<%BANNER%>

Record for a UF thesis. Title & abstract won't display until thesis is accessible after 2010-05-31.

Permanent Link: http://ufdc.ufl.edu/UFE0022127/00001

Material Information

Title: Record for a UF thesis. Title & abstract won't display until thesis is accessible after 2010-05-31.
Physical Description: Book
Language: english
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: Environmental Engineering Sciences -- Dissertations, Academic -- UF
Genre: Environmental Engineering Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Thesis: Thesis (Ph.D.)--University of Florida, 2008.
Local: Adviser: Sansalone, John.
Electronic Access: INACCESSIBLE UNTIL 2010-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022127:00001

Permanent Link: http://ufdc.ufl.edu/UFE0022127/00001

Material Information

Title: Record for a UF thesis. Title & abstract won't display until thesis is accessible after 2010-05-31.
Physical Description: Book
Language: english
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: Environmental Engineering Sciences -- Dissertations, Academic -- UF
Genre: Environmental Engineering Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Thesis: Thesis (Ph.D.)--University of Florida, 2008.
Local: Adviser: Sansalone, John.
Electronic Access: INACCESSIBLE UNTIL 2010-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022127:00001


This item has the following downloads:


Full Text

PAGE 1

1 EXPERIMENTAL AND NUMERICAL ANALYSIS OF STORMWATER UNIT OPERATIONS AND PROCESSES By SUBBU-SRIKANTH PATHAPATI A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2008

PAGE 2

2 2008 Subbu-Srikanth Pathapati

PAGE 3

3 To my grandfather, the late Pathapati Subba Rao

PAGE 4

4 ACKNOWLEDGMENTS First and forem ost, I extend my gratitude to my advisor, Dr. John Sansalone, for his continued support, encouragement and guidance. In addition to academic knowledge, I have learned the elements of a distinguished work ethic from him. I will take these experiences with me, in all my future career endeavors. I extend my sincere appreciation to the memb ers of my graduate committee: Dr. James Heaney, Dr. Ben Koopman and Dr. Jennifer Curtis for their interest, advice and insight. I am forever in debt to them for their guidance. I thank my dear colleagues and friends who have helped me in the lab and in the field: Dr. Jong-Yeop Kim, Dr. Bo Liu, Dr. Gaoxia ng Ying, Natalie Magill, Robert Rooney, Adam Winberry, Dr.Ki-Joon Jeon, Saurabh Raje, Ruben Ke rtesz, Tingting Wu, Sandeep Gulati ,Hwanchul Cho, Giuseppina Garofalo, Josh Dickens on, Paul Indeglia, Dr. Ti anpeng Guo Dr. Xuheng Kuang, Will Barlett, and Matt McGaugh. I thank the many friends I have made duri ng these years, for being there for me throughout and would like to name a few of these: Dr. Srinivas Gopal Krishna, Ashwin Chittoor, Chetan Salimath, Arpit Mathur, Bharath Thiruve ngadachari, Karthik Bhar at, Avinash Rajendran, Kishore Menon, Rupa Nair,.Sruti Ramnath Ad itya, Preeti Bhuvan, Aditya Ramachandran, Praveen Sampath, Jayaram Balasubramania n, Karthik Chepudira, and many others. I thank my mother Mrs. Lakshmi Govindaraju, my aunt Ms. Padma Pathapati, my grandmother, Mrs. Sundari Pathapati, and to my uncle Mr. Kotesh Govind araju for their support. I thank Dr. Phillip Barkley of the University of Florida Health Cent er for taking care of me. I express my sincere gratitude to Dr. Jocelyn Lee of the Univer sity of Florida Health Center, for her tremendous kindness, insight and compassion.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........8 LIST OF FIGURES.........................................................................................................................9 LIST OF ABBREVIATIONS........................................................................................................ 11 ABSTRACT...................................................................................................................................17 1 GLOBAL INTRODUCTION.................................................................................................19 2 EXPERIMENTAL MODIFICATION OF A ST ORMWATER HYDRODYNAMIC SEPARATOR FOR ENHANCED PARTICLE SEPARATION........................................... 25 Introduction................................................................................................................... ..........25 Background.............................................................................................................................27 Objectives...............................................................................................................................28 Methodology...........................................................................................................................29 Experimental Site Setup.................................................................................................. 29 Influent Particle Gradation..............................................................................................30 Field Test Procedure........................................................................................................31 Laboratory Analyses........................................................................................................ 32 Geometric Configurations............................................................................................... 33 Tracer Analysis................................................................................................................35 Results and Discussions........................................................................................................ ..38 Treatment Influent Flow Rate at 25 % of Design Flow Rate .......................................... 38 Treatment Influent Flow Rate at 75 % of Design Flow Rate .......................................... 39 Treatment Influent Flow Rate at 125 % of Design Flow Rate ........................................ 41 Discussion..................................................................................................................... ...42 Tracer Study Results........................................................................................................ 43 Conclusions.............................................................................................................................46 3 CFD MODELING OF A STORMWATER HYDRODYNAMIC SEPARATOR .................63 Introduction................................................................................................................... ..........63 Objectives...............................................................................................................................67 Methodology...........................................................................................................................68 Pilot-scale Testing Setup................................................................................................. 68 Influent Particle Gradation..............................................................................................68 Field Test Procedure........................................................................................................69 Laboratory Analyses........................................................................................................ 70 Computational Fluid Dynamics Methodology................................................................ 71 Modeling Fluid Flow....................................................................................................... 71

PAGE 6

6 Modeling the Static Screen.............................................................................................. 73 Modeling the Particulate Phase....................................................................................... 75 Discretization of Geometry............................................................................................. 76 Discretization of Governing Equations...........................................................................77 Solution Schemes............................................................................................................77 Results and Discussions........................................................................................................ ..78 PSD Results.....................................................................................................................78 CFD Model Results.........................................................................................................78 Post-processing CFD Model Results: Particle Dynam ics and Hydrodynamics.............. 80 Dynamics of a Particle with dp= 450 m........................................................................ 82 Dynamics of a Particle with dp= 25 m.......................................................................... 82 Conclusions.............................................................................................................................83 4 COMBINING PARTICLE ANALYSES AND CFD MODELING TO PREDICT HETERO-DISPERSE PARTICULATE MATTE R FATE AND PRESSURE DROP I N A PASSIVE RAINFALL-R UNOFF RADIAL FILTER........................................................ 94 Introduction................................................................................................................... ..........94 Objectives...............................................................................................................................96 Methodology...........................................................................................................................97 Experimental Setup......................................................................................................... 97 Media Characteristics...................................................................................................... 98 Prototype Test Procedure................................................................................................98 Laboratory Analyses........................................................................................................ 99 Pressure Head Measurements........................................................................................101 Computational Fluid Dynamics Model......................................................................... 101 Modeling Flow in Porous Media................................................................................... 102 Modeling the Particulate Phase..................................................................................... 104 Discretization and Solution Schemes............................................................................106 Results and Discussions........................................................................................................ 106 Experimental Results.....................................................................................................106 CFD Model Results.......................................................................................................107 Particle Separation.........................................................................................................107 Head loss and Pressure Distributions............................................................................ 108 Conclusions...........................................................................................................................110 5 MODELING HYDRAULICS AND PARTICLE DYNAMICS OF A STORMWATER HYDRODYNAMIC SEPARA TOR FOR TRANSIENT INFLUENT LOADS .................. 123 Introduction................................................................................................................... ........123 Objectives.............................................................................................................................126 Methodology.........................................................................................................................127 Experimental Methodology........................................................................................... 127 Multiphase Flow Modeling Methodology..................................................................... 129 Modeling the Particulate Phase..................................................................................... 134 Discretization and Solution Schemes............................................................................136

PAGE 7

7 Results and Discussions........................................................................................................ 136 Conclusions...........................................................................................................................140 6 A PARTICLE SEPARATION MODEL OF A VOLUME TRIC CLARIFYING FILTER FOR SOURCE AREA RAINFALL -RUNOFF PARTICULATE MATTER...................... 148 Introduction................................................................................................................... ........148 Media filtration of rainfall-runoff..................................................................................148 Modeling Approach....................................................................................................... 151 Objectives.............................................................................................................................152 Methodology.........................................................................................................................152 VCF and Watershed Configuration............................................................................... 152 Data Acquisition and Management............................................................................... 154 Influent and Effluent Sampling and Analysis ............................................................... 156 Multiphase flow Modeling Methodology...................................................................... 158 Modeling the Particulate Phase..................................................................................... 161 Discretization and Solution Schemes............................................................................163 Time Discretization....................................................................................................... 164 Results and Discussion......................................................................................................... 165 Event-Based Hydrologic Lo adings and Response ........................................................165 Filter Media Cartridge Head Loss Modeling Results.................................................... 166 Separation of Particulate Matter Modeling Results .......................................................167 Conclusions...........................................................................................................................169 7 GLOBAL CONCLUSIONS.................................................................................................182 LIST OF REFERENCES.............................................................................................................185 BIOGRAPHICAL SKETCH.......................................................................................................192

PAGE 8

8 LIST OF TABLES Table page 2-1 Comparison of target particle size dist ribu tion with calculated and measured gradation utilizing 5 different Silica particle gradations.................................................................... 49 2-2 Experimental matrix for optim ization of the screened HS.................................................... 50 2-3 Clarification response or part icle separation (e xpressed as SSC removal efficien cy) of different configurations to the influent particle gradation................................................. 51 2 Indices for characterizing hydraulic respon se across flow rates representing 25%, 75% and 125% of Qd where Qd is the design flow rate............................................................ 52 3-1 Summary comparison of pilot-scale meas urem ents, CFD modeled results, and overflow rate (Q/A) model results..................................................................................................... 86 4-1 Summary of SSC results for RCF tested with a Sil-co-Sil 106 gradation at a nominal concentration of 200 m g/L. EBCT is the mean fluid empty bed contact time................ 113 5-1 Summary of measured and modeled part iculate m atter (PM) sepa ration by the screened HS for four discrete storm events.................................................................................... 141 5-2 Summary of measured and modeled part iculate m atter (PM) sepa ration by the screened HS for four discrete storm ev ents, using the measured EMC.......................................... 141

PAGE 9

9 LIST OF FIGURES Figure page 2-1 Experimental setup for testing the screened HS.. ..............................................................53 2-2 Geometry of the screened HS............................................................................................ 54 2-3 Plots of target influent particle size distribution. ............................................................... 55 2-4 Clarification response of different configurations across flow rates. ................................ 56 2-5 Particle Size Distributions of separated particles at Q = 0.25*Qd.....................................57 2-6 Particle Size Distributions of separated particles at Q = 0.75*Qd.....................................58 2-7 Particle Size Distributions of separated particles at Q = 1.25*Qd.....................................59 2-8 Tracer study at a flow rate Q = 0.25*Qd............................................................................60 2-9 Tracer study at a flow rate Q = 0.75*Qd............................................................................61 2-10 Tracer study at a flow rate Q = 1.25*Qd............................................................................62 3-1 Full scale experimental setup for testing a screened HS.................................................... 87 3-2 Observed phase shift in particle si ze distribution from infl uent to effluent...................... 88 3-3 Demonstration of grid independence................................................................................. 89 3-4 Measured vs. modeled Mparticles......................................................................................90 3-5 Comparison of measured versus m odeled results as a function of Q................................ 91 3-6 Particle trajectories calculated by a Lagrangian DPM for the screened HS ...................... 92 3-7 Modeled velocity distributions within the screened HS....................................................93 4-1 Process flow diagram for steady flow operation of the radial filter cartrid ge................. 114 4-2 Profile view of the radial cartridge filter (RCF) apparatus .............................................. 115 4-3 Measured media size expressed as a Gaussian frequency histogram ..............................116 4-4 Observed phase shift in particle si ze distribution from infl uent to effluent.................... 117 4-5 Comparison of measured versus modeled re sults as a function of influent flow rate .....118 4-6 Head loss ( H) as a function of influent flow rate (Q). ................................................... 119 4-7 Measured vs. modeled particle m ass in the effluent of the RCF.................................. 120

PAGE 10

10 4-8 Head loss ( H) and pressure distributions in the RCF. ................................................... 121 4-9 CFD predictions of trajectories of fluid and particles inside the RCF ............................. 122 5-1 Plan and side view of detail ed geometry of a screened HS. ............................................142 5-2 Experimental site for monitoring rainfall-runoff from an urban highway....................... 143 5-3 Influent hydrology for four discrete storm events........................................................... 144 5-4 Measured vs. modeled particle size dist ributions separated by the screened HS. ...........145 5-5 Temporal particle trajec tories calcu lated by a Lagrangian DPM for the screened HS.... 146 5-6 Modeled versus measured particle si ze distributions for four storm eventsn.................. 147 6-1 Plan view of experimental site and V olumetric Clarifying Fi lter (VCF) system............ 171 6-2 Location of the pressure transducers inst alled in the Volum etric Clarifying Filter......... 172 6-3 Measured media size expressed as a Gaussian frequency histogram ..............................173 6-4 Profile and plan views of a section of the com putational grid of the VCF system.......... 174 6-5 Hydrographs and hyetographs for 5 real-time rainfall runoff events.............................. 175 6-6 Measured vs. modeled cartridge head lo ss profiles as a functio n of nor malized time.... 176 6-7 Comparison of head loss as a f unctio n of surface loading rate (SLR)............................. 177 6-8 Head loss ( H) and pressure distributions in the VCF. ................................................... 178 6-9 Comparison of measured and modeled tem poral variation in effluent mass................... 179 6-10 Measured versus modeled effluent concentrations as function of storm elapsed time.... 180

PAGE 11

11 LIST OF ABBREVIATIONS A Effective surface area of the filter media (m2) 321,, aaa Empirical constants for smooth spherica l particles as a f unction of Reynolds number (AOCM)p Aluminum oxide coated media with a pumice substrate Ai Effective area of the inlet throat As Effective screen area BMP Best management practices 2C Inertial resistance factor, (m-1) 1C ,2C ,3C Empirical constants in the standard kmodel C(t) Tracer concentration at time t Ceff Effluent concentration (mg L-1) CFD Computational fluid dynamics Ci-IN Concentration of influent during time period i (L s-1) Cj-EFF Concentration of influent during time period j (L s-1) CSO Combined sewer overflow CV Control volume sd Diameter of the screen apertures (m) md Granular media particle diameter (m) pd Particle diameter (m) d15m Particle diameter at which 15% of particle gradation mass is finer d50m Median particle diameter based on mass d50m Particle diameter at which 50% of particle gradation mass is finer

PAGE 12

12 d85m Particle diameter at which 85% of particle gradation mass is finer Deff Duration of effluent (min) Dinf Duration of influent (min) dp Particle diameter DPM Discrete phase model Drain Duration of rainfall (min) E(t) Residence time distribution function EBCT Empty bed contact time (s) ECD Equivalent circular diameter EMC Event mean concentration ER Efficiency ratio (%) F(t) Cumulative Residence time distribution function kG Generation of k due to the mean velocity gradients bG Generation of kdue to buoyancy HS Screened hydrodynamic separator ID Inner diameter (cm) IDs Inner diameter of the screen area (m) IDv Inner diameter of the volute area (m) k Turbulent kinetic energy per unit mass, (m2s-2) mK Physico-chemical property of media )(' iLK Loss coefficient through pe rforated plate in the ith direction (m-1) )( iLK Loss factor through perforated plate in the ith direction (m-1) L Length of the packed bed (m)

PAGE 13

13 Lx Length in the x-direction (m) Ly Length in the y-direction (m) m Number of effluent measurements MDI Morrill Dispersion Index Me Mass of particles in the effluent Mi Mass of particles in the Influent iM Mass of particles in the Influent (g) RCFM Mass of particles captured by the RCF (g) HSM Mass of particles capture d by the Screened HS (g) IN Number of particles injected at the inlet HSN Number of particles that remain in the screened HS RCFN Number of particles that remain in the screened RCF n Number of influent measurements OD Outer diameter (cm) PM Particulate matter PR % removal (%) PSD Particle size distribution PVF Particle volume fraction (%) Q Influent volumetric flow rate (L s-1) QA/QC Quality assurance and Quality control Qd Design hydraulic operating flow rate (L s-1) Qeff-avg Average effluent flow rate (L s-1) Qeff-max Maximum effluent flow rate (L s-1)

PAGE 14

14 Qinf-avg Average influent flow rate (L s-1) Qinf-max Maximum influe nt flow rate (L s-1) Qmax Peak flow rate for the rainfall-runoff event (L s-1) Qmean Average flow rate for the rainfall-runoff event (L s-1) Qmedian Median flow rate for the rainfall-runoff event (L s-1) AQ / Surface loading rate (m s-1) pRe Particle Reynolds number mediaRe Porous media Reynolds number RCF Radial cartridge filter RTD Residence Time Distribution SLR Surface loading rate, (Lm-2m-1) SSC Suspended Sediment Concentration kS, S User-defined source terms in the standard kmodel S Source/sink term in the generali zed scalar conservation equation; iS Source term for the ith momentum equation t Time measured from time 0 t50 Time at which 50 % of trace r had passed through the reactor t90 Time at which 90 % of trace r had passed through the reactor tmean Mean hydraulic residence time from tracer measurements tp Time at which peak concentration is observed U Steady mean value of velocity, (m s-1) u Fluid velocity (m s-1) u Fluid velocity vector, (m s-1)

PAGE 15

15 )(tu Fluctuating component superimposed on it, (m s-1) wvu Vector components of velocity fl uctuations due to turbulence, (m s-1) pu Particle velocity (m s-1) UOP Unit operations and processes sv Discrete particle settling velocity (m s-1) rv Radial velocity component in the screening area (m s-1) tv Tangential velocity component in the screening area (m s-1) magv Velocity magnitude in the cell, (m s-1) 2 xv Square of the velocity through the plate, considering that th e plate is x % open to flow in the given direction, (m2s-2) 2v Square of the velocity through the plate, considering that the plate is 100 % open to flow in the given direction, (m2s-2) sv Superficial velocity th rough the porous media (m s-1) V Volume of the unit VCF Volumetric clarifying filter Vi-IN Volume of influent during time period i (L s-1) Vj-EFF Volume of effluent during time period j (L s-1) vr Radial velocity component in the screen area vs Discrete particle settling velocity vt Tangential velocity com ponent in the screen area Hydraulic residence time (s) Rate of dissipation of turbulent kinetic energy per unit mass (m2s-3)

PAGE 16

16 iv Velocity in the ith momentum equation (m s-1) k Turbulent Prandtl numbers for kand respectively Diffusion coefficient (m2s-1) Fluid density (kg m-3) Fluid property per unit mass Hydraulic residence time (s) p Particle density (kg m-3) Permeability (m2) Porosity (%) p Pressure loss through the plate (Pa) Viscosity (kg m-1s-1) particlesM Particle separation efficiency (%) t Eddy viscosity (kg m-1s-1) 2 Removal efficiency 2 Variance of the RTD function 2/ 2 Peclet number equivalent

PAGE 17

17 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy EXPERIMENTAL AND NUMERICAL ANALYSIS OF STORMWATER UNIT OPERATIONS AND PROCESSES By Subbu-Srikanth Pathapati May 2008 Chair: John J. Sansalone Major: Environmental Engineering Sciences Anthropogenic particulate matter (PM) trans ported in urban rainfall-runoff has been identified as a significant contributor to ove rall deterioration of su rface water in the USA (USEPA 2000). Rainfall-runoff transports an en trained mixture of colloidal PM, non-colloidal PM, dissolved and complexed pollutants (Sansal one et al. 2007, Lee and Bang 2000, Sansalone et al.. 1998, Igloria et al.. 1997, Sansalone and Buchberger 1997) PM transported by runoff act as reactive surfaces for adsorbing, desorbing and leaching organics and metals as well as phosphorus and other nutrients (Sansalone 2002). Separation of PM by unit operations and processes (UOPs) for in-situ treatment of rainfall-runoff is challenged by factors such as the stochastic nature of hydrologic a nd pollutant loads and concerns su ch as availability of land and infrastructural resource s (Liu et al.. 2001). This study aimed at understanding the hydr odynamic and clarification response of innovative UOPs for urban stor mwater management hydrodynamic separators and volumetric filters by means of a coupled experimental and m odeling approach. A screened HS is a type of HS that utilizes a combination of inertial separa tion, discrete particle (Type I) settling, and size exclusion by means of a static scr een to effect particle separation. A passive radial cartridge filter

PAGE 18

18 (RCF) utilizing aluminum-oxide coated media with a pumice substrate (AOCM)p was also tested. All tests were conducted for a known influent particle size distribution and concentration across the entire range of operating flow rates. Computational fluid dynamics (C FD) utilizes numerical methods to solve the fundamental equations of fluid dynamics, i.e. the Navier-Sto kes equations (Versteeg et al.. 1995). This study applied the principles of CFD to predict the part icle separation behavior of a screened HS for dilute multiphase flows, to a scientifically acc eptable degree of accuracy. PM separation of the HS and RCF were modeled for both steady and tr ansient influent flow rates and particulate loading. The Reynolds averaged Navier-Stoke s equations were clos ed by applying a two equation turbulence model, and a Lagrangian di screte phase model was used to examine the particle clarification response. The absolute relative % difference (RPD) between measured and modeled data was less than 10 %. The CFD mode l was stable and accurate, demonstrating grid independence. Post-processing the CFD predicti ons provided an in-depth insight into the mechanistic behavior of the screened HS a nd RCF by means of three dimensional hydraulic profiles, particle trajectories and pressure distributions. A CFD approach is the next step in understanding and designing UOPs for urban stor mwater, and has far-reaching results compared to traditional bl ack-box approaches.

PAGE 19

19 CHAPTER 1 GLOBAL INTRODUCTION Anthropogenic particulate m atter (PM) trans ported in urban rainfall-runoff has been identified as a significant contributor to ove rall deterioration of su rface water in the USA (USEPA 2000). Rainfall-runoff transports an en trained mixture of colloidal PM, non-colloidal PM, dissolved and complexed pollutants (Sansal one et al. 2007, Lee and Bang 2000, Sansalone et al.. 1998, Igloria et al.. 1997, Sansalone and Buchberger 1997) The temporal particle size distribution (PSD) and chemical composition of pa rticulate matter (PM) delivered by runoff from a rainfall-runoff event varies significantly betw een different geographical regions and even with spatial variation in the same watershed. (Sansalone 2002). A major issue that needs to be addressed while designing stormwater unit operations and processes (UOPs) is the dearth of available land for construction, especially in highly polluted urban areas. In light of this, many new devices have been introduced, which have the advantage of a small-footprint and ease of new installation or retrofit to existing infrastructure. This dissertation is a coupled experime ntal and numerical approach to characterizing the particulate matter separation by various UOPs, for screened hydrodynamic separators and small scale as well as large scale granular medium volumetric filtration systems, for controlled and real-time influent hydrology and granulometry. Hydrodynamic separators (HS) are one such cl ass of UOPs that broadly rely on centrifugal forces in addition to gravitational force to separa te particles (Brombach et al.. 1987, Brombach et al.. 1993, Pisano et al.. 1994, USEPA 1999, Andoh et al .. 2003). An attractive feature of a HS is the potential for longer particle trajectories per given unit surface area in comparison with traditional unit operations and processes (UOPs) such as settling basins. A screened HS is a variant on the HS principle, and uses the comb ined separation mechanis ms of vortex induced

PAGE 20

20 inertial separation, screening and se dimentation and has been used in recent years as a device for treating stormwater runoff (Rushton 2004, Rush ton 2006), and oil/grease removal (Stenstrom and Lau 1998). This study examined a simple scr eened HS consisting of two annular cylindrical chambers separated by a static screen consisting of a regular array of apertures. The flow inlet is tangential to the inner cylindrical chamber. The screen apertures allow vortex flow to exit the inner screen chamber and enter the outer volute ch amber. The geometry of the screen results in a weakly reversed flow direction in the outer annu lar area, termed the vol ute area. Particle separation by this HS configuration is a function of the discrete pa rticle settling velocity, particle diameter, influent volumetric flow rate, radi al velocity components in the screening area, tangential velocity components in the screening ar ea, hydraulic residence time, and the diameter of the screen apertures. The design flow rate (Qd) for the screened HS used in this study is 9.5 L/s. Various filtration configurations including infiltration and exfiltration through fixed granular medium (including soil) systems have b een suggested as viable solutions for meeting runoff quantity and quality regulations for watershe ds (Colandini 1999, Sansalone 1999, Li et al.. 1999, Colandini et al.. 1995, Geldof et al.. 1994, Schueler 1987). Recently, Hipp et al. (2006) suggested that removable filter in serts, which are mechanistically different from typical fixed bed granular medium filters, may be viable preliminary unit operations due to easier maintenance. While granular filtration has demonstrated advantag es for improving water quality, there is a requirement for careful design, analysis and prototype testing, as well as regular maintenance to ensure optimal hydraulics for qua ntity control. Performance (mass and size of PM separated) of a granular medium filter depends on the influent surface loading rate, total surface area of the media the influent volumetric flow rate, effective poros ity, particle diameter,

PAGE 21

21 granular media diameter, empty bed contact time (EBCT) (AWWA 1990) and physical, chemical and surficial properties of the media, such as adsorption properties, media (internal) porosity, and morphology. (Tien 1989). Oxide coated media ha s been found to be effective at reducing turbidity, phosphorus and microbes in water and wastewater treatment (Ayoub et al. 2001, Chen et al. 1998, Ahammed et al. 1996). Recen tly, the use of oxide coated media has been extended to runoff treatment (Eri ckson et al. 2007, Sansalone et al. 2004). In this study, aluminum oxide coated media with a pumice substrate (AOCM)p was utilized for physical filtration. Granular filtration mechanisms have been classified into surficial straining, sedimentation, interception, inertial impaction, diffusion, hydrodynamic and electrostatic interaction (Wakeman et al.. 2005). Filtration dyn amics are widely assumed to be dependent on the surface loading rate (SLR) and with increasing SLR, macroscopic mechanisms tend to predominate the separation process, due to redu ced contact times. The radial flow rapid-rate filter used in this study was operated at surf ace loading rates (SLR) ranging from 24 to 189 L/m2-min, which is higher than SLRs of typical rapid sa nd filters (83 L/m2-min) (Reynolds et al. 1995). Typical stormwater design approaches are exte nsions of wastewater tank design principles, which assume ideal to predictable influent quantitative and qualitative loads. It is clear that the rainfall-runoff process and the associated par ticulate and dissolved matter delivery is a highly variable process and therefore unit operations for stormwater PM management have to be designed to operate across rapi dly changing flow and partic le concentrations. Existing stormwater models such as the stormwater management model (SWMM) (Rossman 2007) use idealized influent hydrographs and pollutogr aphs. While these have greatly improved the efficacy of design considering the complex and inter-related design parameters, they are not

PAGE 22

22 equipped to provide an in-depth hydrodynamic and particle clarification profile of the UOP for transient loads. Existing literature does not pr ovide a detailed and fundamental insight into the functioning of UOPs, including hydrodynamic separators. Prev ious studies have ranged from simplified overflow rate theory (Weib 1997) to semi-empiri cal approaches based on broad assumptions of vortex flow behavior (Paul et al.. 1991). This lack of information leads to difficulties in scaling and implementation of these devices. Fenner et al .. (1997, 1998) report that similitude analysis does not yield a single dimensionless group th at can be used in scale-up of a HS. Hydraulic optimization has long been in pr actice in engineering, for instance, in wastewater treatment, particular ly in the design and analysis of grit chambers, primary and secondary clarifiers. For instance, weirs and baffles have been used as energy dissipating devices and to optimize flow paths to minimize short-ci rcuiting and hydraulic in stability (Tchobanoglous et. al, 2003) Improvement of sw irl concentrators/vortex chambers has been an ongoing process. Researchers have utilized baffles, bars a nd other appendages to optimize solid separation (Alquier et al. 1982) by minimi zing eddies, short-circuiting and dead spaces. Furthermore, fundamental design methodologies have been sugg ested to optimize vortex separation, based on the characteristics of the vor tex flow (Paul et al. 1991). Computational fluid dynamics utilizes numer ical methods to solve the fundamental equations of fluid dynamics, i.e. the Navier-Stokes equations (Versteeg et al.. 1995). The applicability and efficacy of CFD techniques have closely followed corresponding leaps in computational power. While traditionally be ing in the realm of Aerospace and Process engineering, CFD is now used in design and optimization of UOPs in environmental engineering. Computational fluid dynamics (C FD) has found applications in environmental

PAGE 23

23 engineering in recent years, specifically in water and wastewater treat ment (Do-Quang et al. 1998). CFD approaches to model particle-laden flows is an active research area (Curtis et al.. 2004, van Wachem et al.. 2003), including hydr odynamic separators. Faram et al.. (2003) utilized a Reynolds Stress Model (RSM) to model turbulent flow and a Lagrangian Discrete Phase Model (DPM) to model behavi or of the particulate phase. Okamoto et al.. (2002) studied particle separation by a vo rtex separator using a kmodel to model flow and an Algebraic Slip Mixture (ASM) model for the particulate phase. Li m et al.. (2002) used CFD-predicted velocity profiles to study behavior of flocs in a vortex separator, with a Renor malization Group (RNG) kturbulence model. Tyack et al.. (1999) comp ared measured velocity profiles in a vortex separator to those predicted by a Renormalization Group (RNG) kturbulence model. Tung et al. (2004) studied deep bed filtrati on for a sub-micron/nano-particle suspension utilizing a microscopic approach wherein diffe rent types of media packing were modeled. However, macroscopic approaches to modeling filter media are needed to model practical systems where packing schemes are most commonl y random and unstructured. Li et al. (1999) applied a 2-D numerical model to simulate vari ably saturated flow in a partial exfiltration system. Sansalone et al.. (2005) applied a 2-D numerical model to simulate the transient hydrodynamics of a partial exfiltration system fo r rainfall-runoff clarification. However, a macroscopic CFD based approach to simulate the 3-D hydrodynamic and clarification response of a granular medium filter is mu ch needed in order to understand passive radial filtration, which may not be fully described by a simplified 2-D ap proach, due to lack of ideal symmetric flow conditions.

PAGE 24

24 It should be noted that albeit studies dealing with the applic ation of CFD in understanding the multiphase dynamics that occur in a hydrodynamic separator are few in number, this is clearly not the case with wastewater unit operations such as sett ling tanks (Jayanti et al. 2004, Naser et al. 2005, Deininger et al 1998, Brouckaert et al. 1999, Brestscher et al. 1992, Zhou et al. 1994, Szalai et al. 1994, Ka rl et al., 1999) and for CSO pollu tion abatement devices such as storage chambers (Stovin et al., 1996, 1998, 2002). In order that an engineer be aided in designing UOPs with complex hydraulics and particle transport, it is imperative that there be a coupled experiment al and numerical modeling method that can: Predict the particle separation behavior of the UOP through the application of an appropriate flow and particulat e phase model, within reasonable computational limits, to a scientifically acceptable degree of accuracy. A targeted experimental matrix that adheres to effective and practical QA/QC constraints, to experimentally characterize the UOP in or der to effectively verify and validate the numerical model and to provide any da ta required for calibrating the model. Apply to steady, as well as temporally transi ent quantitative and qualitative hydraulic and pollutant loads.

