states
nental Protection
Municipal Environmental Research
Laboratory
Cincinnati OH 45268
EPA-600/2-80-005
May 1 980
as. i and Development
Evaluation of
Full-Scale Tertiary
Wastewater Filters
>
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development, U S Environmental
Protection Agency, have been grouped into nine series These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are
1 Environmental Health Effects Research
2 Environmental Protection Technology
3 Ecological Research
4 Environmental Monitoring
5 Socioeconomic Environmental Studies
6 Scientific and Technical Assessment Reports (STAR)
7 interagency Energy-Environment Research and Development
8 "Special" Reports
9 Miscellaneous Reports
This report has been assigned to the ENVIRONMENTAL PROTECTION TECH-
NOLOGY series This series describes research performed to develop and dem-
onstrate instrumentation, equipment, and methodology to repair or prevent en-
vironmental degradation from point and non-point sources of pollution This work
provides the new or improved technology required for the control and treatment
of pollution sources to meet environmental quality standards
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
FOREWORD
The Environmental Protection Agency was created because of increasing
public and government concern about the dangers of pollution to the health
and welfare of the American people. Noxious air, foul water, and spoiled
land are tragic testimony to the deterioration of our natural environment.
The complexity of that environment and the interplay between its components
require a concentrated and integrated attack on the problem.
Research and development is that necessary first step in problem solu-
tion and it involves defining the problem, measuring its impact and search-
ing for solutions. The Municipal Environmental Research Laboratory develops
new and improved technology and systems for the prevention, treatment, and
management of wastewater and solid and hazardous waste pollutant discharges
from municipal and community sources, for the preservation and treatment of
public drinking water supplies, and to minimize the adverse economic, social,
health, and aesthetic effects of pollution. This publication is one of the
products of that research; a most vital communications link between the
researcher and the user community.
Conventional methods for treatment of municipal wastewaters frequently
produce effluents that will not meet local discharge requirements. Granular
media filters are being installed to provide tertiary treatment for increased
removals of suspended solids and BOD. This report provides performance data
for full-scale tertiary filters from eight treatment plants and discusses
effects of various design and operating practices. Semi-empirical mathemati-
cal models in the report relate filter clarification efficiency to character-
istics of secondary effluent particulate matter and to filter operating para-
meters.
Francis T. Mayo, Director
Municipal Environmental Research
Laboratory
iii
-------
ABSTRACT
The clarification efficiency of eight full scale tertiary granular media
filters was characterized using conventional design and operating parameters,
such as influent suspended solids and flow rate, and properties of secondary
effluent suspensions (refiltration, particle size). Clarification efficiency
is only weakly dependent on filter media depth and media grain size. Better
correlations of filter clarification efficiency have been obtained for grab
compared to composite sample data. Composite process variables generally fit
a log normal distribution best and normal distribution second best.
Refiltration parameters were shown to have a very strong correlation with
filter performance for plants where straining filtration was believed to be
the dominant particle collection mode. Semiempirical mathematical models were
developed to characterize "in-depth" and "straining" filtration. If in-depth
filtration was the dominant particle collection mode, inclusion of refiltra-
tion parameters did not improve model prediction. The models developed in
this study may be applied with some caution to predict filter suspended solids
removal or clarification efficiency without pilot-scale tests.
An important design consideration for small scale tertiary wastewater
filters was found to be the ability to handle shock loads caused by secondary
process upsets. This consideration generally favors those designs with slow
rate of headloss development.
This report was submitted in fulfillment of Grant Number R803212 by North-
western University under the partial sponsorship of the Environmental Protec-
tion Agency. This report covers a period from November 9, 1974 - August 18,
1978 and work was completed as of August 1977.
iv
-------
TABLE OF CONTENTS
ABSTRACT iv
LIST OF FIGURES vii
LIST OF TABLES
LIST OF SYMBOLS
ACKNOWLEDGEMENTS
I. INTRODUCTION 1
II. CONCLUSIONS 3
III. RECOMMENDATIONS 6
IV. EXPERIMENTAL PROGRAM AND PROCEDURES 7
Plant Descriptions 7
Secondary Plants 7
Filtration Plant Designs 7
Unique Features of Plants 11
Phases of Study 13
Sampling Equipment and Field Procedures 14
Laboratory Procedures 17
V. PERFORMANCE DATA AND OBSERVATIONS OF PLANT OPERATIONS 23
Removal of BOD and Suspended Solids 23
Design Considerations 27
VI. CLARIFICATION EFFICIENCY MODELS FOR GRANULAR
BED FILTRATION 34
Introduction 34
Current Wastewater Filter Design 34
Particle Collection Modes 34
Suspension Properties 40
Model Development 42
Characterization of Straining and In-Depth
Filtration for Full-Scale Filters 43
VII. FREQUENCY DISTRIBUTIONS OF SECONDARY AND FILTER
EFFLUENT PARAMETERS 45
Introduction 45
Statistical Analyses 45
Analysis of Addison North and South Data 47
Statistical Analyses for All Phase 1 Data • 61
Discussion 65
-------
VIII. RELATIONSHIP OF FILTER PERFORMANCE TO CONVENTIONAL
OPERATING PARAMETERS--PHASE 1 DATA 69
Introduction 69
Relationship to Secondary Plant Operational Variables 69
Analysis of Average Performance Data 69
Analysis of Individual plant Data 77
Techniques for Segmenting Individual Data Sets 80
Headloss and Solids Capture Ratios 91
Development and Testing of Operational Mathematical
Models—Phase 1 Data 94
Discussion 105
IX. RELATIONSHIPS BETWEEN MEASURES OF CHARACTERISTICS
OF SECONDARY EFFLUENT SUSPENDED SOLIDS AND
THEIR EFFECTS ON FILTER PERFORMANCE 113
Introduction 113
Relationship Between Measured Suspension Characteristics 113
Bivariate Relationships of Filter Performance to
Solids Loading and Suspension Characteristics
Discussion 151
X. DEVELOPMENT AND TESTING OF OPERATIONAL MATHEMATICAL
MODELS--PHASES 2, 3, AND 4 152
Introduction 152
Determination of Filtration Mode 152
Development of Operational Mathematical Models 161
Testing Models With Other Data 166
Model Selection and Application 171
LITERATURE CITED 178
APPENDICES 182
A. BIVARIATE CORRELATION COEFFICIENTS 182
B. CALCULATION OF EQUIVALENT MEDIA DIAMETER 186
C. UNITS FOR REFILTRATION AND STRAINING PARAMETERS. 188
D. SURVEY OF U.S. TERTIARY FILTERS 190
vi
-------
LIST OF FIGURES
Figure
1
2
3
4
5
6
7
Filter Design ...... ...«
1/3
Refiltration Date Plot - VC vs. (V/t)
Filter Effluent vs. Influent Soluble BOD -- Addison
Suspended Solids Concentrations vs. Backwash Cycle Time . . .
Cumulative Frequency Distribution of Filter Flow Rate--
Page
12
20
21
25
29
30
Addison North Plant, A) Log Ordinate, B) Arithmetic 50
8 Cumulative Frequency Distribution of Total Suspended
Solids--Addison North Plant, A) Log Ordinate,
B) Arithmetic .......... 51
9 Cumulative Frequency Distribution of Total BOD--
Addison North Plant, A) Log Ordinate, B) Arithmetic 52
10 Cumulative Frequency Distribution of Soluble BOD--
Addison North Plant, A) Log Ordinate, B) Arithmetic 53
11 Cumulative Frequency Distribution of Filter Flow Rate--
Addison South Plant, A) Log Ordinate, B) Arithmetic 54
12 Cumulative Frequency Distribution of Total Suspended
Solids—Addison South Plant, A) Log Ordinate,
B) Arithmetic 55
13 Cumulative Frequency Distribution of Total BOD—
Addison South Plant, A) Log Ordinate, B) Arithmetic 56
14 Cumulative Frequency Distribution of Soluble BOD--
Addison South Plant, A) Log Ordinate, B) Arithmetic 57
15 Relative Frequency Histograms for C and C—Addison
North and South Plants 60
vii
-------
LIST OF FIGURES (Continued)
Figure
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Daily Changes in C and C --Addison South Plant (#2). ....
C/C vs. Food to Microorganism Ratio and Sludge
C vs. Food to Microorganism Ratio and Sludge
Average C/C vs. Media Depth and Equivalent
o
C vs. Solids Loading — Addison South, Phase 1. ........
Cumulative Backwashes by Headloss vs. Solids
Cumulative Backwashes by Headloss vs. Solids
Loading and Time--Addison South, Phase 1 .....
Cumulative Headloss vs. Solids Loading and Time —
C vs. Headloss and Solids Capture/Headloss--
Lake Zurich, Phase 1
C/C vs. Headloss and Solids Capture/Headloss —
Calculated vs. Measured C--Addison South Model,
Page
67
70
71
72
74
75
76
78
81
82
83
84
85
86
87
88
90
95
96
101
viii
-------
LIST OF FIGURES (Continued)
Figure Page
36 Calculated vs. Measured C--Lisle Model, Lisle Data ....... 102
37 Calculated vs. Measured C--Lake Zurich and
Marionbrook. .............. ........ .... 103
38 Calculated vs. Measured C--Addison South Model,
Marionbrook Data ........... ............. 106
39 Calculated vs. Measured C — Lake Zurich Model,
Addison South Data ....... ...... ......... .
40 Calculated vs. Measured C--Lisle Model, Lake
Zurich Data ......... ...... ............ 108
41 Calculated vs. Measured C- -Marionbrook Model,
Lisle Data ........... ................ 109
42 Calculated vs. Measured C- -Marionbrook Model,
Hsiung's Data. ............ ............. HQ
43 Calculated vs. Measured C- -Marionbrook Model,
Published Data ..... .................... Ill
44 Particle Size Histogram — Addison South, Phases 3 and 4 ..... 115
45 Particle Size Histogram — Des Plaines River,
Phases 3 and 4 ..... .... ................ 116
46 Particle Size Histogram — Addison North and
Barrington, Phase 4 ...... . ................ 117
47 Particle Size Histogram- -Lake Zurich and Lisle,
Phase 4. . ........................... 118
48 Particle Size Histogram--Marionbrook and Romeo-
ville, Phase 4 ....... . ................. 119
49 R and E vs. a — Addison South and Des Plaines
River, Phase 9? ........................ 121
50 R vs. a --Phase 4 ....................... 122
nm
51 E vs. a —Phase 4 ....................... 123
nm
52 R vs. C --Hsiung's Data, Phases 2 and 3 ...... , ..... 125
o
53 R vs. C --Phase 4 ....................... 126
o
ix
-------
LIST OF FIGURES (Continued)
Figure iS£§
54 d and a vs. C —Addison South, Phase 4 127
nm nm o
55 E vs. C --Hsiung's Data, Phases 2 and 3 128
o
56 E vs. C --Phase 4 129
o
57 Effluent Suspended Solids vs. Solids Loading—
Addison South, Phases 2 and 3 132
58 Effluent Suspended Solids vs. Headless and
Solids Loading--Des Plaines River, Phase 3 133
59 Effluent Suspended Solids vs. Solids Loading—
Addison North, Barrington, Des Plaines River
and Romeoville, Phase 4 135
60 Effluent Suspended Solids vs. Solids Loading—
Addison South, Lake Zurich, Lisle, and Marion-
brook, Phase 4 I36
61 C/Qj vs. Particle Size—Addison South and
Des Plaines River, Phase 3. 137
62 C/Co vs. Particle Size—Addison North, Barrington,
Des Plaines River, and Romeoville, Phase 4 138
63 C/C vs. Particle Size—Addison South, Lake
Zurich, Lisle and Marionbrook, Phase 4 139
64 C/C (log) vs. R—Hsiung's Data, Addison South,
Phase 2 141
65 C/CQ {l°g) vs. R—Addison South and Des Plaines
River, Phase 3 ....... 143
66 C/C (log) vs. R—Addison North, Barrington,
Des°Plaines River, and Romeoville, Phase 4 144
67 C/C (log) vs. R--Addison South, Phase 4 145
68 C/C (log) vs. E--Hsiung's Data, Addison South,
Phase 2 146
69 C/C (log) vs. E—Addison South and Des Plaines
River, Phase 3 147
70 C/C (log) vs. E—Addison North, Barrington,
Des°Plaines River, and Romeoville, Phase 4 148
x
-------
LIST OF FIGURES (Continued)
Figure Page
71 C/C (log) vs. E — Addison South, Phase 4. . . ........ 149
72 Pore Blocking and Non-Blocking Regions in
Filtration of Stabilized Solids „„..,,.,, .0000... 153
73 L°810 (dnm/d50) VS' L°S10 J^"^ase 4. ........... 154
74 C/C (Log) vs. Log Particle Number Flux (J )--
4o 000°. .0000.0. 00. ..ooooooooo 156
75 Calculated vs. Measured C--Marionbrook In-Depth Filtration
Model, (Eqtn 22) Phase 4 Data. ................ 157
76 Normalized Data Plots--Addison North, Addison
South, Des Plaines River, and Romeoville ... ........ 159
77 Normalized Data Plots--Barrington, Lake Zurich,
Lisle, and Marionbrook .................... 160
78 Calculated vs. Measured C — Models for High and
Low Ncim}, Hsiung's Data. ................... 168
~. blK
79 Calculated vs. Measured C--Models for High and
Low E, Hsiung's Data ....<, . .'...... ........ 169
80 Calculated vs. Measured C--Addison South and Addison
North, Des Plaines River, and Romeoville Models;
Hsiung's Data ..... .................... 17°
81 Calculated vs. Measured C--Model for Hsiung's
Data, High E Data. ...................... 172
82 Calculated vs. Measured C--Model for Hsiung's
Data, Addison South, Addison North, Des Plaines
River, and Romeoville Data ............. ..... 173
83 Calculated vs. Measured C--Addison North, Des
Plaines River, and Romeoville Model, Addison
South Data .......................... 174
84 Calculated vs. Measured C- -Addison South Straining
Model, Addison North, Des Plaines River, and
Romeoville Data ........................ 175
xi
-------
LIST OF TABLES
Number Page
1 Unit Process Design Data ........ 8
2 Filtration Plant Characteristics ..... 9
3 Filtration Plant Media Specifications 10
4 Chronological Summary of Sampling Program
and Equipment Used 16
5 Statistics for BOD Data. ....... 24
6 Summary of Plant Performance Data. ..... . 26
7 Power Function Relationships for Filter
Coefficient—After Wright et al. (1970) 38
8 Power Function Relationships for Filter
Coefficient—After FitzPatrick (1972) 39
9 Standard Probability Models. 46
10 Summary of Simple Statistics for Addison
North and South Plants 49
11 Goodness of Fit Summary—Addison North
and South Plants 58
12 Empirical Fit of Bimodal Data 62
13 Summary of Simple Statistics, Phase 1 Data 63
14 Kolmogorov-Smirnov Goodness of Fit Summary--
Phase 1 64
15 Autocorrelation Coefficients--Phase 1 66
16 Bivariate Correlation Coefficients for
Plant Performance Data 79
xii
-------
LIST OF TABLES (Continued)
Number Page
17 Addison North and South Plants--Mean Performance
Data and Bivariate Correlation Coefficients for
Segmenting Data ..... ................. 89
18 Addison South Plant--Separation of Data Sets for
Low and High Solids Capture Ratios, Regression
Coefficients and Mean Performance Data ........... 92
19 Headless and Solids Capture Data, Phase 1 ......... 93
20 Regression Coefficients, All Phase 1 Data ......... 97
21 Regression Coefficients, Variable Flow
Plants, Phase 1 ............. . ........ 98
22 Regression Coefficients for C = a C Q ,
Phase 1 Data .......... ,° ............ 100
23 Testing of Regression Equations of Form
C = a Cb QC, Phase 1. . ...... . ........... 104
24 Summary of Particle Size and Refiltration Data ....... 114
25 Bivariate Correlation Coefficients for Filter
Performance, Hsiung's Data, Phases 2, 3 and 4 ....... 131
26 Regression Coefficients for C = a C Q° R G&
and C = a C^ Q'26 Rd G6 ..... . ............ 150
27 Segmented Data for High and Low N , C , and E ...... 162
O
28 Regression Coefficients for Log C/C =
f(Q, Cfl, E, d5Q, d^, d^/d^) -Phase 4 .......... 163
29 Regression Coefficients for Models and Intercepts
and Slopes for Plots of Calculated vs. Measured C ..... 165
30 Intercepts and Slopes for Plots of Calculated vs.
Measured C for Testing Models with Independent Data .... 167
xiii
-------
LIST OF SYMBOLS
_0
C -- Filter effluent suspended solids ML
CQ -- Filter influent suspended solids ML"3
C/C- -- Ratio of effluent to influent suspended solids
Cb -- Filter effluent BOD ML"3
C0b -' Filter influent BOD ML"3
C0Q -- Filter solids loading ML"2!'1-
BOD -- 5-day biochemical oxygen demand ML"3
de -- Equivalent media diameter L
dm -- Media diameter L
~~ Number mean suspended particle L
-- Suspended particle size L
-- Effective media size L
-- Geometric mean media diameter L
E -- Refiltration parameter Ml/6Ll/2T-l/6
G -- Refiltration parameter LT^'3
JNQ "" Particle flux to filter, particles/cm2/sec L"2!"1
k -- Parameter of negative binomial distribution
L -- Filter depth L
mx -- arithmetic mean
mx -- geometric mean
NsTR ~~ Dimensional straining parameter L"-'
Q -- Superficial filtration rate LT"1
r -- Bivariate or multiple correlation coefficient
ra -- Autocorrelation coefficient
rt -- Multiple correlation coefficient for log
transformation regression
R -- Refiltration parameter M
RG -- Refiltration parameter--product of R and G MLT'l/3
RL -- Run length T
Sc -- Ibs solids captured/ft2 filter surface area/ft
headless buildup /ft bed depth M L~2
Sch -- Ibs solids captured/ft2 filter surface area/ft ML~^
headloss buildup
SS -- Suspended solids ML"3
t -- Refiltration time T
T -- Temperature °F
V -- Volume of filtrate L3
VCO -- Weight of solids retained on filter paper in
refiltration test M
X5Q "~ Median value of distribution
Z -- Kolmogorov-Smirnov parameter (goodness of fit test)
xiv
-------
LIST OF SYMBOLS (Continued)
Greek
-- Proportional to
-- Filter coefficient L"1
-- Initial filter coefficient IT1
-- Liquid viscosity -- centipoise M
-- Standard deviation
-- Geometric standard deviation
-- Standard deviation of number mean particle size L
-- Specific deposit- -volume solids/volume filter
— Chi square parameter (goodness of fit test)
-- Parameters with indicates arithmetic average,
e.g. Q, C/GO
Filtration Plant Codes
AN -- Addison North
AS — Addison South (Filter No. 2)
AV -- Addison South (Filter Nos. 4,5,6,7)--variable flow
studies
BA -- Barrington
DP -- Des Plaines River
LZ -- Lake Zurich
LI -- Lisle
MB -- Marionbrook
MBD -- Marionbrook- -Dry weather flow, Phase 1
RO -- Romeoville
--L -- The third letter L, M, or H indicates a variable flow
plant and low, medium or high flow rate
— 2 -- A number following the code indicates the phase of the study
xv
-------
ACKNOWLEDGEMENTS
This study was conducted at the Technological Institute, Northwestern
University, Evanston, Illinois, in part through funds made available by
Research Grant No. R803312, Office of Research and Monitoring, U.S. Environ-
mental Protection Agency.
The report was prepared by Joseph A. FitzPatrick and Charles L. Swanson.
The assistance of Dominic Delle Grazie and T. Felten in construction and
installation of field sampling equipment, and the cooperation of treatment
plant superintendents and staffs, is acknowledged.
In addition, assistance of several graduate students in Civil Engineering
at Northwestern University is acknowledged for studies of hydraulic variables
(Scott Springer), secondary treatment and particle properties (Fred Leone),
pilot-full scale comparisons (Bob Minix) and particle properties and filter
performance correlation (Bill Ng). Results of these studies contributed to
the findings reported herein. Financial assistance from Brunswick Foundation
for student support in the latter phase of this project is also acknowledged.
A special note of thanks are due Wesly 0. Pipes, Drew L. Betz Professor
of Ecology, Drexel University, for his able assistance in the preliminary
aspects of this study and assistance in the statistical interpretation of
the data (especially in chapter VII).
The assistance of Jackie P. Headley in typing and preparing this manu-
script is appreciated.
xvi
-------
CHAPTER I
INTRODUCTION
Filtration is considered to be the most important tertiary process in
the implementation of the Federal Water Pollution Control Act Amendments of
1972 and 1977. Many existing secondary wastewater treatment plants cannot
meet the minimum monthly average effluent standard of 30 mg/JL for suspended
solids and biochemical oxygen demand (BOD) established by the Environmental
Protection Agency in 1973. This is particularly true for small plants where
removal efficiency is usually less than that achieved for large plants. The
addition of tertiary filters will enable many plants to meet the standard and
higher treatment levels for water quality limited streams. A survey con-
ducted in 1974 by one of us (J.A.FitzPatrick) indicated only 77 operating U.S.
tertiary filters treating secondary effluents greater than 0.3 mgd (Appendix
D). Based on tertiary filtration needs reported by Lykins and Smith (1974)
over 1,500 plants will be required to achieve water quality standards es-
tablished by the 1972 Act. Approximately 94 percent of plants will be
smaller than 5 million gallons per day (mgd) and 80 percent will be smaller
than 1 mgd. An equivalent number of plants will be required in the future
to meet the goals of the 1972 Act.
Research on tertiary wastewater filtration in the United States began
in the mid-19601s. Since then numerous pilot-scale studies have been con-
ducted to evaluate the effect of media size, media depth, and flow rate on
process performance. It is generally recommended that pilot-scale studies
be conducted to develop design parameters for full-scale plants since fil-
tration models cannot be used to reliably predict performance (Cleasby and
Baumann, 1974). The classical mechanisms for in-depth filtration, that were
developed from studies of potable water filtration and from laboratory
studies using suspensions with controlled properties, suggest a logarithmic
removal of particles with depth, but do not adequately describe suspended
solids removal for direct filtration of secondary effluents because greater
than logarithmic removal of solids occurs due to straining at or near the
surface of the filter medium. In addition, many of the models were developed
with studies on unflocculated suspensions and describe only initial or clean
bed filtration. In previous wastewater filtration studies, insufficient
attention has been given to the effect of secondary effluent suspension pro-
perties on particle collection mode and clarification efficiency.
The objectives of this study have been to gain a clearer understanding
of parameters affecting the clarification efficiency of full-scale granular
media wastewater filters and to develop operational mathematical models to
describe primarily this clarification efficiency. A four-phase study was
conducted in the Chicago metropolitan area at eight plants with design flows
-------
of 0.8 to 2.5 mgd. At four plants the flow to three filters was varied from
approximately 0.5 to 2 times average flow.
The objectives of Phase 1 were to characterize clarification efficiency
as a function of conventional design and operating parameters and to deter-
mine the statistical properties of secondary effluent and filter effluent
parameters. For this phase, all water quality data were obtained from flow-
proportioned composite samples of secondary effluent and filter effluent. No
routine tests were conducted on the properties of influent suspensions. Re-
sults of this phase are described in Chapters VII and VIII.
The results of Phases 2, 3 and 4 are given in Chapters IX and X. The
objective of these phases was to refine prediction of performance by
characterization of properties of filter influent suspensions. Phase 2 was
conducted at the Addison South Plant where composite samples were obtained
for refiltration tests. During Phase 3, grab samples were obtained at the
Addison South and Des Plaines River Plants and particle size distributions
were determined in addition to refiltration tests. Grab samples were ob-
tained at all eight plants during the last phase and analyzed as in Phase 3.
Pertinent filtration literature is reviewed in Chapter VI to provide
the basis for development of operational mathematical models to describe
filter clarification efficiency.
-------
CHAPTER II
CONCLUSIONS
This study of full-scale granular media filters consisted of four phases.
In the first phase, filter clarification efficiency was characterized using
conventional design and operating parameters, and the statistical properties
of secondary effluent and filter effluent parameters were determined. In
later phases, clarification efficiency was characterized using properties of
secondary effluent suspensions (refiltration, particle size).
Frequency Distributions of Secondary Effluent and Filter Effluent Parameters
1. Mean filter influent BOD (C ,) varied from 9 to 44 mg/4, for the eight
plants, and mean effluent BOD (C, ) varied from 3 to 25 mg/t. The average
ratio of effluent to influent BOD (C./C ,) varied from 0.13 to 0.82.
Filters cannot be compared directly for BOD removal efficiency because
of the dissolved fraction that varies for the secondary treatment pro-
cesses.
2. Average secondary influent suspended solids varied from 28 to 62 mg/s and
filter effluent suspended solids from 5 to 20 mg//,. Coefficient of vari-
ation for influent suspended solids ranged from .33 to .60 while for ef-
fluent suspended solids the range was .25 to .80, indicating a wider
range in effluent than influent characteristics during the sampling
periods. The average ratio of effluent to influent suspended solids
(C/C ) varied from 0.17 to 0.53 for the eight plants.
3. The majority of the data obtained in this study appears to fit a log
normal distribution but occasionally a better fit to the normal distri-
bution was obtained.
