A Study on a Novel Sewer Overflow Screening System
Thesis
Submitted in Total Fulfilment of the Requirements for the Degree of
Doctor of Philosophy
By
Md Abdul Aziz
Student ID: 6632475
Faculty of Science, Engineering and Technology (FSET)
Swinburne University of Technology
Hawthorn, Victoria 3122
Australia
2016
Dedicated to my parents and wife
Md Abdul Aziz i
Declaration
The author hereby declares that this thesis, submitted in fulfilment of the
requirements for the Degree of Doctor of Philosophy, contains no material which has
been accepted for the award of any other degree or diploma, except where due reference
is made in the text. To the best of my knowledge, this thesis contains no material
previously published or written by another person, except where due reference is made
in the text. In places where the work is based on joint research or publications, this thesis
discloses the relative contribution of the respective workers or authors.
Part of this thesis have been copyedited and proofread by Dr Jillian Graham
(Articulate Writing Solutions), whose services are consistent with those outlined in
Section D of the Australian Standards for Editing Practice (ASEP).
Md Abdul Aziz
March, 2016
Md Abdul Aziz ii
Abstract
After heavy rainfall, sewer overflow spills into receiving water bodies, causing
serious concern for the environment and public health. This has led to a need to conduct
research in order to develop different types of screening devices. Some of the limitations
in existing screening processes include the expense involved in ongoing operation and
maintenance costs, poor capture efficiency, and blinding on the screens. Most sewage
overflow screens use sophisticated electro-mechanical systems. These systems are
expensive and complex, and the possibility of malfunctions is a cause for concern,
particularly in unstaffed remote locations. In this study, novel sewer overflow screening
devices have been developed and evaluated to overcome the limitations of low capture
efficiency and high initial and maintenance costs.
The set-ups of experimental screening devices involve significant cost and time.
To overcome this issue, the proposed sewer overflow screening device was analysed
using a 3D computational fluid dynamics (CFD) model. Plausibility checking of the CFD
model was done using an analytical model. The CFD model helped in designing the
location of perforations, inlet length and orientation. However, a limitation was
encountered in the attempt to reduce blinding of the perforations. To address this, a
revised design using comb separators instead of perforations was proposed.
Laboratory tests of the revised ‘comb separator’ were performed to determine the
trapping efficiencies for common sewer solids under different experimental conditions.
Comparisons of the ‘comb separator’ with the industry standard Hydro-JetTM suggest that
the former can perform better when there are blockages, and that it has a higher capture
efficiency during periods of low flow.
It was important to conduct modelling analysis to overcome physical limitations,
and to visualise a range of experimental conditions using CFD and experimental
analysis. An artificial neural network (ANN) model was adopted in this study, as it has
the capacity to intelligently predict the outcome of complex, non-linear physical systems
with relatively poorly-understood physio-chemical processes. The model successfully
predicted the experimental results with more than 90% accuracy, with an average
absolute percentage error of about 7%.
Md Abdul Aziz iii
It was also necessary to conduct sensitivity analysis of the comb separator to
understand its performance qualitatively and quantitatively. A multiple linear regression
(MLR) model was developed taking five input parameters (runtime, flows, effective comb
spacing, weir opening and layers of combs) and output capture efficiency for model
generation. The MLR model results suggest that the most significant predictor influencing
sewer solids capture efficiency is effective spacing, followed by flow discharge and
runtime.
The proposed ‘comb separator’ screening system shows good application
potential during testing in situations of low flow. Further research is recommended,
including a testing plan for high flow scenarios (such as flooding) and onsite testing of
the ‘comb separator’.
Md Abdul Aziz iv
Acknowledgments
I would like to express my deepest gratitude to Almighty Allah, the most merciful,
the most benevolent and the master of the judgement day, for giving me patience and
helping me through to completion of my study at Swinburne University of Technology.
I owe my deepest thanks to my principle coordinating supervisor, Associate
Professor Monzur Imteaz, for his continuous support and guidance, without which it
would not have been possible to complete this thesis. It was an honour to work with him.
During the latter stage of my research project, I was working full-time and studying part-
time, which meant that my thesis took longer to complete. I am indebted to my supervisor
for understanding my situation, and for continuing to show great faith in me.
It gives me pleasure to acknowledge my industry supervisor, Dr Don Phillips, who
was inspiring and supportive during this project. I am also greatly appreciative of the
contribution of Dr Shirley Gato-Trinidad, my co-supervisor. I am grateful to Dr Don
Phillips and Dr Shirley Gato for helping me to settle into in my research project, and for
their on-going guidance.
I would like to acknowledge Associate Professor Jamal Naser, Dr Nazmul Huda
and Dr Morshed Alam for their consistent support and contributions, especially in
developing ideas in the field of Computational Fluid Dynamic (CFD) Modelling. I am
indebted to Dr Tanveer Ahmed Choudhury for introducing me to the world of Artificial
Intelligence. I thank Dr Sylvia Mackie, Carol Farr, Dr Samsuzzoha, Dr Mahabubur
Mollah, Taha Mollah, Paul Fennell and Dr Jan for their assistance in reviewing my thesis,
journal articles and conference papers. I would also like to thank my exam review
committee, including Dr Arul Arulrajah, Dr Scott Rayburg and Dr Richard Manasseh, who
reviewed my work and provided valuable suggestions that contributed immensely to my
completion of this research project.
I am indebted to Swinburne University of Technology for granting me the
Swinburne University postgraduate research award (SUPRA) to facilitate my research
and to support me financially in the first two years of my candidature. I am also grateful
to the University for their flexibility in allowing me to switch from full-time to part-time
student.
I would like to thank my parents, sisters and brother for their constant motivation
and encouragement. My father was the first to inspire me towards my PhD research
Md Abdul Aziz v
during my childhood, and must now be a proud man in heaven. To my wonderful mother
I am particularly and profoundly grateful. When others doubted my educational ability,
she stood by me, giving me the support and courage to struggle on against the odds.
Special thanks go to my wife. She has made me feel as if all the pleasure in study
were mine and all the pain hers during this tough seven years. I honour Almighty Allah
for making it possible for her to share this journey. Her patience and tolerance during the
more difficult times are noted and deeply appreciated.
I am very proud to work for the Wimmera Catchment Management Authority, the
best working environment one could imagine. I am grateful for the affection and respect
this organisation has bestowed on me. I thank Mr Dave Brennen (CEO Wimmera CMA)
for allowing me to continue my study while working full-time, and also my manager Mr
Paul Fennell, an admirable and genuine gentleman whose friendship I treasure.
Thanks to Dr Amin and his family, Sadi and his family, Tanveer and his family,
Morshed, Maruf, Nazmul and their families, and all our family friends in Horsham and
Melbourne, who have shared lots of laughter and mateship. Many thanks to my nephews
and nieces (Aurna, Purba, Piyal, Shreya, Arpon and Megh) who have been a great source
of inspiration; they are fine examples of peace, tranquillity and innocence. Last but not
least, thanks to all of those friends who may not be aware how much they helped and
inspired me. If I have forgotten anyone, this is a shortcoming on my part, which I hope will
be forgiven.
Table of Contents
Md Abdul Aziz vi
Table of Contents
Chapter 1 INTRODUCTION ....................................................................... 1
1.1 Background ........................................................................................................ 2
1.2 Problem Statement ............................................................................................ 3
1.3 Objective of this Study ...................................................................................... 4
1.4 Research Contributions .................................................................................... 5
1.5 Thesis Structure and Overview ........................................................................ 6
Chapter 2 LITERATURE REVIEW ............................................................. 8
2.1 Introduction ........................................................................................................ 9
2.2 Summary of Current Screening Applications ................................................ 12
2.2.1 Static screening ............................................................................... 13
2.2.2 Mechanical and Electrical screening devices ................................ 14
2.3 Methods to improve Sewage Overflow Screening ........................................ 23
2.3.1 Hydrodynamics Applications .......................................................... 23
2.3.2 Experimental Investigations ........................................................... 25
2.3.3 Artificial Neural Network (ANN) Applications ................................ 26
2.3.4 Sensitivity Analysis to Model Results ............................................ 28
2.4 Identification of Research Needs.................................................................... 29
2.5 Summary ...................................................................................................... 30
Chapter 3 RESEARCH METHODS .......................................................... 32
3.1 Introduction ...................................................................................................... 33
3.2 Research Questions ........................................................................................ 33
3.3 Research Process ............................................................................................ 34
3.4 Research Design.............................................................................................. 36
3.4.1 Computational Fluid Dynamic (CFD) Analysis .............................. 37
3.4.2 Laboratory Experiments .................................................................. 37
3.4.3 ANN model to complement deterministic approach ..................... 38
Table of Contents
Md Abdul Aziz vii
3.5 Analysis Procedure ......................................................................................... 38
3.5.1 CFD and Analytical Modeling .......................................................... 38
3.5.2 Experimental Investigation ............................................................. 39
3.5.3 ANN Modeling .................................................................................. 39
3.5.4 Sensitivity Analysis ......................................................................... 39
3.6 Summary ...................................................................................................... 40
Chapter 4 HYDRODYNAMIC ANALYSIS ................................................. 42
4.1 Introduction ...................................................................................................... 43
4.2 Screening Concept .......................................................................................... 46
4.3 Development of the Analytical Model ............................................................. 48
4.4 Computational Fluid Dynamics (CFD) Model ................................................. 51
4.4.1 Finite Volume Method (FVM) ........................................................... 53
4.4.2 Multiphase Flow Modelling ............................................................. 54
4.4.3 Approaches to Multiphase Modelling ............................................. 55
4.4.4 Euler-Lagrange Approach ............................................................... 55
4.4.5 Euler-Euler Approach ...................................................................... 55
4.4.6 Model Geometry and Computational Methodology ....................... 58
4.4.7 Boundary Conditions ...................................................................... 64
4.4.8 Explaining CFD Results .................................................................. 65
4.4.9 Plausibility check of the CFD model .............................................. 68
4.5 Discussion of Results ..................................................................................... 69
4.5.1 Discussion of Hydrodynamic results ............................................. 69
4.5.2 Discussing Location of Circular holes ........................................... 75
4.5.3 Discussion of the Inlet performance .............................................. 76
4.5.4 Standard Weir orientation ............................................................... 78
4.6 Limitation of Screening Device ....................................................................... 80
4.7 Summary ...................................................................................................... 80
Table of Contents
Md Abdul Aziz viii
Chapter 5 IMPROVEMENT OF THE SCREENING DEVICE ...................... 82
5.1 Introduction ...................................................................................................... 83
5.2 Methodology used in the Experiment ............................................................ 86
5.2.1 Data Collection................................................................................. 86
5.3 Test Procedure ................................................................................................ 93
5.3.1 Experimental Conditions Used ....................................................... 93
5.4 Discussions of Results ................................................................................. 101
5.4.1 Sewage solids more than 10 mm in diameter .............................. 101
5.4.2 Sewage solids less than 10 mm in diameter ................................ 102
5.4.3 Performance comparison of Comb Separator and Hydro-JetTM 104
5.5 Limitations of the Experiment ....................................................................... 105
5.6 Summary .................................................................................................... 106
Chapter 6 ANN MODEL TO COMPLEMENT CFD AND LABORATORY TESTING ......................................................................................................... 108
6.1 Introduction .................................................................................................... 109
6.2 Artificial Neural Network (ANN) .................................................................... 111
6.3 Description of Network Structure ................................................................. 116
6.3.1 Artificial Neuron Model .................................................................. 116
6.3.2 Multi-Layer Feed Forward Neural Network Structure .................. 117
6.4 Network Learning .......................................................................................... 117
6.4.1 Back propagation algorithm ......................................................... 118
6.4.2 Levenberg-Marquardt Algorithm .................................................. 118
6.4.3 Resilient Back Propagation Algorithm ......................................... 119
6.5 Data Collection and Pre-processing ............................................................. 120
6.5.1 Creation of Database ..................................................................... 120
6.6 Result Analysis and Discussion ................................................................... 123
6.7 ANN Model Validations.................................................................................. 126
6.8 Summary .................................................................................................... 127
Table of Contents
Md Abdul Aziz ix
Chapter 7 SENSITIVITY ANALYSIS OF THE COMB SEPARATOR ....... 129
7.1 Introduction .................................................................................................... 130
7.1.1 Objective ........................................................................................ 131
7.2 Background .................................................................................................... 131
7.2.1 Developing a Multiple Linear Regression (MLR) Model .............. 134
7.2.2 Summary of the Model .................................................................. 139
7.2.3 Development of the Dataset Using Sampling Techniques .......... 145
7.3 Results and Discussion ................................................................................ 149
7.3.1 Relative Significance of the Input Parameters ............................. 149
7.3.2 Selection of the Input Parameters ................................................ 150
7.3.3 Impact of Effective Spacing on Capture Efficiency ..................... 152
7.3.4 Impact of Flow on Capture Efficiency .......................................... 152
7.3.5 Runtime Impact on Capture Efficiency......................................... 153
7.4 Summary .................................................................................................... 154
Chapter 8 CONCLUSIONS .................................................................... 157
8.1 Introduction .................................................................................................... 158
8.2 Research Summary ....................................................................................... 158
8.3 Knowledge Contributions ............................................................................. 159
8.4 Limitations .................................................................................................... 161
8.5 Future Research ............................................................................................ 162
References ........................................................................................... 164
Appendix A:Experimental Data ............................................................ 179
List of Figures
Md Abdul Aziz x
List of Figures Figure 2.1: Frequency of Trash and Litter from Various Sources (Source: [156]) ........ 12
Figure 2.2: Static bar screens in operation (source: [3]) .............................................. 13
Figure 2.3: Brush-Raked fine screen static screen ...................................................... 14
Figure 2.4: Typical mechanical bar screens ................................................................ 16
Figure 2.5: Internally Fed Rotary Screen..................................................................... 18
Figure 2.6: Centrifugal screen ..................................................................................... 19
Figure 2.7: The cut-away view of the Rotary-Jet Screen ............................................. 20
Figure 2.8: Schematic representation of the Hydro JetTM ............................................ 22
Figure 2.9: A typical artificial neuron k ........................................................................ 27
Figure 3.1: Flow Chart of the Current Research Plan36 (Where RQ stands for Research
Questions) .................................................................................................................. 36
Figure 3.2: Flow chart of the current research plan ..................................................... 41
Figure 4.1: Schematic diagram of the proposed sewage overflow screening device ... 47
Figure 4.2: Front views of the proposed device under different phases ....................... 47
Figure 4.3: Breakdown of the flow components of the experimental device ................ 50
Figure 4.4: Geometric details of the screener device .................................................. 59
Figure 4.5: Position 1 (condition 1) is the inlet parallel to the ogee weir ...................... 63
Figure 4.6: Position 2 (condition 2) is the inlet perpendicular to the ogee weir ............ 63
Figure 4.7: Boundary conditions used in the CFD model ............................................ 64
Figure 4.8: Water levels over the weir at different locations ........................................ 66
Figure 4.9: Volume fraction of water at inlet parallel (position 1) to ogee weir ............. 67
Figure 4.10: Volume fraction of water at inlet perpendicular (position 2) to ogee weir . 67
Figure 4.11: Comparison of flow velocities over the top of the ogee weir .................... 68
List of Figures
Md Abdul Aziz xi
Figure 4.12: Comparison of flow velocities 6cm downstream of the ogee weir ............ 68
Figure 4.13 Comparison of water level along the flow for condition 1 .......................... 69
Figure 4.14 Comparison of water level along the flow for condition 2 .......................... 70
Figure 4.15: Velocity vector at the inlets parallel (left) ................................................. 71
Figure 4.16: Velocity vector at perpendicular (right) to the ogee weir .......................... 71
Figure 4.17: Comparison of flow velocities along the width for condition 1 .................. 72
Figure 4.18: Comparison of flow velocities along the width for condition 2 .................. 72
Figure 4.19: Pressure variation at condition 1 ............................................................. 73
Figure 4.20: Pressure variation at condition 2. ............................................................ 73
Figure 4.21: Shear stress distributions for the inlet parallel to the weir width ............... 74
Figure 4.22: Comparison of shearing stress along the bottom of the curved surface... 75
Figure 4.23: Comparison of water levels along the flow for condition 1, water level on the
top of the weir, 3cm and 6 cm downstream ................................................................. 76
Figure 4.24: Comparison of water levels along the flow for conditions 2, water level on
the top of the weir, 3cm and 6 cm downstream ........................................................... 76
Figure 4.25: Impact of device inlet position on the wave reflection viewed from the back
of the weir with a lateral inflow .................................................................................... 77
Figure 4.26: The Waterways Experimental Station (WES) standard spillway shapes .. 78
Figure 4.27: CFD results viewed from the back of the weir with a lateral inflow on four
standard inlet orientations as suggested by the U.S. Army Engineers Waterways
Experimental Station ................................................................................................... 79
Figure 5.1: Experimental set ups for the Comb Separator ........................................... 86
Figure 5.2: Flow diagram of the revised screening experimental works ...................... 87
Figure 5.3: Concept diagram of target capture efficiency curve ................................... 88
Figure 5.4: Experimental set-up for the proposed sewage overflow screening device . 89
List of Figures
Md Abdul Aziz xii
Figure 5.5: Operation procedure of new sewage overflow screening device- Phase 1.89
Figure 5.6: Operation procedure of new sewage overflow screening device-Phase 2 . 90
Figure 5.7: The design parameter of the ball valve chamber ....................................... 91
Figure 5.8: Vertical position of the Comb Separator in the device ............................... 94
Figure 5.9: Top view of the position of the Comb Separator ........................................ 94
Figure 5.10: Capture of sewage solids during an experimental run ............................. 95
Figure 5.11: Capture of sewage solids after an experimental run ................................ 95
Figure 5.12: Mixing of sewer solids on to the Comb Separator device ........................ 96
Figure 5.13: Comb Separator is in operation, nappe clear the retention screen .......... 96
Figure 5.14: Sewage solids used in the test ................................................................ 97
Figure 5.15: Comb Separator in operation .................................................................. 98
Figure 5.16: Capture efficiency of sewer solids at different experimental set ups ...... 102
Figure 5.17: Effective comb spacing (mm) against average capture efficiency (%) and
flow (l/s per metre length of weir) .............................................................................. 104
Figure 6.1: Conceptual diagram showing an analogy of the work principal between the
human brain and the ANN model .............................................................................. 111
Figure 6.2: Conceptual diagram of input-output and weight adjustment .................... 112
Figure 6.3: Demonstration of over-fitting for a function approximating ANN .............. 115
Figure 6.4: A Non-linear model of an artificial neuron................................................ 117
Figure 6.5: Block diagram of proposed ANN model. ................................................. 122
Figure 6.6: Comparison of different node in the 1st and 2nd hidden layer ................... 124
Figure 6.7: Comparison of experimental and ANN predicted capture efficiency ........ 125
Figure 6.8: Regression Value for training, validation and test results ........................ 126
Figure 6.9: Comparison of experimental and model results for validation dataset.
Experiment numbers 1 to 4 were conducted on a specific experimental setup and then
List of Figures
Md Abdul Aziz xiii
the flow was changed for experiments 5 to 8. Solids used for different test numbers: 1
and 5 = cigarette butts, 2 and 6 = condoms, 3 and 7 = tampons, and 4 and 8 = wipe
papers. ..................................................................................................................... 127
Figure 7.1: Flow chart of the methodology adopted in the sensitivity analysis. .......... 133
Figure 7.2 Regression Standardized predicted value against residual ...................... 136
Figure 7.3: Normal distribution plot of the Standardized Residual vs Frequency ....... 137
Figure 7.4: Experimental data shows Observed vs Expected Cumulative probability 138
Figure 7.5: P-P plots for the runtime predictor ........................................................... 146
Figure 7.6: Data on both sides of normal for runtime ................................................ 146
Figure 7.7: P-P plots for the runtime predictor ........................................................... 147
Figure 7.8: Data on both sides of normal for flow ...................................................... 147
Figure 7.9 P-P plot of effective spacing ..................................................................... 148
Figure 7.10 Data on both sides of normal for effective spacing ................................. 148
Figure 7.11: Relationship between Effective spacing (mm) and Capture Efficiency (%)
................................................................................................................................. 152
Figure 7.12: Relationship between the Flowrate (l/s) and Capture Efficiency (%) ...... 153
Figure 7.13: Relationship between Runtime (min) and Capture Efficiency (%) .......... 154
List of Tables
Md Abdul Aziz xiv
List of Tables
Table 2.1: Sewer solids available in the urban sewage treatment [156] ...................... 10
Table 2.2: Regulatory requirement for aesthetics [147]............................................... 25
Table 4.1: Calculation of hydrodynamic parameters from analytical results ................ 51
Table 5.1 Different experimental setups used for sewage solid less than 10 mm diameter
................................................................................................................................... 98
Table 5.2 Comb Separator Testing at Experimental Set up 1 ...................................... 99
Table 5.3 Comb Separator Testing at Experimental Set up 2 ...................................... 99
Table 5.4 Comb Separator Testing at Experimental Set up 3 ...................................... 99
Table 5.5 Comb Separator Testing at Experimental Set up 4 .................................... 100
Table 5.6 Comb Separator Testing at Experimental Set up 5 .................................... 100
Table 5.7 Comb Separator Testing at Experimental Set up 6 .................................... 100
Table 5.8 Experimental set ups at five different conditions for sewage solids more than
10 mm in diameter .................................................................................................... 101
Table 5.9 Capture efficiency with different experimental set ups ............................... 103
Table 5.10 Comparative performance of the Hydro-JetTM and the Comb Separator .. 105
Table 6.1 Comparison of different training paradigms ............................................... 125
Table 7.1 Provides descriptive statistics of the Comb Separator experimental data set
................................................................................................................................. 140
Table 7.2 Correlations of different parameters .......................................................... 140
Table 7.3 Summary for the multiple linear regression (MLR) model .......................... 141
Table 7.4 ANOVA table for the MLR model ............................................................... 142
Table 7.5 Coefficient of different parameters ............................................................ 143
Table 7.6 Schematic diagram of SaSAT data generation .......................................... 145
List of Tables
Md Abdul Aziz xv
Table 7.7: Results using the Latin Hypercube Sampling (LHS) method for 10,000 data
................................................................................................................................. 150
Table 7.8 Comparison between initial and final model results ................................... 151
List of Notations
Md Abdul Aziz xvi
List of Notations
Symbol Unit Description
ρ kg/m3 Density
μ m2/s Dynamic viscosity
P Pascal Atmospheric pressure
gx m/s2 Gravitational force in X direction
gy m/s2 Gravitational force in Y direction
q m2/s Flow per unit width
Pascal Shear stress at the boundary
v m/s Average velocity
Q m3/s Total discharge
L m Lateral crest length or width
He m Total head upstream from the crest
l/s l/s Liter/second
l/s/m l/s/m Litre/second/meter
mm mm Millimetre
u m/s Velocity in the x- directions
v m/s Velocity in the y- directions
w m/s Velocity in the z (vertical)- directions
Hd m The depth of water upstream of spillway
C0
Discharge coefficient
List of Notations
Md Abdul Aziz xvii
Symbol Unit Description
N No Number of phases
k Volume fraction
k Density of phase
vk Phase velocity
k Pa.s Molecular viscosity
tkT Reynolds stress
k Kronecker delta function
tk Turbulent viscosity
SItc
, Sato’s viscosity
k Turbulent kinetic energy
Dissipation rate of turbulent kinetic energy
DC Drag coefficient
Reb Bubble Reynolds number
c Kinematic viscosity
Db mm Bubble diameter
d Dispersed phase
∆t sec Time step increment in the model
w Shear stress
Uτ m/s The friction velocity
C.E Capture Efficiency
kpw Synaptic weight
List of Notations
Md Abdul Aziz xviii
Activation function
px Input signal
H Hessian matrix
J Jacobian matrix
zk Old parameter value
R Regression coefficient
MSm Average Improvement in Prediction by Model
MSr Average Difference between Model and Observed data
df Degree of freedom
List of Acronyms
Md Abdul Aziz xix
List of Acronyms
Acronym Description
AMP2 Asset Management Plan 2
ANN Artificial neural network
AVL Fire Computational Fluid Dynamic software
BP Back propagation algorithm
BR Bayesian regularization algorithm
CFD Computation Fluid Dynamics
CMC Center for Marine Conservation
CSO Combined Sewer Overflow
CVs Control Volumes
EPA Environmental Protection Agency
FEM Finite Element Method
FDM Finite Difference Method
FLUENT Computational Fluid Dynamics software
FVM Finite Volume Method
HDVS Hydrodynamic Vortex separation
LHS Latin Hypercube Sampling
LM Levenberg-Marquardt algorithm
MLP Multi-layer perceptron
MLR Multiple Linear Regression
RP Resilient back propagation algorithm
SaSAT Sampling and Sensitivity Analyses Tools
Sewer Refers to a channel or pipe or collection of them which carry liquid waste
Sewage The liquid waste (also containing solids) which flows along the sewer(s)
Sewerage The pipes, pumps, and infrastructure and is frequently be referred to as a sewerage system
List of Acronyms
Md Abdul Aziz xx
SM Spectral Methods
SPSS Statistical Package for the Social Science
SSO Sanitary Sewer Overflow
STAR-CD CD-adapco's legacy CFD package
TM Testing Method
UKWIR UK Water Industry Research
VOF Volume of Fluid
VOF Volume of Fluid
ZPRED Z Predicted
ZRESID Z Residual
Outcome from this Research
Md Abdul Aziz xxi
Outcome from this Research
Journal Papers
1) Aziz, M. A., Imteaz, M., Choudhury, T. A., & Phillips, D., 2013a,
‘Applicability of artificial neural network in hydraulic experiments using a new sewer
overflow screening device’, Australian Journal of Water Resources, vol.17, no.1,
pp.77-86.
2) Aziz, M. A., Imteaz, M., Naser J., & Phillips, D., 2013b, ‘Hydrodynamic
Characteristics of a New Sewer Overflow Screening Device: CFD Modelling and
Analytical Study’, International Journal of Civil and Environmental Engineering, vol. 7,
no.1, pp.71-76.
3) Aziz, M. A., Imteaz, M., Huda M., & Naser J., 2014a, ‘Optimising inlet
condition and design parameters of a new sewer overflow screening device using
numerical modelling technique‘, Journal of Water, Sciency and Technology vol.70,
no.11, pp.1880-1887.
4) Aziz, M. A., Imteaz, M., Rasel, H.M., & Phillips D., 2015a, Development
and Performance Testing of ‘Comb Separator’, A Novel Sewer Overflow Screening
Device. International Journal of Environment and Waste Management, vol.16, No.3,
2015.
5) Aziz, M. A., Imteaz, M., Samsuzzoha, M. A., and Rasel, H.M., 2015b,
Sensitivity Analysis on the Pollutant Trapping Efficiencies of a Novel Sewerage
Overflow Screening Device. Revising (March 2016) Journal of Hydro-informatics
6) Aziz, M. A., Imteaz, M., 2016a, -‘A Literature Review on Research
Methodologies on Sewerage Overflow Screening Devices’. To be submitted
International Journal of Water.
Peer reviewed conference papers
1) Aziz, M. A., Imteaz, M., J. Naser, Nazmul H., & Phillips, D. 2010,
‘Hydrodynamic Characteristics of a proposed sewer overflow screening device’, at
The 5th Civil Engineering Conference in the Asian Region, 8-12 August, Sydney
Australia.
2) Aziz, M. A., Imteaz, M., Choudhury, T. A. & Phillips, D. I. 2011, ‘Artificial
Neural Networks for the prediction of the trapping efficiency of a new sewer overflow
screening device’, 19th International Congress on Modelling and Simulation, Perth,
Australia. Indexed in Scopus
Outcome from this Research
Md Abdul Aziz xxii
3) Aziz, M. A., Imteaz, M., Samsuzzoha, M., & Phillips, D., 2013c,
‘Sensitivity analysis for a proposed sewer overflow screening device’, 20th
International Congress on Modelling and Simulation, Adelaide, 2 to 5 December,
Australia.
4) Aziz, M. A., Imteaz, M., Nazmul H., & Jamal N., 2013d, ‘Understanding
functional efficiency of a sewer overflow screening device using combined CFD and
analytical modeling’, 20th International Congress on Modelling and Simulation,
Adelaide, 2 to 5 December, Australia.
5) Aziz, M. A., Imteaz, M., Rasel, H.M., Phillips, D., 2015c, ‘Performance
Testing of ‘Comb Seperator’ –A Novel Sewerage Overflow Screening Device’,
ASEAN- Australian Engineering Congress on Innovative Technologies for
Sustainable Development and Renewable Energy 11-13 March.
6) Aziz, M. A., Imteaz, M., Rasel, H.M., and Samsuzzoha, M., 2015d,
‘Parameter Sensitivity Using Sampling Technique for a proposed ‘Comb Separator’,
A sewer overflow screening device’, ASEAN- Australian Engineering Congress on
Innovative Technologies for Sustainable Development and Renewable Energy 11-
13 March.
Chapter 1: Introduction
Md Abdul Aziz 1
Chapter 1
Introduction
Chapter 1: Introduction
Md Abdul Aziz 2
1.1 Background
During heavy, long-lasting downpours of rain, the urban sewer system is not able
to carry the excess water; hence some of this excess water flows into the open creek
system, carrying with it a lot of sewer solids. These sewer solids are dispersed,
suspended or washed into the creeks and rivers. They eventually settle, creating odours
and a toxic/corrosive atmosphere in the mud deposits on riverbeds. The solids create
additional problems, either through their general appearance (increasing dirtiness) or
through the actual presence of specific, objectionable items, such as floating debris,
sanitary and faecal matter, scum, or even parts of car tyres. These sewage overflows
have physical, chemical and biological effects on the surrounding environment. The raw
sewage has the potential to carry pathogens, including bacteria, viruses, protozoa,
helminths, inhalable moulds and fungi. These pathogens can cause life-threatening
ailments, including diseases such as cholera, dysentery, infectious hepatitis and severe
gastroenteritis [121]. Sewage overflow also has an adverse impact on the environment
by increasing pollutants, including plastic and paper products. This sewage sludge is
also a key concern for environmental scientists and engineers [65].
Some mitigating options to segregate sewage solids have been adopted include
temporary holding tanks at sewage treatment plants, real-time control of sewerage
systems, enlarged upstream sewers to provide transient storage, separation of storm
and sewage flows, and various screening devices in separate and combined sewage
overflow (CSO) chambers. In most cases screening is the only economically-viable
method [60].
Screening is a process that should be automated in order to ensure operational
safety and to reduce aesthetic pollutants. Moreover, a floatable control is preferred by
most o proposed and existing environment regulation agencies. This requirement has
triggered the need for research into the construction of an efficient and effective
screening devices and screening handling systems. To optimise their use in the actual
environment, especially at unmanned locations where there is a requirement for
minimum maintenance.
Chapter 1: Introduction
Md Abdul Aziz 3
1.2 Problem Statement
Past studies have used a number of different screening systems in sewer
overflow locations. The initial screens used were stationary, which caused problems
because sewage solids clogged the screens. Most ‘conventional’ screening systems
utilise electro-mechanical components to facilitate the screening process [131].
However, given the harsh unmanned remote environment of many sewer overflow
device locations, this is clearly not ideal. Blocking and seizure of moving parts are
common maintenance problems, and in many cases, electrical failure necessitates an
onerous maintenance commitment [10]. Moffa [118] used a rotary screen consisting of a
large rotating angled-drum to maximise dewatering, with the screenings travelling up to
the drum, where they are removed from the unit. Metcalf and Eddy [117] proposed a
series of screens attached to a cage that rotates around a vertical axis creating
centrifugal force. The flow enters from the base and flows upward to a deflection plate,
where pollutants are collected from outside the cage. Hydrodynamic vortex separation
(HDVS) was another popular screening concept developed in the early 1960s. In the first
generation, HDVS was found to be effective in retaining 70% of the pollution load
Smisson [151]. A second-generation HDVS developed by the American Waterworks
association and the Environmental Protection Authority (EPA) was reported by Field [64].
The third-generation device was commercially patented in the 1980s as the Storm King®
overflow. The HDVSs went through a series of performance evaluations in Europe,
North-America, and Japan [20]. Unfortunately aesthetic solids of neutral buoyancy were
not trapped in HDVSs [16].
A state-of-the-art review of different screeners is provided by Saul [144] & [146].
A recent update on this literature can be found in the work of Madhani and Brown [106].
The literature suggests that screens need to have the ‘self-cleansing’ mechanism. To
overcome this challenge, a non-powered self-cleansing screening system that can
capture neutrally-buoyant aesthetic solids greater than 6mm in two dimensions was
tested by Smith and Andoh [152]. Faram et al. [59] tested a hydro jet device installed in
the USA, Australia and mainland Europe. However, in most cases the device was directly
associated with blockages in the sewerage system. The most conventional screening
systems usually utilise electro-mechanical components to facilitate such a process [131].
However, given the harsh unmanned remote operating environment of the sewer
overflow device, this is not ideal [10].
Chapter 1: Introduction
Md Abdul Aziz 4
Some common drawbacks in the available commercial screening devices include
inadequate screening capacity, the requirements for external power and the high cost
[153]. The aim of this research was to create a novel sewer overflow screening device
free of the existing limitations. The focus in this research was to ensure that the proposed
screen is inexpensive, has no moving parts, and is free of sophisticated electrical-
mechanical switching systems.
Two different types of sewer overflow screens are analysed in the current
research. The first is a gross pollutant screening device, which comprises a solid sewage
trapping device with an ogee weir. The ogee weir spillway possesses excellent hydraulic
features in terms of flow efficiency, and relatively good flow measuring capabilities. In
the device studied, a deviation from the traditional weir flow was considered, with
construction distinctions such as upstream flow conditions, reflection waves due to a
short device boundary, and different shapes and sizes of construction that changed flow
properties. The second screen proposed is the Comb Separator, which uses a series of
combs rather than the circular holes of the first device. Both these screens have no
moving parts, and would work more efficiently in harsh environmental conditions. A
detailed description of the screens is provided in Chapters 4 and 5.
1.3 Objective of this Study
The objective of this research was to design an efficient and effective sewer
overflow screening device. This thesis describes the development of a sewer
overflow screen that can overcome most of the key limitations of existing sewage
overflow screens.
The research investigation included design of a concept screen, testing of the
performance of the screen using computer model and laboratory experiments. This
research endeavor to innovate a novel sewer overflow screen with the following
features:
High efficiency in trapping pollutants
Minimal blockages on the screen
Automation for effective use in remote unstaffed locations
Operational safety with a bypass channel
Chapter 1: Introduction
Md Abdul Aziz 5
Floatable control to meet environmental regulations
Low maintenance and operational costs
No need for a sophisticated electro-mechanical switching system
Suitable for unmanned remote locations
1.4 Research Contributions
This research makes the following contributions to the discipline of sewer
overflow screening applications in the urban sewerage system:
The literature review revealed several gaps in the research relating to sewer
overflow screens. The scope of this research was designed to address these issues.
An analytical solution of the Navier-Stokes equation was developed for the
proposed model, which was then compared with the results obtained from computational
fluid dynamic (CFD) analysis. Results of the analytical solution confirmed the plausibility
of the CFD model developed for this study [10]. Due to lack of proper experimental data
for the validation exercise, an analytical model was used instead to check the
performance of the CFD model [11].