PAGE 25

25 CHAPTER 2 EXPERIMENTAL MODIFICATION OF A ST ORMWATER HYDRODYNAMIC SEPARATOR FOR ENHANCED PARTICLE SEPARATION Introduction Non-point source pollution has emerged as a sign ificant issue of concern in recent years. Various studies have shown that urban stormwater runoff is one of the major reasons for the deterioration of water bodies in the USA (USE PA 2000). The physical and chemical nature of stormwater pollutants has been studied, and it has been reported that a significant fraction of pollution from runoff can be attributed to pa rticulate matter (Sansal one 2002). Control of particulate pollutants is an esp ecially challenging task in an ur ban environment, due to complex watershed characteristics and hydrology. In add ition, land-availability and infrastructure constraints have only compounded the problem and have led to research and development of innovative technologies for in-sit u and ex-situ control of polluti on, such as hydrodynamic vortex separators (Brombach et al. 1987, Brombach et al. 1993, Pisano et al. 1994, EPA 1999, Andoh et al. 2003). A unique feature of hydrodynamic separators is th at the particle trajectory in such systems is many degrees of magnitude longe r than in traditional Unit Opera tions (UOPs) such as settling basins (Field et al. 1996).The probability of pa rticles being separated within a given volume is thus, degrees of magnitude greater. It should be noted that the st udies mentioned previously have dealt with performance evaluation of hydrodynamic separators based only on experimental data. One type of HS utilizes a combination of UOPs, namely, inertial separation due to vortex action, size exclusion by screeni ng, and discrete particle (Type I) settling. This device, referred to as a screened HS or simply HS for this documen t, has been used in recent years as a device for treating stormwater runo ff CSO outfalls (Heist et al.. 2003), and oil & grease removal (Stenstrom and Lau 1998).

PAGE 26

26 Analysis of storage chambers and hydrodynami c separators for CSO treatment has been carried out by both experimental and numerical approaches (Ty ack et al. 1999, Faram et al. 2002, Okamoto et al. 2003). Hydraulic optimization has long been in practice in engineering, for instance, in wastewater treatme nt, particularly in the design and analysis of grit chambers, primary and secondary clarifiers For instance, weirs and baffles have been used as energy dissipating devices and to optimize flow path s to minimize short-circuiting and hydraulic instability (Tchobanoglous et. al, 2003) Improveme nt of swirl concentrators/vortex chambers has been an ongoing process. Researchers have util ized baffles, bars and other appendages to optimize solid separation (Alqui er et al. 1982) by minimizing eddies, short-circuiting and dead spaces. Furthermore, fundamental design methodologi es have been suggested to optimize vortex separation, based on the characteristics of the vortex flow (Paul et al. 1991). Studies agree that the predominant variable s affecting the performance of a HS are the settling velocity characteristics of the influent pa rticles and the flow rates at which the device is loaded, and that improved solid separation can be effected by lengthened part icle trajectories or increased particle residence time (Field et al.. 19 96). This study aims at optimizing the particle removal performance of a screened HS by means of geometric modifications. These modifications were based on hypothe ses built upon increasing particle residence time in the unit, reduction in flow instability and short-circuiting. The study was conducted for an influent particle gradation typically issu ed by regulatory bodies. A concen tration of 200 mg/L was chosen and the tests were conducted for 8 different conf igurations, across three flow rates representing 25, 75 and 125 %, respectively, of the maximum hydr aulic operating flow ra te or design flow rate (Qd) of the unit under test. The tests were conducted under steady-state flow regimes, to better distinguish the effect s of different configurations on solid separation.

PAGE 27

27 Background A schematic of the screened HS is provide d in Figure 2. The device consists of two concentric cylindrical chambers, the inner scr een area, and the outer annular volute area. The inlet is tangential to the inner cylindrical cham ber. The inner cylindrical chamber is equipped with a static separator screen, with perforations elongated in shape and are aligned with the longer axis in the vertical direct ion. Furthermore, the shape of th e apertures, resu lts in reversed flow direction in the outer volut e chamber. The screen area open to flow is designed in a manner that would cause the radial veloc ity through the screen to be of an order of magnitude lower than the inlet velocity. The resultant higher ratios of tangential to radial velocities prevent particles from exiting through the apertures (Wong, 1997), and allow them to settle to the sump, via circular helical paths of progressively decreasing radii. This mechanism is at its most favorable level when the energy of the driving vortex is optim al and is subsequently seen to be higher at flow rates closer to the design flow rate (Qd). The flow path through the screened HS is continuous and steady. A unique characteristic of a screened HS is th at it does not rely on secondary currents due to vortex action, as with traditional vortex se parators. The action of secondary currents is countered by providing a filtr ation mechanism in the form of a perforated screen. This has been suggested to provide higher particle separati on rates as opposed to traditional vortex separators (Schwarz and Wells 1999). Overall, the performance of a screened HS can be summarized as being dependent on th e following variables, as shown in the following expression. ),,,,,,(strps particlesdvvQdvfM (2-1) In this expression,particlesM is the particle sepa ration efficiency (%);sv is the discrete particle settling velocity [LT-1];pdis the particle diameter[L];Q is the influent volumetric flow

PAGE 28

28 rate [L3T-1];rv is the radial velocity com ponent in the screen area [LT-1];tv is the tangential velocity component in the screen area [LT-1]; is the hydraulic residence time [T]; and sd is the diameter of the screen apertures [L]. The ratio of the radial to tangential velocities in the screen area was expressed as follows. i s t rA A v v (or) i s t rA A k v v (2-2) In this expression,sA is the effective screen area [L2];iA is the effective area of the inlet throat [L2]; and k is a dimensionless coefficient proportional toQ. Previous studies in particle be havior in vortices (Julien et al ..,1986) have suggested that in steady vortices, there exists a equal diffusive fl ux acting opposite to the centrifugal flux pushing particles toward the outside of the vortex, for both free and forced vortices, and at equilibrium, these two fluxes are equal. Other studies (Davil a et al..2000) have sugge sted the dependence of particle motion in a vortex on the particle Froude number, and terminal settling velocity. Studies in modeling the performance of hydrodynamic separa tors (Fenner et al. 1 997, Fenner et al. 1998) have suggested that no single di mensionless group can be used in describing, and scaling the solid separation performance. Other such studi es used a simplified overflow rate theory approach(Weib, 1997).For this st udy, the overall performance of th e screened HS was evaluated on the basis of equations 1 and 2, keeping sdconstant at 2400 m, and with pre-determined values of pd and sv. Objectives The first objective of this study was experime ntal characterization of the clarification response of a prototype sc reened HS to an influent particle gradation, at a concentration of 200 mg/L, commonly observed in urban runoff, as a f unction of geometric modifications that affect

PAGE 29

29 performance indicating variables described in the previous sec tion. A second objective was to study the effect of varying th e dimensionless ratio of isAA/ on solid separation performance, and to develop preliminary design crit eria for creating optimal ratios oftrVV/.The third and major objective of this study was to suggest a geometric modification that would optimize the overall solid separation performance of the scr eened HS, across flow ra tes representing 25, 75 and 125 % respectively, of the maximu m hydraulic operating flow rate (Qd),with the unit filled to its residual treatment capacity, and with an allowable error rate of + 10% by mass. The fourth objective was to characterize va riations in hydrodynamic respons e of the modified geometric configuration in comparison with a baseline unit, from Residence Ti me Distribution (RTD) curves obtained by a tracer study. Methodology Experimental Site Setup Figure 3-1 contains a schematic plan view of the instrumented experimental setup. A storage tank (Storage tank A) w ith an approximate capacity of 4000 L was used as a reservoir for influent potable water. Three sludge pumps locat ed in Storage Tank A, with capacities of 6.94 L/s each were used to generate influent flow The sludge pumps were used in appropriate combinations to generate flow across the entire treatment range with an error of less than + 2%. Flow rates were measured in a 5.08 cm (2-inc h) Parshall flume located downstream of the storage tank. Depth of flow in the Parshall flum e was measured with a 70 KHz ultrasonic sensor manufactured by American Sigma Inc, and data was recorded real-time by an American Sigma data logger. The Parshall flume was re-calibrated volumetrically over the entire flow rate range, with triplicate measurements, at an error less than or equal to + 1%, and this re-calibrated relationship between depth of water in the flum e and the corresponding volumetric flow rate was

PAGE 30

30 used in place of the standard relationship for a 5.08 cm (2-inch) Parshall flume. This was done to account for any errors that might arise due to the comp licated upstream plumbi ng at this specific experimental site. Influent partic les and Sodium Chloride (NaCl) tr acer were injected at the dropbox located immediately downstream of the 5.08 cm (2-inch) Parshall flume and immediately upstream of the screened HS. Flow was routed to the screened HS utilizing a system of gate valves located immediately downstream of the drop-box and immediately upstream of the screened HS. The overall diameter of the screened HS was 89.3 cm and the diameter of the inner screen area was 49.5 cm. The height of the unit was 96. 5 cm. The residual stor age volume of the unit was approximately 408 L, based on the depth of liquid in the unit under quiescent conditions, and the maximum hydraulic operating volumetric flow rate(Qd) was 9.4 L/s. The static separator screen consisted of apertures with a sd of 2400 m. A storage tank (S torage tank B) with an approximate capacity of 4000 L was used as a storage reservoir for effluent from the Screened HS, and was located immediately downstream of the unit. Influent Particle Gradation The clarification response of a screened HS has been iden tified to be dependent on the characteristics of the influent pa rticle gradation (Kim et al., 2 005). Particle gradations commonly seen in urban runoff vary significantly base d on the demography, geography, and land-use characteristics of a watershed. The influent partic le gradation that was used in this study is one that is typically seen in urban runoff. The particle size distribution of th e influent solids by mass is shown in Figure 3. It is noted that approximately over 50 % by ma ss of this gradation, is finer than 75 m. By a widely accepted standard of cla ssification, over 50 % by mass of this gradation can be said to consist of se ttleable-suspended particles.

PAGE 31

31 Five different test gradations of silica particles were obtained from the US Silica Company and each of the 5 test gradations were first sieved to separate out particles too coarse and too fine between specific sieve increments. These sieved gr adations were combined to meet the specified test gradation. Once combined and mixed, the five sieved final particle components of the final gradation were sieved again down to 25 m, and the fraction finer than 25 m was analyzed using laser diffraction. Measured results were then compared to the target gradation and to the calculated gradation. Results of the target grad ation, the calculated gradation and the measured gradation are depicted in Figure 3 and Table 1. The mass of particle mixture to be added was chosen so as to achieve a 200 mg/L influent concentration, based on 4000-L of influent. QA/QC checks were periodically made to ensure cons istency of the influent particle gradation. Field Test Procedure The following describes the field test procedure. The storage tanks, the recirculation pipes and the screened HS were cleaned thoroughly by pumping potable water through the system. This was done to ensure that any previously depo sited particles or any ex traneous particles were removed from the system prior to starting the test. Storage tank A was then filled with about 4000 L of water, and the screened HS was filled to its residual storage cap acity of approximately 408 L. The valves on the recirculation system we re calibrated with tap water and set to the desired flow rate (25, 75, and 125 % of Qd). The desired test flow rates were achieved through the use of either one or two 110-gpm recirculatio n pumps. Calibrated flow measurements were made throughout each run. The moment of time at which flow was diverted to the screened HS, was noted as Time 0. Addition of influent particles at the dr op-box was started immediately at Time 0. The run time for each test was calcu lated based on available influent volume of approximately 4000 L, and flow rate tested. Flow was diverted to the screened HS and influent particle addition into the dropbox upstream of the Screened HS commenced immediately at

PAGE 32

32 time 0.The rate of particle injection was consistent ac ross the test running time, on a gravimetric and granulometric basi s. Ten replicate 2-L effluent samples were collected for water quality and SSC analyses, at the outfall of the effluent pipe into storage tank B. Separately, 100 gallons of effluent (10 of 5-gallon with repli cate) were captured and recovered in 55-gallon polypropylene tanks, at the same sampling frequenc y as the 2-L samples, in order to recover sufficient particle mass to carry ou t effluent particle gradations. Laboratory Analyses Two 20-L composite effluent samples (A a nd B) were prepared from 10 individual replicate (A and B) 2-L samples. The entire measured volume of each replicate composite sample was then filtered using a 1.2m glass fiber filter. Total susp ended solids (TSS) were thus measured as suspended sediment concentration (SSC). Particles captured by the screened HS from each run were completely recovered as wet sl urry from the unit. Effluent particles captured in two 55-gallon polypropylene tanks were also re covered as wet slurry. After each run, a known amount of coagulant ferric chloride was adde d to the screened HS and the two 55-gallon polypropylene tanks, in order to acc elerate the settling process of particles. At least 5 hours of settling time was allowed before slurry recovery. The recovered particle slurry was then transpor ted to the laboratory and dried at 40 C. The dried particle mass was weighed and the SSC rem oval efficiency was calculated as effluent mass was computed from the measured effluent SSC corresponding to total vol ume. A mass balance check was performed on the SSC removal efficiency as, Mass Balance Error = %10 )( 100 i e HS iM MMM (2-3)

PAGE 33

33 In this expressioniM is the mass of particles in the Influent [M]; HSMis the mass of particles captured by the screened HS [M]; and eM is the mass of particles in the effluent [M]. If the mass balance error criterion was satisfie d, the SSC removal efficiency was calculated as, i HSM M 100 (2-4) The dried particle mass was then sieved us ing graded sieves to obtain particle size distributions according to ASTM D422 for siev e analysis (ASTM 1993). Sieve numbers were chosen appropriate to the influent gradation. The sieve numbers used were #30, #50, #100, #140, #200, #270, #400, #500 and the pan. Mass balances we re determined for each mechanical sieve analysis and mass balance errors were in the range of 1 to 2 % by mass. Geometric Configurations The screened HS is a variation on the hydrodynamic separation principle, and its predominantly distinguishing feature is that this UOP, unlike trad itional vortex separators, does not rely on secondary currents created by a vortex, to effect particle separation. It ensues that the rotary flow patterns that occur in the screened HS are different. The presence of a static screen, and the absence of an underflow, creates a stead y and continuous flow path through the device. The device functions closest to optimal operati on under steady flow conditions. Therefore, a main aspect of the modifications in the inner separation/screening chamber was to not impede the steady and continuous flow through the device by providing baffles as attenuating devices. Modifications involving the area of the inlet, and the effectiv e screen area, did not aim at impeding flow. These modifications were aimed at increased ratios of tangential to radial velocities, in terms of magnitude alone, and did not interfere w ith directional aspects of flow

PAGE 34

34 within the device. Modifications in the outer volu te chamber were aimed at lengthening particle trajectory out of the system. With these cons traints in mind, eight different geometric modifications were made to the existing unit. The first modification involved fixing a 5 cm (2-inch) effluent weir baffle to the unit. The effluent weir baffle is a sheet metal baffle placed at the outlet of the device. The principal goal of this was to reduce possible short-circuiting and to lengthen the path of travel of particles out of the unit. This does not al ter the volume of the unit. The e ffluent baffle was hypothesized to improve solid separation performance mainly at higher flow rates. The second modification was placing a sump baffle. The sump baffle is a baffle placed on top of the sump portion of the inner screen area. The hypothesis tested here wa s that this would reduce probable re-suspension of deposited particles by scouring. The third modification was a 5 cm (2-inch) submergence baffle. This was fixed at the outlet, to increase the height of water inside the unit, thereby increas ing the volume of water inside the unit by a small amount. The dimensionle ss ratio of screen area to inlet throat area, isAA/ was tested as a viable de sign parameter. By modifying iA by friction-fitting hard Styrofoam wedges, the inlet flow velocity wa s increased, by creating a nozzle-like effect. The hypothesis tested here was that th is would decrease the number of particles exiting the screen via the apertures, due the increased tang ential velocities. The screen area sA was changed by blocking the screen with high-streng th adhesive non-permeable tape. Various combinations of the above modificatio ns were tested. Overall, eight different configurations from A to H were tested. Configuration A in cluded only the effluent weir baffle. Configuration B incl uded the effluent weir baffl e and the sump baffle. Configuration C included the effluent weir ba ffle, sump baffle and submergence baffle.

PAGE 35

35 Configuration D included the effluent weir baffle, submergence baffle and screen area reduced by 37 % from the top. Configuration E in cluded the effluent we ir baffle, screen area reduced by 37 % from the top, and throat area reduced by 17 %. Configur ation F included the effluent weir baffle and the submergence baff le. Configuration G included the effluent weir baffle, and throat area reduced by 26 %. Finally, Configuration H included a effluent weir baffle, screen area reduced by 18 %, a nd throat area reduced by 17 %. The choice of configurations was an iterative process, based on the results of the prev ious configuration. A detailed description of the experimental matrix and dimensions of modifications is provided in Table 1 and the accompanying description. Tracer Analysis A tracer study was conducted to determine hydr aulic residence time distributions (RTD). Concentrated Sodium Chloride (NaCl) solution was used as a tracer. A known volume of tracer was injected as a single pulse into the drop box upstream of the screened HS. The methodology for tracer injection followed the exact same steps as for influent particle injection, described in the previous section, with the tracer slug inject ion taking the place of the influent particles. A calibration curve was developed to establish the relationship between concentrations and conductivity. The concentrations were thus calcul ated from conductivity measurements. Prior to tracer injection, the conductivity of the potab le water used for testing was measured. For continuous real-time measurement of conductivity at the effluent sampling point, a calibrated conductivity probe, manufactured by YSI Inc., wa s used. The conductivity probe was placed so it was fully submerged at the outfall of the effluent pipe. The tracer study was run at a constant flow rate, until the conductivity dropped back to the background (potable water) conductivity. The screened HS was thoroughly cleaned with pota ble water, after every run. Each tracer run was run in triplicate and validated by a mass balance check, with th e allowable error rate set at +

PAGE 36

36 5% by mass. Discrete measurements were taken every one second, and the mean residence time was approximated as (Tchobanoglous et. al, 2003), 0 0)( )( dttC dtttC tmean (2-5) In the expression, meantis the mean detention time from tracer measurements [T]; t is time measured from time 0 [T] ;)(tC is tracer concentration at time t [LT-1].The theoretical residence time was calculated as Q V (2-6) In the expression, V is the volume of the unit [ L3]; and Q is the Influent flow rate [L3T-1]. The residence time distribution f unction E(t) was calculated as 0)( )( )( dttC tC tE (2-7) And consequently the cumulative RTD function F (t) was calculated as tdttEtF0)()( (2-8) Previous studies on Residence Time Di stributions in hydrodynamic separators (Alkhaddar et al. 2001) suggest that the hydrodynamic separator behaves as a incompletely mixed plug flow reactor, and different models, su ch as the Axial Dispersion Model (ADM), and Tanks-in-series Model (TISM) were used to describe the data. In this study, the aim was to compare the RTD of the optimized configuration re lative to the baseline c onfiguration. Therefore the following parameters were used (Levenspiel, 1972; Letterman, 1999),

PAGE 37

37 Morrill Dispersion Index (MDI) = 1090/tt (2-9) Volumetric Efficiency = MDI /1 (2-10) Index of short-circuiting = /it (2-11) Index of modal retention time = /pt (2-12) Index of average retention time = /50t (2-13) Hazens N = (t50 ) / (t50 tp) (2-14) In the above expressions, itis the time at which tracer first appears;ptis the time at which peak concentration is observed;50tis the time at which 50 % of tracer had passed through the reactor; and 90tis the time at which 90 % of tr acer had passed through the reactor. The Morrill Dispersion Index is cl ose to 1 for a plug-flow reac tor. Higher values indicate deviations from plug-flow towards complete-mixi ng, and result in lower volumetric efficiency. The indices of short-circuiting and modal retention time vary fr om 0 to 1 from complete-mixing to plug flow. The index of averag e retention time provides a measure of the skew of the curve. Values greater than 1 indicate skew to the right and values less than 1 in dicate skew to the left. The variance (2) was calculated from equation (4) as, 0 0 2 2)( )()( dttC dttCttmean (2-15) A Peclet number equivalent was adopted (Alq uier et al. 1982), and defined as equal to 22/. This was used to distinguish between th e modified and baseline configurations. A higher value of the Peclet number equivalent would indicate improve d plug-flow behavior.

PAGE 38

38 Results and Discussions Treatment Influent Flow Rate at 25 % of Design Flow Rate Figure 4 depicts particle separa tion performance removal at 25% of design flow rate across configurations. It was no ticed that, at this flow rate, the sc reened HS functions primarily as a circular tank, and the primary mechanism of soli d separation is gravitatio nal sedimentation. The improvement in performance over a baseline unit at this low flow rate was seen to be minimal, and not very significant. This response was consistent across configurations. The %age of separated solid captured by the screen area and by the volut e area did not vary significantly across all configurations excep t Configuration B and Configur ation C, which exhibited a noticeable tendency for increased ca ptured of solids in the Volute area as opposed to the screen area. This can be attributed to the reason that the Sump Baffle, was a geometric modification common to these two configurations. The baffle was intended to prevent re-suspension of settled particles, by reduci ng the area of the sump exposed to flow. However, at low flow rates, such as 25% of design flow rate, the benefits of reduced scouring are not significant At this flow rate, particles tend to have more vertical trajectories as opposed to vortex-driven circular trajectories seen at higher flow rates. This results in a fair ly uniform deposition of pa rticles over the entire surface area of the sump. In this case, this resulted in reduced distance from the screen apertures to the particles deposited on the baffle, as opposed to particles deposit ed in the sump, and subsequently these particles were re-suspended and pushed out of the screen, where they were captured in the volute area. This is supported by the part icle size distributions by mass of recovered so lid particles from the screen area, volute area, and from the samp led effluent, as seen in Figure 5. It is seen that, as this flow rate, the particle size distri butions by mass across configurations exhibit similar trends. Any deviations in parameters such as d15m, d50m, and d85m are minor and do not

PAGE 39

39 significantly alter the overall SSC removal efficiency. It would be useful to note that, at this flow rate, the solids captured by the Screened HS tend to be closer to the Influent particle size distribution as compared to par ticle size distributions at higher flow rates (Figure 6 and Figure 7), i.e., the d15m, d50m, and d85m are noticeably finer at 25% of de sign flow rate as compared to 75% an 125% of design flow rate The overall SSC removal efficiency ranged from 48 to 52.7%, across configurations. Detailed results are listed in Table 2. Treatment Influent Flow Rate at 75 % of Design Flow Rate Figure 4 depicts particle se paration performance at 75% of design flow rate across configurations. The noticeable difference is that the difference in functionality of the screened HS starts to surface at this flow rate as compared to lower flow rates. At 75% of design flow rate, the energy of the driving influent flow is sufficient to create a vorte x in the screened HS. However, the significance of vortex flow as a performance affecting variable can be quantified as being higher than at lower fl ow rates, and lower than at highe r flow rates. A lthough there is sufficient energy to create and su stain a vortex, the distribution of associated surf ace velocities, tangential and radial velocities is of a magnitude lower than at higher flow rates. That is, the static screening separation mechanism that is de pendent on higher ratios of tangential to radial velocities is not pronounced or fully utilized. With particle separa tion being directly influenced by velocity distributions within the unit, the ove rall solid separation mechanism at an influent flow rate of 75% of design flow rate can be said to be a combination of screening and sedimentation, with a greater emphasis on grav itational sedimentation mechanisms than on mechanisms of hydrodynamic separation. There is improvement over a baseline setting by about 6 %, for Configuration A. Other configurations do not differ significantly from one anot her. Figure 6 depicts the particle size distributions by mass across configurations for a flow rate of 75 % of design flow

PAGE 40

40 rate. The most noticeable trend is that in co mparison with the lower flow rates discussed previously, the particle size distribution tends to capture a lesser fraction of the influent gradation, and it tends to capture a coarser portio n of the influent gradation. This is supported by the visibly coarser values taken by d15m, d50m, and d85m. Configuration A performed best, capturing about 45 % of influent gradation, in th e screened HS. Conseque ntly, the particle size distribution curve for Configurati on A is seen to capture a higher number of fine particles as compared to the baseline setting. It is seen that the %age of particles captured by the screen area as opposed to the volute area varies significantly across some configuratio ns, unlike with a lower flow rate. The reasons for the variation in Configuration B can be attribut ed to the deleterious effect of the Sump baffle in preventing particles from settling to the botto m of the sump, as expressed in detail in the preceding section. Configuration C resulted in de creased particle capture in the screen area, partly due the same reason as with Confi guration B. An additional reason for lowered performance is described as follows. Under a ba seline condition in itself, there was a slightly higher liquid depth in the Volute area as opposed to the Screen area, at steady state. These measurements were made with a 2.5 psi pressure transducer, manufactured by Druck Inc, as part of a different study. Under the condition where a submergence baffl e was used, the overall depth of liquid in the Screened HS was increased further. The corresponding increase in the effective volume meant that, a higher amount of energy woul d be necessary to create and sustain a vortex equivalent in magnitude to that seen with c onfigurations that did not cause increased liquid depth. Therefore, any appreciative effect of in creased liquid depth was negated by the absence of sufficient energy to create and sustain a vortex. Th is is an observation co mmon to Configurations C and F, both of which employed a submer gence baffle. A glance at the particle size

PAGE 41

41 distributions in Figure 6 reveals that Configuratio n B, C and F captured less fine particles than Configuration A. On the other hand, Configurations D, E, G and H do not show any significant deviation from the Baseline configuration, despite reduced screen area and inlet throat area. This is because, the effect of these parameters comes in to play only in a situation where there exists a vortex that creates flow regime s that are conducive to hydrodynami c separation, such as flow rates close to the design flow rate. These effects are examined in further detail in the succeeding section on Tracer analysis. Treatment Influent Flow Rate at 125 % of Design Flow Rate Figure 4 depicts particle se paration performance at 125% of design flow rate across configurations. For the given vol ume of the screened HS, there is a significant improvement in particle separation performance over a baseline unit, across configurations. The experimental results presented in Table 3 clearly indicate the change in solid separation mechanisms in that they tend predominantly towards hydrodynamic sepa ration and screening as opposed to at lower flow rates, where the effects of gravitational sedimentation were predominant. The particle size distributions of the partic les captured in the screen area, volute area and the efflue nt are depicted in Figure 7. An observation that is consistent across configurations is that there is a clear shift in the particle size distribution of capture particle s towards a coarser particle diameter as is evidenced by the d15m, d50m, and d85m. All configurations with th e exclusion of D, E and H showed a marked improvement in capture of fine pa rticles compared to a ba seline setting, at an almost equal level. Configurations B and C, which did not function as well as the other configurations, at lower flow ra tes, did not have any deleteri ous effects on SSC removal. The location and geometry of the sump baffle allows it to work optimally, when particles do not settle on the baffle itself. At lower flow rates, particle trajectories were not predominantly influenced by hydrodynamic separation. Therefore, particles tended to settle on the baffle itself

PAGE 42

42 and hence be re-suspended. On the other hand, at the highest flow rate the particle settling trajectories are circular helical pa ths of progressively decreasing radii, enabling particles to settle into the sump through the center region of the inner screen area. Thus, particles were not deposited on the baffle, instead, se ttled to the bottom of the sump. Configurations D, E and H invol ved modifications to the scre en area, which resulted in increased differential pressure through the scre en and higher radial velocities as opposed to tangential velocities in the screen area, thus caus ing particles to exit the screen area. It is interesting to note that Configur ation E performed s lightly better than Configuration D, at solid separation. This indicates the effect of increased tangentia l velocities due to the nozzlelike effect of the decreased throat area. A consistent characteristic of particle size di stribution curves for the tested configurations is that almost all configurations showed a clear increase in the capture of finer particles, in the volute chamber, compared to a baseline setting. This was conclusive evidence to identify the beneficial effects of the Effl uent weir baffle, which was the common element to all these configurations. The effluent weir baffle clearly increased the capture of finer particles in the volute chamber, by lengthening the trajectory of particle s out of the screened HS. Discussion From the point of view of assessing the performance of the scr eened HS, equation (2) provides a preliminary method that can be extend ed to design. This ha s been ratified by the improved performance of Confi guration G which the highest isAA/among the configurations that involved modifications to sA and iA, and consequently high solid separation at flow rates close to dQ.Another observation is that Configuration H, perfor med at a significantly lower level as opposed to other configur ations, while maintaining the same isAA/.

PAGE 43

43 This leads to the conclusion that isAA/is a non-singular paramete r, and dependent largely on the dimensions of a given unit. It is seen that, for this pa rticular Screened HS, designing a screen which allows 40 % of the inner volume of flow to pass through, is best to promote ideal ratios of rtVV/. Further reduction of the inlet throat areatAhowever, leads to excessive and deleterious head build-u p upstream of the screened HS. It was also noted that increasingisAA/, by reducing tA did not have significantly enhancing effects on performance compared to the baseline isAA/ at a flow rate of 1.25*dQ, while causing lowered perfor mance at a flow rate of 0.75*dQ, due to excessive head buildup upstream a nd inadequate energy of the incoming flow. Tracer Study Results At the beginning of the run, the vortex is not formed and the process of hydrodynamic separation has not reached a stea dy state. At a given flow rate which is not conducive to the creation of a vortex, short circuiti ng does occur at the in itial portion of the tr acer study and this is largely dependent on the location of the outlet with respect to the inlet. Th e outlet is located to provide the longest path for an individual particle out of the system, under a condition of hydrodynamic separation in steady state. However, the location of the outlet will result in shortcircuiting under conditions other than those c onducive to steady state hydrodynamic separation. Such conditions exist in the period of time that is required for an ideal vortex to develop, measured from Time There is a clear indi cation of short-circuiting even for a baseline unit as is observed in the tracer C (t ) and F curves in Figures 8, 9 a nd 10. However, even with shortcircuiting in the initial portion of the RTD curv e, the screened HS has mean residence times almost twice in magnitude to the theoretical resi dence time based on the volume of the screened HS.

PAGE 44

44 At an influent flow rate of 25 % of de sign flow rate, there was an increase of approximately 5 % in average hyd raulic residence time for the best modified configuration (Configuration A) over the baseline configuration. However, this di d not have a significant effect on the % SSC removal efficiency of the unit. This is a clear indication of the influent particle size distribution as a predominant factor in dete rmining the performance of the screened HS. A large fraction (50 %) of the influent particle gradation consists of pa rticles that would be classified as settleable and suspended particles, and their removal would require an increase in residence time that is disproportional to the given unit, at th e given volume. Therefore, the particle separation process tende d towards that predicted by overflow rate theory. However, it should be noted that for the given volume, the me asured hydraulic residen ce time is greater than the theoretical residence time. Based on this, it w ould be apt to state that the performance of the screened HS at this flow rate is marginally bette r than that of a circular tank of similar volume, owing to the geometry of the unit. An interesting point to note about the C (t) curve at a flow rate of 25 % of Qd is that the peak concentra tion is seen to be lower than that of tracer responses at higher flow rates, and there is a tailing effect. This is explai ned by comparing the F curves for these three flow rates. It is s een that the deviation from plug fl ow increases with decreasing flow rate. At the lowest flow rate, the entire volume is not utilized, due to th e presence of dead spaces. This is also indicated by the fact that the residence time increase at the lowest flow rate for the screened HS in comparison to the theoretical resi dence time is lower in magnitude at lower flow rates, than at higher flow rates. Also to be noted is that any difference between commonly used indices for analyzing tracer curves, presented in Table 4, do not translat e into any significant difference in solid separation performance, for the given influent particle size gradation.