4. Extreme values of the data are generally not well predicted by either
the normal or the log normal distributions. Truncated or censored dis-
tributions may give a better fit. Alternatively larger data sets may be
required to give more stable frequency plots.
5. Visual fitting of effluent quality data on probability plots generally
corroborates quantitative goodness of fit tests but still may be mis-
leading, particularly regarding extreme value prediction.
6. There is an indication that some of the parameters have bimodal distri-
butions. Over longer periods of time these might become multimodal and
-------
with enough data the total distribution may still go over to a log normal
distribution.
7. Hie extent of autocorrelation of the data was low enough so that the basic
assumption that the data are independent was reasonable.
Design Considerations
1. Hie most important single design consideration for small tertiary fil-
tration plants is the ability to handle secondary process upsets. This
consideration favors all filters evaluated except the Environmental
Elements design.
2. Emergency overflows are needed for backwash equalization basins.
3. Gravity backwashing is not complete for dual media filters and results
in 25 to 50% washable solids in the solids inventory. No apparent dete-
rioration in performance occurs because of the high solids inventory.
4. In the absence of control system malfunctions, media loss is normally
5 to 10% per year for the dual media designs.
5. Mudball formation or filter bed cracking has never been observed in the
dual media filters evaluated in this study.
Characterization of Clarification Efficiency Using Conventional Design and
Operating Parameters
1. For plants where flow to individual filters was varied, average clarifi-
cation efficiency decreased linearly with average flow rate.
2. Correlation coefficients between filter solids removal efficiency and
operating parameters such as flow, solids loading and run length are
poor for data obtained over long time frames. Improved correlations
were obtained for data analyzed over shorter time periods which minimized
the effect of seasonal and other variations in secondary plant effluent
suspension characteristics.
3. Differences in characteristics of secondary effluent suspended solids
rather than media grain size and depth are most likely the major reason
for variations in clarification efficiency between plants.
4. The following semiempirical model was developed for in-depth filtration:
C = 0.40 C °'8V34
o
where: C » filter effluent suspended solids - mgAt
C = secondary effluent suspended solids - mg//,, and
Q = filtration rate - gpm/sf
In testing with independent data, the in-depth model was able to predict
-------
C within ± 25 percent for 85 percent of the data for three plants.
Characterization of Clarification Efficiency Using Properties of Secondary
Effluent Suspensions
1. Very poor simple correlations were obtained between refiltration para-
meters and particle size data.
2. Number mean particle size correlated with filter suspended solids removal
efficiency within but not between plants.
3. Excellent simple correlations were obtained for filter suspended solids
removal efficiency and the refiltration parameter E if straining was
believed to be the dominant particle collection mode.
4. If in-depth filtration was believed to be the dominant particle collection
mode, inclusion of E in semiempirical models did not significantly improve
correlations.
5. Number mean particle size and E were generally smaller for plants where
in-depth filtration was believed to be the dominant particle collection
mode.
6. The following model developed from data from three plants can now be
applied with caution for high values of E (>1.7):
7. A nearly perfect fit of the data was obtained in testing another model
developed from the data for a plant where flow was varied to 3 filters
with data from three other plants if only data with high values of E
were included. For inclusion of all data, C was predicted within ± 30
percent for 80 percent of the data.
8. Both the straining and in-depth models tend to overpredict effluent
suspended solids if the other particle collection mode apparently
dominates.
-------
CHAPTER III
RECOMMENDATIONS
Additional research on prediction of clarification efficiency of granu-
lar media filters is needed. The models developed in this study need to be
tested with additionel full-scale plant data. Research is needed on methods
to determine a priori whether in-depth or straining filtration will be the
dominant particle collection mode. The initial approach should include a
pilot-scale study under controlled conditions.
-------
CHAPTER IV
EXPERIMENTAL PROGRAM AND PROCEDURES
PLANT DESCRIPTIONS
Secondary Plants
A summary of unit process data for the full scale secondary plants and
filtration systems is given in Table 1. The Addison South and Lake Zurich
plants have contact stabilization activated sludge and trickling filter pro-
cesses in parallel. Other plants utilize contact stabilization activated
sludge processes, with the exception of Marionbrook which also has one com-
plete mix system. Aeration tank detention time based on raw sewage flow
varies from about 6 to 10 hours and final sedimentation tank overflow rate
from about 400 to 700 gpd/fts. These are considered to be typical design
specifications for small-scale activated sludge plants.
Although only the Addison South and Harrington plants serve sewer systems
that contain some combined sewers, large increases in flow occur for all
plants during rains. Increases in plant flow also occur during the spring
months when the ground water table is higher. This indicates that infiltra-
tion and inflow are severe problems. Many of the filtration process failures
observed in this work occurred during rains when the final sedimentation tanks
failed at high flows.
For the small plants in this study, the most troublesome unit operation
is sludge processing and disposal. Except for anaerobic digestion for the
Addison South and Lake Zurich trickling filter processes, and the Harrington
plant, all plants utilize aerobic digestion. Frequent problems, such as
equipment breakdowns, create a situation for the activated sludge plants
where the only immediate solution is to increase the process solids inventory
with a resulting reduction in removal efficiency of final clarifiers. There
may also be a tendency for secondary process operation to be less precise
since the operator knows that the filters will act as a "stop-gap.11
Filtration Plant Designs
A summary of filtration plant design data is shown in Tables 2
and 3. Chemical treatment of secondary effluent prior to filtration, except
for chlorination, was not practiced at any of the plants. Geometric mean
media sizes (d,-n) varied from 1.4 to 2.7 mm. for anthracite and 0.6 to 1.4
mm. for sand. Design depths ranged from 12 to 24 in. for anthracite and 10
to 24 in. for sand. Actual depths when sampling was conducted were consider-
-------
CO
4J
CO
CO
(U
CO
CO
jjcn
CJ ±J
e -<
0 0
u >
e
su ts
» S
ti i
— o*
^
o O cnm cn inm
CM ^^O CM CM OO CM ~> ~*
ZA c/3Cb cn cn cn>£d cn ens
U UH O O CJr- CJ U U
CMCM CMCM^- cncM^
m
cn
C
cn
o
o
o
CM
^
o
o
o
en
O
CM
cn
u
^.
00
c
9)
SO
•O
3
c.
§
u
•3
S
_4 C
o
.
C
O Ji
o =
a
(U
i-H
CO
o
Z
S i
u -J
u tn ^
fl « e3
S £ -J
-------
CO
u
4J
CO
9)
4J
U
(Q
M
CO
•g
CO
r-<
Q,
JJ
r-l
•H
to
H
-c
w
S.
A
.
a
L>
_, Q
*«
3 3
E*
1-
£
i
C
•^
i
OS
i
T
-
C,
X
c
o
w
§4)1- C r,
^ cc — ^
U U > w
c ^
i 2
ffl
u « ^
Z C ""*
n
* y«.
|| =|c
j
K
* s £ s-
§ - « « 2
c "« « c- S ^
> 00 C. to >—
« e jj L.
w c a a
- -» -g o- ^.
B °° U L. 0 w
e e at v o
3 -< a. c-«*-
|- 1|^
« -^ >2 u ^ ®
£ 4) C U C
Z J •* 0 —
t O **•
«-» e -* c C1
ai at u o £,
o -* a **
C_^
=
Ic
«l ft) "^
h. >
S ^ Zt "n jj "^
o* • «e.
tssr
^ £
5 —i
c*
U"
O
eo
«
**
u^
-
^
ij —
60 J^
^^
£
O
.
"
r^
*
I
O
f*J
S
~
w
•z.
=
—
<
5S^S
o c c o
r*. r^ (•*«£.
0 -• — ^
-31
O
•o-
ff> r-. r- r*.
----
£
-H
jN-COjn
0 C O O
es
*"
cn
—
r>4
b. ^«
sn '-'
JC
tn
e
R •> »
— **i
< *• *
rn
O
S
03
-S M
V 0
rj
O
O
(M
c
11
o
<^
tn
c
^
-
(M (-1 IN
000
-3- eo rsi
i
n
-
^
u J=
-*
^^
-
i i i
tA
(N
IE
eo
O
Ckt
W *-'
^C
V.
N
0)
-
O -tf <^
o o o
ao r- ic
-------
CO
C
O
CD
O
•H
M-l
•H
O
0>
CL
CO
13
0)
e
C
CO
CL
4J
CO
CO
H
()
tliracite
•s «
1
•9 £
C -(
W 3
e
S
TS
C
t)
C
•o
iracite
Uniform!-
C
•e
c
•5
o7?
•5 «
0) r"
£ "**
"i s~
C <
i «s
* 1=
l=s=
i 09
"ti-
-1 L.
3 £ £"
.-Hi
i n
-4 L.
— M.-J
-11 =
l~^-
C u^
<
C
.£ X
I|IS
it* — i
S~-£
u eC
>. — i
u u
R=|s
' K
w4 U
-. - «• —
* 6 SC
sr S
~ *e
j -
a.
^. r~ ^- 0 — C i*.
'"" V
i - -
•J ^'-31»»M(N(^'"- O
(M f«gp^»M-rf-N»-w* -^
O — O OC — • 0 pg * *^
eo ^•cD^aO'C^-* *o
9 rsi eo N >« O
CM eocM0*9tcoeofM »
rw O CM O O O C ^ O
f-» m "^ CO >O >£ N
^ sOJ-sC^>^%Off< P*
— 0 — 0 O 0 O O O
j> — i \o «i (M o ^e o ^o
o ooooooo o
-T -ff
-. ff» irt o — ^ ^
9* fl P^ •* •*
^ i-» — ^ ec »
go N 00 ^ « ^
f^l N m CN
K »• — -3- -- i C ft. 01 C
--^^uai-- "8
-= •? « "s - c a et £
<< e -: _: x - ec
£3 5 3 2
' t „ « fc
O u
sc e
•c «
-i u 0
c >-
s
' g o
00V
g *
E-e .u ^ «
O C 4J 1-1
£ f i i £
10
-------
ably less for many plants because of media loss during backwashing since the
plant was constructed or the media last replaced.
Six plants have filter configurations shown in Figure 1 with a backwash
storage' tank located above the sand and anthracite filter media, and a flow-
splitter box for proportioning flow to individual filters (Smith and Love-
less, General Filter, and Eimco designs). Adjustments were made to weir
lengths in the splitter boxes for the Addison South, Lake Zurich, Lisle, and
Marionbrook plants to vary the flow from approximately 0.5 to 2 times average
for three filters.
During the filtration cycle, valves A and C in Figure 1 are open and
valve B is closed. The level of liquid in the vertical influent pipe above
the backwash storage tank level provides the head for filter operation.
Secondary effluent flows through the filter media, into the underdrain sys-
tem, and then to the backwash storage tank where it discharges over a hori-
zontal weir. A sensor automatically initiates the backwash cycle when the
headless reaches a preset level. Valve B then opens and valves A and C close.
A valve on a drain line (not shown) opens to lower the water level above the
filter media prior to air scour. After air scour is completed, valve C opens
and the media is fluidized by flow from the storage tank. The backwash cycle
is terminated by a level sensor in the tank when the water level reaches a
preset point and valves then return to their original positions.
The Neptune Microfloc design at Barrington utilizes a tri-media of
anthracite, garnet, and silica sand with a total design depth of 30 in. A
conventional gravity filter configuration is used similar to the historic
designs for potable water filters. A surface wash system consisting of a
rotating arm within the filter media is utilized in lieu of air scour. A
pumped backwash cycle is initiated when the headless reaches a preset level.
The Des Plaines River plant utilizes a Hardinge filter supplied by Envi-
ronmental Elements Corporation. This design is unique in that an 11 inch
deep single medium sand filter is divided into 8 inch wide compartments formed
by a series of partitions. The backwash mechanism is a motor driven carriage
assembly that moves dfawn the length of the filter and backwashes one compart-
ment at a time. Backwash water is pumped and air scour is not utilized.
The installation at Romeoville is a 10 inch sand filter supplied by
Hydro-Clear Corporation. This design has two unique features to prolong fil-
ter runs. When the liquid level reaches a preset point in a filter run, an
"air-mix" system with diffusers located several inches above the filter medium
is actuated and serves to keep solids in suspension. A "pulse-mix" system
causes air to rise periodically through the filter medium to dislodge parti-
cles from the surface and redistribute solids in the bed. The "air-mix"
system continues in operation during a pumped backwash cycle.
Unique Features of Plants
The activated sludge and trickling filter process effluents at the Addi-
son South plant are combined in a flow equalization basin ahead of the fil-
ters. Filters Nos. 1, 2 and 3 were constructed in 1969 and Nos. 4, 5, 6 and
7 in 1973. Each filtration system has its own elevated flow splitter box
11
-------
or LU
X
o
CQ
GO
e
60
•H
CO
0)
12
-------
which is connected to a common influent pump discharge header. Hie backwash
cycle is initiated when the headloss reaches 13 feet for the Smith and Love-
less filters. Springer (1975) conducted a study on the effect of flow surging
on clarification efficiency during Phase 1. Springer found that by providing
a larger wet well capacity, thus decreasing pump cycle frequency, he was able
to improve average filter suspended solids removal efficiency about 10% (from
80 to 90%). Since the plants are served by a sewer system that has some com-
bined sewers, raw sewage may be bypassed after chlorination during rains. The
Addison North Plant has a basin that is used to store peak raw sewage flows
for subsequent pumpback to the secondary plant. The basin serves as a primary
sedimentation tank when the capacity is exceeded.
Harrington is the only plant that is manned 24 hours per day, seven days
per week. All other plants are staffed for the day shift only and for a few
hours on weekends and holidays.
At the Lake Zurich plant, the activated sludge and trickling filter pro-
cess effluents are discharged to a lagoon ahead of the filters. The filter
influent pumps are constant flow with on-off pump cycles. One large capacity
pump operates during peak flows. During Phase 1 the filter media consisted
of approximately 1 inch of anthracite intermixed with 11 in. of sand. The
filter media was replaced with 12 in. of anthracite and 12 in. of sand prior
to Phase 4.
The Lisle and Marionbrook plants have lagoons ahead of the filters. Dur-
ing Phase 1 at Marionbrook, the effluent from the contact stabilization plants
was discharged to the lagoon. The effluent from the complete mix plant was
discharged directly to the filters. During Phase 4 all plants discharged to
the lagoon. The Eimco filter design at these plants has a timer mechanism
that can be set to initiate the backwash cycle.
PHASES OF STUDY
Phase 1
In the initial phase of the study to characterize performance as a func-
tion of design and operational variables, all analyses were made on flow-pro-
portioned composite samples of secondary effluent (filter influent) and filter
effluent. Most sampling periods were 24 hours, with occasional periods of 4,
8, and 16 hours. Samples were analyzed for suspended solids and five-day BOD
on a routine basis. Average run lengths were determined from backwash fre-
quency counters or operating time for backwash mechanisms. For plants that
did not have backwash frequency counters, mercury switches attached to valve
operating mechanisms and clocks were used to determine frequency. Headloss
at the end of runs was determined if filters were equipped with altitude
gauges.
Sampling was conducted during the period from May, 1975, to August, 1976.
Sampling programs were generally scheduled for two months at each plant to
provide sufficient data for statistical analyses. Because filters were out
of operation frequently, only 36 samples were obtained for the Romeoville
plant. Fifty-eight or more samples were obtained for other plants. Flow rate
was varied to three filters at the Addison South (Nos. 4-7), Lake Zurich,
13
-------
Lisle, and Marionbrook plants.
Phase 2
Phase 2 was conducted at the Addison South plant in December, 1976, and
January, 1977. Composite samples were obtained but were not proportioned
according to flow. Flow was varied to three filters and Filter No. 2 was
also sampled. Samples were analyzed for suspended solids only and refiltra-
tion tests were conducted.
Phase 3
During Phase 3, conducted during March and April, 1977, grab samples were
obtained at the Addison South (four filters) and Des Plaines River plants.
Refiltration tests and particle size counts were made on-site. At Addison
South, grab samples were obtained every half-hour and composited over a three-
hour period. Samples were composited each hour at the Des Plaines River plant.
Filter headloss data were obtained for both plants.
Phase 4
Grab samples were obtained at all eight plants during Phase 4 in May and
June, 1977. Suspended solids tests were conducted, as in Phases 2 and 3.
Samples were transported to the laboratory for refiltration tests and parti-
cle size counts. Headloss measurements were made.
SAMPLING EQUIPMENT AND FIELD PROCEDURES
SERCO Samplers
Four SERCO automatic samplers (Model NW-3-DC), manufactured by Sonford
Products Company, were used during Phase 1 along with flow splitter samplers
described below. The SERCO sampler has 24 sample bottles with individual
tubing connected to a metal sampler head, and a battery-operated trip arm.
To set up the samplers, the sampler head is connected to a vacuum pump and
bottles are vacated to about -25 in. Hg., and the tubing for each sample
bottle is pinched off by a set switch. Hie movement of the trip arm can be
adjusted to disengage a set switch at intervals from 2 min. to several hours.
One sample per hour was obtained during Phase 1 and the units were not used
during later phases.
Flow Splitter Samplers
Several flow-proportioned sampling systems were designed and fabricated.
The basic design consists of a "weir box", two cylinders with V-notch weirs
on their peripheries connected in series, and a timer and solenoid.
The sampler device is attached to the weir and intercepts a relatively
constant percentage of the total flow. Spacers were installed at the entrance
to the weir box to obtain the desired flow rate. The effluent from the weir
box is piped to the center of the first flow splitter. Flow from one V-notch
is isolated and piped to the second flow splitter, where the flow from one V-
14
-------
notch is again isolated and piped to the timer-solenoid device. The final
design utilized ten V-notch weirs on each cylinder for an approximate overall
flow reduction of 1/100. The effluent line from the second flow splitter
discharges to a plastic box and then to waste. The timer-solenoid device
diverts the sample line for a few minutes per hour to an adjacent plastic box
that is connected to a sample container.
A chronological summary of the sampling program and equipment used is
given in Table 4.
Sample Preservation
During May to August, 1975, ice was placed in a container inside of the
SERCO samplers at the Addison North and South plants. The samplers were not
housed at these plants. A timer-solenoid device was not used for the flow
splitter samplers used for filter effluents and samples were collected in
covered 50-gal. drums.
A decision was made to not preserve samples with ice for the remainder
of the study because the major thrust of the study was on suspended solids
removal, where changes in concentration would be expected to be less pro-
nounced as compared to other parameters, and because of the additional field
labor that would have been required.
Flow Measurement
Each plant is equipped with indicating, recording, and totalizing raw
sewage flow meters, except for the activated sludge portion of the Lake Zurich
plant where flow is approximately one-third of total plant flow. Average
plant flow during runs was determined from totalized flow readings. When the
SERCO samplers were used, data from flow charts were used to flow proportion
the composite samples, except at the Harrington plant where operators recorded
the flow at two-hour intervals.
At Lake Zurich, where flow was varied to individual filters, rise rates
in backwash storage tanks were measured and used to estimate flow (i.e.,
1 ft/min. = 7.5 gpm/ft3). Since the filter influent pumps operate at constant
flow, a float-switch device was installed in the flow-splitter box and con-
nected to a battery-operated clock to determine operating time during runs.
Rise rates in backwash storage tanks at the Addison South plant could not
be used to estimate flow because of the flow equalization tank and on-off
cycles for one large capacity influent pump. The flow to each filter was
determined from measured total head on the effluent weirs using published
graphs of head vs. discharge for weirs with end contractions. Measurements
were made for a minimum of six days and the flow for each filter was totaled.
Flows for all seven filters were totaled and the percentage of flow to each
filter was calculated. These percentage factors were then applied to the
metered plant flow to estimate average individual filter flow rates. Filters
receiving the low, medium, and high flow rates were changed several times to
minimize possible bias in the data. Each time a change was made, the above
procedure was repeated.
15
-------
CM
£>
co
H
4J O
C 1-1
CU4J OOOr-lr-Hr-lr-IHOr-lr-IM
3 tfl
r-l O
M-l O
w
S-4 ^
(0 _x
4-> S-I /-N/-S
r-l CO CO CO
•r~l i— 1 ^»w^ N*-''
ff j O FT i pr . y} FT i fy | TT | ££J y^ ^J |V | J~ y . IV |
g
CO
CO
/"— S
Csj
N^ X
fi
4J O
C -H
CMCNCM--~-~^- — — . ' — -~~---
co oooOr-rHf3O C
4J co coco-Hp^ o o PL* 1-1 en cu o
C CO -r-l -r-l S-I CO T-l CO S-I 'r-l r-l -r-l
cfl co 'd'ds-ico^is-igcos-i'dcas-i
i — i ji 'd'ococucocoocucO'dT-ico
P-I p^ <x> ^
t"*- lQ
^-. W
r— 1 P—
1 l-i P—
VO tfl ^
r- s-i co
^^ 60 1
CSJ ^-* CM
r— i
cj\ r— m f»
CM CN
tfl rC
4J 4J
3 r-* 3 r-
O 1 O 1
60
C
•r-l
•d
s-i 1-1
CO T-l
s~*. r— 1 3
C a«.a
co E
V. tfl CU
tfl CO T)
4J -r-l
S-I CO
CO CO C!
CO 4J -H
1 1J
CX T-I S-i
g r-l CO
Cfl Oi r-l
to to cx
r— r— s g
i — xi P— O to
~-» tfl ^^ r-l CO .
-------
At the Marionbrook and Lisle plants, the flow to filters was estimated
from the ratio of individual filter weir length to total weir length in the
flow splitter box. At Marionbrook, a portion of the secondary effluent was
bypassed when flow was high. Since the filter effluent weirs were not readily
accessible, a manometer was installed to measure the head in the flow splitter
box. The manometer readings were compared to readings when the filters were
not bypassed to estimate flows. This approximate procedure was used for 26
runs.
LABORATORY PROCEDURES
Laboratory and field data are not reported here. They are included in the
1978 Ph.D. dissertation entitled "Evaluation of Full-Scale Tertiary Wastewater
Filters" by Charles L. Swanson. Copies may be obtained from University Micro-
films, Ann Arbor, Michigan.
Standard Laboratory Analyses
All laboratory analyses for BOD and suspended solids were conducted in
accordance with Standard Methods for the Examination of Water and Wastewater
(1971 Edition). Dechlorination procedures were used prior to the BOD test
for several samples from the Marionbrook plant in Phase 1 when chlorine resi-
dual was 0.2 to 0.4 mg/£. It was found that the differences in BOD were within
the standard error of measurement and dechlorination procedures were discon-
tinued.
Suspension Properties
Methods used to characterize secondary effluent suspensions were sus-
pended solids concentration, refiltration parameters, and particle size
measurements.
Hsiung (1972) was the first to utilize refiltration techniques to charac-
terize performance of tertiary wastewater filters. The refiltration technique
was first developed utilizing the Kozeny-Carmen equation by LaMer and Smellie
(1956). This technique involves filtration of three or more samples with dif-
ferent volumes through microporous filter paper and measurement of the time
1/3
for refiltration of the filtrate. When VC is plotted vs (V/t) , a linear
relationship is obtained, where
V = volume of filtrate - m^,
C = filter influent suspended solids - mgAf,, and
t = refiltration time - sec.
Hsuing defined the refiltration parameters R and G as the intercepts on the
ordinate and abscissa axes, respectively. He proposed an empirical factor,
1/3 1/2
E = (R G) , and obtained data for pilot filtration of secondary effluents
from 4 plants at a constant rate of 5 gpm/sf. An excellent correlation of E
with percentage removal of suspended solids was obtained. In Chapter VI it .
is shown that for unisized particles,E is proportional to the cube root of
17
-------
the particle diameter. In this study an excellent correlation of C/C with E
was obtained for data sets where straining was believed to be the dominant
particle collection mode.
Particle size counts were made using a compound microscope (lOOx) and a
Porton graticule. The number mean particle diameter (d ) varied from 8 to
38 urn. Some correlation was obtained for C/C vs d for individual plant
^ o nm
data with particle collection efficiency increasing with larger particle sizes
as would be expected. However, the particle number mean did not correlate
well with C/C between plants.
Refiltration Tests
Refiltration tests were conducted using 0.45 jj,m Millipore filter paper
(47 mm. diameter) and a standard Millipore Filtration Apparatus at a vacuum
of 25 in.Hg. The procedure used by Hsiung (1972) for determining the refil-
tration parameters R (intercept on ordinate axis) and G (intercept on abscis-
sa) was modified .to account for differences in volume and temperature between
the filtered sample and filtrate. The correction applied was:
(1)
where t = refiltration time--sec.,
t = corrected time—sec.,
V = volume of sample filtered—m£,
s
Vf = volume of filtrate--m{,,
|j, = viscosity of sample-centipoise, and
S
)j,f = viscosity of filtrate-centipoise.
Viscosities were determined from sample and filtrate temperatures using Hazen's
approximation (Fair, Geyer and Okun, 1968):
H = 78.6/(T + 10) (2)
where T is in °F.
During Phase 2, at least four refiltrations were made for each sample in
order to improve accuracy. Frequently, three points fell in a straight line
with one outlying point. In those cases, the outlying point was excluded in
estimating the line of best fit by eye. The filter paper was not prewashed
with demineralized water during Phases 2 and 3.
1/3
During Phases 3 and 4, plots of VCQvs.(V/t) were prepared during the
tests and additional filtrations were made if a reasonable straight line fit
was not obtained for three samples. The line of best fit was then determined
using regression techniques on all data points.