The construction of experimental set ups involved significant cost and time;
moreover, a proposed concept for a novel gross pollutant screen had to be proven. To
overcome these problems, a state-of-the-art CFD model was developed using the Euler-
Euler approach to study the hydrodynamic characteristics of the sewer overflow
screening device. The results of the CFD model predicted a number of important design
parameters such as flow distribution, velocity distribution, pressure distribution and
sloshing behaviour across the device. The analysis helped to understand the device
orientation, as well as some other critical design parameters. Results obtained from the
CFD model formed the basis for optimising design parameters of the laboratory scale
experimental set up [11].
The physical experiments allowed for a certain number of trials for
experimental set ups. To visualise a range of different conditions within and outside the
physical limitations of the experiment, it was important to do modelling analysis. In the
current research, the sewage solids tested had different densities, which made it too
complex to model using the CFD model. These challenges were overcome by using an
Artificial Neural Network (ANN) model. The ANN can successfully predict the
experimental results with more than 90% accuracy, with an average absolute percentage
error of around 7% [9].
Chapter 1: Introduction
Md Abdul Aziz 6
A novel screening device called the Comb Separator was developed and
tested. The comb separator can capture larger sewerage solids more than 10mm with
over 95% capture efficiency. Two key improvements were recorded in the comparison
analysis. Firstly, there was minimal blinding effect on the Comb Separator, which is a
key improvement over the previous static screening or electro-mechanical switching
concept. Secondly, the research device produced improved sewage solids capture
efficiency in low flows (up to 70 l/s), compared to the industry standard Hydro-JetTM [12].
Analysing the parameter sensitivity for hydraulic devices such as the ‘Comb
Separator’ is a necessary check to understand qualitatively or quantitatively sources of
variation during practical application. Sensitivity analysis can provide insights that guide
informed decision-making for the management of different sewer overflow events using
this screening device [13].
1.5 Thesis Structure and Overview
The thesis is divided into seven chapters. Chapter 2 to 7 present and discuss the
theories, methods, modelling and results of this research into the usage of proposed
sewer overflow screens.
Chapter 2 reviews the relevant literature regarding sewer overflow screens used
in both separate and combined sewage overflow systems. The limitations of the existing
screens are detailed, and the aims and objectives of this research are established. A
brief discussion about the proposed screening concept and the design considerations in
achieving that concept is also presented in this chapter.
Chapter 3 gives a detailed description of the methodology adopted in this thesis.
The research questions are outlined, and the methodology was adopted based on
answering those questions. A flow chart of the research methodology is also provided in
this chapter.
Chapter 4 details the purpose of the modelling investigation. This chapter
discusses the development of both the analytical and CFD models, and describes the
Chapter 1: Introduction
Md Abdul Aziz 7
plausibility check of the CFD model using the analytical model. Results of the CFD model
are also discussed, and an explanation is given on how to optimise the proposed gross
pollutant device. This chapter also discusses the limitations of the gross pollutant
screening device.
Chapter 5 gives an overall description of the laboratory experiments leading to
the design of the proposed Comb Separator device. The design of this device is based
on previous analysis of CFD results, and on the limitations observed in relation to the
earlier gross pollutant device. This chapter also discusses experimental data collection,
test procedures and the results of the experiments. The limitations of the experimental
device are reported at the end of this chapter.
Chapter 6 describes the development of an Artificial Neural Network (ANN) to
overcome some of the limitations of the experimental work and CFD analysis. A detailed
description of the development of the Artificial Neural Network model is also included in
this chapter. It concludes with a discussion of how the ANN results could improve the
optimisation process of the Comb Separator device to obtain maximum capture
efficiency of sewage solids overflow.
Chapter 7 describes the sensitivity analysis of the Comb Separator. The
performance of the comb separator is compared with the industry standard Hydro JetTM
screening device. To analyse sensitivity, a linear regression model was developed. As
there is only a small set of experimental data available for analysing the sensitivity of the
input parameters of the comb separator, sampling techniques were used to expand the
dataset. Results of the input parameters are also included in the discussion.
Chapter 8 summarises the key findings of the study, and offers suggestions for
further research.
Chapter 2: Literature Review
Md Abdul Aziz 8
Chapter 2
Literature Review
Chapter 2: Literature Review
Md Abdul Aziz 9
2.1 Introduction
A review of the scientific literature on sewage overflow screening systems is
outlined in this chapter. The literature was sourced from professional societies, research
organisations, local councils, government agencies, published journals, reports and
conference papers. The review covers the following issues:
Effect of sewage solids overflows on the environment
Types of sewage solids found
Three major types of screening systems available in current sewage
overflow systems
Key methods to improve sewage solids overflow screening
Identification of research needs
Key consideration of the proposed sewage solids overflow screening
device
Scope of this research
During wet weather conditions, urban sewerage systems are not capable of
carrying all the excess water; hence it flows into open creeks or wetlands, often carrying
sewage solids. These sewage solids are dispersed, suspended or washed into rivers,
creeks, wetlands and the ocean. These solids further create an aesthetic nuisance, either
by their general appearance or through the actual presence of specific, objectionable
items such as floating debris, sanitary and faecal matter, scum, and even condoms. Due
to increasing public complaints, the focus of scientists and engineers is on the retention
of entrained sewer solids within sewerage overflow devices.
To overcome the issue of sewage solids overflow, different types of screening
systems are available, most of which use floatable controls at wastewater treatment,
sanitary sewer overflow (SSO), and combined sewer overflow (CSO) locations. The
preliminary treatment step is screening, which helps to segregate visible objectionable
materials, and also protects downstream equipment. In recent years, the frequency of
Chapter 2: Literature Review
Md Abdul Aziz 10
sewage overflows has increased, both in separate and combined sewer systems, due to
population growth and impervious areas. The untreated overflow causes concern during
storm events because of the level of pollution. Sewage overflow results in visually
offensive littering of the receiving watercourse, and has negative physical, chemical and
biological effects on the receiving environment. The impacts on human health and on
the environment can be acute and cumulative, and there are also aesthetic
considerations. Screening and handling are unpopular processes for plant staff because
of odours, aesthetics and health concerns. Moreover, historically the screening system
has not been a piece of equipment that functions efficiently for long periods without
maintenance. The remote location placement of these screens typically makes the
operation and maintenance of screening systems costly and labour intensive. Thus it is
important to design an effective and efficient sewer overflow screening system.
While replacement of the pipe with one that has a greater diameter will increase
the capacity of sewerage systems, it is a very costly measure that is not economically
viable. Some earlier drainage plans did not consider the impact of flow pollution. The US
EPA estimated that the cost to overcome the sewage overflow problem in the USA alone
would be around 100 billion dollars. Considering the massive cost to replace and enlarge
the existing sewerage system, most research focus has been on improving the screening
capacity of the sewer overflow by introducing different types of screening systems. The
data summary of a sewage treatment works in central England [156], including the
composition of sewage solids, is shown in Table 2.1:
Table 2.1: Sewer solids available in the urban sewage treatment [156]
Component % by Weight (Dry Basis)
Rags 15 - 30
Paper 20 - 50
Rubber 0 - 5
Plastic 5 - 20
Vegetable Matter 0 - 5
Fecal Matter 0 - 5
Chapter 2: Literature Review
Md Abdul Aziz 11
To design a screening system to improve screening efficiency, it is important to
understand what types of sewage solids are being collected in the system. The
composition of sewage solids captured by screening varies from one location to another.
An updated review quantity and type of litter in a drainage system in South Africa
[19] found the following types of litter:
Plastic 62%
Polystyrene 11%
Paper 10%
Cans 10%
Glass 2%
Other 5%
These data were collected over a period of 122 days, and involved a total of 106
cubic metres of litter, transported by 32 separate storm events. Similar data is available
for debris found on beaches, both in terms of mass and number of items observed per
length of beach. These results can be found in the reports of CMC [36], HydroQual [87],
1993, HydroQual [88] and [123].
Varieties of sewage solids reported in Australia include condoms, tampons,
cigarette buds, wrap papers and bottle caps [9]. To segregate these sewer solids,
different types of screening systems are used. Wastewater treatment facilities and CSOs
apply different types of screens. These screens can be classified based on type, function
or size of screen opening.
The aim of this literature review is to understand the gaps in research, and to
identify potential areas for improvement of sewer overflow screening technology. The
methodology followed includes recognition of the reported limitations in hydrodynamic
investigation, followed by experimental or on site investigations, heuristic modelling
approaches such as Artificial Neural Network (ANN) models, and sensitivity analysis of
the different types of model used.
Chapter 2: Literature Review
Md Abdul Aziz 12
The frequencies of different sewer solids coming from various sources are shown
below:
Figure 2.1: Frequency of Trash and Litter from Various Sources (Source: [156])
2.2 Summary of Current Screening Applications
Designing sewage solids overflow screens is challenging, considering the masses
and loading rates of different sewage solids. The amount of sewage solids in an overflow
varies widely in a wet weather event. It will be particularly high if significant storm water
events happen after a drought. However, if the storm happens on consecutive days, the
mass of floatables and solids can reduce on the second day. Screening systems for
floatable control are used both in separate and combined sewerage systems. A survey
of screening devices shows that there are three types available in the current market:
2.2.1 Static screening
2.2.2 Mechanical and Electrical screening
2.2.3 Hydraulic screening
Chapter 2: Literature Review
Md Abdul Aziz 13
2.2.1 Static screening
A static bar screen is a typical example of a static screening device. This type of
screen is one the cheapest forms of screening technology available. The static bar
screen consists of sturdy bars aligned parallel to one another. These static screens are
fixed and capable of capturing solids and floatable material. Static bars constitute a
stand-alone system without any mechanical, electrical parts or automated cleaning
mechanism. As there is no self-cleaning process available with static screens, it is
important that manual cleaning of solids and floatables is done periodically. So as to
avoid restriction of flow, maintenance crews need to make visits to ensure that the screen
does not become clogged. This is one of the key limitations of this type of screen.
Figure 2.2: Static bar screens in operation (source: [3])
These screens also take up significant space at the points of installation into the
existing sewerage system. There is also no provision for any bypass channels, which
increases the risk of clogging with solids and floatables, eventually leading to failure of
Chapter 2: Literature Review
Md Abdul Aziz 14
the screen to serve its purpose. This flow restriction limitation necessitates the
installation of new screening chambers.
Figure 2.3: Brush-Raked fine screen static screen (source: [118])
2.2.2 Mechanical and Electrical screening devices
A number of different types of electrical and mechanical screening devices are
available on the market. Descriptions of the four most common types of electrical-
mechanical bar screens are provided below:
2.2.2.1 Vertical mechanical bar screens
2.2.2.2 Horizontal mechanical bar screens
2.2.2.3 Rotary drum screens
2.2.2.4 Centrifugal screens
Chapter 2: Literature Review
Md Abdul Aziz 15
2.2.2.1 Vertical Mechanical Bar Screens
These types of screens have both below and above water surface components.
The mechanical arm is above the water surface, whereas the submerged portion has a
vertical, inclined, static bar screen rack. To clean the bar periodically, the rake arm
moves down below the water surface and onto the bar rack. When the rake arm
continues to move upwards on the screen near the discharge chute, the solids and
floatables are dumped in the loading container. The key benefits of using vertical
mechanical bar screens are:
Well known and well understood technology that has been in practice in
wastewater treatment facilities for over a decade.
When the water level in the chamber is high, the rake arm mechanism stops the
bar screen from blockage.
The bar screens are heavy duty bars that are structurally more sturdy than wire
mesh type screens.
The system can be adjusted with the addition of a flushing water system that
allows flush solids and floatables to get back to the interceptor.
The performance of these screens is based on bar spacing; however these
screens are effective in the removal of sewer solids and floatables up to 12.5mm
in diameter or greater.
However, these bar screens have limitations, some of which are listed below:
The mechanical and electrical components require more options for operational
and maintenance screening.
Maintenance of mechanical screening bars operating in remote unstaffed
locations is expensive.
A high clearance height is involved, which creates problems at some overflow
locations.
The initial cost to build such screens is more than to build concrete or other
structures; hence capital costs are higher.
Chapter 2: Literature Review
Md Abdul Aziz 16
2.2.2.2 Horizontal Mechanical Bar Screens
The horizontal mechanical bar screens (refer Figure 2.4) are rigidly constructed
and well mounted, using materials that are free from corrosion such as stainless steel
bars, which are then equally spaced apart. These bars are designed in such a way that
there are no intermediate supports to collect solids. A level sensor is activated
spontaneously when storm water rises to such a level that it overflows the weir of the
screen. The hydraulically driven rake system moves back and forth across the screen to
keep the screen clean. The advantages of horizontal mechanical bar screens are listed
below:
When a high level is detected in the chamber, a programmed instruction to the
mechanical screening is activated.
The rake arm assembly effectively protects the bar screen from blockages.
Heavy duty bars are more durable than mesh type screens; therefore the bar
screens perform better in remote unstaffed locations.
Figure 2.4: Typical mechanical bar screens (Source: [117])
Chapter 2: Literature Review
Md Abdul Aziz 17
This type of set up requires less maintenance and personnel costs, as sewer
solids are well managed through being pushed back into the wastewater channel.
Based on bar spacing, the horizontal mechanical bar screens are operative for
the removal of sewer solids and floatables up to 12.5mm and greater in size.
There are certain limitations in using the horizontal mechanical bar screens:
The mechanical and electrical components require more operational and
maintenance necessities than most non-mechanical screening systems.
If placed in remote isolated locations maintenance of the sophisticated electrical
and mechanical system can be difficult and expensive.
2.2.2.3 Rotary Drum Screens
A wide variety of industries apply rotary drum screens (refer Figure 2.5), including
the municipal wastewater, processed flood, and pulp and paper industries. These
screens have wedge wire wrapped around a drum screen that is open on both sides.
The drum screen is adjusted to the carriage of mechanical rollers so that it can rotate
around a horizontal axis. This horizontal axis is parallel to the sewage flow. The
screening occurs inside the drum. This type of screen has the following advantages:
This technology is accepted by a variety of industries; hence it is well-known and
understood.
The revolving action and an inner spray cleaning system prevent the drum screen
from blocking.
This screen is able to remove small sewage solids up to 12.5 mm and greater.
The wedge wire used in the drum screen has crossbars, which means the slots
are smaller than those of mechanical bar screens.
This screen has a clearance height lower than that of bar screens.
Although this screen works well in certain conditions, however it got the following
limitations:
Chapter 2: Literature Review
Md Abdul Aziz 18
The water spray systems using mechanical and electrical components have more
operational and maintenance requirements than do the mechanical bar screen
and the non-mechanical screening system.
To house the screening facility, additional concrete or other structures are
required, resulting in increased capital costs.
These devices contain more mechanical parts, which could potentially cause
failure of this device.
The maintenance and personnel costs of collection, transportation and disposal
of this screen are high.
The wedge wires for the drums are not constructed of thick, heavy duty bars.
Thus there is the potential that the wedge wire construction may not withstand
the force of repeated high flows.
Figure 2.5: Internally Fed Rotary Screen (Source: [117])
2.2.2.4 Centrifugal Screens
This type of screen considers a series of screens that is attached to a cage which
revolves around a vertical axis, refer to Figure 2.6.
Chapter 2: Literature Review
Md Abdul Aziz 19
Figure 2.6: Centrifugal screen (source: [117])
The sewage overflow enters the inside of the screen cage at the bottom, and
flows upwards to a deflection plate mounted at the top of the unit. The deflected flow
passes through the screens, and pollutants are collected outside the cage as shown
above.
Hydraulic Screening
Hydrodynamic vortex separation (HDVS) dates back to the early 1960s, and is
the most commonly used hydraulic screening device. The full scale device was tested at
Bristol in the UK. One of the key attributes of the HDVS separator is that it creates
tangential flow into a cylindrical vessel (refer Figure 2.7).
Figure 2.7 shows a cut-away view of the Storm King ® Overflow HDVSs with a
number of internal components highlighted. This is an industry standard screening
device and developed through a throw research process.
Chapter 2: Literature Review
Md Abdul Aziz 20
Figure 2.7: The cut-away view of the Rotary-Jet Screen (Source: [59])
This tangential flow creates a complex rotating flow regime. In the configuration
of the HDVS, the inlet deflector plate minimises headloss by streamlining the incoming
flow as it enters the main vessel body, and joins with the mass of fluid circulating within
the vessel. The device has proven to be more effective and efficient than conventional
chambers or fixed screens [21] & [20]. Over time, different types of HDVS configurations
have evolved. The development HDVSs is discussed in the following paragraphs.
The first generation HDVS separator could retain up to 70% of the pollution load
[151]. This second generation device was introduced in the 1970s, and is reported in
Field’s work [64]. During the early 1980s, a third generation of the HDVS was developed
in the UK. This device overcame shortcomings that had been identified in the previous
EPA Swirl Concentrator. With this upgrade, the device screening system was patented
and commercialised, carrying the trade name Storm King® Overflow. Although the
device was used commercially, there was a need to reduce turbulence in the Swirl
Chapter 2: Literature Review
Md Abdul Aziz 21
concentration at high flow. In the mid to late 1980s, the HDVS-FluidsepTM [30] was
developed in Germany.
Alongside the development and upgrade to commercially available devices during
the 1980s, the Hydrodynamic vortex separation was subject to different performance
assessments in Europe, North America and Japan [30]; [80]; [21]; [20] and [125]. A
performance evaluation of a number of these devices based on influent solids and their
settling characteristics especially highlighted the velocity distribution [167]; [14].
The HDVS separator became more sophisticated by the early 1990s, as the
system was used for water quality control for CSOs (e.g. Storm King® overflow, Swirl
Concentrator and FluidsepTM), and stormwater treatment (eg. Downstream Defender ®
and the Vortechs TM system). In the later years of the 1990s, HDVS technology advanced
further in response to the Asset Management Plan 2 (AMP2) requirement in the UK. This
required a non-powered self-cleaning screening system to address the issue of total
capture of neutrally buoyant solids greater than 6 mm in two dimensions [152]; [16].
More than 1,500 HDVSs have been installed around the world for managing
sewer solids in stormwater, combined and separate sewers, and waste water treatment.
Although many HDVSs are used, confirming the application potential of the device, there
are mixed views regarding their efficiency and effectiveness.
Another thoroughly researched device is the Hydro-JetTM Screen. This device has
a self-cleaning mechanism, and its suggested use is in combined sewer overflows that
utilise a purely hydraulic cyclic backwashing mechanism. The National Rivers Authority
[120] in the UK set the standard for intermittent wet weather discharge and removal of
pollutants. The most stringent condition requires the segregation of solids greater than
6mm diameter in any two dimensions. Previous experience has shown that if the device
does not use a screen self-cleaning mechanism, it will be subject to blinding. Some of
the conventional systems use electro-mechanical components to facilitate such a
process; however, these sophisticated systems are subject to seizure or jamming of
moving parts, which means that maintenance requirements are high.
Chapter 2: Literature Review
Md Abdul Aziz 22
Figure 2.8: Schematic representation of the Hydro-JetTM (Source: [59])
To overcome such challenges, the UKWIR [145] developed 15 proprietary
screens. Nine had an external power source; 11 had moving parts. The remaining four
screening systems had neither moving parts nor power requirements. Two were
stationary screens, while the others used a self-cleaning mechanism. Of these devices,
the Hydro-JetTM screen is the only one that features all the key attributes of no moving
parts, no electrical-mechanical switching system, and incorporating a self-cleaning
mechanism. Moreover, the system was developed as a cost-effective contender, with a
hydraulically-sophisticated backwashing mechanism.
The Hydro-JetTM screen has been subjected to a rigorous evaluation process [15];
[16] & [58]. Having studied the available documentation on this evaluation process, the
Hydro-JetTM, is considered the benchmark device for a comparative analysis of the
screening systems in this research. A comprehensive discussion of the performance
analysis of the Comb Separator and Hydro JetTM is provided in section 5.4.3.
Chapter 2: Literature Review
Md Abdul Aziz 23
2.3 Methods to improve Sewage Overflow Screening
2.3.1 Hydrodynamics Applications
Experimental measurements are probably the best way to understand the
capture efficiency of sewage solids for any proposed device. However, this process
cannot be undertaken before the device has built [51]; [105]. However, the CFD model,
after validation, could offer an alternative method to predict the performance of the
proposed device. The hydrodynamic investigations of a sewage overflow screening
device can contribute significant knowledge without the need for the physical set up of
the screening system. The key advantages of CFD modelling are suggested by Harwood
and Saul [77]:
CFD can simulate experimental conditions without physical laboratory
facilities.
The structural geometry of the CFD model can be changed quickly, which can
avoid the significant time and costs involved in reconstructing a physical model.
Flow parameters of shear stress, velocity and pressure are calculated at all
points, providing more insight than the physical model [83]; [160].
The recent development of powerful computational facilities allows the CFD
model to simulate complex flow dynamics. Best practice hydrodynamic applications
using the CFD model are well documented in the work of Casey and Wintergerste [33].
The CFD model contributes to improvements in the efficiency of sewage overflow
screening systems, especially CSO chambers [143]. A Storm King ® hydrodynamic
separator was modelled by Svejkovsky and Saul [163] using 3D FLUENT. Pollert [132]
and Hrabak et al. [84] modelled and evaluated the hydraulic performance of CSOs. The
complex hydraulic flow features such as erosion, containment or mobilisation of
pollutants were studied by Harwood [78], and the deposition of solid particles was studied
in by Stovin et al., [160]. Similar work by Stovin [158], Stovin and Saul [159;161], and
Adamsson et al., [1] highlighted difficulties with modelling particles transported in
physical models, and showed the application of the CFD model for particle transport in
sewerage systems. An explanation on how to model suspended solids separation and
vortex separation can be found in the work of Pollert and Stransky [133] and Tyack and
Fenner [167].
Chapter 2: Literature Review
Md Abdul Aziz 24
The ogee weir spillway possesses excellent hydraulic features in terms of flow
efficiency, as well as relatively good flow measuring capabilities. In the device studied,
the differences from a traditional weir flow considered were the upstream flow conditions,
reflection waves due to a short device boundary, and the change in shape and size of
the flow properties. These slight changes need to be thoroughly researched to identify
whether they have a negative effect on the evaluation of the performance of the spillway.
A detailed study had been carried out to determine the standard shape and size of the
crest of the overflow spillway; the relative height and the upstream slope of the spillway
were also considered [114]; [41]. Similar objectives have been reported in the work of
the US Army Corp of Engineers [169]; [170].
Most previous investigations were confined to physical models. In recent years,
with the advent of powerful computational advances, research is focused on flow
simulation using numerical modelling. An early attempt to model the spill overflow using
potential flow theory was made by Cassidy [34], and with limited experimental data, good
agreement with experimental data was noted. Better accuracy with experimental data
was found in studies of [90] and [24] using linear finite element approximation. In
addition, 2D irrotational gravity flow over the curved water surface was successfully
modelled. Xie and Chen [103] and Guo et al. [66] extended on the potential flow theory
while applying an analytical functional boundary. A 2D finite volume based numerical
model for flow over a spillway was validated using water level and pressure data on the
physical model [27].
Most of the existing literature reported either experimental works or numerically
simulated flow phenomena over an ogee weir in ideal conditions with no wave reflections
and much larger upstream and downstream boundaries. Although such assumptions
simplify the problem, they cannot be incorporated into the existing sewerage drainage
system, where space constraints in the urban drainage system would be a major issue.
This research gap was identified, and the current research has focused on understanding
the space restriction in experimental set ups and in the analysis.
The current research also takes into account discussion regarding the inlet
orientation, the effect of reflected wave on a small dimension screening system,
Chapter 2: Literature Review
Md Abdul Aziz 25
optimisation of the inlet length to maximise device performance, and the best ogee weir
orientation based on a previous study by the US Army Corp [169].
2.3.2 Experimental Investigations
The capture or retention onsite of sewage solids in different gross pollutant
devices has been a common challenge for the water industry. Phillips [129] mostly used
real floating litter items for his proposed screen. Armitage and Rooseboom [19] tried to
capture an artificial pollutant. Overflows occur more often in combined sewer overflows
than in a separate sewer system. An attempt to address this issue has led to active
research on CSOs over the last 50 years. The initial work done Sharpe and Kirkbride
[150], followed by a series of works during the 1960s, 1970s and 1980s, formulated the
report of guidelines [23]. Further laboratory testing on the topic identified some limitations
in the gross retention performance. This work also updated the user guide in the report
published in 1994 [69] that explains how to design the combined sewer overflow
structures. Following a thorough investigation, a report entitled ‘Predicting aesthetic
pollutant loadings at CSOs’ was completed and published for the water industry in 2002.
In addition, full scale research at the National CSO test facility Wigan WwTW highlighted
the need to use screen technology [165]. The regulatory requirement for aesthetics can
be found in the work of Saul and Blanksby [147].
Table 2.2: Regulatory requirement for aesthetics [147]
Amenity Use Category Expected Frequency of Spills Standard
High Amenity > 1 spill per year 6 mm solids separation
<= 1 spill per year 10 mm solids separation
Moderate Amenity > 30 spills per year 6 mm solids separation
<= 30 spills per year 10 mm solids separation
Low Amenity and Non- Amenity Good Engineering design
Chapter 2: Literature Review
Md Abdul Aziz 26
A comprehensive assessment of different screen types was conducted by UK
Water Industry Research (UKWIR). The screening performance is a combination of
different parameters, however two key parameters are, design flow rate and how many
pollutant particles are present in the sewage overflow.
When used in isolation, most sewage overflow screens tend to blind. To
overcome this problem, many screens use the electro-mechanical switching system.
However, considering the remote locations and harsh environments of some sewers,
these screens are not ideal. To overcome these problems, an effective and low-
maintenance screening system Hydro JetTM screen was developed. The general
concentration device was subject to a series of tests carried out at the University of
Sheffield, which concluded that the system could capture solids of 6mm in two
dimensions as required. This device was given a long-term site trial at Wessex Water,
and has been further tested at 27 sites with 13 rotary Hydro Jet TM screens.
The Hydro JetTM screening system was developed with a rigorous testing and
evaluation process in order to meet the industry requirement, and represents a high-
class solution to many problems that existed before. The performance of the device
proposed in this research will therefore be compared with the Hydro JetTM screen. Further
discussion in this regard can be found in Chapter 5, Section 5.4.3.
2.3.3 Artificial Neural Network (ANN) Applications
The Artificial Neural Network (ANN) was inspired by biological neural networks,
and has the unique ability to learn and generalise knowledge. An ANN can be considered
as a massive parallel-distributed information processing system. This system has certain
performance characteristics that resemble the biological neural networks of the human
brain [79]. ANNs constitute a non-linear data modelling tool used to model complex
relationships between inputs and outputs without any prior assumptions or any available
mathematical relationship between them. ANNs comprise a group of interconnected
artificial neurons, which are simple and fundamental processing units. A neural network
is characterised by its architecture, which represents the pattern of connection weights
and the activation function [61].
Chapter 2: Literature Review
Md Abdul Aziz 27
Artificial Neural Networks have already been successfully used to simulate flood
forecasting in urban drainage systems [31]; [96], real-time flood forecasting [126], annual
run off predictions [67]; [68], rainfall forecasting [100], real time control in combined
sewerage systems in Germany [173], and real time water level predictions of sewerage
systems covering gauged and ungauged sites [47]. As ANNs have been successfully
applied to simulate water quality and flow prediction applications [108]; [42], they have
been adopted in the current study. Some fundamental challenges in developing ANNs
include structure identification, parameter estimation, generalisation performance
improvement with proper choices of algorithms, over fitting, and finally, model validation
[109].
Figure 2.9: A typical artificial neuron k
p
jkjkjk xwy
1 (2.1)
Each artificial neuron is basically a computer processor (refer Figure 2.9), where
the output yk is a function of the weighed sum of the inputs, In Eqn. (2.1)., x1,x2,…,xp are
the input signals; wk1, wk2,…, wkp are the assigned weights; θk is the threshold value and
is the transfer function.
Chapter 2: Literature Review
Md Abdul Aziz 28
The experimental work in this research was restricted by the physical limitations
inherent in laboratory studies. These include the limited number of trials that can be
achieved, the limited number of experimental set ups possible, and the significant cost
and time required for running the trials. To overcome these problems, the experimental
results were analysed and used to train an Artificial Neural Network (ANN) model. ANN
had already been used successfully in similar kinds of environmental problems such as
water level predictions, flood forecasting and control in combined sewers [47]; [31]; [173].
Willems and Berlamont [174] have demonstrated a number of uncertainties
involved in deterministic models for sewerage systems. In particular, model
simplifications of the physical system make it difficult to adopt physical law based models
such as CFD. In the problem studied, the features which are difficult to model are: (i)
physical characteristics of different sewage particles; (ii) multi-fluid sewerage systems
with changing velocity due to different viscosity of fluids, and (iii) interaction between
liquid and solid particles. However, an ANN has the capability to effectively extract
significant features and trends from complex systems, even if the underlying physics is
either unknown or difficult to recognise [47]. Moreover, it can reduce computational time
and cost, unless completely new sets of experimental conditions are used [135]. They
also have the capability to predict complex input output relationships with little
understanding of the physio-chemical system. This makes the model the obvious choice
among a wide range of urban drainage systems.
In the case of sewage solid capture efficiency under study, the neural network
modelling was able to learn the existing non-linear input-output relationships. A multi-
layer feed forward artificial neural network using a back propagation algorithm was used.
Such networks have been used almost exclusively in environmental modelling [109]. The
current research implemented ANN modelling to overcome the physical limitations of the
experiments. Further discussion in this regard can be found in Chapter 6.
2.3.4 Sensitivity Analysis to Model Results
In the current research, the sewage overflow screening system required hydraulic
testing and modelling. Analysing the hydraulic device parameter sensitivity has been a
standard practice of hydraulic engineers over the years [97]; [116]; [176]. Sensitivity
Chapter 2: Literature Review
Md Abdul Aziz 29
analysis, quantitative or qualitative, provides a guide to different sources of variation
[139]. Some of the key reasons for analysing sensitivity are listed below:
To identify the most influential factors in the test result of the model analysis
To detect factors that require further testing and research to improve confidence
in the model output
To recognise areas in the spacing of inputs where the maximum variation occurs
in the model output
To categorise any factors that interact with each other
To develop a robust understanding of the meaningful input parameters
To comprehend the impact of experimental design parameters on sewage solids
capture efficiency
Sensitivity analysis is a standard, accepted methodology for any modelling
investigation, or for analysing expanded data series. Extended analysis from basic
sensitivity analysis can be found in the work of Hall and Solomatine [70] & [71]. A
comprehensive review of the application of sensitivity analysis in environmental models
is presented by Hamby [74]. Other works discussing sensitivity analysis include [81];
[49]; [95]; [94] and [75].
As sensitivity analysis of the model result is a necessary aspect of responsible
model use, the current research has analysed experimental results using this model.
Analysing the parameter sensitivity for the Comb Separator as a proposed hydraulic
device is an essential check done to understand sources of variation, qualitatively or
quantitatively, during practical application of the screen.
2.4 Identification of Research Needs
This section summarises issues identified as requiring further investigation, and
are the points that frame the scope of this research project. The scope of this research
includes:
Understanding the space restriction in an existing urban sewerage system
Chapter 2: Literature Review
Md Abdul Aziz 30
Understanding device performance based on the analysis of screening
orientation, wave reflection on the screen, and different device optimum
dimensions
Hydraulic analysis of sewage solids carry over on the screen
Fineness of screens or openings
Bar and perforated or punched hole openings
Performance of the self-cleaning effect
Overcoming physical limitations of the experimental conditions
Sensitivity analysis to develop meaningful and simplified input models, while
considering how key input parameters have an influence on output capture
efficiency
Comparison of the performance of the proposed screening system with the Hydro
JetTM screen
Wash water and power requirements
2.5 Summary
This chapter has provided a brief overview of the relevant scientific literature.
Three types of screening systems were identified: static screening, mechanical and
electrical screening, and hydraulic screening in existing CSO or separate sewer systems.
The limitations of the earlier static screens include the potential for blockage and high
maintenance costs. Mechanical and electrical screening performs better in overcoming
blockages; however, they are not ideal for use in remote locations. Sophisticated
mechanical electrical switching systems can create more issues, due to the isolated,
rough locations of these screens. Hydraulic screens usually have a self-cleaning effect
which makes them desirable in isolated locations.
There are four key methods found in the literature that improve the understanding
of sewage overflow screens:
Chapter 2: Literature Review
Md Abdul Aziz 31
a. Hydrodynamic applications such as CFD models
b. Experimental investigations
c. Heuristic approaches such as the ANN model
d. Sensitivity analysis of the model results
Further details of these key methods are discussed in the Chapter 3.
Chapter 3: Research Methods
Md Abdul Aziz 32
Chapter 3
Research Methods
Chapter 3: Research Methods
Md Abdul Aziz 33
3.1 Introduction
Background information, a review of relevant literature, and an analysis of the gaps
in research were provided in Chapters 1 and 2. To address these gaps, a set of fundamental
research questions has been formulated. To answer these research questions, a series of
research studies was designed, and the research methodology was derived based on
answering the research questions.
3.2 Research Questions
The key question of this research is: what is an efficient and effective sewer overflow
screen that overcomes the key limitations of existing sewer overflow screens? Some
common drawbacks with the existing sewer overflow screening include the following:
Inadequate screening capacity
Blockages on the screen
Sophisticated electro-mechanical switching system
High operational and maintenance costs
The need for a self-cleansing device due to location in remote unmanned places
To address the issues described above, the current research was planned with the
aim of answering the following key research questions identified:
RQ1: How to improve sewage solids overflow screening capacity?
RQ2: How to reduce blockage on the screen?
RQ3: How to avoid sophisticated electro-mechanical switching system in the screen?
RQ4: How to make the screen self-cleansing?
RQ5: How to optimise the proposed sewer overflow screening device?
RQ6: How to reduce the operational and maintenance cost?
RQ7: Is the performance of the proposed screen any better than current practice?
Chapter 3: Research Methods
Md Abdul Aziz 34
3.3 Research Process
Individual research studies were designed to find answers to each of the research
questions above. These are briefly described below:
Study 1: How to improve sewage solids overflow screening capacity?
Comprehensive computational fluid dynamic modelling was completed to
understand the orientation of the screening device and the inlet pipe diameter, in order to
maximise sewer overflow capture efficiency [10]. The performance of the proposed device
was tested in a series of laboratory experiments [12]. The optimum device set-up was
simulated using an ANN model, and good agreement was achieved with the experimental
results [9]. Finally, sensitivity of the capture efficiency was tested at different inlet conditions
to understand which input parameters have the most sensitivity for the proposed device,
and how to use the device efficiently [11]. A more detailed description of the screening
capacity can be found in Chapters 4, 5 and 6.
Study 2: How to reduce blockage on the screen?
The literature shows that blockage on the screen has been a common problem. To
overcome this, a series of experimental set ups was tested using different comb spacing. It
was found that the blockage on the comb separator was minimal, and that if the device is
run for 10 to 15 minutes, there is hardly any blockage seen on the comb spacing [12]. A
detailed description of these experimental results can be found in Chapter 5.
Study 3: How to avoid sophisticated electro-mechanical switching system in the
screen?
Most recent devices use a sophisticated electro-mechanical switching system,
which at times does not work in remote unmanned locations. To overcome this issue, the
sewage solids overflow screening devices described in Chapter 4 [11] and Chapter 5 [12]
did not include an electro-mechanical switching system. A detailed design of the screening
concept can be found in Chapters 4 and 5.