PAGE 45

45 At and inflow rate of 75 % of Qd, there is a difference in hydrodynamic response, immediately indicated by the hi gher peak concentra tion in the C (t) curve. As discussed previously in the section describing SSC removal results, this flow rate can be designated as a threshold flow rate for transition to steady st ate hydrodynamic separation from solid separation influenced more by gravitational forces than by vortex action. The variances ( 2), of the C (t) curves for an optimized setting (Configura tion A) and the baseline setting do not differ significantly. A glance at the F curve shown in Fi gure 9 reveals the mechanistic behavior of the unit at 75 % of design flow rate. There was an in crease in hydraulic resi dence time of about 8 % over a baseline setting. The normalized cumu lative tracer concentration was noted for Configuration A and the baseline setting, at values of the inde pendent variable, t, representing multiples of the theoretical residence time ( ). This was chosen as the time allowed for the flow regime to tend towards steady state hydrodynamic separation, in a magnitude proportional to the magnitude of the vortex. A look at values of F (n* ) with n taking values 1 to 5, indicates a clear separation between the F curves of an optimized se tting versus a baseline setting, with increasing flow rate. For instance, for an n value of 2, the correspondin g (1-F) value for the optimized setting indicates that 37 % of tracer mol ecules remained in the reactor for time 2* or greater whereas for a baseline setting, 31 % remained in th e reactor. This clearly indicates the effect of the effluent weir baffle as th e flow reaches steady-state. For an influent flow rate of 125 % of Qd, the trend towards plug fl ow behavior is more pronounced than with the lo wer two flow rates. The tracer res ponse curves for this flow rate are shown in Figure 10. At this flow rate, there is adequate energy to generate a vortex conducive to hydrodynamic separation. At this flow rate, th e primary and predominant solid separation mechanism is hydrodynamic separation. The variances ( 2), of the C (t) curves for an optimized

PAGE 46

46 setting (Configuration A) and the baseline setti ng do not differ significantl y, as with 75 % of design flow rate. The normalized cumulative tra cer concentration was noted for Configuration A and the baseline setting, at values of the independent variable, t, representing multiples of the theoretical residence time ( ), as described in the previous section. Values of F (n* ) with n taking values 1 to 6, indicates a more pronoun ced separation between the F curves of an optimized setting versus a baseline setting, with increasing flow rate. For instance, for an n value of 3, the corresponding (1-F) valu e for the optimized setting i ndicates that 15 % of tracer molecules remained in the reactor for time 2* or greater whereas for a baseline setting, 11 % remained in the reactor. Overa ll, there was an increase of ab out 10 % in residence time for Configuration A as compared to a baseline setting. From Table 4, it is noted that whil e commonly used indices such as the 2/ 2, Morrill Dispersion Index, Hydraulic e fficiency based on MDI, and t50/ do not show differ significantly between the best modified confi guration and the baseline configur ation, across flow rates, there is a clear lag introduced in the RTD curve by usin g an effluent baffle, as is evident from the t10, t50, t90 and tp The modal retention time index tp / shows a tendency for the modified configuration to tend towa rds plug flow as compared to the baseline setting. Although most other indices do not differ significantly between the modi fied and baseline confi gurations, it should be noted that they do indicate favorable conditions in the modified configuration. In summary, the provision of the effluent weir baffle in Configuration A resulted in an increase in hydraulic residence time compared to a baseline setting, with increasing flow rate. Conclusions The clarification response of a screened HS to an influent particle gradation, at a concentration commonly observed in urban runoff, was studied across flow rates representing 25, 75 and 125 % of the maximum hydraulic operational flow rate. This study concludes that

PAGE 47

47 although the screened HS is primarily a hydrodynamic separator, the va riables that affect particle separation differ significantly from that of a typical vortex separator. This can be attributed to the absence of an underflow, and the presence of a static screen. Therefore, the approach to optimizing the performance of the screened HS should be one suited to its mechanisms, and geared towards satisfying the constraints set by the given influent particle size gradation. The solid separation performance of the screened HS was seen to vary significantly with changes in the dimensionless ratio of isAA/(Screen area/Inlet throat area) was tested. It was concluded that, for this particular screened HS, designing a sc reen which allows 40 % of the inner volume of flow to pass through, is best to promote ideal ratios of rtVV/. The best modified configuration for enhanced particle separation wa s Configuration A. The resulting increased residence time was conf irmed by means of a trace r study. The effluent weir baffle had a significant impact on the reside nce time at the higher flow rate of 125 % of the design flow rate, with residence time increase d by approximately 10 %. Although the average residence time shows a marked increase even at lo wer flow rates, particle removal was mainly enhanced at the higher flow rate. Using the e ffluent baffle increased the residence time of particles which may have otherwis e exited the system at high flow rates, thereby increasing the chances of them remaining trapped in the unit. There was an increase of about 10% in SSC removal efficiency at 125% of design flow rate under the optimized setting. It would be beneficial for st ormwater treatment facilities that employ such devices to optimize them for the targeted removal performan ce, for the characterized influent particle size gradation, across the entire range of operating flow rates, and with the device filled to its residual treatment capacity, as is the case in real s ituations, in between period ic cleanouts of the device. These would ensure a more structured and defens ible approach to utilizing these devices for

PAGE 48

48 treating real-time storm events The importance of hydraulic op timization is all the more imminent considering that this provides a resourceful and scientific approach to maximizing the functionality of a hydrodynamic separator, wi thout expensive geometric modifications.

PAGE 49

49 Table 2-1 Comparison of target particle size di stribution with calculated and measured gradation utilizing 5 different Silica particle gradations Silica particle gradations Mass Fraction Particle Size 20/30 H-85 SCS 250 SCS 51 MUS 10 SUM Calculated Target Measured (microns) (g) (g) (g) (g) (g) (g) % % % 500-1000 40.4 3.8 0 0 0 44.2 5 5 5 250-500 0 40 0 0 0 40 5 5 5 100-250 0 160 76 0 0 235.6 28 30 35 50-100 0 36 92 13.2 0 141.2 17 15 15 8-50 0 0.4 32 166.8 8 207.2 25 25 22 2-8 0 0.4 0 75 52 127.4 15 15 13 1-2 0 0 0 21 20 41 5 5 5 SUM(g) 40.4 240 200 276 80 836.6 100 100 100

PAGE 50

50 Table 2-2 Comparison of target particle size distribution with calculated and measured gradation utilizing 5 different Silica particle gradations. Experimental matrix for optimization of the screened HS Details of Geometric Modifications Run # Setting Q EBa SPBb SMBc sAd [m2] iAe [m2] i sA A O-1 0.25* Qd N N N 0.5334 0.0106 50 O-2 0.75* Qd N N N 0.5334 0.0106 50 O-3 Baseline 1.25* Qd N N N 0.5334 0.0106 50 3 0.25* Qd Y N N 0.5334 0.0106 50 1 0.75* Qd Y N N 0.5334 0.0106 50 2 A 1.25* Qd Y N N 0.5334 0.0106 50 6 0.25* Qd Y Y N 0.5334 0.0106 50 5 0.75* Qd Y Y N 0.5334 0.0106 50 4 B 1.25* Qd Y Y N 0.5334 0.0106 50 8 0.25* Qd Y Y Y 0.5334 0.0106 50 7 0.75* Qd Y Y Y 0.5334 0.0106 50 9 C 1.25* Qd Y Y Y 0.5334 0.0106 50 14 0.25* Qd Y N Y 0.3340f 0.0106 31 13 0.75* Qd Y N Y 0.3340f 0.0106 31 15 D 1.25* Qd Y N Y 0.3340f 0.0106 31 17 0.25* Qd Y N N 0.3340f 0.0089g 38 18 0.75* Qd Y N N 0.3340f 0.0089g 38 16 E 1.25* Qd Y N N 0.3340f 0.0089g 38 19 0.25* Qd Y N Y 0.5334 0.0106 50 12 0.75* Qd Y N Y 0.5334 0.0106 50 20 F 1.25* Qd Y N Y 0.5334 0.0106 50 23 0.25* Qd Y N N 0.5334 0.0078h 68 22 0.75* Qd Y N N 0.5334 0.0078h 68 21 G 1.25* Qd Y N N 0.5334 0.0078h 68 26 0.25* Qd Y N N 0.4437i 0.0089g 50 25 0.75* Qd Y N N 0.4437i 0.0089g 50 24 H 1.25* Qd Y N N 0.4437i 0.0089g 50

PAGE 51

51 Table 2-3 Clarification response or particle separation (expressed as SSC removal efficiency) of different configurations to the influent particle grada tion (Figure 2.3), across flow rates representing 25%, 75% and 125% of design flow rate, for an influent concentration of 200 mg/L. Details of conf igurations are provided in Table 2-2. Details of Experimental Setup Particle separation Flow Rate Influent Conc. Screened Particles Mass Volute Particles Mass Particle Mass Mass Balance Error Run # Setting % Qd [mg/L] [g] [g] [%] [%] O-1 0.25*Qd 200 225.8 160.1 47.1 7.8 O-2 0.75*Qd 200 149.2 154.6 38.9 9.6 O-3 Baseline 1.25*Qd 200 67.3 136.4 28.7 7.1 3 0.25*Qd 200 210.5 208.8 52.5 -6.3 1 0.75*Qd 200 164.9 200.2 45.6 2.1 2 A 1.25*Qd 200 107.8 197.6 38.2 3 6 0.25*Qd 200 177.7 209.5 48.4 -0.7 5 0.75*Qd 200 134.2 205.3 42.4 1.8 4 B 1.25*Qd 200 114.7 196.5 38 1.5 8 0.25*Qd 200 174.8 229.2 50.5 -5.8 7 0.75*Qd 200 105.6 202.1 38.5 3.5 9 C 1.25*Qd 200 89.1 199.6 36.1 2 14 0.25*Qd 200 221.4 184.3 50.7 -8.1 13 0.75*Qd 200 142.9 182 40.6 -3.7 15 D 1.25*Qd 200 86.9 118.1 25.6 7.4 17 0.25*Qd 200 210.5 199.9 51.3 -2.7 18 0.75*Qd 200 156.8 156.8 39.2 0 16 E 1.25*Qd 200 76.9 160.3 29.6 5.4 19 0.25*Qd 200 214.6 206.6 52.7 -0.5 12 0.75*Qd 200 135 188.8 40.5 1.8 20 F 1.25*Qd 200 144 177.9 37.2 0.4 23 0.25*Qd 200 205.2 179.1 48 -5.8 22 0.75*Qd 200 145.5 155.4 37.6 -2 21 G 1.25*Qd 200 140.5 168.6 38.6 7.4 26 0.25*Qd 200 205.8 205.6 51.4 -9.1 25 0.75*Qd 200 128.7 176.7 38.2 -2.7 24 H 1.25*Qd 200 95.5 76.6 21.5 4.9

PAGE 52

52 Table 2. Indices for characterizing hydraulic response across flow rates representing 25%, 75% and 125% of Qd where Qd is the design flow rate Q Q=0.25* Qd Q=0.75* Qd Q=1.25* Qd Configuration A(Optimal) BaselineA(Optimal) Baseline A(Optimal) Baseline 192 192 64 64 38.4 38.4 tmean (s) 320 308 119 111 77 70 ti(s) 26 20 28 25 20 21 tp (s) 215 178 47 42 31 28 t10 (s) 110 107 45 45 31 29 t50(s) 292 274 103 95 62 57 t90(s) 614 530 218 198 133 119 MDI 5.58 4.95 4.84 4.40 4.29 4.10 1/MDI 0.18 0.20 0.21 0.23 0.23 0.24 ti / 0.14 0.10 0.44 0.39 0.52 0.55 tp / 1.12 0.93 0.73 0.66 0.81 0.73 2/ 2 0.31 0.32 0.30 0.30 0.29 0.29 t50/ 1.52 1.43 1.61 1.48 1.61 1.48 Hazen's N 3.79 2.85 1.84 1.79 2.00 1.97

PAGE 53

53 Figure 2-1 Experimental setup for testing the screened HS. Th e diameter of the HS was 89.3 cm and the diameter of the inner screen ar ea was 49.5 cm. The height of the unit was 96.5 cm. The storage capacity of the sc reened HS under quiescent conditions was 408 L. The screen aperture size for the screened HS used in this study was 2400 m. 70 KHz ultrasonic sensor Drop-box (Influent particle injection) Diversion valve HS Recirculation system Re-circ. pumps Storage tank B (4000 L) Storage tank A (4000 L) Effluent Inf. OD=15 cm ID=20 c m

PAGE 54

54 Figure 2-2 Geometry of the screened HS. A) Pl an and side view of detailed geometry of a screened HS. B) Typical fluid flow profile in side a screened HS. Outer Annular Chamber (Volute Area) Inner Chamber (Screening Area) Inle t Inner Chamber (Screening Area) Outle t Outer annular chamber (Volute Area) Cross-flow Screen Conical Sump A B

PAGE 55

55 Figure 2-3 Plots of targ et influent particle size distribution, calculated and measured gradation utilizing a mixture of five different US Silica Sand s (20/30,H-85,Min-U-Sil 10, Sil-Co-Sil 51 and Sil-Co-Sil 250). Details of the granulometric characteristics of the silica sands are provided in Table 2.1. The specific gravitie s of particles across the gradation were not seen to significantl y deviate from 2.65 g/cm3.The resulting gradation was verified by mechanically dry sieving the prep ared mixture (ASTM method D422). Particle diameter ( m) 1 10 100 1000 Percent finer by mass (%) 0 10 20 30 40 50 60 70 80 90 100 Target Influent Calculated Influent Measured Influent

PAGE 56

56 Percent of Q d (L/s) 0255075100125 C Effluent [mg/L] 60 80 100 120 140 160 180 Particle mass (%) 0 20 40 60 80 100 Baseline A B C D E F G H (A) (B) Figure 2-4 Clarification response of different configurations to for the non-cohesive sandy silt influent particle gradatio n across flow rates representing 25%, 75% and 125% of design flow rate, for an influent concentration of 200 mg/L. A) Measured separated particle mass in the HS. particle mass denotes particle mass expressed as % of mass in the influent that was captured in the HS. B) Effluent concentrations measured as SSC [m g/L]. SSC is suspended sediment concentration.

PAGE 57

57 Percent finer by mass (%) 0 20 40 60 80 100 Influent Baseline A B C D E F G H Screen area Particle diameter ( m) 10 100 1000 Percent finer by mass (%) 0 20 40 60 80 100 Volute area Figure 2-5 Particle Size Distri butions by mass, of separated part icles as a function of modified configurations Baseline, A, B, C, D, E, F, G, H, for an influent concentration of 200 mg/L, a screen aperture size of 2400 microns, at a flow rate of Q = 0.25*Qd. A) Screen area. B) Volute area. A B

PAGE 58

58 Percent finer by mass (%) 0 20 40 60 80 100 Influent Baseline A B C D E F G H Screen area Particle diameter ( m) 10 100 1000 Percent finer by mass (%) 0 20 40 60 80 100 Volute area Figure 2-6 Particle Size Distri butions by mass, of separated part icles as a function of modified configurations Baseline, A, B, C, D, E, F, G, H, for an influent concentration of 200 mg/L, a screen aperture size of 2400 mi crons, at a flow rate of Q = 0.75*Qd. A) Screen area. B) Volute area. A B

PAGE 59

59 Percent finer by mass (%) 0 20 40 60 80 100 Influent Baseline A B C D E F G H Screen area Particle diameter ( m) 10 100 1000 Percent finer by mass (%) 0 20 40 60 80 100 Volute area Figure 2-7 Particle Size Distri butions by mass, of separated part icles as a function of modified configurations Baseline, A, B, C, D, E, F, G, H, for an influent concentration of 200 mg/L, a screen aperture size of 2400 microns, at a flow rate of Q = 1.25*Qd. A) Screen area. B) Volute area. A B

PAGE 60

60 0 0 0 0 0 0 0 0 Tracer (NaCl) [mg/L] 0 20 40 60 80 100 120 140 160 Configuration [A] Baseline Q=0.25*Q d Figure 2-8 Tracer study of Confi guration [A] (modified) and the baseline configuration at a flow rate Q = 0.25*Qd. Elapsed Time (sec) 0 200 400 6 00 800 100 0 1200 1 400 Cumulative RTD F 0.0 0.2 0.4 0.6 0.8 1.0

PAGE 61

61 Figure 2-9 Tracer study of Configuration [A] (modified) and the baseline configuration at a flow rate Q = 0.75*Qd. Tracer (NaCl) Concentration [mg/L] 0 50 100 150 200 250 Elapsed Time (sec) 0100200300400 Cumulative Residence Time Distribution F 0.0 0.2 0.4 0.6 0.8 1.0 Q=0.75*Q d Q=0.75*Q d Configuration [A] Baseline Configuration [A] Baseline

PAGE 62

62 Figure 2-10 Tracer study of Conf iguration [A] (modified) and the baseline configuration at a flow rate Q = 1.25*Qd Tracer (NaCl) Concentration [mg/L] 0 50 100 150 200 250 Elapsed Time (sec) 050100150200250300 Cumulative Residence Time Distribution F 0.0 0.2 0.4 0.6 0.8 1.0 Q=1.25*Q d Q=1.25*Q d Configuration [A] Baseline Configuration [A] Baseline

PAGE 63

63 CHAPTER 3 CFD MODELING OF A STORMWATER HYDR ODYNAMIC SEPARATOR Introduction Particulate matter (PM) transported in urban rainfall-runoff has been identified as a significant contributor to overall deterioration of surface waters (USEPA 2000). Rainfall-runoff transports a hetero-disperse distribution of PM that ranges from colloidal to debris-size PM as well as PM-associated contaminants (Sansalone et al. 2007, Lee and Bang 2000, Sansalone et al.. 1998, Igloria et al.. 1997, Sansalone and Buchberg er 1997). Particle size distribution (PSD) of PM delivered by runoff varies with flow rate an d spatial position along a flow path, with urban source areas providing the more hetero-dispers e yet coarser PSDs (Kim and Sansalone 2008). One of the most significant advances in the field of urban runoff management is with respect to modeling hydrologic and hydraulic processes in these urbanized systems. Existing urban runoff models such as the Stormwater Management Model (SWMM) (Rossman 2007, Huber 1988), utilized worldwide, an d variations thereof, have gr eatly improved the efficacy of hydrologic and hydraulic design and an alysis in highly complex urba n drainage systems. While this has been a major advancement in the field of urban drainage, it is ou tside the current scope of such models to provide in-depth hydrodyna mic and particle clarif ication profiles for a geometrically and hydrodynamically-complex un it operation loaded by hetero-disperse PM. Many conventional urban runoff analysis methods for clarification of PM are extensions of historical settling or wastewater tank principles, which often a ssume ideal overflow theory and reasonably steady influent hydraulic as well as PM characteristics with time (Cristina et al. 2003). However, differing flow rates generated by rainfall-runoff processes, hetero-disperse PM not amenable to a single gravimetric aggregat e index such as TSS, and the geometric and hydrodynamic complexity of unit operations require more representative models and

PAGE 64

64 measurements; with the commensurate complexity added by such representations. Wastewater and drinking water practices ha ve incorporated models and me asurements of representative complexity. However, the development of representative models and measurements for stormwater treatment unit operations have lagged behind the proliferation of best management practices (BMPs) installations largely designe d, analyzed and permitted based on index or empirical concepts (Sansalone 2005). The terminology, BMP, generally elicits a more qualitative and empirical eval uation of treatment that has been based on %-removal and examination of unit operations as lumped b lack-box systems (Sansalone 2005). While treatment BMPs evolved from larg e surface area detenti on/retention basins designed originally to mitigate flow and volume, a class of preliminary treatment manufactured BMPs have focused on small-footprint unit operations minimize land cost s without hydrologic co ntrol. As a result, small-footprint unit operations without hydrologic c ontrol have proliferated over the last decade. Screened hydrodynamic separators (HS) as pr eliminary treatment for coarse PM (> 75 m) and trash/debris; fall into this category of unit operations (Rushton 2004, Rushton 2006, Stenstrom and Lau 1998). In a HS gravitational settling an d to a lesser extent inertial forces and size exclusion separates coarse PM (Brombach et al .. 1987, Brombach et al.. 1993, Pisano et al.. 1994, USEPA 1999, Andoh et al.. 2003). The HS provides the potential for longer particle trajectories per given unit su rface area in comparison to traditional BMPs such as detention/retention basins. Despite thousands of HS installations in North America over the last decade, existing literature does not provide funda mental insight into performa nce of an HS other than %reduction information with an HS treated as a lumped black-box. Previous studies in the literature have ranged from overflow rate theory (Weib 1997) to semi-empirical approaches

PAGE 65

65 based on broad assumptions of vortex flow behavior (Paul et al.. 1991). Th e lack of mechanistic evaluations and use of lumped index methodologies have led to difficulties in HS scaling and implementation. Fenner et al.. (1997, 1998) repo rt that similitude analysis does not yield a single dimensionless group that can be used in scale-up of a HS. These approaches can be contrasted with CFD approaches to examine PM fate in treatment unit operations such as the HS. CFD approaches to model particle-laden flows are an active research area (Curtis et al.. 2004, van Wachem et al.. 2003) for many hydraulic systems including a HS. Faram et al.. (2003) utilized a Reynolds Stress Model (RSM) to mode l turbulent flow and a Lagrangian Discrete Phase Model (DPM) to model PM fate. Lim et al.. (2002) used CFD-predicted velocity profiles to study behavior of flocs in a vortex separator with a Renormalization Group (RNG) kturbulence model. Tyack et al.. (1999) compar ed measured velocity profiles in a vortex separator to those predicted by a Renormalization Group (RNG) kturbulence model. However, many of these CFD mode ling studies did not couple representative PM measurements with the numerical approach. Limiting the modeling approach, for example with an overflow rate evaluation of an HS, or limiting the representa tiveness of measurements, for example with indices such as TSS and auto sampling has resulted in most HS studies as %-removal evaluations of a black-box unit. This study examined a HS with two cylindrical chambers illustrated in plan in Figure 3-1. The flow inlet is tangential to the inner cylindr ical chamber and the two chambers are separated by a static screen. The static screen consists of a regular array of apertures. These apertures allow flow to exit the inner screen chamber and enter the volute chamber. The inner cylindrical chamber, along with the screen and the sump area are designated the screen area. The outer

PAGE 66

66 less turbulent annular area is the volute area. PM separatio n by this HS configuration is primarily a function of various parameters, as expressed in the following equation. ),,,,,,(strps particlesdvvQdvfM (3-1) In this expression,particlesM is the mass of particles separated, sv is the discrete particle settling velocity, pdis the mass-based particle diameter, Q is the influent volumetric flow rate, rv is the radial velocity com ponent in the screening area, tv is the tangential velocity component in the screening area, is the hydraulic residence time, and sd is the diameter of the screen apertures. The screened HS has no underflow, and therefore the mechanisms of hydrodynamic separation are different from typical vortex-based separators, wher e secondary currents play an important part in PM separation. The screen of th e HS consists of apertures of diameter 2.4 mm. These apertures are not expected to screen particles wherein there is interaction with the screen wall in the form of reflection, interception or stra ining, for particles with diameters less than the aperture size of 2.4 mm. In this study the mass-based d50 of the influent PSD is 78 m). The PM separation mechanism in the screening area or the inner chamber of the screened HS is summarized as follows. The aperture geometry allo ws for increased tangential fluid velocities as opposed to increased radial fluid velocities. Particles that enter the screening area have a tendency to exit the screen, advectin g with velocities proportional to the radial fluid velocity. As the particles travel outwards from the center and towards the screen, the radial fluid velocity component is significantly weaker in comparis on to the tangential flui d velocity component. Therefore, a fraction of partic les do not possess adequate moment um to pass through the screen. This phenomenon has been termed as inertial separation in the rest of this paper.

PAGE 67

67 Objectives This study has four objectives. The first ob jective is to coupl e a pilot-scale HS configuration utilizing representative PSD sampling and anal ysis as well as measured PM material balances with the development of a CF D-based HS model. In this first objective the CFD-based HS model is validated by concentrat ion, mass and PSD measurements for pilot-scale testing conducted across the opera ting flow rate range of the HS The second set of objectives evaluates the role of flow rate and PSD, acr oss the hydraulic operating range of the HS, on the separation of PM as a function of particle size (important for examining trapped and exported PSDs for O & M and chemical part itioning during storage). The th ird objective is to examine spatially-distributed particle trajectories in th e HS and settling velocity frequency distributions within the HS to Newtonian and Stokesian models. In this objective the HS is examined as a unit operation where the multiple (screened and volute) chambers have unique sedimentation roles for PM in contrast to a conventional examin ation of the HS as a lumped black-box. The fourth objective is to quantify differences between validated CFD-based model results with results based on examining the HS as a lumped sy stem with a conventional overflow rate model. As part of this fourth objective, PM is represen ted as the entire measured PSD (providing a direct comparison to the CFD-based model at each flow rate), or based on a gravimetric index of the PSD (the mass-based median of the PSD; the d50m) and therefore providing an evaluation of how the modeled performance utilizing an index meas urement deviates from measured PSD results. (It is further noted that most index-based measurements such as TSS provide no indication of the d50m).

PAGE 68

68 Methodology Pilot-scale testing setup Figure 3-1 illustrates the schematic plan view of the instrumented full-scale HS as part of the pilot-scale testing configur ation. The flow control valves and pump were calibrated (with volume and time) across the entire flow rate range of 0.16 to 23.4 L/s. A Parshall flume was also calibrated volumetrically over the entire flow rate range, with triplicate measurements, at an error less than + 1%. A known and constant gradation of in fluent particles were injected at the free-flow drop-box located immediately downstream of the Parshall flume and immediately upstream of the HS. The overall diameter of the HS is 1.524 m and the diameter of the inner screened chamber area is 0.65 m. The height of the HS is 1.82 m. The empty HS sump storage volume is approximately 1580 L. The design hydraulic operating volumetric flow rate (Qd) is 15.89 L/s based on the manufactured specification. The HS tested was hydraulically-sized based on hydrologic loadings from the impervious 1088 m2 watershed at the pilot-scale facility located in Baton Rouge, LA. The static cross-flow cy lindrical screen with a diameter of 0.65 m consisted of apertures, each with a d50 of 2.4 mm. The overall screen area that was open to flow (through these non-clogging apertures) was approxi mately 40 % of the total screen area of 0.5334 m2. Influent Particle Gradation The clarification response of the HS is highly dependent on the influent PSD (Kim et al.. 2004); in this study the textural classification is noncohesive sandy silt. The influent mass-based PSD is illustrated in Figure 3-2. It is noted that approximately over 50 % by mass of this gradation, is finer than 75 m and therefore, over 50 % by ma ss of this gradation consists of settleable (nominally ~25 to 75 m) and suspended par ticles (nominally 1 to ~25 m) (Kim and Sansalone 2008). The influent PSD was prepared by a weighted combination of commercially-

PAGE 69

69 available gradations of solid sub-rounded silica pa rticles. The dry mass of particle mixture was added to each experimental r un based on approximately 38,000 L of influent ( 25 4 C) to achieve a 200 mg/L influent con centration, which is a typically observed concentration in runoff at this paved source area waters hed (Sansalone et al.. 2005). Pe riodic QA/QC checks of influent PSD and concentration were conducte d to ensure consistency of the influent particle gradation. Field Test Procedure Before each treatment run the storage tanks, the recirculation pipes and the HS were cleaned and washed thoroughly by pumping potable water through the system. The desired test flow rates and velocities were achieved through the use of the influent pump and the recirculation system to vary flows. Calibrated flow control valves were operated to achieve the desired flow rate. Flow measurements were validated throughout each run and a cumulative volumetric balance was based on a Druck pressure sensor in each influe nt supply tank. Druck pressure sensors (2 psi, resolution of 0.5 mm) were installed on the influent (sump) side and volute side of the screen as well as at the outer diameter of the volute chamber before effluent discharge from the volute area. PM with the infl uent gradation shown in Figure 3-2 was injected just upstream of the inlet of the HS, after achievi ng a steady influent flow rate with fluctuations of less than + 1 % of the target flow rate. The rate of particle injection was constant on a gravimetric and granulometric PSD basis. 10 duplicated 2 L efflue nt samples were collected for water quality and suspended sediment concentration (SSC) analyses, at the outfall of the effluent pipe. Approximately 400 L of effluent, repres enting 10 duplicated 20 L samples, was collected and composited into several 210 L polypropylene tanks, at the same sampling frequency as the 2 L samples, in order to recover sufficient particle mass to carry out efflue nt particle gradations.

PAGE 70

70 Laboratory Analyses Two 20 L composite effluent samples (A and B) were prepared from the 10 individual replicate (A and B) 2 L samples. The entire measured volume of each replicate composite sample was then filtered using a nominal 1 m glass fiber filter. Since the entire volume was examined of each replicate composite sample was analyzed the gravimetric index of total suspended solids (TSS) was equivalent to suspe nded sediment concentra tion (SSC) without the inherent sampling and analysis errors associated w ith TSS. Particles captured in the screened HS from each run were completely recovered as wet slurry from the HS screened area and volute area separately, and these areas cleaned for the next run. Effluent PM captured in the 210 L polypropylene tanks was also recovered as separate slurries. At least 5 hours of quiescent settling was allowed before slurry recovery. The recovered PM slurry was transported to the laboratory, dried at 40C and the mass was recove red. Total effluent mass was computed from the measured sample effluent SSC for the total volume treated. A flow volume totalizer was used as a check of the incremental flow volum e measurements for each treatment run. A mass balance check requiring at leas t a 90+% recovery across the HS was performed based on SSC. Mass Balance Error = %10100 )( i e HS iM MMM (3-2) In this expression, iM is the mass of particles in the influent ,HSMis the mass of particles captured by the screened HS and eM is the mass of particles in the effluent computed from the measured effluent SSC corresponding to total treated volume. All gravimetric particle measurements were carried out on a dry mass basi s. With the mass balance error criterion ( 10 %) satisfied, the particle removal efficiency (particlesM ) is calculated.

PAGE 71

71 100 )( )( i HS particlesM M M (3-3) The dried PM mass was mechanically sieved to obtain a PSD according to a modified method for ASTM D422 (sieve analysis) (ASTM 1993, Sansalone et al.. 1998). Sieves utilized included the #30, #50, #100, #140, #200, #270, #400, #500 and the pan. Mass balances were determined for each analysis and mass balance erro rs were in the range of 1 to 2 % by dry mass for all PSDs. Computational Fluid Dynamics Methodology The hydrodynamics and particle dynamics of the HS vary as a function of x, y and z spatial coordinates. As a result, a three dimensional (3-D) approach was required. The generalized 3-D scalar conservation equation for a co ntrol volume (CV) with volume ) (zyxV is utilized. (Versteeg et al.., 1995) Sgraddivudiv t )()( )( (3-4) In this expression, is any fluid property per unit mass, such as mass fraction, velocity, etc; uis the fluid velocity, is the diffusion coefficient, and Sis the source/sink term. z w y v x u udiv ) ( (3-5) The mass continuity equation is obtained by assigning a value of 1 to 0 udiv t (3-6) Modeling Fluid Flow The dominant flow regime is turbulent in the HS. The Reynolds numbers range from 102 to 105, indicating the requirement of a turbulence model that can effectively model flows that range from laminar to transitional to fully turbulent. A Reynolds averaged Navier-Stokes or

PAGE 72

72 RANS approach (Ferziger et al.. 2002) was uti lized to resolve turbulent flow. The RANS approach is based on time averaging the Na vier-Stokes equations. This process yields additional terms in the standard transport e quations, termed the Re ynolds stresses. A kmodel (Launder and Spalding 1974) accounted for th e effects of turbulence on mean flow. A closed solution is obtained for th e turbulent transport equations by relating Reynolds stresses to an eddy viscosity (). Newtons law of viscosity is applied to illustrate the relationship between viscous stresses and Reynolds stresses. It should be noted that eddy viscosity (t) is a nonphysical quantity, and is expressed by the following equation. 2k ft (3-7) In this expression, kis the turbulent kinetic energy per unit mass. ) (5.0222wvuk [L2T-2] (3-8) In this expression, wvu ,are vector components of velocity fluctuations due to turbulence and is the dissipation rate of turbulen t kinetic energy per unit mass. Vortexinduced flows in the HS did not exhibit secondary currents and flows are not highly swirling. The standard kmodel has been suggested as a robust a pproach in situati ons where the swirl velocities are not large (Mohammadi and Pironneau 1994) but has also been applied in the case of hydrocyclones with highly swirling flows (Now akowski et al.. 2004, Statie et al.. 2001, Petty et al.. 2001). The full Reynolds Stress Model (RSM), the realizable and renormalized kmodels were tested, and did not s uggest any improvement that might warrant the increased computational power and time. The standard kmodel equations are expressed for k and respectively.