18
-------
Filter paper was prewashed with 100 m{, of demineralized water during
Phase 4. During this phase, filtrate from the suspended solids tests (glass
fiber filters) for each plant was used for refiltration in order to stream-
line the test procedure. Before adopting this procedure, tests were con-
ducted to compare results with use of filtrate from the Millipore filter
paper. No significant differences (less than 1 or 2 percent at 2 min.) in
refiltration time were obtained.
The accuracy of the refiltration test in determining the parameter E was
estimated in Phase 4. Four filtrations were used for 21 of the 38 samples in
this phase. In order to eliminate bias, these 21 samples were used for the
following analysis. The assumption was made that the true value of E was that
1/3
obtained for the regression of VC vs. (V/t) with the four data points.
Four additional regressions were made for three data points with a different
data point excluded each time. The method used to estimate error bounds is
illustrated below for a sample obtained at the Barrington Plant (5/26/77).
E Error--%
Four data 1.403
Three data -- Run 1 1.418 +1.1
-- Run 2 1.411 + 0.6
-- Run 3 1.404 + 0.1
-- Run 4 1.374 - 2.1
Error Range 3.2
Error Bounds ±1.6
One refiltration test had error bounds of ± 18 percent. For the remain-
ing 20 samples, the error bounds ranged from ± 0.4 percent to ± 5.4 percent,
1/3
with an average of ± 1.8 percent. A plot of VC vs. (V/t) is shown in
Figure 2 for the Barrington plant sample, and the samples with error bounds
of ± 0.4 and ±5.4 percent. For the data in this analysis, it can be con-
cluded that E is determined with an average accuracy of less than ± 2 percent
for 95 percent of the samples.
Particle Size Determinations
Wet mount particle size counts were made using a compound microscope
(100 X) and a Porton graticule, shown in Figure 3. The graticule has cir-
cles with diameters increasing by a factor of J~2. The diameters vary from
approximately 4.2 p^i for class 1 to 68 \j?a. for class 9. Grids are used for
particles larger than cldss 9. The lower one-half of the graticule, with an
-3 2
area of 10 cm was used as the counting field. A counting cell with a depth
-4 3
of 1 mm. was used giving a viewing field volume of 10 cm .
Sample bottles were shaken by hand to break up large floe particles that
tended to form during the time from collection and during transport to the
laboratory. Samples were placed in the counting cell using a wide tipped 10
ntf, pipette.
19
-------
I
o
O
0
\
\
Phase 4 Data
Plant Dote C0 E Error-± %
OAS 5/11 47" 1.79 5.4
DBA 5/26 9 1.40 1.6
A DP 5/15 12 1.67 0.4
\\
\\
\\
\\
(V/t)1
0 0.4 0.8 1.2 1.6 2.0
J/3
Figure 2. Refiltration data plot, V C vs. (V/t)
1/3
20
-------
o
•o
to
o
o
«o
«M
I . I . I , I , I , I , I , I . I.. I .
8
oo
CM
4-1
tO
O
•H
a
o
•H
(-1
00
§
§
M
O
00
•H
21
-------
The number of fields used to determine the particle counts in each size
class varied from 6 to 40, depending on the particle concentration. In order
to eliminate bias in the analysis, the counting cell was moved while the field
was not viewed. The standard method of viewing enough fields to count a speci-
fied minimum number of particles in the largest size class was not used because
of the excessive time requirements for dilute samples.
22
-------
CHAPTER V
PERFORMANCE DATA AND OBSERVATIONS OF PLANT OPERATIONS
REMOVAL OF BOD AND SUSPENDED SOLIDS
Tertiary granular media filters are designed primarily for removal of
suspended solids and the BOD associated with the suspended solids. However,
removal of dissolved BOD occurs to some extent due to bacterial activity with-
in the filter media. Dissolved BOD is defined here as that BOD associated
with the filtrate from a standard glass fiber filter. Some colloidal solids
measured in that filtrate may be transported to the full-scale filter medium
by Brownian diffusion (Cookson, 1970) and removed or metabolized by attached
microbial populations.
A summary of BOD removal data is shown in Table 5. Mean filter influ-
ent BOD (C , ) varied from 9 to 44 mg/t for the eight plants, and mean effluent
BOD (C, ) varied from 3 to 25 mg/;,. The average ratio of effluent to influent
BOD (C,/C , ) varied from 0.13 to 0.82. Filters cannot be compared directly for
BOD removal efficiency because of the dissolved fraction that varies for the
secondary treatment processes.
Dissolved BOD tests were conducted for a portion of the sampling program
at the Addison North and South plants and were not made routinely thereafter.
Plots of effluent vs. influent dissolved BOD are shown in Figure 4. It is
apparent that a significant removal of dissolved BOD occurs for these filters.
Average influent and effluent dissolved BOD is shown below.
Addison Addison
North South
Dissolved CQb --mgAf, 7.5 8.6
Dissolved C, --mg/^ 3.8 6.0
% Removed 49 30
This observed removal also probably reflects the fact that biofilms attached
to the filter media serve as adsorption sites for some components of the non-
colloidal organics. Dissolved oxygen levels in filter influents are high due
to turbulence in the flow splitter boxes and filter influent lines, providing
sufficient oxygen for aerobic bacterial activity.
A summary of average flow and suspended solids data is shown in Table 6.
23
-------
r-l
$
CO
X*
P-1
,
to
4-1
to
*O
Q
O
O
M-l
CO
O
^_(
4J
CO
t-l
4J
CO
4->
OT
•
in
CU
r-l
ft
CO
H
.£
c
^
4J
C r-l
CU ^
3 60
1-1 B
C !n
M 0
u
4-1
C r-l
CJ ^
3 >\n
r3 |P
m 1
W u*
CO
0)
CO
«
O
O
•
O
s
4J
C
CO
r-l
PL,
rti
vu
60
C
cfl
Pi
b
c
CO
cu
S
cu
60
C
CO
OH
b
C
to
cu
CU
£^
CO
OS
'g
CO
CU
S
XI
O
0
,
o
0 I—I
r-l A
"^ / \
1 1
CM r^
i i
CM O
CM r-l
CO CM
1-4 r-l
<)• CO
<}• CM
m co
CO CM
1 1
r- O
i— i i— i
r^- cy.
m co
in vo
CM r-l
oo o
r-l CO
j~
i
0
i—*
-*
r-l
^J-
CM
ON
r-l
1
CM
<*
in
r-l
r-l
OO
VD
Of
T)
CU
S
i
CU
r-l
CO
•r-l
rJ
m
oo
i*
VO
•— 1
°
CO
-------
CP
I
.c
o
10
CD
S 5
o
CO
0
0
15
o>
e 10
o*
-2 5
13
MBMM
o
0
Fig. A
0 0
0 Addison North
o°
o o o o
o o o o o o
p o o oo o
5 10 15 20
Soluble Cob-mg/l
o Fig. B
•Slope= I
oo/o
o o o o
/o ° °o Addison South
° - ° /P« ^° o ^
O 0°c&°0 0°
0
5 10 15
Soluble
20
Figure 4. Filter effluent vs. influent soluble BOD, Addison North and
South - Phase 1.
25
-------
Table 6. Summary of plant performance data.
Plant
Addison North
(1)
Addison S. («r '
Addiaon S. (#4-#7/
Low Flow
Medium Flow
High Flow
Harrington
Des Plaines River
Lake Zurich
Low Flow
Medium Flow
High Flow
Lisle
Low Flow
Medium Flow
High Flow
Marionbrook
Low Flow
Medium Flow
High Flow
Romsoville
No. of
Data
73
70
61
51
64
58
60
24
48
22
32
29
35
105
100
55
36
Flow-gpm/sf
Meant? )
3.8
2.1
0.87
1.8
3.4
3.6
0.76
1.3
2.3
3.9
1.4
3.0
5.3
1.2
2.4
4.4
1.5
CV(*J
0.16
0.22
0.21
0.28
0.25
0.23
0,39
--
--
--
0.26
0.26
0.17
0.36
0.31
0.36
0.26
Suspended Solids-mg/A
In fluent (Co)
Mean
44
38
52.
49
51
37
31
28
31
29
62
59
61
38
34
32
31
cv(*)
0.43
0.45
0.42
0.43
0.43
0.32
0.61
0.39
0.35
0.38
0.61
0;61
0.61
0.53
0,50
0.59
0.55
Effluent(C)
Mean
6.6
9.0
11
14
20
12
5.0
9.5
14
15
9.0
12
18
7.1
8.4
7.2
7.4
CV
0.56
0.43
.0.65
0.57
0.60
0.34
.0.60
0.31
0.31
0.25
0.51
0.62
0.72
0.76
0.79
0.64
0.50
C/Co
Mean
0.17
0.26
0,24
0.30
0.40
0.32
0,21
0.34
0.48
0.53
0.18
0.24
0.34
0.20
0.26
0.27
0.26
CV
0.65
0.46
0.63
0.53
0.38
0.34
0.76
0.26
0.33
0.21
0.50
0.54
0.59
0.65
0.65
0.67
0.38
(1) Secondary process - activated sludge and trickling filters in parallel.
Other plants activated sludge.
(2) Filters operate at constant flow with on-off influent pump cycles.
(3) Arithmetic mean
(4) Coefficient of Variation - standard deviation/arithmetic mean
26
-------
Average secondary influent suspended solids varied from 28 to 62 mgAf, and fil-
ter effluent suspended solids from 5 to 20 mg/t. Coefficient of variation for
influent suspended solids ranged from .33 to .60 while for effluent suspended
solids the range was .25 to .80, indicating a wider range in effluent than
influent characteristics during the sampling periods. The average ratio
effluent to influent suspended solids (C/C ) varied from 0.17 to 0.53 for
the eight plants.
DESIGN CONSIDERATIONS
Process Selection
Historically, optimal design for potable water treatment has been defined
as the condition where effluent turbidity breakthrough occurs at the time the
terminal headloss is reached (Ives, 1972). This concept has also been ex-
tended to tertiary wastewater filtration, along with optimizing run length
by determining the effect of recycled backwash on secondary treatment costs
(Baumann and Huang, 1974).
In the authors' opinion, these concepts are not applicable to design of
small-scale ( < 5 mgd) tertiary wastewater filters. As a practical matter,
the incremental capital and pumping costs for changes in splitter box ele-
vation over the normal design range for the filter configuration shown in
Figure 1 are insignificant. In addition, the incremental secondary treat-
ment costs for the additional capacity to handle backwash over the range of
run lengths of 8 to 24 hours are not significant, particularly if one or two
backwashes are set by timers to occur during diurnal periods when plant flow
is low.
The most important single design consideration for small tertiary fil-
tration plants that are not staffed 24 hours per day is the ability to handle
secondary process upsets. This consideration appears to favor dual media
designs with relatively large media and high terminal headloss since particle
collection efficiency showed a weak dependence on media size, as discussed
in Chapter VIII.
Process Failures
Process failure of the Environmental Elements sand filter installation
occurred at influent suspended solids concentrations of approximately 60 mg/£.
When failure occurred, the filter influent backed up and flowed into the back-
wash return channel, eventually overflowing into the filter effluent channel.
At the time of sampling no automatic bypass was provided. In the subsequent
plant expansion bypass occurs if the head on the filters exceeds 16 inches
and is returned to the head of the plants.
The Hydro Clear design has a bypass weir to handle process upsets. Pro-
cess failure for the Romeoville installation corresponds to exceeding the
capacity of the backwash equalization basin ("mudwell"). This occurred under
extended secondary process upsets because of a combination of unrelated drain-
age into the basin and inadequate basin drain pipe capacity. This design has
an air pulse system and can handle considerably higher suspended solids levels
27
-------
than the Environmental Elements design if backwash basin capacity is adequate.
Multimedia designs with anthracite and sand d5Q of 1.4 - 1.7 mm
and 0.58 - 0.71 mm respectively can generally function with influent sus-
pended solids levels of approximately 80 to 100 mg/£. No process failures
were observed for the Harrington plant (Neptune Microfloc), which is staffed
24 hours per day. For this plant the filters are bypassed during secondary
process upsets. Filter Nos. 4-7 at the Addison South Plant, with a d5Q of 2.7
mm for anthracite and 1.4 mm for sand, were able to filter influent suspended
solids concentrations in excess of 200 mgAt. It should be pointed out that
under these conditions effluent suspended solids exceeded 30 and even 50
and may still be unacceptable. Maximum available head is approximately
13 feet for the Smith and Loveless design, and 7 feet for the General Filter
and Eimco designs. For the design shown in Figure 1 failure is considered
to occur when the filter influent backs up into the flow splitter box.
In all 6 filter designs with an elevated splitter box, an overflow pipe
is provided. The overflow piping system should be designed to handle the
maximum capacity of the filter influent pumps. This is not the case at three
plants where the splitter box overflows during secondary process upsets when
the filters become clogged and backed up. The overflow drenches electrical
and mechanical controls and causes malfunctions, in addition to flooding of
the filter building. This also compounds the corrosion problem on electrical
control mechanisms.
Equalization of Backwash Water
Many state regulatory agencies require flow equalization facilities for
return of backwash to the secondary treatment process in order to minimize
flow surges. This requirement causes severe problems during secondary pro-
cess upsets at three of the plants. At these plants, the equalization basins
are located under the filters below grade level.
When filters backwash frequently, water backs up and floods the filter
buildings. At one plant, doors on the filter building are left open to pre-
vent submergence of automatic valve operating mechanisms. Depths in excess
of 2 feet were observed in the filter building at one plant. This situation
also occurs when valve control systems malfunction and backwash valves remain
open. In this situation, secondary effluent discharges to the equalization
basin and floods the building.
This problem could be eliminated by providing an overflow in the equali-
zation basin to return backwash water directly to the secondary process. An
alternative would be automatic shutdown of flow to the filters and bypassing
directly to chlorination facilities, if permitted by regulatory agencies.
Filter Media Backwashing
Backwash rate tests were conducted for all filtration plants with the
backwash storage tank located above the filter media compartment. A plot of
backwash flow rate vs cycle time for 3 filters is shown in Figure 5. For
the large media at Addison South (No. 4) and Addison North the backwash rate
28
-------
cr
\
\
en
o
LU
cr
o
u.
14
12
,0
o 8
Addison South
Fifter No. 4
Addison North
Rlter No. 2
Lake Zurich
Filter No. 4
2 4 6 8 10 12 14
BACKWASH CYCLE TIME (minutes)
Figure 5. Flow rate vs. backwash cycle time.
29
-------
100
Lake Zurich
Filter No. 4
BACKWASH CYCLE TIME (minutes)
Figure 6. Suspended solids concentration vs. backwash cycle time.
30
-------
appears to be insufficient to develop complete fluidization. The flow rate
varies during the cycle because of the declining head in the backwash storage
tank. During the latter part of the cycle, flow rate is insufficient to main-
tain fluidization for the filters with smaller media at Lake Zurich.
At the Lisle and Marionbrook plants (Eimco design), initial rate varied
from 13 to 18 gpm/sf and after 3 to 4 min. the rate dropped to 10 gpm/sf.
For these filters, the backwash rate can be adjusted by changing the opening
size on the discharge line.
Tests were made at the Addison South (No. 4) and Lake Zurich plants to
determine the effectiveness of the backwash cycle in cleaning the media.
After an initial backwash, filters were backwashed 2 more times as *oon as
the storage tank refilled. A plot of backwash suspended solids concentration
as a percent of the maximum concentration in the first wash vs time is shown
in Figure 6 for 3 consecutive washes for Lake Zurich. The suspended solids
concentration decreases rapidly with time during the first cycle. The solids
yield per backwash cycle was determined by summing the product of flow times
solids concentration per unit of time. For Lake Zurich, the ratio of per-
centage of solids recovered during the 1st, 2nd, and 3rd washes was 75:15:10,
respectively. Thus, about 25% or more washable solids are carried in inven-
tory after one backwash.
Filter media cleaning was much less effective for the large media at
Addison South (No. 4) where little fluidization takes place. For this filter,
the ratio of percentage solids recovered during the 3 cycles was 50:25:25,
respectively. Thus, 50% of the washable solids remained in inventory after
one backwash.
These dual media filters are not cleaned as effectively as unstratified
coarse single medium sand filters as reported by Dahab and Young (1972), as
evidenced by the large solids inventory after one backwash cycle. However,
data obtained on filter performance indicate that the high solids inventory
does not appear to result in decreased particle collection efficiency. The
removal efficiency for the small diameter media filters at Addison South,
which are cleaned more effectively, was similar to the large diameter media
filters at comparable flow rates during fliase 1. A similar result was ob-
tained in later phases of the study.
Dahab and Young (1977) contended that mudball formation is prevalent when
wash conditions are not conducive to breaking up and washing mudballs from
the freeboard space. They considered poor freeboard solids washout to be
one undesirable characteristic of dual media filters, since the solids may
provide nuclei for mudball formation.
From examination of both pilot and full scale dual media filters in this
study, it was found that freeboard solids washout is not complete and as much
as 1/8 to 1/2 inch of solids settle back onto the media surface after a wash.
In pilot column tests run in parallel with the full scale filters at Addison
South and Marionbrook, Minix (1979) found these solids formed a porous struc-
ture that contributed little to accumulation of residual headloss.
Presumably, since each of the filters has air scour or surface wash these
31
-------
solids did not get the chance to contribute to mudball formation. With the
above mentioned characteristics and wash volumes of only 30 to 40% of those
normally used, mudball formation or filter bed cracking has never been ob-
served in these dual media filters. However, evidence of media packing
(mudballing) was found in the Neptune Microfloc installation, particularly
in the corners of the rectangular basin where the rotary surface wash was in-
effective. This was in spite of the 17 gal/min/ft3 wash rate and wash volumes
three to four times the other dual media designs (Table 2). Also as expected
all designs that use auxiliary air (or surface wash) experienced rapid deteri-
oration of performance if the auxiliary malfunctioned.
Media Integrity
Loss of media integrity, i.e. the loss of media, change in gradation or
intermixing, occurs to a varying degree in these dual media filters. Anthra-
cite loss, the most dramatic phenomenon, is noted on Table 3 for most dual or
multi-media filters. Media loss is somewhat inexplicable based on the rela-
tively low backwash rates (subfluidization in some cases) in all the gravity
wash filters. In fact, losses are associated with mechanical or control sys-
tem failures. Control systems and valve mechanisms often suffer corrosion
damage causing malfunctions, such as continuation of air scour during back-
wash and resulting media loss. In one filter, a structural failure in the
bottom of the backwash storage tank caused leakage and loss of media during
air scour. Attrition loss in the absence of such incidents is normally only
5 to 10 percent per year for most filters, except for one system where anthra-
cite losses are large due to high backwash rates. The least amount of media
loss occurred for the relatively large media at Addison North and South.
Since there is little difference in particle collection efficiency for
various multimedia configurations, design considerations should generally
favor large diameter media.
Cleasby and Baumann (1974) state that for dual media filters the dgo an-
thracite : dio sand ratio is the most important factor in determining the amount
of intermixing of anthracite and sand at the dual media interface. The cal-
culated ratios of 4.5 - 5.0 suggest that a relatively mixed interface should
result. Pilot columns paralleling full scale operation at two sites de-
veloped up to 8 inches intermix zones but did not contribute excessive
headloss, Minix (1979). A possible explanation is that most of the intermixing
occurred during air scour rather than backwashing (subfluidization) and was
correspondingly less dramatic. Core samples from the full scale filters showed
that the sand tended to accumulate near the filter walls to some extent, and
that the depth of anthracite was largest near the center.
No specific detailed data on media cleaning and integrity were obtained
for the shallow bed single media sand filters. Even though able to produce
a satisfactory effluent at 60 mg/^, influent solids, the Environmental Elements
filter developed an associated problem. Grease and high solids carryover to
the filters from the secondary under these conditions provided mudball nuclei
that resulted in several inches of media loss over a period of such incidents.
Eventually, the "clean bed" headloss increased enough that complete media
replacement was required. With a high or higher grease and solids the Hydro
Clear installation never experienced either of these problems because of a
32
-------
detergent and hypochlorite media wash system.
Mechanical and Electrical Control Systems
It is difficult to make a direct comparison of control systems for the
plants because of the variable age of equipment. However, some general ob-
servations can be made. No problems occurred during the study for the Hydro
Clear and Environmental Elements designs. Frequent problems occurred for
the Neptune Microfloc plant; this equipment is scheduled for retirement from
service soon.
Frequent maintenance is a necessity for the plants with the backwash
storage tank located above the filter media compartment. Corrosion problems
are severe after about 5 years due to the high humidity in the filter build-
ings. The air cylinder operated backwash and air valves were observed to
stick frequently in the open position. However, the increased maintenance
for these designs is partially offset by the reduction in operating labor
required as compared to the Environmental Elements design which required
frequent washdown and cleaning of the filter walls.
33
-------
CHAPTER VI
CLARIFICATION EFFICIENCY MODELS
OF GRANULAR BED FILTERS
INTRODUCTION
Filtration of secondary effluents through granular media is considered
to be one of the most cost effective processes in upgrading effluent quality
to meet more stringent discharge standards. Suspended solids, consisting of
flocculated microorganisms or cell debris and inorganic particulates, are
separated from the liquid phase by collection at the surface and within the
pores of an unconsolidated porous medium.
Most recent studies of tertiary wastewater filtration have been directed
toward filter media selection and least cost design and optimization; little
attention has been given to the effect of suspension characteristics on filter
clarification efficiency. No method has previously been developed to determine
filter clarification efficiency a priori without pilot-scale testing.
The purpose of this chapter is to review pertinent filtration and clari-
fication literature and provide the basis for development of operational math-
ematical models to describe wastewater filter clarification efficiency. These
models are intended to provide a uniform basis for examination of data on
full-scale filter clarification efficiency.
CURRENT WASTEWATER FILTER DESIGN
Wastewater filter design involves specification of filter type, process
arrangement, flow rate, and media type, size and depth. Chemical addition to
promote aggregation/destabilization of the colloidal fraction ahead of filters
is generally not practiced, particularly for small plants.
Where economics dictate or permit, design criteria are developed from
pilot-scale tests. In many cases design is based on empiricism and rules of
thumb derived from potable water filtration and ten or more years of experi-
ence in tertiary wastewater filtration in the United States (Cleasby and Bau-
mann, 1974) and Great Britain (Oakley and Cripps, 1969; Guiver and Huntingdon,
1971; Tebbut, 1971; Michaelson, 1971). Filter suspended solids removal effi-
ciency is specified with tongue in cheek since reliable, simple correlations
of filter efficiency have yet to be developed for wastewater suspensions.
PARTICLE COLLECTION MODES
Particle collection in granular media filtration involves complex inter-
34
-------
actions between a number of physical and chemical forces. These are mani-
fested in terms of a number of variables: media type, size, and depth, fil-
tration rate, influent suspended solids concentration, influent particle size
distribution, type of secondary process, and flocculation characteristics of
solids. The latter three will also be a function of the type of secondary
process.
Much of the work on filtration theory prior to 1960 can be classified as
physical, and was concerned primarily with media size and depth, deposit dis-
tribution within the filter, and headloss development.
Beginning approximately 15 years ago, emphasis was placed on both the
physical and chemical aspects. Beginning around 1964 (Gregory, 1964), parti-
cle collection was considered to be a two-step process consisting of trans-
port to the vicinity of the collector (grain) and attachment to the grain or
already deposited solids. The former was considered physical and the latter
chemical. Classical particle transport mechanisms were identified as gravity
settling, interception, and Brownian diffusion; and attachment forces as van
der Waal attraction, electrical double layer, and specific chemical inter-
actions (Ives, 1969; O'Melia and Stumm, 1967).
Depth Filtration Models
The bases for most physical, depth filtration models describing evolu-
tion of clarification efficiency with depth and time are the first order
deposit and mass continuity equations proposed by Iwasaki (1937).
- = XC , and (3)
" 9L Q dt '
where C = concentration of suspended material,
L = filter depth,
\ = filter coefficient,
t = time,
o"s= specific deposit — volume solids/volume filter, and
Q = superficial filtration velocity.
For initial conditions, integration and rearrangement of Equation 3 gives
where C^, = initial concentration of suspended material.
As filtration and deposition of suspended material progresses, the value of
the filter coefficient X changes. A variety of theoretical and empirical
35
-------
models have been proposed to relate \ to specific deposit (Herzig, Le Clerc,
and Le Goff, 1970). This research has been summarized elsewhere and will not
be reviewed here (Ives, 1971).
FitzPatrick (1972) and FitzPatrick and Spielman (1973) presented a model
of initial filtration (\Q) derived from extending and solving the trajectory
equations of Spielman and Goren (1970) for flow past a spherical medium parti-
cle. By this theory, which contains no adjustable constants, they were able
to reproduce to a factor of 2 the collection efficiency (Xo) over a range of
four orders of magnitude.
Recently, Ghosh, et al. (1975), have applied the Spielman and FitzPatrick
(1973) model for particle capture to analyze wastewater filter performance.
They obtained good correlation of collection efficiency over the 2 to 10 pfli
particle size range. However, particle sizes in secondary effluent suspen-
sions are not usually confined to this limited range.