Study 4: How to make the screen self-cleansing?
As the sewer overflow screen devices are located in remote and unmanned
locations, it is important to design the device with a self-cleansing effect [10]. The self-
cleansing effect also reduces operational and maintenance costs. A detailed CFD
investigation was carried out to understand how to maximise the self-cleansing effect for
Chapter 3: Research Methods
Md Abdul Aziz 35
the proposed sewer overflow screening device. The performance of the self-cleansing effect
was also verified in laboratory testing of the device [12]. Detailed descriptions of
investigations aimed at maximising the screening self-cleansing effect are set out in
Chapters 4 and 5.
Study 5: How to optimise the proposed sewer overflow screening device?
The CFD model was too complex to derive an outcome when sewer solids of
different density were tested. To overcome this challenge, laboratory testing was designed.
The laboratory experiments had limitations in the extent of the trial, meaning that the
experimental set up may not lead to the optimum result. To complement this information,
an Artificial Neural Network (ANN) was designed and tested to find the optimum condition
of the experimental set up [9]. A more detailed description can be found in Chapter 6.
Study 6: How to reduce the operational and maintenance cost?
Most previous sewer overflow screening devices used a sophisticated electro-
mechanical switching system. These systems are expensive to design and maintain in
remote unmanned locations. Moreover, such devices do not perform well in harsh remote
physical environments. The proposed device did not have a sophisticated switching system
or electrical equipment. Instead, an automated valve was used of the type recommended
by most environmental agencies. The device was also tested for self-cleansing effect to
reduce the operational and maintenance cost [11].
Study 7: Is the performance of the proposed screen any better than current
practice?
The last research question is of significant importance in highlighting the contribution
of the current research to the actual performance enhancement of the sewer overflow
screen. The performance of the ‘Comb Separator’ was compared with the industry standard
‘Hydro-JetTM’ [12]. Finally, sensitivity analysis of the laboratory experiments was undertaken
so that the ‘Comb Separator’ can be used more effectively in actual sewer overflow
situations [13].
Chapter 3: Research Methods
Md Abdul Aziz 36
3.4 Research Design
The current research was designed based on the research questions. A checklist of
different research strategies for seven key research questions was prepared. The initial
concept design was modified based on research experience. The strategies applied in the
current research are shown in the flow chart below:
Figure 3.1: Flow Chart of the Current Research Plan
(Where RQ stands for Research Questions)
Research Design
Define Research Questions (RQ1 to RQ7)
Concept Sewer Overflow Screen Design
CFD Model Investigation (RQ1, RQ3 & RQ4)
Revised Sewer Overflow Screen Design
Laboratory Experiments (RQ1, RQ2, RQ3, RQ4, RQ6 & RQ7)
ANN Modeling (RQ 5) Sensitivity Analysis
Report Writing
Chapter 3: Research Methods
Md Abdul Aziz 37
3.4.1 Computational Fluid Dynamic (CFD) Analysis
The key research questions answered using the CFD model are RQ1, RQ3 and
RQ4. The CFD modelling and lab experiments were complementary to each other, with the
aim of finding the answers to the research questions. However, the CFD investigation was
carried out first to get a better insight into the problem so as to try to find solutions
accordingly. The RQ2 could not be tested under the CFD model, as sewer solids differed in
density, making it too complex to model. Where the CFD model could not provide the
answer to research questions, alternative research methods were adopted.
3.4.2 Laboratory Experiments
The next discussion concerns onsite and experimental work on the sewer solids
screening system. The experimental work for this research is addressed in Chapter 5. Out
of the literature review came the following discussion points to explore further in the current
research:
Fineness of screen or openings
Bar or perforated punched hole opening
Performance of the self-cleansing effect
To be more specific, the laboratory experiment tried to answer most of the research
questions, with the exception of RQ5 regarding device optimisation. The CFD model helps
immensely in designing experimental parameters such as device orientation, input pipe
diameter and weir opening. Lab experiments try to answer most research questions
regarding screening capacity, blockage on the screen, self-cleansing mechanism, and so
on. However, there are physical limitations of the experimental set-ups, which necessitated
the use of a different research method/strategy: ANN modelling.
Chapter 3: Research Methods
Md Abdul Aziz 38
3.4.3 ANN model to supplement deterministic approach
To overcome the limitations of experimental investigations, it is important to do ANN
modelling. A more in-depth discussion about the ANN model can be found in Chapter 6. A
review of literature on ANN modelling uncovered the following research needs:
Hydraulic analysis of sewage solids with limited experimental set ups
Overcoming physical limitations of physical experimental conditions
This method helps to answer RQ5 regarding optimisation of the experimental set up
using the ANN method. The final research question regarding comparison with the industry
standard device is presented in Chapter 5. The performance of the ‘Comb Separator’ was
tested against the industry standard ‘Hydro-JetTM’ [12].
Sensitivity analysis is an integral and necessary part of any proposed hydraulic
device, including the sewer overflow screening device. Chapter 7 will discuss in detail the
sensitivity analysis adopted in the current research [13]. The sensitivity analysis of the
model results were important contributions in developing meaningful and simplified input
models while considering the key input parameters.
3.5 Analysis Procedure
Different research methods were adopted in this research to achieve the desired
outcome. Key methods adopted are outlined below:
3.5.1 CFD and Analytical Modeling
CFD is a proven modelling approach to understand dynamic behaviour of flows and
analysis of different design parameters. One of the significant benefits of CFD is that a
series of trials can be done before physical set up of the screening device. As experimental
set-ups involve significant cost and time, the initial approach in testing the performance of
the proposed sewer overflow screening device was to use a CFD model. To justify the use
of the CFD model, an analytical model was developed, which was used to check the
plausibility of the CFD model. The CFD model provides detailed insight into different
experimental parameters, orientation of the device, and so on. However, the CFD model
was not an alternative to the laboratory experiments; rather it complemented them,
Chapter 3: Research Methods
Md Abdul Aziz 39
improving the performance of the laboratory experiments. In the current search, sewer
solids of different densities were tasted. This is a very complex task for CFD analysis; hence
alternative methods were adopted. A detailed discussion of the development of the CFD
model and analysis of the results are provided in Chapter 4.
3.5.2 Experimental Investigation
The experimental set-up was developed based on learning from the CFD model.
The screening concept was also revised based on CFD results. The key objective of the
laboratory experiments was to improve capture efficiency of sewer solids. A series of trials
with different set-ups were undertaken to test methodologically. The target efficiency of
sewer solids was 80% for smaller sewer solids less than 10mm diameter with minimal
blinding on the screen, and to perform up to one overflow event in one year. There are some
physical limitations in the experimental set-ups. A further investigation considering all
possible scenarios was carried out using ANN modelling analysis. A detailed discussion of
the laboratory experiments is found in Chapter 5.
3.5.3 ANN Modeling
To supplement the limitations of CFD and laboratory experiments, an ANN model
was developed. The benefit of the ANN model over CFD analysis is that it can accurately
predict the input/output of a process, without having the physics explicitly provided.
Moreover, ANN can overcome physical limitations inherent in laboratory studies. Proper
care was taken in developing structural identification, parameter estimation, network
optimisation and ANN model validation. A detailed description of the ANN development and
an analysis of the results are provided in Chapter 6.
3.5.4 Sensitivity Analysis
Sensitivity analysis of hydraulic devices such as the sewer overflow screening device
is a standard procedure. Sensitivity analysis develops meaningful and simplified inputs for
the model taking into consideration key input parameters. The experimental dataset was
extended using the Latin Hypercube Sampling (LHS) technique, without changing the
relationship between input and output parameters. In-depth discussions about the
development, hypothesis and analysis of results are reported in Chapter 7.
Chapter 3: Research Methods
Md Abdul Aziz 40
3.6 Summary
In developing the scope for the current research along with the literature review gap
analysis, it was prudent to follow best practice guidelines for standard screening attributes
as suggested in the literature. These include the following stipulations:
The device should be built so that minimum inspection and maintenance are
required, and so that it can be constructed above ground level.
The treatment capacity required for the drainage system must be able to cope with
a maximum of one overflow event a year of 70 litre/second (l/s), and a minimum of
one flow event in four months of 20 l/s.
It should not use sophisticated electrical or mechanical systems, or signals that may
become ineffective during extreme events such as floods or lengthy droughts, and
which incur significant costs in maintenance and re-running.
In the event of a device failing, there must be a bypass option that causes no serious
damage to the device or the environment.
Taking all these points into consideration, the following flow chart illustrates the plan
in the current research. To address the research gaps identified, this research will follow
the following flow chart.
Chapter 3: Research Methods
Md Abdul Aziz 41
Figure 3.2: Flow chart of the current research plan
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 42
Chapter 4
Hydrodynamic
Analysis
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 43
4.1 Introduction
Nature cannot make sewage solids disappear; rather, it can transport them from
one system to the next, that is, from the drainage system to the creek system. Most urban
sewerage systems built in the 18th century used a simplified one-way material flow.
These sewerage systems had limited focus on long term environmental outcomes like
floating debris, creating issues with aesthetics and public health. The main consideration
for design was structural durability and to a lesser extent flexibility in response to
changing needs. In recent years the number of sewage solids overflows has increased,
due to exponential population growth and the increase of impermeable areas. Sewage
solids disperse, float or wash into rivers and eventually settle on the bottom, creating
odours and toxic/corrosive conditions.
This sewage solids overflow is more visible after heavy rainfall. Sewage overflows
to receiving water bodies raise serious environmental, aesthetic and public health
concerns. To address these problems, much research has been carried out into different
types of screening devices to remove these pollutants. The screening of sewage solids
is a controlled process that is desirable in the sewerage system. Floatable control is
preferred by most of the proposed and existing environmental regulations as it is effective
for use in un-manned remote locations. In most cases screening is the only economically
viable method, according to Faram [60]. These issues trigger the need to research
different types of screening devices and screening handling systems. A state of art
review of these screeners can be found in the work of Saul [146] & [148] and recent
literature updates can be found in the work of Madhani [107].
In past investigations, a number of different screening systems were used in
sewage overflow locations. One of the most common was the rotary screen proposed by
Moffa [118], which consists of a large rotating drum that is slightly angled to maximize
dewatering. The angle of the drum ensures effective dewatering as the screenings travel
up to the drum where they are removed from the unit. Metcalf and Eddy [117] proposed
a centrifugal screen with a series of screens attached to a cage that rotates around a
vertical axis. The flow enters from the bottom and flows upward to a deflection plate at
the top of the unit and screenings are collected from outside the cage. Hydrodynamic
vortex separation (HDVS) was another popular screening concept developed in the early
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 44
1960s. The first generation HDVS devices were found to be effective in retaining 70% of
the pollution load [151]. A second generation HDVS developed by the American Water
Works Association and EPA was reported by [64]. The third generation device in the
1980s was commercially patented as Storm King® overflow. The HDVSs had been
through a series of performance evaluations in Europe, North-America, and Japan.
Unfortunately aesthetic solids of neutral buoyancy were not trapped in HDVSs according
to Saul [144].
To overcome this challenge a non-powered self-cleaning screening system which
can capture neutrally buoyant aesthetic solids greater than 6mm in two dimensions was
tested [152] and [16]. Despite more than 1,500 HDVSs being installed worldwide for
storm water, there are still mixed views regarding their effectiveness [17]. Faram et al.,
[59] tested hydro jet devices installed in the USA, Australia and mainland Europe.
However, in most cases the devices were directly associated with blockages of the
sewerage system.
Reported literature suggests that screens need to be self-cleaning otherwise if
they are placed in remote un-staffed locations, they are subject to blinding [9]; [154].
Usually most conventional screening systems utilise electro-mechanical components to
facilitate such a process [131]. However, given the harsh unmanned remote environment
of many sewage overflow device locations, this is clearly not ideal [10]. Blocking and
seizure are common maintenance problems of moving parts and electrical failure will
dictate an onerous maintenance commitment in many cases [10]. To overcome such
problems a novel self-cleaning, less expensive, low maintenance sewage overflow
screening device with no moving parts is proposed. The aim of this chapter is to
investigate the optimum inlet and design parameters of the novel sewage solids overflow
screening device using CFD modelling.
With the advancement of computational power in recent years Computational
Fluid Dynamics (CFD) has become a proven technology to investigate hydraulic
behaviour of hydraulic structures and CSO design [161]; [113]; [50]. One key advantage
of the CFD based approach is that three-dimensional solid-liquid two phase flow
problems under a wide range of flow conditions can be evaluated rapidly, which is almost
impossible experimentally [164]. This technique has been successfully used for accurate
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 45
prediction of the flow pattern in storage tanks [51] and the solids separation of CSOs and
storage structures [77]; [10].
The proposed sewage overflow screening device has overcome common
drawbacks in available devices such as: inadequate screening capacity, external power
requirement and high cost [10]. Simon and Phillips [154] developed a sewage solids
overflow screening device with temporary holding tanks which provides transient storage
and real time control of sewerage systems. The device has no moving parts, it has a
robust stop/start operation, it works as self-cleaning device and it has no sophisticated
electrical mechanical circuit. The gross pollutant trapping device is a novel self-cleaning
sewage overflow screening device with a sewage overflow chamber, a rectangular tank
and a slotted ogee weir to capture the gross pollutants (Figure 4.1). The device does not
require any power source containing mechanical or electrical components; moreover the
device is a self-cleaning device which is a key requirement for use in remote un-staffed
locations. The proposed gross pollutant trapping device improves most of the reported
limitations of commercial sewage solids overflow screening devices. Those limitations
are:
Limited screening efficiency
Blockages on the screen of the sewerage system
Use of a sophisticated (electrical/mechanical) switching system
No self-cleaning effect for use in remote un-staffed locations
A comprehensive modelling investigation was of paramount importance. Firstly it
was necessary to check the concept design of the proposed screening device. Secondly,
it was essential to get a detailed understanding of the screening device design
parameters and finally, it was compulsory to test it under a series of different conditions;
which are very difficult to carry out in a physical experimental situation. In addition,
experimental work involves significant cost and time. To overcome these challenges and
to design an efficient, effective and optimised experimental sewage solids overflow
screening device, a 3D computational fluid dynamics (CFD) model was used. The main
purpose of the modelling investigation was to optimise and design the sewer overflow
screening system. To achieve this objective a plausibility check was performed between
an analytical and a CFD model.
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 46
The CFD model has adopted the following method or approach for analysing the
results
1. Once the plausibility check was completed for the CFD model the following
analysis was perform with the CFD model.
2. It is important to identify which inlet position will maximize the capture efficiency.
To understand this phenomena to compare two different inlet orientations (inlet parallel
to the ogee weir and inlet perpendicular to the ogee weir) for the proposed screening
device to obtain the maximum self-cleaning effect.
3. In design different experimental parameter it is important to understand the
pressure, velocity and water level information to optimise the experimental device. To
understand hydrodynamic flow properties such as velocity, water level, shear stress to
obtain the maximum self-cleaning effect and to find an effective location for the screening
device to maximise self-cleansing effect.
4. Due to the small device dimension it is important to understand the reflection of
wave on such device. To get a better understanding on the wave reflection it is important
to get the proper inlet length for the experimental device. To determine the optimum inlet
length to reduce wave reflections and improve functionality of the self-cleaning effect
5. There are different weir designs suggested in the literature so it is important to
understand which of these design will work better in the current setting. To optimise ogee
weir design orientation based on standard U.S. Army Engineers Waterways
Experimental Station design best practice guidelines.
4.2 Screening Concept
Figure 4.1 shows the overflow sewage device in the first phase. As sewage builds
up in the left chamber (A), water pressure will push the floatable ball upward in the right
chamber (B), which is connected to the left chamber via a pipe (C). As the floatable ball
goes upward, it will block the hole on the upper surface. The plan view of the proposed
device shows the left and right box chambers with the vertical dotted lines on the plan
view representing the screening device.The thick horizontal dotted lines on the plan view
represent the pipes connecting the left and right chambers. The thick smaller circle is the
hole at the bottom of the right chamber (B) and the dotted circle is the floatable ball. The
sewage builds up in the left chamber (A) until it becomes full, at which time the sewage
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 47
overflow will pass over a weir type structure. Figure 4.2 shows the second phase of the
scenario with the overflowing sewage.
Figure 4.1: Schematic diagram of the proposed sewage overflow screening device
Figure 4.2: Front views of the proposed device under different phases
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 48
Towards the bottom of the sloping weir, the screen will exclude the solids while
allowing water to pass through the screen, bypassing the right chamber (B) to exit to the
creek or waterway through two bypass channels (D). Figure 4.2 also shows the third
phase of the scenario, when the flow has subsided and the sewage level in the left
chamber (A) recedes. Once the sewage level drops down to a certain level, the buoyancy
pressure on the ball will reduce and the ball will drop, allowing the trapped pollutants to
exit into the right chamber (through the pipe C) and then to be flushed back into the
sewerage system using valve (E). This proposed overflow sewage device is designed to
be installed downstream of an existing sewage overflow location.
4.3 Development of the Analytical Model
Analytical models are sometimes also called mathematical models. This model
attempted to explain the behaviour of a system with mathematical equations. The power
of analytical models is that the mathematical function can provide information without
graphical or tabular presentation. In the current research an analytical solution of the
Navier-Stokes equations were derived [10]. A description of this model development
process is explained below.
To find an analytical solution using the Navier-Stokes equations some simple
assumptions are made [55]. Firstly, the flow is considered to be steady and uniform,
flowing under the influence of gravity and parallel to the surface bottom while the effect
of air viscosity at the free surface is negligible. As the surface is inclined we have to
consider the body force. Therefore with constant viscosity the Navier Stokes equation
becomes,
)()( 2
2
2
2
yu
xug
xP
yuv
xuu
tu
x
(4.1)
)()( 2
2
2
2
yv
xvg
yP
yvv
xuu
tv
y
(4.2)
The variables u, v, and w represent the velocities in the x-, y-, and z-directions; ρ
= density; μ = dynamic viscosity of water; P = defined as pressure; gx , gy are the
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 49
gravitational force in x and y directions. As we consider 1D flow the z-direction does not
need to be considered as the flow is steady δ/δt =0. Moreover the flow is parallel to the
inclined surface, i.e. X-axis, so δux /δx =0 and u=0. As the flow is uniform the flow takes
over a constant depth ‘h’ which leads to pressure gradient δP/ δx =0. If z is the vertical
direction, the potential per unit mass due to body force is gz
Therefore the components of body force in the X and Y directions are:
singxzgx
gX z
(4.3)
cosgyzgy
gY z
(4.4)
After incorporating all assumptions, components of body forces Equations. 4.1
and 4.2 reduce to:
0sin 22
dyudg (4.5)
01cos
yPg
(4.6)
Taking the value in Equation 4.6
δP/δy = -ρg cosθ leads to
CgyP cos (4.7)
at y =h, P = 0 atmospheric pressure i.e. C = ρgh. Therefore the expression for
pressure becomes,
)(cos yhgP (4.8)
Now integrating Eq. (4.5) twice with respect to y yields,
21
2
2sin CyCygux
(4.9)
At boundary condition y = 0, ux = 0 i.e. C2 = 0; again at y = h, duv / dy =0; C1 = gh
sinθ/ v;
)2(sin2
2yhygu x
(4.10)
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 50
Figure 4.3: Breakdown of the flow components of the experimental device
To get the shear stress at the boundary, applying Newton’s law of viscosity-
00)(sin
yy
xxy yhgdy
du
, which gives
singhxy (as y = 0) (4.11)
q is the flow per unit width,
sin31)2(sin
232
00
ghdyyhygdyuqhh
x
(4.12)
and average velocity,
sin31 2ghh
qv (4.13)
In the above equations, substituting inflow and weir surface angle with the
horizontal, unit width of the weir, different hydrodynamic parameters were calculated.
The US Army Corp [168] had developed standard shapes for the downstream profile of
the ogee weir defined by equation,
X1.85 = 2.0 Hd0.85 Y (4.14)
X is the horizontal axis and Y is the vertical axis. The depth of water upstream of
the spillway Hd is calculated from the non-dimensional equation for discharge given by,
Y
X2/A2 + Y2/B2 = 1
X1.85 = 2.0 Hd0.85 Y
cosg sing
g
X
h
Flow
Z
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 51
23
0 232
eLHgCQ (4.15)
Where, Q = total discharge; L = Lateral crest length or Width; He = total head
upstream from the crest; g = gravitational constant; and C0 = discharge coefficient. As
the velocity head is relatively very small with total head Hd considered equal to He which
is 7.16 cm (that is, h/Hd greater than 1.33 and He = Hd, for the approach velocity head is
negligible) [41]. Moreover the effect of slope or roughness did not change the average
value of He. The curve of the ogee weir surface is drawn from the equation
Y = 1.744 X1.85 (4.16)
The position (0, 39.5) is the starting coordinate over the ogee weir and different
parameters are calculated based on different points taken on the curve. The slope angles
are used to calculate the analytical results for velocity using Equation 4.13, flow using
Equation 4.12 and shear stress from Equation 4.11. Table 4.1 provides all the analytical
value derives in this regard.
Table 4.1: Calculation of hydrodynamic parameters from analytical results
4.4 Computational Fluid Dynamics (CFD) Model
CFD modelling is the analysis of a system involving fluid flow [171]. This CFD
modelling technique is very powerful and covers a wide range of application areas. Some
of the applications include:
Aerodynamic shape design of aircraft and vehicles
X Y Slope sin h (cm) )(vVel m/s xy )/( 2mN
0.00 39.50
1.98 39.84 0.17 0.169 3.97 0.87 65.62
3.00 39.62 -0.15 0.129 4.33 0.79 54.97
6.00 39.124 -0.17 0.232 3.57 0.97 81.07
11.98 33.00 -0.67 0.555 2.67 1.295 144.93
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 52
Hydrodynamic behaviour analysis of ships
Distribution of pollutant and effluents analysis
Flow analysis in rivers, estuaries and oceans etc.
The CFD codes are structured around numerical algorithms which can tackle fluid
flow problems. Some of the key steps in developing a CFD model include the CFD
modelling approach, discretization methods, schemes, turbulence modelling and
boundary conditions. CFD models use an iterative approach to obtain the solution of
Navier-Stokes equations. The Navier-Stokes equations are comprised of the
fundamental principle of conservation of mass and momentum. Hence, the cornerstone
of CFD is the fundamental governing equations of fluid dynamics – the continuity,
momentum and energy equations. They are the mathematical statements of three
fundamental principles upon which all of fluid dynamics is based:
The mass of fluid is conserved
Momentum is conserved, i.e. the rate of change of momentum equals the sum of
the forces on a fluid particle (Newton’s second law)
Energy is conserved, i.e. the rate of change of energy is equal to the sum of the
rate of heat addition and the rate of work done on a particle (first law of
thermodynamics)
CFD models describe the behaviour of fluid in terms of macroscopic properties,
such as velocity, pressure, density and temperature, and their space and time derivatives
[86]. From the 1960s onwards, the aerospace industry has integrated CFD techniques
into design, research and development of aircraft and jet engines. Increasingly CFD
became a vital component in the design of industrial products and processes. CFD has
entered into the wider industrial community since the 1990s [171]. With the rapid
improvement of the development of high performance computing facilities, CFD
modelling techniques have evolved as a powerful tool for researchers working on product
designs. CFD models can predict flow phenomena from single phase flows to complex
multiphase flows involving different fluid mixtures. Successful and efficient development
of a CFD model can predict fluid flow behaviour, provide insight into different product
orientations and help efficient and effective product design.
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 53
Fluid flow behaviour in a system can be represented with a series of non-linear
partial differential equations. Analytical solutions of these equations are very difficult to
derive, except in some special cases. To get an approximate solution numerically, a
discretization method is applied. This method approximates the differential equations by
a system of algebraic equations. These equations can then be solved using a computing
scheme which provides a description of the flow field at discrete locations in space and
time. The accuracy of experimental data depends on the quality of tools used, and
similarly the accuracy of numerical solutions is dependent on the quality of discretization
used [62]. There are a number of different computing schemes describing fluid flow which
can be solved using computational methods. Most commercial and research codes rely
on the following:
Finite Volume Method (FVM)
Finite Difference Method
Finite Element Method
Spectral Methods Method
All of these schemes above need the definition of discrete points in space at which
variables such as velocity, pressure and temperature will be computed. Although the
governing equations are always the same, the particular geometry with initial and
boundary conditions determines a unique solution for each particular problem. The
current research is based on the finite volume method. Most of the popular CFD codes
currently available use this scheme.
4.4.1 Finite Volume Method (FVM)
The finite volume method (FVM) is comprised of the integral form of the governing
equations involving surface integrals (e.g. convective and diffusive fluxes) and volume
integrals (e.g. those describing sources and sinks). There is also a rate of change term
which applies in the case of transient flow (i.e. unsteady flow that changes over time).
The FVM represents the integration of the governing equations over contiguous control
volumes (CVs) representing the solution domain. As variable values are computed only
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 54
at discrete points, approximations must be used to express the integrals in terms of
unknowns at discrete locations. The algebraic equations are obtained by linking the
variable value at the centroid of the CV with those at a neighbour CV.
A large system of algebraic equations is obtained for the whole solution domain.
As these equations are in general, non-linear and coupled, the solution must be sought
using iterative process. Which means repeating a sequence of operations over and over,
until changes in computed variables becomes negligible and the process is declared as
converged.
The current research used the academic CFD code of AVL Fire [22], however
most of the main commercial CFD codes, such as FLUENT, STAR-CD are based on the
FVM scheme. One of the key reasons FVM has succeeded over the other methods is
that it is inherently conservative. Although there are some errors in various
approximations, the discretised equations still fulfil the conservation laws exactly. In other
words, the errors introduced through various approximations affect only the distribution
of variables within the solution domain without violating conservation principles.
Moreover, FVM is easier for engineers to understand as other schemes involve more
complex mathematical methods.
4.4.2 Multiphase Flow Modelling
If the flow of interest involves more than one phase or component then the
problem needs to be considered as a multiphase flow in CFD. A phase can be defined
as an identifiable class of material that has a particular inertial response to and
interaction with the flow and the potential field in which it is immersed. For example,
different sized solid particles of the same material can be treated as different phases
because each collection of particles with the same size will have a similar dynamic
response to the flow field.
Two phase flow is the simplest case of multiphase flow. Multiphase flow can be
classified according to the state of the different phases or components. Therefore they
can refer to gas-liquid flows, liquid-solid flows or gas-particle flows or bubbly flows, and
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 55
so on. In the studied problem the interaction of flow in the screening chamber can be
categorised as liquid-solid flows and interaction with the atmosphere as gas-liquid flows.
4.4.3 Approaches to Multiphase Modelling
Computational fluid mechanics (CFD) modelling in recent years has provided the
basis for further insight into the dynamics of multiphase flows. Currently there are two
approaches for the numerical calculation of multiphase flows: the Euler-Lagrange
approach and the Euler-Euler approach.
4.4.4 Euler-Lagrange Approach
The Lagrangian discrete phase model follows the Euler-Lagrange approach. A
fundamental assumption made in this model is that the dispersed phase occupies a low
volume fraction, even though high mass loading is acceptable. This approach is
generally used for highly dispersed flows where the volume fraction of the dispersed
phase is small. The time-averaged Navier-Stokes equation is solved for the fluid phase
which is treated as a continuum, while the dispersed phase is solved by tracking a large
number of particles, bubbles, or droplets through the calculated flow field. There is an
exchange of interfacial momentum, mass, and energy between the dispersed and the
continuous phase. The particle or droplet trajectories are computed individually at
specified intervals during the fluid phase calculation. The model is appropriate for the
modelling of spray dryers, coal and liquid fuel combustion, and some particle-laden flows.
However, it is inappropriate for the modelling of liquid-liquid mixtures, fluidized beds, gas-
liquid flow or any application where the volume fraction of the secondary phases is not
negligible. In the current research we had to consider interaction between gas-liquid
flows hence the current research did not adopt the Euler-Lagrangian approach.
4.4.5 Euler-Euler Approach
In the Euler-Euler approach the fluid phases are treated mathematically as
interpenetrating continua. Fluids are treated in every computational cell using the
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 56
concept of phasic volume fraction. For a two phase flow situation each of the phases is
considered to occupy a fixed volume fraction in a computational cell. This is because the
volume of one phase cannot be occupied by the other phase. These volume fractions
are assumed to be continuous functions of space and time and their sum is equal to one.
Conservation equations for each phase are derived to obtain a set of equations, which
have a similar structure for all phases. These equations are closed by providing
constitutive relations that are obtained from empirical information, or, in the case of
granular flows, by application of kinetic theory.
The present study is based on the Euler-Euler approach, because the research
deals with the interaction between gas and liquid flows. The present research was carried
out by using the commercial CFD package AVL FIRE [22]. The FIRE Eulerian Multiphase
Module allows for the use of the following models based on the Euler-Euler approach.
These are listed below in the order of increasing accuracy:
Homogeneous (Equilibrium) Model
Multi-fluid Model
Volume of Fluid (VOF) Free-Surface Model
Homogeneous Model
The homogeneous model is the least accurate multiphase model based on the
Euler-Euler approach. A volume fraction equation is calculated for each phase. However,
only a single momentum equation is calculated for the phases in momentum equilibrium.
Multi-fluid Model
All conservation equations are solved for each phase. Since the multi-fluid model
requires by default the calculation of the complete set of the conservation equations for
each phase, it represents the basis for the Euler-Euler schemes in the FIRE Eulerian
Multiphase Module. The commercial software AVL FIRE’s user-defined subroutines
(UDF) allow for customizing the calculation of the mass, energy and momentum
exchange.
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 57
VOF Model
The VOF model is a surface tracking technique applied to a fixed Eulerian grid
proposed by Hirt and Nichols [82]. This model is designed for two or more immiscible
fluids where the accurate prediction of the interface between the fluids is of interest. In
the current research we had to consider the interaction between air and liquid flow, hence
the concept of VOF was important to explain the interaction between air and water in the
cell volume. In the VOF model, a single momentum equation is shared by the fluids, and
the volume fraction of each of the fluids in each computational cell is tracked throughout
the domain. Applications of the VOF model include stratified flows, free surface flows,
filling, sloshing, the motion of large bubbles in a liquid, the motion of liquid after a dam
break, the prediction of jet break-up (surface tension), and the steady or transient
tracking of any liquid-gas interface.
From the numerical perspective the volume of fluid model is very similar to the
homogeneous model. A single momentum equation is calculated for all phases that
interact using the VOF model. However, the calculation of volume fraction equations
using the VOF model is considerably more accurate allowing for the sharp resolution of
the interfaces. One of the common defects of the VOF calculation can occur when the
interface is not resolved sharply despite the use of the high-order discretization
techniques for the volume fraction equation – in that case the VOF model degenerates
into the homogeneous model. This is quite common in many practical calculations. It
happens due to the very high resolution requirements of the VOF model that can be often
hard to fulfil.
In the following sections, details of the modelling procedures including model
geometry, solution procedures and governing equations solved for each geometry will
be discussed further.
The Navier-Stokes equations can predict fluid flow behaviour in its general form.
The hydrodynamic characteristics of the overflow sewage screening device were
investigated using a CFD model by adopting the finite volume method in the Euler–Euler
approach. The 3D multiphase flow numerical model was developed using a commercially
available CFD package, AVL Fire [22] to predict the flow over the ogee weir.
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 58
4.4.6 Model Geometry and Computational Methodology
The screening device has a rectangular tank (1m X 0.2m), an ogee weir and an
inclined surface. The height of the ogee weir bottom is 0.75m and the diameter of the
inlet pipe is 0.2m. A design flow rate of 40 litres/second was considered for this analysis.
The outflow was assumed to be free flowing and perpendicular to the outlet surface at
the edge of weir. A 3D CAD model, similar to the schematic diagram shown in Figure
4.4, of the proposed sewage overflow screening device was developed using a CAD tool.
The volume mesh generated by using the CAD model for the CFD analysis is
shown in Figure 4.5 and Figure 4.6. Two different inlet positions were chosen to analyse
water level, flow velocity and shear stress distribution for the proposed screening device.
The model which was developed included the following features and assumptions:
Unsteady state multiphase solution for momentum and continuity was considered
Standard k-ε turbulence model for the turbulence modelling was employed
A cell centred finite volume approach was used to discretise the governing equations
and the resulting discretised equations were solved iteratively using a segregated
approach
Pressure and velocity were coupled using the Semi Implicit Method for Pressure Linked
Equations (SIMPLE) algorithm, [128].
Least squares fit approach was used for the calculation of the derivatives. Some other
important works include [32] and [124] on turbulence modelling and [130] on the self-
cleaning approach.
For momentum and turbulence, a first order upwind differencing scheme was used
whereas a central differencing scheme with second order accuracy was used for the
continuity equation
Screening device walls were treated as standard wall functions with no slip condition
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 59
Air was considered to be the dispersed phase and water was considered to be the
continuous phase.
Figure 4.4: Geometric details of the screener device
4.4.6.1 Governing Equations
Basic Eulerian equations (Equations. 4.17 and 4.18), describing multiphase flow
were expressed by the conservation equations for mass and momentum as follows:
Continuity:
t
kk
+ . kk v
k = 0 k= 1,……,N (4.17)
Where N is the number of phases, k is volume fraction of phase k, k is
density for phase k, vk is phase k velocity and 1
1
kk
Momentum Conservation:
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 60
tkk
kv+ . kk v
k v
k = - k p + . k ( k + t
kT ) + k fk +
kll ,1
Mk
(4.18)
kll ,1
Mkl represents the momentum interfacial interaction between phase’s k and l, f is
the body force vector which comprises gravity (g), p is pressure. Pressure is assumed to
be identical for all phases:
p =kp k = 1, ……., N
The phase k viscous stress integral is divided into non-transposed and
transposed terms. It can be expressed as:
k = k (v k + v Tk ) (4.19)
Where, k is the molecular viscosity. For incompressible flow, Reynolds stress,
tkT , takes into account the effect of turbulence which can be expressed by the
Boussinesq hypothesis:
tkT = - k
kkvv = tk (v k + v T
k ) - 32
kkkk (4.20)
Where k is the Kronecker delta function and t
k is the turbulent viscosity. For
the continuous phase, turbulent viscosity was calculated by adding shear induced
turbulent viscosity with Sato’s viscosity due to bubble induced turbulence [142].
tc = SIt
c, + BIt
c, (4.21)
Where shear induced turbulent viscosity for continuous phase can be expressed as,
SItc
, = Cc
c
2ck
(4.22)
Sato’s viscosity due to bubble induced turbulence can be expressed as [142],
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 61
BItc
, = Csato c D
bvr d (4.23)
where C = 0.09 and C
sato = 0.6 are dimensionless constants, k is the turbulent
kinetic energy and is its dissipation rate which can be obtained by solving equations
for the standard k- turbulence model put forward by Launder and Spalding [102]. The
turbulent kinetic energy (k) equation can be expressed as:
N
kllklkkkkkk
k
tk
kkkkkkkkk KPkk
tk
,1.v.
(4.24)
k = 1, ……… , N
N
kllklK
,1 is the interfacial turbulence exchange between phases and Pk is the production
term due to shear. Turbulence dissipation (ε) equation is,
k
kkk
k
kkk
N
kllklk
tk
kkkkkkkkk
kC
kPCD
t2
2
1,1
.v.