PAGE 73

73 k bk jk t j i iSGG x k x ku x k t )() ( (3-9) S k CGCG k C x x u xtb k j t j i i 2 2 3 1) ( )() ( (3-10) kG represents generation of k due to the mean velocity gradients; bG is generation of kdue to buoyancy; 1C ,2C and 3C are constants ;k and are the turbulent Prandtl numbers for kand respectively; kS and S are user-defined source terms. The constants were determined by Launder and Spaulding (1974). It was hypothesized that isotropy of Reynolds stresses can be assumed reasonably in the case of the HS. Values of1C ,2C ,3C ,k and in the model were 1.44, 1.92, 0.09, 1.0 and 1.3 respectively (Launder et al.. 1974). The no-slip boundary condition, along with effects of viscous blocking and kinematic damping creates large gradients in variables near the HS walls. To account for these phenomena semi-empirical wall functions for the kmodel were used (L aunder et al.. 1974). Modeling the static screen The shape of the screen apertures results in a weakly reversed flow direction in the volute chamber. The screen area open to flow is designed so that screen radial velocity is approximately an order of magnitude lower than the inlet ve locity. The non-cohesive, inorganic and granular nature of stormwater PM, and large 2.4 mm aper tures results in a screen that non-clogging, at least for stormwater. Meshing the screen with structured or unstru ctured meshing schemes places severe constraints on time and computational power. To overcome this difficulty, the static screen was modeled as a porous perforated plate with th e addition of a momentum source term to the standard fluid flow equations. Th is momentum sink contri butes to the pressure gradient in the porous computati onal cell, creating a pre ssure drop that is pr oportional to the fluid

PAGE 74

74 velocity in the cell. The source term is composed of two parts: a viscous loss term and an inertial loss term. For simple homogeneous media, the si nk term is expressed by the following equation. imag i i ivvCvS 2 12 (3-11) In this equation, iS is the source term for the ith momentum equation, is the permeability, iC2is the inertial re sistance factor, iv is the velocity in the ith momentum equation, magv is the velocity magnitude in the computati onal cell. The pressure drop through the screen was measured experimentally with pressure tran sducers and it was found that the head loss was 38.1 mm (150 mm across the entire HS) at design flow rates. Given the HS height of 1.82 m, these head loss values are low. The viscous resi stance term was assumed to be negligible and the source term was then modeled considering only the inertial resistance term, and the inertial resistance factor per unit thickness of the plate, i.e. )(2iC was calculated for the x, y and z directions as described by th e following equation (FLUENT 2005). n K CiL i )(' )(2 (3-12) In this equation, nis the plate thickness, and )(' iLK is calculated in the following equation. 2 2 )()('*n m iLiLv v KK (3-13) In this equation, 2 mv is the square of the velocity through the plate, considering that the plate is m % open to flow in the given direction and 2nv is the square of th e velocity through the plate, considering that the apertures are 100 % open to flow in the given direction.)( iLK is the loss factor in the ith direction and is calculated from the following equation. 2 )( oiL screenvKp (3-14)

PAGE 75

75 In this expression, p is the measured pressure lo ss through the stat ic screen. 2C is calculated based on the %ages of sc reen area open to flow described previously for the x and y directions. The inertial resistance in the z-directi on is theoretically infinite. However, to specify 2C in the z-direction, a value sufficiently larg e enough (twice the order of magnitude of the largest value of2C) was used. The assumption is that the porous cells (apertures) are 100% open, indicating that there is no physical/geometric obstruction to flow through the apertures. Modeling the Particulate Phase Typically, multiphase flows are modeled with an Eulerian-Eulerian approach or an Eulerian-Lagrangian approach (van Wachem et al.. 2003). The latter is commonly accepted to be suited to modeling dilute flows, defined as flow s with a particulate volume fraction (PVF) less than 10% (Brennen 2005), which is the case in th is study (PVF << 0.1 %). In this approach, the flow field was solved using the Eulerian approach Following this, particles were tracked using a Lagrangian Discrete Phase Model (DPM). The DPM is derived from force balances based on classical Newtons (Turbulent and transitional regimes) and Stoke s (Laminar regimes) laws describing particle motion, and is summarized by the following equation. x p px p D pF g uuF dt du )( ) ( (3-15) In this expression, dF 24 Re 182pD ppC d (3-16) 2 3 2 1Re Rep p Da a aC (3-17)

PAGE 76

76 In the above expressions uis the fluid velocity, pu is the particle velocity, is the fluid density, p is the particle density, pd is particle diameter, is the viscosity, 321,, aaa are empirical constants that apply to smooth spherical particles as a function of the Reynolds number (Morsi et al.., 1972) andpReis the particle Reynolds number. Particle trajectories are obtained by integration. For this study, particles of cons istent morphology were injected across the entire flow cross-section immediately above the inlet to the HS to obtain comparable particle trajectories across varied flow ra tes, by reducing uncertainty in initial spatial location. Particles were defined as silica particles of diameters associated with the sieves utilized and a measured specific gravity of 2.65 as determined by helium pycnometry (Sansalone et al. 1998). Particles were tracked for a specific length for each fl ow rate, based on hydraulic residence times calculated by injecting neutrally buoyant submicron tracer particles. Part icles that remained in the HS after integrating over the specified length were considered to have been separated by the HS. Particle removal was thus de fined by the following equation. 100 I HSN N p (3-18) In this expression, HSN is the number of particles that remain in the screened HS, and INis the number of particles injected at the inlet. Discretization of Geometry The computational geometry of the screen ed HS was generated using GAMBIT, a preprocessing software for FLUENT. Boundary cond itions were specified in this step. It is important to note that the free surface of the flow was specified as a shear-free wall, while assigning boundary conditions. The computati onal geometry was discretized using an unstructured mesh with tetrahed ral elements. The TGrid (Qi et al.. 2006) algorithm was used to

PAGE 77

77 generate the mesh. Progressively finer meshes were tested, with 0.4, 1.96 and 2.10 million cells respectively, to examine the effects of com putational mesh sizes on particle separation predictions. Localized mesh refinements were not utilized. The HS mesh is shown in Figure 31. Discretization of Governing Equations A Finite Volume Method (FVM) was used to convert the governing equations to algebraic equations. The governing equations were integrat ed over each control volum e to yield discrete equations. A cell-centered scheme was used in the process of discretization. Values of cell faces were computed using a Second-Order Upwind Scheme (Barth et al.. 1989). Upwinding implies that the face values are obtained from upstream cell quantities relative to the normal velocity. The Second-Order Upwind Scheme is t ypically suggested as a requirement to procure accurate results with unstructured meshing schemes. Solution Schemes Pressure-velocity coupling is an issue that must be addressed during the process of obtaining a sequential solution to the momentum and continuity equations. The SIMPLE (SemiImplicit Method for Pressure Linked Equations) al gorithm (Patankar 1980) was used to introduce a pressure term in the continuity equation. Various other algorithms have been devised for effecting pressure-velocity c oupling, such as the SIMPLEC (Vandoormaal et al.. 1984.) and PISO (Issa 1986) algorithms, but did not provi de any significant improvement in convergence from the SIMPLE algorithm for the HS. The SI MPLE algorithm was chosen as the consistent approach to pressure-velocity coup ling across the range of operating flow rates. The criterion for iterative convergence was set at 1x10-3, which is a typical constraint for multiphase flows which do not consider chemical kinetics (Ranade 2002).

PAGE 78

78 Results and Discussions PSD results Figure 3-2 illustrates the phase shift in particle size distribution from influent (left) to effluent (right) as transformed by the HS. Particle s in the effluent are finer than in the influent, as a result of the separation mechan isms of the HS. As flow rates increase, effluent particles are progressively coarser as compared to the finer e ffluent PSDs for the lower flow rates. Results also indicate the finer effluent PSDs (at lower flow rates) are significantly less hetero-disperse. The median particle size based on mass (d50m) is commonly selected as an index for a PSD and this index is plotted in upper plot of Figure 3-3-2. The d50m of the effluent PSD is significantly finer than that of the influent d50m across the entire flow rate range. The overall separation mechanisms of the HS can be categorized into three primary mechanisms. These mechanisms are discrete particle (Type I) settli ng, size exclusion and inertial separation. While gravitational settling and to a lesser extent size exclusion are the predominant separation mechanisms at lower flow rates, inertial separation co ntribute to overall particle separation at higher inflow rates. CFD model results Refining the HS mesh from 0.4 million to 1.96 m illion cells, resulted in an improvement in model predictions, as illustrated in Figure 3-3. However, further refinement of the mesh had little effect on improving overall model performance, confirming that the solution is grid independent. Model results are pres ented in two parts. The first part consists of modeled particle separation efficiencies at given flow rates, for individual partic le sizes, and these results are summarized in Figures 4. The second results se t is depicted in Figure 3-5, where results are compared on the basis of overall separation efficiency expressed as particle mass (gravimetric mass of SSC in the effluent) a nd effluent concentrations (Ceff) expressed as SSC.

PAGE 79

79 The first presentation of results of the CFD model is further grouped into three categories based on flow rates. The first category consists of the lowest flow rates (approximately 1 to 10 % of the HS design flow rate, Qd = 15.89 L/s), where gravitational sedimentation effects predominate. The second category consists of in termediate flow rates (approximately 25 to 75 % of Qd), where gravitational sedimentation and exclusion by the deflective screen are the predominant separation mechanisms, and inertial sepa ration effects are low. It is observed that up to approximately 50% of Qd that the coarse fraction (> 75 m) of the PSD is effectively separated. The third category consists of high flow rates (approximately 100 to 150 % of Qd, where the effects of inertial separation and incr eased particle residence times as a result of lengthened particle trajectories are increasi ngly important with respect to gravitational sedimentation. Absolute relative % difference (RPD) was chosen as a parameter to compare measured and modeled data. RPD was calculated from the following equation. 100 data Measured data) Modeled-data (Measured RPD Absolute (3-19) Results in Figure 3-4 indicate that for the first category of flow rates, 0.16 L/s, 0.79 L/s and 1.59 L/s, there is agreement between measured and modeled data, with absolute RPDs lesser than 4 %. It is noted that the CFD model predic ts particle separation behavior effectively across the entire PSD. In the second category of flow rates, 3.97 L/s, 7.94 L/s and 11.92 L/s, results demonstrate agreement between measured and m odeled with an RPD < 10%, although the model tends to slightly overestimate removal of coarser PM. In the high flow rate category, 15.89 L/s, 19.87 L/s and 23.34 L/s, results demonstrate agre ement but there is a tendency for CFD to over predict removal of coarser PM. However, the overall RPD still remained < 10%. In the field of urban drainage these results are noteworthy given that the flow rates range over two orders of magnitude and the hetero-disperse influent PSD ranges by almost three orders of magnitude in

PAGE 80

80 particle size. Results are also of signifi cant value given that the current paradigm of measurement for design, analysis and permitting for BMP performance is the index parameter TSS; although only occasionally a specifi c gravity or size index (such as d50m) are ascribed. The second set of results, Figure 3-5, compar es model and measured results of effluent PM mass and concentration. Effluent measured and modeled concentrations are compared, for an influent SSC of 200 mg/L, across the entire range of inflow rates. Range bars associated with measured data represent the standard deviation between replicated effluent samples. Results indicate that the model predicti ons of SSC reproduce the measured data with an absolute RPD < 10 %. Figure 3-5 also compares the overall particle separation by the screened HS is expressed in terms of gravimetric SSC mass in the effluent, particle mass. Model and measured results agree with an absolute RPD < 10%. Range bars indicate the overall mass balance error (MBE) associated with the SSC measurements for each pilot-scale test at a given flow rate. CFD results are compared and contrasted to conventional methods in Table 3-1 for the HS and loadings of this study. While the table summarizes the ag reement between measured and CFD results, the table also illustrates deviati on for the conventional overflow rate model (Q/A). Furthermore results indicate that despite the use of a lumped black-box model such as overflow rate, knowledge of a PSD for the PM loading provides a far more reasonable (albeit less accurate than CFD) set of results than the convention use of a lumped index such as a d50m for PM. It is further noted that in most conven tional analyses, even a d50m is not available for gravimetric indices such as TSS. Post-processing CFD model results: Pa rticle dynamics and hydrodynamics Figure 3-6 illustrates the particle trajectori es calculated by a Lagrangian DPM for the screened HS, for an influent flow rate (Q) of 3. 97 L/s. This flow rate was chosen as a typical flow rate on the basis of its frequency of occu rrence in typical stormwater flows and on the HS

PAGE 81

81 hydraulic design capacity for watershed at this pilot-scale facility (Sansalone et al.. 2005). Three particle sizes were chosen for illustration, 450 m in the sediment size range (medium sand), 75 mm as the nominal particle size that differentiates coarse and fine PM (differentiates sediment and settleable PM), and 25 m as the nominal size that differentiates settleable and suspended PM size ranges. Plot (a), (b) and (c) depict partic le trajectories for a particle diameter (dp) of 450, 75 and 25 m respectively. The dependence of par ticle separation on partic le size is clear. It can be observed for the coarse end of the si ze spectrum that particles are influenced predominantly by gravitational forces whereas for the fine end of the size spectrum the suspended particle behavior is largely a function of inertial hydrodynamic forces. Figure 3-7 illustrates HS velocity distribut ions, for particles injected at the inlet geometric centroid. Plots (a) and (b) illustrate the spatial variation of particle velocity. A neutrally buoyant tracer particle was chosen to represent fluid velocity in a comparable Lagrangian frame of reference. Distance traveled by a tracer particle, from the injection point (geometric centroid of the inlet) to the outlet of the HS, was recorded as the overall path length of a fluid particle. The distances traveled by discrete PM sizes were normalized to the overall path length. Plots (c) through (f) il lustrate the mech anistic aspects of partic le separation in the HS through normalized frequency distributions of pa rticle velocities. Due to the difference in mechanisms within the screen and volute area, results are presented separately. Plot (c) and plot (d) represent particle velocities in the screen area for a 450 m and 25 m particle respectively, while plot (e) and plot (f) represent particle settling velocities (vs) within the volute area for the 450 m and 25 m particle respectively. Qualitativ e variations in the histograms are highlighted by dotted numbered circles. In explaining partic le behavior, it should be noted from Figure 3-6

PAGE 82

82 path lines that the inlet velocity vector is exac tly aligned in the y-direction. Thus, the inertial force is predominantly represented by th e y component of the velocity vector. Dynamics of a particle with dp= 450 m Plot (a) depicts the par ticle velocities of a 450 m particle relative to fluid velocity, as a function of the normalized path length. The initial fluctuations in particle velocity of the 450 m particle are attributed to the sliding of the particle along the turbulent boundary layer at the bottom of the inlet. The point where the inlet pipe opens tangentially into the inner screening chamber is located at a distance close to 18% of the total normalized path length. Following this, the particle falls into the screen area, and its motion is almost entirely affected by gravitational forces and correspondingly, in plot (c), number indicates that the particle velocity that occurs with the highest frequency does not diverge si gnificantly from Vs predicted by Newtons law (indicated by the bulls-eye symbol in the plot). In addition, it is noted that the geometric centroid of this distribution lies close to Vs predicted by Newtons law. As the particle settles in the inner screening chamber, the particle trajectory tends to be closer to the scr een due to higher radial velocities compared to tangential velocities and it escapes the screen, to finally settle at the bottom of the volute area. This co rresponds to the number in pl ot (e) which indicates that the predominant velocities of a 450 m particle in the volute area are close to zero, indicating that the particle has been separated by gravitational settling. Having es caped the screen, the particle settles into the volute area with a velocity in the range of th at indicated by number Dynamics of a particle with dp= 25 m Plot (b) shows the relative particle velocities of a 25 m particle relative to fluid velocity, plotted as a function of the normalized path length It is seen that a 25 m particle essentially follows the fluid pathlines and is not separated by the HS. Plots (d) a nd (f) indicate a wider distribution of particle veloci ties in the HS for the 25 m pa rticle as opposed to a 450 m

PAGE 83

83 particle, leading to the inference that a 25 m particle is largely affected by the characteristics of the flow regime. The following observations were made regarding th e spatial location of occurrence of velocities. Firstly, in the screen area, in Plot (d ) the number indicates the low velocities occurring at the center of the screen area and the nu mber indicates the increase in velocities as the particle approaches the scre en. Number occurs at the screen, and is dominated by the y-component of velocity. Number shows the increase in velocity towards screen. This occurs closer to the screen than num ber Number occurs at the inlet to the screened HS. This is inferred from the magnitude of the velocity, which is equal to the inlet velocity of 0.08 m/s. In plot (f), number indi cates velocities at the outlet of the HS, where the y-component is less predominant. Number indicates the decrease in velocities from the screen towards the outlet, and hi ghlight indicates the veloc ity at the screen, for the 25 m particle. It is noted that the cen troid of the distributions for both the screen and volute areas for a 25 m particle diverges signi ficantly from a value of Vs predicted by Newtons law. The fate of PM in the HS can be influenced by the point of injection. The trajectory analysis dealt with a geometrica lly representative injection point, mainly for the purpose of illustrating how CFD results are utilized to examine a geometrically-complex unit operation. Conclusions In the field of urban drainage treatmen t, the conventional pract ice is to evaluate treatment BMPs as lumped black-box units ex amined by overflow rate models for PM indices, sampled by automated samplers, examined based on indices such as TSS, with results characterized by %-removal. In contrast, this study utilized PSDs, mass balances, CFD, unit operation concepts and spatially-distributed ev aluation of a commonly-deployed screened HS unit operation instead of these conventional treatment paradigms of urban drainage. This study applied the principles of CFD, using a Finite Volume Method (F VM), to model PM behavior of

PAGE 84

84 the geometrically and hydrodynamically -complex HS, using a standard kmodel to account for effects of flow turbulence and resolve the R ANS equations, and a Lagr angian discrete phase model (DPM) to track particles. The static scr een was modeled as a porous cylinder by adding appropriate source terms to the flow equations. The CFD model was validated with pilot-scale data across the operating range of flow rates and PSDs. PM data were collected from representative manual sampling and resolved into PSDs, concentration and mass; with mass balances conducted for each pilot-scale test a nd each PSD analysis; in contrast to more commonly utilized indices such as TSS. Each te st was further evaluated with pressure sensor, flow and volume measurements. CFD model valid ation results for PM co ncentration, mass and PSDs were within 10% of the pilot-scale measured data. Model grid stability and independence were demonstrated. Post-processing results prov ided insight into the mechanistic HS behavior by means of 3-D hydraulic profiles and particle trajectories and allows re-designs to further mitigate short-circuiting. Results demonstrate that while coar se PM on the order of several hundred microns are separated by the HS at design flows, settleable and suspended particles are largely eluted from the HS even at flow rates less th an design flow and under clean unit conditions of this study (therefore scouring of previous sedi ments did not occur). The ability of the CFD model to reproduce treatment results across a rang e of flow rates and PSDs suggest that the validated model can serves as a foundation upon which design alternatives can be proposed. A validated CFD-based iterative approach to design of this HS as a preliminary unit operation has the potential to provide reduced prototyping cost s with improved performa nce, as a result of carefully designed experimental matrices, focuse d on meeting coarse PM control requirements for downstream treatment units. It is noted that for this study, mo st parameters required for the

PAGE 85

85 CFD model were measured. The only parameters that were provided in the model were the inertial resistance coefficients. These were obtai ned by physical measurement of aperture areas of the screen, these areas being open to flow. The pressure drop across the static screen was also measured. The apertures geometry and dist ribution was uniform throughout the screen. This study demonstrated the ability of a CFD approach to model effluent mass, concentration and PSDs of PM from the HS. In th e field of urban drainage treatment, it has been argued that this level of agreement between meas ured and modeled is not achievable. None the less, this level of agreement was achieved through the combination of a validated CFD model and representative measurement of flow, granul ometric and geometric quantities coupled with material balances; as well as resolution of PM into a PSD through the common practice of mechanical sieve analysis with the companion ma ss balances. Clearly the need to resolve PM into a PSD is an important conclusion from Table 3-1. Table 3-1 also illustrates that despite the simplicity of an overflow rate model, that even for a complex screened HS a significant fraction of the PM separation performance of the HS is si mple discrete (Type I) gravitational settling. Even with a simple overflow model, resolution of PM into a PSD allows gravitational settling to be identified as the most significant mechanism for coarse PM separation of a screened HS. Study results indicate that a screened HS is a preliminary unit operation that may have potential to protect downstream primary unit operations, units capable of more significant PM concentration reductions and finer PM capture. While the HS unit was evaluated under clean sump and volute conditions, the capture of PM by HS units require regular maintenance and management to prevent scour and mitigate the lab ility of separated coarse PM to repartition PMbound contaminants back to captured runoff stor ed in such units between runoff events.

PAGE 86

86 Table 3-1 Summary comparison of pilot-scal e measurements, CFD modeled results, and overflow rate (Q/A) model results utilizing either the event representative influent PSD (an NJCAT gradation), or utilizing the d50m(78.1 m) (median particle size of influent gradation based on mass) of ove rall % mass separated or effluent PM concentration [mg/L] by the screened HS (2.4 mm apertures) as a function of influent flow rate. The influent concentration and dry PM mass were 200 mg/L and 3145 g respectively for each pilot-scale run at each flow rate shown. % (%) of influent PM mass separated by the screened HS Effluent PM concentration [mg/L] Overflow Model (Q/A) Overflow Model (Q/A) Influent Flow Q (L/s) Measured CFD Model PSD d50m Measured CFD Model PSD d50m 1.59 65.8 71.3 100.0 100.0 64.5 63.2 0.0 0.0 3.97 64.1 65.5 81.8 100.0 68.0 75.9 40.0 0.0 7.94 62.2 58.8 73.7 100.0 92.4 90.6 57.7 0.0 11.92 47.9 50.0 63.7 100.0 90.1 109.9 79.7 0.0 15.89 41.3 46.4 56.5 100.0 125.3 117.8 95.7 0.0 19.87 36.5 42.2 52.5 79.5 117.6 127.0 104.4 45.2

PAGE 87

87 Figure 3-1 Full scale experimental setup for testing a screened HS (drawing not to scale) and computational grid of the HS (Inset). 5.486 m OD = 0.304 Sump Inlet reservoir (V = 37854L) 0.71 m Influent reservoir (V = 37854 L) Fl ow co ntr o l v al ve Effluent sampling 6.0 m 15cm Parshall flume (75 KHz ultrasonic) 0.76 m Pump (Capacity 69.39 L/s) 3.8 m Drop Box OD = 0.304 m 5.5 m IDs = 0.65 m Lx = 0.87 m OD = 0.304 m OD = 0.152 m Ly=0.76 m 0 1 2 0 1 2 ID v = 1.524 m Recirculation valve -1 0 Volute -2

PAGE 88

88 Particle diameter ( m) 1 10 100 1000 Percent finer by particle dry mass (%) 0 20 40 60 80 100 0.16 L/s 0.79 L/s 1.59 L/s 3.97 L/s 7.94 L /s 11.92 L/s 15.89 L/s 19.87 L/s 23.34 L/s 31.17 L/s 38.93 L/s Flow rate, Q (L/s) 010203040 Effluent d 50 ( m) 0 20 40 60 80 Influent d 50 (78.1 m)F l o w Q + 50 Figure 3-2 Observed phase shift in particle si ze distribution from influent to effluent via the screened HS. The influent gradation is that of the NJCAT protocol.

PAGE 89

89 Figure 3-3 Demonstration of grid independe nce. Results depict measured vs. modeled particle mass separated by the HS expressed as % fi ner by mass at an influent flow rate (Q) of 15.89 L/s, as a function of progressively finer meshes. RPD1= 25.77 Particle diameter ( m) 10 100 1000 % Finer by mass 0 20 40 60 80 100 Measured Modeled1 4e+05 cells Modeled2 1.96e+06 cells Modeled3 2.78e+06 cells RPD2= 2.25 RPD3= 5.29 Q = 15.89 L/sAbsolute RPD

PAGE 90

90 Figure 3-4 Measured vs. modeled Mparticles separated by the HS expressed as % finer by mass as a functoin of Q, [L3T-1] for a volute area diameter of 1.524 m, at an influent concentration of 200 mg/L. The mass-base d error associated with each measured PSD is in the range of 1 to 2% of the entire gradation mass and is distributed across the gradation; although the distri bution of error was not measured. Therefore the individual error bars are not shown. % Finer by mass 0 20 40 60 80 100 Measured Modeled % Finer by mass 0 20 40 60 80 100 Particle Diameter ( m) 10 100 1000 % Finer by mass 0 20 40 60 80 100 Q = 15.89 L/s Q = 19.87 L/s Q = 23.34 L/s Absolute RPD = 2.25 Absolute RPD = 6.64 Absolute RPD = 9.48 % Finer by mass 0 20 40 60 80 100 Measured Modeled % Finer by mass 0 20 40 60 80 100 Particle Diameter ( m) 10 100 1000 % Finer by mass 0 20 40 60 80 100 Q = 3.97 L/s Q = 7.94 L/s Q = 11.92 L/s Absolute RPD = 8.43 Absolute RPD = 9.52 Absolute RPD = 1.06 % Finer by mass 0 20 40 60 80 100 Measured Modeled % Finer by mass 0 20 40 60 80 100 Particle Diameter ( m) 10 100 1000 % Finer by mass 0 20 40 60 80 100 Q = 0.16 L/s Q = 0.79 L/s Q = 1.59 L/s Absolute RPD = 1.41 Absolute RPD = 3.81 Absolute RPD = 0.48

PAGE 91

91 Q (L/s) 04812162024 C effluent [mg/L] 0 25 50 75 100 125 0 25 50 75 100 125 [C] Measured Modeled M Measured Absolute RPD = 7.67 M particles (%)Absolute RPD = 8.40 Figure 3-5 Comparison of measured versus modeled results as a function of Q, [L3T-1] for noncohesive sandy silt influent particle gradation.

PAGE 92

92 Figure 3-6 Particle traj ectories calculated by a Lagrangian DPM for the screened HS, for an influent flow rate (Q) of 3.97 L/s. Plot (a ) (b) and (c) depict particle trajectories for particles with a diameter (dp) of 450 m, 75 m and 25 m respectively. Particle density( p) is 2.65 g/cm3. C Screen Volute Sump A Screen Sump B Screen Volute Sump

PAGE 93

93 Figure 3-7 Modeled velocity distri butions within the screened HS for an influent Q of 3.97 L/s. Plots (a) and (b) are particle velocities relative to fluid velocities integrated across the computational domain for a particle in jected at the geometric centroid of the inlet, for a particle with dp of 450 m and for a particle with dp of 25 m respectively. Plots (c) and (d) are frequencies of particle velocities in the screen area for particle diameters (dp) of 450 m and 25 m respectively. Plots (e) and (f) are frequencies of particle velocities in the volute area for dp of 450 m and 25 m respectively. Screen Normalized path length 0.00.20.40.60.81.0 Particle velocity (m/s) 0.00 0.03 0.06 0.09 0.12 Fluid 450 m Normalized path length 0.00.20.40.60.81.0 Particle velocity (m/s) 0.00 0.03 0.06 0.09 0.12 Fluid 25 m Screen V s,Newton V s,Newton Normalized frequency 0.0 0.1 0.2 0.3 0.4 0.5 450 m screen area Particle Velocity (m/s) 0.000.030.060.090.12 Normalized frequency 0.0 0.1 0.2 0.3 0.4 0.5 450 m volute area Normalized frequency 0.00 0.04 0.08 0.12 0.16 Particle velocity (m/s) 0.000.030.060.090.12 Normalized Frequency 0.00 0.04 0.08 0.12 0.16 25 m screen area 25 m volute areaV s,Newton V s,Newton V s,Newton V s,Newton (a) (b) (c) (d) (e) (f) 1 2 3 4 5 6 7 8 9 10 11

PAGE 94

94 CHAPTER 4 COMBINING PARTICLE ANALYSES AND CFD MODELING TO PREDICT HETERODISPERSE PARTIC ULATE MATTER FATE AND PRESSURE DROP IN A PASSIVE RAINFALL-RUNOFF RADIAL FILTER Introduction Anthropogenic particulate matter (PM) trans ported in urban rainfall-runoff has been identified as a significant contributor to ove rall deterioration of su rface water in the USA (USEPA 2000). Rainfall-runoff transports an en trained mixture of colloidal PM, non-colloidal PM, dissolved and complexed pollutants (Sansal one et al. 2007, Lee and Bang 2000, Sansalone et al.. 1998, Igloria et al.. 1997, Sansalone and Buchberger 1997). PM tr ansported by runoff act as reactive surfaces for adsorbing, desorbing and leaching organics and metals as well as phosphorus and other nutrients (Sansalone 2002). Separation of PM by unit operations and processes (UOPs) for in-situ treatment of rainfall-runoff is challenged by factors such as the stochastic nature of hydrologic a nd pollutant loads and concerns su ch as availability of land and infrastructural resources (Liu et al.. 2001). Empirical or bla ck-box approaches are based on gross assumptions and thus their application to treatment unit pr ototyping may mask the actual mechanistic behavior of a UOP. Various filtration configurations including infiltration and exfiltration through fixed granular medium (including soil) systems have b een suggested as viable solutions for meeting runoff quantity and quality regulations for watershe ds (Colandini 1999, Sansalone 1999, Li et al.. 1999, Colandini et al.. 1995, Geldof et al.. 1994, Schueler 1987). Recently, Hipp et al. (2006) suggested that removable filter in serts, which are mechanistically different from typical fixed bed granular medium filters, may be viable preliminary unit operations due to easier maintenance. While granular filtration has demonstrated advantag es for improving water quality, there is a requirement for careful design, analysis and prototype testing, as well as regular

PAGE 95

95 maintenance to ensure optimal hydraulics for qua ntity control (Keblin et al.. 1997). Performance (mass and size of PM separated) of a granular medium filter depends on the following broad parameters (Tien 1989). ),,,,,/( mmp particlesKddAQfM (4-1) In this expression,particlesM is the mass of particles separated, AQ/ is the influent surface loading rate, A is the total surface area of the media Q is the influent volumetric flow rate, is the effective porosity, pd is the mass-based particle diameter, md is the diameter of the granular media, is the empty bed contact ti me (EBCT) (AWWA 1990) and mK includes physical, chemical and surficial properties of the media, such as adsorption properties, media (internal) porosity, and morphology. Oxide coated media has been found to be effective at reducing turbidity, phosphorus and microbes in water and wastewater treat ment (Ayoub et al. 2001, Chen et al. 1998, Ahammed et al. 1996). Recently, the us e of oxide coated media has been extended to runoff treatment (Erickson et al. 2007, Sansalone et al. 2004). In this study, aluminum oxide coated media with a pumice substrate (AOCM)p was utilized for physical filtration. Granular filtration mechanisms have been classi fied into surficial straining, sedimentation, interception, inertial impaction, diffusion, hydrod ynamic and electrostatic interaction (Wakeman et al.. 2005). Filtration dynamics are widely assu med to be dependent on the surface loading rate (SLR) and with increasing SLR, macroscopic m echanisms tend to predominate the separation process, due to reduced contact tim es. The radial flow rapid-rate filter used in this study was operated at surface loading rates (SLR) ranging from 24 to 189 L/m2-min, which is higher than SLRs of typical rapid sand filters (83 L/m2-min) (Reynolds et al. 1995). Computational fluid dynamics (CFD) has found a pplications in environmental engineering in recent years, specifically in water and wast ewater treatment (Do-Qua ng et al. 1998). CFD is a

PAGE 96

96 method to solve the Navier-Stokes equations for coupled fluid and particle dynamics by a series of discretization techniques and algorithms (Anderson 1995). Tung et al. (2004) studied deep bed filtration for a sub-micron/nano-particle suspen sion utilizing a microscopic approach wherein different types of media packing were modeled. However, macroscopic approaches to modeling filter media are needed to model practical sy stems where packing schemes are most commonly random and unstructured. Li et al. (1999) applie d a 2-D numerical model to simulate variably saturated flow in a partial exf iltration system. Sansalone et al .. (2005) applied a 2-D numerical model to simulate the transien t hydrodynamics of a partial exfiltr ation system for rainfall-runoff clarification. However, a macroscopic CFD ba sed approach to simulate the 3-D hydrodynamic and clarification response of a granular medium filter is much needed in order to understand passive radial filtration, which may not be fully described by a simplified 2-D approach, due to lack of ideal symmetric flow conditions. Objectives CFD approaches to modeling granular filtrati on of rainfall-runoff are relatively new. Due to the water quality and hydraulic requirement s encountered in building rainfall-runoff PM control systems in urban areas, th ere is the opportunity to utilize recent advances to examine the behavior of rainfall-runoff filte rs. This study hypothesized that computational tools such as CFD, and measurement tools such as laser diffraction for PM analysis can provide examination and allow accurate prediction for separation of PM as a function of particle size and head loss behavior of filtration systems. Moreover, passiv e rapid rate filters have not been well understood mechanistically. The goal of this study was to demonstrate that a calibrated and validated CFD model can simulate the prototype performance of a passive radial cartridge filter (RCF) that utilizes (AOCM)p for treating a representative rainfall-r unoff particle size di stribution (PSD) and concentration. The efficiency of the RCF is hypothe sized to decrease as a function of flow rate,

PAGE 97

97 under maintainable head loss conditions for such a system deployed in the field. The first objective is to apply CFD to pred ict particle separation behavior and head loss response of the RCF through application of an appropriate flow and particulate phase m odel, within reasonable computational limits, to a scientifically acceptabl e degree of accuracy. The second objective is to compare the results of the numerical model to paired experimental results. A process of verification and validation was undertaken, in order to confirm the validity of the numerical model, as the third objective. The fourth objective was to examine the head-loss behavior and pressure distributions in the RCF based on the validated particle separation model. Methodology Experimental Setup Figure 4-1 illustrates the schema tic plan view of the instrumented RCF as part of the experimental setup. A storage tank with a capacity of approximately 40,000 L was utilized as a reservoir for influent potable water. A centr ifugal pump with a capacity of 414 L/min was used to generate influent flow. The flow measuremen t system consisted of a calibrated 5 cm multi-jet water velocity meter (DLJ Multi-jet). The flow control valve system facilitated flow rates ranging from 11.34 L/min to 90.7 L/min, which represent 11% and 114% of the design flow rate (Qd) respectively. The RCF was contained in a cyli ndrical test tank which served as a container to stabilize the inflow and ensure saturated fi ltration. The total volume of the RCF was 72.9 L and the additional volume of the test tank around the RCF was approximately 57 L to the top of the RCF. The diameter and height of the RCF are 0.4572 m and 0.5588 m, respectively. The media was contained between an outer mesh and non-reactive impervious polypropylene discs on top and bottom of the cartridge. Tank di mensioning was based on the representative elemental volume of an RCF under field depl oyment conditions in single or multiple applications. The scaled geometry of the RCF is shown in Figure 4-2.