Minix (1979) conducted on-site pilot scale studies in parallel with the
full-scale sampling program at the Addison South and Marionbrook plants during
Phase 1. He found clarification efficiency a weak function of run time and
filter depth. Comparisons of his data with the Spielman and FitzPatrick (1973)
model showed significant underprediction for anthracite layers but slight over-
prediction for sand layers.
Based on pilot-scale experimental results for a relatively limited range
of secondary effluent properties, Tchobanoglous and Eliasson (1970) developed
a depth filtration semiempirical clarification model, viz.
ff = - [ ] \oC , and (6)
dX L (1 + a X)n J
C = concentration—mg/
-------
studies by numerous investigators who related the initial filter coefficient
as a power function of several variables with constant coefficients
= a Q d d£|) . Table 7 summarizes only the functional relationships
they reported for variables measured in this study. Both discrete and floc-
culent suspensions but no secondary effluents were used in the reported
studies. The coefficients varied for Q from -0.3 to 4, for dp from -1 to
0.3, and for d^ from -1 to -3.
FitzPatrick (1972) derived functional forms for the dependence of \ on
system variables. These are shown in Table 8, along with dependencies
derived from his filtration experiments using suspensions and media with well-
defined properties. He found that the coefficient varied for Q from 0 to
-1, for d from -.67 to 2, and for djjj from -1 to -3.
Straining
A large portion of the mass of the suspended solids in secondary efflu-
ents is of sufficiently large particle size so that straining will occur.
This has been suggested as a significant particle collection mode (Craft and
Eicholz, 1970; Tchobanoglous and Eliasson, 1970). Straining can occur at and
near the surface of the filter medium and deeper in the filter in the crevices
between adjacent medium particles (Payatakes and Brown, 1977). In design of
potable water filters, straining is avoided in order to utilize the full depth
of the filter medium and maximize run length. Straining does not have as
large a tendency to occur for potable water filtration as compared to direct
wastewater filtration because of lower solids loading and absence of coarse
particulates. A good understanding of the straining filtration process is
still not available at this time (Payatakes and Brown, 1977).
Hall (1957), using solely geometric arguments, developed a relation for
the filter coefficient indicating a -1.5 power dependence on media diameter
if straining was dominant. Craft and Eicholz (1970) indicated the filter co-
efficient should be proportional to (dp/d^)3'3 if straining dominates.
Heertjes (1973) and Tiller (1975) have developed models relating to formation
of cakes and precoats at the surface of filter septums, which may be closely
analogous to straining at the surface of granular media.
Kavanaugh (1974) explored possible relationships between cTp/djQ and the
number flux (Jjjn) °^ particles to the filter beds that would experience strain-
ing. His own data included both unflocculated and flocculated suspensions of
latex microspheres and TiOfe colloids, and data from the literature included
only unflocculated suspensions and unisize filter media. From a plot of the
data, he developed a semiempirical correlation for unflocculated suspension
particle size dp that would experience straining when challenging a unisize
granular bed of diameter ^ at concentration N0 and flowrate Q (or applied
particle flux of Q No = JNQ) :
dp * 0.25 dm JNQ -°'2 (7)
His correlation showed that, as should be expected, interstitial straining
depends not only on geometric variables (dm) but also on flowrate which is
37
-------
Table 7. Power function relationships for filter coefficient (\ )
After Wright, et al. (1970) (\Q = -Log £- = aQbdpCdmd)
Investigator
Iwasaki (1937)
Stein (1940)
Fair (1951)
Ling (1955)
Stanley (1955)
Hall (1957)
Straining
Settling
Mintz and Kristhul (1960)
Mackrle and Mackrle (1961)
Maroudas and Eisenklam (1965)
Ives and Sholji (1965)
Ives and Ison (1967)
Power Function Coefficient
Q
-0.3
-
-
-
-1.6
-
-
-0.7
-1
-1
-1
-4
and Pearson,
First Annual
d d
-1
-3
-1.7
-1.5
-2.5
-1.5
-1 -1
-1.7
-2
-
-1
0.3 -1.4
E., "Filtration Kineti<
Progress Report - SERL
College of Engineering, University of California, Berkeley,
1970. (References keyed to bibliography in report.
38
-------
Table 8. Power function relationships for filter coefficient (\ ) -
After FitzPatrick (1972) (\ = -Log £- = aQbd Cd d)
o C p ffl
Power Function Coefficient
Investigator Q d d
p tn
Experiment
NQrs* < 1.0 -.1 to -.5 1-5 -2.3 to -1.7
N_ > 10 -1 1.8-2.2
(jr s
Classical Theories
Interception 02-3
Gravity -1 2 -1
Diffusion -.67 -.67 -1.67
Straining-Hall(1957) 0 1.5 -1.5
Dimensionless gravity settling group, the larger the value of Np ,
the greater the effect of particle sedimentation in the capture0'"
process.
39
-------
proportional to >%fy
Shea and Mails (1971) developed a computer model of the microscreening
process for treatment of primary and secondary effluents from laboratory and
field studies. For the rotary drum units tested, straining across the screen
openings and intercepted mat of solids is clearly the dominant particle re-
moval mode. They found that the particle size distribution of the process
influent was the key parameter in determination of particle removal efficiency.
Flocculation properties of influent suspensions were not measured.
At present, no definitive theory has been developed and tested for strain-
ing in granular media filtration of secondary effluents. In addition, no the-
ory exists to determine a priori whether straining or in-depth filtration will
be the dominant particle collection mode. However, as explained in subsequent
sections, hypotheses are advanced that relate suspension properties to semi-
empirical clarification models.
SUSPENSION PROPERTIES
Little information is available in the literature on the role of suspen-
sion properties in the tertiary wastewater filtration process. Secondary
effluent suspensions may have a wide range of solids concentrations as well
as particle size and surface chemistry. Particle size distributions are dif-
ficult to obtain experimentally, particularly for the larger particles because
the number concentrations are so low and particle size may be altered in the
sampling or counting method used (Treweek and Morgan, 1977a).
Hsiung (1972) was the first to utilize refiltration techniques on second-
ary effluent suspensions to characterize performance of tertiary wastewater
filters. The refiltration technique was first developed by Lamer and Smellie
(1956), utilizing the Kozeny-Carman equation. This technique involves filtra-
tion of three or more samples of different volumes through microporous filter
paper and measurement of the time for refiltration of the filtrates.
Hsiung utilized the Carman equation for flow through a filter cake of
constant and known thickness:
V = APgAe3
(8)
where V = volume of sample filtered,
t = refiltration time,
AP = pressure drop across cake,
g = acceleration of gravity,
e = porosity of cake = 1 - (W/ALp) ,
A = filter cross sectional area,
W = weight of solids in cake,
L£ = thickness of cake
p = bulk density of cake,
40
-------
p, = viscosity of fluid passing through cake,
K = constant, and
S = surface area of particles in unit volume of cake.
If AP, g, A, K, and p, are constant, Equation 8 reduces to
V ' 3
Assuming all solids are removed by filtration, then
W = V C0 (10)
where C0 is secondary effluent suspended solids. From Equations 9 and 10:
AT n T J-1 /3
-------
or
d = k"" R1/2 G3/2 (16)
or
dp a E3 (17)
Then E is really a semiempirical factor proportional to the cube root of the
particle diameter. It follows then that it may be possible to characterize
the extent of straining by the refiltration parameter E or the measured mean
particle size. If this hypothesis is correct, inclusion of E in power func-
tion relationships for forms of the filter coefficient should result in a sig-
nificant improvement in correlations if straining is the dominant particle
collection mode.
Treweek and Morgan (1977b) measured the extent of flocculation of an IE.
coli suspension produced by the addition of cationic polymers. Methods used
to characterize the extent of flocculation were turbidity, electrophoretic
mobility, light scattering, Coulter counting, and refiltration methods. Of
the techniques used, they found that the refiltration method was the best
technique for determining the effectiveness of the polymers in flocculating
E. coli. Their results confirm that refiltration parameters characterize the
extent of flocculation but do not imply that they will also characterize fil-
ter performance.
MODEL DEVELOPMENT
It would be expected that particle attachment would depend on both parti-
cle and media surface properties. As shown in Chapter V, the full-scale
dual media filters evaluated in this study carry a large solids inventory
after backwashing. Pilot-scale filters can be backwashed much more effectively
although any wastewater filter cannot be completely cleaned since a microbial
film always stays on the media. It follows that particle-particle surface
chemical interactions may well be more important than particle-media inter-
action in defining particle capture and clarification efficiency.
A large number of suspension properties could be included in a correla-
tion of filter clarification efficiency. Possibilities include solution chem-
istry, total suspended solids, particle size, zeta potential, density, shape,
floe strength, and biological identification.
Inclusion of too many parameters makes correlation unwieldy and may
require an excessive number of measurements and a large data base. Several of
the above parameters which may be important are difficult or impossible to
quantify at present (e.g., biology, floe strength). Most will be distributed
and may not show a single central tendency. This suggests that in the short
run a more practical approach may be to use lumped or mean parameters to char-
acterize suspension properties. Examples include refiltration parameters,
mean particle size and variance, mean electrophoretic mobility, and surface
charge. The latter two properties are measured according to methods presented
by Novak, et al. (1977).
42
-------
If in-depth filtration is the mode of capture, theory suggests correla-
ting the filter coefficient \ as:
x a ' L Log ir = f(Q> dmj dp or E) ' (18)
where f is a power function with constant coefficients.
If filters are overdesigned with respect to depth and the bulk of the
solids capture occurs in the upper layers, even though in-depth filtration
may still be the mode of capture, it may be hypothesized that:
Log £- = f(C0, Q, dm, d or E). (19)
^o
In this study, the largest measured variation of C/CQ is a factor of 10 (0.06
to 0.6 for Phases 2, 3 and 4), thus it is just as reasonable to hypothesize
that:
C = f(Co, Q, dm, dp or E). (20)
If straining filtration dominates, it is not clear what dependence the
filter coefficient should follow since a theoretical interpretation is not
available at present. Nevertheless, straining or surface capture should be
confined to a smaller depth of the media as compared to in-depth particle cap-
ture. In that case, it would then be expected that Equation 18 would provide
a poorer correlation of the data than Equations 19 and 20, and that the
exponent for E in Equations 19 and 20 would be higher for straining as com-
pared to in-depth particle collection.
These hypotheses are tested in Chapters VIII and X, using multiple regres-
sion techniques with log transformations. In Chapter VIII, models with C, CQ,
Q, L, and dm are tested. Additional approaches for segmenting data based on
dominant in-depth or straining particle collection modes are discussed in the
following section.
CHARACTERIZATION OF STRAINING AND IN-DEPTH FILTRATION FOR FULL-SCALE FILTERS
From theory, headloss buildup vs. time should be linear or polynomial
for in-depth filtration (Ives, 1971) and logarithmic for straining or sieving
(Tiller, 1975; Cleasby and Baumann, 1974). Where total headloss data during
runs or backwash frequency data were obtained in this study, filter perform-
ance can be analyzed by postulating that straining is the dominant particle
collection mode for high headloss buildup and in-depth filtration for low
headloss buildup. In addition, straining is expected to be the dominant sus-
pended solids collection mode for the single medium sand filters at the Des
Plaines River and Romeoville plants.
Because of the inherent limitations of full-scale plant studies, the re-
sources required to monitor headloss buildup and particle collection at inter-
mediate depths in the filter media would have been prohibitive, within a rea-
sonable study budget. This precluded a more exhaustive study to characterize
particle collection mode. However, a significant effort to characterize col-
lection mode was undertaken in Phase 1 and reported elsewhere (Minix, 1979).
43
-------
As shown in Equation 17, E is proportional to the cube root of parti-
cle size. Since the role of straining increases with larger particle size
and E, both can be used to aid in separation of data for analysis. In
Appendix D, a dimensional straining group (Ngnro) is developed to use as an
aid in segmenting data (Chapter X).
44
-------
CHAPTER VII
FREQUENCY DISTRIBUTIONS OF SECONDARY
AND FILTER EFFLUENT PARAMETERS
INTRODUCTION
This chapter describes the results of statistical studies of secondary
effluent and granular media filter effluent parameters. Parameters included
were flow rate, BOD, and suspended solids. All data are based on flow pro-
portioned composite samples. The normal, log normal, rectangular, Poisson,
and negative binomial distributions were examined to determine their utility
for describing parameters for the Phase 1 data in 1976 for the Addison North
and South (Filter No. 2) plants. Kolomogorov-Smirnov (K-S) tests were used to
examine all Phase 1 data and the extent of autocorrelation was determined.
Insufficient data were obtained for these statistical analyses in Phases 2,
3 and 4.
Knowledge of frequency distributions is essential for the proper design
of data collection programs for survey and enforcement monitoring, and for
establishment of effluent limitations. There are several reports in the lit-
erature in which the frequency distributions for effluent parameters have been
characterized to some extent (Hoogerhyde, 1973; Eckenfelder and Roth, 1974;
Dean and Forsythe, 1975; Crandall, 1975). Hoogerhyde (1973) indicated that
tertiary filter effluent suspended solids are normally distributed. Dean and
Forsythe (1975) have argued that effluent parameters from a wide variety of
treatment processes should have a log normal distribution. There are also
other frequency distributions which might adequately characterize effluent
parameters.
STATISTICAL ANALYSES
In the context of this chapter, a frequency distribution is a tabulation
or a graphical representation of the relative frequency of occurrence of an
event corresponding to a continuous random variable. It may be represented
as a probability density function or as a cumulative distribution function.
Several common theoretical frequency distributions were examined to deter-
mine their utility in describing effluent quality parameters. These distri-
butions are the normal (Gaussian), the log normal, the rectangular (uniform),
the Poisson and the Negative binomial. The probability density functions for
these distributions are given in Table 9. In this table 0^ is the arith-
metic mean, ax is the standard deviation, mx is the geometric mean and 0*1^
is the geometric standard deviation. The Poisson parameter, v, is equal to
the mean and variance. The k parameter for the negative binomial distribution
is related to the mean and variance, as shown in the table, and is a measure
45
-------
Table 9. Standard probability models
Model
=F
*
Negative
l 1
Rectangular :
**
Probability Density Function
X -V
r,/-r\ V 6
P(x) - x ,
i r i fx -"Scl2!
FW"aVST^L 2K 1 J
X
rf*\ l ton -f * r l In I^X "l]2^
p(x) ' VT w p t 2 L», vfi JJ J
x V2TT a. lnx x
lnx
TCk + x) /mx NX / m \'k
p(3° ' r(x + i) rooVm + k ; • V1 ' r^y
X K.
, Jb^ . a s x s b
p(*H
[elsewhere 0
Parameter (s)
V
m , o
X X
£ rr
m , o
x lnx
,2 .°2
X X^
^ b-a
mx=a+ —
2 (b-a)2
ax - 12
x - a continuous random variable
k - denoted as a measure of extreme variance (Thomas et. al., 1956)
m and a - respectively the mean and standard deviation of the natural
logarithms of the raw data.
I i
x - defined for arbitrary range a to b.
46
-------
of extreme variance.
It will be shown later that data are sufficiently independent so the
application of theoretical frequency distributions is valid. A major objec-
tive of this portion of the study was to determine if effluent quality data
fit one of these common one or two parameter frequency distributions. It is
quite likely that better fits could be obtained if three or four parameter
distributions were tried. However, using additional parameters requires addi-
tional data and, in general, complicates any statistical tests to be made.
The multiparametric distributions are normally not used unless there is some
specific advantage to be gained.
The selection of the frequency distribution giving the best fit is dis-
cussed later. The selection was made using the Chi-square (x3) test, the
Kolomogorov-Smirnov (K-S) test, and a visual comparison. The parameters
for each distribution were calculated from the data. Then, using the proba-
bility density functions and the calculated parameters, expected frequencies
for observations were calculated. The x2 test indicates that the data actu-
ally fit the theoretical distribution at a given level of significance if the
sum of the differences between the observed and expected frequencies squared
divided by the expected frequencies is less than or equal to a given value of
a •£ distributed variable with degrees of freedom equal to the number of class
intervals minus one more than the parameters estimated for the theoretical
distribution.
The K-S test is also a quantitative goodness of fit test and is based
upon the deviation between the hypothesized cumulative distribution function
and the observed cumulative histogram. The cumulative distribution functions
for the observed data and the theoretical distribution are computed as well
as the difference between them at each level. The largest difference (posi-
tive or negative) is determined and, from that, the Kolmogorov-Smirnov Z.
The larger the value of Z, the less likely it is that the observed and theoret-
ical distributions are the same. A visual fit of the data is obtained by
comparison to a line generated from the parameters on normal and log normal
cumulative probability paper (probit scale). Data are compared to a line with
the same Xg0 (median) and CT or o~].nx as the case may be. Agreement may be
good near the central tendency of the data for either distribution but such
plots permit a judgment of which model fits the data best in the "tails" of
the distribution.
Autocorrelation is a measure of the relative dependence of the value of
a parameter to its value at an earlier time in a time series. To test for
the extent of autocorrelation, a plot of the value of the variable after a
time interval vs. the value at the beginning of the interval is prepared. The
autocorrelation coeffient (ra) is the bivariate correlation coefficient for
the plot. Thus an ra of plus or minus one indicates a linear increase or
decrease with time, respectively, and a value of zero indicates a total inde-
pendence on the previous value.
ANALYSIS OF ADDISON NORTH AND SOUTH DATA
Cumulative Frequency Plots
47
-------
The variables tested as well as sample means, medians and standard devia-
tions are shown in Table 10. Figures 7A and 7B show the distributions of
daily average filter flow rates (gpm/sf) for the Addison North Plant on a log
probability plot and a normal probability plot, respectively. The straight
line on each plot indicates where the data would be if the fit were perfect.
Both probability plots (Figures 7A and 7B) show a reasonable fit at inter-
mediate values of flow. Discrepancies exist between the data and the theo-
retical distribution at both high and low flows on Figure 7B. Superficial
filtration rates greater than 4 gpm/ft3 are associated with precipitation
events and could well indicate a separate distribution. The data are trun-
cated at 5 gpm/ft3 because higher flows are diverted to storage facilities for
subsequent pumpback to the plant. The actual raw sewage flow would cause this
value to be exceeded. The log normal probability plot indicates a better fit
throughout the entire range of the data but the break between flows less than
4 gpm/ft2 and greater than this value is still apparent.
In Figures 8A and 8B, the filter effluent suspended solids (C) data
from the North Plant are plotted as squares and the secondary effluent sus-
pended solids (C0) are plotted as triangles. From the log probability plot
( 8A ), it appears that both the C and C0 data give a similar level of fit.
Both C and CQ appear to show deviation in the tails on the normal probability
plot ( 8B )• C0 values above perhaps 70 mg/£ correspond to "upset" conditions
in the contact stabilization process and thus may be representative of a dif-
ferent distribution. From the normal probability plot it is clear that the
CQ shows much greater variability (slope of the line) than the C, but this is
not apparent from the log probability plot, i.e., the geometric but not the
arithmetric standard deviations of the two data sets are similar. Figures
9A and 9B present the plots of total secondary effluent BOD (Cob) an
-------
Table 10. Summary of simple statistics for Addison North
and South plants.
Geometri c
Plant Parameter*
South Flow
C0b
Sol.Cob
cb
Sol.Cb
co
C
North Flow
cob
Sol.Cob
Cb
Sol.Cb
Co
C
Cob* -
Cb
C
Co
Sol. -
Standard Standard
Units Mean Deviation Median Deviation
(gal/min/ft2) 2.06 0.46
(mg/1) 41.8 17.1
(mg/1) 8.6 3.6
(mg/1) 9.7 4.9
(mg/1) 6.0 3.1
(mg/1) 37.5 16.9
(mg/1) 9.1 3.9
(gal/min/ft2) 3.76 0.62
(mg/1) 33.90 12.7
(mg/1) 7.5 2.7
(mg/1) 5.1 3.9
(mg/1) 3.8 2.9
(mg/1) 43.9 18.9
(mg/1) 6.6 3.7
1.95
39.5
8.0
10.0
4.9
32.0
8.2
3.60
31.0
7.0
3.5
2.2
39.0
5.3
Secondary Effluent Biochemical Oxygen Demand
Filter Effluent Biochemical Oxygen
Filter Effluent Suspended Solids.
Secondary Effluent Suspended Solids
Soluble, passing glass filber Reeve
Demand (5
•
Angel R
0.21
0.45
0.45
0.58
0.49
0.40
0.52
0.17
0.39
0.47
0.78
0.79
0.44
0.61
(5 day).
day).
filter
Coefficient
of Variation
0.22
0.40
0.42
0.50
0.53
0.45
0.43
0.16
0.37
0.37
0.77
0.77
0.43
0.56
49
-------
Q_
CD
O
.01 .1 1.0 10.0 50-0 90.0 99.0 99.9 39.99
PERCENTfiGE LESS THflN iNDICflTED VflLUE
Figure 7.
.1 1.0 10.0 SO.O 90.0 99-0 93.399.99
PERCENTflGE LESS THflN INOICRTED VPLUE
Cumulative frequency distribution of filter flow rate,
Addison North plant.
50
-------
SUSPENDED SOLIDS (MG/L)
12 S 10 20 60 10
X
<*
ef
X
-fl-
it
Ax
t\\
s
flu
^*
ent
/
jnt
^
f*
^
S*
J"
.«*
^0
V
x
^
**^
f.
y
,6
— i-r
A
.01 .1 l.Q 10.0 50.0 90-0 99-0 99-9 99-99
PERCENTflGE LESS THflN INDICflTED VflLUE
Q
®
inf'. u<
e£
nt
x ent
.1 1.0 10.0 50.0 90.0 99.0 99-9 99.39
PERCENTflGE LESS THflN INDICRTED VRLUE
Figure 8. Cumulative frequency distribution of total suspended
solids, Addison North plant.
51
-------
§ F
s
a
ef
fl
rf
*^"i
ue
it
la
4
1 3i oy-^
«rf2
sa-|
\
J
^*f
B.
Figure
.01 .1 1 .0 1Q.Q 50.0 90.Q 39.0 99-999.99
PERCENTfiGE LESS THfiN INDICfiTED VflLUE
9. Cumulative frequency distribution of total BOD,
Addison North plant.
52
-------
a
cu
llf
a
o
CD
eff l|uen(l
.01 .1 1.0 10-0 SO.O 90.0 99-0 99.999.99
PERCENTRGE LESS THRN INDICRTED VflLUE
CD
o
§ 2
co
fluent
^r
e if lien
.01 .1 1.0 10-0 50.0 90.0 99.0 99-9 c».j9
PERCENTRGE LESS THRN INDICRTED VflLUE
Figure 10. Cumulative frequency distribution of soluble BOD,
Addison North plant.
53
-------
co
CL.
CD
.1 1.0 10.0 50-0 90.0 99-0 99.9 99.99
PERCENTRGE LESS THRN INDICflTED VflLUE
B
.01 u i.g to.a so.o 90-0 99-0 99.999.99
PERCENTfiGE LESS THPN INDICfiTED VflLUE
Figure 11. Cumulative frequency distribution of filter flow
rate, Add!son South plant.
54
-------
o
SUSPENDED SOLIDS (MG/L)
12 5 10 20 50
•
>
X
s*
s.
ir
^
r^
^
tlu
f*
<
snt
r*
r
*•
^
La
f
P4I4
5^
ff
*^
IK
U€
^
ot
a*-
ix^1
X
(
^^?
,X
X
a
A
.01 .1 1.0 10.0 50.0 90.0 99.0 99.9 99.99
PERCENTflSE LESS THflN INOICflTED VflLUE
8
C3
s
O
I
A
In
:lue
A
^\
nt
•JfT
,,T
p*
p*
V^
-f
f*-^
^
A
***
.01 .1 1.0 10-0 50.0 90.0 99.0 99-999.99
PERCENTAGE LESS THflN INDICflTED VflLUE
figure 12. Cumulative frequency distribution of total suspended
solids, Addison South plant.
55
-------
g
g
QD
x*
infl lent
„•<••
:*
tt-
.01 .1 1.0 10.0 50.0 90.0 99.0 39.9 99-99
PERCENTflGE LESS THflN [NOICflTED VflLUE
o
00
s
o
T
a
•*!
^
r
ii
,>^
1
fid
^-a
sni
I—1 '
**1
r**
1 i a
'*•-
tf-n
•
^
A
r^1
!
1^-1
r
!
t
i
1
a
e
o
OJ
e::flienl
B
• 01 .1 t.Q 10.0 50.0 90.0 99.0 99.399-99
PERCENTRGE LESS THfiN INDICRTED VflLUE
Figure 13. Cumulative frequency distribution of total BOD,
Addison South plant.
56
-------
en
co
in
(\j
a
I a
a
§ 2
o
CO
irfl
lent
flient
.01 .1 1.0 10-0 50-0 90.0 39.0 99.399.99
PERCENTAGE LESS THflN INDICATED VflLUE
influeit
ef
snt
B
.01 .1 t.Q 10.0 50-0 90-0 99.0 99.999.99
PERCENTRGE LESS THflN INDICflTED VRLUE
Table 14. Cumulative frequency distribution of soluble BOD,
Addison South plant.
57
-------
Table 11. Goodness of fit summary, Addison North and
South plants.
