Table 3 4. P robabilities of failure of the d eterministic optimum areas Element P f det 1 2.13E 03 2 1.06E 02 3 4.80E 04 4 2.19E 03 5 4.04E 04 6 1.07E 02 7 1.69E 03 8 1.89E 03 9 5.47E 13 10 1.07E 02 Total 4.08E 02

PAGE 32

32 Table 3 5. Results of the probabilistic optimization of the ten bar truss Elements A i det (in) A i ( P f det ) i ( P f ) i 1 7.9 7.192 2.13E 03 5.88E 03 2 0.1 0.3243 1.06E 02 3.07E 03 3 8.1 7.162 4.80E 04 8.26E 03 4 3.9 3.701 2.19E 03 2.15E 03 5 0.1 0.451 2 4.04E 04 3.18E 05 6 0.1 0.3337 1.07E 02 2.14E 03 7 5.798 5.1697 1.69E 03 1.02E 02 8 5.515 4.9782 1.89E 03 3.75E 03 9 3.677 3.5069 5.47E 13 4.70E 13 10 0.141 0.4325 1.07E 02 5.46E 03 Total 1497.6 lbs 1407.13 lbs 4.08E 02 4.08E 02

PAGE 33

33 Tabl e 3 6. Results of the ECARD optimization Element Determ. Des. Iter 1 Iter 2 Iter 3 Iter 4 AREAS (in 2 ) 1 7.9000 7.4487 7.4787 7.4841 7.4849 2 0.1000 0.1000 0.1000 0.1000 0.1000 3 8.1000 7.0752 7.0406 7.0401 7.0402 4 3.9000 3.9382 3.9666 3.9710 3.9716 5 0.1000 0.1000 0.1000 0.1000 0.1000 6 0.1000 0.1000 0.1000 0.1000 0.1000 7 5.7980 5.0457 5.0440 5.0442 5.0441 8 5.5150 5.3538 5.3873 5.3941 5.3951 9 3.6770 3.8416 3.9657 3.9873 3.9908 10 0.1410 0.1314 0.1310 0.1309 0.1309 Weight (lb) 1497.60 1407.1 6 1415.94 1417.71 1418.00 MEAN STRESSES (ksi) 1 16.6667 17.7656 17.7047 17.6934 17.6918 2 16.6667 14.479 14.0059 13.9276 13.9147 3 16.6667 18.9868 19.0693 19.0688 19.0684 4 16.6667 16.5606 16.4537 16.4378 16.4354 5 0 4.462 4.7514 4.7913 4.7 977 6 16.6667 14.479 14.0059 13.9276 13.9147 7 16.6667 18.966 18.951 18.9472 18.9468 8 16.6667 17.3456 17.2576 17.2391 17.2363 9 25 24.0093 23.2747 23.1515 23.1313 10 16.6667 15.5797 15.1199 15.0471 15.0354

PAGE 34

34 Table 3 7 Results of th e ECARD optimization Element Determ. Des. Iter 1 Iter 2 Iter 3 Iter 4 APPROXIMATE P F 1 2.13E 03 5.65E 03 5.26E 03 5.21E 03 5.20E 03 2 1.06E 02 2.16E 03 2.11E 03 2.10E 03 2.10E 03 3 4.80E 04 7.44E 03 7.51E 03 7.51E 03 7.50E 03 4 2.19E 03 1.97E 03 1.77 E 03 1.74E 03 1.74E 03 5 4.04E 04 4.04E 04 1.72E 03 1.86E 03 1.88E 03 6 1.07E 02 2.17E 03 2.09E 03 2.09E 03 2.09E 03 7 1.69E 03 1.23E 02 1.20E 02 1.20E 02 1.20E 02 8 1.89E 03 3.59E 03 3.22E 03 3.17E 03 3.16E 03 9 5.47E 13 3.09E 14 2.50E 15 1.67E 15 6. 66E 16 10 1.07E 02 5.17E 03 5.21E 03 5.22E 03 5.23E 03 SYSTEM 4.08E 02 4.08E 02 4.08E 02 4.08E 02 4.08E 02 ACTUAL P F 1 2.13E 03 5.53E 03 5.26E 03 5.21E 03 5.20E 03 2 1.06E 02 3.09E 03 2.25E 03 2.13E 03 2.11E 03 3 4.80E 04 6.95E 03 7.51E 03 7.51E 03 7 .50E 03 4 2.19E 03 1.96E 03 1.77E 03 1.74E 03 1.74E 03 5 4.04E 04 1.72E 03 1.86E 03 1.88E 03 1.88E 03 6 1.07E 02 3.09E 03 2.23E 03 2.11E 03 2.09E 03 7 1.69E 03 1.21E 02 1.20E 02 1.20E 02 1.20E 02 8 1.89E 03 3.49E 03 3.22E 03 3.17E 03 3.16E 03 9 5.47E 13 2.77E 14 2.44E 15 1.59E 15 1.48E 15 10 1.07E 02 7.14E 03 5.50E 03 5.27E 03 5.24E 03 SYSTEM 4.08E 02 4.51E 02 4.16E 02 4.10E 02 4.08E 02