PAGE 98

98 Media Characteristics Engineered (AOCM)p was utilized as the granular media. The distribution of media diameter (dp) was obtained by digital image analysis a nd utilization of Image Pro Plus v 6.2 Image Analysis Software. Image analysis results were utilized to calculate an equivalent circular diameter (ECD) as follows (Holdich 2002). The EC D is defined as the diameter of a sphere having the same projection area as the object. The resolution of the imaging method was 10.1 mega-pixels. /4 AECD (4-2) In this expression, A is the projected area of the media. Figure 4-3 illustrates a frequency histogram for the media tested. A three-paramete r Gaussian distribution was fit to the data (R2 = 0.93). The additional parameter accounts for the ar ea under the curve, which is not equal to 1. The mean media size is 3.56 mm 0.8 mm for a sample size (n) of 2551. The specific gravity of the pumice based media with was 2.35 (Farizoglu 2003). The media was pluviated into the cartridge so that the cartridge contained 49830.4 g of dry medi a at full capacity for each experimental run. Triplicate measurements of total clean bed porosity ( ) of the RCF with media produced a mean of 0.71 and standard deviation of 0.041 by volume. Of this total porosity the internal porosity of the medi a was measured in triplicate by mercury porosimetry (ASTM 2003) with a mean of 0.37 and a standard deviation of 0.022. The total porosity includes all available pore areas in the media and media bed, but not ne cessarily the effective po rosity available for a given flow rate. Prototype Test Procedure Each filtration experiment was started with a clean filter bed. As a secondary unit operation filters are operated under steady flows a nd steady flows were achieved through the use

PAGE 99

99 of the influent pump and the recirculation system to establish differing leve ls of influent flow. Flow control valves were calibrated to reach flow rates ranging from 0.16 to 23.36 L/s. Flow measurements were validated throughout each run and a cumulative volumetric balance was performed after each run. The PM suspension was prepared with Sil-co-Sil 106, a non-cohesive silt manufactured by US Silica and used as a common filtration test material. The influent particle gradation had a specific gravity of 2. 65 as determined by helium pycnometry. Figure 43 depicts the particle size distri bution (PSD) by mass of the influent; typical of a gradation that is largely settleable and suspended in rainfall-runoff (Sansalone and Kim 2007). The mass-based d50 of the influent par ticle gradation was 16.3 m. Over 85 % of the PSD by mass consists of settleable particles, as determined by laser diffraction. A slurry delivery system was de signed to inject the influent particle size distribution on a temporally consistent mass and granulometric basi s. The slurry delivery system consisted of a polypropylene slurry tank with a capacity of 20 L, an electronic mixer system in order to maintain the homogeneity of the influent suspensi on, and a variable flow-r ate peristaltic pump. The appropriate dry mass of particles to be adde d to the slurry tank and the rate of injection into the RCF was calculated depending on the influent flow rate being tested in order to deliver a constant nominal influent concentration of 200 mg/L (Sansalone et al. 1998, Sansalone 2002). Upon achieving the desired influent flow rate, par ticles were injected at the inlet of the RCF. 10 discrete duplicated 2 L influent and effluent samples were collect ed at regular sampling intervals designed for each individual experiment based on flow rate and run times. Laboratory Analyses For each run 10 discrete samples were subsequen tly combined to form duplicate (A and B) 20 L composites for water quality and suspended sediment concentrati on (SSC) analyses. SSC analysis was conducted in accordance with Standard Method 2540 D (ASTM 1997). PSDs of the

PAGE 100

100 influent and effluent were measured using a laser diffraction analysis based on Mie scattering theory (Finlayson-Pitts et al. 2000). The particle size range was from 1.0 m to 250 m, and was applicable to sediment concentration range of 1 to 1000 mg/L with a resolution < 1 mg/L. The particle separation efficiency of the filter is defined in te rms of the average event mean concentration (EMC) for PM over the time pe riod of the influent flow (Huber 1993). t ttq dttqtc V M EMC0 0)( )()( (4-3) In this expression, M is the total effluent mass load over the entire duration of the test, V is the total volume of flow over th e entire duration of the test, C is the flow weighted mean concentration, c (t) is the time variable particulate-bound concentra tion, q (t) is the time variable flow rate and t is time. The overall efficiency ratio was calculated as follows. EMCinlet average EMCoutlet average EMCinlet average ER (4-4) Urbonas (1995) suggested the following equation for calculating the % removal by using inflow and outflow loads. 100 ) ( ) () (1 1 1 INi n i INi EFFj m j EFFj INi n i INiCV CV CV PR (4-5) In this expression, Vi-IN and Vj-EFF are the volume of influent flow and effluent flow during the sampling periods i and j respectively; Ci-IN and Cj-EFF are average concentrations associated with periods i and j respectively; and n and m ar e the total number of influent and effluent measurements taken during event, respec tively. The effluent concentration (Ceff) and effluent

PAGE 101

101 particle mass load have been used to represen t the separation efficiency of the RCF. For the purpose of QA/QC, a mass balance erro r constraint of 10% was imposed. Mass Balance Error = %10100 ) ( i e RCF iM MMM (4-6) In this expression, iM is the influent mass of particles,RCFM is the mass of particles captured by the RCF and eM is the effluent mass of particle s, computed from the measured effluent SSC across the total treated volume. Al l gravimetric measurements were carried out on a dry mass basis. A maximum mass balance error of 10% was established for each experimental run. Pressure Head Measurements The variations in the height of the wate r column in the central discharge pipe (hi) of the RCF and the outer manometer (ho) was measured by the use of 1 psi pressure sensors manufactured by Druck Inc. Real-time da ta was acquired via a CR1000 datalogger, manufactured by Campbell Scient ific Inc. Thus, head loss ( H) in the radial direction was calculated as follows. iohhH (4-7) In this expression, h0 is the water column height in the outer manometer and hi is the water column height in the central discharge pipe, normalized to the same reference location. Computational Fluid Dynamics Model While the geometry of the RCF suggested a simplified two dimensional CFD model, the hydrodynamics and particle dynamics vary as a functi on of x, y and z spatial coordinates. As a result, a three dimensional (3-D) approach was required. The generalized 3-D scalar conservation equation for a control volume (CV) with volume )(zyxV is utilized. (Ferziger et al.., 2002)

PAGE 102

102 Sgraddivudiv t )()( )( (4-8) The RCF was thus modeled by adding a momentum sink term to the flow equations. This sink contributes to the pressure gradient in th e porous computational cell, creating a pressure drop that is proportional to the fluid velocity in a computati onal cell. The source term is composed of two parts: a viscous loss term and an inertial loss term. 3 1 3 12 1jj j ij jij ivvCvD S (4-9) In this expression, iS is the source term for the ith momentum equation, is permeability, iC2is the inertial re sistance factor, iv is the velocity in the ith momentum equation, v is the velocity in a computational cell. For isotropic porous media the above equation can be expressed in the following form. 2 22 1 vCvSi (4-10) In this expression, is the permeability [L2] and C2 is the inertial resistance coefficient. Modeling Flow in Porous Media The dominant flow regime is laminar to transitional in the RCF. The Reynolds number for flow in the media was calculated as follows. )1( Re sm mediavd (4-11) In this expression, md is media particle diameter, is the fluid viscosity, is the porosity of the packed bed (not media porosity), sv is the superficial veloc ity through the packed bed and is fluid density. The Reynolds numbers were found to range from approximately 5 to 30.

PAGE 103

103 Porous media may be modeled with a macroscopic or a microscopic approach (Ranade 2002). In the latter approach, the micro-scale pore structure is taken into account. A microscopic approach demands tedious computational resour ces and is typically used as a benchmark problem, as opposed to modeling practical systems. On the contrary, the macroscopic or lumped approach considers the entire filte r bed as isotropic or non-isotropic porous media. This approach is characterized by pertinent lumped para meters such as the effective porosity ( ), inertial and viscous resistance coefficients. The geometry of the media can be accounted for by utilizing bed porosity distributions, such as Muellers distribution (Mueller 1992), or by assuming homogeneous distribution of pores between medi a. With uniform pluviation in the RCF and uniform size gradation for the media a uniform po re distribution was assumed in this model. Previous studies have shown that the standard kmodel (Launder and Spalding 1974) for turbulent flow has worked well in both micro a nd macroscopic approaches to solving flow in porous media (Antohe et al. 1997). The tr ansport equations of the standard kmodel are expressed as follows. k bk jk t j i iSGG x k x ku x k t )() ( (4-12) For : S k CGCG k C x x u xtb k j t j i i 2 2 3 1) ( )() ( (4-13) In these expressions kG represents generation of k due to the mean velocity gradients; bG is generation of kdue to buoyancy; 1C ,2C and 3C are constants ;k and are the turbulent Prandtl numbers for kand respectively; kS and S are user-defined source terms. The constants were determined by Launder and Spaulding (1974). The values

PAGE 104

104 of1C ,2C ,3C ,k and used in the model were 1.44, 1.92, 0.09, 1.0 and 1.3 respectively (Launder et al.. 1974). The free surface of the flow was modeled as a shear-free boundary. Flow in porous media has been traditiona lly modeled analytically by comparison to pipe/conduit flow by specifying analogous parame ters such as the hydraulic diameter and roughness coefficient Laminar flow (Re < 10) through por ous media has been successfully modeled by applying Darcian-type equations. Models such as Blake-Plummer and CarmanKozeny equations were developed to account to transitional flow regimes. These models were extended by Ergun (1952) to account for turbulen t flow. The Ergun equation for packed beds applies to flow regimes from laminar to turbul ent and is expressed by the following equation. 2 3 3 2 2)1(75.1)1(150s m s mv d v d L p (4-14) In this expression, p is the pressure drop across the medi a, L is the length of the packed bed, is the fluid viscosity, md is media particle diameter, is the total porosity of the packed bed, vis the superficial velocity through the packed bed and is fluid densit y. Comparing equations 9 and 14, and C2 can be expressed as follows. 2 32)1(150 md (4-15) 3 2)1(5.3 md C (4-16) Modeling the Particulate Phase Brennen (2005) suggests a Eulerian-Lagrangian approach (van Wachem et al.. 2003) to model multiphase flows with a particulate vol ume fraction (PVF) less than 10% which is the case in this study (PVF << 0.1 %). In this appr oach, the flow field is first solved using a continuum approach and subsequently particles are tracked using a Discrete Phase Model

PAGE 105

105 (DPM). The DPM is derived from force balanc es based on Newtons (Turbulent and transitional regimes) and Stokes (laminar regimes) laws for particle motion, and summarized by the following equation. x p px p D pF g uuF dt du )( ) ( (4-17) DF 24 Re 182 pD ppC d (4-18) 2 3 2 1Re Rep p Da a aC (4-19) In the above expressions uis the fluid velocity, pu is the particle velocity, is the fluid density, p is the particle density, pd is particle diameter, is the viscosity, 321,, aaa are empirical constants that apply to smooth spherical particles as a function of the Reynolds number (Morsi et al.., 1972) andpReis the particle Reynolds number. Sub-spherical silica particles of a specific gravity of 2.65 were injected at regular intervals across the entire cross section of the inlet to the RCF. Neutrally buoyant tracer particles were utilized to calculate an unb iased particle tracking length. Particles that remained in th e RCF system after integrating over the specified length were considered to have been separate d. Particle removal was defined by the following equation. 100 I RCFN N P (4-20) In this expression, RCFN is the number of particles that remain in the RCF, and INis the number of particles injected at the inlet.

PAGE 106

106 Discretization and Solution Schemes The computational domain was discretized usin g an unstructured mesh with tetrahedral elements, generated by TGrid (Qi et al.. 2006). Numerical solutions were obtained using a Finite Volume Method (FVM) and a cell-centered scheme used for discretizat ion. A Second-Order Upwind Scheme (Barth et al.. 1989) was used solve for flow parameters. The SIMPLE (SemiImplicit Method for Pressure Linked Equati ons) algorithm (Patankar 1980) accounted for pressure-velocity coupling. The criteri on for iterative convergence was set at 1x10-3 (Ranade 2002). Results and Discussions Experimental Results Figure 4 illustrates the phase shift in mass-based PSD from influent (left) to effluent (right) as transformed by the RCF as a function of flow rate. The d50 of the effluent PSD is visibly finer than that of the influent, and the phase shift incr eases with decreasing flow rate. This leads to the hypothesis that at lower flow rates and lower head required to facilita te complete radial filtration, gravitational settling is an important mechanis m of particle separation in the RCF system whether deployed as a single cartridge or in an vo lumetric filter system. Effluent concentrations ranged from 30 to 58 mg/L for flow rates ranging from 11.3 L/min to 90.1 L/min respectively, for a steady influent concentration of 200 mg/L as shown in plots A and B of Figure 4-5. The SSC removal results are presented in Table 4-1. Figure 4-6 illust rates the overall clean bed head loss for the RCF as a function of flow rates. The head loss ranged from 1.2 cm to 3.9 cm for flow rates of 22.7 L/min (SLR = 48 L/m2-min) to 90.8 L/min (SLR = 189 L/m2-min). The maximum standard deviation was 1.9 cm for a flow rate of 79.5 L/min. These head loss values are relatively low due to the media d50, the reasonable media uniformity and the clean bed total porosity of 0.71.

PAGE 107

107 CFD Model Results Grid independence was achieved for the solutions at a mesh size of 3.01 million cells. Therefore model results were run with a mesh size of 3.01 million cells. Modeled results were compared with experimental data by means of an absolute relative % difference (RPD). Absolute RPD was calculated from the following equation. 100* data Measured data) Modeled-data (Measured (%) RPD Absolute (4-21) Particle Separation In order to interpret the overall PM separa tion (as SSC) by the RCF, two approaches were followed, the first is a comparison on the basis of effluent concentratio ns, and the second is a comparison on the basis of effluent mass load. Figure 4-5 is a comparison of measured and modeled clarification response of the RCF. Plot A is a comparison of measured and modeled effluent concentrations. The range bars on the m easured data represent the standard deviation between replicate samples. Plot B compares modeled and measured effluent gravimetric dry mass loads. Range bars are used to include the overall mass balance error (MBE) from the experiment. In summary, there is good agreement between measured and modeled data. The absolute RPD is well under 5% in both cases. All experimental runs yielded a mass balance error that was less than 10% with the recovery of all particles filtered, settled or eluted. Modeled and measured effluent PSDs are compared in Figure 4-7 for flow rates ranging from 11.4 L/min to 90.4 L/min. Model results are in good agreement with measurements as a function of dp ( Maximum RPD < 4 %). The three lower flow rates on the left in Figure 4-7 correspond to a range of Reynolds numbers that indicate a predominantly laminar flow regime, while the three higher flow rate s on the right in Figure 4-7 co rrespond to a transitional flow regime. One important aspect of the model is th at it accounts for the vari ation of porosity as a

PAGE 108

108 function of flow rate by implicit ly accounting for the available flow path across a given section of the porous media. This is due to the solution of the complete N-S equations across the entire computational domain, as opposed to black -box type calculations. The modeled and experimental results agree demons trating that finer particles are not separated by the RCF, with increasing flow rate. Head loss and Pressure Distributions Figure 4-6 compares measured and modeled head loss in the radi al direction of the filter ( H). Results indicate agreement between meas ured and modeled head loss (absolute RPD = 9.1%) illustrating that the model is capable of predicting head loss despite representing the media with a spherical geometry. This can be attrib uted to the predominantly laminar to mildly transitional nature of the flow, where the visc ous resistance component is negligible (Darby 2002). With the aid of the calibrated and vali dated CFD model, pressure distributions represented as head loss along the radial and vertical directions were computed and are illustrated in Plots (A) and (B) respectively of Fi gure 4-8. The head loss varies inversely with distance from the center of the RCF as illustrated in Plot (A). This variation displays a linear trend with the slope increasing as a function of flow rate. The ove rall head loss across the entire radial distance is equal in magnitude to the overall modeled head loss. Plot (B) illustrates pressure distributions along a lin e parallel to the z-axis and located at the geometric mid-point of the RCF in the radial direction, at one-half the annular thickness of the media section of the cartridge (r/2). Results indicate that for low flow rates the pressure distribution is fairly uniform. However, beyond a flow rate of 45.4 L/min, there is a parabolic profile for pressure distributions as a function of depth, with the minimum pressure at the top and bottom ends (no-flow boundaries) of the RCF. A check was performed for conservation of

PAGE 109

109 energy across the entire RCF using Bernoullis theorem, with local velocities obtained from the CFD model. This check supporte d the validity of the modeled distributions. The pressure distributions may be explained as follows. The center effluent outlet of the RCF never flowed full and remained open to the atmosphere. Therefor e, the RCF configuration is analogous to an inverted well, with flow pumped out in the direction of gravitational force. At the top surface, the parabolic profile of the drawdow n is due the head-loss created by the porous media. At the bottom of the RCF, there is also a pressure differe ntial from inside to outside of the cartridge. This is in line with typical drawdown in pumped wells as expressed by the Theis equation (Freeze et al.. 1979). The maximum differential occurred at 90.9 L/min and is 0.3 kPa. CFD was used to examine not only the pre ssure distributions but also the hydrodynamic and particulate loadings under which the RCF was tested and is operated given that manufacturers of such systems claim a uniform pr essure distribution and treatment from top to bottom of the RCF. Figure 4-9 provides a 3-D picture of particle dynamics within the RCF system for a flow rate of 45.4 L/min. Results in dicate that the overall surface area of the filter is not loaded uniformly, even in a single cartridge configuration. For instance, sediment-sized particle (dp of 75 m) results in Plot (D) demonstrate th at more than 75% of the overall RCF surface area is not being u tilized. With decreasing particle si ze, a larger portion of the overall available surface area of the filter is being utilized. Plot (B) shows the trajectories of a suspended particle (dp=10 m), the particle is strongly coupled with the hydrodynamics depi cted in plot (A) and a larger portion of the filter surface is available for filtration. However, as illustrated in Plot (C), for a slight increase in particle diameter (dp = 25 m), approximately 50% less surface area is utilized. This provides insi ght into the actual mechanisms in the RCF system. Gravitational effects are important even for this fine PSD and uniform filtration with depth does not occur. At

PAGE 110

110 low flow rates, gravitational settling is the pred ominant mechanism of particle separation in the system whether deployed as a single cartridge in a tank or in an volum etric filter system of multiple filters. Results also indicate that depe nding on flow rate and particle size, a significant fraction of the RCF surface area may not be utilized and not utilized uniformly. Pressure or head distributions along the axes of a granular filter are often difficult to measure and are often approximated based on various assumptions. These assumptions range from Darcian flow regimes to modeling flows with ca pillary tube analogs as in the Carman-Kozeny model. With the aid of the calibrated and validated CFD model, pressure distributions represented as head loss along the radial and vertical dire ctions were computed and this served to better understand the mechanistic behavior of the porous bed. Conclusions This study applied laser diffraction measurements of particle size distributions (PSDs) and the principles of CFD using a Finite Volume Method (FVM) to model the particle separation behavior of a passive radial cartridge filte r (RCF) using aluminum oxide coated media (AOCM)p,. The CFD approach employed a standard kturbulence model to resolve turbulent flow and a Lagrangian discrete phase model (DPM) to track particles. The uniform subspherical porous media with a d50 of 3.56 mm, a total bed porosity of 0.71 and a media (internal) porosity of 0.37 was modeled by adding a sink term to the momentum equations. The loading to the RCF was a fine silt gradation with a d50 of 16.3 m and a concen tration of 200 mg/L and flow rates ranging from 11.4 L/min (SLR= 24 L/m2-min) to 90.8 L/min (SLR = 189 L/m2-min). The model was validated with data across a range of flow rates, PSDs, and head-loss distributions. Predictions from the CFD model were within 10% of measurements and the CFD solution was grid independent.

PAGE 111

111 This study demonstrated that a CFD model c ould reproduce the fate of a hetero-disperse PSD as a function of particle size for mass, concentration and head loss behavior of a RCF for typical runoff loading rates, PSDs, and susp ended sediment concentration (SSC) levels. A steady state solution was obtained for flow through the RCF and clean bed conditions were implemented in the experiment and in th e model. CFD predictions provided an in-depth insight into the mechanistic behavior of th e RCF by means of three dimensional hydraulic profiles, particle trajectories and ra dial and axial pre ssure distributions. Results demonstrate that while the overall RCF particle removal for the silt sized heterodisperse influent gradation (SSC = 200 mg/L) ranged from 88% (effluent SSC = 66.7 mg/L) to 72% (effluent SSC = 50.2 mg/L) for flow ra tes ranging from 11.4 L/min to 90.9 L/min. The maximum head loss did not exceed 5 cm even at the highest flow rate for the porous (AOCM)p with a mean diameter of 3.56 mm. The low head loss allows a smaller media diameter to be utilized that would further reduce effluent concentrations at the expense of a higher head loss. Pressure distributions obt ained from the CFD indicate that effects of gravity on particle trajectory and separation are significant across the en tire range of flow rates, specifically for the settleable fraction of PM. Results demonstrate that the pressure distribu tions and loadings were not uniform with depth in the RCF system. The ability of the CFD model to reproduce tr eatment results across a range of flow rates and PSDs suggest that the cal ibrated/validated model can serve as a foundation upon which design alternatives can be proposed. A calibra ted/validated CFD-based iterative approach to design of this RCF as a unit operation has the pote ntial to provide reduced prototyping costs with improved performance, as a result of carefully designed experimental matrices, focused on PM control requirements for effluent discharges.

PAGE 112

112 This study combined PSD measurements using laser diffraction, media porosimetry, image analysis and material balances as well as more conventional gravimetric SSC measurements of PM and pressure sensor measurement. Such da ta are needed in the calibration and validation process for a defensible porous media CFD model of a RCF.

PAGE 113

113 Table 4-1 Summary of SSC results for RCF tested with a Sil-co-Sil 106 gradation at a nominal concentration of 200 mg/L. EBCT is th e mean fluid empty bed contact time. Q Surface Loading Rate Influent SSC Effluent SSC SSC Removal Fluid EBCT (L/min) L/m2-min [mg/L] [mg/L] (%) min 11.4 24 202.8 50.2 86 4.42 22.8 48 203.7 54.4 85 2.21 34.2 72 195.5 52.5 80.7 1.47 45.6 95.4 193.4 64.7 78.4 1.11 68.4 142.8 202.4 69.3 76.7 0.74 79.2 165.6 202.4 68.8 75.1 0.64 90.6 189.6 209.3 66.7 71.4 0.56

PAGE 114

114 Figure 4-1 Process flow diagram/ experimental setup for steady flow operation of the radial filter cartridge (not to scale). PM slurry tank [C = 200 mg/L] Mixer Centrifugal pump (Capacity: 6.9 L/s) 5 cm velocity water meter (DLJ multi-jet) Influent reservoir (V=40,000 L) Variable flow peristaltic pump Constant head tank Filter Cartridge (V = 72.9 L) Effluent Influent

PAGE 115

115 Figure 4-2 Profile view of the radial cartridge filter (RCF) apparatus. Z X X (m) Inlet Filter cartridge 0 0.15 0.30 0.45 0.60 0.75 0.90 1.05 1.20 0 0.4 0.2 -0.2 -0.4 Z (m) Outlet H

PAGE 116

116 Figure 4-3 Measured media size expressed as a Gaussian frequency histogram. The average media size was found to be 3.56 .8 mm. A three parameter Gaussian distribution was used to model th e data; a = 5.283, b= 0.6392. x0=3.435. dp (mm) 01234567 Normalized Frequency 0 2 4 6 8 R 2= 0.93 2 ))(5.0(0*b xxeay

PAGE 117

117 Figure 4-4 Observed phase shift in particle si ze distribution from influent to effluent via the radial cartridge filter (RCF). Particle diameter ( m) 1 10 100 1000 Percent finer by particle dry mass (%) 0 20 40 60 80 100 Influent 11.4 L / min 22.8 L / min 34.2 L / min 45.6 L / min 68.4 L / min 79.8 L / min 91.2 L / min Q + 50 Influent Effluent RCF

PAGE 118

118 Q (L/s) 0153045607590 C effluent [mg/L] 0 25 50 75 100 0 25 50 75 100 [C] Measured Modeled Measured Absolute RPD = 2.1 M particles (%)Absolute RPD = 6.25 Figure 4-5 Comparison of measured versus modele d results as a function of influent flow rate (Q) for the non-cohesive influent particle gradation for an influent concentration of 200 mg/L. Absolute RPD is the absolute relative % difference be tween experimental and numerical model data.

PAGE 119

119 Figure 4-6 Comparison of measured versus modeled filter head loss ( H) as a function of influent flow rate (Q) for the non-cohesi ve influent particle gradation at an influent concentration of 200 mg/L; Error bars represent the standard deviation between replicated measurements. Q is the influent flow rate, [L3T-1]; Absolute RPD is the absolute relative % differe nce between experimental and numerical model data. Q (L/min) 020406080100 0 5 10 15 20 Measured Modeled H (cm)Absolute RPD = 9.1

PAGE 120

120 % Finer by mass 0 20 40 60 80 100 % Finer by mass 0 20 40 60 80 100 Particle Diameter ( m) 10 100 1000 % Finer by mass 0 20 40 60 80 100 Q = 11.4 L/min Absolute RPD =0.08 Measured Mass = 141 g Modeled Mass = 141.1 g Q = 22.7 L/min Absolute RPD = 1.17 Measured Mass = 116.5 g Modeled Mass = 117.7 g Q = 34.1 L/min Absolute RPD = 1.95 Measured Mass = 149.6 g Modeled Mass = 152.5 g Figure 4-7 Measured vs. modeled particle mass in the effluent of the RCF expressed as % finer by mass at influent flow rates (Q), [L3T-1] of 11.4 L/s, 22.7 L/s, 34.1 L/s, 68.1.4 L/s, 79.5 L/s and 90.9 L/s at an influent concentration of 200 mg/L. Absolute RPD is the absolute relative % difference on a mass basis between experimental and numerical model data. % Finer by mass 0 20 40 60 80 100 % Finer by mass 0 20 40 60 80 100 Particle Diameter ( m) 10 100 1000 % Finer by mass 0 20 40 60 80 100 Q = 68.1 L/min Absolute RPD =2.6 Measured Mass = 179.4 g Modeled Mass = 184.1 g Q = 79.5 L/min Absolute RPD = 3.65 Measured Mass = 194 g Modeled Mass = 186.9 g Q = 90.9 L/min Absolute RPD = 3.84 Measured Mass = 221.8 g Modeled Mass = 230.3 g Measured Modeled

PAGE 121

121 Figure 4-8 Head loss ( H) and pressure distributions in the RCF. Plot (A) illustrates predictions of head-loss across the RCF from the CFD model. Plot (B) illus trates predictions of pressure distributions along the vertical (Z) axis of the RCF. Q1, Q2, Q3, Q4, Q5, and Q6 represent flow rates of 22.7, 34.1, 45.4, 68.1, 79.6, and 90.9 L/min respectively. The order of flow rates in (B) is the same as in Plot (A); (r, ) and (0, ) represent the radius and center of the RCF respectively, in radial coordinates; (r/2, 0, zn) and (r/2, 0, zo) represent the top and the bottom of the mid-point of the RCF respectivel y, in Cartesian coordinates Radial distance (cm) 05101520 H (cm) 0 1 2 3 4 5 6 (r, ) (0, ) (r/2, 0, zn) Cartridge depth (cm) Q6 Q5 Q4 Q3 Q2 Q1 Q+ (A) Q+ (B) 0 0.05 0 10 20 30 40 50 0.10 0.15 0.20 0.25 0.30 (r/2, 0, z0) Gage Pressure (kPa)

PAGE 122

122 Figure 4-9 CFD predictions of tr ajectories of fluid and particle s inside the RCF, shaded on the basis of residence time, [T]; Plot (A) describes the pa thlines of fluid particles of negligible mass; Plots (B) through (D) desc ribe the pathlines of inert spherical particles of diameters (dp) 10 m, 25 m and 75 m respectively. Particle density( p) is 2.65 g/cm3.Qd is 45.4 L/min. (A) Fluid pathlines (B) dp = 10 m Influent Effluent Influent Effluent Influent Effluent Influent Effluent (C) dp = 25 m (D) dp = 75 m Filter media Filter media (s) (s)

PAGE 123

123 CHAPTER 5 MODELING HYDRAULICS AND PARTICLE DYNAMICS OF A STORMWATER HYDRODYNAMIC SE PARATOR FOR TR ANSIENT INFLUENT LOADS Introduction Anthropogenic particulate matter (PM) trans ported in urban rainfall-runoff has been identified as a significant contributor to ove rall deterioration of su rface water in the USA (USEPA 2000). Rainfall-runoff transports an en trained mixture of colloidal PM, non-colloidal PM, dissolved and complexed pollutants (Sansal one et al. 2007, Lee and Bang 2000, Sansalone et al.. 1998, Igloria et al.. 1997, Sansalone and Buchberger 1997). The temporal particle size distribution (PSD) and chemical composition of particulate matter (PM) delivered by runoff from a rainfall-runoff event varies significantly betw een different geographical regions and even with spatial variation in the same watershed. (Sansalone 2002). A major issue that needs to be addressed while designing stormwat er unit operations and processes (UOPs) is the dearth of available land for construction, especially in highly polluted urban areas. In light of this, many new devices have been introduced, which have the advantage of a small-footprint and ease of new installation or retrofit to existing infrastructure. Hydrodynamic separators (HS) are one such cl ass of UOPs that broadly rely on centrifugal forces in addition to gravitational force to separa te particles (Brombach et al.. 1987, Brombach et al.. 1993, Pisano et al.. 1994, USEPA 1999, Andoh et al.. 2003). An attractive feature of a HS is the potential for longer particle trajectories per given unit surface area in comparison with traditional unit operations and processes (UOPs) such as settling basins. A screened HS is a variant on the HS principle, and uses the comb ined separation mechanis ms of vortex induced inertial separation, screening and se dimentation and has been used in recent years as a device for treating stormwater runoff (Rushton 2004, Rush ton 2006), and oil/grease removal (Stenstrom and Lau 1998).