Parameter
North Plant
Flow
cob
Sol.Cob
Cb
Sol.Cb
Co
c
South Plant
Flow
Cob
Sol.Cob
r ++
cb
Sol.Cb
Co
C++
K-S*
.41
.63
.86
.41
.18
.76
.52
.60
.66
.17
.21
.52
.81
.74
LOG NORMAL
CHI-SQ"1" PLOT
XX
.41(2) X
X
.20(2) XX
.37(2) X
.41(4) X
.36(8) X
.39(5)
.25(2)
.13(3)
.45(2) X
.42(3) X
.03(2)
NORMAL
K-S*
.13
.45
.63
.05
.05
.34
.10
.17
.89
.22
.29
.07
.24
.37
CHI-SQ+
.17(2)
.40(2)
.50(2)
.04(2)
.33(4)
.50(2)
.10(3)
.21(2)
PLOT
X
X
XX
X
X
Kolmogorov Smirnov (K-S) - 2 Tailed Test Significance Level.
* Chi-squared test significance Level for degrees of freedom
in parentheses.
++ Bimodal
X Good visual fit to line with same nfx and Ox or Cf£nx
XX Better visual fit of two distributions, but still not very good.
58
-------
soluble Cg^ for Addison South and North), a somewhat higher but still low sig-
nificance fit for the Poisson distribution was obtained by the -^ test. C at
Addison South showed a weak fit (29 percent significance level). Soluble BOD
parameters at both plants showed the best fit to the Poisson distribution
(66 percent significance at 2 degrees of freedom for the South plant and 73
percent significance at 2 degrees of freedom at the North plant). In most
instances, when a greater than 20 percent significance is recorded for both
the K-S and Chi-squared test, the visual fit is in agreement. If the K-S
and Chi-squared significance predictors are widely discrepant, the distribution
showing the higher K-S significance corroborates the visual fit. These lines
of observation may be seen by examining Table 11.
Relative frequency histograms of C0 and C0^ data for the North and South
plants are presented in Figure 15. This figure helps, to some extent, to
interpret the results given in Table 11. By and large, the data that fall
in the tails of the distributions are most important in determining the good-
ness of fit, and those data sets that show a reasonable fit of the tails on
the appropriate probit plot corroborates whether the fit is better to normal
or log normal statistics. In two cases in particular, this seems not to be
followed: for the North Plant total C^ and C are concluded to show a better
fit to the log normal distributions although relatively significant Chi-
squared values were obtained. This may be explained by looking at the fre-
quency histogram on Figures 15C andlSD. Both are skewed to the right. The
absence of observations at negative values of the parameters result in com-
paring observed and expected frequencies for the normal distribution only in
the range where agreement is good. Thus the resulting Chi-squared value tends
to be abnormally small and the corresponding significance of fit is too high.
In this case the fit to a normal distribution has 50 percent significance at
2 degrees of freedom for total C^ and 33 percent significance at 4 degrees
of freedom for C.
Bimodal Distributions
Filter effluent data for Addison South show generally poor fit to either
normal or log normal statistics. This is because both of these distributions
appear bimodal, as shown in Figures 15A and 15B. To test the hypothesis that
the data are bimodally distributed, the data were truncated with the aid of
collateral data, i.e., data other than the parameters to be examined were cut
at a given level. Corresponding filter effluent data were truncated according
to chronology of data (several different ways) or mean values of process flow
(Q), solids and BOD loading (C^Q and CO^Q) and C^ and C levels. Strongest
evidence that the bimodal character is the combination of two unimodal dis-
tributions derives from truncation of Cofo and Co data at their mean values.
First mode C and C^ appear to be associated with respective values of Co and
Coij at less than their mean values. Likewise, second mode C and Cj, are asso-
ciated with respective Co and CQJJ at greater than their mean values. In gen-
eral, the number of data were insufficient to permit meaningful quantitative
goodness of fit tests on the truncated distributions. Furthermore, simple
bivariate correlation coefficients of q, and Cob (0.53) and C with Co (0.38)
do not provide sufficient information to assess whether Cob and CQ are the
dominant determinants of tertiary effluent parameters for the South Plant.
59
-------
a FREQUENCY FREOIIENCY
FREQUENCY
o S 8
)
i
\
1
i i i 1
3 MG/L PER CELL
—
i i
-
0 5 10 15 20 25 3D
C, -MG/L
b
10
5
0
1 /
/F
/
,
, N
x. 2 MG/L PER CELL
X^ V *
(A) ADD I SON SOUTH PLANT - C,
/ x
1
/
/\.
V
\
FREQUENCY FREQUENCY
t— i— t— ro ^
> >^1CS *J1O C3 CDO
/i
/f;'
\J' ,
^ 3 CiG/L PER CEU
r^ \
\s
^^
5 10 15 20 25
II
rT.
2 MG/L PER CELL
&> # 30 % "" 5 10 15 20 25
C-MG/L
HISTOGRAM ^ ADDISON SO(JTH pLANT . C HISTOGRAM
3 NR/L PER CELL
\ LOG NORMAL .
^ unnutt "
vV •
POISSON
x^T^^"^^sL i i
0 5 10 15
C, -MG/L
B
FREQUENCY
S E
/
1
1
i
i
^
/
20 2!
\ \
"Vvi 2 MG/L PER CELL
0 5 n Wun/T15
20 25
20
LU
"• 10
0
C
20
3 10
u.
0
/ / \
' ) ™' "\
'.'/
i
\ '
. \ 3 MG/L PER CELL
\N\ LOG NORMAL -
\ \ FORMAL
_1\ PO i «nN
\ N
5 10 15 20 2
C-MG/L
//$
^/J7
t .
Ni 2 MG/L PER CELL
\\
XX^u^
5 10 15 20 25
(0 ADOISON NORTH PLANT - C, HISTOGRAW
b
C-MG/L
(D) ADDISON NORTH PUNT - C HISTOGRAM
Figure 15. Relative frequency distributions for C, and C,
Addison North and South plants.
60
-------
An attempt to fit the data to a bimodal distribution might be tried with
a three parameter empirical model developed for particle size analysis and
used in coagulation studies (Dallavalle, et al., 1951). The frequency distri-
bution would then be given by an equation similar to those shown in Table 12.
Several methods indicated in Table 12 could be used to evaluate the parameters,
but many more data than the 60 or 70 we obtained would be needed to obtain a
good fit. It must be stressed that the distribution so derived would be
completely empirical and without any relation to the moments of the data.
STATISTICAL ANALYSES FOR ALL PHASE 1 DATA
A summary of simple statistics for all Phase 1 data is given in Table 13.
For the variable flow plants, statistical analyses were conducted using raw
sewage flow rather than filtration rate. The results from statistical analy-
ses for plants other than Addison North and Addison South must be interpreted
with caution because data were not obtained during process upsets or when the
filters were not in operation because of high flow. If data had been taken
during these periods, it could be expected that a better fit would be obtained
for the log normal distribution. A chronological summary of the sampling pro-
gram is given in Table 4.
Data for Harrington, Des Plaines River, Marionbrook and Romeoville were
obtained during two or more seasons of the year. From Table 13, the high
coefficient of variation for flow for the Des Plaines River and Marionbrook
can be attributed in part to seasonal variations. The high coefficient for C
of 0.79 for Marionbrook suggests that particle collection mode may have changed
during the study. For several plants, the coefficient of variation was higher
for C than for Co. Again, this suggests possible changes in particle collec-
tion mode during the sample period. The high coefficients of variation for Cj,
of 0.79 and 1.1 for the Addison North and Des Plaines River plants, respec-
tively, may be caused by the accuracy of the BOD test since effluent BOD is
very low.
Kolomogorov-Smirnov Tests
A summary of K-S goodness of fit data for the normal and log normal dis-
tributions for all Phase 1 plants is shown in Table 14. Data are not shown
for the Poisson distribution which gave a very poor fit. For plants where flow
was varied to individual filters, K-S and autocorrelation tests were made only
for flow data from 24-hour runs. Tests for other parameters were made using
all data. Insufficient BOD data (less than 20) were obtained for the variable
flow studies at Addison South, Lisle, and Marionbrook plants for statistical
analyses.
The 1975 Addison South flow data show a higher significance for the normal
as compared to the log normal distribution, the opposite result obtained for
the 1976 data. Sampling periods for each data set cover approximately the
same months. Similar results were obtained in 1975 as compared to 1976 for
Co and C, with the log normal distribution providing a better fit.
The Barrington and Lisle data show a good fit for flow to both distri-
butions. The flow data for the Des Plaines River, Lake Zurich, and Marion-
brook plants do not show a good fit for either distribution. As discussed
61
-------
Table 12. Empirical Fit of Bimodal Data
f/432 \ "I
Frequency = exp I ( ~ SA x +a_x H-a^x + a x a )
r / 24 2 \ 1
or Frequency = k exp \ ~ a x +px +qx)
where k = exp
- ar J
The parameters a, p, q and k may be obtained by the 1) method of
moments, 2) least squares method or 3) five ordinate method.
62
-------
t
CO
CO
•o
l-<
cu
to
CQ
1
CO
0
4J
00
4J
CO
4-1
OB
CU
O.
S
T-I
CD
VM
O
£
S3
c
3
cn
•
en
1-1
cu
i—4
J2
a
H
-^ u
00
g
1
cj a
^^ ^>
1?
f
o
U S
•5 cj
CO
E
i
.0.
U 2
•5 &
M)
E
A
en
^— '
^ CJ
£
f^t S
14-1
O CO
» o •* O CM
i-i r-4 CN f-l
CM cn cn CM
-d- * en
CM CM
r-i cn
i— 1 r-l <• OO
Or o- *
• r-li—l «f-li—( a • • •
o^ JON ^^ oo P^ r***
CTN m oo r-i r-i I-H cn o ON in
cncnensososomininm
00-— (ONCMO>r-iCX3
CO O O
CD rJ T-I CD
as T-I 03 5
rJ r3 S Oi
o
b
*~^
^
CO
1
ao
CD
03
2
c
3
CD
4_l
C
03
r-l
a
3
-4
IW
o
T-4
93
T-I
9t
O
8
u
^_^
en
^X
at
O
u
1-4
4-1
J-*
14
CO
CM
^
63
-------
0)
en
05
3
M-l
O
01
CO
01
C
T)
O
O
00
o
M
O
!-<
S
Cl)
o
c
CJ
•1-1
Cf
en
CO
•^
m
H
cn
i
^
_j
CJ
,-3
CJ
z
fj
u
z
kJ
0
z
* *
-I: ,J
o
[fa *
z
M
0)
• cc
Z CJ
JJ
c
1^
CN -j- *> i -^ uO
en r* en CM
^H -H i r- oo
en o <~* ^H >r^co-d'COCN
fO^-i
ej o O
p— i -.-I D
« u e
J X as
01
c
o.
3
0
0)
I-l
ea
T-l
ct)
>
0
r-l
2 3
S o
U HI
OC 00
J 3
0)
h-J M
a
S o
H T3
z n
' 3
* *
64
-------
earlier, those plants were not always sampled during process upsets and high
flow periods.
The secondary effluent and filter effluent BOD data are about equally
split between providing better fits to the normal as compared to the log
normal distributions. Except for the Harrington plant, all data sets provide
a better fit for C0 to the log normal distribution. Ten of the 17 data sets
for C show higher significance for the log normal distribution; five are about
equal and two show a better fit for the normal distribution. In addition, the
C data show a better fit for lower values of the coefficient of variation.
Extent of Autocorrelation
In autocorrelation tests, the time interval lag must be selected. Auto-
correlation coefficients are shown in Table 15 for a time lag of one composite
sample period. For an interval of one, it would be expected that ra would be
positive unless a change occurred in the operating mode. In general, Q shows
the highest autocorrelation as would be expected because of seasonal changes
during the sampling periods caused by variations in sewer system infiltration
and inflow. From Table 15, influent and effluent BOD have the lowest auto-
correlation.
A plot of daily changes in C and Co for the Addison South Plant is shown
in Figure 16. The values of ra are 0.34 and 0.20 for C and Co, respectively.
Figure 16 shows a decreasing trend for both C and CQ. The decrease in Co can
be attributed in part to the decrease in plant flow (ra=.58) and increase in
efficiency of the final sedimentation tanks during the sampling period. The
values of ra for C and C0 for other plants were generally less than 0.6.
It can be concluded that the extent of autocorrelation is low enough so
that the basic assumption that the data are independent is reasonable.
DISCUSSION
The results of the statistical tests to determine the best frequency dis-
tribution to use for these data were rather encouraging. It is clear that the
Poisson and negative binomial distributions do not appear to offer any advan-
tages for analysis of this type of'data. What choice exists is between the
normal and log normal distributions. If the spread of data is small, there
is not much difference between these distributions. For most sets of data,
the log normal distribution should give a better fit because the logarithmetic
transformation greatly condenses the range of the data.
The results of the ^3 and K-S tests did show a clear cut preference for
the log normal distribution for all parameters at Addison North except soluble
CK. This may reflect a more stable operation at the North plant. Based on
X and K-S tests, the South plant data appear split between normal and log
normal statistics. However, closer examination shows that two of the para-
meters C], and C are bimodal distributions and not really best fit by the normal
distribution. Therefore, the log normal distribution appears to fit the same
parameters at the South as the North plant, excluding the two that are bimodal.
No explanation is offered as to why soluble C^ should be the only parameter
with a better fit to the normal than log normal distribution.
65
-------
Table 15. Autocorrelation coefficients - Phase 1.
Plant
Non-Variable Flow
Addison North
Addison South
Harrington
Des Plaines R.
Romeoville
Variable Flow
Addison South
Low Q
Med Q
High Q
Lake Zurich
Low Q
Med Q
High Q
Lisle
Low Q
Med Q
High Q
Marionbrook
Dry Weather-
Jan-Feb. '76
Low Q
Med Q
1st half
Wet Weather -
Low Q
Med Q
High Q
2nd half
Wet Weather -
Low Q
Med Q
High Q
No. of
Cases
73
70
58
60
36
61
51
64
24
48
22
32
29
35
21
31
38
24
26
41
32
16
Q C , C, C
x ob b o
.69 .09 .17 .48
.58 .05 .03 .20
.54 .29 .11 .07
.13 .13 .26 .37
.54 .35 .02 .36
-.02 - .51
.72 -.04 -.02 .52
.20 .35 .53
.36
.34 .26 .20 0
.51
.21
.52 - - .53
.24
.52
.57
.25
.04** - - .14
.06
.35
.37
.19
C
.23
.34
.30
.21
.25
.54
.31
.54
.59
.02
.05
.67
.55
.80
.60
.52
.43
.40
.44
.21
.03
.16
*Based on raw sewage flow for variable flow plants
**Based on all data
66
-------
-------
The distribution of effluent quality parameters is the result of complex
interactions among a number of factors. It seems reasonable to assume that If
the flow rate, wastewater strength, and operating conditions were about the
same for several days, the effluent quality should be about the same for that
period of time. "About the same" in this context means that small variations
from day to day will make the effluent quality parameters slightly higher or
slightly lower than the average values. This suggests that over short periods
of time, the effluent quality parameters should have normal distributions.
Wastewater flow and strength and operating conditions do not remain "about
the same" for very long. A rainfall will change many factors, seasonal tem-
perature differences change the biota of the secondary treatment processes,
and operational conditions change abruptly due to many occurrences such as
digester upsets, failure of mechanical equipment, etc. An upset may occur
for a short period of time producing one or two days of high effluent BOD and
suspended solids. On the other hand, a change in conditions may persist a
long enough period of time for the processes to stabilize and start producing
"about the same" effluent quality from day to day. This line of reasoning
suggests that over a long period of time, the effluent quality data would fit
a series of normal distributions each having different means with perhaps a
few extremely high values from upsets. Data from a series of normal distri-
butions with different means very well could appear to have a log normal
distribution (Aitchison and Brown, 1957).
From the parameters for a fitted theoretical distribution, the frequency
and corresponding values of secondary effluent suspended solids can be deter-
mined directly. However, the particle collection mode for granular media
filters may not be the same over the range of Co. Thus it is not possible to
calculate performance directly from the fitted distribution with a high degree
of precision, although the models developed in Chapters VIII and X for in-depth
filtration and straining can be used to estimate an expected range of effluent
quality if suspension properties are known.
68
-------
CHAPTER VIII
RELATIONSHIP OF FILTER CLARIFICATION
EFFICIENCY TO CONVENTIONAL OPERATING PARAMETERS--
PHASE 1 DATA
INTRODUCTION
This chapter compares filter clarification efficiency with conventional
parameters such as secondary plant operational variables, flow, influent solids
concentration, influent solids loading, and filter media specifications. Oper-
ational mathematical models are developed and tested using data from Phase 1
of the study.
RELATIONSHIP TO SECONDARY PLANT OPERATIONAL VARIABLES
Several of the operational variables used for design and control of the
activated sludge process relate to some extent to the characteristics of mixed
liquor suspended solids. The sludge volume index (SVI), defined as the volume
in m£, occupied by 1 gm of solids after settling for 30 min. , is a measure of
the resistance of solids to thickening. The food to micro-organism ratio (F/M)
is a measure of organic loading. F/M is defined as the ratio of BOD added
per day to mixed liquor suspended solids under aeration.
Kreissl (1974) reported that for studies at the EPA Blue Plains Pilot
Plant in Washington, D.C., both secondary effluent and dual media filter ef-
fluent suspended solids (SS) increased when SVI increased. Gulp and Gulp
(1971) reported that higher effluent SS are expected for high F/M ratios.
Data on plant operational parameters were obtained from monthly reports
to the State of Illinois and included raw sewage BOD, aeration tank suspended
solids concentration, sludge volume index, and aeration tank dissolved oxygen
concentration. Plots of C and C/CO vs. F/M and SVI are shown in Figures 17
and 18, respectively, for the Addison North and South plants. No correlation
of filter performance with these plant operational variables is apparent.
Multivariate correlations including filtration rate and filter solids loading
yielded the same result and are not reported here. Mean values of plant oper-
ational variables and a bivariate correlation matrix with filter performance
is given in Appendix A (Tables A-l and A-2).
ANALYSIS OF AVERAGE PERFORMANCE DATA
Plots of C/C0 vs. average flow rate (Q) and average influent solids load-
ing (C0Q) are shown in Figures 19A and 19B respectively, for plants where flow
was varied to individual filters. Although the independent variables Co and
69
-------
0.6
0.4
o
O
O
0.2
0
Phase I
o Addison North
• Addison South
9
o
0^0
°0
o o
0
.05
0.6
0.4
o
O
O
0.2
0
50
,10 .15
F/M
.20
.25
Phase I
o Addison North
• Addison South °
o o
0 Q 00°
o
oooo0
100 150 200 250
SVI
Figure 17. C/C vs. food to microorganism ratio and sludge
volume index, Addison North and South - Chase 1.
70
-------
i
O
20
15
10
0
o Addison North
• Addison South
o
o
ho
coo
o o
0 .05 .10 .15
F/M
.20 .25
20
15
I1 10
i
O
5
0
B
O O
V
•••
o o
CO
ooo
CO O
o o
>
o o oo o
ooo o oo o
lO CO OOOO
O CO
oo o o
oo o o
0 50 100 150 200 250
SVI
Figure 18. C vs. food to microorganism ratio and sludge volume
index, Addison North and South - Phase 1.
71
-------
0,6
0,4
0°
0,2
0
0
2
o MB
t
3
Q-GFWSF
6
0,6
0,4
0,2
0
0
i i
• LI 0,01 -I
o LZ -0,01
• AV 0,24
o MB 0,11
CQQ-LB/SF/D
Figure 19. Ave. C/C vs. ave. flow and solids loading.
72
-------
Q varied widely^ as did C/C0, the average performance (C/CO) and effluent sus-
pended solids (C) appear linear with flow rate. For individual data, it is
not possible for both relationships to be linear unless Co is constant and Co
is constant for each data point for Lisle and Addison South plants. For those
plants, data were obtained for low, medium and high filtration rate simulta-
neously. The more nonlinear dependence apparent for the Lake Zurich (LZ) and
Marionbrook (MB) plants may result from the fact that the three filters (low,
medium and high rate) were not always in operation at parallel times and thus
C0 is not constant for each.
Figures 20A and B show average effluent suspended solids (C) as a function
of flow and solids loading for the Addison South and Lisle plants. Data for
the Lake Zurich and Marionbrook plants are excluded from this plot because Co
was not constant, a_s already mentioned. As with average data for C/CO, a lin-
ear dependence of C with Q and C with COQ is apparent. Linear correlation
coefficients for all pairs of data in each of the data sets is shown on
Figures 19 and 20. The Addison plant (AV) shows highest (r = 0.70) corre-
lation of C with C0Q. Remaining data sets on Figure 20 show marginal corre-
lation (r = 0.55) and some data sets in Figure 19 show no correlation indivi-
dually, e.g., Lake Zurich and Lisle on Figure 20.
In this phase of the work, no routine quantitative measurements were made
to determine particle size distributions. However, it was observed that the
suspended solids for the Lake Zurich and Marionbrook plants secondary efflu-
ents were more finely divided compared to the other plants. This is not sur-
prising since both plants have ponds ahead of tertiary filtration. The secon-
dary processes for the Lake Zurich and Addison South plants consist of activated
sludge and trickling filters in parallel. It would be expected that suspended
solids in trickling filter plant effluents would be less flocculent compared
to activated sludge plant effluents.
Interpretation on theoretical grounds (gravity settling and low hydro-
dynamic shear) suggests that performance (1 - C/CO) should increase more than
the linear relations suggested at low flow or solids loading. Even so, a
finite intercept C or C/CO is expected at zero flow rate suggesting that a
portion of the secondary effluent suspended solids cannot be removed by granu-
lar media filtration without alteration of suspension characteristics, e.g.,
particle size and charge by chemical addition. However, this point is aca-
demic since filters are only economic at higher flow rates.
Graphs of average C/CO as functions of average temperature, total media
depth, and equivalent media diameter (de) are shown in Figures 21, 22A and B
respectively, where de is the equivalent media diameter (dso) that would give
the same clean bed headloss as the sand and anthracite media sizes at the same
total depth. Values of de were calculated from Kozeny's equation for flow
through porous media, as shown in Appendix B. For plants where flow was
varied, only data for the medium flow rates are shown.
If one excludes the single medium sand filters at the Des Plaines River
(DP) and Romeoville (RO) plants, Figure 21 shows a strong dependence of filter
performance on temperature (broken line) over the relatively narrow tempera-
ture range of 55° to 67°F. It is believed that this is an artifact resulting
73
-------
30
20
10
0
i I
FIG, A
• AV
• LI
0,52
0,49
0
Q-GFWSF
0
Figure 20. Ave. C vs. ave. flow and solids loading.
74
-------
0,5
i I I i i i 1 r
Vz
\
\
\
\
\
0,3
c^°
0,2
0,1
0
RA\
Vv
RO "MB
\ "LI
DP
\
\
•AN\
\
50 55 60 65 70
TEMP-OF
Figure 21. Ave. C/C vs. ave. temperature.
75
-------
0,6
0,4
0,2
0
I i I I I I I I
FIG, A
• LZ
•BA BAV
RO m*u "MB "AS
DP BAN
i I
0 10 20 30 40 50
TOTAL feiA DEPTH - INCHES
0,6
0,4
BA
0,2
0
1141111
FIG, B
• LZ
AS " AV
LI
DP
AN
J i I | i i i i
0 0-5 1.0 1,5 2,0 2,5
EQUIV, feiA DIA, - M M
Figure 22. Ave. C/CQ vs. media depth and equivalent media
diameter.
76
-------
from "chance" in the selection of plants and times of the year for sampling.
Alternatively, if one realizes that the Lake Zurich and Addison North plants
had the finest and coarsest particle size, respectively, in the filter influ-
ent, then their average C/CO ordinate is fixed basically on this factor.
Eliminating those plants and realizing that other plants had comparable filter
influent particle size, C/C0 is independent of water temperature.
If suspension properties are constant, an inverse dependence of perfor-
mance on media depth would be expected if in-depth, rather than surface fil-
tration, is the dominant particle collection mode. Surface cake and/or siev-
ing filtration are believed to be the dominant modes for the sand filters at
the Des Plaines River and Romeoville plants. If suspension properties are
constant, higher removal efficiency should be obtained for finer media whether
straining or depth filtration is operative. However, if the bulk of the removal
is of coarse solids, little dependence of removal efficiency on media size or
depth will be apparent, and residual suspended solids will be a relatively
constant fraction of influent solids. With the information that the Lake
Zurich and Addison North filter influent suspensions were coarsest and finest,
respectively, for any plant tested, then Figures 23A and B show that average
removal of the remaining plants (with comparable influent suspensions) is inde-
pendent of media depth and equivalent media diameter.
ANALYSIS OF INDIVIDUAL PLANT DATA
Bivariate correlation coefficients for plant performance and selected
operating parameters are shown in Table 16. Although performance showed a
strong correlation with flow and solids loading for average data where flow
to individual filters was varied, much poorer correlations were obtained for
individual data.
Correlation coefficients for C to Co varied from 0.18 to 0.71. Figures
23A and B show C as a function Co for the Addison South (Filter No. 2) and
the Romeoville plants, respectively.
If solid breakthrough occurs before limiting headloss is exceeded or
before filters are backwashed (manually or by timers), it would be expected
that clarification efficiency would decrease with increasing run length.