PAGE 35

35 Figure 3 1. Geometry and loadings of the ten bar truss

PAGE 36

36 CHAPTER 4 PRACTICAL EXAMPLE: APPLICATION OF ECARD T O A WINGBOX MODEL Introduction The second demonstration example is the wingbox model problem This example will demonstrate the risk allocation between different failure modes of stress and displacement while minimizing the total weight of the wingbox Pub lically available information was used to model the wingbox as a Boeing 767 wing in a 2.5 g dive First, a description of the problem will be given which will cover model geometry, and loading calculations Then, a deterministic optimization of the problem will be presented followed by the probability of failure calculation using Monte Carlo simulations. Finally, the ECARD optimization will be performed, and the efficiency and accuracy of the ECARD method will be discussed Problem Description Geometry The geometry of the wingbox was modeled using the open source Boeing 767 wing schematics shown in Figures 4 1 and 4 2. The root dimensions have a width of 170 inches with a height of 50 inch. The tip dimensions are 20 inches wide, 10 inches tall. The skin thi ckness at the root is 1.0 inches and decreases in steps linearly to 0.1 inches at the tip of the wing. The wing is 745 inches long from root to tip. The wingbox is modeled using ANSYS Parametric Design Language (APDL) and analyzed using ANSYS Finite Eleme nt Analysis software. Since the APDL is a parametric language, it can generate finite element models with different design values. The APDL code consists of a set of commands that instructs ANSYS to build the model, add the loading and boundary conditions, and conduct the FEA analysis. The design variables of the model are the skin thicknesses at the root and midpoint of the wingbox. The tip thickness is fixed to 0.1 in.

PAGE 37

37 The thickness varies linearly between the root and midpoint, and between the midpoint and the tip. Figure 4 3 shows the ANSYS model of the wingbox. The model contains a total of 5,105 nodes, 5,652 elements and was meshed using SHELL 63 Elastic Shell elements. The element contains four nodes, four thicknesses, and orthotropic material proper ties. It also has six degrees of freedom at each node, translations in the x, y, and z directions and rotations about the nodal x, y, and z axes. The boundary conditions are that all nodes at the root of the wing are constrained in all directions. Figure 4 4 shows the meshed wingbox mode l The material properties of the model are based on 7150 T77 aluminum, which were taken from Metallic Materials Properties Development and Standardization (MMPDS) [21 ]. The se material properties are listed in Table 4 1. Loa ding C alculations The stresses and deflections of the wingbox model are calculated using mechanics of materials methods. Determination of these stresses and deflections requires simplifying assumptions on the geometry of the structure. Without losing gene rality, the following simplifying assumptions have been made: 1. The cross section of the wing is made of thin hollow rectangular box section and the platform is trapezoidal. The wing is modeled as a cantilever beam clamped at its root section and free at its ends. 2. The lift over the surface of the wing area is replaced by a line load (lift per unit length) elliptically distributed over a sweep of a line whose locus of pointes are one quarter of the chord from leading edge (c/4 line), as seen in Figure 4 5 Wing lift distribution is directly related to the wing geometry and determines such wing performance characteristics as induced drag, structural weight, and stalling characteristics. The distribution of the aerodynamic lift along the span of a wing is com monly regarded as elliptical