PAGE 124

124 This study examined a simple screened HS consisting of two cylindrical chambers as illustrated in plots (a) and (b) of Figure 5-1. The flow inlet is tangential to the inner cylindrical chamber. The chambers are separated by a static screen. The static screen consists of a regular array of apertures. The inner cylindrical chamber, along with th e screen and the sump chamber are designated as the screen area. The screen apertures allow vortex flow to exit the inner screen chamber and enter the outer volute chambe r. The geometry of the screen results in a weakly reversed flow direction in the outer annu lar area, termed the vol ute area. Particle separation by this HS configura tion is a function of various pa rameters, as expressed in the following equation. ),,,,,,(strps particlesdvvQdvfM (5-1) In this expression,particlesM is the mass of particles separated, sv is the discrete particle settling velocity, pd is the mass-based particle diameter, Q is the influent volumetric flow rate, rv is the radial velocity com ponent in the screening area, tv is the tangential velocity component in the screening area, is the hydraulic residence time, and sd is the diameter of the screen apertures. The design flow rate (Qd) for the screened HS used in this study is 9.5 L/s. Typical stormwater design approaches are exte nsions of wastewater tank design principles, which assume ideal to predictable influent quantitative and qualitative loads. It is clear that the rainfall-runoff process and the associated partic ulate and dissolved matter delivery is a highly variable process and therefore unit operations for stormwater PM management have to be designed to operate across rapi dly changing flow and partic le concentrations. Existing stormwater models such as the stormwater management model (SWMM) (Rossman 2007) use idealized influent hydrographs and pollutogr aphs. While these have greatly improved the efficacy of design considering the complex and inter-related design parameters, they are not

PAGE 125

125 equipped to provide an in-depth hydrodynamic and particle clarification profile of the UOP for transient loads. Existing literature does not provide a detailed and fundamental insight into the functioning of hydrodynamic separators. Previous studies have ranged from simplified overflow rate theory (Weib 1997) to semi-empirical approaches based on broad assumptions of vortex flow behavior (Paul et al.. 1991). This lack of information leads to difficulties in scaling and implementation of these devices. Fenner et al.. (1997, 1998) report th at similitude analysis does not yield a single dimensionless group that can be used in scale-up of a HS. Computational fluid dynamics utilizes numer ical methods to solve the fundamental equations of fluid dynamics, i.e. the Navier-Stokes equations (Versteeg et al.. 1995). The applicability and efficacy of CFD techniques have closely followed corresponding leaps in computational power. While traditionally be ing in the realm of Aerospace and Process engineering, CFD is now used in design and optimization of UOPs in environmental engineering. CFD approaches to model particle-laden flows is an active research area (Curtis et al.. 2004, van Wachem et al.. 2003), including hydr odynamic separators. Faram et al.. (2003) utilized a Reynolds Stress Model (RSM) to resolve turbulent flow and a Lagrangian Discrete Phase Model (DPM) to model behavi or of the particulate phase. Okamoto et al.. (2002) studied particle separation by a vo rtex separator using a kmodel to resolve flow and an Algebraic Slip Mixture (ASM) model for the particulate phase. Li m et al.. (2002) used CFD-predicted velocity profiles to study behavior of flocs in a vortex separator, with a Renor malization Group (RNG) kturbulence model. Tyack et al.. (1999) comp ared measured velocity profiles in a vortex separator to those predicted by a Renormalization Group (RNG) kturbulence model.

PAGE 126

126 It should be noted that albeit studies dealing with the application of CFD in understanding the multiphase dynamics that occur in a hydrodynamic separator are few in number, this is clearly not the case with wastewater unit operations such as sett ling tanks (Jayanti et al. 2004, Naser et al. 2005, Deininger et al 1998, Brouckaert et al. 1999, Br estscher et al. 1992, Zhou et al. 1994, Szalai et al. 1994, Ka rl et al., 1999) and for CSO pollu tion abatement devices such as storage chambers (Stovin et al., 1996, 1998, 2002). An experimentally validated CFD approach was used previously to accurately describe the performance of a screened HS, across a range of steady flow rates and ge ometric configurations for an influent of known particle size distribution at a concentration typically observed in realtime storm events. The next step lies in extendin g this model to real-time rainfall-runoff events with coupled and transient flow rates, influent particle size distribu tions and influent PM concentrations. Objectives The first objective of this study was to mode l the overall particulate matter separation across a real-time rainfall-runoff event for a hydrodynamic separator by th e application of an appropriate turbulent flow mode l. The subsequent objective was to compare predicted results with measured PM removal data. The applicab ility of the model across different transient quantitative and qualitative loads was to be test ed as the third objective. Typical stormwater basin designs are based on influent flow rates calculated from re turn storms, and on event mean concentrations (EMC) of water qua lity parameters. Therefore, as the fourth and final objective, this approach was compared to the real-time performance of the UOP by a CFD simulation with the same model, for a steady influent flow rate, for the EMC of the PM and the event-mean (representative) influent PSD.

PAGE 127

127Methodology Experimental Methodology The experimental site is illustrated in Figure 5-2. The catchment area was a stretch of elevated urban highway (Interstate-10) over Ci ty Park Lake in Baton Rouge, Louisiana, constructed from Portland cement concrete (PCC ). The overall drainage area was approximately 1088 m2. A detailed description of the catchment is available elsewhere (Sansalone et al.. 2005). The overall experimental setup consists of the following components a system of pipes and troughs to deliver rainfall-runoff from the catchment a 5.08 cm (2 inch) Parshall flume equipped with a 70 KHz ultrasonic sensor for measuring flow rates, a gate valve to divert flow to the screened HS. Four discrete rainfall-runoff events were monitored and treated by the screened HS. All storms were routed through a pre-cleaned syst em in order to accurately monitor PM mass balance. The static screen aperture size used wa s 2.4 mm for all storms with the exception of the 20 Aug 2004 storm, where an aperture size of 1.2 mm was used. Sampli ng started immediately when runoff was observed, and was performed on a flow-weighted basis, across the entire duration of the rainfall-runoff event. Influent was sampled at the drop-box upstream of the screened HS and effluent was sampled at th e outflow of the HS. All samples were taken manually, across the entire cross-section of the outf all, in order to obtain a truly representative and complete particle size distribution. The number of discrete samples of influent and effluent was chosen appropriately to provi de a reasonable estimate of tem poral particulate concentrations (McBean et al.. 1997). The total particulate matter was obtained as a sum of the mass of sedi ment, settleable and suspended particle fractions. Sediment particles ar e defined as particles with a diameter greater

PAGE 128

128 than 75 m (ASTM 1993). The settleable fraction (25 m
PAGE 129

129 In this expression, iM is the influent mass of particles,HSM is the mass of particles captured by the screened HS and eM is the effluent mass of particles, computed from the measured effluent SSC across the total treated volume. All gravimetric measurements were carried out on a dry mass basis. A maximum mass balance error of 10% was established for each experimental run. The event mean concentration (EMC) for PM over the time period of a rainfall-runoff event is typically used as an indicator of ove rall water quality and to calculate % removal for stormwater UOPs (Huber 1993). t ttq dttqtc V M EMC0 0)( )()( (5-4) In this expression, M is the total effluent mass load over the entire duration of the test, V is the total volume of flow over th e entire duration of the test, C is the flow weighted mean concentration, c (t) is the time variable particulate-bound concentra tion, q (t) is the time variable flow rate. Multiphase Flow Modeling Methodology The behavior of the screened HS is modeled more accurately in three dimensions, due to the simultaneous effects of the complex sta tic screen geometry, the vortexing flow and gravitational forces on the instantaneous motion of particles. The fluid fl ow equations that are solved by CFD are based on fundamental laws of conservation of mass, momentum and energy. CFD provides a group of numerical techniques to solve non-linear partial differential equations of flow, i.e. the Navier-Stokes equations. Versteeg et al. (1995) define the general conservation equation for any fluid property for a control volume )( zyxV is as follows.

PAGE 130

130 Sgraddivudiv t )()( )( (5-5) In this expression, u is the fluid velocity, is the diffusion coefficient, and Sis the source/sink term. z w y v x u udiv ) ( (5-6) In this expression, u, v and w are the velo city vectors in the x, y and z directions respectively. The momentum equations for three dimensions are obtained by incorporating u, v and w into equation 2. X-momentum: MxSugraddiv x p uudiv t u ))(( )( )( (5-7) Y-momentum: MySvgraddiv y p uvdiv t v ))(( )( )( (5-8) Z-momentum: MzSwgraddiv y p uwdiv t w ))(( )( )( (5-9) In the above expressions, p is the pressure, is the viscosity, and MzMyMxSSS are source/sink terms to account for surface forces such as viscous and pressure forces, and body forces such as gravitational and centrifugal forces in the x, y and z directions respectively. The inlet Reynolds number was found to vary significantly as a function of time for the four rainfall-runoff events. There was high tem poral variability (1e+ 02 to 9e+05) in the magnitude of Reynolds numbers calculated for the 20 Aug 2004 rainfall-runoff event. On the other hand, the magnitude and temporal variation of Reynolds numbers was low for the 14 Oct 2004 storm (1e+02 to 2e+03). Therefore, there is the need for a turbulence model that can provide stable and accurate flow simulations across widely ranging and often rapidly changing inlet Reynolds number.

PAGE 131

131 The standard kmodel was used to resolve turbulen t flow for the HS, based on previous studies with hydrocyclones (Nowakowski et al.. 2 004, Statie et al.. 2001, Petty et al.. 2001) and the screened HS at steady flow rates. Mohammadi et al. (1994) pr ovide a detailed analysis into the kmodel and confirm the applicability across a wi de range of flow regimes. In the standard kmodel (Launder and Spalding 1974) for turbulen t flow, a closed solution is obtained for the turbulent transport equations by relating Reynolds stresses to an eddy viscosity (). Newtons law of viscosity is applied to illustrate the relationship between viscous stresses and Reynolds stresses. It should be no ted that eddy viscosity (t) is a non-physical quantity, and is expressed by the following equation. 2k ft (5-10) In this expression, kis the turbulent kinetic energy per unit mass, calculated as follows. ) (5.0222wvuk [L2T-2] (5-11) In this expression wvu ,are vector components of velocity fluctuations due to turbulence and is the dissipation rate of tu rbulent kinetic energy per unit mass. The transport equations of the standard kmodel are expressed in the following equations. For k: k bk jk t j i iSGG x k x ku x k t )() ( (5-12) For : S k CGCG k C x x u xtb k j t j i i 2 2 3 1) ( )()( (5-13) In these expressions kG represents generation of k due to the mean velocity gradients; bG is generation of kdue to buoyancy; 1C ,2C and 3C are constants ;k and are the

PAGE 132

132 turbulent Prandtl numbers for kand respectively; kS and S are user-defined source terms. The constants were determined by Launder and Spaulding (1974). It was hypothesized that isotropy of Reynolds stresses can be assumed r easonably in the case of the HS. The values of1C ,2C ,3C ,k and used in the model were 1.44, 1.92, 0.09, 1.0 and 1.3 respectively (Launder et al.. 1974). The no-slip boundary conditi on, along with effects of viscous blocking and kinematic damping creates large gradients in solution variables near the walls of the HS. To account for these phenomena in the turbulence mode l, semi-empirical wall functions for the kmodel were used (Launder et al.. 1974). The shape of the screen apertures results in a weakly reversed flow direction in the outer volute chamber. The screen area open to flow is designed so that radial velocity through the screen is approximately an orde r of magnitude lower than the inlet velocity. Approximately 40 % of the screen area is effectively open to flow. Meshing the static screen, with structured or unstructured meshing schemes places severe constr aints on time and computational power. To overcome this difficulty, the static screen was mo deled as a porous perforated plate with the addition of a momentum source term to the standa rd fluid flow equations. This momentum sink contributes to the pressure grad ient in the porous computationa l cell, creating a pressure drop that is proportional to the fluid velocity in the cell. The source term is composed of two parts: a viscous loss term and an inertial loss term. For simple homogeneous media, the sink term is expressed by the following equation. imag i i ivvCvS 2 12 (5-14) In this expression, iS is the source term for the ith momentum equation, is the permeability, iC2is the inertial re sistance factor, iv is the velocity in the ith momentum equation,

PAGE 133

133magv is the velocity magnitude in the computational cell. It is important to note that the primary assumption is that the porous cells are 100% open, i.e. the actua l geometry of the screen is represented by a resistance component. The pressure drop through the screen was meas ured experimentally for steady flow rates ranging form 1 to 150 % of the design flow rate w ith pressure transducers and it was found that the head loss was less than 150 mm even at high er flow rates, across the static screen. The viscous resistance term was assumed to be neg ligible and the source term was then modeled considering only the inertial resistance term, a nd the inertial resistance factor per unit thickness of the plate, i.e. )(2iC was calculated for the x, y and z dire ctions as described by the following equation (FLUENT 2005). n K CiL i)(' )(2 (5-15) In this expression, nis the thickness of the plate, and )(' iLK is a coefficient calculated from the following equation. 2 2 )()('* v v KKo iLiL (5-16) In this expression, 2xv is the square of the ve locity through the plate, considering that the plate is o % open to flow in the given direction and 2v is the square of th e velocity through the plate, considering that the plate is 10 0 % open to flow in the given direction.)(iLK is the loss factor in the ith direction and is calculated from the following equation. 2 )( oiLvKp (5-17) In this expression, p is the measured pressure lo ss through the stat ic screen. 2C is calculated based on the %ages of sc reen area open to flow described previously for the x and y

PAGE 134

134 directions. The inertial resistance in the z-directi on is theoretically infinite. However, to specify 2C in the z-direction, a value sufficiently large enough to not cause numerical instability (twice the order of magnitude of the largest value of2C) was used. Modeling the Particulate Phase Multiphase flows are modeled with an Eule rian-Eulerian approach or an EulerianLagrangian approach (van Wachem et al.. 2003 ) depending on the extent of coupling between phases. Elgobashi (1991) proposed a regime ma p for appropriating the degree of inter-phase coupling, by analyzing length and time scales. Subs equently, it was determined that a Lagrangian approach to tracking the secondary phase is mo st appropriate for flows with a low influent particulate volume fraction which is the case in this study (0.2 %< PVF < 3.2 %). The transient flow field was modeled utilizi ng the Eulerian approach, for each time step. Subsequently, the influent partic les were injected at times that correspond to influent sampling times during storm data collection. The Eulerian-Lagrangian approach was chosen to model the behavior of particles in the computational domain of the HS. In this approach the flow field was solved using the Eulerian approach. Following this, particles were track ed using a Lagrangian Discrete Phase Model (DPM). The DPM is derived from force balanc es based on classical Newtons (Turbulent and transitional regimes) and Stokes (Laminar regime s) laws describing par ticle motion, and is summarized by the following equation. x p px p D pF g uuF dt du )( ) ( (5-18) dF 24 Re 182pD ppC d (5-19)

PAGE 135

135 2 3 2 1Re Rep p Da a aC (5-20) uudpp p Re (5-21) In equations 5-18 through 5-21, uis the fluid velocity, pu is the particle velocity, is the fluid density, p is the particle density, pd is particle diameter, is the viscosity, 321,, aaa are empirical constants that apply to smooth spherical particles as a function of the Reynolds number (Morsi et al.., 1972) andpRe is the particle Reynolds number. Particle trajectories are obtain ed by integration of equation 21. For this study, particles of consistent morphology were inject ed across the entire flow crosssection immediately above the inlet to the HS to obtain comparable particle trajectories across varied flow rates, by reducing uncertainty in initial spatial location. Particle s were defined as silica particles of diameters associated with the sieves utilized and a meas ured specific gravity of 2.65 as determined by helium pycnometry (Sansalone et al. 1998). Particles were tracke d for a specific length for each flow rate, based on hydraulic residence times calculated by injecting neutrally buoyant submicron tracer particles. Particles that remained in the HS after integrating over the specified length were considered to have been separated by the HS. Particle removal was thus defined by the following equation. 100*I HSN N p (5-22) In this expression, HSN is the number of particles that remain in the screened HS, and INis the number of particles injected at the inlet.

PAGE 136

136Discretization and Solution Schemes The computational domain was discretized usin g an unstructured mesh with tetrahedral elements, generated by TGrid (Qi et al.. 2006). The volume of the HS was di vided into total of 3.86 million cells, after ensuring grid convergence. Numerical solutions were obtained using a Finite Volume Method (FVM) a nd a cell-centered scheme used for discretization. A SecondOrder Upwind Scheme (Barth et al.. 1989) was used solv e for flow parameters. A first-order fully implicit scheme was used for time discretization. The fully implicit scheme is unconditionally stable and is suggest ed for transient CFD operations which do not involve chemical reactions (Ranade 2001). The fl ow data were measured for discrete time intervals. In order to test th e sensitivity of the solution to smaller time steps, continuous functions were utilized to model the measured flow data. For simple hydrographs such as the 20 Aug 2004 event, a Weibull function was utilized For more complicated storms, the runoff hydrograph was modeled by splitting the hydrograph into piecewise continuous segments which allowed for accurate curve fitting. For exampl e, the 05 June 2005 storm was modeled with a combination of quadratic and e xponential functions All the storms were modeled successfully with a time step t of 30 seconds. Smaller time steps did not significantly affect the flow solutions. The transient SIMPLE (Semi-Implic it Method for Pressure Linked Equations) algorithm (Patankar 1980) accounted for pressure-velocity coupling. The criterion for iterative convergence was set at 1x10-3 (Ranade 2002). Results and Discussions The influent hydrographs and hyetographs of th e four discrete storm events are presented in Figure 5-3. The varying hydrology resulted in a distinct particle transport signature for each storm. Among the four storms analyzed, the screen ed HS had a total PM removal efficiency that ranged from 48% to 57%. Influent and effluent mass loads are pr ovided in Table 5-1. The total

PAGE 137

137 particulate clarification response of the screened HS is greatly dependent on the influent particle size gradation. Coarse particles are separated by gravitational settling at low flow rates, while finer particles are separated by vor tex-induced inertial separation, which is more effective at high flow rates close to the design fl ow rate of the screened HS. Figure 5-4 illustrates the modele d and measured gravimetric pa rticle size distributions of the total PM separated by the screened HS. The modeled results agree very well with the experimental data, across the four discrete rainfall-runoff events. The absolute relative % difference was utilized to compare measured a nd modeled results, and wa s calculated as follows. data Measured data) Modeled-data (Measured *100 RPD Absolute (5-23) CFD simulations reproduced the overall separate d particle size distri bution within 10 % of the measured value. The overall mass and relative % difference for measured and modeled PM is provided for each storm in Table 5-1. The temporal variations in the position of a given particle are illustra ted in Figure 5-5. The 20 Aug 2004 storm was chosen, and particle trajectories in the HS were obtained from postprocessing CFD results. The part icle diameter chosen was 300 m, which is the median particle diameter by mass (d50m ) of the influent PSD. The peak flow rate of the 20 Aug storm was twice in magnitude of the design operational flow rate. The hydraulic residence tim e is not sufficient to offset the effect of advective transport, and prom ote gravitational settling for the coarse particle. This type of illustration provides an in-depth in sight into the particle transport of any given particle of known specific gravity, within a turbulent flow regime. Bertrand et al.. (1998) classifi ed rainfall-runoff events into mass-limited and flow-limited events based on temporal pollutant mass deli very. For mass-limited events, mass delivery is skewed toward the initial portion of the event, while mass delivery tends to follow the

PAGE 138

138 hydrograph for flow-limited events. This classification was applied to the storms in this study, in order to facilitate a discussion of their PM clar ification responses to di fferent steady influent flow rates, chosen to test the practical usability of a supposedly representative singular value for influent flow rate, in lieu of simula ting an entire rainfall-runoff event. Figure 5-6 illustrates the result s of simulating different influent flow rates. The 20 august 2004 storm was a high intensity storm with the peak flow rate being close to twice the design flow rate of 9.5 L/s. At higher flow rates, th ere is enough driving force to create and sustain a forced vortex. This suggests the predominance of inertial or separation as opposed to gravitational settling. Moreover, there was a di sproportionate amount of mass delivered in the initial portion of the storm, when the flow ra tes were high. Further, the influent particle distribution was predominat ed by coarser particles (d50m = 300 m). Due to the aforementioned reasons, it is not surprising that the peak flow rate reproduces the overall PM separation better than the mean and median flow rates. The 03 October 2005 storm had a ma ximum flow rate that was close to 125% of the design flow rate. However, the mass delivery was proportiona l to the flow rate, ac ross the entire event, and the influent PSD was finer (d50m = 58 m). In light of this, there was no significant difference between PM separation predicted by the mean and median flow rates. However, there is a greater digression from measured PM removal utilizing the peak flow rate. The 05 June 2005 storm had a maxi mum flow rate that was almo st identical to the design flow rate of the screened HS. The partic ulate transport was mass-limited and the d50m of the influent was 247 m. With this coarser influent, we find that neither the mean nor the median influent flow rates are able to accurately re produce the particle dynamics of the screened HS across the storm. However, simila r to that observed w ith the 20 August 2004 storm, we find that

PAGE 139

139 the peak flow rate provides the best predic tion, thus bolstering the hypothesis of increased inertial separation as opposed to gravitational settling at flow rates close to or greater than the design flow rate. The 14 Oct 2004 storm was a very low intens ity and long duration storm. Under these conditions, the screened HS functions more as a circular sedimentation tank. However, the particle delivery was mass limited, owing to a dr y deposited particle gr adation that was fine enough to advect with the low intensity sheet flow on the paved surface of the watershed. However, we notice that the mean, median or peak flow rate fall short of predicting the actual PM performance. The reason for this can be attri buted to the fine influe nt particle gradation (d50m = 45 m), which still leaves a large fraction of the influent mass in the suspended particle size range, which are separated more as a function of velocity gradients than on increased hydraulic residence times. Table 5-2 provides a comparison of the PM separation at different flow rates. From a quantitative design point of view, it beco mes apparent that no one flow rate can be considered representative, as the predictions of the PM separation behavior of the UOP vary significantly using the mean, median and peak in fluent flow rates. The influent hydrographs range from slightly skewed Gaussian distri butions to extended ra ndom pulses across a long period of time. Any intuitive assumption that the median flow rate is likely to be most representative was proved to be inapplicable across stor ms with varying hydrographs. This ties in to the prevailing question of how much of a rainfall-runoff even t needs to be treated. While practical design initiatives call for specific qua ntitative and qualitative loads, the actual functionality of the UOP is heav ily influenced by the coupled vari ations in flow and influent particle size and mass distributions.

PAGE 140

140Conclusions Particulate matter separation by a screened hydrodynamic separator fo r transient hydraulic and particulate loads observed in a realtime rainfall-runoff event was modeled by the application of the standard kturbulence model and a Lagrangian discrete phase model. Four discrete rainfall-runoff events were modeled in dividually and the modele d results agreed very well with the measured data (Absolute RPD < 10%). The CFD model was applicable across the entire range of flow rate and influent PSD variations and wa s able to accurately model both mass-limited and flow limited rainfall-runoff ev ents. Modeling the unsteady flow across the entire duration of the storm was tested against us ing a single design flow rate and the event mean concentration of PM for each event. It was obser ved that the PM separation behavior of the UOP varies significantly using the m ean, median and peak influent flow rates. Accurate modeling calls for including the flow variations across the entire treated volume of runoff.

PAGE 141

141 Table 5-1 Summary of measured and modele d particulate matter (P M) separation by the screened HS for four discrete storm events. Hydrologic and PM indices Experimental measurements and model predictions Separated mass % removal Qp Qave V Inf. mass Measure Model Measure Model RPD Storm Event (L/s) (L/s) (L) (g) (g) (g) (g) (g) (%) 20Aug04 17.5 5.1 12286 10592 6249.3 6691.7 59.0 63.2 -7.1 3Oct05 12.1 3.1 2615 738 354.2 397.3 48.0 53.8 -3.7 5Jun05 9.4 1.9 5856 4758 3187.9 3449.3 67.0 72.5 -8.2 14Oct04 0.6 0.1 1672 544 277.4 310.2 51.0 57.0 -5.6 Table 5-2 Summary of measured and modele d particulate matter (P M) separation by the screened HS for four discrete storm events using the measured EMC for influent PM, for Qmean, Qpeak and Qmedian Effluent mass load predicti ons for given Influent EMC Q = f(t) Qmean Qmedian Qpeak MeasuredModel RPD Model RPD Model RPD Storm Event (g) (g) (%) (g) (%) (g) (%) 20-Aug-04 4313.9 3595.5 16.7 3864 11.6 4646.15 -7.2 3-Oct-05 381.9 339.9 11 342.2 11.4 448.8 -14.9 5-Jun-05 1581.8 1039.7 34.3 755.2 109.4 1921.8 -17.7 14-Oct-04 267.8 126.8 52.6 178 52.2 203.4 31.7

PAGE 142

142 (a) (b) Figure 5-1 Plot (a) Plan and side view of detailed geometry of a screened HS. Plot (b) Typical fluid flow profile in side a screened HS. Outer Annular Chamber (Volute Area) Inner Chamber (Screening Area) Outer annular chamber (Volute Area) Static Screen Inlet Inner Chamber (Screening Area) Outlet Static Screen Conical Sump

PAGE 143

143 Figure 5-2 Experimental site for monito ring rainfall-runoff from an urban highway. t Q I dp m dp m Influent Effluent 70 KHz ultrasonic sensor Drop-box (Influent samples) Diversion valve OD=15 cm ID = 20cm 5 cm Parshall flume Runoff from catchment Runoff from catchment

PAGE 144

144 t/t max 0.00.20.40.60.81.0 Q/Q max 0.0 0.2 0.4 0.6 0.8 1.0 Imax = 121.9 mm/hr Qmax = 17.5 L sec-1 I/Imax 0.0 0.5 Q/Q max 0.0 0.2 0.4 0.6 0.8 1.0 20 Aug 2004 event tmax = 50 min I/Imax 0.0 0.5 Imax = 2.2 mm/hr Qmax = 0.6 L sec-1 14 Oct 2004 event tmax = 200 min I/Imax 0.0 0.5 t/t max 0.00.20.40.60.81.0 Q/Q max 0.0 0.2 0.4 0.6 0.8 1.0 Imax = 45.7 mm/hr Qmax = 9.4 L sec-1 05 Jun 2005 event tmax = 56 min I/Imax 0.0 0.5 Q/Q max 0.0 0.2 0.4 0.6 0.8 1.0 03Oct 2005 event tmax = 15 min Imax = 61.0 mm/hr Qmax = 12.1 L sec-1 Figure 5-3 Influent hydrology for four discrete storm events.

PAGE 145

145 % Finer by mass 0 20 40 60 80 100 Measured Modeled % Finer by mass 0 20 40 60 80 100 Particle Diameter ( m) 10 100 1000 10000 % Finer by mass 0 20 40 60 80 100 Absolute RPD =7.1 Mexperiment = 6249 g Mmodel = 6669 g Absolute RPD = 3.7 Mexperiment = 354 g Mmodel = 367 g Absolute RPD = 8.2 Mexperiment = 3188 g Mmodel= 3449 g Measured Modeled Particle Diameter ( m) 10 100 1000 10000 % Finer by mass 0 20 40 60 80 100 Absolute RPD = 5.6 Mexperiment = 277 g Mmodel = 310 g 20 Aug 04 03 Oct 05 05 Jun 05 14 Oct 04 Figure 5-4 Measured vs. modeled particle size distributions of particles separated by the screened HS.

PAGE 146

146 Figure 5-5 Temporal particle tr ajectories calculated by a Lagrangian DPM for the screened HS, for the 20 August 2004 storm. The peak flow rate Qmax was 17.5 L/s and the total duration of the storm, tmax was 50 minutes. Trajec tories are calculated for the particle diameter corres ponding to d50m and equal to 300 m. Particle density ( p) is 2.65 g/cm3. a) ti = 0.30*tmax, Qt=0.25*Qmax b) ti = 0.36*tmax, Qt=0.50*Qmax d) ti = 0.64*tmax, Qt=0.10*Qmax c) ti = 0.44*tmax, Qt=1.0*Qmax Inflow Inflow Inflow Inflow Outflow Outflow Outflow Outflow

PAGE 147

147 Figure 5-6 Modeled versus measur ed particle size dist ributions for four storm events utilizing Qmean, Qpeak and Qmedian. % Finer by mass 0 20 40 60 80 100 % Finer b y mass 0 20 40 60 80 100 Particle Diameter ( m) 1 10 100 1000 10000 % Finer by mass 0 20 40 60 80 100 Particle Diameter ( m) 1 10 100 1000 10000 % Finer b y mass 0 20 40 60 80 100 20 Aug 04 03 Oct 05 05 Jun 05 14 Oct 04 QMean QMedian QPeak Influent Effluent

PAGE 148

148 CHAPTER 6 A PARTICLE SEPARATION MODEL OF A VOLUME TRIC CLARIFYING FILTER FOR SOURCE AREA RAINFALL-RUNOFF PARTICULATE MATTER Introduction Anthropogenic particulate m atter (PM) trans ported in urban rainfall-runoff has been identified as a contributor to deterioration of surface water in the USA (USEPA 2000). Rainfallrunoff transports an entrained mixture of colloidal suspended, settleable and sediment fractions of PM, with the associated chemical constituents distributed across this PM (Sansalone et al 2007, Lee and Bang 2000, Sansalone et al. 1998, Igloria et al. 1997, Sansalone and Buchberger 1997). PM transported by runoff act as reactive surfaces for adsorbing, desorbing and leaching these chemical constituents which include organi cs, metals as well as nu trients (Sansalone 2002). Separation of PM by unit operations and processes (UOPs) for in-situ treatment of rainfall-runoff is challenged by factors such as the stochastic nature of hydrologic loads, variability and complexity of PM and chemical constituents and practical constraints such as land area and infrastructure (Liu et al. 2001). Media filtration of rainfall-runoff Various filtration config urations including infiltration and exfiltration through fixed granular media such as sand, so il or engineered media systems ha ve been suggested as viable solutions for meeting hydrologic and chemistry regulations for watersheds (Colandini 1999, Sansalone 1999, Li et al. 1999, Colandini et al. 1995, Geldof et al. 1994, Sansalone and Teng 2005). Engineered media such as oxide coated me dia has been found to be effective at reducing chemical constituents in wastewater and sour ce area rainfall-runoff (Ayoub et al. 2001, Teng and Sansalone 2004, Sansalone and Teng 2004,). Recently, the use of oxide coated media has been extended to in-situ and ex-situ runoff treatment (Erickson et al 2007, Sa nsalone et al. 2004, Liu

PAGE 149

149 et al. 2005a, Liu et al. 2005b). In this study, aluminum oxide coated media with a pumice substrate (AOCM)p was utilized for ex-situ physical filt ration of source area rainfall-runoff. Recently, Hipp et al. (2006) suggested that removable filter inserts may be a viable preliminary unit operation due to ease of maintenance. Wh ile granular filtration has demonstrated advantages for improving water chemistry through PM capture, such advantages require careful design, analysis and prototype testing, as well as regular maintenance to ensure acceptable hydraulic capacity (Keblin et al. 1997). Performance (mass and size of PM separated) of a granular medium filter depends on the following broad parameters (Tien 1989). ) ,K,d,d ,f(Q/A, Mmmp particles (6-1) In this expression, particles is the mass of particles separated, Q/A is the influent surface loading rate, A is the total surface area of the media Q is the influent volumetric flow rate, is the effective porosity, dp is the mass-based particle diameter, dm is the diameter of the granular media, is the empty bed contact time (EBCT) (AWWA 1990) and Km includes physical, chemical and surficial properties of the media, such as adsorption properties, media (internal) porosity, and morphology Granular filtration mechanisms can be classifi ed into surficial straining, sedimentation, interception, inertial impaction, diffusion, hydrod ynamic and electrostatic interaction (Wakeman et al. 2005). Filtration dynamics are widely assu med to be dependent on the surface loading rate (SLR) and with increasing SLR, macroscopic mechanisms tend to dominate the separation process, due to reduced contact tim es. The radial flow rapid-rate filter used in this study was operated at surface loading rates (SLR) ranging from 24 to 189 L/m2-min, which is higher than SLRs of typical rapid sand filters (83 L/m2-min) (Reynolds et al 1995).