Except for the Lake Zurich plant, very poor linear correlations were obtained
for C and C/CO with run length (RL"). This is partly explained by the fact
that the Lake Zurich filter influent had very fine suspended solids and fully
developed depth filtration took place. Thus the active deposition zone (clog-
ging front) advances through the filter with time giving rise to a break-
through phenomena. Furthermore, run length was varied over a wider range for
this plant compared to other plants where filters were manually backwashed at
the start of a run. Also, the filter influent pumps at Lake Zurich cycled on
and off, thus possibly providing a greater opportunity that solids could be
sloughed from previous deposits and breakthrough experienced.
Except for plants where flow to individual filters was varied, linear
correlation coefficients for C and C/CO to Q were less than 0.3. This suggests
that, although flow is an important design parameter, day to day changes are
too small compared to other variables, e.g., suspension properties, to result
in measurable changes in filter performance. For plants where flow was varied
77
-------
20
15
10
0
ADD, SOUTH PLANT
= 0,38
FIG, A
•* •
• • • •
• • •
0 20 40 60 80 100
CO-MG/L
20
15
10
0
RMVILLE PLANT
R = 0,66 •
FIG, B
0 20 40 60 80 100
Figure 23.
CO-MG/L
C vs. C , Addison South (#2) and Rosieoville.
78
-------
I
CO
•u
co
•d
a)
a
c
V
a
-u
CO
8
•H
U
•H
0)
O
O
4J
CO
r-l
01
M
rl
O
u
CO
vO
rH
0)
CO
H
o2
CJ
--- o
U 4J
Of
o
00
0
-- o
O *J
o
^- o
CJ *J
o
s
o
o
Q
0 0
o
4J
„ o-
o
o
4J
o
o o
u
g
s
Si" CM r-l
O CM O
• • •
1 1
r-l r» sj
•* •* CM
• • •
1 1
VO 00 00
CM CO "tf"
^ 1
JS X! *
4J 4J ^
P 3
O O •
Z C/3 CO
a c c
o o o
OQ co co
•r4 T4 i-l
^3 ^3 ^3
^4 ^n ^^
m r-i oo CM -* m
r-l r-l rH O O r-l
• •••••
1 1
m O r-l r-l l-l VO
CO CO O O r-l CM
• •••••
1 1 1
r^ o r-i s* r-i in
CM r-l sj
iH di (U O O
U (U rH i-l fi
IH CO J2 CO S-l S
CO 0} C0 *r4 Cd O
79
-------
to individual filters, correlations for flow increased significantly for most
data sets. A plot of C/CO vs. Q for the variable flow studies at the Addison
South plant is shown in Figure 24.
A plot of G/CO vs. solids loading is shown in Figure 25 for the Lisle
plant. No correlation of these two parameters is apparent. A similar result
was obtained for other plants, as shown in Table 16, where bivariate corre-
lation coefficients are less than 0.5.
The highest correlations of filter clarification efficiency were obtained
for C vs. solids loading for the variable flow plants. Plots of C vs. COQ
are shown in Figures 26 to 29 for the Addison South, Lake Zurich, Lisle and
Marionbrook plants, respectively. The data for the Marionbrook plant are for
a period of dry weather flow early in the testing program where the correlation
coefficient is 0.71. The Addison South data also show good correlation with a
coefficient of 0.70. Wider scatter was obtained for the Lake Zurich and Lisle
plants.
TECHNIQUES FOR SEGMENTING INDIVIDUAL DATA SETS
The relatively poor correlations obtained between clarification efficiency
and operating parameters can be attributed in part to other variables including
characteristics of secondary effluent suspended solids and possible changes in
the dominant particle collection mode during the sampling periods. In an at-
tempt to improve correlations, data were segmented by: (a) a cumulative data
analysis technique, (b) grouping by CQ ranges, and (c) examining the relation-
ship of solids capture to headloss increase.
Cumulative Data Analysis
Since straining can be characterized by high headloss buildup and in-depth
filtration by relatively lower headloss accumulation with time, cumulative
plots of headloss or number of backwashes vs. time and solids loading may indi-
cate when a change in particle collection mode occurs. These plots for the
Addison North and South plants are shown in Figures 30 and 31, respectively.
From Figure 30, a marked- point of inflection is apparent at the approxi-
mate midpoint of the test program for both the solids loading and time cumula-
tive plots for the Addison North plant. Inflection points also occur for the
Addison South plant in Figure 31 but are less pronounced. Backwash frequency
is higher for the Addison North plant because of the higher flow rate.
Mean performance data and bivariate correlation coefficients for segment-
ing the Addison plant data into two time periods are shown in Table 17. Dur-
ing the first time period, flow and influent suspended solids concentrations
were higher for both plants. No significant improvements in correlation of the
data were obtained by use of this technique.
Plots of cumulative headloss vs. solids loading and time are shown in
Figure 32 for the Marionbrook plant. If a filter backwashed during a run, the
number of backwashes was multiplied by 6 ft. (maximum headloss) and added to
the headloss at the end of the run to obtain total headloss. During the ini-
tial portion of the sampling program, headloss buildup was very low and in-
80
-------
*
»•
• t
o
•
o
II
^1
*
4
e
00
*
*
*
4
4
*
'
*
*
*
4s
4
4
*
*
*
«
0
*
$
* * *
£
*
•* .#
*
*
A.
^
«
£.
*
^i
C
4
o
0/0
(V 4
41
*
* 0
4
* 4
*
cvt
*
4s P
* 4s
*
* .
(V
* P
4
f"
*
0
>
1
*
4>
4>
ev
*
*
^
i*
"
4
* -4
'»* *
* 4
* *
(V
*
tf
*
*
*
• * 41
r
4s CV
^|
o
(M
41
*
"*
41
(M
«
4s
* «
* 9
I PJ IV 4
» 1
•c
**^
-------
, ' i
i
!
i
1 !
I
1
t
i
!
' !
,
I
i
'• \
!
!
I
1
1 '
1
I
;
i i
1
. 1
i
'
1 1
r-l
O
t *
: o
^ "
i
! K
!
1
j
1
1
i
i
i
i
i
i
i
i
I
1
! *
i
i
1
1
i
•
1
i
1:
!
;
-^•4 ••< ^^ ^^ ^ ^
o i o
C i CO
1 ^^
4 ! *
,_
1
\
•
>
1
i
1
I
i
j
i
!
I
1
i
•
1
'
1*
} !>"4 +"4 ^
1
«
*
*
*
*
*
«
4
1 + ^^ H
0
>0
•
c
i
i
i
, !
'
1 i
• i
!
i
i
i !
i
5 j
« i
i i
I
I
< I
i i
!
i
t '
: i
; 1
' j
i
i !
; i
!
!
i
i
j
;
[
i
j
1
-* i
!
' *
- i *
; * *
I
: 1
* *
. i
* 1 *
i
i * *
* ! (VI
i
* 1
i
: , *
-* i
' 1
i
i ^^ ^^A ~fA ^»d fca4 fcv
P W^ w*4 ^J ^^ ^^ ^™
- i
-c
•««•
•
j
o/a !
*<** *•* \
*
i
* . i
: i
i
|
i
i
i
i
i
!
i
]
j
j
i
-* ',
i
* !
!
*
*;
i *
i *
i
i
i
i
\
i
*,
! W
*.
*!
* *
*
*
*
! * *
* i *
* • * «v
'* *
*
IVt *
*
fU
«n*
tnivf * * *
* * evi cvi *
«•> ivi
F
| 01 P-(
^ ^^*
1 ' 5 '
• i i— i
1
1 00
I 1 c
* i -ij
• i £
i ; «
1 i 0
1-0 -1
ts*±.
•*'.' CO
* '» -O
» -*•; -H
• ! "2
t °
»
* !
1 ! m
1 1 >
10 0
* I e» o
^4 -^
* i«J »
* «
« 10
* esj
* „,
« I a)
H a
* Si
4A. ~* Pbl
^F 'w ' *^
i
0 !
j
i
i
i
i
i
82
-------
4
*
C
9
»
^
0
*
*
O
O
Jl
J-)
c
Ul
*
>•
>
*
O
O
O
•*
*
ft
-
if
4
*»
i
*
*
*
• ••
*
*
£
*
*
0
0
0
l"l
/Sra-0
'
i
i
;
^
•
^
i
1
*
j
i
*
*|
i
*
1
*
* * i «
* * «
* *
*
* *
1 *
*i
*
* * * *
f ^ 4
r
* (V
* * *
IT
* * ft
* *
(V. *
* *
0
c>
oil
i
i
)
i
i
i
j
i
!
,
I
i
j
<
i
j
1
*
*
*
(M *i
» * *j *
i *
* CM
* ft* | *
* m *
m « * *
f\3 IVI 1*) ^
* * (VI
en (vj
*
* « « >c «r r
* *j (VI* *
* ftt * (V
0
e
*
o
"*
c
O
' *
.
cj
eg
<
>fi
a
i
t*
:
a
0
«
*
1C
p^
e
a
i
M
j
* !
c
o
*
^
i
i *
i
i
1C
t
»^
• -
e
.
CO
•a
CO
.0
§
en
00
13
(0
O
01
13
vO
0)
l-i
83
-------
*
i
1
1
1
1
1
*
'
*
r
t
IT
m
i
t
i
i
*
n
«
*
e
0
•
o
CM
1
t
i
t
i
i
(
j
*
*
* *
<
**
ft*
* *
*
j
)
™
*
(M * *
* *
*.:
* * *
:*
*
*
* * *
*
(M
O
O
»
in
r-t
•
i
!
i
*
!
'i
i
*. I
0 i
« ;
M
]
* «i
••
)Qt
*
* . .
CV3
* *
* (V
* :*
4t
*•
* *
C
C
C
f
f
f
)
1
t
i
i
j
i
i
j
I
i
i
i
*
*
*
*
o
o
•P
Ift
1
i
i
1
!
i
i
i
i
1
i
i
j
i
1 C
• 0
f 1*1
1
^
^''
1
t
1 0
% tf)
*•• *
f /\H
1
1
*-
1
* c»
l-ct
• •
I CM
t
1
^
i ;
i
1
t
1 i
*•
1
t
10
141
t i
1 ;
1 i
1
t
IS
i"i
i
t i
t !
*
l
*•
|1
e
S *
o* .
CO
•O
P
CO
*
CM
0)
84
-------
,
-
:
i
i
j
i
! * -
i
:
i I
t '
• : Jd
1*0
1 {
1 '
•*• •
• ^i
1 c -or
S 1
' •
j
j 1
\
i ! *
i i
;
; J
i
1 »^
1 ;
1
i T '-
1 O
i i *• •
i : r rvl
* .
i
•* i
m
0 !
,
i
i
H : |
u
i
i
'* 1
1
1
|. «
* '
T '
1
t-
1 0
10
i * t
i t ft >«pJ
i • •*•!
; •) 1*3
;
!
;
: * i
i i *
: i!i
I
1
» '<
t
1 «S «>
! ' 1 ; 1 ' *"*? **i
)
'
!
J
1
j *
i
j
!
j
j
i
4r •• '• ^
o
o
o
1 "
to ^
+ • O
i 1 5
1 -r4
+ ; ""•
1 n
t , *
» O J?
« * >
1 (VI u
r
* *
t i •
r I
fWPI 11
-------
o
un
1 i
'
' '•!'*•'
* '
•
* *
» .
• . *
! , *
'
• ' * ' »
! . ! ' ! « * , * *
1 'I,** ;
1 * « *
*
' * * i . a * CN!
i « rj
r-l ' * *
^ 1 ' ^
' «: # *
0 , « *,
:'•!:'< * •
t ' * 1
i ' ' ' ' i i
' ' 1 '
: ' i i i ' ! i '
i ! ! : ! ! ' ^
: . ! ! '
*T i i '
'O O O 'O O O C
le o o 'e o 10,
o ; in e ! IT o IT.
«>1 **
86
-------
150
GO
q—
O
100
50
0
o
o
150
QQ 100
o
o
50
0
B
oo.
00
0 50 100 150 200
2CnCHb/sf/d
Figure 30.
0 20 40 60 80
Time (Days)
Cumulative backwashes by headless vs. solids loading
and time, Addison North - Phase 1.
87
-------
150
O
-2 50
0
.00
00°°°
oo
00
OD
150
100
50
0
B
,0°
0 20 40 60 80
2CnQ-lb/sf/d
0 20 40 60 80
Time (Days)
Figure 31. Cumulative^ backwashes by headless vs. solids
loading and time, Addison South - fliase 1.
88
-------
Table 17. Addison North and South Plants - Mean performance data and
bivariate correlation coefficients fqr segmenting data, Phase i
No. of cases
Q-gpm/sf
No. of BW
CQ - mg/j&
C - mg/j&
C/C
o
C/C to :
o
Q
Qt
t
C to:
C
o
Q
CoQ
t
No. of cases
Q-gpm/sf
No. of BW
CQ - mg/jfc
C - mg/4
C/C
o
C/C to ;
©
Q
Qt
t
c to;
C
o
Q
C0Q
t
ADDISON
All
Data
73
3.8
3.3
44
6.5
.173
-.07
.05
.04
.26
.06
.29
-.05
ADDISON
70
2.1
2.8
37
9.0
.262
0
.19
.18
.40
.02
.23
-.27
NORTH
May 20-
Julv 1
37
4.2
4.5
50
6.5
.121
.20
.16
.07
.21
.10
.30
-.07
SOUTH
34
2.3
4.1
42
9.4
.220
.14
.16
.12
.38
-.11
,12
-.34
July 2-
Aue. 7
36
3.3
2.1
39
6.6
.174
-.01
-.24
-.26
.32
.09
.33
-.05
36
1.9
1.7
32
8.7
.259
-.04
.16
.13
.42
.21
.39
-.11
89
-------
300
200
100
0
1/18
oO 0, 00
2/12 5/12
c
..•^
i i
8/1 o
0
o
o -
o
o
o
>
A
1
0
20 40 60
2CnQ-lb/sf/d
300
200
100
0
80
1
1/18
_ o
1
2/12 5/12
0°
o
o
o o
1
1 1
8/1 o
o
o
-
B
1 1
20 40 60 80
Time (Days)
Figure 32. Cumulative headless vs. solids loading and time,
Marionbrook, Medium flow - Hiase 1.
90
-------
depth filtration was considered to be the dominant particle collection mode.
On February 11, 1976, heavy rains along with melting snow occurred, headloss
buildup was very rapid, and straining was considered to dominate. Greatly
improved correlations were obtained for segmenting the dry weather data on
this basis (Figure 29). Filters backwashed frequently during a second testing
period in August, 1976. These backwash data are suspect because of malfunc-
tions in control systems.
Grouping by Ranges of Filter Influent Suspended Solids
Data for each plant were grouped into three CQ ranges; less than 35 mg/^,,
35-50 mgAf,, and greater than 50 mgAt,. No significant improvements in corre-
lation of the data were obtained by this approach.
Ratio of Solids Capture to Headloss Increase
The ratio of solids capture to headloss increase, measured as weight of
suspended solids captured per unit of total bed headloss buildup, excluding
the clean-bed headloss, is a measure of the amount of material that can be
stored in a filter bed during a run (Dahab and Young, 1977). This ratio
should also be indicative of the type of particle collection mode since a
large solids capture ratio should be characteristic of in-depth filtration.
Data for the Addison South plant were segmented for the solids capture
ratio (Scft) less than or equal to and greater than 0.1 lb/ft3. This value
was selected to segment the data into two approximately equal groups.
Regression coefficients for a model of the form C = CQ QC> where a, b, and
c are constants, are shown in Table 18 along with mean performance data.
Multiple correlation coefficients for the logarithmic transformed equation for
segmenting the data are nearly identical. The data, except for similar perfor-
mance at the medium flow rate, indicate higher removal efficiency for the high
Sch range. This suggests that removal efficiency may be higher for in-depth
as compared to straining filtration. This concept is explored further in the
next section.
HEADLOSS AND SOLIDS CAPTURE RATIOS
Headloss and solids capture data are shown in Table 19 for the Addison
South, Lake Zurich, and Marionbrook plants for runs where an intermediate back-
wash did not occur. The accuracy of the data is poor at high solids capture
ratios because headlosses are less than 0.1 ft. Mean solids capture varied
from 0.15 to over 2 lb/ft2/ft.
Although correlation coefficients are generally low, the Addison South
and Lake Zurich data show a decrease in C/CO with increasing solids capture
ratios. This is also shown in Table 18 for Addison South. No explanation can
be given why there is no apparent trend for the Marionbrook data. The highest
bivariate correlation coefficients occur for C and C/CO to headloss for low
flow for Addison South and Lake Zurich. At lower flows, there is less pene-
tration of particles into the bed as compared to high flows. The decrease in
clarification efficiency may possibly be caused by deeper floe penetration at
high headloss. Plots of C vs. headloss and solids capture ratio are shown in
91
-------
Table 18. Addison South Plant - separation of data sets for low and high
solids capture ratios - Phase 1.
Regression Coefficients for C = aC Q
Phase 1
All Data
Sch = 0.1
Sch > 0.1
All Phase 1 No.of
Q-gpm/sf
C-mg/jl
C0-mg/;&
C/Co
Sch s 0.1 No. of
Q-gpm/sf
GO -mg / A
Co~mg/ &
C/Co
Sch > 0.1 No.of
Q -gpm/sf
C-mg/jfc
Co-mg/jfc
C/GO
No. of
Cases t a
176 .58 1.13
73 .63 .88
103 .60 1.29
Mean Performance Data
All Low Q
cases 176 61
2.06 0.87
15.3 11.2
50.9 52.3
.314 .235
cases 73 28
2.08 0.87
17.3 14.3
50.5 50.8
.362 .310
cases 103 33
2.05 .86
13.9 8.6
51.3 53.5
.283 .176
b
.57
.69
.49
Med Q
51
1.76
13.7
48.9
.300
17
1.67
13.4
49.3
.299
34
1.81
13.9
48.7
.301
c
.41
.29
.52
High Q
64
3.43
20.5
51.3
.401
28
3.53
22.7
50.8
.451
36
3.36
18.7
51.7
.365
92
-------
(U
CO
CO
CO
JJ
CO
13
V
I*
3
4J
a
CO
o
CO
T)
o
CO
•a
c
CO
CO
o
r-l
T3
CO
0)
..
O 1C
U 0
0M
o
K C
8 c
o
O CO
lj C
«
IH CU
^ 00
D C
u a
IW ^
r-l
i a
•gs
U]
o
00
II C
U CO
IH OS
I-l
1 C
U CO
en 53
to
c
CO
• pri
£
c
CO
£
IH
O a
4)
• CO
Z U
C
a
PL,
SO
•*
i
CM
in
in
1
r-i
so
rH
ON
I
CM
q
q
i-i
SO
rH
1
'-O
q
CO
CO
CM
CO
rH
•
O
SO
lH
r*.
en
u
3
o
to
C
o s
en o
1
CM CM
CM CO
1 1
CM q
1
m co
rH rH
1
SO SO
q •*
i
CM
ON
• rH
1 1
CM SO
q q
ON CM
i-H CM
" •
CM CM
CM in
1 1
SO CO
q
«3- 00
m
• CM
UO rH
1 1
co !••
q
so so
CM
q
CO
CO
ON
CM
1
CM
CO
O
r
-------
Figure 33 for the Lake Zurich plant. The Lake Zurich filters were not always
backwashed at the start of runs in order to obtain data on the effect of run
length on filter performance. Figure 33 shows that C generally increased with
higher headless and decreased with higher solids capture ratios, although the
high solids capture ratios (0.1 ft. headloss) tend to distort the plots. A
similar result is shown in Figure 34 for plots of C/CQ vs. headloss and solids
capture ratio.
DEVELOPMENT AND TESTING OF OPERATIONAL MATHEMATICAL
MODELS--PHASE 1 DATA
Data from this phase were analyzed using the three models shown in
Equations 18, 1% and 20, using multiple regressions and log transformations.
Selected models were then compared with data trom other plants.
Regression Coefficients for Models
A convenient method for analyzing the effect of a number of independent
variables on a dependent variable is to conduct a series of multiple regres-
sion analyses and add a new variable each time. The regression coefficients
in the assumed power law relations can then be compared with results expected
from theory.
Regression coefficients for the three models using all Ihase 1 data are
given in Table 20. For the Model I, the regression coefficient for viscosity
has the wrong sign since, from theory, it would be expected that C would in-
crease at higher viscosity. The significance level, or probability that the
result occurred by chance, for equivalent media diameter is high. The coef-
ficient for de, however, has the correct sign. From Figure 21. the values
for de do not vary widely except for the large diameter media at the Addison
plants. This result may be caused in part by the high removal efficiency for
the Addison North plant. Correlation coefficients are not improved signifi-
cantly by the use of variables other than Co and Q.
For Models II and III, the coefficient for viscosity also has the wrong
sign. For the Model II inclusion of L and de does not improve correlations.
The regression coefficients for Q of about -0.25 are in the lower range of
those reported in the literature. For Model III, a very poor correlation is
obtained when only Co and Q are included. This suggests that the filter clari-
fication efficiency is relatively independent of depth for these filters, as
shown from analysis of the average plant data.
Regression coefficients for the four variable flow plants, excluding the
wet weather flow data for Marionbrook, are shown in Table 21. The model for
C yielded a positive exponent for L and a negative exponent for de, neither of
which would be expected from theory. However, only four different media speci-
fications, with ranges of de of 0.8 to 1.7 mm. and L of 12 to 45 in., were
included in the analysis.
It is clear that the inclusion of independent variables other than Co
and Q in models cannot be justified without inclusion of other variables dis-
cussed in Chapter X. The model of choice without such variables then becomes
94
-------
13
- 10
V*
cn
E
i
o _
5
O
o1
o •
o
-
o
.
o
o
1
• ' 0 ' • '
-
Lake Zurich
Low Flow Rate
O BW at start of run
• Not BW
A
I 1 1
2 3
h-ft
ID
- 10
O>
E
° 5
n
^ i i i i
-*^ Lake Zurich
* o Low Flow Rate
O O BW at start of run
0 • Not BW
h=O 1 h = 0.l
on u-' o
•
I I I i
h = 0.l
0 -
B
0 0.5 1.0 1.5 2.0 2.5
Sctrdb/ftVft
Figure 33. C vs. headless and solids capture/headloss, Lake Zurich -
Phase 1.
95
-------
0.6
0.4
o
O
\
O
0.2
00
o
\o
o
0.6
0.4
o
o
o
0.2
00
Lake Zurich
Low Flow Rate
O BW at start of run
• Not BW
2 3
h-ft
&•'•'
h«0.l
°
h=0.l
0
B
h-o.
o
0 0.5 1.0 1.5 2.0 2.5
Sch-(lb/ft2)/ft
Figure 34. C/C vs. headless and solids capture/headloss. Lake Zurich
Phase 1.
96
-------
Table 20. Regression coefficients - all Phase 1 data.
Model
I
II
III
Variables
Included
C = a C VjA,ed f
o x e
Co,Q
C >Q>M<
O
CO,Q,^,L
co,Q,n,L,de
(Significance)
C be d_e f
Log - = aCo Q A de
o
Co,Q
Co'Q'^
CO,Q,M,,L
C°,Q,^,L,de
(Significance
Q,u-,L,de
(Significance)
1 _ C _ brtc dj f
E Log ~ = aco Q M- de
o
Co,Q
C0,Q,H
C ,Q,M-,d
o e
(Significance)
Q,Me
(Significance)
r
t
.42
.44
Regression Coef
b
c I d
.56 JF .36
.53 ! .34
1 i
.45
.46
.43
.45
.46
.47
.34
.25
.45
,61
.57
.56
.54
(0)
.34
.36
.34
.35
(0)
-
(0)
.11
.19
.31
(0)
-
•
.34
.36
(0)
-.26
-.25
-.25
-.26
(0)
-.23
(0)
-.27
-.22
-.27
(0)
-.24
-
-.98
-1.32
-1.07
(0)
-
.75
1.03
.85
(0)
.74
(0)
-
2.87
1,37
(0)
1.26
icient
e
-
_
-.16
-.30
(0)
-
-
.14
.24
(0)
-27
(.01)
-
-
_
-
f
-
_
-
.36
(0)
-
-
-
-.26
(P)
-.17
(.07)
-
-
-1.01
(0)
-.90
97
-------
Table 21. Regression coefficients, variable flow plants - Phase 1.
Model
IV
V
VI
Variables
Included
C = aCobQCAedef
Co'Q
C ,Q,M-
C°,Q,n,L
C°,Q,M,,L,d
(Significance)
Log 7;- - aC bQCM-dLed f
\j O G
o
Co'Q
CQ ,Q , p.
CO,Q,M-,L
o Q
(Significance)
Q,M-,L,d
(Significance)
o
Co'Q
Co,Q,ti
C0,Q,M-,d
(Significance)
Q'MQ
(Significance)
r
t
.54
.55
.55
,56
.53
.53
.53
.55
.34
.17
.25
.72
.62
Regression Coefficients
b
.46
.47
.46
.43
(0)
.45
.46
.46
,49
(0)
""
.22
.27
.50
(0)
-
c
.33
.33
.35
.36
(0)
-.28
-.28
-.29
-.31
(0)
-,27
(0)
-.019
-.028
-.29
(0)
-.26
(0)
d
.32
.44
,67
(.0.6)
.061
.021
-.20
(.51)
-.53
(.12)
1.51
-.33
(.31)
-.67
(.06)
e
-
.084
-.29
(.03)
-
.028
-.34
(0)
.27
(.04)
-
-
-
f
-
-
.62
(0)
-
-
-.61
(0)
-.33
(.09)
-
-1.48
(0)
-1.29
(0)
98
-------
C = a (£ Qc (21)
where a, b, and c are constants.