(4.25)
Closure coefficients used in the model are k =1.0, =1.3, C1=1.44, C2=1.92,
C
=0.09
N
kllklD
,1 represents interfacial dissipation exchange between phases.
Momentum interfacial exchange between gas and liquid was modelled by
implementing interfacial momentum source at the interface which includes drag and
turbulent dispersion forces (AVL 2008)
Mc= C
D 81
c iA vr
vr + C
TD c ck d = dM (4.26)
Where c denotes continuous (water) and d denotes the dispersed phase (air).
The first term in Equation (10) represents mean contributions due to drag force and the
second term takes into account the turbulence effect.
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 62
The drag coefficient,DC , is a function of the bubble Reynolds number, Re
b. The
following correlation for drag coefficient,DC , was used [22]
687.0Re15.01Re24
bb
DC Reb 1000 (4.27)
Bubble Reynolds number, Reb, and can be defined as:
c
brb
D
vRe (4.28)
Where c is the kinematic viscosity for the continuous phase relative velocity is
defined as:
vr = v
d - v
c
The interfacial area density for bubbly flow can be expressed as [63]:
b
di DA 6 (4.29)
Where Db = 0.01 mm is the bubble diameter and d is dispersed phase volume
fraction. Bubble dispersion coefficient used in Equation [3.26] was,
CTD= 0.1
The flow was initialised in the simulations with small initial values assigned to k
and ε, which made the initial turbulent viscosity roughly equal to the kinematic viscosity
for water. The fluid properties for air and water were taken as the properties at NTP (T
= 293.15 K, P = 1 atm). Typical turbulence quantities at the inlet of the domain were
calculated from inlet velocities by considering turbulence intensity, I = 0.05 where,
81-
inlet 0.16(Re)UuI / .
The simulation was carried out considering an unsteady state condition with time
steps of ∆t = 0.05 second. Total time period for each run was 180 seconds which was
adequate to obtain time averaged steady state results and numerical stability. Three grid
resolutions were tested for the grid independency test. The purpose of a grid
independency test is to determine the minimum grid resolution required to generate a
solution that is independent of the grid used. Starting with a coarse grid, the number of
cells was increased in the region of interest until the solution from each grid was
unchanged for successive grid refinements. All the cells in the calculation domain were
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 63
polyhedral with a large number of hexahedral cells. As the computational domain
consisted of hybrid unstructured meshes in a curvilinear non-orthogonal coordinate
system with artesian base vectors and refined regions in some locations, calculating the
number of cells in each direction was complicated.
Figure 4.5: Position 1 (condition 1) is the inlet parallel to the ogee weir
A view of the coarse computational grid is shown in Figure 4.5 and Figure 4.6,
which consists of a total of 38619 cells in the whole computational domain. Meshing
procedure was done by Fame Advanced Hybrid meshing technique [22].
Figure 4.6: Position 2 (condition 2) is the inlet perpendicular to ogee weir
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 64
4.4.7 Boundary Conditions
To simulate the numerical model for designed flow, it is important that boundary
conditions accurately represent what is physically happening in the device, refer to
Figure 4.7.
Figure 4.7: Boundary conditions used in the CFD model
Inflow Condition: Inlet mass flow (designed) was 40 kg/s and it was assumed that
flow direction was perpendicular to the inlet surface (parallel to inlet pipe) shown in red
(Figure 4.7).
Atmospheric Boundary: Atmospheric pressure boundary condition was
considered for the upper curved surface as shown in blue in Figure 4.7.
Wall Boundary: Roughness height of the ogee weir surface was assumed as
0.001m and wall functions are based on the assumed logarithmic velocity distribution
(mark with pink in Figure 4.7). The friction velocity (Uτ) is defined by;
wU (4.30)
Where, w- shear stress and -fluid density. For mean velocity the following wall
functions are used:
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 65
63.11, *** <yyU (4.31)
63.11,ln1 *** >yEyK
U (4.32)
Where,
pp UUk
CU
2141* (4.33)
and
pp ykCy
2141* (4.34)
Index “p” denotes the values at the centre point of the wall-nearest the control
cell, k is kinetic energy, μ is viscosity for mean velocity and y is the normal distance from
the wall. Here, coefficient C = 0.09, E = 9 is the constant in the law-of the-wall and K =
0.41 is von Karman constant. The near wall viscosity (μw) can be expressed by;
*
*
Uy
w (4.35)
And the wall shear stress (τw) defined by;
ptwt
p
ppw UU
EykKC
*
2141
ln
(4.36)
Where,
wwpwpwptwt nnUUUUUU (4.37)
The subscript’t’ denotes the tangential direction, parallel to the wall surface.
Outflow Condition: Outflow was assumed to be a free flow and perpendicular to
the outlet surface at the edge of the weir, shown in green in Figure 4.7.
4.4.8 Explaining CFD Results
Time periodic results were taken at the end of every 2 seconds for a total time of
180 seconds for every simulation run. Then, the values obtained for velocities and
turbulence properties were time averaged for all the time steps. To obtain a converged
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 66
solution, the approach used was to reduce the normalised sum of absolute residuals to
a value of 1.0 x 10-4 for each time step. The water level was determined by considering
the flooding and drying concept reported by Stelling [157], which suggested a value ≥
50% of the volume fraction to be considered as water level. Flow reflection dominated
the cross section profile of the sewage overflow device.
Figure 4.8: Water levels over the weir at different locations
To analyse the 3D numerical results for water levels, velocities and to show
changing profiles, three distinct sections across the width of the weir were selected, left,
middle and right. The position of the left, middle and right sections in the simulation result
are shown in Figure 4.8. The first set over the weir, and the second and third sets, are 3
cm and 6 cm downstream of the ogee weir respectively. For the presentation of
numerical results, locations parallel to the weir inlet are numbered as points 1, 2, 3 and
perpendicular to the weir inlet are numbered as points 4, 5, 6.
A multiphase modelling approach was selected to model the air and water.
Volume fraction of the water as phase 2 was selected to determine the free surface
profile. The results of the final time steps volume fraction are shown in the Figure 4.9. In
determining water level, assumptions were made based on the flooding and drying
concept
Figure 4.9 suggest there are some wave reflections on the right sight and the
splashing of water will be within the device. Figure 4.10 suggest wave reflections on both
Left Water Level Middle Water Level Right Water Level
Width of weir opening 1000 mm
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 67
edges of the weir. The wave reflection of such small device is very important aspect in
designing the device to maximise self-cleansing effect at the bottom.
Figure 4.9: Volume fraction of water at inlet parallel (position 1) to ogee weir
Figure 4.10: Volume fraction of water at inlet perpendicular (position 2) to ogee weir
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 68
4.4.9 Plausibility check of the CFD model
Plausibility check of the CFD model was done using a simplified analytical
solution. To find an analytical solution using the Navier-Stokes equations some simple
assumptions were made.
Figure 4.11: Comparison of flow velocities over the top of the ogee weir
Figure 4.12: Comparison of flow velocities 6cm downstream of the ogee weir
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 69
Firstly, the flow was considered to be steady and uniform, flowing under the
influence of gravity and parallel to the bottom surface while the effect of air viscosity at
the top surface was negligible. Analytical equations for shear stress, unit flow, average
velocity and curve of the ogee weir were calculated and reported [10]. Plausibility of the
numerical water levels was checked against the analytical water level. The (one-
dimensional) analytical solution matches the trend of the 3D CFD water level reasonably
well, (Figure 4.11 and Figure 4.12). In the absence of experimental data this provides a
decent plausibility check for the numerical model.
4.5 Discussion of Results
4.5.1 Discussion of Hydrodynamic results
To understand flow reflections, CFD simulated water levels at the left, middle and
right sections along the flow were extracted. These results were compared with the one-
dimensional analytical solution considering steady, incompressible fluid. Analytical
formulation is unable to include flow reflections from the wall. A model result shows the
dominant effect of flow reflection in the relatively small sewage overflow device as shown
in Figure 4.13 and Figure 4.14.
Figure 4.13 Comparison of water level along the flow for Condition 1
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 70
Figure 4.14 Comparison of water level along the flow for Condition 2
Figure 4.15 and Figure 4.16 shows how the velocity changes over the weir. As
the flow propagates downstream of the weir, the velocity at the bottom increases three
times that of the velocity over the ogee weir in both cases. This increase in velocity will
effectively increase the self-cleaning capacity of the device.
If screens are provided near the top of the weir surface at condition1, only the
right-hand side strip will get efficient self-cleansing while the left side holes are likely to
be blocked by larger pollutants in the sewer water.
However as flow becomes uniform (across the width) near the bottom of the weir
surface, the self-cleansing capability can be achieved. Keeping this fact in mind, it is
proposed to provide perforations (circular holes) near the bottom of the weir surface.
Analysis from flow velocity and pressure will explain further in regards to perforations
location.
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 71
Figure 4.15: Velocity vector at the inlets parallel (left)
Figure 4.16: Velocity vector at perpendicular (right) to the ogee weir
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 72
To understand how the velocities change due to the reflection effect at different
sections, comparison of velocities at different inlet orientations are shown in Figure 4.17
and Figure 4.18 for inlet condition 1 and condition 2.
Due to varying water levels (high water level near the right side and low water
near the left side), near the top of the weir level surface, the self-cleaning property will
not be as effective near the top region at condition 1. There is not much variation of
pressure in both condition 1 and condition 2 as both cases atmospheric pressure was
used as boundary condition, refer to Figure 4.19.
Figure 4.17: Comparison of flow velocities along the width for condition1
Figure 4.18: Comparison of flow velocities along the width for condition 2
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 73
Figure 4.19: Pressure variation at condition 1
Figure 4.20: Pressure variation at condition 2
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 74
If screens are provided near the top of the weir surface at condition 1, only the
right-hand side strip will achieve efficient self-cleaning while the left side holes are likely
to be blocked by larger pollutants in the sewage water. However as flow becomes
uniform (across the width) near the bottom of the weir surface, the self-cleaning capability
can be achieved towards the bottom. Keeping this fact in mind, it is proposed to provide
perforations (circular holes) near the bottom of the weir surface.
As the water flows down, its velocity increases due to gravitational acceleration.
CFD simulation shows the formation of flow separation near the ogee weir. The CFD
simulated shear stress is lower than the analytical value as it is unable to consider flow
undulations. Moreover analytical calculation assumed an in-viscid fluid domain without
having any boundary layer effect while CFD simulation considered viscous boundary
effects, refer to Figure 4.21. The additional shear stress will be beneficial for removing
the sewer solids sticking near the perforations. Figure 4.22 shows distribution of shearing
stress on the sewer overflow screening system.
Figure 4.21: Shear stress distributions for the inlet parallel to the weir width
Analysis of the shear stress along the flow path was performed to identify efficient
self-cleaning location. From CFD simulation it was shown that shear stress increases
significantly towards the bottom of the inclined surface, which suggests that, the location
Screening area
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 75
of the screen should be towards the bottom. Moreover, the flow becomes uniform across
the width in this area, which will also help effective screening.
4.5.2 Discussing Location of Circular holes
Due to varying water levels (higher at the right than the left side), near the top of
the weir surface, the self-cleaning property will not be as effective near the top region at
condition 1. If screens are provided near the top of the weir surface as at condition 1,
only the right-hand side strip will have efficient self-cleaning while the left side holes are
likely to be blocked by larger pollutants in the sewage water. However as the wave
becomes uniform (across the width) near the bottom of the weir surface, the self-cleaning
capability can be achieved. Additionally as the water flows down, its velocity and shear
stress increases. Velocity of the sewage flow increased up to three times near the
perforated holes from around 1m/s to 3m/s for an inlet flow of 40 l/s. Shear stress also
increased substantially from around 100 N/m2 to 300 N/m2 close to the perforated holes
[10]. Keeping this in mind, it is proposed to provide perforations (circular holes) near the
bottom of the weir surface.
Figure 4.22: Comparison of shearing stress along the bottom of the curved
surface
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 76
4.5.3 Discussion of the Inlet performance
The CFD simulated results showed that due to wave reflection under condition 1
(inlet parallel to the weir, refer to Figure 4.23), water level on the right side overrode the
water levels in the middle and left sections.
Figure 4.23: Comparison of water levels along the flow for conditions 1, water
level on the top of the weir, 3cm and 6 cm downstream
Figure 4.24: Comparison of water levels along the flow for conditions 2, water
level on the top of the weir, 3cm and 6 cm downstream
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 77
Whereas, under condition 2 (inlet perpendicular to the weir, refer to Figure 4.24),
the reflected wave contributed to elevated water levels towards both the left and right
sides of the device.
Towards the bottom of the inclined surface, the wave became an equally
distributed turbulent flow (across the width).The reflected water level on the right side
reduced as the wave traveled downstream of the ogee weir (condition 1). Due to higher
water level, higher velocity was found downstream near perforations which were a
favorable condition for self-cleaning. Condition 1 can provide a better screening effect
on the right side because high water levels generate higher velocity and shear stress
(Figure 4.25).
Figure 4.25: Impact of device inlet position on the wave reflection viewed from
the back of the weir with a lateral inflow
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 78
4.5.4 Standard Weir orientation
Since ogee weirs provide excellent hydraulic features, in-depth research has
been carried out to determine the standard shape and size of the crest of the overflow
spillway [41]. With extensive physical modelling tested by the U.S. Corp of Engineers,
they suggested a standard design parameter for the ogee weir (US Corp of Army [169]
& [170]. The proposed experimental design conditions were tested using the CFD
modelling technique to maximize uniform flow for better screening. Inlet conditions P &
Q with 0H: 3V and 2H: 3V slope demonstrate similar types of water level variations due
to wave reflections from the wall (Figure 4.26).
Figure 4.26: The Waterways Experimental Station (WES) standard spillway
shapes [41]
P Q
R S
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 79
In fact orientation P produced more wave reflection than the rest. Orientation S
with 3H: 3V slope also demonstrated scanty water level available on the left of the weir.
Inlet condition R with 1H: 3V slope for the rectangle with the ogee showed less wave
reflection compared with others.
Orientation R provided the best self-cleaning effect and maximum screening
efficiency downstream. With orientation R flow become equally distributed across the
weir ensuring more effective self-cleaning (Figure 4.27).
Figure 4.27: CFD results viewed from the back of the weir with a lateral inflow on four
standard inlet orientations as suggested by the U.S. Army Engineers Waterways
Experimental Station
P
Q
R
S
Starting point of Equally distributed
Wave
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 80
4.6 Limitation of Screening Device
The device self-cleansing effect largely depends on falling water but after some
modification it could not reduce blinding of the perforations. This is a major limitation as
the annual maintenance requirement exceed regularly due to blinding of the perforated
screening during CSO events [131].
4.7 Summary
Some of the key finding from the CFD model are summarised below:
The inlet parallel to the ogee weir was considered a better inlet option as water
level over the ogee weir was higher due to wave reflection on the right side which can
provide higher velocity and shear stress. Condition 1 can produce a better self-cleaning
mechanism due to higher velocity and shear stress near the perforated holes.
The flow becomes uniform near the bottom of the inclined surface with higher
velocity and shear stress. This suggests that the perforations should be placed near the
bottom of the inclined surface to achieve an effective self-cleaning capability for the
device. Uniform flow towards the bottom of the inclined surface will help to remove any
pollutants adhered to the perforations.
As the sewage overflow screening device is small, the wave reflection effect
was found to be dominant for this device. It is suggested that a 1.5m long inlet will reduce
wave reflection up to 10% compared to a 0.3m inlet.
Four standard ogee weir orientations were analysed to reduce wave reflection.
Orientation R with an inclined slope 1H: 3V from the rectangular device to the ogee weir
was found to be the most efficient based on the best practice guidelines provided by the
U.S. Army Corp of Engineers.
The study provided valuable insights into designing an efficient and effective
gross pollutant device. The experimental work was restricted by the physical limitations
of cost and time which are inherent in laboratory studies. CFD modelling provided an
excellent opportunity to design the gross pollutant screening device. Study findings will
help to maximize its functionality and its effectiveness in trapping sewage solids with high
efficiency, both in terms of the capture of sewage solids and the development of a self -
Chapter 4: Hydrodynamic Analysis
Md Abdul Aziz 81
cleaning mechanism. Chapter 5 will discuss an improvement to the proposed self-
cleaning device for it to work efficiently in remote un-staffed locations.
Work presented in this chapter has been published in the following journal and
conference papers:
Aziz, M. A., Imteaz, M., Huda M., & Naser J., 2014a, ‘Optimising inlet
condition and design parameters of a new sewer overflow screening device using
numerical modelling technique‘, Journal of Water, Sciency and Technology vol.70,
no.11, pp.1880-1887
Aziz, M. A., Imteaz, M., Naser J., & Phillips, D., 2013b, ‘Hydrodynamic
Characteristics of a New Sewer Overflow Screening Device: CFD Modelling and
Analytical Study’, International Journal of Civil and Environmental Engineering, vol. 7,
no.1, pp.71-76.
Aziz, M. A., Imteaz, M., J. Naser, Nazmul H., & Phillips, D. 2010,
‘Hydrodynamic Characteristics of a proposed sewer overflow screening device’, at The
5th Civil Engineering Conference in the Asian Region, 8-12 August, Sydney Australia.
Aziz, M. A., Imteaz, M., Nazmul H., & Jamal N., 2013d, ‘Understanding
functional efficiency of a sewer overflow screening device using combined CFD and
analytical modeling’, 20th International Congress on Modelling and Simulation, Adelaide,
2 to 5 December, Australia.
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 82
Chapter 5
Improvement of the
Screening Device
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 83
5.1 Introduction
Under wet weather conditions, sewage overflows are of serious concern to the
environment, aesthetics and public health. Sewage solids are dispersed, suspended or
washed into rivers. They eventually settle, creating odours and a toxic/corrosive
atmosphere in bottom mud deposits. The solids also are aesthetically unpleasant either
in their general appearance or because of the actual presence of specific, objectionable
items, such as floating debris, sanitary discards/faecal matter, scum or even parts of car
tyres. In order to reduce these sewage solids, different types of screening devices are
used in existing networks. According to Faram et al., [60] screening is the only
economically viable method to segregate sewer solids in most cases.
Although the sewage solids overflow screener, discussed in Chapter 4, worked
well and had reasonable capture efficiency, the self-cleaning mechanism only gave
reasonable results. Blockages on the screener reduced its screening capacity which
reduced the capture efficiency of the gross pollutant device. Similar findings [9] report
that hairs from fibrous material wrap around the wires of bar screens and reduce
screening efficiency. This blockage effect occurred more often with small sewage solids
less than 10 mm in diameter.
This triggered the need to research efficient and effective self-cleaning screening
devices and screening handling systems. Past studies used a number of different
screening systems in sewage overflow locations. Hydrodynamic vortex separation
(HDVS) was a popular screening concept developed in the early 1960s. First generation
HDVSs were found to be effective in retaining 70% of the pollution load [151]. To improve
the pollution load second generation HDVSs developed by the American Water Works
Association and EPA were reported by Field [64] and the third generation device in the
1980s was commercially patented as Storm King® overflow. The HDVSs went through
a series of performance evaluations in Europe, North-America, and Japan [20].
Unfortunately lightweight sewer solids of neutral buoyancy were not trapped in HDVSs
[16]. A comprehensive review of different screeners was provided by Saul [144] & [146].
The recent update of these research papers can be found in the work of [106] and [107].
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 84
Reported literature suggests that screens need to have a self-cleaning mechanism to
work efficiently.
To overcome this challenge a non-powered self-cleaning screening system that
can capture neutrally buoyant light weight sewage solids greater than 6 mm in two
diameters was tested by [152]. Later [59] tested six hydro jet devices installed in USA,
Australia and mainland Europe. However, in most cases the devices were directly
associated with blockages of the sewerage system. Usually the most conventional
screening systems utilise electro-mechanical components to facilitate such a process
[131]. However, given the harsh, unstaffed, remote operating environment of many
sewage overflows this device is not ideal [10]. Some of the common drawbacks in the
available commercial sewer overflow screening system include inadequate screening
capacity, requirements of external power and high cost [153]. To overcome such
problems a novel self-cleaning sewage overflow screening system is proposed [11].
Some of the key attribute of the proposed sewer overflow screening device include: less
expensive, has low maintenance and contains no moving parts, no moving parts and
robust stop/start operation. The proposed screening system consists of temporary
holding tanks that provide a transient storage and real time control of sewerage overflow.
The aim of this proposed device was to overcome key limitations in existing
commercial screening devices. The key attributes of the proposed Comb Separator
device are listed below:
To minimise blockage on the screen of solid sewage materials at the Comb
Separator the device should be built so that minimal inspection and maintenance
are required. The device should be able to be constructed above ground level.
The treatment capacity required for the drainage system must be able to cope
with a maximum one year overflow event i.e., 70 l/s per meter width of the device
and a minimum of a one in four months flow of 20 l/s per meter width.
The screening device should capture more than 90% of solids exceeding 10 mm
in diameter and return them to the sewerage system for all events up to and
including the one year overflow event (70 l/s). It should provide high capture
efficiency (more than 80%) for solid sewage materials less than 10 mm in
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 85
diameter. If the proposed screening device can capture such high efficiency than
the ongoing operation and maintenance cost will be minimum.
It should not use sophisticated electrical, mechanical systems or signals which
may become ineffective during extreme events, such as floods or lengthy
droughts. In the event of a device failing there must be a bypass option causing
no serious damage to the device or the environment.
The present research focused on long-term operational performance, durability
and low maintenance requirements. In comparison to the existing electro-mechanical
screeners, the proposed Comb Separation device has negligible blinding during
screening and better applicability for remote locations. As there are no moving parts,
there will less maintenance and operational cost for the device as these screening
located in harsh environmental conditions [8]. This research also focused on optimising
the performance of the proposed comb separator to ensure high capture efficiency.
A series of laboratory tests were carried out to evaluate steady state, short
duration (varying from 6 minutes to 32 minutes) flow conditions with flow variations from
20l/s to 70l/s. Instead of using static screens which have a high maintenance cost, a self-
cleaning comb separation device was used (Figure 5.1). Analysis of the result will
discuss the key experimental conditions of flow, comb spacing, effective spacing
(spacing between different layers of comb), comb layers, weir opening, and runtime.
The key objectives of the experiments are listed as follows:
To analyse capture efficiency of sewage solids more than 10 mm in diameter
To comprehend the impact of experimental design parameters (flow discharge,
weir opening, comb spacing and layers) on sewage solids capture efficiency
To check reproduction of the experimental results and ensure generation of
consistent results
To compare performance of the proposed Comb Separator with the standard
Hydro JetTM screen under low flow (up to 70 l/s) conditions.
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 86
5.2 Methodology used in the Experiment
5.2.1 Data Collection
A series of experimental trials were conducted to monitor capture efficiency of the
experimental device.
Figure 5.1: Experimental set ups Comb Separator
AA = Width of the sharp crested weir,
BB = Distance between 1st and 2nd comb spacing,
EE= Distance between sharp crested weir to the outlet chamber,
FF = Height of the sharp crested weir from sewer screening chamber,
GG= Distance from the face of the sharp crested weir to the angle screener,
HH = Distance from the sharp crested weir to the angle screener,
JJ = Height of the angle screener, KK = Diameter of the screening chamber,
MM = Distance of the angle screener from the outlet chamber,
NN= Diameter of the valve ball, PP = Fall distance of the valve ball,
QQ= Height of the sharp crested weir from bottom of the chamber,
RR = Height of the device screening chamber from setup location,
TT= Height of the outlet chamber.
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 87
Six different experimental set ups with twenty two different experimental
conditions were tested in the experimental facility at Swinburne University of Technology,
Melbourne, Australia. The flow diagram of the experimental process is shown in Figure
5.2.
Figure 5.2: Flow diagram of the revised screening experimental works
The experimental conditions were varied by changing the critical experimental set
ups such as flow discharge, weir opening, comb spacing and comb layers. A total of 51
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 88
sets of experimental data were collected. The schematic diagram of the experimental
set-up is provided in Figure 5.2. To test and validate the performance of the Comb
Separator, critical model parameters, such as runtime, flow discharge, weir opening,
spacing and layers of combs were varied. The aim was to achieve capture efficiency at
or above 80% with minimal blockage on the combs. Overflow event was on a one in one
year overflow for sewer solids less than 10mm like cigarette butt, refer to Figure 5.3. The
robustness of the optimum experimental conditions was validated by repeating the
experiment several times to ensure reasonable consistency of results.
Figure 5.3: Concept diagram of target capture efficiency curve
5.2.2 Screening Mechanism
The laboratory device was connected to an inlet pump and inlet pipe. Two outlets
were mounted on the device - one to convey overflow water away and the other to drain
the sewage water remaining in the storage chamber [9]. A series of combs, to segregate
sewage solids from the sewage overflow, were mounted next to the sharp crested weir
(Figure 5.4).
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 89
Figure 5.4: Experimental set up for the proposed sewage overflow screening device
Phase 1: After the start of precipitation overflow, the storage chamber fills with
sewage. A floating ball, at the bottom of the sewage solids holding chamber, closes at
this point as shown in Figure 5.5. As the overflow continues, the storage chamber
overflows above the sharp crested weir. The captured sewage solids are intercepted by
the parallel combs and fall into the holding chamber (pollutant capture chamber).
Figure 5.5: Operation procedure of new sewage overflow screening device- Phase 1
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 90
Figure 5.6: Operation procedure of new sewage overflow screening device-Phase 2
Phase 2: After cessation of precipitation, the water level within the storage
chamber falls below the valve level. The low pressure of the liquid in the sewage solids
holding chamber allows the ball to drop and flushes the entire captured sewage solids
back into the storage chamber which is shown in Figure 5.6.
Ball Valve Details
Efficient functioning of the lower ball valve chamber was the key aspect of the
device design. The device needs to work in unstaffed remote locations so self-cleaning
activity was the key functionality of the device. The detailed designs of ball valves used
in the sewage chamber, are shown in Figure 5.7.
Common sewage solids, like condoms, tampons, cigarette butts, wraps, cotton
balls and bottle caps, were used in the experiment. The experimental conditions were
varied with different flow volumes and the number spacing of comb layers. The sharp
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 91
crested weir effectively responded to the device failed condition and bypassed the
outflow chamber.
Figure 5.7: Showing the design parameter of the ball valve chamber
To obtain the optimum comb spacing, an experiment was set up using various
comb spacing’s was varied from 25 mm centre to centre to 10 mm centre to centre. Weir
openings were adjusted from 460 mm to 970 mm across the width of a 1000 mm weir.
To confirm blinding at the comb, separation was minimal. Two and three comb layers
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 92
were tested. In addition to these, flow volumes varied from a one in four month sewage
overflow condition (20 l/s/m) to a one in one year overflow condition (70 l/s/m). Sewage
solids tested in the experiment are classified as sewage solids more than 10mm in
diameter (toilet paper, bottle tops, dish wipes, tampons, condoms, cotton balls) and
sewage solids less than 10 mm (cigarette butts and artificial cigarette butts). Capture
efficiency of the sewage solids was calculated using Equation 5.1.
𝐶𝑎𝑝𝑡𝑢𝑟𝑒 𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦 (𝐶. 𝐸) =𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑜𝑙𝑖𝑑 𝑚𝑎𝑡𝑒𝑟𝑖𝑎𝑙𝑠 𝑟𝑒𝑡𝑎𝑖𝑛𝑒𝑑
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑜𝑙𝑖𝑑𝑠 𝑖𝑛𝑠𝑒𝑟𝑡𝑒𝑑 𝑤𝑖𝑡ℎ 𝑖𝑛𝑓𝑙𝑜𝑤∗ 100 (5.1)
Sample Data Collection
Data was collected for a series of different experimental conditions. Data was
collected using two specific diameter criteria. These were sewage solids more than
10mm in diameter and sewage solids less than 10mm in diameter. A sample data
collection processes for sewage solids less than 10mm in diameter is shown below which
helped to improve the cigarette butt capture efficiency in the Comb Separator screening
device.
This data collection exercise was conducted on 20th October 2010 at Swinburne
University Hydraulic Experimental Laboratory. For each data collection experimental run
the following standard reporting process was followed see appendix for further
information.
Aim: To test the cigarette butt capture efficiency of a two-comb arrangement.
Test set-up:
Two overlapped comb screens, the first with wires 25mm centre to centre
The second with wires 10mm centre to centre, combs 20mm centre to centre
Front comb 65mm from crest of weir
Retention screen 140mmm behind the crest of the weir
Concrete block placed on the floor of the model sewage chamber to distribute flow
Crest length reduced to 470mm
Retention screen screwed to the floor
19 cigarette butts wrapped in duct tape with a mean sample diameter of 8.82 mm.
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 93
Test 1
Head on weir 50mm giving a flow of 20L/m/s.
Nappe easily cleared the retention screen, refer Figure 5.13.
Test items:
10 artificial cigarette butts
Test run commenced 11:50 AM, Finished at 12:05 PM.
Capture efficiencies:
Cigarette butts: First pass, 1/10, second pass 0/1. 2 butt in down pipe.
Hence = (8/9) x100 = 89%.
Comments:
It was intended to close the bar spacing of comb two to 9 mm but time constraints
prevented this for that days tests. To offset this, the butts were wrapped in duct tape to
increase their diameters accordingly in order to simulate accurately the closer bar
spacing.
Further discussions about experimental data process can be found from Table 5.2 to
Table 5.7.
5.3 Test Procedure
A series of experimental test runs were completed at Swinburne University
Hydraulic Laboratory. The valve operation on the previous design was effective; hence
no alteration of the valve operation was suggested in the revised design. The blockage
on perforations was the key issue which was updated in the revised design.
5.3.1 Experimental Conditions Used
A series of different sewage solids were tested. Figures 5.8 and 5.9 show the
arrangement of the comb separator in the sewage overflow screening device. How the
sewer solids are capture during and after an experimental run is shown in Figure, 5.10
and Figure 5.11. Figure 5.12 shows inserting of sewer solids in the ‘comb separator’
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 94
whereas Figure 5.13 shows the comb separator in operation with three layers of combs
to capture sewer solids.
Figure 5.8: Vertical position of the comb separator in the device
Figure 5.9: Top view of the position of the Comb Separator
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 95
The device tested the following sewage solids: condoms, tampons, wrap papers,
tissue papers, cigarette butts, cotton buds, cotton balls, bottle tops, cans etc. The main
improvement achieved with this device was on blinding or blocking of perforations.
Figure 5.10: Capture of sewage solids during an experimental run
Figure 5.11: Capture of sewage solids after an experimental run
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 96
Figure 5.12: Mixing of sewer solids on to the Comb Separator device
Figure 5.13: Comb Separator is in operation, nappe clear the retention screen
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 97
The concept of the comb separator worked well against blinding or blocking of the
screening device. After some initial runs, it was found that the device could trap more
than 90% of sewage solids more than 10 mm in diameter. There was no reported
blockage on the comb separator. Figure 5.14 shows sewer solids used in the experiment.
Figure 5.14: Sewage solids used in the test
5.3.2 Optimising Experimental Results
Most of the sewage solids such as condoms, tampons, bottle tops, cans, wrap
papers etc. were trapped relatively easily. However, trapping cigarette butts was not
achieved at a satisfactory level. To improve the trapping efficiency of this relatively small
diameter sewage solid a series of adjustments were made in the experimental set up.
Primarily four key experimental parameters were changed to improve capture efficiency.
The four key parameters were:
Number of comb separator layers
Spacing of comb separator
Inflow volume; and
Weir opening
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 98
A series of six different experimental setups were used in the current investigation
for sewage solids less than 10mm in diameter as reported below:
Table 5.1 Different experimental setups used for sewage solid less than 10 mm diameter
Exp. Set-Up
Max Flow
(L/s/m)
Effective Spacing
(mm)
Spacing of 1st comb
Spacing of 2nd comb
Spacing of 3rd Comb
Layer of combs
Weir opening
(mm)
1 46 3.50 25 25 25 3 970 2 46 1.50 20 20 20 3 470 3 71 1.50 20 20 20 3 970 4 30 4.50 15 15 N/A 2 970 5 67 3.25 15 12.5 N/A 2 470 6 67 4.80 25 10 N/A 2 470
To obtain the optimum comb spacing a trial and error methodology was adopted
where comb spacing was varied from 25 mm centre to centre to 10 mm centre to centre.
The effective spacing is the spacing between two layers of comb as shown in Figure
5.15. Weir openings were adjusted from 460 mm to 970mm across the width of a 1000
mm weir length. Results of the trial and error methods and summaries are in the Tables
5.2 to Table 5.7:
Figure 5.15: Comb Separator in operation
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 99
Experimental Sewage Solids Materials less than 10 mm in diameter Table 5.2 Comb Separator Testing at Experimental Set up 1
Experimental Set-Up
Name of material
Flow (Litre/s)
Spacing of 1st comb
Spacing of 2nd comb
Weir opening
mm
Layer of
combs
Capture efficiency
%
1 Cigarette butts 45 25 25 970 3 60
Cigarette butts 45 25 25 970 3 85
Cigarette butts 40 25 25 970 3 75
Cigarette butts 45 25 25 970 3 55
Art Cig butts 45 25 25 970 3 67
Key Learning Capture efficiency increases but not up to desired level, artificial cigarette butts need to soak in water to simulate actual conditions, also used dry cigarette butts
Proposed Revision Reduce comb spacing to improve capture efficiency
Table 5.3 Comb Separator Testing at Experimental Set up 2
Experimental Set-Up
Name of material
Flow (Litre/s)
Spacing of 1st comb
Spacing of 2nd comb
Weir opening
mm
Layer of
combs
Capture efficiency
%
2 Cigarette butts 36 20 20 470 3 50
Cigarette butts 22 20 20 470 3 52
Cigarette butts 15 20 20 470 3 75
Key Learning Surprisingly, despite using 20mm centre to centre combs, the capture efficiency remains much less than expected.
Proposed Revision
Important to know each comb spacing layer performance on capture efficiency. Further reduce first comb spacing to 12.5mm and second comb spacing to 15mm
Table 5.4 Comb Separator Testing at Experimental Set up 3
Experimental Set-Up
Name of material
Flow (Litre/s)
Spacing of 1st comb
Spacing of 2nd comb
Weir opening
mm
Layer of
combs
Capture efficiency
%
3 Cigarette butts 45 20 20 970 3 69
Cigarette butts 45 20 20 970 3 69
Cigarette butts 45 20 20 970 3 77
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 100
Key Learning Cigarette butts are very buoyant, soaking for 24 hours provided decent improvement, no blockage in the screener
Proposed Revision
As blockage in the screener further reduces, the comb spacing also reduces. It was suggested to use two comb layers instead of three combs for further testing.
Table 5.5 Comb Separator Testing at Experimental Set up 4
Experimental Set-Up
Name of material
Flow (Litre/s)
Spacing of 1st comb
Spacing of 2nd comb
Weir opening
mm
Layer of
combs
Capture efficiency
%
4 Cigarette butts 15 15 15 970 2 91
Cigarette butts 30 15 15 970 2 75
Cigarette butts 20 15 15 970 2 85
Key Learning Positive improvement in capture efficiency, no blockage observed
Proposed Revision
Flow decreases lead to high capture efficiency. However, it is important to perform on high flow. Weir opening reduces to increase flow capacity.