PAGE 150

150 Primary issues faced of designing a filtration system for stormw ater PM control is the issue of excessive head-loss due to long-term retention of trapped particles in the packed filter bed and unsteady flow rates if there is not upstream hydraulic attenua tion of flow. Farizoglu (2003) reported that head-losses increase proportional to the square of the influent flow rate. Given the large variability of flow rates within and between hydrologic events head loss considerations are a challenge in the design of field scale filtrati on systems to achieve a pre-determined level of clarification. Reddi (1997) repor ted that fine PM in runoff is a contributor to clogging of filters, and eventual head buildup. As the filter progr essively clogs, filter ripening occurs and progressive clogging has the potential to cause dete rioration of the filtration mechanism, due to the modified and increasingly non-ideal hydrodynamics. Many urban stormwater modeling approaches are extensions of wastewater treatment principles which assume steady flow and PM load s. This is the case for many stormwater unit operations and regulatory testing and modeling protocols assuming steady flow and load. While controlled testing, required in many regulatory te sts, is a necessary building block to examine and compare unit operation behavior to controlled lo ads in lieu of the uncertainty and stochastic nature of field loadings; ultimately unit operations are loaded by highly variable and uncontrolled runoff processes. The associated PM and chemical constitu ent delivery in runoff events are highly variable processes. Theref ore stormwater unit opera tions and their models must support rapidly changing flows and PM ch aracteristics even when upstream volumetric clarification is provided. Th e stormwater management mode l (SWMM) (Huber 1988) has made very significant advances over simplified peak flow models and greatly improved the design and modeling of volumetric controls even for complex hydrology and urban drainage systems. However while the current version of SWMM ca n model complex hydrologic and variable PM

PAGE 151

151 (as a lumped parameter, TSS) loads and examine volumetric attenuation, it is beyond the scope of the current version to provide in-depth hydrodynamic, head lo ss and PM clarification profiles of a filtration unit operation for transient loads an d variable particle size distributions (PSDs). Modeling Approach Num erical modeling of water and wastew ater treatment processes has found many applications in the past seve ral decades (Do-Quang et al. 199 8). For treatment systems where hydrodynamics play an important role, numerical methods have been developed to solve the Navier-Stokes equations for coupl ed fluid and particle dynamics by a series of discretization techniques and algorithms (Ande rson 1995). Such approaches to model dilute particle-laden flows and granular media filtration of such flows is an active research ar ea (Curtis et al. 2004, van Wachem et al. 2003). Tung et al. (2004) studied deep bed filtration for a sub-micron/nano-particle suspension utilizing a microscopic approach wherein diffe rent types of media packing were modeled. However, macroscopic approaches to modeling filter media are needed to model practical systems where packing schemes are most commonl y random and unstructured. Li et al. (1999) applied a 2-D numerical model to simulate vari ably saturated flow in a partial exfiltration system. Sansalone and Teng (2005) applied a 2-D numerical model to si mulate the transient hydrodynamics of a linear partia l exfiltration system containing oxide-coated media and cementitious permeable pavement subject to direct unsteady rainfall-runoff loadings in order to examine the PM clarification in this in-situ sy stem. In the approach of the present study a macroscopic-basis is utilized to simulate th e 3-D hydrodynamic and clarif ication response of a series of clustered radial filte r cartridge without resorting to the assumptions of symmetric loading and response within the clustered system. Furthermore, the modeling combines the unsteady hydrodynamic inputs and corresponding va riable PSD characteristics of PM; such

PAGE 152

152 unsteady and variable behavior be ing characteristic of actual rainfall-runoff events. Therefore this study not only examines the 3-D spatial comp lexity of the clustered filter VCF system, but also examines the VCF response subject to uncontrolled and coupled hydrologic and PM loadings. Objectives This study focuses on th e PM filtration and head loss response of a VCF system to direct and unsteady rainfall-runoff loadings from an urban paved source area watershed dominated by traffic loadings. There are three sets of objectives in this study for the five events monitored and modeled in this study. Between these events the VCF was off-line and was not subject to any loading. With the VCF directly loaded by uns teady direct runoff and variable PM, the first objective was to examine the VCF hydraulic resp onse including flow changes, head loss and surface loading rate (SLR) this series of five events. The second objective was to examine the temporal response of the VCF for the d50m of the PSD. The final set of objectives was to examine the event-based behavior of the VCF in modifying the influent concentration and mass of PM. Methodology VCF and Watershed Configuration The instrumented VCF received d irect unst eady runoff from a source area urban paved watershed in urban Baton Rouge, Louisiana. The upstream urban drainage system was designed to intercept lateral pavement sheet flow from a concrete-paved watershed consisting of two identical eastbound and westbound catchments, each having a contributing area of 544 m2. The watershed was dominated by traffic and the av erage daily traffic (ADT) for eastbound and westbound I-10 was 142,000 vehicles. Rainfall was reco rded with a tipping bucket rain gage and data-logger in increments of 0.254 mm (0.01 inch). Mean annual precipitation in Baton Rouge,

PAGE 153

153 Louisiana is 1460 mm/year. Further details of the watershed, h ydrology and water chemistry and loads can be found elsewhere (Dean et al. 2005, Sansalone et al. 2005). During 2006 the watershed was loaded by 59 rainfall-runoff events with five events monitored and treated in a series in April and May by the VCF. Since the VCF was an off-line unit, events not treated bypassed the VCF and were directly discharged into City Park Lake. Prior to treating the first event of 21 April 2006, the VCF had only been loaded by potable water for hydrodynamic testing. The main VCF components were a clarif ication vault and five radial downflow media filtration cartridges, shown in Figure 6-1 and 6-2. The five cartridges with aluminum-coated media are housed in a 116.8 cm by 212.2 cm vau lt surface area. This structure contained the influent and effluent pipes as well and internal manifold delivering treated effluent runoff to the effluent drop box. Pressure transducers are installe d at the influent Parshall flume, system vault, cartridge center pipe effluent box and effl uent V-notch weir. Runoff was collected through dire ct piping from the watershed catch basins and expansion joint, and entered the VCF thr ough an influent delivery system which includes a 5.08 cm (2inch) Parshall flume and a 36.6 cm by 79.2 cm drop box for sampling. Runoff that entered the VCF influent box was transported directly to the bottom of the vau lt beneath the radial cartridges. When the water surf ace elevation in the vault reache d the operating level (slide gate was in the closed position), a float valve was triggered through buoyancy and the orifice plate opened gradually. Runoff then flowed through the f ilter cartridges driven by the differential head and drained into the perforated dr ain tubes located at the cartridge center and then to the collector manifold. The manifold was plumbed to a floatcontrolled slide gate that sets the VCF flow control to achieve a balance between flow and driving head level. After an event, detained runoff continued draining through each cartridge, manifo ld and the slide gate until the vault water

PAGE 154

154 surface was below the bottom of th e filter cartridges. When runoff ended, the float controlled slide gate closed until the next runoff event. Runoff was retained in the VCF up to the bottom of the filter cartridges. Data Acquisition and Management Druck pressure transducers connected to a Cam pbell-Scientific CR 1000 were used in the VCF for real-time water level monitoring. Water levels inside and outsi de the media cartridges were monitored using two 2.5 psi (17.2 kPa) transducers. A single 5.0 psi (34.5 kPa) transducer was located at the bottom of effluent drop box. Each transducer wa s calibrated, assigned appropriate multipliers and offset s; with pressure data acquired every 10 seconds. Transducer data were converted to flow us ing a calibrated head-discharge re lationship for the influent 50.8 mm Parshall flume and the effluent 60o V-notch weir summarized in equation 1 and 2, respectively and the VCF stage-st orage provided in equation 3. Components are illustrated in Figures 1 and 2. 8619.28.304 62.74 2 tan2 15 8 Z hgCQu d (6-2) 6548.18.304 07.31 Z KhQu (6-3) ZAVe Ae = 2.48 m2 (Z < 0.1780 m) ZA Ve 061. 0 Ae = 2.13 m2 (0.1780 m < Z < 0.5840 m) ZA Ve 540. 0 Ae = 1.31 m2 (0.5840 m < Z< 1.1428 m) (6-4) ZA Ve 397. 0 Ae = 2.13 m2 (Z > 1.1428 mm)

PAGE 155

155 In these equations Cd is the effective discharge coefficien t, g is the gravit ational constant [L/T2], is the v-notch angle, h is the head [L] ab ove the v-notch invert, u is an exponent, Z is the head [L] above the VCF invert (datum), Q is the discharge [L3/T], K is an index of VCF storage surface area, V is VCF storage volume (m3), and Ae = effective VCF surface area (m2). An event volume balance is conducted to ensure volume conservation. A volume balance error (VBE) criterion of 10%. VINF, VEFF, VS and Vs represent influent, e ffluent, storage and sampling volumes, respectively. 100 (%) INF sS EFF INFV VVVV VBE (6-5) Radial cartridge filters (RCF) and (AOCM)p The total volume of each RCF was 72.9 L with the diameter and height of the RCF are 0.4572 m and 0.5588 m, respectively. The media wa s contained between an outer mesh and nonreactive impervious po lypropylene discs on top and bottom of the cartridge. Engineered (AOCM)p was utilized as the granular media. Th is media had a pumice-substrate with an aluminum oxide coating. The distribution of media diameter (dp) was obtained by digital image analysis and utilization of Image Pro Plus v 6.2 Image Analysis Software. Image analysis results were utilized to calculate an equivalent circul ar diameter (ECD) as follows (Holdich 2002). The ECD is defined as the diameter of a sphere havi ng the same projection area as the object. The resolution of the imaging method was 10.1 mega-pixels. /4 AECD (6-6) In this expression, A is the projected area of the media. Figure 6-3 illustrates a frequency histogram for the media tested. A three-parameter Gaussian distribution fit the data (R2 = 0.93). The additional parameter accounts for the area unde r the curve, which is not equal to 1. The

PAGE 156

156 mean media size is 3.56 mm 0.8 mm for a sample size (n) of 2551. The specific gravity of the pumice-based media was 2.35. The media was pluvia ted into the cartridge so that the cartridge contained 49830.4 g of dry media at full capac ity for each experimental run. Triplicate measurements of total clean bed porosity ( ) of the RCF with media produced a mean of 0.71 and standard deviation of 0.041 by volume. Of th is total porosity the in ternal porosity of the media was measured in triplicate by mercury por osimetry (ASTM 2003) with a mean of 0.37 and a standard deviation of 0.022. The total porosity includes all av ailable pore areas in the media and media bed, but not necessari ly the effective porosity available for a given flow rate. Influent and Effluent Sampling and Analysis Five discrete rainfall-runoff events in a seri es were monitored and treated by the VCF. For each event sampling started immediately when runoff was observed, and was performed on a time basis for the entire duration of runoff. Infl uent was sampled at the influent drop box of the VCF and effluent was sampled at the outflow of the VCF. All samples were taken manually, across the entire cross-section of the influent and e ffluent to ensure representative sampling. The number of discrete samples of influent and e ffluent was chosen appr opriately to provide a reasonable estimate PM concentrations. PM was measured by two methods; gravimetrically as the sum of the sediment (> 75 m), settleable (~25 to 75 mm) and suspended (1 to ~ 25 m) fractions and utilizing laser diffraction for part icle volume % for each particle size; generating PSDs. The total PM was obtained as a sum of the mass of sediment, settleable and suspended PM fractions which also generated the susp ended sediment concentration (SSC). The methodology is described elsewhere (AST M 1993, ASTM 1998, APHA 1998, Kim and Sansalone 2008a, Sansalone and Kim 2008b). PSDs of the influent and effluent were measured using a laser diffraction analysis and Mie scattering theory (Finlayson-Pitts et al 2000). The

PAGE 157

157 particle size range was reported from 0.02 m to 2000 m, in 100 increments and was applicable to PM concentration range of 1 to 1000 mg/L w ith a resolution < 1 mg/L The contribution of coarser particles to the temporal influent PSD was calculated from the separated VCF particles and the temporal effluent PSD. The PSD measured for each influent and effl uent discrete sample by laser diffraction. The relative percent difference (RPD ) is used to compare model a nd experimental results and is follows. For the purpose of QA/QC, a mass balance error constraint of 10% was imposed on PM data. Mass Balance Error (MBE) = %10100 ) ( i e VCF iM MMM (6-7) In this expression, iM is the influent mass of PM,VCFM is the mass of PM captured by the VCF and eM is the effluent mass of PM, computed fr om the measured effluent PM across the total treated volume. All gravim etric measurements were carrie d out on a dry mass basis. The event mean concentration (EMC) for PM over th e time period of a rainfall-runoff event is typically used as an indicator of overall water quality and to calculate percent removal for stormwater unit operations (Huber 1993). t ttq dttqtc V M EMC0 0)( )()( (6-8) In this expression, M is the total effluent mass load over the entire duration of the test, V is the total volume of flow over th e entire duration of the test, C is the flow weighted mean concentration, c (t) is the time variable particulate-bound concentra tion, q (t) is the time variable flow rate.

PAGE 158

158Multiphase flow modeling methodology While the geometric symmetry of a single RCF suggested a simplified two dimensional model, the hydrodynamics and particle dynamics vary as a function of x, y and z spatial coordinates, as does the physical clustering of five RCF in pa rallel. As a result, a three dimensional (3-D) approach was required. The fluid flow equations were solved based on fundamental laws of conservati on of mass, momentum and energy. The Navier-Stokes equations were solved using numerical tech niques of computational fluid dynamics. Versteeg et al (1995) define the general conservation equation for any fluid property for a control volume )( zyxV is as follows. Sgraddivudiv t )()( )( (6-9) In this expression, u is the fluid velocity, is the diffusion coefficient, and Sis the source/sink term. z w y v x u udiv ) ( (6-10) In this expression, u, v and w are the velo city vectors in the x, y and z directions respectively. The momentum equations for three dimensions are obtained by incorporating u, v and w into equation 6-9. X-momentum: MxSugraddiv x p uudiv t u ))(( )( )( (6-11) Y-momentum: MySvgraddiv y p uvdiv t v ))(( )( )( (6-12) Z-momentum: MzSwgraddiv y p uwdiv t w ))(( )( )( (6-13)

PAGE 159

159 In the above expressions, p is the pressure, is the viscosity, and MzMyMxSSS are source/sink terms to account for surface forces su ch as viscous and pressure forces, and body forces such as gravitational and centrifugal forces in the x, y and z directions respectively. The RCF was modeled by adding a momentum sink term to the flow equations. This sink contributes to the pressure gradient in the porous computationa l cell, creating a pressure drop that is proportional to the fluid velocity in a comp utational cell. The source term is composed of two parts: a viscous loss term and an inertial loss term. 3 1 3 12 1jj j ij jij ivvCvDS (6-14) In this expression, iS is the source term for the ith momentum equation, is permeability, iC2is the inertial re sistance factor, iv is the velocity in the ith momentum equation, v is the velocity in a computational cell. For isotropic porous media the above equation can be expressed in the following form. 2 22 1vCvSi (6-15) In this expression, is the permeability [L2] and C2 is the inertial resistance coefficient. The dominant flow regime is laminar to tr ansitional in the RCF. The Reynolds number for flow in the media was calculated as follows. )1( Re sm mediavd (6-16) In this expression, md is media particle diameter, is the fluid viscosity, is the porosity of the packed bed (not media porosity), sv is the superficial veloc ity through the packed bed and is fluid density. The Reynolds numbers were found to range from approximately 5 to 30.

PAGE 160

160 Previous studies have shown that the standard kmodel (Launder and Spalding 1974) for turbulent flow has worked well in both micro a nd macroscopic approaches to solving flow in porous media (Antohe et al 1997). The tr ansport equations of the standard kmodel are expressed as follows. k bk jk t j i iSGG x k x ku x k t )() ( (6-17) For : S k CGCG k C x x u xtb k j t j i i 2 2 3 1) ( )() ( (6-18) In equations 6-17 and 6-18, kG represents generation of k due to the mean velocity gradients; bG is generation of kdue to buoyancy; 1C ,2C and 3C are constants ;k and are the turbulent Prandtl numbers for kand respectively; kS and S are user-defined source terms. The constants were determined by Launder and Spaulding (1974). The values of1C ,2C ,3C ,k and used in the model were 1.44, 1.92, 0.09, 1.0 and 1.3 respectively (Launder et al. 1974). The free surface of the flow was modeled as a shear-free boundary. Porous media may be modeled with a macroscopic or a microscopic approach (Ranade 2002). In the latter approach, the micro-scale pore structure is taken in to account. A microscopic approach demands tedious computational resour ces and is typically used as a benchmark problem, as opposed to modeling pract ical systems. On the contrary, the macroscopic or lumped approach considers the entire filter bed as isotro pic or non-isotropic porous media. This approach is characterized by pertinent lumped para meters such as the effective porosity ( ), inertial and viscous resistance coefficients. The geometry of the media can be accounted for by utilizing bed porosity distributions, such as Muellers distribution (Mueller 1992), or by assuming

PAGE 161

161 homogeneous distribution of pores between medi a. With uniform pluviation in the RCF and uniform size gradation for the media a uniform po re distribution was assumed in this model. Flow in porous media has been traditiona lly modeled analytically by comparison to pipe/conduit flow by specifying analogous parame ters such as the hydraulic diameter and roughness coefficient. Laminar flow (Re < 10) through por ous media has been successfully modeled by applying Darcian-type equations. Models such as Blake-Plummer and CarmanKozeny equations were developed to account to transitional flow regimes. These models were extended by Ergun (1952) to account for turbulen t flow. The Ergun equation for packed beds applies to flow regimes from laminar to turbulent and is expressed by the following equation. 2 3 3 2 2)1(75.1)1(150s m s mv d v d L p (6-19) In this expression, p is the pressure drop across the medi a, L is the length of the packed bed, is the fluid viscosity, md is media particle diameter, is the total porosity of the packed bed, vis the superficial velocity through the packed bed and is fluid densit y. Comparing equations 9 and 14, and C2 can be expressed as follows. 2 32)1(150 md (6-20) 3 2)1(5.3 md C (6-21) Modeling the particulate phase Multiphase flows are modeled with an Eulerian-Eulerian approach or an EulerianLagrangian approach (van Wachem et al. 2003) depending on the extent of coupling between phases. Elgobashi (1991) proposed a regime ma p for appropriating the degree of inter-phase coupling, by analyzing length and time scales. Subs equently, it was determined that a Lagrangian

PAGE 162

162 approach to tracking the secondary phase is mo st appropriate for flow s with a low influent particulate volume fraction which is the case in this study (PVF < 3.0 %). The transient flow field was modeled utilizi ng the Eulerian approach, for each time step. Subsequently, the influent partic les were injected at times that correspond to influent sampling times during storm data collection. The Eulerian-Lagrangian approach was chosen to model the behavior of particles in the computational domain of the VCF. In this approach the flow field was solved using the Eulerian approach. Following this, particles were track ed using a Lagrangian Discrete Phase Model (DPM). The DPM is derived from force balanc es based on classical Newtons (Turbulent and transitional regimes) and Stokes (Laminar regimes) laws descri bing particle motion, and is summarized by the following equation. x p px p D pF g uuF dt du )( ) ( (6-22) dF 24 Re 182 pD ppC d (6-23) 2 3 2 1Re Rep p Da a aC (6-24) uudpp p Re (6-25) In equations 6-22 through 6-25, uis the fluid velocity, pu is the particle velocity, is the fluid density, p is the particle density, pd is particle diameter, is the viscosity, 321,, aaa are empirical constants that apply to smooth spherical particles as a function of the Reynolds number (Morsi et al., 1972) andpReis the particle Reynolds number.

PAGE 163

163 Particle trajectories are obtain ed by integration of equation 622. For this study, particles of consistent morphology were injected across th e entire flow cross-section immediately above the inlet to the VCF to obtain comparable partic le trajectories across varied flow rates, by reducing uncertainty in initial spatial location. Particles were defined as silica particles of diameters associated with the sieves utilized and a measured specific gravity of 2.65 as determined by helium pycnometry (Sansalone et al 1998). The cumulative gamma distribution function described in the previous section was used to model the relative mass fractions as a function of particle diameter. Particles were tracked for a specific length for each flow rate, based on hydraulic residence times calculated by injecting neutrally buoy ant submicron tracer particles. Particles that remained in the VCF after integrating over the specified length were considered to have been separated by the VC F. Particle removal was thus defined by the following equation. 100*I HSN N p (6-26) HSN is the number of particles that remain in the VCF, and INis the number of particles injected at the inlet. Discretization and Solution Schemes The computational domain was discretized usin g an unstructured mesh with tetrahedral elements, generated by TGrid (Qi et al. 2006). The volume of the VCF was divided into total of 5.18 million cells, after ensuring grid convergence. Numerical solutions were obtained using a Finite Volume Method (FVM) a nd a cell-centered scheme used for discretization. A SecondOrder Upwind Scheme (Barth et al. 1989) was used solve for flow parameters. The spatial discretization is shown in Figure 6-4 for the VCF.

PAGE 164

164Time Discretization The flow data were measured for discrete time in tervals. In order to test the sensitivity of the solution to smaller time steps, continuous functions were utilized to model the measured flow data. The runoff hydrograph was modeled by spli tting the hydrograph into piecewise continuous segments which allowed for accurate curve fitting. An important consideration to make while c hoosing a time discretiza tion approach was to account for the high variability between the magnit udes of influent flow as a function of time. In a given storm event, the flow rates can change multiple times, and this change can be rapid, or gradual in time. In light of this, two time discretization approaches were tested. The first approach was to use a first-order fully implicit scheme for time discretization. The fully implicit scheme is unconditionally stable and is suggested for transient operations which do not involve chemical reactions (Ran ade 2001). The second approach was to use an advanced adaptive time stepping approach, which involved modification of the time step based on the truncation error of the time integration scheme, estimated fr om a predictor-corrector type of algorithm (Gresho et al 1980). At each time step, a predicted solution is obtained using an explicit method such as the Adams-Bashforth met hod (Ferziger et al 2002), and this is used as the initial condition for the next time step. The correction is computed using the implicit formulation, and the truncation error is the difference between the predicted and the corrected solutions. A preset truncation error tolerance is chosen, and is used as the constraint within which the time step can be changed. The transient SIMPLE (Semi-Implicit Method for Pressure Linked Equations) algorithm (Patankar 1980) accounted for pressure-veloc ity coupling. The criterion for iterative convergence for the flow field solution was set at 1x10-3 (Ranade 2002).

PAGE 165

165Results and Discussion Event-Based Hydrologic Loadings and Response Five consecutive rainfall-runoff events lo ading the VCF were captured, monitored, analyzed and the result s modeled. The event hyetographs and corresponding hydrographs are shown in Figure 6-5. With the exception of the very low intens ity event of 06 May 2006 with an initial lag time of 39 minutes th ere was a rapid runoff response to rainfall loadings characterized by an initial lag time of less than 3 minutes for all other events. These results are similar to previous results reported for this watershed (Sansalone et al. 2005). Irrespective of rainfall intensity, the runoff response (the hydraulic loading to the VCF) was highly unsteady for all events and for two events multiple distinct hydrograph peaks were generated. Event runoff durations loading the VCF ranged from 50 to 170 minutes. All volume balance errors (VBE) were in the range of 10% except for the 06 May event which only generated 495 L (14.7% VBE) of runoff with flow depth monitoring data approaching the lower detection limit in the Parshall flume. Traffic is the dominant anthropo genic activity in the watershed that influenced the rainfall-runoff relationship and PM loadings to the VCF. Therefore traffic was measured as vehicles during storm (vds) to provide an index for the influence on the rainfall-runoff relationship and also PM loadings (as SSC) (Sansalone et al. 1998, Sansalone et al. 2005). The vds values ranged from 4429 for the 21 April 2006 event to 8938 for the 29 April 2006 event. Traffic was also the dominant sour ce of abstractions during a rainfall-event causing deflection of rainfall and re-entrainment and deflection of the runoff. The initial watershed losses accounted for approximately 0.5 mm of rainfall and the rainfall-runoff relationship for the watershed required rainfall depths to be greater than 1.0 mm to provide sufficient runoff volume for initial sampling.

PAGE 166

166Filter media cartridge head loss modeling results From the pressure transducer data collected fo r inflow, outflow, inside of cartridge, outside of cartridge and the effluent drop box a comp lete hydraulic profile across the VCF treatment system during each event is constructed. As an in-situ unit operation the VCF is not backwashed by intent during the period of monitoring. A primary interest is the head loss across the filter cartridges as a function of flow rate and surf ace loading rate for each rainfall-runoff event. These results are summarized in Figure 6-6 for the five events illustrated. Head loss values remained below 200 mm during all events and th e head loss profiles generally mimicked the hydrographs. Measured and modeled head loss pr ofiles were similar with a RPD generally less than 15%. As shown in Figure 6-7, the unsteady event mean head loss is nominal; less than 40 mm and generally following a linear trend as a function of event mean surface loading rate (SLR). The unsteady event mean head loss is comp ared to head loss results for steady flow rate conditions of this same system under pilot scale te sting. The steady flow rate head loss results are consistently lower th an the unsteady event he ad loss below 150 L/min-m2, but appear to converge beyond this level. With the aid of the calibrated and validated CF D model, pressure dist ributions represented as head loss along the radial and vertical directions were comput ed and are illustrated in Plots (A) and (B) respectively of Figure 6-8. The head loss varies inversely with distance from the center of the RCF as illustrated in Plot (A). This variation displays a linear trend with the slope increasing as a function of flow rate. The overa ll head loss across the entire radial distance is equal in magnitude to the overall modeled head loss. Plot (B) illustrates pressure di stributions along a line parallel to the z-axis and located at the geometric mid-point of the RCF in the radial direction, at one-half the annular thickness of the media section of the cartridge (r/2). A check was performed for conservation of energy

PAGE 167

167 across the entire RCF using Bernoullis theorem, with local velocities obtained from the CFD model. This check supported the validity of the modeled distributi ons. There is a clear parabolic drawdown towards the center of the cartridge, proportional to increasing peak flow rates. This is in contrast with steady flow model resu lts for a single radial ca rtridge filter (RCF), discussed in detail in Chapter 4. Whereas for a steady state simulation, beyond a flow rate of 45.4 L/min, there is a parabolic pr ofile for pressure distributions as a function of depth, with the minimum pressure at the top and bottom ends (no-flow boundaries) of the RCF. The center effluent outlet of the RCF never flowed full and remained open to the atmosphere. Therefore, the RCF configuration was analogous to an inverted well, with flow pu mped out in the direction of gravitational force. At the top surface, the parabolic profile of the drawdo wn is due the head-loss created by the porous media. At the bottom of the RCF, there is also a pressure differential from inside to outside of the cartridge. This is in line with typical drawdown in pumped wells as expressed by the Theis equa tion (Freeze et al. 1979). However, in the case of the VCF, the pressure differential from the inside to the outside of the cartridge is not severe, as th e effluent, which is open to the atmosphere, is located in the farfield, sufficiently distant to eliminate any draw down at the bottom of the cartridge. An important result was that the head loss for each individual cartridge remained similar across the entire duration of each individual storm event. This clea rly indicates that irresp ective of the asymmetric cartridge layout in the tank, the tank does provide uniform hydr aulic loading for each cartridge. Separation of particulate matter modeling results The varying hydrology resulted in a distinct particle transport signature for each storm.. Among the five storms analyzed, the VCF was su ccessful at separating a large portion of the influent PM. The overall particle separation is most effective for sediment sized particles. However, even for the finer settleable and su spended particles, the VCF proved effective in

PAGE 168

168 separating the influent PM. The model was able to accurately predict th e overall PM separation across each individual storm event, with the RPD remaining below 15%. Figure 6-9 and figure 6-10 provide a compar ison of measured and modeled temporal effluent mass loads and concentrations respectivel y. Bertrand et al. (199 8) classified rainfallrunoff events into mass-limited and flow-lim ited events based on temporal pollutant mass delivery. For mass-limited events, mass delivery is skewed toward the initial portion of the event, while mass delivery tends to follow the hydrograph for flow-limited events. The five storms accommodate both mass and flow-limited kinds of particle transport. For the 21 April 2006 event, the effluent mass load, and SSC remain ed relatively invariant as a function of time, irrespective of the change in influent SSC, suggesting that PM separation mechanisms in the VCF are independent of the influe nt concentration. This behavior is noticed for all but the 29 April 2006 and 06 May 2006 events, suggesting the po ssible difference in in fluent and effluent PSDs. Figure 6-10 is a plot of measured and modeled effluent d50m as function of time. In this figure, we notice that for the 29 April 2006 storm, the particle si ze distribution in the effluent changes as a function of time. This can be attr ibuted to the complex multiple-peaked hydrograph, with a very high flow rate (25.3 L/s) and an influent volume of almost 50,000 L The 06 May 2006 storm on the other hand was a very low intensity storm (Qmax=0.3 L/s), and delivered approximately 500 L of water. These two storms represent two extremes, both in flow rates and volume of influent delivered. Th ese results suggest that the infl uent concentration is truly a function of the particle size distribution transpor ted by the storm, wherein the higher flow rates transport a PSD that is inclusive of sediment, settleable and suspended fr actions, while low flow rates mainly deliver suspended fractions. The CFD model is able to accurately predict the

PAGE 169

169 temporal variation of effluent mass, effluent concentration and effluent d50m across varying influent flows and particle concentrati ons, for both mass and flow-limited events. Conclusions In-situ control of hydrology and PM are major considerations for many watersheds and receiving water systems. Volumetric clarifica tion-type BMPs that can provide storage through detention/retention operational management are more frequently incorporating media filtration as a unit operation. These systems can provide some degree of volumetric and peak flow attenuation, provide preliminary or primary cl arification of PM; and with filtration provide secondary treatment to separate PM a nd nutrients associated with PM. This study examined the behavior of a vol umetric clarifying filter (VCF) loaded by a paved source area watershed dominated by transportation, in which some fraction of the influent runoff is stored in the VCF between events. In this study five source area rainfall-runoff events are monitored and treated real time; and the role of runoff storage and fi ltration on the separation of PM is examined and modeled. Despite the significant variability between the five events, delivery of PM mass (as SSC) generally resulted in a mass-limited (first-order) transport based on runoff volume for this source area watershed. In this study the VCF is operated in an undr ained condition; where some fraction of runoff from the previous event remains as stored volu me until mixed with influent from a new event; typical of many below-grade in-s itu BMPs. As a result, hydrol ogic attenuation by the VCF is muted. Mean head loss across the filters of the VCF is low (< 40 mm), due to the coarse uniform size gradation of the engineered media. Eventbased mean head loss increases nearly linearly with the corresponding event mean surface loading ra te for the VCF. The temporal variation of filter head loss across an event can be clearly discerned as a function of flow rate. Since there is no backwashing of the VCF system during the st udy period and any increase in head loss is not

PAGE 170

170 discernable in comparison to head loss due to flow rate variations. This result suggests that the vertical orientation of the radi al-flow cartridges and utilization of the coarse uniform media did not generate a significant schmutzdecke on the ve rtical inflow surfaces of the filter. Despite the media remaining in a moist condition (but ab ove the storage water surface) for the entire study period results do not indicat e significant biological clogging. Results indicate that the VCF is capable of significant load reductions of PM as SSC with effluent concentrations at or below 30 mg/L on an event basis. For both head loss and PM treatment results, the numerical model was capable of reproducing the behavior of this VCF sy stem to highly variable and unsteady flow during a series of five consecutive events.