Models for Individual Plant Data
Regression coefficients were generated from data sets for the model
C = a GO Qc and are shown in Table 22 with the multiple correlation coeffi-
cients (rt) for the assumed relationship. For plants where flow was not varied
to individual filters, the coefficient c on superficial flow had a negative
value for two plants; negative values are not expected since filter performance
should deteriorate at higher flow rates. For variable flow plants, the coef-
ficient c ranged from 0.32 to 0.5 for most data and multiple correlation coef-
ficients varied from 0.26 to 0.72. The coefficient for Co varied from about
0.2 to 0.9.
Data for the Marionbrook plant were obtained during the period from Janu-
ary to August, 1976. Data were analyzed separately for both dry and wet wea-
ther periods. Prior to melting snow and heavy rains in mid-February, head-
loss buildup was less than 2 ft/day and in-depth filtration was believed to
be the dominant particle collection mode. For this period, the bivariate
correlation coefficient between C and COQ was 0.71. From Table 22, the
highest multiple correlation coefficient of 0.77 was obtained for an equation
of the form:
C = 0.40 C0-86Q-34 (22)
During the wet weather flow periods at the Marionbrook plant, flow generally
was higher by a factor of two, headloss buildup was very rapid, and straining
or semi-cake filtration was considered to dominate.
Scattergrams of calculated vs. measured C are shown in Figures 35,
36, 37A and 37B for the Addison South, Lisle, Lake Zurich, and Marionbrook
(dry weather) models, respectively. If a model provides an accurate descrip-
tion of the filtration process, although some scatter of the data would be
present, it would be expected that a line of best fit would have an intercept
close to zero and a slope close to 1. The intercepts and slopes for the
scattergrams are given in Table 23. In each case, a positive intercept at
zero measured C and a slope considerably less than 1 are present. Thus the
models tend to overpredict C at lower values of measured C and underpredict at
higher values. The Marionbrook model provides the best fit with an r of 0.84,
an intercept of 4.4 mg/£, and a slope of 0.57.
Testing of Models
Each of the models developed from variable-flow plant data was tested
using other data sets, and scattergrams of calculated vs. measured C were
prepared. Correlation coefficients, intercepts, and slope are given in
Table 23. Typical scattergrams are shown in the following figures.
99
-------
b c
Table 22. Regression coefficients for C = aC Q - Phase 1 data,
Plant
Non-Variable Flow
Addison North
Addison South
Barrington
Des Plaines River
Romeoville
Variable Flow
Addison South
Lake Zurich
Lisle
Marionbrook
Dry weather
1st half wet weather
2nd half wet weather
Mult
.26
.34
.67
.39
.57
.58
.65
.56
.77
.44
.37
a
1.49
2.34
.70
2.37
.65
1.13
1.73
1.46
.40
1.52
.62
b
.28
.38
.87
.24
.64
.57
.49
.38
.86
.35
.44
c
.25
-.11
-.29
.50
.55
.41
.41
.50
.34
.32
.16
100
-------
rH
II
M
5*
Q
~_j
f™l
CO
i
X
*
*"**** *"
•f
j
•>
1
1
1
*
j
]
; *
i
V
\.
N.
>^
4
*
O
CO
•
r-
CV»
«
*
*
^F
£
*
*
'
g
&
CVJ
4
* *
x
X «
* X
X
™
*
in
CO
•
o
(VJ
!
;
i
(
j ;
: '
t
'
\ \
i I
,
1 i
! ,
i
i
1 i
*! 1
1 i !
* '• i i
4 :
!
' ,
< , |
i
4» * 1
'
* *i
* . :
:
* i
i i
» 41 * {
* CM ! CV
* 1
. ! * i
4l (Vt
V • *| * 41
4t 4f|
* * X
(V 1 (V * # * i
# «t *X>J* *;
J tM4t\*
* « (M\l
9 « « »k cvtVI
* « (\j (vi* V*
* v
-4| r|> w . »
J * \
N
!
:
^
o in
cw o>
fi >o
^^ i
5 pQ;|Bin
r t
1 » .
t o
t O
9 ^D
t 10
1 •
f ,
* 1
r !
i
t o
t o
* ob
i -*
i >
i
» i
1 0
1 0
*- »
no
i <»
* ,
i i
* i
1 !
1
I a
* *i
1 (Vt
,
i i
*
t :
t 1
t to
•» •
i fv)
i
i '
+ i
1 :
t '
1 C>
10
• »
t «-i
t ,
0.
«. {
• i
i !
1 O
» c?
Tact
p i
i :
t !
*
1
I J>
\ M
; t '
i
i
o • \
1
1
1
1
j
a
CD
•U
CD
T)
O
CO
V
•o
o
CO
T)
-------
I !«
CO
4J
co
a
t-i
to
"A
(U
*< «
T)
(U
g
t)
(V
4J
VO
CO
102
-------
LZ MODEL, LZ DATA
31 _L?O * —
J FIG. A y/
«* I
•^. i Slope = 1 -i, A
£P *" X^
*"'*•' y;
"S I */^ 2 * *
w T 2 »P7 «« «
3 10.^3 * *« 33§335*2 *»
to T «*j/*«e A
o r yr 5 ^
i X*
r /
i x
o *X
ft ^2.4TO 24 Oo ^3*n
MEASURED C-mg/j6
_1 ' ' ' MB MDTIET. Mft flATA
_ft _
5 FIG. B /
^! I Slope = I*/ «
"SI X *
S i X*
^ J . «* «
.S I X^ **
^12,17 * »3< #***«
I 2*K^2* *J?
I .?^*22* *
T X
0 ;'
I
,'
i
Jio. _J
_ftfl. -
MEASURED C-mg/j&
Figure 37. Calculated vs. measured C, Lake Zurich and Marionbrook.
103
-------
(U
CO
(0
£
o-
O
14-1
ca
g
to
3
cr
(U
CO
CO
-------
Figure Model Data Set
38 Addison South Marionbrook (dry weather)
39 Lake Zurich Addison South (var. flow)
40 Lisle Lake Zurich
41 Marionbrook Lisle
In every case, positive intercepts and slopes less than one were obtained.
Plots of calculated vs. measured C using the Marionbrook Model (dry wea-
ther flow) are shown in Figure 42 for Hsiung's data (1972) and in Figure 43
for other published data. A poor fit of the model is apparent for Hsiung's
data. In Figure 43 , the data reported by EPA (1975) provide the best fit.
Most of the EPA (1975) data are from full scale sand filters in Great Britain.
DISCUSSION
Performance of tertiary filters is the result of complex interactions be-
tween a number of variables. Changes in the characteristics of secondary efflu-
ent suspended solids, believed to be a major variable, can occur on a seasonal
or daily basis, or over a shorter time frame during filter runs. Causes include
changes in secondary process biota, changes in process operating modes, process
upsets, and flow surges causing temporary loss of floe from secondary sedimen-
tation tanks. Parameters of importance that were not measured routinely in
this phase include particle size characteristics, floe strength, and particle
surface charge properties. Even so, it was found that if filtration process
data are analyzed over shorter time frames where suspended solids character
is more constant, improved correlations can be obtained with simple correlations.
In pilot- scale studies on filtration of activated sludge process efflu-
ents, Tchobanoglous (1970) reported that media grain size had a significant
effect on removal efficiency for single medium sand and anthracite filters at
depths from 18 to 30 in. Influent suspended solids levels for his studies
varied from approximately 14 to 24 mg/j, . Small removals of suspended solids
occurred below depths of 16 to 20 in. Baumann and Huang (1974) reported that
variations in performance over a wide range of media sizes were not significant
for dual media pilot-scale filtration of trickling filter process effluents.
Trickling filter effluent suspended solids varied from 15 to 50 rag//,. It was
reported that no significant removal occurred below depths of 12 in. for both
the sand and anthracite media. In pilot-scale studies using unstratified bed
filters, Dahab and Young (1977) concluded that removal efficiency is not re-
duced greatly by increasing the effective size (dlo) of media from 1 to 2 mm
at flow rates of 2 to 4 gpm/sf. Influent suspended solids levels for their
studies ranged from approximately 20 to 50
In the studies reported by Tchobanoglous (1970), influent solids levels
were much lower than the averages of 28 to 62 mg/£ observed for the plants in
this full-scale investigation. At lower solids levels, it would be expected
that mean particle size would be smaller and media grain size would have a
greater influence on removal efficiency than in this study.
Media grain size and depth did not have an apparent significant effect on
filter clarification efficiency for the full-scale plants in this study. Al-
105
-------
'
!
i
j
4
0
r
c
!
i
i
^
• >•
B
O
•
«•
U
:
^
(-4
II
S.
O
1-1
XI
>
"
'
!
I
(U
4
N
in
00
•
e
CXI
r/sm-
!
;
i
*
t
i
j
i
i
i
!
^
% 4
4
4
\y *
^
£
4
-
- C
01
I
f*
v
~
,
CM
*
*
4
CM 4
k *
• V1*
CVI MV
4 X
* *
c
1 ? 0
* o w
* -1^
1 Q ^§
*? •? "
J *^ *
-------
C
f
9
•4
1^
u
to
-t
t
^J
^^
II
li
a*
o
^v
>
«
*
*
*
4
v
\
\
<
<*•
(XI
£
.
*
*
*
*
t
tv
^
c
\
V
\
»^k
»
4
*
4
4
a
<
c
D pa**
£
£
/.,
V*
(X) »
IVft A .rf
i C%§ "^ *w
N*
\^
* ^S
*
**«
0
0
»•«& ti
* >o o
1 CO
>
*• i S
t ' °
! ' CO
» : -rl
t O TJ
• o ^
1 CD
1 < r-l
§ 0)
| 1 TJ
• o
* g
» ! J=
1 O "
r?^ g
r* •** JO
JS
' CO
* o ^
i j Ij "
1 ci 5 °
IE '*4t ^ M
1 - AJ ^
»l ^
en
t 01
* e
t
* i m
1 0 >
1 C»
» • T)
1 M 0}
»^^ 4J
^^ CD
i ;3
» s •
•*> i U co
| • i-l 4J
* co co
v lo
Ml ^
* * en
: u>
0 H
1-1
107
-------
J J
108
-------
73»02
48.68
"•»•.
&Q
i
4J
J
O
o
I
v T . ...
I
I *
T * Slope = \S
J • x
; * /
1 v * ' /
I * /
I « . . X
* * / *
I /
I * v^
1 •'•* * / *
* #/ »
I « « «« * y
T «* «* .3 * «
12.17
I « 2 2 «,x»» » »
•T *«*t^3« «?
I y?
T /
0
*,
•
1
___ 9
14.00
---- 5*^00 — I
MEASURED C-
Figure 41. Calculated vs. measured C, Marionbrook Model,
Lisle data.
109
-------
70
60
50
E
6 4°
-a
0)
30
_
a 20
O
10
(111
O
O
i r
Slope =
I I I I
0 10 20 30 40 50 60 70
Measured C-mg/l
Figure 42. Calculated vs. measured C, Marionbrook Model, Hsiung's data.
110
-------
25
— 20
O
-a
-------
though some improvement in efficiency would be expected for larger depths, the
improvement is apparently too small as compared to the influence of other vari-
ables. The differences in characteristics of suspensions are most likely the
major reason for the variations in performance between plants. However, based
on reported pilot studies at equivalent and lower influent suspended solids
levels (Baumann and Huang, 1974; Dahab and Young, 1977), it is probable that
media size and depth do not have a large effect on mass removal efficiency over
the ranges encountered in this study. However, media size in particular has
an important effect on clogging rate.
112
-------
CHAPTER IX
RELATIONSHIPS BETWEEN MEASURES OF CHARACTERISTICS
OF SECONDARY EFFLUENT SUSPENDED SOLIDS AND THEIR EFFECTS
ON FILTER CLARIFICATION EFFICIENCY
INTRODUCTION
This chapter describes the relationships between measured suspension
properties and their individual effects on filter clarification efficiency
using data from Phases 2, 3, and 4. Methods used to characterize secondary
effluent suspensions were suspended solids concentration, refiltration para-
meters, and particle size measurements. Attempts to obtain meaningful data
on electrophoretic mobility and the optimum cationic polymer dosage necessary
to neutralize particle surface charge were not successful.
RELATIONSHIP BETWEEN MEASURED SUSPENSION CHARACTERISTICS
A summary of mean and ranges for Co, number mean particle size, and re-
filtration parameters is given in Table 24. Filter influent suspended solids
concentrations were, in general, considerably lower during Phase 4 as compared
to other phases. A total of 38 samples of filter influent was obtained during
Phase 4. Including filters operating at variable flow, 66 data on filter clari-
fication efficiency were obtained. Nine filter influent samples each were ob-
tained at the Addison South plant during Phases 2 and 3, and 19 samples were
obtained for the Des Plaines River plant during Phase 3.
Particle Size Measurements
Typical particle size histograms, expressed as number of particles per
cm3 for each size class, are shown in the following figures:
Figure Phase Plants
44 3, 4 Addison South
45 3, 4 Des Plaines River
46 4 Addison North, Harrington
47 4 Lake Zurich, Lisle
48 4 Marionbrook, Romeoville
Very few particles were larger than 100 jj,m. Larger floe particles were ob-
served at the secondary sedimentation tank overflow weirs but were broken up
by turbulence in flow to the filters and by the sample preparation technique
used. Secondary effluent flows to the filters by gravity at the Harrington,
Des Plaines River, and Romeoville plants, and is pumped for other plants.
113
-------
CD
•o
CD
M
4J
a
to
os|
a
0
»
X
till
1
i
a
1
£|
.
i
JiN(n
C
CO
JCN
r^c»>!^oooO.» >00
r~cn-»c'>»
«-l CM p-i »^ ^^ r»s0<^esf
01
m
tl
B
a
a.
00
o
£
O
z
i §
J- Q a
«
ea
n
JS
A.
•i4
"^ "O
^ 4!
x:
1
e
o
a
•*<
•o
^
B
O
1
u
V
CO
n
«
oi
I 1
a a
•- N
PM
a Jt
5 J
|
c
« Q
^. MI
a u
j jg
V
•^
•H
§
i
at
114
-------
IID.&J
t
6
10
£ 4
\
CO
"o
r 2
D
CL
O
n
LJ
« /•
-
L J
^ ^
L
j
\ I 1
A
Addison South
Phase 3
C0 = 60mg/l
i
10 Fields
-
i 1 i
10
30
100
300
I04 Particles/cm3
D — ro oi
-
i
i i
Addison South B _
Phase 4
C0 = 24mg/l
15 Fields
i 1 i
30
100
300
Figure 44. Particle size histogram, Addison South -Phases 3 and 4
115
-------
"> 1.5
o
_o>
o
£0.5
0
10
E
•SJ
t/>
_0>
o
o
Q_
I.!
>.5
0
figure 45.
10
10
Des Plaines River
C0 = 48mg/l
5 Fields
30
r/u.m
100
Des Plaines River
Phase 4
C0 = 26mg/l
28 Fields
30
r/xm
100
Particle size histogram, Des Plaines River
Phases 3 and 4.
300
B
300
116
-------
0.5
fO
£ 0.4
cn
£0.2
O.
0
1.0
10
§ 0.8
CO
"5 0.6
S. 0.4
0.2
0
10
10
Addison North
Phase 4
C0= 16 mg/l
30 Fields
30
100
Barrington
Phase 4
C0 s 22mg/l
30 Fields
30
dp-/am
100
300
B
300
Figure 46. Particle size histogram, Addison North and Harrington -
Phase 4.
117
-------
I.U
"p 0.8
o
8 0.6
o
fe 0.4
Q_
2 0.2
0
i i i
Lake Zurich
Phase 4
C0 - I5mg/l
30 Fields
•»
i
-
i 1 i
3 10 30 100 300
dn-LLm
2.0
ro
E
••M
,0 , -
^ 5
(/>
o
fe 1.0
CL
2 0.5
0
K r
1 I I
Lisle
Phase 4
C0=l4mg/l i
-
-
i
15 Fields
—
"
B
F 1
3 10 30 100 300
Figure 47. Particle size histogram, Lake Zurich and Lisle - Phase 4.
118
-------
10
I
en
o
CL 2
0.4
10
E 0.3
^e
to
_o>
•S 0.2
O O.I
10
10
Marionbrook
Phase 4
C0= 39 mg/l
10 Fields
30
100
Romeoville
Phase 4
C0=l2mg/l
40 Fields
30
300
100 300
Figure 48. Particle size histogram, Marionbrook and Romeoville,
Phase 4.
119
-------
The Addison South data in Figure 44 show a unimodel distribution for this
size range. A large number of particles smaller than 3 pm were usually present.
This can be attributed in part to the trickling filter process effluent which
is less flocculent than activated sludge process effluent. Average number
mean particle size was 8 pm for Phase 3 and 12 pm for Phase 4.
Most of the Des Plaines River samples during Phase 3 exhibited a bimodal
distribution, shown in Figure 45, with peaks in the range from 3 to 5 pm and
at about 30 pm. Tchobanoglous and Eliassen (1970) found a similar distri-
bution for secondary effluent with the second peak at about 80 pm. Average
^nm was 18 pjn for Phase 3 and 16 pjn for Phase 4.
Samples for the Addison North and Barrington plants generally showed weak
multimodal distributions (Figure 46). Most samples for Lake Zurich, Lisle, and
Marionbrook showed unimodal distributions (Figures 47 and 48). Three of the
five samples for Romeoville exhibited a random or rectangular distribution,
shown in Figure 48. For these samples, 40 fields were counted and fewer than
10 particles were found in each size class.
A bivariate correlation matrix for particle size statistics, refiltration
parameters, and filter performance is given in Table A-3 in Appendix A. Parti-
cle size statistics are shown below.
d Number mean particle size
dam Area mean particle size
d^ Mass mean particle size
d4« 4th moment of number mean particle size
NT Mean number of particles per field
a2 Variance of number mean particle size
nm
a Standard deviation of number mean particle size
nm
a /x Coefficient of variation of number mean particle size
nm
%S1 Percent of total number of particles in size classes
1 and 2 (4.2 and 6 pm)
%S2 Percent of total number of particles in size classes
3 and 4 (8.5 and 12 p»)
%S3 Percent of total number of particles in size classes
5 and 6 (17 and 24 pm)
7e,S4 Percent of total number of particles in size classes
7 to 12 (30 to 192 pm)
Bivariate Relationships Between Suspension Properties
As would be expected, the mean number of particles per field had a rela-
tively strong correlation with Co. Very poor correlation coefficients were
obtained for other parameters. Plots of R and E vs. crnm» which generally
showed a better correlation than dnm, are shown in Figures 49 for ftiase 3, and
in Figures 50 and 51 for Phase 4, respectively.
120
-------
12
8
en
E
i
o:
0
0
Phase 3
oAS
• DP
A
10
15 20
1.8
1.7
LU
1.6
1.5
0
Phase 3
oAS
• DP
B
10
15 20
Figure 49. R and E vs.
Phase 3
Addison South ,and Des Plaines River,
121
-------
1 I 1>I I '
«^ NJ — . 00
•a ,QQ _J _} 2E
§ • • 4 ^
J3 ZCO GLO
O D < > -
s >
0 .»
O A
o 3*
D < ^* ^ -
D * ^
OD ^*
*
i r i i i
*
to
to
0
10
ID
CM £
f
0 E
W tf
to
g
to
O
VJ O CD (D ^ CM O
^
ta
£
09
>
erf
o
a*
»4>
3
a
122
-------
1 p. 1
o o
D
D
DO
< N — CD
CD _J _I ^
• • « *
ZCO CLO
< < o or
O D
in
to
O
to
in
C\J
oi
ID
in
en
CO
b
en
m
CD
CD
CD
m
123
-------
For Phase 3 data, both R and E appear to increase for larger values of
°nnr Ihi-8 would be expected since large values for anm correspond to higher
clnm> and theory suggests that both R and E should increase with particle size.
No trend is apparent for the Des Plaines data for Phase 3.
Wide scatter is apparent for the plot of R vs. Olim for all Phase 4 data
in Figure 50, although a trend of increasing R with CTnm is present (r = 0.40).
Individual plant data for Addison North and Romeoville, and except for one out-
lying point for Harrington, show a stronger correlation.
From Figure 51, no trend is apparent for the plot of E vs. o_ for all
Phase 4 data. However, the Addison South plant data show a strong correlation
over the narrow range of E from 1.7 to 1.8. The Addison North data also show
a definite trend of increase in E with increase of anm- As shown in Chapter IV
E is determined with a precision better than ± 2 percent. Considering the
error bounds, the trends present for the Addison North and South are not as
strong.
A plot of R vs. CQ is shown in Figure 52 for Hsiung's data and Phases 2
and 3. A very strong correlation is apparent for the Addison South plant for
Phase 2 where CQ exceeded 200 mgAf, for three of the nine runs, and exceeded
100 mg/£ for six runs. These high SS levels resulted from failure of the acti-
vated sludge process secondary sedimentation tanks where the sludge blanket
reached the level of the effluent weirs. Although no particle size measure-
ments were made, large floe particles were observed in the secondary effluent.
A similar situation with high sludge blanket levels occurred at the Des Plaines
River plant during Phase 3, although the sedimentation tanks did not fail com-
pletely. For Des Plaines River, R is considerably higher than for the other
data but shows no trend with Co. No trend is apparent for the other data.
Except for the Addison South and Harrington plant data, no trend is appar-
ent for the plot of R vs. Co for Phase 4 data, shown in Figure 53 . Plots of
dnm and anm vs. C are shown in Figure 54 where a relatively strong correlation
is apparent for the few data. Therefore, Figure53 actually shows the corre-
lation between R and CT for Addison South. It can be concluded then that R is
not related to C0 except when higher values of C0 indicate carryover of large
floe particles from secondary sedimentation tanks.
Plots of E vs. Co are shown in Figure 55 for Hsiung's data and Phases 2
and 3 are in Figure 56 for Phase 4. Phase 2 data for Addison South show a
strong correlation of E to C0 which can be attributed to carryover of large
floe from secondary sedimentation tanks, as discussed for R vs. CQ. Hsiung's
data also show an increase of E with larger values of C0.
In Figure 56, E shows a very strong correlation with Co for the Addison
South data. This may also be attributed to the relationship of Co to d^ and
an_ in Figure 54. The Des Plaines River and Harrington data also show an
increase of E with Co.
If the error bounds for E are considered, the correlations present in
Figures 55 and 56 would be weakened. The consistent trends for the plots of
dnm and crnm vs. CQ and E vs. C for the Addison South Phase 4 data suggest
124
-------
o
to
ejforo
^ M ~
|£ff
'55 COCOQ.
O
00
C\J
O
CJ
O
s
en
1
N
g e
O
C\J
«
O
O
CD
«
3
00
§
«
CD
00
ol
-------
0
1 1 1 1 1
D D
a
^
w
0°
>- -
<• <
<0_ .
0 .0 ^
O • <4
0 1«^
• ^ •
»•
• >> •
-
i i i i i
in
m
t-b.
>j
0
^~
m
ro
O
to
m
CM
0
CM
in
o
m
0
0>
£
i
o
O
0)
TO
as
o
in
>
OS
-------
30
E 20
:!.
J 10
0
30
20
10
Fig. A
o
Fig.B
o o
Co-mg/l
o -
0 10 20 30 40 50
C0-mg/l
o
o
0 10 20 30 40 50
Figure 54. d and cr vs. C , Addison South - Phase 4.
^ Titn ttflrt O
ntn
127
-------
0
1 ] 1 '1
9 CJ ro to
o> as a>
t/5 eft to
o o o
CQ. Q. Q.
3 1 1 1
'w CO CO £L
X< < Q
O« < ^
•
9
.
•
O
o
9
0 ^_ OJ
ro
T3
C
CO
W
IT)
m
0)
3
So
•H
128
-------
1
CD
D
O
O
D
0*
c
0>
< N — CD
• m< >
« o a:
o D < >
O
in
in
in
ro
O
ro —
in ^~
C\J I
o
O
\^^ w
0 m* ^ ~
o • « *
9 B
_ .. ^^
1^^ ^^ ^p
t>
**^
P> •
-
i i i i i
in
•*
0)
(O
CO
2 fi
o
0
CO
^ «
vO
CD
O £
CD
m
ro
129
-------
that E may, in fact, be determined to a precision considerably better than ± 2
percent. The method used to estimate the error bounds in Chapter IV is
considered to be conservative.
KIVARIATE RELATIONSHIPS OF FILTER PERFORMANCE TO SOLIDS
LOADING AND SUSPENSION CHARACTERISTICS
A bivariate correlation matrix of filter performance vs. refiltration para-
meters, flow rate, influent suspended solids, solids loading, and number mean
particle size is given in Table 25. If straining was believed to be the dom-
inant particle collection mode, good correlations of log C/CO and C/C with R
and E would be obtained. The parameter G exhibited poor correlations. Number
mean particle size correlated well with some individual plant data but did not
show good correlations between plants. Effluent suspended solids (C) showed a
strong dependence on solids loading for data sets where straining was not con-
sidered to be the dominant particle collection mode.