Table 5.6 Comb Separator Testing at Experimental Set up 5
Experimental Set-Up
Name of material
Flow (Litre/s)
Spacing of 1st comb
Spacing of 2nd comb
Weir opening
mm
Layer of
combs
Capture efficiency
%
5 Cigarette butts 18 12.5 15 470 2 94
Cigarette butts 25 12.5 15 470 2 53
Cigarette butts 34 12.5 15 470 2 59
Cigarette butts 33 12.5 15 470 2 59
Key Learning High efficiency which is consistent with previous results except with cotton buds. Cigarette butt capture efficiency increases but remains unacceptably low at high flow.
Proposed Revision
Reduce second comb spacing and increase first comb spacing as velocity increases when solids pass first comb spacing which reduces capture efficiency.
Table 5.7 Comb Separator Testing at Experimental Set up 6
Experimental Set-Up
Name of material
Flow (litre/s)
Spacing of 1st comb
Spacing of 2nd comb
Weir opening
mm
Layer of combs
Capture efficiency
%
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 101
6 Cigarette butts 10 25 10 470 2 89
Cigarette butts 29 25 10 470 2 47
Art Cig butts 29 25 10 470 2 100
Art Cig butts 34 25 10 470 2 100
Key Learning No blockage in the screener observed
Observation As the second comb spacing is much closer than first comb spacing, capture efficiency increases to desired level.
5.4 Discussions of Results
5.4.1 Sewage solids more than 10 mm in diameter
Five different experimental set ups (considering set up 2 and 3 as same) were
used and sewage solids capture efficiency was recorded very high. The experimental
conditions used are shown in Table 5.8.
Table 5.8 Experimental set ups at five different conditions for sewage solids more than 10 mm in diameter
Exp. Set ups
Max Flow
(L/s/m)
Effective Spacing
(mm)
Spacing of 1st comb
Spacing of 2nd comb
Spacing of 3rd comb
Layer of
combs
Weir opening
(mm)
1 53 3.50 25 25 25 3 510
2 46 3.50 25 25 25 3 970
3 46 1.50 20 20 20 3 970
4 50 3.25 12.5 15 N/A 2 470
5 67 4.00 25 10 N/A 2 470
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 102
Analysing the performance of the proposed Comb Separator device, for sewage
solids larger than 10mm in diameter, a high capture efficiency of more than 95% was
obtained. The sewage solids used in the testing included toilet paper, bottle tops, cans,
dish wipes, tampons, condoms, cotton balls. Blinding on the combs was negligible during
the tests. The results were consistently capture sewer solids over 95% in all different
experimental set-ups (Figure 5.16), further experiments only considered smaller sewage
solids less than 10 mm in diameter.
Figure 5.16: Capture efficiency of sewer solids at different experimental set ups
5.4.2 Sewage solids less than 10 mm in diameter
5.4.2.1 Increasing the effective comb spacing
Blinding on the screener was a key limitation in designing the sewage overflow
screening devices in most reported literature studies [131]. The comb separator spacing
was reduced to make sure minimal blinding occurred on the combs. As the effective
comb spacing was reduced to less than 2mm, minor blinding occurred on the combs.
Minimal blinding was observed when effective screening was more than 3.5mm. It was
found that the flow had actually taken all the sewage solids either into or outside the
capture chamber without anything sticking against the comb separator bars. There were
90
91
92
93
94
95
96
97
98
99
100
Set up 1 Set up 2 Set up 3 Set up 4 Set up 5 Set up 6
Aver
age
Cap
ture
Effi
cien
cy
Experimental Set ups (Sewer Solids more than 10mm diameter)
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 103
no blockages at the end of most experimental runs, to get optimise consider more trials
were given till blockage were found by reducing comb spacing. Two layers of combs
were found to be more efficient than three layers.
Two different experimental set ups satisfied the design criteria of minimal
blockage on the comb and more than 80% capture efficiency. To confirm robustness of
the experimental set ups, the same experiment was repeated on high flow to validate
minimum variations of the experimental results. Setup 4, which performed well on low
flows, could not perform well on high flow. Setup 6 produced consistent results in
robustness checking within reasonable variations of 2% to 5% in total capture efficiency.
The optimum design criteria was achieved for the device set-up which is an effective
comb spacing of 4.8mm where the first comb spacing was 25mm, whereas the second
comb spacing was 10mm. The optimum design set-up is shown as experimental Setup
6 in Table 5.9.
Table 5.9 Capture efficiency with different experimental set-ups
Experimental Sewage Solids Materials less than 10 mm in diameter
Exp.
Setup
Max
Flow
(L/s/m)
Effective
Spacing
(mm)
Spacing
of 1st
comb
Spacing
of 2nd
comb
Spacing
of 3rd
Comb
Layer
of
combs
Weir
opening
(mm)
Average
Capture
Efficiency
%
1 46 3.50 25 25 25 3 970 68.4
2 46 1.50 20 20 20 3 470 66.5
3 71 1.50 20 20 20 3 970 71.0
4 30 4.50 15 15 N/A 2 970 83.7
5 67 3.25 15 12.5 N/A 2 470 68.7
6 67 4.80 25 10 N/A 2 470 90.7
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 104
As the effective spacing gap reduced to 2mm or less there was minor blinding
occurs on the ‘Comb Separator’ which reduces the capture efficiency of sewer solids
[12]. As the effective spacing increases the capture efficiency also increases. It is
important to note that flow volume also plays an important role in explain capture
efficiency. Figure 5.17 compiling average capture efficiency (%) and average flow (l/s) in
Y axis to give a better understanding about the performance of the effective comb
spacing (mm) in X axis.
Figure 5.17: Effective comb spacing (mm) against average capture efficiency (%) and
flow (l/s per metre length of weir)
5.4.3 Performance comparison of Comb Separator and Hydro JetTM
The Hydro JetTM screen had been subjected to a series of development, testing
and evaluation programs [15]; [17]. The screen used in the Hydro JetTM was a static one
with a specific focus on capturing sewage solids 6mm in diameter. The performance of
the Hydro JetTM screen focussed around screening capacity and effectiveness. The
proposed Comb Separator device used comb spacing which had minimal impact on
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 105
screening capacity during most of the experimental runs with different sewage solids.
The comparative performances of both the devices are tabulated in Table 5.10
In comparison to the Hydro JetTM, the Comb Separator performed with higher
efficiency and negligible blockage on the combs. In the current study, the experiments
were conducted on low flows (up to 70 l/s/m). Further improvement in the experimental
set up to generate high flows (90 l/s to 120 l/s) could provide better insight into this
screen’s suitability for managing higher flows [12].
Table 5.10 Comparative performance of the Hydro-JetTM and the Comb Separator
Hydro-JetTM Screen Mesh Aperture Size (mm)
Static 6 4 2
Average total Efficiency (%) 51 67 69
Number of Observations 17 20 12
Flow Range (l/s) 17 - 60 18 - 60 17 - 45
Average Flow (l/s) 45 43 33
Comb Separator Effective Comb Spacing (mm)
Effective Gap between combs 4.8 4.5 1.5
Average total Efficiency (%) 90.7 83.7 66.5
Number of Observations 6 14 6
Flow Range (l/s/m) 20 - 67 15 -30 30-70
Average Flow (l/s/m) 48 22 47
5.5 Limitations of the Experiment
The experimental facility can generate only up to 70 l/s flow hence the facility
would not be able to test on higher flows, for example 120 l/s. There are always some
physical limitations in the experimental set up. In this research the closest spacing
between combs that could be drilled was 10mm, therefore, the test could not be
conducted for spaces smaller than 10mm. It was recommend that further analysis and
sensitivity testing of the input parameters needs to be performed using standard
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 106
sensitivity testing. Chapter 6 will discuss in detail an ANN application in the current
research to overcome physical limitations of experiments, data generation, training and
testing of ANN model. Chapter 7 will discuss in detail the sensitivity analysis of the input
parameters of the proposed Comb Separator.
5.6 Summary
A new sewage overflow device with improved capture efficiency, low
maintenance, and a self-cleaning mechanism was tested at the hydraulic laboratory of
Swinburne University of Technology. The proposed device overcame most of the
common limitations in existing screening systems such as blinding (minimal blockage),
high maintenance requirements and the installation of an electrical-mechanical switching
system [154]. A series of trials with different numbers of combs, the spacing of combs,
flow volume and weir opening were tested. Some of the key findings from the
experiments are summarised below:
The proposed device can capture larger sewage solids of more than 10mm
diameter with greater than 95% capture efficiency
Capture efficiency is dependent on selected experimental parameters such as
flow, weir opening, comb spacing and layers which all vary sewage solids capture
efficiency. Two layers of combs were found to be more efficient than three layers
Increasing comb spacing improves capture efficiency. Robustness of the
optimum set up was tested to generate consistent results
Comparisons with the Hydro-JetTM suggest that the Comb Separator shows
minimal blockage and higher capture efficiency on low flows
The hydraulic experiments suggested good application potential of the proposed
device in urban sewerage systems. Further experiments with an improved pumping
device to generate higher flow (up to 120 l/s) could provide better insight especially with
smaller solids with a diameter less than 10 mm.
Work presented in this chapter has been published in the following journal and
conference papers:
Chapter 5: Improvement of the Screening Device
Md Abdul Aziz 107
Aziz, M. A., Imteaz, M., Rasel, H.M., & Phillips D., 2015a, Development and
Performance Testing of ‘Comb Separator’, A Novel Sewer Overflow Screening Device.
Accepted with minor review International Journal of Environment and Waste
Management, Vol 15, No 3, 2015.
Aziz, M. A., Imteaz, M., Rasel, H.M., Phillips, D., 2015c, ‘Performance Testing
of ‘Comb Seperator’ –A Novel Sewerage Overflow Screening Device’, ASEAN-
Australian Engineering Congress on Innovative Technologies for Sustainable
Development and Renewable Energy 11-13 March
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 108
Chapter 6
ANN Model to
Complement CFD
and Laboratory
Testing
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 109
6.1 Introduction
It is too complex to model different density sewer solids using the CFD modelling.
The laboratory experiments also has its limitations due to limited number of experimental
set ups are physically possible to conduct the experiments. To overcome these
limitations it was important to adopt an approach which can simulate the complex input
output relationship without knowing the underline physical characteristics. An ANN
model was adopted to perform this task. ANN models were suitable considering mixing
of different solids in the sewer system which leads to different density fluid flow
conditions. Moreover, the nonlinear relationship between input and output variables, and
complex physio-chemical interaction leads to difficulties in formulating mathematical
model based on physical laws.
The experimental work discussed in Chapter 5, the physical experiments allow
for a certain number of trials for experimental set ups. To visualize a range of different
conditions with and outside physical limitations of an experiment, it was important to do
modelling analysis. Moreover, experimental work involves significant cost and time. To
overcome these problems, current research set up a computational model using
experimental results. The initial focus was on the Computational Fluid Dynamics model
(CFD), which also provided reasonable results in developing and optimising the earlier
proposed gross pollutant device as discussed in Chapter 4. However, the studied
problem had some unique challenges which are difficult to model using physical law
based models such as CFD. These include:
Physical characteristics of different sewage solids particles
Multi-fluid sewerage systems with changing viscosity of fluid
Interaction between liquid and solid particles.
Considering these limitations, it was important to adopt an alternative method for
analysis. There are alternative heuristic approaches such as multivariate regression,
fuzzy logic and the artificial neural networks (ANNs) model which could provide
satisfactory results. Adamowski and Karapataki, [2] performed a comparative
investigation between ANN and multivariate regression forecasting of peak urban water
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 110
demand. They found that the ANN model approximations are better for peak weekly
water demand compared with multiple linear regression models. Moreover, ANNs have
the following model attributes:
ANNs can approximate the complex relationship between the input and output
parameters without knowing their physical characteristics
At times noise cannot be avoided in the simulations, even so the ANN can
produce good results
The ANN follows an adaptive approach to solutions over time to compensate
for changing circumstances
Once the model is trained it is easy to use
ANNs have already been used successfully in similar kinds of environmental
research like water level predictions, flood forecasting and control in combined sewers
as presented by [47]; [31] and [173]. Willems et al. [174] demonstrated a number of key
challenges in the use of an ANN model for a sewerage system. Chiang et al. [47]
demonstrated that ANNs have the capability to effectively extract significant features and
trends from complex systems even if the underlying physics is either unknown or difficult
to recognize. Moreover, ANNs greatly reduce the computational time and cost, unless
completely new sets of experimental conditions are used [135].
In the current research there are sixteen input parameters and one output
parameter. Moreover, a few input parameters can change their physio-chemical
behaviour once mixed with sewage output. Considering all these complexities; the ANN
model was the obvious choice in analysing the current Comb Separator sewage overflow
screening device. In this research of sewage solid capture efficiency, neural network
modelling is capable to predict any nonlinear input-output relationship. A multi-layer feed
forward artificial neural network, using a back propagation algorithm was used. Such
networks have been widely used for several environmental problems and modelling
[109]. A series of 47 sets of experimental data were collected to develop the ANN model.
A separate validation set of 8 experimental data were collected to validate the trained
ANN model.
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 111
6.2 Artificial Neural Network
An ANN was designed to simulate the functionality of a human brain where
millions of neurons are connected to each other. The reasoning used in ANN is more like
human brain which than learns the attitude and stores the information based on a
representative dataset. The power of ANN is the ability to process that has a massively
parallel distributed information processing system. This activity is same as human brain
has millions of neurons and trillions of interconnection between neurons so the ANN
resembling biological neural networks of the human brain [79]. It is inspired by the
structural, functional and computational aspect of a biological neural network as shown
in Figure 6.1.
Figure 6.1: Conceptual diagram showing an analogy of the work principal between the
human brain and the ANN model (Source: [149])
The development of ANN model is based on the following rules:
The information processing units are known as neurons also known as nodes,
units or cells
Neurons are interconnected using signals known as connection links
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 112
Connection strength is represented with associated weight between two neurons
All neurons are applied with an activation function to determine its output signal.
The inter-connection between the two neurons is represented by the term weight [122].
A basic concept in evaluating the parameter relationship can be derived by these weights
as shown in 6.2.
Figure 6.2: Conceptual diagram of input-output and weight adjustment (Source: [40])
Proper optimisation of the weight matrix is the key aspect to forming relationships
between neurons. One of the important aspects of optimization involves modification of
the synaptic weights to generate minimal error between actual outputs and predicted
from the model. This modification process is known as a learning algorithm. The most
potent algorithm, used in this work and which is well accepted, is the back propagation
neural network [56]. Werbos [172] developed the back propagation algorithm in his PhD
dissertation at Harvard University. However, work from [137] made the algorithm popular
as he demonstrating how to train the hidden neuron for a complex mapping problem.
The parallel computational process and the ability of the network to learn and generalize
a process, makes ANNs highly popular in solving problems that are currently intractable.
In addition, an ANN provides the following benefits and capabilities [79].
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 113
Nonlinearity: An ANN is a non-linear computational tool. This feature is useful in
modelling non-linear problems that are challenging to model by existing mathematical
techniques.
Input-output mapping: From a given dataset containing input and output samples;
the network demonstrates the ability to learn examples to do input-output mapping. This
is achieved by weight optimization using a learning paradigm during the network training.
Adaptation process: An ANN can adopt its weights with changes in the
surrounding environment. With small changes in the operating condition the ANN can be
easily retrained. This feature is particularly useful to change weights in real time for
implementation of an on line control system using the ANN.
Evidential response: An ANN can be designed to generate information on both
the choice and reason behind the selection of a particular pattern for a pattern
classification problem. This helps in improving the classification performance through the
rejection of ambiguous patterns.
Contextual information: The parallel distributed structure allows every neuron in
the network to hold some knowledge of the problem and be influenced by the global
activity of the other neurons in the model.
Fault tolerance: An ANN is integrally fault tolerant due to the presence of massive
parallelism. In case of a fault, instead of a catastrophic failure, the network’s performance
reduces [28]. Two other key functions of ANNs are categories of recognition and function
approximation [138]. One of the previous seen inputs are trained in the recognition
category. For the function approximation, network need to approximate unseen inputs
from training to model complex input output relationships.
Neural networks have a wide variety of applications and have been implemented
in many practical applications, it is more powerful when the physical law based
mathematical techniques are difficult to implement or it become too complex to develop
input and output relationship between parameters. ANNs have been successfully used
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 114
for solving a wide range of practical problems in hydrology and water quality such as
forecasting the rainfall [45]; [46], operation of a reservoir [37]; [43], forecasting flow in the
stream [6]; [44], water quality forecasting in rivers from predicted pesticide leaching
through turf grass-covered soil [4]. Starrett et al.[155]; Sandhu and Finch [140]
investigated the flow condition and gate positions in the Sacramento San Joaquin Delta
salinity levels between the interior and along the boundary of the delta using ANN.
The current research concentrates on a function approximation network, where
the term, generalization, indicates the capability of the network to acquire the complex
input output relationships and predict capture efficiency of sewage solids. Generalization
is the key strength of the ANN that makes it desirable from other methods of
approximation by having the network trained to be responsive to an unseen environment.
However, it is important to make sure the trained ANN model is protected from over
fitting, which is a noticeable problem which could led to poor generalization for a function
approximation. In these cases, the output generalization fails to respond well with an
unseen dataset. Instead of learning the data the network actually memorizes the
samples which led to cause this issue.
One common aspect of over fitting is when we use few dataset for training in
comparison to the total number of network parameters. In the current research we have
used 60% of the dataset for training. On the flip side if the dataset is too big it creates
more complex functions. Thus, it is important to use adequate data to improve the
generalization ability of the model; the dataset should not be too small not too big [40].
An example of a typical over fitting problem is shown in Figure 6.3. Here the blue boxes
represent the data of a noisy sine function, whereas the red dashed line represents the
response of a trained ANN. The result indicates that the network has over-fitted the data
and, thus, the network would not generalize well in an unknown environment. The
network actually memorizes each data point instead of trying to map the input-output
relationship. The trained network, without over-fitting, should be able to ignore the noise
and learn the underlying function, which is the sine function for this case. The black line
represents the ideal output of such a type of network without over-fitting.
Several authors [18]; [112]; [177] have tested how to improve the generalization
ability of a trained ANN by injecting noise into the inputs [38]. Karystinos and Pados [99]
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 115
tried by expanding training samples which were generated randomly in agreement with
probability distribution function (PDF) generated by the Parzen-Rosenblatt estimate [98].
The intent of this procedure was to overcome the problem of over-fitting and to improve
the generalization ability of the ANN. The generalization performance of the trained
network was evaluated from the error generated by the network on the data that are
training dataset which is known as the generalization error. Cross validation [101]; [104]
and early stopping [134]; [175] are other statistical techniques to overcome the problem
of over fitting; which helps to reduce the generalization error that led to better
performance of the ANN. The database is divided, into training, testing and validation
datasets. In the current problem we have used 60% of the data for training, 20% for
testing (which is similar to calibration of hydrologic model) and 20% for validation (also
known as cross training which is similar to model validation in hydrologic models).
Figure 6.3: Demonstration of over fitting for a function approximating Artificial
Neural Network (Source: [40]).
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 116
During the training process the connection weights of an ANN are consistently
and continuously stimulate by the environment where the network is embedded. The
training goal of the model was to minimize the error function by continuously update the
connection strengths and threshold values to make the output close to the targets [5].
The training set all the time trying to minimise the error gradient and update the network’s
weights and biases. Cross training (validation) dataset is recommended to prevent
overfitting. The network’s error on the validation set is calculated and monitored during
the training process. This set is not used to update the network’s weights and biases.
The validation error generally decreases when the network’s training starts. A rise in the
validation error for a certain number of iterations (also described as epochs) indicates
over fitting of the network. When the network realise the overfitting happening it led to
stop the training and the weights and bias values are saved at the minimum validation
error. It is important to test with a separate test set to check the performance of the
trained network through calculation of the generation error. Few different independent
data splits are performed, followed by extensive training to get good statistical results.
6.3 Description of Network Structure
6.3.1 Artificial Neuron Model
An artificial neuron is the fundamental non linear information processing unit.
Three key components of a neuron are given below:
Weights: Weights represent the strength or value assigned to each of the
connecting links or synapses. An input signal px to the neuron k is multiplied by the
synaptic weight kpw . The synaptic weight kpw defines the strength of the connection
between the input px and the neuronk .
Adder: A linear adder was used to gather the weighted input signal.
Activation function: The summed output of the weighted input signal kv is limited
to some finite value within a permissible amplitude range of the output signal set for the
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 117
model. This is done by the activation function . The function defines the neuron
output ky relative to the activity level at the function’s input kv .
Figure 6.4: A Non-linear model of an artificial neuron k (Source: [40])
6.3.2 Multi-Layer Feed Forward Neural Network Structure
The structure is made up of three main sections: the input layer, the hidden layers
and the output layer. Current research consider multilayer feed forward neural network
structure as it is most common network structure to use in environmental modeling, a
detail description of the model can be found in the work of Churchland & Sejnowski [48].
6.4 Network Learning
How the ANN optimises the weights within the network parameters is referred to
as a paradigm. The neural network operates using a learning paradigm which refers to
a model of the environment [162]. There are three classes of learning paradigms:
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 118
supervised learning, reinforcement learning and self-organized or unsupervised learning.
Further details about these learning paradigms can be found in the work of [48]; [26].
6.4.1 Back propagation algorithm
Back propagation algorithm is the most commonly used neural network algorithm
used in the environmental model. In the current research a model considering multi-layer
perceptron, based on back propagation algorithm was used [9]. A detail description of
the algorithm can be found in the work of [56]; [40]; [5] & [7].
6.4.2 Levenberg-Marquardt Algorithm
The Levenberg-Marquardt (LM) algorithm use Newton’s method to approximate
and is selected to reach the second order training speed without calculating the Hessian
matrix. The Hessian matrix (H) approximation and error gradient (g) is calculated as per
Equation 6.1 and Equation 6.2.
TH J J (6.1)
Tg J e (6.2)
J denotes the Jacobian matrix designed with the first derivatives of the network
errors, e, on the training set so that the network’s weights and biases and can be
considered using the typical back propagation method [76]. JT is the rearrange of the
Jacobian matrix, J.
The LM algorithm uses the estimated calculation of the Hessian matrix to adjust
and alter the parameters. If zk signifies the old parameter value; whereas the new
parameter value after calculation of the network errors is given by zk+1 (Equation 6.3)
1
1T T
k kz z J J I J e
(6.3)
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 119
The parameter µ is set to a precise value at the beginning of the training. After
each epoch, the performance function is calculated. When the performance function is
less than the previous epoch, the value of µ is decreased by an exact value. On the other
hand, if the performance function rises, the value of µ is also increased by an exact value.
The value of µ equal to zero turns Equation (6.3) into Newton’s method. The aim is to
return to Newton’s method quickly since it is faster and more accurate near when the
error is less.
A highest value of µ is set beforehand the training. If µ touches its highest value,
the training ends. This activity suggests the network has failed to get a converging
solution. The training is also ends when the error gradient (Equation 6.2) drops below a
precise set value or when the goal set for the performance function is met.
The network training steps using the LM algorithm are as follows:
1. All the inputs to the network are presented. The corresponding network
outputs, errors and the sum of square errors over all inputs are computed.
2. The Jacobian matrix, J, is computed.
3. Equation 6.3 is computed to obtain the new parameter values.
4. The sum of squares of errors is recomputed with the updated parameter
values.
5. If the new sum of squares is smaller than the previous value, μ is reduced by
a specific factor β and the process is re-started from step 1.
6. If the new sum of squares is increased, the value of μ is increased by α and
the process is re-started from step 3.
The network is assumed to have converged when the error gradient is less than
some predetermined value or when the sum of squares has been reduced to some
specific error goal. Another popular algorithm used in the environmental model is known
as Resilient back propagation algorithm.
6.4.3 Resilient Back Propagation Algorithm
The Resilient Back Propagation (RBP) algorithm eliminates the detrimental
effects of the levels of the partial derivatives. The sign of the derivative governs the
direction of weight update. The size of the weight change is determined by a discrete
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 120
parameter update. The update parameter value for each weight and bias is enlarged by
a specific factor. This increase occurs if the derivative of the performance function, with
respect to that of weight, has the identical sign for two consecutive iterations. On the
other hand the update value is reduced by a factor if the derivative, with respect to that
of weight, changes sign from the previous iteration. When the derivative is zero, the
update value remains the same. As the weights fluctuate, the change in the weight is
reduced. When the weight continues to change in the same direction for little iteration,
the levels of the weight change rises. A detailed study of the algorithm is provided in
[136].
6.5 Data Collection and Pre-processing
6.5.1 Creation of Database
The database for ANN analysis was created from the hydraulic experimental data
of the Comb Separator, a sewage solids overflow screening device. In total there were
sixteen input parameters which had an impact on the output sewage solids capture
efficiency. The input parameters were considered with all the logically possible input
conditions or materials that could have an impact on the output capture efficiency. Out
of these sixteen parameters there were nine parameters related to the experimental
setup and seven parameters related to the experimental materials used. The input
parameters considered in the device setup were:
1. Number of comb layers
2. Spacing of the first comb layer
3. Spacing of the second comb layer
4. Centre to centre spacing of combs
5. Position of comb from crest of weir
6. Length of the crest over the weir
7. Height of water over the weir head
8. Flow (measured rate per meter length of the weir)
9. Position of the retaining screen
The input parameters considered in the experimental sewage solids testing were:
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 121
1. Cigarette butts (actual)
2. Cigarette butts (artificial used for experimental purpose)
3. Cotton buds
4. Condoms
5. Tampons
6. Wrap papers
7. Toilet paper
All this experimental data was collected and normalised using feature scaling
method also known as unity based normalization that bring all values into the range [0,1]
before they were ready for analysis using the ANN method. MATLAB [111] software
package was used for data analysis and reporting.
6.5.2 Network Architecture
6.5.2.1 Selection of Input Output Parameters
To ensure a good model approximation the number of input-output data pairs
used for training should be equal to or greater than the number of parameters (weights)
in the network [35]. Like most environmental modelling approaches, some of the key
steps in ANN modelling include data collection (from experimental results), pre-
processing of the data (normalizing experimental data) and assessment of the output. A
robust and sufficiently large database is essential to construct a network which
generalizes well. Moreover, a clear understanding of the hydraulic process is required
for successful modelling of this nature. For instance, physical insight into the problem
being studied can lead to better choices of input variables for proper mapping [7]. This
will lead to effective and efficient modelling, avoiding loss of information due to
inappropriate choice of input parameters. In the present study 40 sets of experimental
data were collected.
Eight different types of experimental conditions were tested by changing the
volume of flow, the number and spacing of combs, weir opening and different sewage
solids materials to improve the capture efficiency of sewage solids. Based on these
experimental results an input-output relationship was assumed, which showed that 16
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 122
input parameters influence the pollutants capture efficiency [8] & [9]. While conducting
the experiment, one of the experimental conditions (inflow volume, number of combs,
spacing of combs etc.) was varied, while the others were kept constant to their reference
values. This allows examining how different experimental conditions could affect the
output pollutant capture efficiency, referring to Figure 6.5.
Figure 6.5: Block diagram of proposed ANN model.
6.5.2.2 Parameter Estimation and Network Optimisation
Optimal network architecture should retain a simple and compact structure while
providing best performance in terms of error minimization. The network should be neither
too small; as it will have insufficient degrees of freedom to capture the underlying
relationship in the data, nor too large; as it may fail to generalise, memorising fluctuations
in the training data that are not representative of the system being modelled. A model
considering multi-layer perceptron (MLP), based on the back propagation algorithm, is
used in this work. The multi-layer ANN architecture comprises three main parts: the input
layer, the output layer and the layer in-between termed the hidden layer. The number of
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 123
neurons required to describe each parameter is dependent on the nature of the
parameter. A real valued parameter requires one neuron to represent the value, while x
neurons are required to describe 2x categories for parameters representing
classifications. The flexibility lies in selecting the number of hidden layers and in
assigning the number of neurons to each of these layers. Maren [110] suggested two
hidden layers when the outputs need to be continuous functions of the inputs. Two
hidden layers were thus used in this study and the number of neurons in each layer was
ascertained from the network training and optimization process.
In the back propagation algorithm, all the weights within a network are adjusted
simultaneously. Every neuron learns a feature, defined by the back propagated error
signal, through weight changes. The weight changes in each neuron are independent
from one another. The parameters are constantly redefined until a set error minimum is
reached [42]. The studied network optimization uses the experimental database to fix
the number of neurons in the hidden layers as well as optimising the weight population
to produce a minimum output error (Table 6.1). The training process for ANNs can be
considered to be similar to the idea of calibration which is an integral part of most
hydrologic modelling studies. The purpose of training is to determine the set of
connection weights and nodal thresholds using the back propagation algorithm that
cause the ANN to predict outputs that are sufficiently close to target values [7]. To avoid
the problem of over-fitting and to improve the generalization ability of the trained network,
the method of cross-validation and early stopping were implemented. The available
dataset was divided into three parts: the training set, the validation set and the test set.
For the current study, we used 60% of the data for the training set and 20% for each of
the validation and test sets. A detailed description of the data split technique is given by
Sarle [141].
6.6 Result Analysis and Discussion
In the studied case 40 weight trials were tested for all the nine algorithms. Error
performances for these algorithms are listed in Table 6.1. Based on the error
performance and regression coefficient (R) value the Levenberg Marquardt and the
Resilient Back Propagation algorithms were chosen as suitable for this study. A trial and
error approach had to be taken as the error back propagation required that the number
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 124
of hidden layers needed to be specified prior to network training [42]. The number of
neurons in the first and second hidden layer was varied from a combination of 5/4
neurons to 23/22 neurons to identify the optimum hidden layer for the problem, refer to
Figure 6.6. Standard regression analysis was performed between the predicted and
experimental pollutant capture efficiency values. From the performed regression
analysis, it was found that the ANN structure with 5/4 neurons in the first and second
hidden layer respectively.
Figure 6.6: Comparison of different node in the 1st and 2nd hidden layer
Nine different standard algorithms were tested to ensure that the Levenberg
Marquardt algorithm and the Resilient Back Propagation algorithm were the optimal
training algorithm for the problem of modelling sewage solid capture efficiency. Forty (40)
different network weight trials were given in each step, where different random initial
weights were used in each trial and the best values for the regression co-efficient (R)
were collected. The performance of different algorithms was judged with respect to the
root mean square error using the following equation.
R2= Sum of Squared Errors / Total Sum of Squares (6.4)
The proposed model responded well to both the Levenberg Marquardt and the
Resilient Back Propagation algorithm with regression values close to unity, refer to Table
0
0.2
0.4
0.6
0.8
1
5.4. 8.7. 11.10. 14.13. 17.16. 20.19. 23.22.
Reg
ress
ion
Valu
e 'R
'
Combination of Neuron in 1st and 2nd Hidden Layer
Levenberg-Marquardt Paradigm Resilient Backpropagation Paradigm
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 125
6.1. The choosing of the paradigms are not limited to the regression value as minimum
and maximum error terms also consider in considering the best paradigms as shown
below, Table 6.1. It is important to note Levenberg-Marquadt and Resilient
Backpropagation algorithm showing less variation between minimum and maximum
error.
Table 6.1 Comparison of different training paradigms
Order Name of Paradigms Tested
Minimum Error
Maximum Error
Regression Value 'R'
A Levenberg-Marquardt 0.08 0.21 0.87 B BFGS Quasi Newton 0.06 0.76 0.61 C Resilient Backpropagation 0.08 0.21 0.87 D Scaled Conjugate 0.11 0.71 0.66 E Conjugate Gradient 0.10 0.40 0.70 F Fletcher-Powell 0.10 0.40 0.70 G Polak-Ribiere 0.15 0.16 0.72 H One Step 0.07 0.65 0.72 I Variable Learning 0.08 0.52 0.63
The regression analysis was performed on the training, validation and test sets
for the ANN with five and four neurons in the first and second hidden layer respectively;
and then trained with the Resilient Back Propagation algorithm. Both the Levenberg-
Marquardt and the Resilient Back Propagation algorithms revealed a promising
regression co-efficient (R) value of 0.862. Comparison of the trained network and
experimental capture efficiency showed good agreement between model results and
experimental data refer to Figure 6.7.
Figure 6.7: Comparison of experimental and ANN predicted capture efficiency.
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 126
Figure 6.8: Regression Value for Training, Validation and Test results
The regression analysis performed on the training, validation and test sets for
the ANN with five and four neurons in the first and second hidden layer respectively
and trained with Resilient Back propagation algorithm is shown in Figure 6.8
6.7 ANN Model Validations
A separate set of validation data was used to judge the overall performance of
the trained network. In the studied problem eight new sets of experimental data were
collected and validated against the trained network to obtain the predicted values. The
generalization performance of a trained network is measured on the error it produces
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 127
with the new dataset. The smaller the error, the better the generalization ability of the
trained network is. It was found that the trained ANN successfully predicted the new
experimental results with an average absolute percentage error of about 7%, which
reveals that the network was trained properly with good generalization ability, refer to
Figure 6.9.
Figure 6.9: Comparison of experimental and model results for validation dataset.
Experiment numbers 1 to 4 were conducted on a specific experimental setup and then
the flow was changed for experiments 5 to 8. Solids used for different test numbers: 1
and 5 = cigarette butts, 2 and 6 = condoms, 3 and 7 = tampons, and 4 and 8 = wipe
papers.
6.8 Summary
Limited understanding of physics of non-Newtonian fluids made it quite
complicated and time consuming to model the physio-chemical interaction of different
sewage particles with water. In order to assess the trapping efficiency of the developed
device under different experimental conditions, a neural network modelling approach
was proposed. The empirical knowledge gained through a series of experiments helped
the formulation of various assumptions which led to a successful ANN model. ANNs
routinely simulate the non-linearity of the physical process without solving complex
partial differential equations. Unlike any other form of mathematical or regression based
Chapter 6: Developing an Artificial Neural Network Model
Md Abdul Aziz 128
modelling there is no need to make assumptions about the mathematical form or the
relationship between input and output parameters.
The ANN’s powerful modelling approach, when trained with input-output data (a
separate set of test data), shows that the model can mimic the underlying complex
physio-hydrodynamic processes (involving different types of materials) that otherwise
would be extremely difficult to model. Such a model also overcomes some common
experimental drawbacks such as different scales and structures and provides a
significant reduction in the time and cost involved in the experimental processes [135].
All of these attributes, along with the nonlinear nature of the activation function,
truly enhance the generalization capabilities of ANNs in the studied problem. Special
attention was given to the generalization of errors during test cases with different
algorithms which significantly contributed to the ANNs performance in predicting
experimental pollutant capture efficiency as shown in Figure 6.9. Separate validation
datasets were used to judge the robustness of the trained network. Promising results
were observed with more than 90% accuracy in eight different experimental results with
an average absolute error of about 7%. This demonstrates the ability of the model to
predict sewage solid capture efficiencies of the device in real-world conditions.
Work presented in this chapter has been published in the following journal and
conference papers:
Aziz, M. A., Imteaz, M., Choudhury, T. A., & Phillips, D., 2013a, ‘Applicability
of artificial neural network in hydraulic experiments using a new sewer overflow
screening device’, Australian Journal of Water Resources, vol.17, no.1, pp.77-86.
Aziz, M. A., Imteaz, M., Choudhury, T. A. & Phillips, D. I. 2011, ‘Artificial Neural
Networks for the prediction of the trapping efficiency of a new sewer overflow screening
device’, 19th International Congress on Modelling and Simulation, Perth, Australia.