PAGE 38

38 and depends (among other things) on the taper ratio For an elliptical lift distribution shown in Fig ure 4 6 the lift w on the total span length of both wings, L is defined by the equation of an ellipse. (4 1) Where, w o is the maximum load per unit length at the center of the fuselage ( x = 0). The value of w o is determined from the gross weight of t he aircraft at landing. For straight and level flight, the total aerodynamic lift is equal to the area of the ellipse and must be equal to the gross weight of the aircraft. Hence, lift per wing is Weight of Airplane x g force (4 2) Now solving for w o in Equation 4 2 a t 2.5 g, (4 3) where W g is the aircraft gross weight. The lift per unit length w at any section of the wing is, hence, given by (4 4) The lift distribution va ries from 747.21 lbs/in at the root to 0 lbs/in at the wingtip. The plot of the lift di stribution is seen in Figure 4 7 The elliptical lift distribution can now be converted into a pressure load, p on the wing and the force per unit length F acting on th e leading edge of the chord as shown in Figure 4 8 Equilibrium of forces in the vertical direction requires (4 5) Equilibriu m of the moments about the quarter chord requires

PAGE 39

39 (4 6) Solving for p and F from Eq. 4 5 and Eq. 4 6 yields (4 7) Using these equations, the pressure at the root was calculated to be 2.198 lbs/in 2 to 0 lbs/in 2 at the tip, while the force at the leading edge was calculated to be 373 .61 lbs/in to 0 lbs/in at the tip. Plot of the pressure and force can be seen in Figures 4 9 and 4 10 Half the distribution obtained from Eq. 4 7 is applied to the top of the model while the other half is applied to the bottom of the model. This was done in order to reduce errors in the FEA analysis of the wingbox model. Deterministic Optimization The deterministic optimization problem can be formulated as (4 8) where V i i and A basis ) are the volume, stress and A basis allowable stress of the material respectively The design variables are the skin thicknesses at the root and midpoint of the wingbox. Maximum allowable values for the problem are the A basis value of 7150 T77 aluminum for stress and 68.2 inches for displacement. In i and the calculated tip displacement A basis ) and a llowable tip displacement to the capacity. T he loads are multiplied by a safety factor of 1.5 in order to consider various uncertainties involved in the geometric parameters applied load, and computational errors The

PAGE 40

40 MATLAB code will input the initial design variables into ANSYS, which will conduct the analysis. ANSYS then sends the maximum von M ises element stress, tip displacement data and total volume of the structure back to MATLAB, which will work to find the design parameters that minimizes the structur al weight, while satisfying stress and displacement constraints The optimization problem converged in 7 iterations with 36 function evaluations. Table 4 2 lists the results of the deterministic optimization. It should be noted that the maximum stress occu rs at the root of the wing model. A contour plot of the element stress can be viewed in Figure 4 11. Probability of Failure Calculation Using MCS In this section, the probability of failure of the wingbox at the deterministic optimum design is evaluated u sing Monte Carlo Simulation (MCS). As stated in C hapter 3, the purpose of the probability of failure is to evaluate the level of safety of the deterministic optimum design. It will also be used for the design criterion in the probabilistic optimization. In the calculation of the probability of failure, the factor of safety will not be considered. Instead, uncertainties related to variability in material properties, manufacturing tolerances, and applied loads will be considered in calculating the probability of failure. Failure of the wingbox is predicted to occur when its maximum von Mises stress and tip displacement is greater than its A basis failure stress value and allowable displacement. Knowing this, the performance function corresponding to the failu re stress mode can be written as (4 9) where and are the capacity an d response of g 1 Similarity, the performance function corresponding to the displacement failure mode can be written as

PAGE 41

41 (4 10) where and are the capa city and response of g 2 Variabilities are introduced into the performance function through random variables A basis (Tip Displacement) allowable and a load factor. The means and standard deviations of random variables are listed in Table 4 3. As in the ten bar truss example, the probabilities of failure were calculated using separable MCS. After calculating the probabili ties of failure the total probability of failure of the system can be approximated as (4 11) where P F is the syste m probability of failure, P f1 is the probability of failure of the stress failure mode, and P f2 is the probability of failure of the displacement failure mode. Using separable MCS with 10 6 samples, the probabilities of failure for each element and the syst em are listed in Table 4 4. The results show that the probability of failure of the displacement failure mode is an about six times larger than the probability of failure of the stress failure mode. Since the stress is a local performance, it depends on th e thickness of the root. On the other hand, the displacement is a global performance and its value depends on the t hickness of entire wingbox. Thu s, it is possible to increase the thickness of the midpoint design variable, which will reduce the probabili ty of failure of the displacement failure mode, while simultaneously decreasing the thickness of root design variable, which will increase the probability of failure of the stress failure mode. This will make it possible to reduce the weight of the wing mo del while maintaining the same level of system probability of failure. Approximate Probabilistic Optimization Using ECARD Starting from the deterministic design, the approximate probabilistic optimization problem can be formulated such that the weight of the structure is minimized, while maintaining the same