PAGE 171

171 Figure 6-1 Plan view of experi mental site and Volumetric Clarifying Filter (VCF) system, CB represents catch basin and 304.8 in the equation represents unit conversion (mm to ft) factor. Data lo gg e r 5.1 cm Parshall Flume Watershed PCC pavement 1088 m2 (2 x 544 m2) 2% surface slope ADT=142,000 (east and west) Tee 10.2 cm PVC pipe from east CB @ 10 % slope Dro p box 6 % 31 cm sloped open PVC trough under expansion joint 10.2 cm PVC pipe from west CB @ 10 % slope 15 cm PVC pipe @ 6 % slope 212.1 cm 45.7 cm Effluent 1 2 Influent 9 3 4 5 6 7 8 10 11 116.8 cm 1. Influent box 2. Radial flow cartridge 3. Baffle 4. Vault drainage pipe 5. Float valve 6. Effluent drop box 7 Orifice 8 Effluent pipe 9 Influent delivery pipe 10. Effluent V-notch weir 11. Effluent drainage pipe 5.1 cm Parshall flume Discharge (L/s) 0 2 4 6 8 10 6548.18.304 07.31 X Y60 o V-notch weir Stage (mm) 0.00.10.20.30.40.5 Discharge (L/s) 0 2 4 6 8 10 0306090120150 8619.28.304 62.74 X Y VCF Stage-Storage 030060090012 0 Storage (L) 0 500 1000 1500 2000 2500 V overflow = 2180 L Stage (mm)

PAGE 172

172 H1 (Parshall flume pressure head) mm H2 (Outside cartridge pressure head) mm H3 (Inside cartridge pressure head) mm H4 (Outlet box pressure head) mm H5 (V-notch pressure head) mm Note (1) & (5) measured by CS420-L Pres sure Transducer (6.9 kPa) (2) & (3) measured by CS420-L Pres sure Transducers (17.2 kPa) (4) measured by CS420-L Pressure Transducers (34.5 kPa) Real-time Data acquisition: CR1000 Datalogger (Cambell Scientific) Media size (d50) : 3.56 0.25 mm Media hydraulic conductivity (k) : 1.44 0.17 cm/s Media macro porosity ( m) : 0.37 0.018 Figure 6-2 Location of the pressure transducers installed in the Volumetric Clarifying Filter (VCF) system. d50, k and m represent the median size, hydraulic conductivity and macro porosity of the media ( AOCM)P used in this study. 393.7 mm H5 184 mm H4 177.8m 558.8mm H3 H2 H1 Data logger Parshall flume System vault Cartridge center draina g e pipe Effluent box V-notch weir

PAGE 173

173 Figure 6-3 Measured media size expressed as a Gaussian frequency histogram. The average media size was found to be 3.56 .8 mm. A three parameter Gaussian distribution was used to model th e data; a = 5.283, b= 0.6392. x0=3.435. dp (mm) 01234567 Normalized Frequency 0 2 4 6 8 R 2= 0.93 2 ))(5.0(0*b xxeay

PAGE 174

174 Figure 6-4 Profile and plan view s of a section of the computational grid of the VCF system. Y Z Y X

PAGE 175

175 Figure 6-5 Hydrographs a nd hyetographs for 5 real-time rainfall runoff events I/Imax 0.0 0.5 Q/Q max 0.0 0.2 0.4 0.6 0.8 1.0 Imax = 91.4 mm/hr Qmax = 13.3 L sec-1 21 April 2006 tmax = 70 min I/Imax 0.0 0.5 Q/Q max 0.0 0.2 0.4 0.6 0.8 1.0 Imax = 259.1 mm/hr Qmax = 25.3 L sec-1 29 April 2005 tmax = 170 min Imax = 1.0 mm/hr Qmax = 0.3 L sec-1 I/Imax 0.0 0.5 Q/Q max 0.0 0.2 0.4 0.6 0.8 1.0 06 May 2006 tmax = 92 min I/Imax 0.0 0.5 t/t max 0.00.20.40.60.81.0 Q/Q max 0.0 0.2 0.4 0.6 0.8 1.0 Imax = 61.0 mm/hr Qmax = 9.1 L sec-1 07 May 2006 tmax = 68 min I/Imax 0.0 0.5 Imax = 30.5 mm/hr Qmax = 6.5 L sec-1 27 May 2006 tmax = 31 min t/t max 0.00.20.40.60.81.0 Q/Q max 0.0 0.2 0.4 0.6 0.8 1.0 27 May 2006 tmax = 50 min

PAGE 176

176 Figure 6-6 Measured vs. modeled cartridge head loss profiles as a function of normalized time. RPD is the relative percent differenc e between measured and modeled data. 27 May 2006 t/t max 0.00.20.40.60.81.0 Q/Qmax 0.00 0.25 0.50 0.75 1.00 Head loss (mm) 0 50 100 150 200 Hydrograph Measured Modeled RPD = 15. 4 % 21 April 2006 Head loss (mm) 0 50 100 150 200 RPD = 12. 3 % 29 April 2006 Q/Qmax 0.00 0.25 0.50 0.75 1.00 RPD = 11. 3 % 07 May 2006 t/t max .00.20.40.60.81.0 Q/Qmax 0.00 0.25 0.50 0.75 1.00 RPD = 13. 2 % 06 May 2006 Head loss (mm) 0 50 100 150 200 RPD = 10. 4 %

PAGE 177

177 Event mean SLR (L/m2-min) 050100150200 Measured mean head loss (mm) 0 20 40 60 80 100 Pilot-scale steady flow testing Unsteady runoff flows in VCF Figure 6-7 Comparison of head loss as a functio n of surface loading rate (SLR) for steady flow testing and for unsteady event flows for the same radial flow cartridges. Transient flow SLR and head loss is determined co rresponding to the m ean flow rate of real-time storm events and arranged in an ascending fashion corresponding to 6 May 2006 (lowest SLR), 21 April 2006, 27 May 2006, 7 May 2006 and 29 April 2006 event (highest SLR), respectively.

PAGE 178

178 Figure 6-8 Head loss ( H) and pressure distributions in the VCF. Plot (A) illustrates predictions of head-loss across the VCF from the CFD model. Plot (B) illustrates predictions of pressure distributions along the vertical (Z) axis of the VCF. S1, S2, S3, S4, and S5 represent storms corresponding to 21 April 2006, 27 April 2006, 06 May 2006, 07 May 2006 and 27 May 2006 storms respectively. Qp is the peak flow rate of the storms. The order of flow rates in (B) is the same as in Plot (A); (r, ) and (0, ) represent the radius and center of the VC F respectively, in ra dial coordinates; (r/2, 0, zn) and (r/2, 0, zo) represent the top and the bottom of the mid-point of the RCF respectively, in Cartesian coordinates. Gage pressure (kPa) Radial distance (mm) 050100150200 H (mm) 0 10 20 30 40 50 (r, ) (0, ) (r/2, 0, zn) Cartridge depth (mm) S2 S4 S5 S1 S3 Qp + (A) (B) 0 0.1 0 0.2 0.3 0.4 0.5 0.6 (r/2, 0, z0) Qp + 500 400 300 200 100

PAGE 179

179 Figure 6-9 Comparison of measured and modeled temporal variation in effluent mass. Error bars represent standard deviation and ar e contained within the symbols. 29 April 2006 Q/Qmax 0.0 0.2 0.4 0.6 0.8 1.0 Measured M max =387 g Modeled M max =360 g 21 April 2006 00 02 04 06 08 10 Effluent M/Mmax 0.0 0.2 0.4 0.6 0.8 1.0 Hydrograph Modeled effluent Measured effluent Measured M max =14 g Modeled M max =12 g 06 May 2006 Effluent M/Mmax 0.0 0.2 0.4 0.6 0.8 1.0 Measured M max = 14.8 g Modeled M max =16.7 g 07 May 2006 Q/Qmax 0.00 0.25 0.50 0.75 1.00 Measured M max = 35 g Modeled M max = 32 g 27 May 2006 t/tmax 0.00.20.40.60.81.0 Q/Qmax 0.0 0.2 0.4 0.6 0.8 1.0 Effluent M/Mmax 0.0 0.2 0.4 0.6 0.8 1.0 Measured M max =29 g Modeled M max =26 g

PAGE 180

180 Figure 6-10 Measured versus mode led effluent concentrations as function of storm elapsed time. RPD is relative percent difference between measured and modeled data. 29 April 2006 Q/Qmax 0.00 0.25 0.50 0.75 1.00 t max =170 min Q max =25.3 L/s RPD = 10.5 % 06 May 2006 log10Effluent SSC [mg/L] 10 100 1000 t max =92 min Q max =0.3 L/s RPD = -15.2 % 07 May 2006 t/tmax 0.00.20.40.60.81.0 Q/Qmax 0.00 0.25 0.50 0.75 1.00 t max = 68 min Q max = 9.1 L/s RPD = 9.1 % 27 May 2006 t/tmax 0.00.20.40.60.81.0 Q/Qmax 0.00 0.25 0.50 0.75 1.00 log10Effluent SSC [mg/L] 10 100 1000 t max =50 min Q max =6.5 L/s RPD = 14.4 % 21 April 2006 00 02 04 06 08 10 log10Effluent SSC [mg/L] 10 100 1000 Hydrograph Modeled effluent Measured effluent Measured influent t max =70 min Q max =13.3 L/s RPD = 8.0 %

PAGE 181

181 Figure 6-11 Measured vs. modeled effluent d50 as a function of time for 5 storm events 21 April 2006 Effluent d50m (m) 0 100 200 300 400 500 tmax=70 min Qmax=13.3 L/s 29 April 2006 Q/Qmax 0.0 0.2 0.4 0.6 0.8 1.0 tmax=170 min Qmax=25.3 L/s 06 May 2006 Effluent d50m (m) 0 100 200 300 400 500 tmax=92 min Qmax=0.3 L/s 07 May 2006 t/tmax 0 .00.20.40.60.81.0 Q/Qmax 0.0 0.2 0.4 0.6 0.8 1.0 tmax=68 min Qmax=9.1 L/s 27 May 2006 t/tmax 0.00.20.40.60.81.0 Q/Qmax 0.0 0.2 0.4 0.6 0.8 1.0 Effluent d50m ( m) 0 100 200 300 400 500 Hydrograph Measured Modeled tmax=50 min Qmax=0.3 L/s

PAGE 182

182 CHAPTER 7 GLOBAL CONCLUSIONS This dissertation focused on a coupled experim ental and numerical approach to characterize particulate matter se paration by stormwater unit opera tions and processes for steady and transient hydrologic, hydraulic and pollutant loadings. A screened hydrodynamic separator and a radial cartridge filter were the two UOPs that were tested. The CFD model was validated with experimental data across a range of flow rates and particle size distributions, PSDs. Predictions from the numerical model were found to lie within 10 % of the measured data. Grid independence for the numerical model was demonstrated. This study demonstrated that a CFD model of the behavior of an HS for typical st ormwater flow rates, particle size gradations and levels of SSC could reproduce the PSD and SSC response of a screened HS. Post-processing the CFD predicti ons provided an in-dep th insight into the mechanistic behavior of the screened HS by m eans of three dimensional hydraulic profiles and particle trajectories. Results demonstrate that wh ile coarse particles are separated, settleable and suspended particles are largely eluted from the HS, even under clean sump and conditions of this study. The ability of the CFD model to reproduce treatment results across a range of flow rates and PSDs suggest that the calibrated and validated model can se rves as a foundation upon which design alternatives can be proposed. A calibrate d/validated CFD-based iterative approach to design of this HS as a preliminary unit operation has the potential to prov ide reduced prototyping costs with improved performance, as a result of carefully designed experimental matrices, focused on meeting coarse particulate contro l requirements for downstream treatment units. Particulate matter separation by a screen ed hydrodynamic separa tor for transient hydraulic and particulate loads observed in a r eal-time rainfall-runoff event was modeled by the application of the standard kturbulence model and a Lagrangian discrete phase model. Four

PAGE 183

183 discrete rainfall-runoff events were modeled in dividually and the modele d results agreed very well with the measured data (Absolute RPD < 10%). The CFD model was applicable across the entire range of flow rate and influent PSD variations and wa s able to accurately model both mass-limited and flow limited rainfall-runoff ev ents. Modeling the unsteady flow across the entire duration of the storm was tested against us ing a single design flow rate and the event mean concentration of PM for each event. It was observed that the PM separation behavior of the UOP varies significantly using the m ean, median and peak influent flow rates. Accurate modeling calls for including the flow variations acr oss the entire treated volume of runoff. This study demonstrated that a CFD model c ould reproduce the fate of a hetero-disperse PSD as a function of particle size for mass, conc entration and head loss behavior of a RCF for typical runoff loading rates, PSDs, and suspe nded sediment concentration (SSC) levels. A steady state solution was obtaine d for flow through the RCF a nd clean bed conditions were implemented in the experiment and in the model. CFD predictions provided an in-depth insight into the mechanistic behavior of the RCF by means of three dimensional hydraulic profiles, particle trajectories and radial and axial pressure distributions. The ability of the CFD model to reproduce tr eatment results across a range of flow rates and PSDs suggest that the cal ibrated/validated model can serve as a foundation upon which design alternatives can be proposed. A calibra ted/validated CFD-based iterative approach to design of this RCF as a unit opera tion has the potential to provid e reduced prototyp ing costs with improved performance, as a result of carefully designed experimental matrices, focused on PM control requirements for effluent discharges. This study combined PSD measurements using laser diffraction, media porosimetry, image analysis and material balances as well as more conventi onal gravimetric SSC

PAGE 184

184 measurements of PM and pressure sensor measurement. Such data are needed in the calibration and validation process for a defensib le porous media CFD model of a RCF. Overall, a CFD approach to modeling the pollutant removal characteristics of UOPs is a state-of-the-art approach to reducing the uncertainty that re sults from assuming ideal conditions, thus providing a more effective method for pollution control.

PAGE 185

185 LIST OF REFERENCES Aha mmed, M.M. and Chaudhuri, M. (1996). Sand-based filtration/adsorption media. J. Water Supply Res. T. 45, 67-71. Alkhaddar, R.M., Higgins, P.R ., Phipps, D.A, Andoh, R.Y.G. (2001). Residence Time Distribution of a Model Hydr odynamic Vortex Separator. Urban water, 2, 17-24. Alquier, M, Delmas, D, Pellerej, M. ( 1982). Improvement of Swirl Concentrator. J. Environ. Eng.-ASCE 108(2) 378-386 American Society for Testing and Materials, (2003). Automated Pore Volume and Pore Size Distribution of Porous Substan ces by Mercury Porosimetry. ( ASTM Designation: UOP57802) West Conshohocken, PA.. American Society for Testing and Materials (ASTM). (1993). Standa rd practice for dry preparation of soil samples for particle size analysis and determination of soil constants, in Annual book of standards Designation: D 421-85, Vol. 04.08, Philadelphia, 8. Anderson, J.D. (1995). Computational fluid dyn amics The basics with applications. 1st Ed., McGraw-Hill Inc.., USA, 23-31. Andoh, R.Y.G. and A.J. Saul (2003). The use of hydrodynamic vortex separators and screening systems to improve water quality. Water Sci. and Technol. 47(4), 175. Antohe, B.V. and Lage, J.L. (1997). A general two-equation macroscopic turbulence model for incompressible flow in porous media. Int. J. Heat Mass Tran., 40(13), 3013-3024. AWWA (1990). Water quality and tr eatment A handbook for community water supplies. 4thed. American Water Works Association, McGraw-Hill Inc, USA, 13.25-13.27. Ayoub, G.M., Koopman, B. and Pandya, N. (2001) Iron and aluminum hydroxyl (oxide) coated filter media for low-concentration phosphorus removal. Water Environ. Res ., 73(4), 478-485. Barth, T.J and Jespersen, D. (1989). The design and application of upwind schemes on unstructured meshes. T echnical Report AIAA-89-0366, AIAA 27th Aerospace Sciences Meeting Reno, Nevada, USA. Bertrand J. L., G. Chebbo, and A. Saget. (1998). Distribution of pollutant mass vs. volume in stormwater discharges and the first flush phenomenon. Water Res. 32(8), 2341-2356 Brennen, C. (2005). Fundamentals of Multiphase Flow (1st Ed.). Cambridge University Press, New York, 1-29. Bretscher, U., P. Krebs, and W.H.Hager (1992). Improvement of Flow in Final Settling Tanks. J. Environ. Eng.-ASCE 118(3), 307-321.

PAGE 186

186 Brombach, H. (1987). Liquid-solid separation at vortex-storm-overflows. Proc., 4th International Conference on Urban Storm Drainage (ICUD) Lausanne, Switzerland.,103108. Brouckaert, C.J., and C.A. Buckeley (1999). T he Use of Computationa l Fluid Dynamics for Improving the Design and Operation of Wate r and Wastewater Treatment Plants. Water Sci. and Technol., 40, 4-5, 81-89. Chen, J., Truesdail, S., Lu, F., Zhan, G., Belvin, C. and Koopman, B. (1998). Long-term evaluation of aluminum hydroxide-coated sand fo r removal of bacteria from wastewater. Water Res. 32(7), 2171-2179. Colandini, V. (1999) Effects of a porous paveme nt with reservoir structure on runoff water: Water quality and fate of heavy metals. Water Sci. and Technol., 39(2), 111-117. Curtis, J.S and van Wachem, B. (2004) Modeling Particle-Laden Flows: A Research Outlook. AIChE J. 50, 26382645. Darby, R. (2002). Chemical E ngineering Fluid Mechanics (2nd ed.). Marcel Dekker Inc, New York, USA, 394-396. Davila, J. and Hunt, J.C.R. (2000). Settling of Sm all Particles near Vorti ces and in Turbulence. J. Fluid Mech., 440, 117-145. Deininger, A., E. Holthausen, and P.A. Wilder er (1998), Velocity an d Solids Distribution in Circular Secondary Clarifiers: Full Scal e Measurements and Numerical Modeling, Water Res ., 32(10), 2951-2958. Do-Quang, Z., Cockx, A., Line, A. and Rousta n, M. (1999). Computational fluid dynamics applied to water and wastewater treatment facility modeling. Environmental Engineering and Policy 1, 137-147 Elgobashi, S.E. (1991). Particle laden turbulent flows: Direct simulati on and closure models. Appl. Sci. Res., 48, 301-314. Ergun, S. (1952). Fluid Flow through Packed Columns. Chem. Eng. Prog., 48(2), 89-94. Erickson, A.J., Gulliver, J.S. and Weiss, P.T. (2004). Enhanced sand filtration for stormwater phosphorus removal. J. Environ. Eng.-ASCE., 133(5), 485-497. Faram, M.G and Harwood, R (2003). A method for the numerical assessment of sediment interceptors. Water Sci and Technol., 47(4), 167-174. Farizoglu, B., Nuhoglu, A., Yildiz, E., Keskinle r, B. (2003). The performance of pumice as a filter bed material under rapid filtration conditions. Filtr. Separat ., 40, 41

PAGE 187

187 Fenner, R.A. (1998). Physical modeling of hydrodynamic vortex separato rs operating with underflow. J. Environ. Eng.-ASCE., 124(9), 881-886. Fenner, R.A. and Tyack, J.N. (1997). Scali ng laws for hydrodynamic vortex separators. J. Environ. Eng.-ASCE., 123(10), 1019-1026. Ferziger, J.J, Peric, M. (2002). Comput ational Methods for Fluid Dynamics. 3rd Ed., SpringerVerlag, Berlin, Germany. Field, R. and OConnor, T.P. (1997). Swirl Technology-Enhan cement of Design, Evaluation and Application. J. Environ. Eng.-ASCE., 122(8), pp 741-748 Finlayson-Pitts, B.J and Pitts, J.N. (2000). Che mistry of the Upper and Lower Atmosphere Theory, Experiments and Applications. 1st Ed., Academic Press, CA, USA, 365-368. Fluent 6.2 users manual. (2005) Fluent Inc. Lebanon, NH Freeze, A.R and Cherry, J.A. (1979). Groundwater. 1st Ed., Prentice Hall Inc., NJ, USA, 317320. Geldof, G., Jacobsen, P., and Fujita, S. (1994) Urban storm water infiltration perspectives. Water Sci. and Technol., 29(1-2), 245-254. Heist J. A., Davey A., Hawkins, R., Fitzgerald, J., a nd Warren, P. (2004) C DS use for treating sanitary wet weather flows. (Paper presente d in the proceedings of the World Water and Environmental Resources Congre ss, Salt Lake City, UT, USA) Hipp, J.A., Ogunseitan, O., Lejano, R. a nd Smith, C. S. (2006). Optimization of stormwater filtration at the urban/watershed interface. Environmental Sci Technology, 40, 4794-4801. Holdich, R. (2002). Fundamentals of Partic le Technology. Midland Information Technology and Publishing., UK, 5-17. Huber, W. (1993). Contaminant transport in surface water. D.R.Maidment, ed., Handbook of Hydrology, McGraw-Hill Book Co., Inc., New York, NY, USA. Igloria, R. A., Hathhorn, W. E., and Yonge, D. R. (1997). NOM and trace metal attenuation during storm water infiltration. J. Hydrol. Eng.-ASCE., 2(3), 120 Issa, R.I. (1986). Solution of im plicitly discretized fluid flow equations by operator splitting. J. Comput. Phys., 62, 40-65. Jayanti, S., and Narayanan, S. (2004). Comput ational Study of Particle -Eddy Interaction in Sedimentation Tanks. J. Environ. Eng.-ASCE 130(1), 37-49.

PAGE 188

188 Julien, P.Y. (1986). Concentration of Very Fine Silts in a Steady Vortex. J Hydraul Res, 24( 4) 255-264. Karl, J.R., and S.A.Wells (1999). Numerical Mode l of Sedimentation/Thickening with Inertial Effects. J. Environ. Eng.-ASCE 125(9), 792-806. Kim, J.-Y., Ma, J., Pathapati, S., Sansalone, J. ( 2004). Continuous deflective separation of non-colloidal particulate ma tter in rainfall-runoff. Proc. StormCon The 3rd Annual North American Surface Water Quality Conference & Exposition Palm Desert, CA, USA. Launder, B. E., and Spalding, D. B. (1974). The numerical computation of turbulent flows. Comput. Method. Appl. M., 3, 269-289. Lee, J. H., and Bang, K. W. (2000). Char acterization of urban stormwater runoff. Water Res ., 34(6), 1773. Letterman, R. D. (1999). Water Quality and Treatment: a handbook of community water supplies. McGraw Hill, New York Levenspiel, O. (1998). Chemi cal Reaction Engineering. 3rd Ed., Wiley, New Jersey. Li, Y., Buchberger, S.G. and Sansalone J.J. (1999 ). Variably saturated flow in a storm water partial exfiltration trench. J. Environ. Eng.-ASCE. 125(6), 556-565. Lim, M.T.N., Lam, S.W., Amal, R., Cathers, B., Pinson, D. (2002) Computational and experimental studies of floc be havior in a Vortex Separator. Proc.9th International conference on urban drainage (ICUD), Portland, Or, 303. Liu, D., Teng, Z., Sansalone J.J and Cartledg e, F. K. (2001) Surface characteristics of sorptive-filtration stormwater me dia II: higher specific gravity (s > 1.0). J. Environ. Eng.ASCE., 127(10), 879-888. McBean, N. R., W. Snodgrass, and I. Mostrenko (1997), Sample size needs for characterization pollutant concentratio ns in highway runoff J. Environ. Eng.-ASCE 123(10), 1061-1065. Mohammadi, B., and O.Pironneau (1994), Analysi s of the k-epsilon tu rbulence model. John Wiley and sons, New York. Morsi, S.A., Alexander, A.J. (1972) An investigation of particle trajectories in two-phase flow systems. J. Fluid Mech., 55(2), 193-208. Mueller,G.E. (1992). Radial void fraction distri butions in randomly packed fixed beds of uniformly sized spheres in cylindrical containers. Powder Technol., 72, 269.

PAGE 189

189 Naser, Gh., B.W. Karney, and A.A.Salehi ( 2005). Two-Dimensional Simulation Model of Sediment Removal and Flow in R ectangular Sedimentation Basin. J. Environ. Eng.-ASCE 131(12), 1740-1749. Nowakowski, A.F., Cullivan, J.C ., Williams, R.A., Dyakowski, T. (2004). Application of CFD to modeling of the flow in hydrocyclonesIs this a realizable options or still a research challenge? Miner. Eng., 17, 661-669. Okamoto, Y., Konugi, M. and Tsuchiya, H. (2002) Numerical simulation of the performance of a hydrodynamic separator. Proc., 9th International c onference on urban drainage (ICUD ), Portland, Or, 9. Patankar, S. (1980). Numerical h eat transfer and fluid flow (1st Ed.). Hemisphere Publishing Corporation, USA, 126-131 Paul, T. C., Sayal, S. K., Sakhuja, V. S., and Dhillon G. S. (1991). Vortex-settling basin design considerations. J. Hydraul. Eng.-ASCE., 117(2), 172-189. Petty, C.A. and Parks, S.M. (2001). F low predictions within hydrocyclones. Filtr. Separat., 38(6), 28. Pisano, W.C., and Brombach, H. (1994). Operati onal experience with vortex solids separators for combined sewer overflow (CSO) Control. Water Sci and Technol., 27(5-6). Qi, D., Lin, W (2006). TGrid : A New Grid Environment. Proc. First International MultiSymposiums on Computer and Computational Sciences Conference (IMSCCS) Hangzhou, Peoples Republic of China, 611-618. Ranade, V. V (2002). Computational flow mo deling for chemical reactor engineering (1st Ed.). Academic Press, London, UK. Reynolds, T. D and Richards, P.A. (1995). Unit Operations and Processes in Environmental Engineering. 2nd Ed., PWS Publishing Company, USA., 115-117. Rossman, L. A. (2007). Stormwater Mana gement Model Users Manual, Version 5.0, USEPA/600/R-05/040, Cincinnati, OH. Rushton, B. (2004). Broadway outfall stor mwater retrofit incor porating a CDS unit. Proc. World Water and Environmen tal Resources Congress Salt Lake City, Ut. Rushton, B. (2006). Broadway outfall stormwat er retrofit project (Phase II Monitoring CDS unit and constructed pond). Draft report to EPA and Southwest Florida Water Management District (SFWMD), 124. Sansalone, J. J, Hird, J.P., Cartledge, F.K. and Tittlebaum, M.E. (2005). Event-Based Stormwater Quality and Quantity Loadings from Elevated Urban Infrastructure Affected by Transportation. Water Environ. Res., 77(4), 348-365.

PAGE 190

190 Sansalone, J.J, Koran, J.M, Smithson, J.A, and Buchberger, S.G (1998). P hysical characteristics of urban roadway solids transported during rain events. J. Environ. Eng.-ASCE., 124(5), 427-440. Sansalone, J.J. (1999). Adsorptive-infiltration of metals in urban drainage media characteristics. Science of the Total Environment 235, 179-188. Sansalone, J.J. (2002). Physi cal and chemical nature of stormwater pollutants. Wet weather flow in the urban watershed, R.Field and D.Sullivan, eds., Boca Raton, Fl, 43-66. Sansalone, J.J. and Buchberger, S.G. (1997). Par titioning and first flush of metals in urban storm water. J. Environ. Eng.-ASCE., 123(2), 134-143. Sansalone, J.J. and Kim, J.Y ( 2007). Suspended Particle Destab ilization in Retained Urban Stormwater as a Function of Coagulant Dosage and Redox Conditions (in press). Water Res Sansalone, J.J. and S.G. Buchberg er (1997), Partitioning and first flush of metals in urban storm water, J. Environ. Eng.-ASCE 123(2), 134-143. Sansalone, J.J. and Teng, Z. (2004). In-situ storm water treatment and recharge through infiltration: Quality and Quantity Attenuation. J. Environ. Eng.-ASCE., 130(9), 990-1007. Sansalone, J.J. and Teng, Z. (2005) Transient Rainfall-Runoff Loadin gs to a Partial Exfiltration System: Implications for Water Quantity and Quality. J. Environ. Eng.-ASCE., 131(8), 1155-1167. Schueler, T.R. (1987) Controlling Urban Runoff: A Practical Manual fo r Planning and Designing Urban BMPs. Metropolitan Washington Counc il of Governments, Washington, DC. Schwarz T. and Wells S. (1999). Storm Water Particle Removal using Cross-Flow Filtration and Sedimentation. Advances in Filtrati on and Separation Technology, 12, ed. by W. Leung, American Filtrations and Separations Society, 219-226. Statie, E.C., Salcudean, M.E., Gartshore, I.S. (2001). The influence of hydrocyclone geometry on separation and fibre classification. Filtr. Separat., 38(6), 36. Stenstrom M.K., Lau S.-L. (1998). Oil and gr ease removal by floating sorbent in a CDS device. Technical Report, Dept. of Civil Eng. UCLA, Los Angeles, CA, Stovin, V.R., Saul, A.J. (1998). A Computational Fluid Dynamics (C FD) Particle Tracking Approach to Efficiency Prediction. Water Sci. and Technol. Vol. 37, No. 1, pp285-293. Stovin, V.R., Saul, A.J. (1996). Efficiency Prediction for Storage Chambers using Computational Fluid Dynamics. Water Sci. and Technol. Vol. 33, No. 9, pp163-170.

PAGE 191

191 Szalai, L., P. Krebs, and W. Rodi (1994), Simulation of Flow in Circular Clarifiers with and without Swirl, J. Hydraul. Eng.-ASCE., 120(1), 4-21. Tchobanoglous G., Burton F. L. and Stensel H. D. (2002). Wastewater Engineering: Treatment, Disposal, and Reuse. 4th Ed., McGraw Hill, New York. Tung, K.L., Chang, Y.L., Lai, J.Y, Chang, C. H and Chuang, C.J. (2004). A CFD study of the deep bed filtration mechanism for s ubmicron/nano-particle suspension. Water Sci. and Technol., 50(12), 255-264. Tyack, J.N. and Fenner, R.A. (1999). Computat ional fluid dynamics modeling of velocity profiles within a hydrodynamic separator. Water Sci and Technol., 39(9), 169-176. Urbonas, B. R. (1995). Recommended paramete rs to report with BMP monitoring data. J. Water Res. Pl.-ASCE 121(1), 23-34. USEPA (1999). Stormwater tec hnology fact sheet Hydrodynami c Vortex Separators, EPA 832F-99-017. U.S. Environmental Protection Agency (US EP A). (2000). Water quality assessment database 2000 305(b) Data, < http://oaspub.epa.gov/waters/w305b_report.nation > (June 14, 2006). Van Wachem, B. G. M., and A. E. Almstedt.(2003). Methods for multiphase computational fluid dynamics. Chem. Eng. J., 96, 81-98. Vandoormaal, J. P. and Raithby, G. D.(1984). Enhancements of the SIMPLE method for predicting incompressible fluid flows. Numer. Heat Transfer., 7,147-163 Versteeg, H., Malalasekera, W. (1995). An Intr oduction to Computationa l Fluid Dynamics: The Finite Volume Method Approach (1st Ed.). Prentice Hall, London, UK. Wakeman, R., Tarleton, S. (2005). Solid/Liquid Separation: Princi ples of Industrial Filtration. 1st Ed., Elsevier Ltd., UK, 38-43. Weib, G (1997). Vortex Separator Proposal of a Dimensioning Method. Water Sci and Technol., 36(8-9), 201-206. Wong, T.H.F. (1997). Continuous Deflective Se paration: Its Mechanism and Applications. (Paper presented at the Water Environm ent Federation Technical Exhibition and ConferenceWEFTEC Chicago, IL, USA) Zhou, S., J.A. McCorquodale and A.M. Godo (1994) Short Circuiting and Density Interface in Primary Clarifiers. J. Hydraul. Eng.-ASCE., 120(9), 1060-1080.

PAGE 192

192 BIOGRAPHICAL SKETCH Subbu-Srikanth Pathapati received his Bachel ors degree in electri cal and electronics engineering from the University of Madras, Chennai, India in 2001 and came to United States of America in Spring 2003 to pursue a graduate de gree Sri Pathapati will receive the degree of Doctor of Philosophy in environm ental engineering from the University of Florida in May 2008. His doctoral research was focused on experi mentation and numerical modeling of urban stormwater particulate matter control unit oper ations and processes. He worked under the guidance of Dr. John J. Sansal one in the Department of En vironmental Engineering and Sciences.