Indexed in Scopus
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 129
Chapter 7
Sensitivity Analysis
of the Comb
Separator
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 130
7.1 Introduction
Analysing the parameter sensitivity of a hydraulic device such as a Comb
Separator has been a standard practice by hydraulic engineers for many years [97];
[116]; [127]; [176]. Such analysis qualitatively or quantitatively explains different sources
of variation [139]. Extended analysis from basic sensitivity analysis can be found in the
works of Hall [70] and Hall and Solomatine [71]. A comprehensive review regarding
application of sensitivity analysis in environmental models is presented by Hamby [73].
Sensitivity analysis of the input parameters, of the Comb Separator device provides a
better understanding of those input parameters, This understanding includes their
influence on the outcome capture efficiency, identification of which parameter is the most
important, the relative importance of each input parameter and identification of which
parameter requires further research.
Our proposed Comb Separator, a sewer overflow screening device, consists of a
rectangular tank and a sharp crested weir. In front of the weir are a series of vertical,
parallel combs to separate entrained sewer solids from the overflow. The studied device
was tested with a series of sewer solid materials including condoms, tampons, cigarette
butts, cotton buds, bottle caps, wrap papers, etc.
A detailed discussion of the experimental work is provided in Chapter 5. Larger
sewer particles (greater than 10mm in diameter) can be captured relatively easily with a
capture efficiency of more than 90%. This capture efficiency was tested in different input
conditions, such as flow, effective spacing, weir opening, comb layers and run time. The
output of this testing is the sewer overflow capture efficiency.
Comparison with the industry standard Hydro JetTM screen shows the capture
efficiency of ‘Comb separator’ performs better on low flow. However there was a variation
of output capture efficiency. Hence the focus of this chapter is to understand parameter
sensitivity on the capture of smaller sewer solid particles.
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 131
7.1.1 Objective
The key objectives of the sensitivity analyses of the Comb Separator device are
listed as follows:
To develop a robust understanding of the meaningful input parameters
To undertake a performance comparison of the proposed Comb Separator with
a standard Hydro JetTM screen under low flow (up to 60 l/s) conditions
To comprehend the impact of experimental design parameters (runtime, flow
discharge, effective comb spacing, weir opening and comb layers) on sewer
solids capture efficiency
To understand the relative significance of the input parameters and to identify
which parameter is the most influential in development of output results.
7.2 Background
A total of 42 sets of experimental data were collected on six different sets of
experimental setups for Comb Separator. Based on the experimental experience, five
input parameters (runtime, flow volume, effective comb spacing, weir opening and layers
of combs) were identified as being influential on output sewer capture efficiency. A
sensitivity analysis of such a device is necessary to ensure optimisation and validation
that also serve as a guide to future improvement opportunities for other proposed
experimental devices. The sensitivity analyses of the experimental data can help to
understand the following:
Which input parameter can be neglected and removed from the model
Which input parameter requires additional research for strengthening knowledge
and understanding to reduce output uncertainty
Which input parameter contributes the most to the output variability
Parameters correlation with output capture efficiency
Once the device is in practical use, what would be the best approach to manage
effectively and efficiently the device performance
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 132
The experimental work was restricted by the physical limitations of conducting
experiments with limited trial combinations of the experimental conditions. This is
inherent in any laboratory study. In addition, the device performance was tested against
industry standard Hydro JetTM screen against industry standard device also identified the
importance of input parameters sensitivity testing. Both ‘Comb Separator’ and Hydro
JetTM were developed to improve screening efficiency of sewer overflow screening
device. The Hydro JetTM screen had subjected to a series of development, testing and
evaluation program [15], [17]. The screen used in Hydro JetTM was a static screening with
specific focus to capture sewer solids 6mm in dimensions. Performance of Hydro JetTM
screening focussed around screening capacity and effectiveness. As the proposed
‘Comb Separator’ device used comb spacing which had minimal impact on screening
capacity during most of the experimental runs with different sewer solids.
The Comb Separator can produce better screening efficiency in capturing sewer
solids with low flows. There was hardly any blinding effect on the Comb Separator which
is a key improvement from the previous static screening concept. However, the capture
efficiency for Comb Separator varies more than the Hydro JetTM screen. It is important
to understand the performance of the input parameters influencing the capture efficiency.
To investigate this issue it was important to perform parameter sensitivity testing.
Sensitivity of these parameters is of paramount importance in considering the ability of
this device to function property in remote, unstaffed locations. Understanding and
analysing model sensitivity and uncertainty has been an active theme of research for
hydraulic engineers for many years [72]. Sensitivity analysis is predominantly used in
design for hydraulic experimental parameters.
In the current investigation a model was developed using Multiple Linear
Regression (MLR) method. As the dataset is small for detail sensitivity analysis, the Latin
Hypercube Sampling (LHS) technique was used to expand the data series without
compromising the input-out relationship. Methodology used for this purpose is
schematised in Figure 7.1.
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 133
7.2 Methodology used in Sensitivity Analysis
The methodology adopted in the sensitivity analysis is shown in the flow chart below:
Figure 7.1: Flow chart of the methodology adopted in the sensitivity analysis.
Developing regression
Model for input output
Relationship Yes
No
No
Improving Screening Efficiency
Comparison between Comb
Separator and Hydro Jet TM
Selection of input Parameters
Latin H sampling for data generation
Check for assumptions
Remarks on input parameters sensitivity
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 134
Some of the key points in developing the methodology are given below:
Define the research problem regarding the sewer overflow screening device and
why we need to do the sensitivity analysis using a comparative analysis between
the Hydro JetTM and Comb Separator
Develop meaningful and simplified inputs to the model considering the key input
parameters influence on the output capture efficiency
Develop a regression model and check for the necessary assumptions
Use sampling techniques to develop large input data sets. In this case the Latin
Hypercube sampling (LHS) technique which is highly recommended by scientific
literature for parameter sampling will be used. LHS was used to generate 10,000
units of data of the input parameter considering their distribution type taken from
experimental results.
7.2.1 Developing a Multiple Linear Regression (MLR) Model
Multiple linear regressions (MLR) are a statistical method which uses some
explanatory (independent) variables to predict the outcome of a response (dependent)
variable. So MLR is to model the relationship between independent (input or predictor
variables) and dependent variables.
The model for MLR, given total ‘n’ observations, is:
Yi = (b0+b1X1i +b2X2i+ ………….. +bnXni) +ᶓ (7.1)
Y is the dependent variable, b1 is the coefficient of the first input X1, b2 is the
coefficient of the second input X2, bn is the coefficient of the nth input (Xn), and ᶓ is the
difference between the input and the observed value of Y for the ith participant. In our
studied case the model become
(CaptureEfficiency)i=b0+b1(runtime)i+b2(flow)i+b3(effectivespacing)i+b4(weiropening)i+b5
(comb layers)i + ᶓ (7.2)
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 135
This model is based on equation (7.2) and includes a b-value for both inputs and
the constant. If we calculate the b-values, we could make predictions about capture
efficiency based on all five input (predictor) variables. It is important to efficiently design
the model input (predictors) parameters as the values of the regression coefficient
depend upon the variables in the model. Predictor (or input) variables should be selected
based on past research to make sure all the key inputs variables are included in the
model generation.
7.2.1.1 Validity of Model Assumptions
There are a few assumptions that need to be satisfied to use MLR model [25]. They are:
Variable Type: All input parameters must be quantitative or categorical and the
output parameter (outcome variable) must also be quantitative and continuous. For
example, in the studied case the three input parameters, runtime (minutes), flow (l/s/m)
and effective spacing (mm) are all quantitative predictor variables. The outcome variable
is the capture efficiency (%) which is also a quantitative variable. Hence the model
satisfies this assumption.
Non-zero variance: The input parameters should vary in value and not have
variances of zero. The experimental data suggest that all the input parameters are non-
zero values so the predictor variables satisfy this criterion.
No perfect multicollinearity: Among the three predictor variables, they should not
have perfect linear relationship two or more of the predictors. For the current study the
predictor variables did not show strong correlation as shown in table 7.3.
Predictors are uncorrelated with ‘external variables’: External variables such as
weir opening and comb layers that were not included in the regression model had little
influence on other predictors or the outcome variable. This assumption means that weir
opening and comb layers should not correlate with runtime, flow and effective spacing
predictors or influence capture efficiency. Both weir opening and comb layers did not
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 136
influence the other predictor variables or outcome variable, hence this assumption was
satisfied.
Homoscedasticity: At each level of the input variables the variable of the residual
terms must to be constant. At every level the residual predictors must have the same
variance (homoscedasticity). However, if these variances are not unequal the data said
to be heteroscedasticity. During the model run two parameters were plotted. One was
ZRESID or residual along Z and the other was ZPRED which is prediction along Z. The
ZRESID and ZPRED should look like dots points are randomly dispersed around zero
as shown in the graph below in figure 7.2:
Figure 7.2 Regression Standardized predicted value against residual
When there is some sort of curve in this graph that indicates the data has broken
the assumption of linearity. On the other hand when graph funnels out there are
possibilities that the data have heteroscedasticity. The points from experimental data are
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 137
randomly and located evenly distributed across the plot. The pattern satisfies the
assumptions of linearity and homoscedasticity have been met.
Independent Errors: The residual terms should not be correlated between any two
observations which suggest a lack of autocorrelation. This assumption is tested using
the Dubin-Watson test [53 & 54]: a very conservative rule of thumb, values less than 1
or greater than 3 are definitely cause for concern. In the studied case the value was
observed at less than 2 which show a positive correlation but not a cause for concern.
Normally Distributed Errors: The residuals used in the model are random and
normally distributed with a mean of zero. This assumption assumes that most of the time
the observed data are zero or very close to zero with only occasional differences much
greater than zero. The histogram and normal probability plot as shown in figure 7.3 was
assessed to examine this assumption.
Figure 7.3: Normal distribution plot of the Standardized Residual vs Frequency
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 138
The bell shape curve on the histogram shows the shape of the distribution. The
curve follows a satisfactory bell shape which suggests the experimental data for the
Comb Separator satisfies the normality assumption in the data set.
Figure 7.4: Experimental data shows Observed vs Expected Cumulative probability
The line in the plot in figure 7.4 represents a normal distribution where the dot
points are observed residuals. In a perfect normally distributed data set, all points will lie
on the line. The data sets are reasonably close to the straight line so the normal
distribution assumption is reasonably satisfied with this data set.
7.2.1.2 Test of Sample Size and Input Parameters
There are a number of different rules of thumb available about how big a data set
should be to obtain a reasonable representation between input (predictor) and output
(outcome) variables. The simplest rule of thumb would be the bigger the data set, the
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 139
better the model. A sampling data set of 10,000 was chosen on this basis. However, for
any practical experiment there are limitations of resources and time. This is a common
restriction embedded in most experimental research. In the current research comb
spacing was varied in the comb layers and up to six different set ups were possible to
test.
The reason is that the regression coefficient ‘R’ is linked with a number of input
parameters ‘k’ and the number of datasets ‘N’ as in following relationship. In the MLR it
was used five input (predictor) variables than expected ‘R’ (regression coefficient) for
random data is k/(N-1) for example with five predictors with 21 sets of data can appear
to have a strong effect, R =5/(21-1) = 0.25. Obviously ‘R’ is expected to be close to zero
so that random data can have no effect. The two most common are 10 cases of data for
each predictor in the model, or 15 cases of data per predictor. This reduces the value of
randomness within 0.1 to 0 which was a reasonable reduction of the effect of
randomness on the dataset. In the studied case we tried two different models. In the first
model we used 5 input parameters with 42 sets of data. However, the data set showed
a randomness of more than 0.1 and with the elimination of two input parameters it had
only a minor variation to the model output. In the second case, 3 input parameters with
42 sets of data were used with the results showing the data randomness effect within 0
to 0.1. This was reasonable under the model assumption. Hence the data set used in
the model was sufficient to produce a decent result using the MLR model.
7.2.2 Summary of the Model
The first table shows the mean and standard deviation of each variable in our
data set, so we understand that average capture efficiency of sewer solids was
72.95~73%. This table was not necessary to interpret the regression model, but it was a
useful summary of the data. The table 7.2 shows three things, firstly, the value of
Pearson’s correlation coefficient between every pair of variables. For example, capture
efficiency had positive correlations with runtime and effective spacing but has a negative
correlation with flow. This also explains that effective spacing has the most significant
positive correlation with capture efficiency, r =0.538. Secondly, the one-tailed
significance of each correlation was displayed. For example, both effective spacing and
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 140
flow are significant (as p<0.001). Thirdly, the number of experimental data used in the
analysis was 42.
Table 7.1 Provides descriptive statistics of the Comb Separator experimental data set
Descriptive Statistics
Sensitivity Parameters Mean Standard
Deviation No of data (N)
Unit
Capture Efficiency 72.95 14.49 42 %
Run time 16.67 6.90 42 Minutes
Flow 43.38 15.83 42 Litre/Sec
Effective Spacing 3.13 1.27 42 mm
Table 7.2 Correlations of different parameters
Correlations
Capture
Efficiency (%)
Runtime (minutes)
Flow (Litre/sec)
Effective Spacing(mm)
Pearson Correlation
Capture Efficiency (%) 1.00 0.32 -0.50 0.54
Run time (minutes) 0.32 1.00 -0.77 -0.71
Flow (litre/sec) -0.51 -0.77 1.00 -0.24
Effective Spacing (mm) 0.54 -0.71 -0.24 1.00
Sig. (1-tailed)
Capture Efficiency (%) - 0.02 0.00 0.00
Run time (minutes) 0.02 - 0.31 0.33
Flow (litre/sec) 0.00 0.31 - 0.07
Effective Spacing (mm) 0.00 0.33 0.07 -
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 141
The correlation matrix is extremely useful for getting a rough idea of the
relationships between inputs (predictors) and output (outcome), and for a preliminary
look for multicollinearity. There was no multicollinearity in the data set; hence there was
no substantial correlation (r> 0.9) between input parameters. If we look at the input
parameters or predictors the highest correlation was between predictors flow and
effective spacing (r=-0.235, p>0.005). This correlation was not significant and the value
was also small, so little collinearity exits in the predictors. The predictor effective spacing
correlates best with the outcome (r=0.538, p<0.001); so it is likely this predictor will best
predict capture efficiency.
The model summary section is the key section to discuss about the overall model
performance of the MLR model. Column ‘R’ lists the values of the multiple correlation
coefficients between inputs (predictors) and output (outcome). The model suggests that
three input parameters or predictors account for 54.9% of the variation in capture
efficiency.
Table 7.3 Summary for the multiple linear regression model
Model Summary
Model R R Square
Adjusted R Square
Std Error of the Estimate
Durbin- Watson
1 0.741 0.549 0.513 10.11 1.14
Change Statistics
R Square Change
F Change df1 df2 Sig. F Change
0.549 15.419 3 38 0
a. Predictors (input parameters): (Constant), Effective spacing (mm), Runtime (min), Flow (litre/sec)
b. Dependent variable: Capture Efficiency (%)
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 142
The adjusted R2 gives us understanding of the model performance and preferably
its value is very close to R2. In the studied dataset the difference between R square and
adjusted R square for the final model was small (0.54-0.52 =0.03 or 3%). This shrinkage
suggests that if the model were developed from the population rather than a sample it
would account for approximately 3% less variance in the outcome. The significance of
R2 can be tested using an F-ratio. Fchange was 15.419 which is significant (p<0.001) as
shown in the table 7.3. The Durbin-Watson statistic in the last column verifies the
assumption of independent errors. As a conservative rule Durbin-Watson value must be
within 1 to 3 and close to 2 is better. The model results satisfy this criterion.
The next table contains an ANOVA, which tests whether the model is significantly
better at predicting the outcome than using the mean as a ‘best guess’. The chosen
model has three predictors and one constant so the model has 38 degrees of freedom.
The F ratio is calculated by dividing the average improvement in prediction by the model
(MSm).
𝐹 =𝑀𝑆𝑚
𝑀𝑆𝑟=
Average Improvement in Prediction by Model
Average Difference between Model and Observed data
Model results greater than 1 suggest that the improvement due to fitting the
regression model is much greater than the inaccuracy. For this model F is 15.419, which
is also highly significant (p<0.001).
Table 7.4 ANOVA table for the MLR model
ANOVA
Model Sum of Squares df Mean
Square F Sig
Regression 4731.19 3 1577.06 15.419 0
Residual 3886.71 38 102.28
Total 8617.91 41
a. Dependent Variable: Capture Efficiency (%)
b. Predictors: (constant), Effective spacing (mm), Runtime (mins), Flow (Litre/sec)
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 143
The next discussion table concerns Model parameters. The first part of the table
gives us an estimate for which b-values (Equation 7.2) and these values indicate the
individual contribution of each predictor to the model. The other two input parameters,
weir opening and comb layers were not considered in the final model as they provided
an insignificant contribution to the model outcome.
The final regression equation is given below:
Capture Efficiency=59.312+0.69(runtime)-0.339(flow) +5.3(effective spacing) (7.3)
The ‘b’ values show the relationship between the capture efficiency and model
input parameters.
Table 7.5 Coefficient of different parameters
Runtime (b =0.69): This value indicates that in the experiment runtime increases
by one unit and capture efficiency increases by 0.69 of a unit. For example, if the Comb
Separator device operates/runs for one additional minute the sewer overflow capture
efficiency may increases around 0.69%. This explanation is true only if effects of flow
discharge and effective spacing are held constant and the device operates till it captures
all sewer solids.
Flow (b = -0.339): This value indicates that flow discharge has a negative
correlation with sewer overflow capture efficiency, so if one more unit of flow passes the
Comb Separator device it means that there is around 0.339 unit reduction in sewer solid
Standardized Coefficients
B Std. Error Beta(Constant) 59.312 8.250 7.189 0.000
Runtime (min) 0.690 0.230 0.329 2.994 0.005
Flow (Litre/sec) -0.339 0.103 -0.370 -3.288 0.002
Effective
Spacing (mm)5.390 1.280 0.474 4.211 0.000
Lower Bound Upper Bound Zero-order Partial Part Tolerance VIF
(Constant) 42.611 76.014
Runtime (min) 0.223 1.157 0.324 0.437 0.326 0.986 1.015
Flow (Litre/sec) -0.548 -0.13 -0.507 -0.471 -0.358 0.936 1.069
Effective
Spacing (mm) 2.799 7.982 0.538 0.564 0.459 0.937 1.068
95.0% Confidence Interval for B Correlations Collinearity StatisticsModel
ModelUnstandardized
Coefficients t Sig.
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 144
capture efficiency. For example, if flow is increasing by 1 cubic meter the sewer overflow
capture efficiency will reduce by around 0.339%. This phenomenon is true when the
other two parameters of runtime and effective spacing are constant.
Effective Spacing (b = 5.390): Effective spacing of the comb has a positive
correlation with sewer overflow capture efficiency. In other words, if the effective spacing
increases, capture efficiency also increases. For example one unit increase in effective
spacing may increase capture efficiency up to 5.39%. Therefore one millimetre increased
in effective spacing will crease 5.39% sewer solids capture by the screen. This
interpretation is true only when runtime and flow discharge are constant.
Each of these beta values has an associated standard error showing how much
these values would vary across different data series or sample. These standard errors
were used to determine whether or not the b-value differed significantly from zero. A t-
statistic test can be performed, which tests whether a b-value is significantly different
from zero. Rule suggests if in the t-test associated with a b-value is significant (if the
value in the column labelled sig. is less than 0.05) then the predictor is making a
significant contribution to the model.
In this case all three predictors are significant at less than p<0.05. In addition, the
smaller the value of significance (and the larger the value of t), the greater contribution it
has on the predictor variables. For example effective spacing has the highest t value of
4.2 and also lowest significance of 0.000 so effective spacing is the most significant
predictor in the discussed model, followed by flow and runtime.
Multicollinearity: Tolerance and VIF statics are the key guidelines to check for
multi-collinearity. If the largest VIF is greater than 10 then there is cause for concern [29];
[119]. If the average VIF is substantially greater than 1 then the regression may be
biased. In this study the VIF values are all well below 10 and the tolerance statistics all
well below 0.2; therefore there was no collinearity within the experimental data set. The
average VIF is close to 1 and this confirms that collinearity is not a problem for the data
set.
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 145
7.2.3 Development of the Dataset Using Sampling Techniques
One of the key challenges for model sensitivity analysis where there are few
parameters which involves the multi-dimensional space in an equitable and
computationally efficient manner. All model input parameter for this model can be defined
in a way that each input parameter have an approximate probability density function
associated with it. The next step would be to simulate by sampling a single value from
each parameter’s distribution. A sample and sensitivity analysis tools (SaSAT) was used
in the model [85]. SaSAT produces input parameter samples for an external model.
These samples, in conjunction with outputs (responses) created from the outer model
(for example regression model), perform the sensitivity analysis. The data generation
process using SaSAT is shown in the table below:
Table 7.6 Schematic diagram of SaSAT data generation
In the studied case the input parameters distribution checks are given below:
It was important to check the input parameter for the sampling model using LHS
technique.
Runtime: The P-P plots suggest that the model data set is close to the inclined
horizontal line, so it was a reasonable consideration to consider Runtime as having a
linear distribution.
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 146
Figure 7.5: P-P plots for the runtime predictor
Figure 7.6: Data on both sides of normal for runtime
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 147
Flow: The P-P plots suggest that the model data set is lying close to the diagonal
line, so it was a reasonable to assume that the flow predictor has a linear distribution,
refer to Figure 7.7 and 7.8.
Figure 7.7: P-P plots for the flow predictor
Figure 7.8: Data on both sides of normal for flow
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 148
Effective Spacing: The P-P plots suggest that the model data set lying close to
the diagonal line, so it was a reasonable assumption that the effective spacing predictor
has a linear distribution, refer to Figure 7.9 and Figure 7.10.
Figure 7.9 P-P plot of effective spacing
Figure 7.10 Data on both sides of normal for effective spacing
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 149
7.2.3.1 Random Sampling: In random sampling each parameter’s distribution is
used to consider N values randomly. This is usually superior to drawing N values
arbitrarily. This is usually better to univariate approaches to sensitivity analysis; however
this is not the most efficient way to sample the parameter space.
7.2.3.2 Latin Hypercube Sampling (LHS): LHS is more efficient and refined
statistical technique being a type of stratified Monte Carlo sampling [91& 92]. Monte
Carlo analysis is an extension of Latin square sampling as proposed by McKay et al.
[115] and further developed and familiarized [91]; [92]; [93]. For each parameter a
probability density function is defined and stratified into N equal-probable serial intervals.
A solitary value is then selected arbitrarily from every interval and for every parameter.
This process provides an input value from each sampling interval and is used only
once in the analysis. This ensures that the whole parameter space is equally sampled in
an efficient manner. The outcome variables can be derived by running the model N times
with each of the sampled parameter sets. A more insight about the Latin Hypercube
sampling methodology can be found in the work [91]; [93]; [94].
7.3 Results and Discussion
7.3.1 Relative Significance of the Input Parameters
Pearson’s correlation coefficient suggests that effective spacing has a large
positive correlation (r=0.684) with the outcome, capture efficiency. Flow discharge has a
negative correlation (r =-0.548) with capture efficiency and runtime has a positive
correlation, (r =0.476) and all these results are statistically significant with p<0.001, refer
to Table 7.7.
The final regression model equation after simulating the individual dataset for
10,000 times using the LHS method shows:
CaptureEfficiency=59.442+0.702(runtime)-0.341(flow)+5.317(effectivespacing) (7.4)
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 150
Table 7.7: Results using the LHS method for 10,000 data
Correlations
Capture
Efficiency (%)
Runtime (minutes)
Flow (Litre/sec)
Effective
Spacing(mm)
Pearson Correlation
Capture Efficiency (%) 1.00 0.49 -0.53 0.68
Run time (minutes) 0.49 1.00 0.01 0.01
Flow (litre/sec) -0.53 0.01 1.00 0.01
Effective Spacing (mm) 0.68 0.01 0.01 1.00
Sig.
(1-tailed)
Capture Efficiency (%) 0.00 0.00 0.00
Run time (minutes) 0.02 0.12 0.31
Flow (litre/sec) 0.00 0.12 0.18
Effective Spacing (mm) 0.00 0.31 0.18
7.3.2 Selection of the Input Parameters
For this research, SPSS Version 22 [89] was used as a tool for MLR modelling.
In the current research forced entry (known as Enter in SPSS) was used as the method
so that all input parameters or predictors are forced into the model concurrently. Unlike
the hierarchical method, the forced entry method makes no decision regarding how the
variables are entered. Table 7.8 shows the results for two MLR models with five and
three input parameters. In developing the MLR model, initially all input parameters that
could have any influence on the output capture efficiency were considered. Trials were
done in the MLR analysis in such a way as all predictors are entered into the model and
their outputs are examined to see which input parameters or predictors contributed
significantly to the model’s capability to predict capture efficiency. In the initial model all
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 151
the five input parameters being run time (minutes), flow (litre/second), weir opening
(millimetre), effective spacing (millimetre) and layers of combs (number) were
considered.
After few different trials with the input parameters, it was found the ‘weir opening’
and ‘comb layers’ were insignificant input (predictor) parameters as they had little
influence on the output sewage solid capture efficiency. ‘R’ is the value of the multiple
correlation coefficients between the predictors and the outcome. ‘R’ values vary from
0.753 to 0.741, from the first model to the second model, which is an insignificant
difference between the two datasets. For the first model ‘R square’ had a value of 0.567
and in the second model with three parameters used its value was 0.549. Therefore two
input parameters; weir opening and comb layers account for only 1.8% of the prediction
accuracy. So, the final MLR considered three input parameters; runtime (minutes), flow
(litre/second) and effective spacing (millimetre), and excluded comb layers (number) and
weir opening (mm). The ‘adjusted R square’ indicates the performance of the model and
in an ideal condition its value will be very close or same as ‘R square’.
Table 7.8 Comparison between initial and final model results
Model 1: Considering Five Parameters Change Statistics
Predictors R R
Square Adjusted
R R Square Change
F Change df1 df2 Sig.F
Change
Layers of combs,
runtime, flow,
weir opening and
effective spacing
0.753 0.567 0.507 0.567 9.438 5 36 0
Model 2: Considering Five Parameters Change Statistics
Predictors R R
Square Adjusted
R R Square Change
F Change df1 df2 Sig.F
Change
Runtime, flow,
effective spacing
0.741 0.549 0.513 0.549 15.419 3 38 0
In this case the difference of ‘R square’ between the final model and the initial
model was small (0.549 – 0.513 = 0.036). This reduction highlights that if the model were
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 152
derived from the population rather than a sample it would account for approximately 3.6%
less variance in the outcome.
7.3.3 Impact of Effective Spacing on Capture Efficiency
Effective spacing was the most significant predictor (input parameter) to influence
capture efficiency. The value indicates that for one unit increase in ‘effective spacing’ the
‘capture efficiency’ increases by 5.317 units. The effective spacing is measured in
millimetre; whereas capture efficiency is measured in percentage. Therefore one
millimetre increased in effective spacing will crease 5.317% sewer solids capture by the
screen. This relation of effective spacing with capture efficiency is valid from 1mm to
6mm and also when ‘runtime’ and ‘flow’ are constant, refer to Figure 7.11.
Figure 7.11: Relationship between Effective spacing (mm) and Capture Efficiency (%)
7.3.4 Impact of Flow on Capture Efficiency
Flow has a negative correlation to sewage solids capture efficiency. As the flow
increases in the sewerage overflow system it produces a higher flow velocity over the
ogee weir and hence more sewage solids are likely to escape the traps/combs causing
lower rate of capture efficiency. Flow is measured in litre/second, whereas capture
efficiency is measured in percentage. An increase in flow by 1 litre/second causes to
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 153
reduce the capture efficiency by 0.341%. It is to be noted that for this analysis, other two
parameters i.e. ‘runtime’ and ‘effective spacing’ are kept constant during the analysis.
Flowrate is one of the key understandings in the sewer overflow screening device
as the flow increases in the device (with fixed weir openings), the flow velocity also
increases which leads to a higher velocity of the sewage solids. Faster movement of
sewage solids near the comb separator is likely to reduce trapping efficiency, refer to
Figure 7.12.
Figure 7.12: Relationship between the Flowrate (l/s) and Capture Efficiency (%)
7.3.5 Runtime Impact on Capture Efficiency
The ‘runtime’ of sewage overflows varies positively with sewer solid ‘capture efficiency’
i.e. the longer the device runs, the higher the capture efficiency of the ‘comb separator’.
The runtime is measured in minutes, whereas capture efficiency is measured in
percentage. It is found that if the device ‘runtime’ increases one unit, the capture
efficiency increases by 0.702%. For example if the device runs for 16 minutes instead of
15 minutes, during the additional minute 0.702% more sewage solids are likely to be
trapped. This relation of device ‘runtime’ (minutes) is valid only on experimental cases
where sewer solids are present in flow while ‘effective spacing (mm)’ and ‘flowrate
(litre/second)’ are kept constant.
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 154
Figure 7.13: Relationship between Runtime (min) and capture efficiency (%)
7.4 Summary
A series of trials with different runtimes, flow, effective spacing, layers of combs
and weir openings were tested. Sensitivity analyses of these input parameters were
performed to identify the influence of them on sewage overflow capture efficiency. The
sensitivity analysis was aimed at developing a robust understanding of the relationships
between the input (predictors) and output (outcome) variables. The MLR model was
initially considered using five input parameters. After significant trial and error it was
found that two input parameters, the weir opening and the comb layers, could be
excluded because these two parameters only contribute to 1.8% of the output.
MLR model Equation 7.3 and LHS sampling technique Equation 7.4 are almost
identical which ensured that the model retained the underlying input and output
relationship when expanding the dataset from 42 sets to 10,000 sets. Sensitivity analysis
delivered a cleaner understanding of the relative importance (rank) of the input
parameters. It was found that ‘effective spacing (mm)’ is the most influential parameter
followed by ‘flowrate (litre/second)’ and ‘run time (minutes)’. The sampling technique also
provided better understand of the input output relations; for example a 1 unit increase in
‘effective spacing’ could increase output ‘capture efficiency’ by 5.31%.
Chapter 7: Sensitivity Analysis of the Comb Separator
Md Abdul Aziz 155
These sensitivity analysis results will be immensely valuable for developing a
practice manual for the proposed device. Further effort to understand the performance
of the proposed ‘comb separator’ could focus mainly on three parameters: effective
spacing (mm), flow (litre/second) and runtime (minutes). These sensitivity analysis
results inform decisions made by the device operator and manager in managing different
sewerage overflow events. Further experiments are suggested to improve the
understanding of the input parameters on high flows.
Work presented in this chapter has been published in the following conferences and
journal papers:
Aziz, M. A., Imteaz, M., Samsuzzoha, M., & Phillips, D., 2013c, ‘Sensitivity
analysis for a proposed sewer overflow screening device’, 20th International Congress
on Modelling and Simulation, Adelaide, 2 to 5 December, Australia.
Aziz, M. A., Imteaz, M., Rasel, H.M., and Samsuzzoha, M., 2015d, ‘Parameter
Sensitivity Using Sampling Technique for a proposed ‘Comb Separator’, A sewer
overflow screening device’, ASEAN- Australian Engineering Congress on Innovative
Technologies for Sustainable Development and Renewable Energy 11-13 March.
Aziz, M. A., Imteaz, M., Samsuzzoha, M. A., and Rasel, H.M., 2015b,
Sensitivity Analysis on the Pollutant Trapping Efficiencies of a Novel Sewerage Overflow
Screening Device. Revising (March 2016) Journal of Hydro-informatics
Conclusions
Md Abdul Aziz 157
Chapter 8
Conclusions
Conclusions
Md Abdul Aziz 158
8.1 Introduction
During continuous downpours, the urban sewer system is not able to carry all the
excessive water; hence some of this excess water flows into the open creek system,
carrying a lot of sewer solids. These sewer solids are dispersed, suspended or washed
into the rivers. They eventually settle, creating odours and toxic/corrosive atmospheres
in the mud deposits on riverbeds. The solids also create aesthetic problems, either
through their general appearance (increasing dirtiness) or through the actual presence
of specific, objectionable items, such as float debris, sanitary discards/faecal matter,
scum or even parts of car tyres. The current study took a holistic view in order to
understand the limitations of existing sewage overflow screens, and to analyse the
research gaps in this field. Some common limitations in existing screening devices
include high on-going maintenance and operational costs, low capture efficiency of
sewage solids, and the use of sophisticated electrical-mechanical switching systems. To
overcome these issues, the objective of the current research was to develop a new
sewage overflow device with improved capture efficiency, low maintenance, and a self -
cleansing mechanism.
8.2 Research Summary
This research focussed on the innovation of a novel sewer overflow screening
device, a ‘Comb Separator’ that can overcome common limitations such as lower sewer
solids capture efficiency and blockage on the screen. The initial concept was tested using
a CFD model analysis, which included no sophisticated electrical switching system and
designing the device so that it is self-cleansing. This also reduces the operational and
maintenance costs of the proposed screen. To prove the concept of the functionality of
this screen, a CFD model analysis was performed. The CFD model provides much better
insight into the design of the device for laboratory testing, with better understanding
about the design parameters.
A series of laboratory tests was completed as part of this project to achieve the
objectives of high capture efficiency, minimal blockage on the screen and effective
function of the device in remote unstaffed locations. It was found that the proposed
Conclusions
Md Abdul Aziz 159
device was free of most of the common limitations in existing screening systems, such
as blinding (minimal blockage), high maintenance requirements and an electrical-
mechanical switching system [154]. Physical limitations of the experimental work were
overcome by doing an ANN model analysis. ANN has the potential to understand the
complex relationship between input and output parameters without having knowledge of
the physical characteristics of the sewer solids. The ANN model supplements the
limitations of the CFD and laboratory testing. Finally, detailed sensitivity testing of the
hydraulic device was performed: a standard check for any hydraulic device. The
sensitivity analysis also provides a guideline for future application of the device in the
physical environment. The ‘Comb Separator’ showed good application potential for
further research in actual sewer overflow systems.
8.3 Knowledge Contributions
In this section, the contributions made in the context of industry practice and
academic research on sewer overflow screening devices are listed.
8.3.1 CFD application
Physical experimental set-ups involve significant cost and time. To overcome
this issue, a thorough modelling analysis was performed using the CFD model. The
modelling investigation optimised the design parameters and the inlet orientation for the
novel gross pollutant device. Two different inlet conditions were analysed. One was an
inlet parallel to the ogee weir, and the other perpendicular to it. The inlet parallel to the
ogee weir was considered the better option, as the water level over the ogee weir was
higher due to wave reflection, which can provide higher velocity and shear stress. The
location of the perforations was found to work more efficiently at the bottom, as there
was higher velocity and shear stress in that position.
To reduce the reflected wave on the small screening device, the model showed that
providing longer inlet pipes was a solution. An inlet 1.5m long will reduce the reflected
wave by 10% more than a 0.3m inlet. The four standard ogee weir orientations proposed
by the US Army Corp of Engineers [169] were investigated, and it was found that an
inclined slope of 1H: 3V from the rectangular device to the ogee weir was the most
efficient.
Conclusions
Md Abdul Aziz 160
The CFD model analysis was very helpful for designing the experimental parameters,
as well as providing the concept for the gross pollutant device.