PAGE 42

42 system probability of failure with that of the dete rministic optimum design. In stead of the actual probability of failure, the approximate probability of failure from ECARD is used. Thus, the optimizat ion problem can be written as (4 12) where the approximate system probability of failure is the sum of each approximate probability of failure mode. Before the ECARD optim ization, a characteristic response and correction factors are calculated for each failure mode. The results of the ECARD optimization are displayed in Table 4 5. Using 10 6 MCS samples, the ECARD optimization needed only two iterations and 4 reliability ass essments to reach close to an accurate optimum. Overall weight is reduced by 0.22% (42.7 lbs) while maintaining the same system probability of failure as that of the deterministic optimum design. This reduction is achieved by reallocating the risk from the higher risk displacement failure mode to the lower risk stress failure mode. The ECARD optimization slightly increased the midpoint thickness and decreased the root thickness. This resulted in a similar level of probabilities of failure between the displa cement and stress failure modes.

PAGE 43

43 Table 4 1. Material Properties of 7150 T77 Aluminum 7150 T77 Aluminum Yield Strength = 80.5 ksi A Basis = 74 ksi B Basis = 79 ksi Table 4 2. Results of deterministic optimization of the wingbox model Root Th ickness (in) Mid point Thickness (in) Deterministic Weight (lbs) 0.78226 0.44164 19,174.40 Table 4 3. Variability for Wing Model Uncertainties Distribution type Mean Standard Deviation COV Load Factor Normal 1 0.1 10% Failure Stress Normal 80,500 psi 5072 psi 6.30% Displacement Normal 68.2 in 3.41 in 5% Table 4 4 P robabilities of failure of the deterministic optimum design P f1 P f2 P F 5.26 E 06 3.17 E 05 3.69 E 05 Table 4 5. ECAR D optimization results Iteration 1 Prob Thickness (in) Mean Value Approx Pf Actual Pf Root 0.7666 Stress: 50,257 psi 1.110E 05 1.141E 05 Midpoint 0.4533 Displacement: 45.249 in 2.584E 05 2.545E 05 Total Pf: 3.694E 05 3.687E 05 Prob Weight: 19132.6 lbs Iteration 2 Prob Thickness (in) M ean Value Approx Pf Actual Pf Root 0.7668 Stress: 50,246 psi 1.131E 05 1.130E 05 Midpoint 0.4531 Displacement: 45.255 in 2.563E 05 2.564E 05 Total Pf: 3.694E 05 3.694E 05 Prob Weight: 19131.7 lbs

PAGE 44

44 Figure 4 1. Boeing 767 wing dimensions

PAGE 45

45 F igure 4 2. Boeing 767 internal schematic

PAGE 46

46 Figure 4 3. ANSYS model of the wingbox

PAGE 47

47 Figure 4 4. Meshed ANSYS model of the wingbox Figure 4 5 Sweep of the quarter chord

PAGE 48

48 Figure 4 6 Relationship of local lift distribution and taper ratio Figure 4 7 Elliptical lift distribution from the roo t to the tip of the wingbox model

PAGE 49

49 Figure 4 8 Equilibrium of forces on the wingbox model Figure 4 9 Pressure distribution from root to wingtip of the model W F c/2 p c/4 chord, c