8.3.2 Laboratory experiments
Once the design parameters and inlet orientation using CFD models were
understood, a second experimental facility was set up with minor adjustments to the
initial device. This novel device is called the ‘Comb Separator’. The ‘Comb Separator’
was subjected to a series of trials with different combinations of numbers of combs,
spacing of combs, flow volume and weir openings being tested. The proposed device
can capture larger sewage solids of more than 10mm diameter with over 95% capture
efficiency. The capture efficiency is dependent on selected experimental parameters
which can vary the sewer capture efficiency. These parameters are flow (discharge),
weir opening, comb spacing and layers. Two layers of combs were found to be more
efficient than three layers. Increasing the comb spacing improves capture efficiency.
Robustness of optimum set-ups were tested to generate consistent results. The
laboratory testing was especially beneficial in helping to understand the sewer solids
capture efficiency and the level of blockage on the screen.
8.3.3 ANN application
The experimental work was restricted by the physical limitations inherent in
laboratory studies. As sewage solids vary in density, they can be difficult to study using
physical laws based on deterministic models such as CFD. The Artificial Neural Network
(ANN) model has the capacity to accurately predict the outcome of complex, non-linear
physical systems where physico-chemical processes are relatively poorly understood. A
series of laboratory tests was conducted with 55 different sets of data (i.e. varying flows
and combs spacing conditions). Forty-seven sets of experimental data were used with
60% being for training, and 20% each for testing and validation of the model. Separate
validation data sets were used to judge the overall performance of the trained network.
The model is able to successfully predict the experimental results with more than 90%
accuracy, with the average absolute percentage of errors varying from 4% to 7%.
The application of ANN supplemented the CFD and laboratory experiments.
This method is especially beneficial when other deterministic physical modelling or
experimental results are not convincing enough to derive a conclusion.
Conclusions
Md Abdul Aziz 161
8.3.4 Sensitivity analysis
Once an ANN model was established and ready to use for experiments on
capture efficiency, it was tested with different options to identify the best one for the
current settings. Input parameters such as flowrate, effective comb spacing, device run
time, weir opening and the number of comb layers were considered. It was important to
compare with the industry standard device Hydro-JetTM screen. It was found that, in low-
flow situations, the Comb Separator has better capture efficiency and avoidance of
screen blockage than the Hydro-JetTM screen. The comparison highlighted the
importance of understanding parameter sensitivity in the Comb Separator. It is strongly
recommended that sensitivity testing be undertaken for any hydraulic device before
application in the natural environment [97].
The sensitivity analysis provided a more precise understanding of the relative
importance of the input parameters. For example, with the application of the Latin
Hypercube sampling technique using 10,000 data [85], it was found that effective spacing
was the most influential parameter, followed by flow discharge and device run time. The
sampling technique also provided a better understanding of the contribution of each input
parameter. For example, with one unit increase in effective spacing, the capture
efficiency of sewage solids in the Comb Separator can increase by 5.31 units. This
sensitivity information provided better insight about the relative importance for the input
parameters. This information would be highly valuable in managing the device in actual
sewer overflow conditions.
8.4 Limitations
A potential limitation of this research was the small sample size in the laboratory
experiments. This is due to the physical limitations in creating more data realistically from
the experimental series. The smallest spacing tested was 10mm, as it was difficult to
provide set-ups smaller than 10 mm. There was also one limitation with the CFD model.
As the sewer solids were materials of different density, this was difficult for the CFD
model to simulate. To overcome these limitations, the ANN model was used and
statistical sensitivity testing was done in the later phase of this research. The use of such
Conclusions
Md Abdul Aziz 162
methods is recommended in order to overcome the above-mentioned challenges
experienced in this research.
8.5 Future Research
Future study on this topic should consider the following key points:
The ANN model could be revised based on the understanding gleaned from
the sensitivity analysis, which initially considered all 16 input parameters. As the
laboratory data set is small, this could provide better insight about the optimisation
conditions for the laboratory experimental set-ups.
Sensitivity of the different algorithms while using ANN also needs to be tested.
Changing the number of hidden layers or splitting the data set difference between input
and output parameters could provide a better understanding of the best case scenario
for the ANN.
Further experimentation with the Comb Separator device is recommended,
especially when flow rates are higher (up to 120 l/s). It is important to understand the
performance of the device in high-flow conditions such as floods and heavy rains. These
results need to be compared with industry standard devices such as the Hydro-JetTM.
Further experiments should consider small-diameter sewage solids such as
cigarette butts. It was observed that the small-diameter sewage solids showed higher
variation in capture efficiency. Further trials could contribute to a better understanding of
this issue. The current sensitivity analysis will provide a useful guideline for analysis.
Future investigation should also consider onsite testing of the proposed Comb
Separator in actual sewage overflow conditions.
The ‘Comb Separator’ showed good application potential at low flows for
improving sewage solids capture efficiency in the urban sewerage system.
References
Md Abdul Aziz 163
References
References
Md Abdul Aziz 164
References
1. Adamsson A, Stovin V, Bergdahl L., 2003, ‘Bed shear stress boundary
condition for storage tank sedimentation’, Journal of Environmental
Engineering, vol. 129, no.7, pp.651–658.
2. Adamowski J., & Karapataki C., 2010, ‘Comparison of multivariate regression
and artificial neural networks for peak urban water demand forecasting:
evaluation of different ANN learning algorithms’, Journal of Hydrologic
Engineering, Vol. 15, No. 10.
3. Alex Dick, 2013, ‘Campaigners storm parliament and tell them: UK Surfing is
worth British Pound 1.8 Billion’, the surfer’s path,
http://surferspath.com/news/campaigners-storm-parliament-and-tell-them-
uk-surfing-is-worth-1-8billion.html#kY3MTAA8J6dl4W2l.97
4. Alp, M., Cigizoglu, H., 2007, Suspended sediment load simulation by two
artificial neural network methods using hydro-meteorological data.
Environmental Modelling and Software, vol.22, no. 1, pp.2-23.
5. ASCE Task Committee on Application of Artificial Neural Networks in
Hydrology, 2000a, ‘Artificial neural networks in hydrology. I: preliminary
concepts.’ Journal of Hydrologic Engineering, ASCE vol.5, no. 2, pp.115-123.
6. Akhtar, M.C., Corzo, G.A., van Andel, S.J., & Jonoski, A., 2009, ‘River flow
forecasting with artificial neural networks using satellite observed precipitation
pre-processed with flow length and travel time information: case study of the
Ganges river basin,’ Hydro. Earth Syst. Sci., vol.13, pp.1607-1618.
7. ASCE Task Committee on Application of Artificial Neural Networks in
Hydrology, 2000b, ‘Artificial neural networks in hydrology. II: hydrologic
applications.’ Journal of Hydrologic Engineering, ASCE vol.5, no.2, pp.124-
137.
8. Aziz, M.A., Imteaz, M.A., Choudhury, T.A., & Phillips, D.I., 2011, ‘Artificial
Neural Networks for the prediction of the trapping efficiency of a new sewer
overflow screening device’, 19th International Congress on Modelling and
Simulation, Perth, Australia.
9. Aziz, M. A., Imteaz, M., Choudhury, T. A. & Phillips, D. 2013a, ‘Applicability of
artificial neural network in hydraulic experiments using a new sewer overflow
screening device’, Australian Journal of Water Resources, Vol. 17, No. 1,
pp.77-86.
References
Md Abdul Aziz 165
10. Aziz, M. A., Imteaz, J. Naser & Phillips, D. 2013b, ‘Hydrodynamic
Characteristics of a New Sewer Overflow Screening Device: CFD Modelling
and Analytical Study’, International Journal of Civil and Environmental
Engineering, vol.7 no.1, pp 71-76.
11. Aziz, M. A., Imteaz, M. Huda., & J. Naser 2014a, ‘Optimising inlet condition
and design parameters of a new sewer overflow screening device using
numerical modelling technique‘, Journal of Water, Science and Technology,
vol.70, no.11,pp.1880-1887
12. Aziz, M. A., Imteaz, M., Rasel, H.M., & Phillips, D., 2015a, ‘Development and
Performance Testing of ‘Comb Separator’, A Novel Sewer Overflow
Screening Device’, International Journal of Environment and Waste
Management, Vol. 16, No.3, 2015.
13. Aziz, M. A., Imteaz, M., Rasel,H.M., & Samsuzzoha, M. 2015d, ‘Parameter
Sensitivity Using Sampling Technique for a proposed ‘Comb Separator’, A
sewer overflow screening device’, ASEAN- Australian Engineering Congress
on Innovative Technologies for Sustainable Development and Renewable
Energy 11-13 March 2015.
14. Andoh, R.Y.G., and Smisson, R.P.M., 1994, ‘High Rate Sedimentation in
Hydrodynamic Separators’. Proceeding of 2nd International Conference On
Hydraulic Modelling Development and Application of Physical and
Mathematical Models, Stratford, UK, pp. 341. 358.
15. Andoh R.Y.G., Smith B.P. & Saul A.J., 1999, ‘the screen efficiency of a novel
self-cleanshing CSO.’ 8th International Conference on Urban Storm Drainage,
Sydney, Australia, 30 August – 3 September, pp.205-212.
16. Andoh, R. Y. G., and Saul, A.J., 2000, ‘Field Evaluation of Novel Wet-Weather
Screening Systems’, Proc.WEF Speciality Conf., Collection Systems Wet
Weather Pollution Control: Looking into Public, Private and Industrial Issues,
New York, USA, 7-10 May.
17. Andoh, R.Y.G. and Saul, A.J., 2003, ‘The use of hydrodynamic vortex
separators and screening systems to improve water quality’, Water Science
& Technology, Vol. 47 No. 4, pp. 175–183.
18. An, G., 1996, ‘The effects of adding noise during backpropagation training on
a generalization performance’, Neural Computation, vol. 8, pp. 643-674.
19. Armitage, N & Rooseboom, A., 2000, ‘the removal of urban litter from
stormwater conduits and streams: paper 1 the quantities involved and
References
Md Abdul Aziz 166
catchment litter management options’. Water SA (South Africa), vol.26, no.2,
pp.181-188.
20. Arnett, C. J. & Gurney, P. K., 1998, ‘High rate solids removal and chemical
and non-chemical UV disinfection alternatives for treatment of CSO.s’,
Innovation 2000, Conference on Treatment Innovation for the Next Century,
Cambridge, UK.
21. Averill, D., Mack-Mumford, D., Marsalek, J., Andoh, R. & Weatherbe, D.,
1997, ‘Field Facility for Research and Demonstration of CSO Treatment
Technologies’, Water Science & Technology, vol.36, no.8-9,pp. 391 . 396.
22. AVL 2008, AVL Fire CFD Solver v8.5 manual, A-8020 Graz, Austria.
23. Balmforth, D.J., & Henderson, R.J., 1988, ‘ A guide to the design of storm
overflow structures’, Water Research Centre Report, ER304E, WRc,
Swindon, Wilts, England.
24. Betts, P. L., 1979, ‘A variation principle in terms of stream function for free
surface flows and its application to finite element method’, Comp. and Fluids,
vol.7, no.2, pp.145-153.
25. Berry, W.D., 1993, ‘Understanding regression assumptions. Saga university
paper series on quantitative applications in the social sciences’, 07-092.
Newbury Park, CA: sage.
26. Becker, S., 1991, ‘Unsupervised learning procedures for neural networks,’
International Journal of Neural Systems, vol. 2, pp.17-33.
27. Bhajantri, M. R., Eldho, T. I., & Deolalikar, P. B., 2006, ‘Hydrodynamic
modelling of flow over a spillway using a two-dimensional finite volume-based
numerical model’, Sadhana Vol. 31, Part 6, December, pp. 743-754.
28. Bolt, G., 1992, ‘Fault tolerance in artificial neural networks,’ Advanced
Computer Architecture Group, Department of Computer Science, University
of York, Heslington, York, YO1 5DD, U.K.
29. Bowerman, B,L., & O’Connell, R.T., 1990, ‘Linear statistical models: An
applied approach’ (2nd ed.). Belmont, CA: Duxbury.
30. Brombach, H., 1992, ‘Solids Removal from Combined Sewer Overflows with
Vortex Separators’. NOVATECH 92, International Conference on Innovative
Technologies in the Domain of Urban Water Drainage, Lyon (France),
November 3-5, pp. 447-459.
31. Bruen, M., Yang, J., 2006, ‘Combined hydraulic and black-box models for
flood forecasting in urban drainage systems’, Journal of Hydrologic
Engineering, vol.11, no.6.
References
Md Abdul Aziz 167
32. Burgisser, M. F. & Rutschmann, P., 1999, ‘Numerical solution of viscous 2DV
free surface flows: Flow over spillway crests’, Proc., 28 th IAHR Congr.,
Technical University Graz, Graz, Austria.
33. Casey, M. & Wintergerste, T., 2000, ‘Special interest group on quality and
trust in industrial CFD: best practice guidelines’. Switzerland: European
Research Community on flow turbulence and combustion (ERCOFTAC).
34. Cassidy, J. J., 1965, ‘Irrotational flow over spillways of finite height’, Journal
of Engineering and Mechanical Division., ASCE, vol. 91, no.6, pp.155-173.
35. Carpenter, W.C., & Barthelemy, J., 1994, ‘Common misconcepts about neural
networks as approximations.’ J.Comp. in Civ. Engrg. ASCE, vol.8, no.3,
pp.345-358.
36. Center for marine conservation (CMC),1998,‘Coastal clean-up- united states’,
www/cmc-ocean.org/cleanupbro.
37. Chandramouli, V. & Deka, P., 2006, ‘Neural network based decision support
model for optimal reservoir operation,’ Water Resources Management,
vol.19, no.4, pp.447-464.
38. Choudhury, T. A., Hosseinzadeh, N. & Berndt, C. C., 2011, ‘Artificial Neural
Network application for predicting in-flight particle characteristics of an
atmospheric plasma spray process’, Surface and Coatings Technology, vol.
205, pp. 4886-4895.
39. Choudhury, T. A., Hosseinzadeh, N. & Berndt, C. C., 2012, ‘Improving the
Generalization Ability of an Artificial Neural Network in Predicting In-Flight
Particle Characteristics of an Atmospheric Plasma Spray Process,’ Journal of
Thermal Spray Technology, vol. 21, pp. 935-949.
40. Choudhury T.A., 2013 ‘Artificial Neural Networks Applied to Plasma Spray
Manufacturing’. Ph.D. thesis. Swinburne University of Technology, Australia.
41. Chow, V. T., 1959, Open-channel hydraulics, McGraw-Hill, New York,
pp.365-380.
42. Claire E, Imrie, Sevket Durucan, 1999, ‘River flow prediction using the
cascade-correlation neural network learning architecture’. Water 99 Joint
Congress- Brisbane, Australia 6-8 July 1999.
43. Chang, L.C. & Chang, F.J., 2001, ‘Intelligent control for modelling of real time
reservoir operation,’ Hydrol. Process., vol.15, no.9, pp. 1621-1634.
44. Chang, F.J. & Chang,Y.T., 2006, ‘Adaptive neuro-fuzzy inference system for
prediction of water level in reservoir’, Adv. Water Resources., vol.29, no.1,
pp.1-10.
References
Md Abdul Aziz 168
45. Chang, L.C. and Chang, F.J., 2009, ‘Integrating hydrometeorological
information for rainfall-runoff modelling by artificial neural networks,’
Hydrol.Process., vol.23, no.1, pp.1650-1659.
46. Chiang, Y.M., Hsu, K.L., Chang, F.J., Hong, Y., & Sorooshian, S., 2007,
‘Merging multiple precipitation sources for flash flood forecasting’, Journal of
Hydrology, vol. 330, no.3-4, pp.183-196.
47. Chiang, Y.M., Li-Chiu, Chang., Tsai, M.J.,Wang, Y.F., & Chang, F.J., 2010,
‘Dynamic neural networks for real-time water level predictions of sewerage
systems-covering gauged and ungauged sites’, Journal of Hydrology and
Earth System Sciences, vol.14,pp.1309-1319.
48. Churchland, P. S. & Sejnowski, T. J., 1992, ‘the computational brain:’ The
MIT press.
49. Crick, M.J., Hill, M.D. & Charles, D.,1987, 'The Role of Sensitivity Analysis in
Assessing Uncertainty.’ In: Proceedings of an NEA Workshop on Uncertainty
Analysis for Performance Assessments of Radioactive Waste Disposal
Systems, Paris, OECD, pp. 1-258.
50. Dewals, B.J., Kantoush, S.A., Erpicum, S., Pirotton, M., & Schleiss, A.J.,
2008, ‘Experimental and numerical analysis of flow instabilities in rectangular
shallow basins’, Environ. Fluid Mech., vol.8, pp.31-54.
51. Dufresne, M., Vazquez, J., Terfous, A., Ghenaim, A., & Poulet, J.,
2009, ‘CFD Modeling of Solid Separation in Three Combined Sewer Overflow
Chambers’, J. Environ. Eng., vol.135 no.9, pp. 776–787.
52. Dufresne, M., Vazquez,j., Terfour, A., Ghenaim, A., & Poulet, J.-B., 2009,
‘Experimental investigation and CFD modelling of flow, sedimentation, and
solids separation in a combined sewer detention tank.’ Computational Fluid,
vol.8, no.5, pp.1042-1049.
53. Durbin, J., Watson, G. S., 1950, ‘Testing for Serial Correlation in Least
Squares Regression, I’. Biometrika vol.37 (3–4), pp. 409–428.
doi:10.1093/biomet/37.3-4.409. JSTOR 2332391
54. Durbin, J., Watson, G. S., 1951, ‘Testing for Serial Correlation in Least
Squares Regression, II’, Biometrika vol.38 (1–2), pp.159–179.
doi:10.1093/biomet/38.1-2.159. JSTOR 2332325
55. Eldho, T.I., 2008, Lecture 26, Lecture Series on Fluid Mechanics, Department
of Civil Engineering IIT Bombay, URL: http://nptel.iitm.ac.in
References
Md Abdul Aziz 169
56. Fahlman, S. E., 1988, ‘Faster-learning variations on back propagation: an
empirical study,’ Proceedings of the 1988 Connectionist Models Summer
School, pp. 38-51.
57. Faram, M.G., Andoh, R.Y.G. and Smith, B.P. 1999, ‘The Screen Efficiency of
a Novel Self-cleansing CSO’, 8th International Conference on Urban Storm
Drainage, Sydney, Australia, 30 August - 3 September 1999, pp. 205-212.
58. Faram, M. G. & Andoh, R. Y. G., 2000, ‘Application of Simulation and
Predictive Techniques for the Evaluation of Hydrodynamic Separators’,
Wastewater Treatment: Standards and Technologies to Meet the Challenges
of the 21st Century, CIWEM/AETT Millennium Conf., Leeds, UK, 4-6 April, pp
223-230.
59. Faram, M.G., Andoh, R.Y.G. & Smith, B.P., 2001, ‘Optimised CSO screening:
A UK perspective’, 4th Novatech International Conference on Innovative
Technologies in Urban Drainage, Lyon, France, 25-27 June 2001, pp.1031-
1034.
60. Faram, M.G., 2001, ‘Developments in screening technology’ CIWEM
Conference-Combined sewer overflow: The challenge and latest innovations
61. Fausett, L., 1994, Fundamentals of neural networks, Prentice Hall,
Englewood Cliffs, N.J.
62. Ferziger, J.H. and M. Peric, 2002, ‘Computational Methods for Fluid
Dynamics’. 3rd ed: Springerlink.
63. Floyd, J., 2005, ‘Converting an Idea into a Worldwide Business
Commercializing Smelting Technology.’ Metall. Trans. B., vol.36 (B), pp.557-
575.
64. Field, R. 1972, ‘The Swirl Concentrator as a CSO Regulation Facility’, USEPA
Report no. R2-72-008, USA.
65. Gonzalesz-Ubierna, S., Jorge-Mardomingo, I., Cruz, M.T., Valverde, I. &
Casermeiro, M.A., 2013, ‘Sewage Sludge Application in Mediterranean
Agricultural soils: Effects of Dose on the Soil Carbon Cycle’, International
Journal of Environment and Research, vol .7, no.4, pp 945-956.
66. Guo, Y., Wen, X., Wu, C. & Fang, D., 1998, ‘Numerical modelling of spillway
flow with free drop and initially unknown discharge’, J. Hydr. Res., Delft, The
Netherlands, Vol. 36, no.5, pp. 785-801.
67. Guanghua Q, Wang, S., 2011, ‘Annual runoff prediction with a sensitive
artificial neural networks model’, 34th IAHR World Congress – Balance and
References
Md Abdul Aziz 170
Uncertanity, 33th Hydrology and Wagter Resources Symposium, 10th
Hydraulics Conference, 26th -1st July , Brisbane, Australia.
68. Guanghua Qin, Hongxia Li, Xin Wang, Qingyan He,& Shenqi Li, 2015,
‘Annual runoff prediction using a nearest-neighbour method based on cosine
angle distance for similarity estimation’, ICGRHWE14, Guangzhou, China.
69. Guide to the design of combined sewer overflow structures, 1994, Report No
FRO488.
70. Hall, J. W., 2003, ‘Handling uncertainty in the hydroinformatic process’
Journal of Hydro informatics., vol.5, no.4, pp. 215–232.
71. Hall, J. W., & Solomatine, D., 2008, ‘A framework for uncertainty analysis in
flood risk management decisions’ J. River Basin Manage., vol.6, no.2,pp.85–
98.
72. Hall, J.W., Boyce, S.A., Wang, Y., Dawson, R.J., Tarantola, S., Saltelli, A.,
2009, ‘Sensitivity analysis for hydraulic models’ Journal of hydraulic
engineering @ ASCE/November 959.
73. Hamby, D.M., 1995a, 'A Numerical Comparison of Sensitivity Analysis
Techniques', scheduled to appear in Health Phys. 68.
74. Hamby, D.M., 1995b, ‘A review of techniques for parameter sensitivity
analysis of environmental models’, Environmental Monitoring and
Assessment vol.32, pp. 135-154.
75. Hamby, D.M., 1993, 'A Probabilistic Estimation of Atmospheric tritium Dose',
Health Phys. vol.65, pp.33-40.
76. Hagan, M. T. & Mehnaj, M. B., 1994, ’Training feedforward networks with the
Marquardt algorithm,’ IEEE Transactions on Neural Networks, vol. 5, pp. 989-
993, November.
77. Harwood, R. & Saul. A.J., 1999, ‘The influence of CSO chamber size on
particle retension efficiency performance.’ In Proceedings of the 8th
International Conference on Urban Storm Drainage, 30 August-3 September,
Sydney, Australia, pp.1-9.
78. Harwood, R., 2002, CSO modelling strategies using computational fluid
dynamics, In Eric W. Strecker and Wayne C. Huber (Eds.), Urban Drainage
2002: Proceedings of 9th International Conference on Urban Drainage -
9ICUD. Portland, Oregon, USA: American Society of Civil Engineers (ASCE),
pp.8-17.
79. Haykin, S., 1994, Neural networks: a comprehensive foundation, Mac Millan,
New York.
References
Md Abdul Aziz 171
80. Hedges, P.D.,1993, ‘The Relationship between Field and Model Studies of
an Hydrodynamic Separator Combined Sewer Overflow’, 6th International
Conference on Urban Storm Drainage, Niagara Falls Canada
81. Helton, J.C., Iman, R.L. & Brown, J.B., 1985, 'Sensitivity Analysis of the
Asymptotic Behaviour of a Model for the Environmental Movement of
Radionuclides’, Ecol. Modelling. Vol.28, pp. 243-278.
82. Hirt .C.W., & Nichols.B. D.,1981, ‘Volume of fluid (VOF) method for the
dynamics of free Boundaries.’ Journal of Computational Physics, vol.39, no.1,
pp. 201-225.
83. Hilgenstock, A. & Ernst, R., 1996, ‘Analysis of installation effects by means of
computational fluid dynamics CFD vs experiments’, Flow measurement and
instrumentation, vol.7, no.3-4, pp.161-171.
84. Hrabak D, Pryl K, Richardson J, Zeman E., 1999, ‘3-Dimensional modelling –
a new tool for the evaluation of CSO hydraulic performance’, In Proceedings
of the 8th ICUSD, pp. 928–933, Sydney, Australia.
85. Hoare, A., Regan, D.G., & David, Wilson.,D.P., 2008, ‘Sampling and
sensitivity analyses tools (SaSAT) for computational modelling’, Theoretical
Biology and Medical Modelling. doi:10.1186/1742-4682-5-4
86. Huda. M.N., 2012 ‘Computational Fluid Dynamic Modelling of Zinc Slag
Fuming Process.’ Ph.D. thesis. Swinburne University of Technology.
87. HydroQual, 1993, city-wide floatables study, final report-sources, fate and
control of floatable materials in New York Harbor; prepare for the New York
city department of environmental protection, bureau of environmental
engineering, Division of water quality improvements.
88. HydroQual, 1995, City Wide Floatables study, floatables pilot program final
report, evaluation of non-structural methods to control combined and storm
sewer flotable materials; prepared for the new York city department of
environmental protection, bureau of environmental engineering, divisions of
water quality improvements.
89. IBM Corp. Released 2013. ‘IBM SPSS Statistics for Windows’, Version 22.0.
Armonk, NY: IBM Corp.
90. Ikegawa, M. & Washizu, K., 1973, ‘Finite element method applied to analysis
of flow over a spillway crest’, International Journal of Numerical Methods in
Engineering, vol.6, pp. 179-189.
91. Iman RL, Helton JC, Campbell JE, 1981a, ‘An Approach To Sensitivity
Analysis Of Computer-Models. Introduction, Input Variable Selection and
References
Md Abdul Aziz 172
Preliminary Variable Assessment’. Journal of Quality Technology, vol.13,
no.3, pp.174.
92. Iman, R. L., Helton, J. C. & Campbell, J. E., 1981b, ‘An approach to sensitivity
analysis of computer models: Part II- Ranking of input variables, response
surface validation, distribution effect and technique synopsis’. Journal of
Qual. Technol. Vol.13, pp. 232-240.
93. Iman RL, Helton JC, 1988, ‘An Investigation of Uncertainty and Sensitivity
Analysis Techniques for Computer Models’, Risk Analysis vol.8, no.1, pp.71-
90.
94. Iman, R.L., & Helton, J.C., 1991, 'The Repeatability of Uncertainty and
Sensitivity Analyses for Complex Probabilistic Risk Assessments', Risk
Analysis. Vol.11, pp. 591-606.
95. International Atomic Energy Agency (IAEA), 1989, ‘Evaluating the reliability
of predictions made using environmental transfer models.’ Vienna: Safety
Series No. 100. Report No. STI/PUB/835; 1-106.
96. Jayawardena, A. W., & Fernando, D. A. K., 1996, ‘Comparison of multi-layer
perceptron and radial basis function network as tools for flood forecasting’,
Proc., North Am. Water and Envir. Conf., ASCE, New York, pp.457–458.
97. Johnson, P. A., 1996, ‘Uncertainty of hydraulic parameters’, J. Hydraul. Eng.,
122(2), 112–114.
98. Kai, W. Jufeng, Y., Guangshun, S. & Qingren, W., 2008, ‘An expanded
training set based validation method to avoid overfitting for neural network
classifier,’ in Fourth International Conference on Natural Computation (ICNC),
pp. 83-87.
99. Karystinos, G. N. & Pados, D. A. 2000, ‘On overfitting, generalization, and
randomly expanded training sets,’ IEEE Transactions on Neural Networks,
vol. 11, pp. 1050-1057.
100. Kin Choi Luk, James E. Ball and Ashish Sharma, 1999, ‘Integration of
Artificial neural networks and geographical information systems for rainfall
forecasting’, Water 99 Joint Congress- Brisbane, Australia 6-8 July 1999.
101. Kohavi, R., 1995, ‘A study of cross-validation and bootstrap for accuracy
estimation and model selection,’ in International Joint Conference on Artificial
Intelligence (IJCAI), pp. 1137-1145.
102. Launder, B.E. & D.B. Spalding, 1974, ‘The Numerical Computation of
Turbulent Flows.’ Com. Meth. App. Mech. Eng.,. vol.3, no.2, pp. 269-289.
References
Md Abdul Aziz 173
103. Li, W., Xie, Q., & Chen, C. J., 1989, ‘Finite analytic solution of flow over
spillways’, J. Engrg. Mech., ASCE, vol.115, no.12, pp. 2635-2648.
104. Liu, Y., 2006, ‘Create stable neural networks by cross-validation,’ in
International Joint Conference on Neural Networks (IJCNN), IEEE, pp. 3925-
3928.
105. Luyckx, G., Vaes, G., & Berlamont, J., 1999, ‘Experimental investigation
on the efficiency of a high side weir overflow’, Water Science and Technology,
vol. 39 no.2, pp. 61–68.
106. Madhani, J. T. & Brown, R. J., 2011, ‘A literature review on research
methodologies of gross pollutant traps’, The Proceedings of the First
International Postgraduate Conference on Engineering, Designing and
Development the Built Environment for Sustainable Wellbeing, Queensland
University of Technology, Brisbane, Qld.
107. Madhani. J.T., 2010 ‘The Hydrodynamic and Capture/Retention
Performance of A Gross Pollutant Trap’, Ph.D. thesis, Queensland University
of Technology.
108. Maier, H. R., & Dandy, G.C., 1996, ‘The use of artificial neural networks
for the prediction of water quality parameters’, Journal of Water Resources
Research, vol.32, no.4, pp.1013-1022.
109. Maier, H. R., & Dandy, G.C., 2001, ‘Neural network based modelling of
environmental variables: asystematic approach’, Journal of Mathematical and
Computer Modelling, vol.33, pp.669-682
110. Maren, A., Harston, C., & R. Pap R., 1990, ‘Handbook of neural
computing applications,’ academic, San Diego, Cardiff, 448.
111. MATLAB (R2012a: The MathWorks Inc., Natick, MA, USA)
112. Matsuoka, K., 1992, ‘Noise injection into inputs in back-propagation
learning,’ IEEE Transactions on Systems, Man and Cybernetics, vol. 22, pp.
436-440.
113. Matthieu D., Jose V., Abdelali T., Abdellah G & Jean B.P., 2009, ‘CFD
Modeling of Solid Seperation in Three Combined Sewer Overflow Chambers.’
Journal of Environmental Engineering (ASCE), September, pp. 776-789.
114. Maynord, S. T., 1985, ’General spillway investigation’, Tech. Rep. HL-85-
1, U.S. Army Engineer Waterways Experiment Station, Vicksburg, Miss.
115. McKay MD, Beckman RJ, Conover WJ, 2000, ‘A comparison of three
methods for selecting values of input variables in the analysis of output from
a computer code’, Technometrics vol.42, no.1, pp.55.
References
Md Abdul Aziz 174
116. Melching, C. S., 1995, ‘Reliability estimation’, Computer models of
watershed hydrology, V. P. Singh, ed., Water Resources, Littleton,Colo.,
pp.69–118.
117. Metcalf and Eddy, 1991, ‘Wastewater Engineering, Treatment, Disposal
and Reuse’, Third Ed., Boston, MA: McGraw-Hill, Inc. pp. 448-451.
118. Moffa, P.E. 1997, ‘Scarborough CSO project-review of CSO screen
application’, report prepared for the City of Scarborough, Municipality of
Metro, Toronto, ON.
119. Myers, R., 1990, Classical and modern regression with applications (2nd
ed.). Boston: Duxbury Press.
120. National Rivers Authority, 1993, General guidance note for preparatory
work for AMP2, Version 2, October, UK.
121. Natural Resource Management Ministerial Council 2004, ‘National Water
Quality management strategy’, Guidelines for sewerage systems bio solids
management, Australia.
122. Nelson, M. and Illingworth, W., 1991, ‘A practical guide to neural nets,’
ed. United States.
123. Newman, T.L., Leo, W.M., & Gaffoglio, R., 2000, ‘Characterization of
urban-source floatables. In collection system wet weather pollution control
conference’, WEF; Rochester, New York, May 2000.
124. Olsen, N. R. & Kjellesvig, H. M., 1998, ‘Three-dimensional numerical flow
modelling for estimation of spillway capacity’, J. Hydr. Res., Delft, The
Netherlands, Vol. 36, no.5, pp. 775-784.
125. Okamoto, Y., Konugi, M. & Tsuchiya, H., 2002, ‘Numerical Simulation of
the Performance of a Hydrodynamic Separator’, 9ICUD conference, Portland,
Oregon, USA.
126. Pannell, J.C., Daniell, T.M., Walker,D.J., 1996, ‘Rainfall Based Real-Time
flood forecasting using artificial neural networks’, 23 Hydrology and Water
Resources Symposium, Hobart, Australia, 21-24 May.
127. Parsinejad, M., Feng, Y., & Ghanbarian, B., 2006,‘Sensitivity Analysis of
Hydraulic Parameters in the Simulation of Unsaturated Soil Water Dynamics’,
Iran Agricultural Research, Vol. 24, no 2 and Vol. 25, no 1.
128. Patankar, S.V. & D.B. Spalding, 1972, ‘A Calculation Procedure for Heat,
Mass and Momentum Transfer in Three-Dimensional Parabolic Flows.’ Int. J.
Heat. Mass. Trans., vol.15,pp.1787-1806.
References
Md Abdul Aziz 175
129. Phillips, D.I., 1999, ‘A new litter trap for urban drainage system’, Water
Science and Technology, vol.39, no.2, pp.85-92.
130. Phillips, D. I. & Simon, M., 2007, ‘A sewage solids screening system for
CSO chambers. Novatech’, Proc 6th International Conference, Sustainable
techniques and Strategies in Urban Water Management, Lyon, France, Vol 2,
pp. 697-704.
131. Phillips, D.I., Simon, M., 2010, ‘An improved method of screening sewer
solids during CSO events.’ proceedings of the 7th International Novatech
conference, Lyon, 2010, abstracts pp.175.
132. Pollert J., 1999, ‘Free surface modelling in sewer system’, In Proceedings
of the 5th Fluent User Conference, Prague, Czech Republic, pp. 61-69.
133. Pollert J, Stransky D., 2003, ‘Combination of computational techniques –
evaluation of CSO efficiency for suspended solids separation’, Water Science
and Technology, vol.47, no.4, pp.157–166.
134. Prechelt, L., 1998, ‘Automatic early stopping using cross validation:
quantifying the criteria,’ Neural Networks, vol. 11, pp. 761-767.
135. Raduly, B., Genaey, K.V., Capodaglio, A., Mikkelsen, P., & Henze, M.,
2007, ‘Artificial neural networks for rapid WWTP performance evaluation:
methodology and case study’, Journal of Environmental Modelling and
Software, vol. 22, no.8, pp.1208-1216.
136. Riedmiller, M. & Braun, H. 1993, ‘A direct adaptive method for faster
backpropagation learning: the RPROP algorithm,’ in IEEE International
Conference on Neural Networks, vol.1, pp. 586-591.
137. Rumelhart, D.E., Hinton, G.E., & Williams, R. J. 1986, ‘Learning internal
representations by error propagation’, Parallel distributed processing, Vol. I,
MIT Press, Cambridge, Mass., 318-362.
138. Rumelhart, D. E., 1990, ‘Brain style computation: learning and
generalization,’ in An introduction to neural and electronic networks, ed:
Academic Press Professional, Inc., pp. 405-420.
139. Saltelli A., 2004, ‘Sensitivity analysis in practice: a guide to assessing
scientific models’, Wiley, Hoboken, NJ.
140. Sandhu, N., and Finch, R. 1996, ‘Emulation of DWRDSM using artificial
neural networks and estimation of Sacramento River flow from salinity’, Proc.,
North American Water and Environment Conference, ASCE, New York,
pp.4335-4340.
References
Md Abdul Aziz 176
141. Sarle, W.S 1995, ‘Stopped Training and Other Remedies for Overfitting’
in Proceedings of the 27th Symposium on the Interface of Computing Science
and Statistics, pp.352-360.
142. Sato, Y. & Sekoguchi, K. , 1975, ‘Liquid Velocity Distribution in Two Phase
Bubble Flow.’ Int. J. Multiphase Flow,. Vol.2.no.1, pp.79-95.
143. Saul AJ, Ellis DR., 1992, ‘Sediment deposition in storage tanks’, Water
Science and Technology, vol. 25, no.8, pp.189-198.
144. Saul, A.J., 1998, ‘CSO state of the art review: a UK perspective.’ UDM’98,
Fourth International Conference on developments in urban drainage
modelling. September, London, UK.
145. Saul A J, 2000, ‘Screen Efficiency (Proprietary Designs)’, UK Water
Industry Research Limited (UKWIR), Report 99/WW/08/5.
146. Saul, A.J., 2003, ‘CSO: State of the Art Review’ Urban Water
Management: Science Technology and Science Delivery, NATO Science
Series, vol.25, pp. 179-190.
147. Saul, A J, Blanksby, J., 2007, ‘CSO Aesthetics and Static Screening
Technologies’, World Environmental and Water Resources Congress:
Restoring our natural Habitat.
148. Saul, A. J., (2008). CSO: State of the Art Review, ICUD Conference
Edinburgh.
149. Science Clarified, 2015, ‘Mind Versus Metal’,
http://www.scienceclarified.com/scitech/Artificial-Intelligence/Mind-Versus-
Metal.html
150. Sharpe DE and Kirkbride TW, 1959, ‘Storm Water Overflows: The
Operation and Design of a Stilling Pond’, Proc. ICE, Vol. 13.
151. Smisson, B., 1967, ‘Design, Construction and Performance of Vortex
Overflow’, Institute of Civil Engineers, Symposium on Storm Sewage
Overflow, London, UK, pp.99-110.
152. Smith, B. P., and Andoh, R. Y. G., 1997, ‘New generation of
hydrodynamic separators for CSO treatment’. Proc. 2nd Int. Conf. on the
Sewer as a Physical, Chemical and Biological Reactor, Aalborg, Denmark,
25-28 May.
153. Simon. M., 2003, ‘Vorthbach km 2+335 bis 3+500 Hydraulische
Bestandsuntersuchung and Vorschlag zur Veranderung’, Report prepared for
the Emschergenossenschaft, Essen, Germany.
References
Md Abdul Aziz 177
154. Simon. M. and Phillips, D.I. 2008, ‘The development of a sewer solids
screening system for CSO cAhambers’ 11th International Conference on
Urban Drainage, Edinburgh, Scotland, UK.
155. Starrett, S.K., Najjar, Y.M., & Hill, J.C. 1996, Neural networks predict
pesticide leaching. Proceeding American Water and Environmental
Conference, ASCE, New York, pp.1693-1698.
156. Stephenson, J., Gall, B., Mroczek, C., Newbigging & Parker, M. J, 2002,
‘Assessment of Technologies for screening, floatable control, and screening
handling’, Water Environment Research Foundation
157. Stelling, G.S. 1984, ‘On the construction of computational methods for
shallow water equations’, Rijkswaterstaat communication No. 35.
158. Stovin VR., 1996, ‘The prediction of sediment deposition in storage
chambers based on laboratory observations and numerical simulation’. Ph.D.
thesis, University of Sheffield
159. Stovin VR, Saul AJ., 1998, ‘A computational fluid dynamics (CFD) particle
tracking approach to efficiency prediction’ Water Science and Technology,
vol.37, no.1, pp.285–293.
160. Stovin, V., R., Saul, A., j., Drinkwater, A. & Clifforde,I.,1999,’ Field testing
CFD-based predictions of storage chamber gross solids separation
efficiency’, Water Science and Technology, vol.39, no.9,pp.161-168.
161. Stovin VR, & Saul AJ., 2000, ‘Computational fluid dynamics and the
design of sewage storage chamber’, J. CIWEM, vol.14, no.2, pp.103–110.
162. Sutton,R.S.,1984,‘Temporal credit assignment in reinforcement learning,’
8410337 Ph.D., University of Massachusetts Amherst, Ann Arbor.
163. Svejkovsky, K., & Saul, A.J., 1993, ‘Computational modeling of the
stormKing™ hydrodynamic separator using 3D mathematical model fluent’,
Department of Civil and Structural Engineering, University of Sheffield, UK.
164. Thinglas T. & Kaushal. D.R., 2008, ‘Three-Dimensional CFD Modeling for
Optimization of Invert Trap Configuration to Be Used in Sewer Solids
Management.’ Particulate Science and Technology, vol.26, pp.507-509.
165. Thomson. B., 2012, Combined sewer overflows (CSOs) and CSO
screens
166. Tyack, J.N., Hedges, P.D., & Smisson, R.P.M., 1992, ‘The Use of Sewage
Settling Velocity Grading in Combined Sewer Overflow Design’. NOVATECH
92, International Conference on Innovative Technologies in the Domain of
Urban Water Drainage, Lyon (France), November 3-5.
References
Md Abdul Aziz 178
167. Tyack, J. N. & Fenner, R. A., 1999, ‘Computational fluid dynamics
modelling of velocity profiles within a hydrodynamic separator’, Water
Science and Technology, vol.39, no.9, pp.169-176.
168. U.S. Army Corps of Engineers, Waterways Experiment Station,
Vicksburg, Miss., 1952, Corps of Engineers Hydraulic Design Criteria,
prepared for office of the chief of engineers, revised in subsequent years
169. US Army Corps of Engineers, 1970, Corps of Engineers Hydraulic Design
Criteria, prepared for office of the chief of engineers, waterways experiment
station, Vicksburg, MS, USA.
170. U.S. Army Corp of Engineers, 1990, ‘Hydraulic design of spillways’, EM
1110-2-1603, Dept. of the Army, Washington, D.C.
171. Versteeg, H.K. and W. Malalasekera, 2007, ‘An introduction to
Computational Fluid Dynamics.’ Second ed2007: Pearson Education Limited.
172. Werbos, P. 1994, ‘Beyond regression: new tools for prediction and
analysis in the behavioral science,’ PhD dissertation, Harvard University,
Cambridge, Mass.
173. Weyand, M., 2002, ‘Real-time control in combined sewer systems in
Germany-some case studies’, Journal of Urban Water, vol.4, pp.347-354.
174. Willems, P., & Berlamont, J., 1999, ‘Probabilistic modelling of sewer
system overflow emissions’, Water Sci. Tech. vol.39, no.9, pp.47-54.
175. Wu, H. & Shapiro, J. L., 2007, ‘Parameter cross-validation and early-
stopping in univariate marginal distribution algorithm,’ in Proceedings of the
9th annual conference on Genetic and evolutionary computation, ACM, pp.
632-633.
176. Yeh, K. C., & Tung, Y. K., 1993, ‘Uncertainty and sensitivity analyses of
pit-migration model’, Journal of Hydraulic Engineering, vol.119 no.2, pp. 262–
283.
177. Yulei, J., Zur, R. M., Pesce, L. L. & Drukker, K., 2009, ‘A study of the
effect of noise injection on the training of artificial neural networks,’ in
International Joint Conference on Neural Networks (IJCNN), pp. 1428-1432.
Appendix A: Experimental Data
Md Abdul Aziz 179
Appendix A:
Experimental
Data
Appendix A: Experimental Data
Md Abdul Aziz 180
Comb Separator testing program, Swinburne University of Technology
Run 5 Monday PM 9th August 2010
Test results and conclusions:
Test 1.
Crest length reduced to 510mm due to incomplete combs and backup.
3 No combs.
Spacing’s as in above diagram.
Flow 27L/s, equivalent to 53L/m/s. Cleared retention screen satisfactorily.
Test items: 20 No strips of toilet paper
20 No cigarette filters
No bottle tops
1 No drink can.
Test run commenced 3.07PM, completed 3.18PM.
Captured:
20 No strips of toilet paper or 100%
6 No cigarette filters or 20%
4 No bottle tops or 100%
1 No drink can or 100%.
Test 2
Maximum flow test 32.7L/s.
Appendix A: Experimental Data
Md Abdul Aziz 181
Conclusions:
Maximum attainable flow probably 35L/s before backing up drowns retention
screen.
Perfect results again for toilet paper, bottle tops and drink cans.
Slight blinding of the retention screen noted, about 5%.
The drink can readily passed over and down between the weir and first comb.
Cigarette filter capture rate still low.
Comments:
Maximum equivalent flow possible will be about 80L/m/s in present configuration,
or the once-a-year Vorthbach CSO overflow.
The present configuration is continuing to give excellent toilet paper capture
rates.
The cigarette filters will be enlarged to represent cigarette butts and retested but
may not result in a significant capture rate improvement.
Blinding of the retention screen is negligible.
Drink cans will be captured, but may not pass through the valve, unless valve
clearances are increased.
The retention screen was tilted forwards by 25mm without any problems, as the
flush-water chamber is full to above the exit weir crest level. It enhanced the
screen cleaning by the impacting nappe.
The retention screen, for some distance on both sides of the ball valve, should
be solid (no holes) for a height of about 100mm. This would improve flushing by
forcing flows less than 100mm deep, ie, towards the end of flushing, to pass
around the ends of the retention screen. This measure would obviate the need
for segment walls along the filtered water chamber and the possibility of solids
build-up on them.
The tilting forward of the retention screen allows for a larger and heavier ball to
be used (refer to above figure).
Future tests:
To improve cigarette butt capture, the spacing of the wires in the combs could be
reduced to 20mm centre to centre, thus reducing the effective gap from 3.5mm to 2mm.
This could be tried later and its effect on the nappe checked.
Appendix A: Experimental Data
Md Abdul Aziz 182
Conclusion:
There is little further progress to be made by laboratory testing alone, and the
development of the comb separator has reached the stage where testing in real
situations is essential.
Appendix A: Experimental Data
Md Abdul Aziz 183
Comb Separator testing program, Swinburne University of Technology
Present: Dr D Phillips; Mr A Aziz
Run 6 Monday PM 23rd August 2010
Test 1.
The comb screens were carefully straightened and aligned so as to accurately
overlap.
Crest across full length of test box ie, 970mm. Configuration as previously.
3 No full length combs straightened and fixed in position.
Spacing’s as in above diagram.
Flow 45L/s. Cleared retention screen satisfactorily.
Test items:
20 No cigarette filters
Test run commenced 12:20PM, completed 12:30PM.
Capture:
12 No cigarette filters or 60%
Comments:
The artificial butts were a single filter wrapped in a tape cover and smaller than actual
butts.
The flow now passes over the weir at right angles with the full length 970mm crest.
Test 2
Real 20 butts were collected at random and their mean diameter and lengths
measured and recorded. The mean of these were calculated as, width =
7.375mm and length, = 35.75mm.
Test flow 45L/s
Other conditions as per test 1
Appendix A: Experimental Data
Md Abdul Aziz 184
Test items
20 representative butts to the above mean dimensions were prepared
Test run commenced 2:18PM, completed 2:24PM
Capture rate:
17 captured, 3 passed or 85%
Test 3
Repeated the above Test 2 at 40L/s with a 75% capture efficiency.
This result is invalid due to rupture of valve housing seal, leading to loss of 4 cig butts.
Comments
Valve housing to be sealed off and tests repeated later.
Appendix A: Experimental Data
Md Abdul Aziz 185
Run 7 Wednesday 25th August 2010
The valve housing was sealed off, pressure tested, and found to be secure.
The interception screen was refixed with its top 150mm only, instead of 160mm
from the back of the weir.
30 plastic bottle tops of varying sizes were obtained from the recycling bins.
Test1
Commenced 4:54PM and finished at 5:06PM:
Flow-rate, 45L/s.
Test items Number in Number retained capture efficiency%
Dish wipes 20 20 100
Assorted bottle tops 20 20 100
Tampons 20 20 100
Balloons 20 20 100
Ersatz cig butts (t1) 20 11 55
Ersatz cig butts (t2) 9 6 67
Comments:
The capture efficiency for all of the above items is 100% with the exception of cigarette
butts that averaged about 61%.
The test for cigarette butts used relatively dry butts and so, to better represent actual
conditions, will be repeated after allowing them to soak for some time.
Blinding of the retention screen was zero with the full width weir crest nappe now
impacting along the length of the retention screen.
Bringing the top of the interception screen to within 150mm of the back of the weir, and
within 40mm of the last comb, did not appear to influence the results.
Appendix A: Experimental Data
Md Abdul Aziz 186
Future tests:
To improve cigarette butt capture, the spacing of the wires in the combs could be
reduced to 20mm centre to centre, thus reducing the effective gap from 3.5mm to 2mm.
As a first step the front comb could be so modified, as it operates in the sub-
critical zone and so should not affect the hydraulics of the nappe.
Conclusion:
The full weir produced overflow normal to the weir and maintained all captured
materials in slow motion in the holding chamber without any suggestion of adherence to
the retention screen. Some improvement in cigarette butt capture may result from prior
soaking.
Appendix A: Experimental Data
Md Abdul Aziz 187
Swinburne University of Technology
Comb Separator testing program
Wednesday PM 2nd September 2010
Present: Dr D Phillips; Mr A Aziz.
All these tests were conducted using the full length weir of 970mm and tree combs.
Test 1. Cigarette butts
Both real and substitute butts were soaked in water for 24 hrs. The 20 butts of
each type were then drained and collectively weighed.
The substitute butts were squeezed and trimmed until they weighed almost the
same as the real butts.
Test items: 20 No substitute cigarette filters
10 No real cotton balls
29 No real cotton buds
The latter two items were added at request of Herr M Simon.
Test run commenced 10:58AM, completed 11:30AM.
The ruler attached to the box read 65mm below the top of box so that the head on the
weir was 150-65 = 85mm. Francis formula gave an overflow of 45L/s. The flow meter
mass flow over one minute gave 2.56 cu m or 43L/s.
Air bubbles entrained in the flow caused major turbulence in the box. A test
showed that the air was entraining as the flow fell from the overhead tank.
Appendix A: Experimental Data
Md Abdul Aziz 188
Capture rates:
The technique adopted was to readmit to the test item port, all those items that
had escaped capture. This was repeated until all items had been captured.
Cigarette butts
1st admission; 12/20: 1st repeat7/8: 2nd repeat 1/1
Hence capture efficiency = 20/29 or 69 per cent.
Cotton balls
1st admission 10/10. Hence capture efficiency 100 percent.
Cotton buds
1st admission; 6/20: 1st repeat 9/14: 2nd repeat 3/5: 3rd repeat 2/2.
Hence capture efficiency = 20/41 or 49 per cent.
Comments:
The artificial butts were trimmed in length to match the weight of the real butts
and so were a little shorter than the 35.75mm representative length, contributing to their
escape. It was noted that the cotton buds were very buoyant and that the cotton balls
ended up as single cotton mass in the holding chamber.
Test 2
The flow was calculated as before as 45L/s and read on the meter as 41.5L/s.
Test items, 20 cotton buds and 20 cigarette butts. All had been left soaking in water
between tests.
Test procedure conducted as a previously.
Testing commenced at 1:15PM and finished around 1:30PM.
Appendix A: Experimental Data
Md Abdul Aziz 189
Capture rates:
Cigarette butts
1st admission; 14/20: 1st repeat3/6: 2nd repeat 3/3
Hence capture efficiency = 20/29 or 69 per cent.
Cotton buds
1st admission; 13/20: 1st repeat 6/7: 2nd repeat 1/1.
Hence capture efficiency = 20/28 or 71 per cent.
Comments:
It was noticeable that while five of the cigarette butts still floated, all of the cotton
buds remained highly buoyant.
The other observation was that while the cigarette butt capture efficiency
remained the same, the cotton bud capture both within and between the two tests
dramatically increased.
Test 3
For the third test; to better represent real world conditions, all 40 items were kept
submerged for 45 minutes and the 20 substitute butts squeezed under water to expel
trapped air. This reduced their buoyancy without affecting their weight.
For the third and final test of the current series, the following items were tested:
20 No wipes (toilet paper), 20 No tampons, 20 No balloons (condoms), 20 No drink
bottle tops (various sizes), 20 No cigarette butts (representative), 20 No cotton balls, 20
No cotton buds. The 140 items were pre-soaked.
The flow was again adjusted to 45L/m/s, the maximum currently possible.
Testing commenced at 2:15 and finished at 2:40PM.
Appendix A: Experimental Data
Md Abdul Aziz 190
The items were randomly added one at a time over about six minutes.
Test items No 1st cap 2nd cap. 3rd cap No repeats % trapped
Wipes 20 20 20/20 100
Tampons 20 20 20/20 100
Balloons 20 20 20/20 100
Assorted bottle tops 20 19 1 20/21 95
Subst. butts 20 14 6 20/26 77
Cotton balls 20 20 20/20 100
Cotton buds 20 16 1 3 20/27 74
Total 140/154 92.3
Comments:
As in previous tests, the capture efficiency for the wipes, tampons, balloons and
bottle tops was virtually 100 per cent.
The capture efficiencies for cigarette butts and cotton buds increased with
subsequent tests, presumably as they became saturated.
The overall capture efficiency of the above 140 items was 92.3% with that of
cigarette butts and cotton buds 77% and 74 % respectively.
The severe turbulence, due to the entrained air at one end of the overflow
chamber, tended to toss some of these lighter items over the weir, thus reducing their
chance of capture and so contributed to the lower, though acceptable, capture rates.
Blinding of the retention screen was negligible with the full width weir nappe
impacting along the full length of the retention screen.
Appendix A: Experimental Data
Md Abdul Aziz 191
Future testing:
This is the last test run pending further modifications to the rig and model. These
include a 100mm ball valve, a captured solids strainer basket at the tank and the drilling
and tapping of 3mm holes, 5mm apart, in two of the comb holders. This will allow the
testing of different comb wire spacing’s.
Conclusions:
The results of the latest series of full-length weir tests indicate that, at the tested
flow rate, the three-comb configuration produces high capture rates for all the items
tested.
The test flow-rate was similar to a typical once-yearly peak overflow for Europe.
Further testing will be conducted with a two-comb configuration and a varied per
metre flow-rate to optimize the performance and economy of the comb separator.
Appendix A: Experimental Data
Md Abdul Aziz 192
Comb Separator testing program, Swinburne University of Technology
Supplementary tests conducted Thursday 16th September 2010
Present: Dr D Phillips and Abdul Aziz
Test setup:
Two comb screens, wires 15mm centre to centre, combs 25mm centre to centre.
Front comb 75mm from crest of weir, or 63mm from back of weir.
Retention screen 120mmm behind back of weir. Concrete and brick blocks placed
on floor of overflow chamber to better distribute flow.
Crest full length of test box ie, 970mm. Configuration as per tests 09092010.
Test 1.
The 2 No comb screens were checked, accurately overlapped and fixed
in position.
Flow 15L/m/s. Cleared retention screen satisfactorily.
Test items:
20 No cigarette filters
20 No cotton Q buds
Test run commenced 11:48AM, Finished at 12:15PM.
Capture:
Cigarette butts
1st pass 2/20, 2nd pass 0/2. 4No butts remained in overflow chamber.
Hence efficiency = (20-4)/ (22-4) x100 = 91%.
Cotton Q buds
Appendix A: Experimental Data
Md Abdul Aziz 193
1st pass 2/20, 2nd pass 0/1. 1 discarded, 1 remained in chamber.
Hence efficiency = (20-1-1)/ (21-1-1) x100 = 95%.
Comments:
The discarded cotton bud had lost a bud, reducing it to a diameter of about 3mm
Test 2
Test flow 30L/m/s
Other conditions as per test 1
Test items
22 No cigarette butts
Test run commenced 1:30PM, Finished 1:45PM
Capture:
1st pass 4/22, 2nd pass 3/4, 3rd pass 0/3. 1 remained in overflow chamber
Hence efficiency = 21/ (22+4+3-1) x100 = 75%.
Test 3
Test flow 20L/m/s.
Other conditions as per Test 1
Test items
22 No cigarette butts
Test run commenced 1:57PM, Finished 2:15PM
Appendix A: Experimental Data
Md Abdul Aziz 194
Capture:
1st pass 4/22, 2nd pass 0/4,
Hence efficiency = 22/ (22+4) x100 = 85%.
Comments:
Capture efficiency of cotton Q buds was 95 per cent at 15L/m/s, the only flow
tested.
Capture efficiency of cigarette butts decreased with increasing flow.
At the Vorthbach MWSS six-monthly flow of 30L/m/s, the capture rate was 75
per cent, indicating that a two-comb setup would only achieve some 58 percent
capture at the once-annual overflow of 70L/m/s. (See graph below).
This appears inadequate and suggests that a three-comb setup would be
required to achieve acceptable capture rates at higher flows.
Proposal for further testing:
A three-comb set-up for testing cigarette butt capture at higher flows, consisting
of wires at 20m centre to centre within combs and combs spaced 20mm behind
one another. The effective gap would be 3.33mm compared to 4.50mm for the
above two comb arrangement.
Appendix A: Experimental Data
Md Abdul Aziz 195
Comb Separator testing program, Swinburne University of Technology
High flow 2-comb tests, Monday 27th September 2010
Present: Dr D Phillips; Mr A Aziz.
Aim:
To test the efficiency of the two-comb 15mm centre to centre set-up, for
cigarette butts at higher per metre flows based on extrapolation of the supplementary
test results 160902010.
Test setup:
Two comb screens, wires 15mm centre to centre, combs 25mm centre to centre.
Front comb 60mm from crest of weir, or 50mm from back of weir.
Retention screen 120mmm behind back of weir.
Concrete and brick blocks placed on floor of overflow chamber to distribute flow.
Crest length reduced to 460mm.
Test 1.
The 2 No comb-screens were checked, accurately overlapped and fixed in
position.
Head on weir 120mm giving a flow of 35L/s or an equivalent flow of 76.2L/m/s.
Nappe easily cleared retention screen.
Test items:
22 No cigarette artificial butts
12 No wipe clothes (toilet paper)
Test run commenced 11:43AM, Finished at 12:15PM.
Appendix A: Experimental Data
Md Abdul Aziz 196
Capture efficiencies:
Cigarette butts
1st pass 13/22, 2nd pass 3/13, 3rd pass 2/3, 4th pass 0/2. 1No butt not found.
Hence Ѯ = ((22-13)+(13-3)+(3-1)+(2-2)) / (22+13+3+2) = (21/40-1) x100
= 55%.
Wipe clothes
1st pass 12/12.
Hence efficiency = (12)/ (12) x100 = 100%.
Comments:
The missing butt may be stuck under a brick that was knocked over by flow. To
be checked.
Test 2
Head on weir 110mm giving equivalent flow of 67L/m/s
Other conditions as per Test 1
Test items
21 No cigarette butts
Test run commenced 1:05PM, Finished 1:25PM
Capture:
1st pass 11/21, 2nd pass 3/10, 3rd pass 2/3, 4th pass 0/2.
Hence Ѯ = 21/ (21+11+3+2+1) x100 = 57%.
Comments:
Capture efficiency of cloth wipes was 100 per cent at 76L/m/s, the only flow
tested, predicting perfect toilet paper capture efficiencies up to the Vorthbach Q1
overflow.
Appendix A: Experimental Data
Md Abdul Aziz 197
The tests confirmed that the capture efficiency of the 2-comb set-up for cigarette
butts decreases with increasing flow to unacceptably low values.
This allows the extrapolated plot of supplementary test results 160902010 to be
adjusted and completed for cigarette butt capture with a 2-comb set-up. (See
graph below).
This appears inadequate and suggests that a three-comb setup would be
required to achieve acceptable capture rates at higher flows.
Proposal for further testing:
A three-comb set-up for testing cigarette butt capture at higher
flows, consisting of wires at 20m centre to centre within combs and combs
spaced 20mm behind one another. The effective gap would be 3.33mm
compared to 4.50mm for the above two comb arrangement.
Appendix A: Experimental Data
Md Abdul Aziz 198
Comb Separator testing program, Swinburne University of Technology
3-comb tests, Monday 11th October 2010
Present: Dr D Phillips; Mr A Aziz.
Aim: To test the capture efficiency of three-combs for cigarette butts.
Test setup:
Three comb screens, wires 20mm centre to centre, combs 25mm centre to
centre.
Front comb 50mm from crest of weir, or 50mm from back of weir.
The comb screens were checked, accurately overlapped and fixed in position.
Retention screen 120mmm behind back of weir.
3rd comb 30mm in front of retention screen.
Crest length reduced to 470mm.
New, 100mm dia. ball valve installed
Test 1.
Head on weir 113mm giving an equivalent flow of 71.4L/m/s.
Nappe easily cleared retention screen.
Test items:
20 No cigarette artificial butts
Test run commenced 3:15PM, Finished at 3:30PM.
Capture efficiencies:
1st pass 9/20, 2nd pass 5/9, 3rd pass 4/5, 4th pass 1/4.
Hence Ѯ = 19/38 = 50%.
Comments:
Most of the day was spent sealing the new valve housing, leaks around model
sewer chamber and fixing retention screen after causing initial test to be abandoned.
Test 2
Head on weir 80mm giving equivalent flow of 41.5L/m/s
Other conditions as per Test 1
Appendix A: Experimental Data
Md Abdul Aziz 199
Test items
18 No cigarette butts
Test run commenced 4:00PM, Finished 4:15PM
Capture:
1st pass 8/18, 2nd pass 5/8, 3rd pass 2/5, 4th pass 1/2.
Hence Ѯ = (17/33) x100 = 52%.
Test 3
Head on weir 65mm giving equivalent flow of 30L/m/s
Other conditions as per Tests 1 and 2.
Test items
18 No cigarette butts
Test run commenced 4:25PM, Finished 4:35PM
Capture:
1st pass 4/16, 2nd pass 1/4.
Hence Ѯ = (15/20) x100 = 75%.
Comments:
Two butts remained in model sewer chamber due to the low flow.
Conclusions:
The tests surprisingly showed that despite using three 20mm centre to centre
combs, the capture efficiency of cigarette butts remained unacceptably low.
The tests confirmed previous observations that capture efficiency decreases
with increasing flow rate.
Further testing:
The unexpectedly poor results showed that the mean gap of 3.67mm was
ineffective at intercepting 7mm dia. cigarette butts, suggesting that another
approach to this problem is needed.
The next tests will return to a two comb set-up with more closely spaced wires.
Appendix A: Experimental Data
Md Abdul Aziz 200
Comb Separator testing program, Swinburne University of Technology
2 comb tests, Wednesday 13th October 2010
Present: Dr D Phillips, Dr M Imteaz, Herr M Simon, Mr A Aziz.
Aim: To test the cigarette butt capture efficiency of a two-comb close-wire
arrangement.
Test setup:
Two overlapped comb screens, 1st with wires 12.5mm centre to centre, 2nd
with wires 15mm centre to centre, combs 20mm centre to centre.
Front comb 70mm from crest of weir.
Retention screen 140mmm behind crest of weir.
Concrete block placed on floor of model sewer chamber to distribute flow.
Crest length reduced to 470mm.
Retention screen screwed to floor.
Test 1.
Head on weir 50mm giving a flow of 20L/m/s.
Nappe easily cleared retention screen.
Test items:
18 No artificial cigarette butts
20 No wipe clothes (toilet paper)+ 3 quarter pieces
10 tampons
10 bottle tops
19 cotton buds
Test run commenced 11:26AM, Finished at 11:45PM.
Capture efficiencies:
Cigarette butts
1st pass 1/16, 2No butt in overflow chamber. Hence Ѯ = (15/16) x100 = 94%.
Wipe clothes
1st pass 0/23. Hence Ѯ = (23/23) x100 = 100%.
Tampons
Appendix A: Experimental Data
Md Abdul Aziz 201
1st pass 0/10. Hence Ѯ = (10/10) x100 = 100%.
Bottle tops
1st pass 0/10. Hence Ѯ = (10/10) x100 = 100%.
Cotton buds
1st pass 6/19. Hence Ѯ = (13/23) x100 = 63%.
Comments:
The high efficiencies of both the larger and smaller items are consistent with previous
results.
Actual sheets of toilet paper were tested with virtually 100 per cent captured. The paper
tended to disintegrate in the holding chamber but readily passed to the outlet valve.
Test 2
Head on weir 90mm giving equivalent flow of 50L/m/s
Other conditions as per Test 1
Test items:
18 No artificial cigarette butts
20 No quarter pieces of wipe cloth
10 tampons
10 bottle tops
Test run commenced 12:12PM to 12.23PM, Finished 12:35PM
Capture efficiencies:
Wipes, 1st pass 0/20. Hence Ѯ = (20/20) x100 = 100%.
Tampons, 1st pass 1/10. Hence Ѯ = (9/ 10) x100 = 90%.
Bottle tops, 1st pass 0/10. Hence Ѯ = (10/10) x100 = 100%.
The flow was increased to 58L/m/s with a head on the weir of100mm.
Cigarette butts
1st pass 7/15, 3No butts in model sewer chamber. Hence Ѯ = (8/15) x100 = 53%.
Comments:
Appendix A: Experimental Data
Md Abdul Aziz 202
Capture efficiency of cloth wipes, tampons and bottle tops was very high but
that of cigarette butts was low and the same as for the previous tests 11102010 at the
same flow.
Test 3:
The combs were reversed with the 15mm spaced comb 60mm from weir crest
and the 12.5 mm spaced comb 15mm further behind.
Head on weir 110mm giving an equivalent flow of 67L/m/s
Other conditions as per Tests 1 and 2.
Test items:
20 No artificial cigarette butts
18 No quarter pieces of wipe cloth
10 cotton buds
Test run commenced 1:25PM, Finished 1:35PM
Capture:
Cigarette butts, 1st pass 8/20, 2nd pass 2/8, 3rd pass 2/2, 4th pass 1/2.
Hence Ѯ = (19/32) x100 = 59%.
Wipes, 1st pass 2/18, 2nd pass 0/2. Hence Ѯ = (18/20) x100 = 90%.
Cotton buds, 1st pass 7/10, 2nd pass 5/7, 3rd pass 4/5. Hence Ѯ = (6/22) x100 =
27%.
The flow was then reduced to determine the minimum overflow to pass over the
retention screen and was found to be about 15L/m/s at an overflow depth of 40mm.
The retention screen crest was 440mm below the weir crest and 140mm downstream
of it.
Conclusions:
Cigarette butt capture improved by reversing the screens but remains
unacceptably low at higher overflows.
Cotton bud capture is also very low at high overflows.
It was observed that the nappe profile was unaffected by the closer spacing of
the wires in the two combs.
Appendix A: Experimental Data
Md Abdul Aziz 203
Real toilet paper capture is virtually 100 per cent but disintegrates and binds
with other captured solids.
The various comb arrangements trialled to-date are all highly effective at
intercepting the larger common sewer solids, but are less effective with thin items such
as cigarette butts and cotton Q buds.
Proposal for further testing:
From the above tests, the overflow nappe did not appear to be effected by the
closely spaced wires of the two comb arrangement. Hence it appears feasible to
reduce the wire spacing to 9mm, leaving a 6mm clear gap to physically intercept
cigarette butts and cotton Q buds.
A possible arrangement would be to locate this comb immediately downstream
of two 25mm centre to centre wire spaced combs. These would intercept and remove
the larger items that could otherwise foul, or blind the 9mm centre to centre wire
spaced comb.
Appendix A: Experimental Data
Md Abdul Aziz 204
Comb Separator testing program, Swinburne University of Technology
Cigarette butt tests, Wednesday 20th October 2010
Present: Dr D Phillips, Mr A Aziz.
Aim: To test the cigarette butt capture efficiency of a two-comb arrangement.
Test setup:
Two overlapped comb screens, 1st with wires 25mm centre to centre, 2nd with
wires 10mm centre to centre, combs 20mm centre to centre.
Front comb 65mm from crest of weir.
Retention screen 140mmm behind crest of weir.
Concrete block placed on floor of model sewer chamber to distribute flow.
Crest length reduced to 470mm.
Retention screen screwed to floor.
19 No butts wrapped in duct tape giving mean sample diameter of 8.82 mm.
Test 1.
Head on weir 50mm giving a flow of 20L/m/s.
Nappe easily cleared retention screen.
Test items:
10 No artificial cigarette butts
Test run commenced 11:50AM, Finished at 12:05PM.
Capture efficiencies:
Cigarette butts
1st pass, 1/10, 2nd pass 0/1. 2No butts in down pipe. Hence Ѯ = (8/9) x100 = 89%.
Comments:
Appendix A: Experimental Data
Md Abdul Aziz 205
It was intended to close the bar spacing of comb 2 to 9mm but time constraints
prevented this for today’s tests. To offset this, the butts were wrapped in duct tape to
increase their diameters accordingly so as to simulate accurately the closer bar spacing.
Test 2
Head on weir 100mm giving equivalent flow of 58L/m/s
Other conditions as per Test 1
Test items:
19 No artificial wrapped cigarette butts
9 No real butts, mean diameter 8.0mm
Test run commenced 12:35PM. Finished 12:55PM
Capture efficiencies:
Wrapped cigarette butts
1st pass, 0/11, 8No butts in model sewer chamber Hence Ѯ = (11/11) x100 = 100%.
Real cigarette butts:
1st pass 5/9, 2nd pass, 2/5, 3rd pass, ½ 4th pass, 1/1 Hence Ѯ = (8/17) x100 = 47%.
Comments:
Capture efficiency of wrapped butts affected by low number that passed over
weir, possibly as they were heavier with the extra wrapping. Real cigarette butt
capture was low. This was because they were not all physically intercepted by
the combs, as some were less than7mm in one dimension, having been
squashed underfoot by the smoker.
Test 3:
Head on weir 110mm giving an equivalent flow of 67L/m/s
Other conditions as per Tests 1 and 2.
Test items:
19 No artificial cigarette butts
20 No quarter pieces of wipe cloth
10 No full pieces of wipe cloth.
Appendix A: Experimental Data
Md Abdul Aziz 206
Test run commenced 2:58PM, Finished 3:25PM
Capture:
Cigarette butts, 1st pass 3/19, 2nd pass 1/3, 3rd pass 0/1.
Hence Ѯ = (19/23) x100 = 83%.
¼ Wipes, 1st pass 1/20, 2nd pass 0/1. 1 on 2nd comb. Hence Ѯ = (20/21) x100 =
95.2%. The wipe caught on the 2nd comb included as captured.
Full wipes, 1st pass 0/10. Hence Ѯ = (10/10) x100 = 100%.
Overall wipes capture efficiency = 96.8%.
Comments: Wrapped cigarette but capture acceptable while wipes capture,
representing different sized toilet sheets, was very good.
Conclusions:
Cigarette butt capture rates are now satisfactory but tests using a 9mm wire
spacing needed to confirm the above results.
Two 25mm combs are needed in front of the 9mm comb to prevent matting.
It was observed that the nappe profile was little affected by the close wire spacing
of the second comb.
Hence a 9mm comb, having a mean clear spacing of 6mm should intercept
virtually all cigarette butts.
Proposal for further testing:
Test for cigarette butt capture using two 25mm and one 9mm combs. This is the
optimum setting developed from the experimental set ups. Further updates in this regard
are included in Chapter 4.