PAGE 50

50 Figure 4 10 Force distribution from root to wingtip of the m odel

PAGE 51

51 Figure 4 11. Contour plot of maximum stress on the wingbox model

PAGE 52

52 CHAPTER 5 SUMMARY AND CONCLUSI ONS An exact capacity approximate response distribution (ECARD) probabilistic optimization method that dispenses with most of the expensive structural response calculations (typically done via finite element analysis) was proposed in this paper. ECAR D was demonstrated with two examples. First, probabilistic optimization of a ten bar truss problem was performed, where risk was allocated between truss mem bers. Then, probabilistic optimization of a wingbox was performed, where risk was allocated between the different failure modes. From the results obtained in these two demonstration problems, we reached to the following conclusions. 1. In the ten bar truss pr oblem ECARD converged to near optima that allocated risk between failure modes much more efficiently than the deterministic optima. The differences between the true and approximate optima were due to the errors involved in probability of failure estimatio ns, which led to errors in the derivatives of probabilities of failure with respect to design variables that is required in risk allocation problems. 2. The approximate optimum required four inexpensive ECARD iterations and five probability of failure calculations for the ten bar truss example to locate the approximate optimum. In the wingbox example, two ECAR D iterations were required and probabilities of failure of the elements are calculated three times to locate the approximate optimum. This represents substantial reduction in the number of probability calculation require for full probabilistic optimization.

PAGE 53

53 APPENDIX CALCULATION OF MEMBE R FORCES OF THE TEN BAR TRUSS Analytical solution to ten bar truss problem is given in Elishakoff et al. [ 16 ] The member incorrect in the reference. The corre ct expressions are: (A 1) (A 2) (A 3) (A 4) (A 5) (A 6) (A 7 ) (A 8) (A 9) (A 10) where

PAGE 54

54 (A 11) (A 12) (A 13) (A 14) (A 15)

PAGE 55

55 LIST OF REFERENCE S [1] Ben Haim, Y., and Elishakof f, I., Convex Models of Uncertainty in Applied Mechanics, Elsevier, Amsterdam, 1990. [2] Reliability from Stress Conference on Fibrous Composites in Structural Design, Vol. 1, Department of Defense, Lake Tahoe, NV, 1992, pp. 503 521. [3] based Optimal Design Using Advanced First Vol. 15, No. 4, 1987, pp. 523 542. [4] Engineering Mechanics, Vol. 120, No. 5, 1994, pp. 1154 1159. [5] esign 564. [6] index and Target performance nd Structures, Vol. 80, No. 3 4, 2002, pp. 257 269. [7] based Design Optimization Using Probabilistic No.5, 2004, pp. 314 325. [8] Du, 2004, pp. 225 233. [9] Ba ystem Reliability 1087 1096. [10] based Structural Optimization of Frame Structures for Multiple Failure Cri teria Using Topology [11] Optimization Using Exact Capacity Approximate Response Distrib submitted, 48th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 2006.

PAGE 56

56 [12] Oberkampf, W.L., DeLand, S.M., Rutherford, B.M., Diegert, K.V. and Alvin, K.F., Laboratory Report, SAND2000 0824, Albuquerque, NM, April 2000. [13] Oberkampf, W.L., Deland, S.M., Rutherford, B.M., Diegert, K.V., and Alvin, K.F., S afety, Vol. 75, 2002, pp. 333 357. [14] Computing, 2004, pp. 103 118. [15] Acar, E., Kale, A., and Haftka, R.T., "Comparing Effectiveness of Measures that Improve Aircraft Structural Safety," submitted, ASCE Journal of Aerospace Engineering, 2006. [16] Publishers, 3 rd edition, 1992. [17] Uncertainty Optimization with Anti No. 6, 1994, pp. 1401 1405. [18] Taking Advantage of Separable Limit 1632, 47th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, Newport, RI, May 2006. [19] Kreyzig E ., Advanced Engineering Mathematics, Wiley, Ne w York, pp. 848. [20] Publishers, 3rd edition, 1992. [21] Richard C. Rice, Jana L. Jackson, John Bakuckas, and Steven Thompson Material Properties Development and Stand Administration, January 2003.

PAGE 57

57 BIOGRAPHICAL SKETCH Richard Pippy was born in Bronxville, New York in 1972. Upon graduation from high school, he joined the United States Marine Corps where he served for six years. Aft er his honorable discharge, he then attended St. Petersburg College where he received an Associate of A rts degree in 2001. He then joined the University of Florida where he earned a Bachelor of Science degree in mechanical engineering in 2005. In 2006, he returned to the University of Nam