Solids Loading
Plots of C vs. solids loading for Phases 2 and 3 for Addison South are
shown in Figure 57. Wider scatter is apparent for Phase 2 compared to the
strong correlation for Phase 3. Mean d^ for Phase 3 was 8 |jpi with a range of
7 to 10 |jm, the smallest particle diameters measured in Phases 2 and 3. Head-
loss buildup was very rapid during Phase 2. For the three-hour runs in Phase
3, headloss buildup was very low and decreased during several runs because of
decreases in plant flow. Thus the apparent dominant particle collection mode
was straining for Phase 2 and in-depth filtration for Phase 3.
During Phase 3 at the Des Plaines River plant, influent suspended solids
levels were higher than normal due to solids carryover from final clarifiers.
Except for four of the 19 one-hour runs, backwash mechanisms were operating
continuously making it difficult to define an appropriate run length. Back-
wash is initiated for these sand filters when the headloss exceeds 3 in. On
several occasions the maximum headloss limit of 15 in. was exceeded and the
filters failed to handle the process flow.
Plots of C vs. headloss and solids loading are shown in Figures 58A and
58B, respectively. Figure 58A shows a strong correlation of C with headloss.
In Figure 58B, data are equally segmented into "low" and "high" headloss levels.
Although it would be expected that surface straining would be the dominant
particle collection mode for sand filters, the higher C values for "high" head-
loss in Figure 58B suggest that solids breakthrough was also occurring. The in-
crease in C for high headloss is probably caused by sloughing of particles at
or near the top of the filter media and deeper penetration of floe particles.
Another factor that may be significant is the high localized filtration rate in
sections of the filter media that had just been backwashed by the traveling
mechanism.
Plots of C vs. solids loading for Phase 4 are shown in the following
figures:
130
-------
60
§
•i-l
CO
5S
4J
•H to
PQ -o
01
r-l
to
H
0)
w
(fl
J2
04
m
OJ
CO
£
M
0)
18
.C
dl
«
M
C
3
•H
CO
X
q-
Q
S
j
5!
g
<
£
-------
60
r- 40
E
O 20
0
-CO CO
0° O
Qo
Addison South
Phase 2
0
4 8 12
C0Q-lb/sf/d
60
40
20
0
Addlson South
Phase 3
o
°
6°
16
B
0
246
C«Q-lb/sf/d
8
Figure 57.
Effluent suspended solids vs. solids loading,
Addison South - Phases 2 and 3.
132
-------
15
10
E
i
o
0
DP-Phase 3
O O
o
co Q
o
FIG. A
0
4 8 12 16
Headloss-in.
15
10
i
o
0
DP-Phase 3
o Low Head
• High Head
•
•
o
•
• •
0 0
FIG. B
0 0.2 0.4 0.6 0.8
C0Q-lb/sf/d
Figure 58. Effluent suspended solids vs. headloss and solids
loading, Des Plaines River - Phase 3.
133
-------
Figure Plant
59A Addison North, Harrington
59B Des Plaines River, Romeoville
60A Addison South
60B Lake Zurich, Lisle, Marionbrook
The Addison North data show a strong correlation over a narrow range of
solids loading. If the data point at high solids loading is excluded for the
Harrington data, no trend is apparent. Grab samples at the Des Plaines River
plant were obtained during early morning hours when headless was low (2 to 5 in.).
The Des Plaines River data show an opposite correlation for C vs. CQQ, and the
Romeoville data show no apparent trend.
The solids loadings for the Des Plaines River and Romeoville plants are
much lower than for other plants. The trend shown for the Des Plaines River
data may be attributed in part to the smaller particles usually present for low
values of Co which experience deeper penetration of the bed. As solids load-
ing increases, particle size also increases and collection goes through a maxi-
mum. At higher solids loading performance is decreased because previously de-
posited particles are sheared from the bed.
The Addison South data in Figure 60A show wider scatter than for Phase 3.
At a solids loadings of 1 Ib/sf/day, C is about 5 mg/; compared to 10 mg// for
Phase 3 suggesting a different particle collection mode, as discussed later.
Mean dnm was 12 ^m for Phase 4 compared to 8 pm for Phase 3.
Mean drm for the Lake Zurich, Lisle, and Marionbrook data shown in Figure
60B was 12 ym for each plant. These plants all have tertiary ponds ahead of
the filters. At a solids loading of 1 Ib/sf/day, C is about 10 mg/{,, the same
as for the Addison South Phase 3 data. For the Marionbrook dry weather flow
data in Phase 1, C was about 13 mg/^ at a solids loading of 1 Ib/sf/day.
These factors strongly suggest that in-depth filtration was the probable
particle collection mode for these plants.
Number Mean Particle Size
Plots of C/GO vs. number mean particle size are shown in the following
figures :
Figure Phase Plant
6 LA 3 Addison South
61B 3 Des Plaines River
62A 4 Addison North, Barrington
62B 4 Des Plaines River, Romeoville
63A 4 Addison South
63B 4 Lake Zurich, Lisle, Marionbrook
If the data point at 9.5 pjn is excluded for Addison South, the plot in
Figure 61 generally shows a decrease in C/C0 for larger d^ in a very narrow
range as would be expected from theory. For Des Plaines River, the data at
134
-------
10
0
FIG, A
OAN
DBA
°0
" O
D 0°
O
0
1
C0Q -
2
LB/SF/D
15
10
CD
0
0
B
ODP
DRO
0
no
0,1 0,2 0,3
C0Q - LB/SF/D
0,4
Ftgure 59. Effluent suspended solids vs. solids loading, Addison
North, Harrington, Des Plaines River and Romeoville -
Phase 4.
135
-------
30
20
10
0
FIG< A
ADDISON SOUIN
GLOW FLOW
OPED, FLOW
A HIGH FLOW
o
QD
o
oo
A-jA
"°
0
CoQ - LB/SF/D
30
20
c_>
ID
FIG, B
1-
OLZ
nil
0
1 2 3
Cod - UB/SF/D
Figure 60. Effluent' suspended solids vs. solids loading, Addison
South, Lake Zurich, Lisle, and Marionbrook - Phase 4.
136
-------
o
O
0.5
0.4
0.3
0.2
0.
0
Addison South-Phase 3
O Low Flow
• Med Flow A
'A High Flow
A No. 2 A •
A
FIG. A
8
10
0.3
Oo0.2
0.1
Des Plaines R.-Phase 3
FIG. B
I I
12 14 16 18 20 22
Figure 61. C/CO vs. particle size, Addison South and Des Plaines
River - Phase 3.
137
-------
0,6
0,4
0,2
0
FIG- A
OAN
DBA
0
10 20
DNM - MICRONS
30
U,O
5?
0,2
0
FIG, B 0 DP
0 DRO
D D
O
D
a °
o n .
0
o
i i i
0
10
20 30
- MICRONS
Figure 62. C/C0 vs. particle size, Addison North, Harrington, Des
Plaines River, and Romeoville - Phase 4.
138
-------
o
O
0
0.3
0.2
O.I
n
i i i
Addison South- Phase 4
- O Low Flow
• Med. Flow
A High Flow ^£
A No. 2 n°»
A
* * A
°A o
FIG. A
i i i
-
6
.
-
0
0.6
0.4
0°
O
0.2
0
10
15
20
Medium Flow-Pnase 4
OLZ
• LI
AMB
O
FIG. B
10
15
20
Figure 63. C/CQ vs. particle size, Addison South, Lake Zurich, Lisle
and Marionbrook - Phase 4.
139
-------
high headless are erratic but the low headless data show a weak dependence of
on d
In Figure 62 , relatively weak correlations are apparent for Harrington
and Romeoville. Except for one data point, a linear decrease in C/CO for in-
creasing dnm appears to occur for Addison North. The Des Plaines River data
show the strongest correlation over the relatively narrow range of d^ of 12
to 18 ^m.
The Addison South data shown in Figure 63 show a strong correlation except
for the data point at 20 (jm. Insufficient data were obtained for Lake Zurich,
Lisle, and Marionbrook to show any trends.
In order to compare results with those reported in the literature, data
were analyzed to determine constants for:
Log^ad^, (23)
for the data sets that showed the strongest correlation. Results are shown
below.
b Significance Level
Addison South Plant
Low 0 0.9 .20
Medium Q 0.8 .05
High Q 0.4 .23
Addison North, phase 4 0.5 .12
Des Plaines River, Phase 4 2.0 .02
The significance level, or the probability that the result occurred by
chance, varies from 0.02 to 0.23 for the above exponents. The power function
coefficient of 2 for the Des Plaines River data is the same as that reported
by FitzPatrick (1972) for classical interception and gravity theories. He
found by experiment about a 1.5 power dependence for particles smaller than
10 |jan. When gravity settling was significant, the dependence increased to the
2nd power.
Although the significance is not high, the Addison South data, where the
data point at 20 \ya was excluded, show a weaker dependence at higher flow rate,
Refiltration Parameters
Since both R and E increase with increasing particle size, and G/CO
decreases with increasing particle size, it would be expected that C/CO would
decrease with increasing R and E. Plots of C/Co(log) vs. R are shown in the
following figures:
Figure Phase Plant
64A 2 Addison South #2, Hsiung's data
140
-------
0
0
O
I.U
.6
.4
.2
.1
.06
n
-------
64B 2 Addison South, variable flow
65A 3 Addison South
65B 3 Des Plaines River
66A 4 Addison North
66B 4 Harrington, Des Plaines River,
Romeoville
67 4 Addison South
The Addison South Phase 2 data in Figure 64 show a higher correlation with
R than Hsiung's data, with C/CO decreasing for larger values of R, as would be
expected. Wider scatter is apparent in Figure 65 for the Addison South Phase 3
data, where in-depth filtration apparently dominates, as compared to Phase 2.
The Phase 3 Des Plaines River data show wide scatter with a decrease in perfor-
mance for higher headloss.
The Phase 4 Addison North and Harrington data in Figure 66 show a strong
correlation with R. Weaker dependencies are apparent for Des Plaines River
and Romeoville. The Addison South data in Figure 67 generally show a weak
dependence.
The plots of C/C (log) vs. E are shown in the following figures:
Figure Phase Plant^
68A 2 Addison South #2, Hsiung's data
68B 2 Addison South
69A 3 Addison South
69B 3 Des Plaines River
70A 4 Addison North, Harrington
70s 4 Des Plaines River, Romeoville
71 4 Addison South
Hsiung's data in Figure 68 show a stronger correlation with E than that
obtained for R, with C/CO (log) decreasing for larger values of E. In contrast,
the Addison South Phase 2 data show the opposite result. The range of E found
by Hsiung was 1.3 - 2.3 as compared to 1.3 - 1.9 for this study.
The Addison South and Des Plaines River data for Phase 3 show weak, if
any, dependence on E. A similar result was obtained for C/CO (log) vs. R.
Strong correlations are apparent for the Addison North, Harrington, Des Plaines
River, and Romeoville data in Figure 70, suggesting that straining may be the
dominant particle collection mode for these data.
No apparent relationship is present for the Addison South Phase 4 low
flow data in Figure 71. These data cover a very narrow range of E frdm 1.7 -
1.8. The medium and high flow data do, however, show a decrease in C/CO (log)
vs. E over this narrow range.
In order to test the data sets for the relationship between R and G,
regressions were made using an equation of the form C = f (Co, Q, R, G). For
one series of tests, the regression coefficient for Q was fixed at 0.26. Re-
sults are shown in Table 26 . The ratio of the exponents used by Hsiung (1972)
142
-------
o
O
\
O
.0
.4
3
. ^
.2
i
AS-Phase3
O Low Flow
a Med Flow
A High Flow
VNo. 2
A
D
O
-
i
1 1 4
A A
DA FIG' A "
$
V A AQ
O
D
0
°^P V
v* fr
o °
0
, ° ,
2.5
3.5
4.5
o
O
\
O
.3
.2
DP-Phase 3
O Low HL
• High HL
6 8
R
FIG. B
0 0
10
12
Figure 65,
C/CQ (log) vs. R, Addison South and Des Plaines River -
ttiase 3.
143
-------
O
O
I.U
.6
.4
.2
.1
.06
i i i
AN-Phase 4
-
FIG. A
-
0 0
0°
o
-
i n i i
04 8 12 l<
R
1.0
.6
4
O .2
.1
o
O
0
FIG. B
A A
on
Phase 4
DBA
?DP
A RO
a o
a
R
8
Figure 66. C/CO (log) vs. R, Addison North, Harrington, Des
Plaines River, and Romeoville - Phase 4.
144
-------
1.0
.8
.6
.4
.3
o
O
.1
.08
.06
0
A
8
Addison South-Phase 4
O Low Flow
• Med. Flow
A High Flow
A No. 2
8
R
Figure 67. C/C (log) vs. R, Addison South - Hiase 4.
10
145
-------
o
O
O
1.0
.6
.4
.2
.1
.06
.04
FIG. A
o o
o
o o
Legend
O Hsiung
• AS2, Phase 2
.2
1.4
1.6 1.8
E
o
O
O
1.3
1.4
1.5
1.6
1.7
2.0 2.2
IAJ
.6
.4
.2
.1
.06
n/i
i i
A
- o A
w
A
0 •
Legend O
- AV- Phase 2
O Low Flow
~ • Med Flow
A High Flow ,
i i i
FIG. B
A
f A ~
O ij
«• o * -
&
0
o
1 1 1
.8
Figure 68. C/CQ (log) vs. E, Hsiung's data, Addison South - Phase 2.
146
-------
.o
.4
.3
o
O
".2
.1
I
.4
.3
o
O
j
AS-Pha'se 3 '. '
- O LOW Flow FIG. A
• Med Flow A •
A High Flow ^ A A
-A No. 2 9 A
A •
O
•
^ A
•°*?
0 **
0
1 1 1 1
4 1.5 1.6 1.7 1.8 1
E
DP-Phase 3
o LOW HL FIG. B
• High HL • »
9 ° ••
o o
0 •
O 0 O
o
1 1 1 1
9
1.4 1.5 1.6 1.7 1.8 1.9
Figure 69. C/C (log) vs. E, Add!son South and Des Plaihes River -
Phase 3.
147
-------
1,0
,6
3 -2
,1
,06
,04
1,0
,6
,4
1 ,2
,1
,06
FIG, A
ADDISON NORTH
HARRINGTON
1,5
1,6 1,7
E
1,8
FIG, B
a a
• DBS PLAINES R
n ROMEOVILLE
• o
1,5
1,6 1,7
E
1,8
1,9
1,9
Figure 70. C/CQ (log) vs. E, Addison North, Barrington, Des Plaines
River, and Romeoville - Phase 4.
148
-------
,o
,6
,3
O
\
l_>
,2
,1
,08
,06
• i i i i
ADD I SON SOUTH PLANT
•» •
0 LOW FLOW
DMED, FLOW
A HIGH FLOW
r° - ['.
D
a
D 0
O
O
O
1 1 1 1 1
1,7 1,72 1,74 1,76 1,78 1,8
E
Figure 71. C/CO (log) vs. E, Addison South - Phase 4.
149
-------
O
vO
CM
O
CO
II
O
CO
O
o
CO
U
O
<4H
CO
0)
•H
O
0)
O
O
O
•H
CO
CO
(U
M
00
vO
CM
a)
cO
H
— ' • • •
•u 1/1
-------
for R and G in developing the parameter E was d/e = 1/3. The ratios for d/e
for the data in Table 26 for the exponent for Q fixed at 0.26 are 0.56 for
Hsiung's data, 0.51 for Addison South phase 2, and 0.44 for the AN, DP, RO
data. Higher ratios of 0.76 and 0.82 were obtained for the Addison South Phase
3 and the BA, LZ, LI, MB data sets, respectively. The inconsistent data for
Addison South Phase 4 are probably caused by the strong correlation between
Co and E, as discussed previously. The above ratios of d/e are closer to 1/3
for plants where straining was believed to be the dominant particle collection
mode.
DISCUSSION
The physical meaning of the refiltration parameters R and E is somewhat
obscure when applied to secondary effluent because of the wide variations of
particle size and surface chemistry. Clearly the septum used for the refil-
tration test does not directly simulate particle removal in a granular bed
filter. However, there are some similarities.
Small particles below the size counted plugging the 0.45 y#i openings of
the Millipore filter paper would reduce both R and E. The same particles will
be only poorly removed in a granular filter. Thus low values of R and E may
then correlate with high values of C and C/C0.
Treeweek and Morgan (1977b) found that the refiltration test was the best
measure of the effectiveness of cationic polymers in flocculating an E. coli
suspension. The refiltration test simulates to some extent the packing of
several particles in the crevices in the filter medium. Since it is a measure
of the extent of agglomeration of particles, a good correlation with removal
efficiency might be expected because particle-particle interactions are prob-
ably more important than particle-media interactions for filtration of second-
ary effluents.
Refiltration parameters may also be a measure of particle shape. Rough
shaped particles would tend to increase the fluid drag and the refiltration
time. However, suspended particles in this study appeared to have the same
general shape.
Clearly the refiltration test lumps together a number of suspension prop-
erties. This study has shown that there is no relationship betwen Co and refil-
tration parameters except where high C is an indication of carryover of large
floe particles from final clarifiers or the Co range is large. Weak correle-
tions of refiltration parameters and particle size characteristics were obtained
for nearly all data. It is highly likely that particle sizes smaller than the
range counted may have an effect on the refiltration test. Particle size
showed a good correlation with filter performance for several individual plant
data but did not correlate well between plants. In general, refiltration para-
meters showed a strong correlation with C/CO except for plants where in-depth
filtration was believed to be the dominant particle collection mode. For most
situations, E showed a stronger correlation with performance than R.
151
-------
CHAPTER X
DEVELOPMENT AND TESTING OF OPERATIONAL
MATHEMATICAL MODELS--PHASES 2, 3, AND 4
INTRODUCTION
This chapter tests the hypothesis that filter clarification efficiency
can be better characterized by including properties of secondary effluent
suspensions in semiempirical models rather than conventional parameters only.
Methods used to test hypotheses that straining or in-depth filtration was the
dominant particle collection mode for individual runs and for individual plant
data are described. Finally, models describing filter clarification efficiency
are developed and tested using independent data sets.
DETERMINATION OF FILTRATION MODE
Since a preliminary cut of the data between straining and depth filtra-
tion is necessary, more refined attempts to segment and correlate the data were
necessary. Methods used are described below.
Relationship Between dp/dm and Number Flux of Particles
Kavanaugh's (1974) relationship (Equation 7) for the ratio of particle
size to media size (dp/dm) and the number flux (Jjjn) °f particles to the filter
bed was tested using Phase 4 data. A reproduction of his plot of Iogj0 (d-fd^)
vs. log 1O JJJQ (particles cm"2 sec~^_) is shown in Figure 72. For data where
unisize suspensions were not used, dp is characterized by number mean particle
size and the particle number flux is based on total number of particles.
For this analysis, dp was characterized by d^m and dm by the dgo of an-
thracite for dual media filters and the dgO of sand for the single medium fil-
ters. Particle number flux was calculated using the total number of particles.
The range of the data is shown on Kavanaugh's plot and tabulated below.
Phase 4 Kavanaugh
(dp/dm) -1.3 to -2.5 -1.1 to -3.4
JN() 3.9 to 5.8 3.3 to 8.4
Individual data are shown in Figure 73.
The results are influenced to a large extent by the range of d50 of 0.7
to 2.7 mm. Except for Des Plaines River and Romeoville, the ratio of drm/d5Q
152
-------
N
£ c
•° ~u
CO Q
w ' 2
CO *"N «*
.-»• 2 3
£ S *a|
3 g * ' *
£ - 11
« c2 <£
C
•H g ••
« 2 £
C M-l
•i-i c
60 0
01 -r4
^•1 CO
CO
60 -H
^ Q)
u a
o
^"^ t-^t
X) 4J
1 v-l
C ?
0
C3 *O
o
t3 4J
C C
CO T-l
60 a
c cu
Lj t,
•r^ >H
_i^> ^^x
O
0
X) CO
0) -H
£0
CO
tN
CU
gj
h
c
~ ^37
2 •— t
*" UJ ^ .^
uj z <
_ 3 5 x
2 S 01 vi co §
o O o ^ u. ^
I
*• J
j
•1
1
S 1 i
x a 1 L
5 « 1
9 *a !«.
CO vf
« S J* /
52 ••-/
5 Jl
ua a. ^
S S 0 *
55 o -i
_ 8 "" *^
£ c.
* I -* 1
W ^^V. / /
I/ /
lo^
^
ot§
"*l
1
3 T f>
i
O
o *
1
„
0 0 0
«o
' ' 0 -
1
' «:-, -
00 * v
•
, ~ < UJ X
< S = S s: <" ^
*£oS* *£
J S < j UJ UJ <
< 5 a d > 2 3
o a < # o • •
i
j ro t
1 1
03
7
u
i!
N-
-------
CC _J _J
•o ,
03
o>
o>
2 CO Q_ O
< < O IT
O D < >
in
in
in
'E
o
V)
~u
O
Q_
o
O>
o
in
ro
0)
ro
to
00
3
bO
a
1
1
in
i
o
T
in
~
in
ro
I
154
-------
is about an order of magnitude below that proposed by Kavanaugh as the dividing
line for possible pore blocking. It is important to note that his data are
based on unflocculated suspensions. No attempt was made to segment data on
this basis because the data (solid data points in Figure 73), where in-depth
filtration was considered to be significant in subsequent analyses, were clos-
est to the proposed dividing line.
Kavanaugh (1974) also did an analysis for the case of straining of desta-
bilized particles, and tested the relationship of filter coefficient or Log!0
C/C0 and particle number flux using the values of C/CO for the top several
inches of filter media. He found no apparent relationship and noted that a
mat of solids did not form at the filter media surface. A plot of C/CO (log)
vs. Log JN for the Phase 4 data is shown in Figure 74. The data in Figure 74
show wide scatter with no apparent trend.
Testing Data with In-Depth Filtration Model
If straining is the dominant particle collection mode, then a poor fit
with in-depth filtration models should result unless the models predict similar
dependencies on variables. This hypothesis was tested using the Marionbrook
model for dry weather flow (Equation 22). A plot of calculated vs. measured
C is shown in Figure 75 for Jfoase 4 data.
From Figure 75A, calculated C appears to be somewhat independent of mea-
sured C. This result is not surprising since straining is expected to be the
dominant particle collection mode for the fine-grained sand filters. Although
the Addison North and South data appear to follow a linear relationship, the
model greatly overpredicts C suggesting that straining is significant.
An excellent fit is apparent for the Lake Zurich, Lisle, and Marionbrook
data in Figure 75B, where C is predicted within ± 25 percent for 85 percent of
the data, strongly suggesting that in-depth filtration is the mode of particle
capture for these situations. The model greatly overpredicts C for the Barring-
ton data, suggesting that straining is significant.
Evaluation of Data on a Common Basis
A method was necessary to evaluate the data on a common basis. This was
done by assuming the following relationship for straining:
C/C0 = a C0 (f Ed , (24)
where a, b, c, and d are constants.
Rearrangement of Equation ( 24 ) gives:
C/C,
(25)
and taking logs of both sides yields
155
-------
1.0
0.8
0.6
0.4
o
O
0.2
O
O.I
.08
.06
/A
Ao
.?
Legend
oAN • LZ
a AS «LI
ABA A MB
vDP TRO
3.5 4 4.5 5 5.5
Log JNolParticles-cm^sec")
Figure 74. C/C (Log) vs. log particle number flux (J«_) -
Phase 4. °
156
-------
20
15
10
0
20
15
S 10
5
5
0
/
a*
/
• AS
a AN
5 10 15 20 25
MEASURED C - MG/L
' /
V
A /
• «QB
/*
• BA
A MB
/
B
0
5 10 15 20
MEASURED C - MG/L
25
Figure 75. Calculated vs. measured C, Marionbrook in-depth filtration
model, (Eqtn. 22) - Phase 4 data.
157
-------
c/c0
Log -g-^- = Log a + b Log E. (26)
Rather than compress the data by plotting E on a log scale, data were "normal-
C/C0
ized" by plotting Log —- vs. E.
GO Qc
Regression coefficients obtained for Equation ( 24 ) for the Addison South
Plant data, where flow to individual filters was varied, were 0.49 for b, 0.26
for c, and -10.6 for E. Regressions were also run for a number of combinations
of individual data sets. If flow to individual filters was not varied, the
coefficient c varied over a wide range with some negative values. A similar
result was obtained for the Phase 1 data. For the Addison North, Des Plaines
River, and Romeoville plant data, where straining was believed to be the domi-
nant particle collection mode, regressions were performed using a fixed coef-
ficient at 0.26 for c. The resulting coefficient for b was 0.4. Values of
0.26 for c and 0.4 for b were then used to normalize the data.
Plots of (C/C0)/C0°-4 Q0-26 vs. E are shown in Figures 76 and 77 for two
groupings of the data. Trends for the other groupings are shown as a dashed
line on each figure. For the Addison North, Addison South, Des Plaines River,
and Romeoville plant data, shown in Figure 76, straining is considered to be
the dominant particle collection mode.
In Figure 77, the individual data for Lake Zurich, Lisle, and Marionbrook
show the opposite dependence on E. This further suggests that in-depth fil-
tration is the dominant particle collection mode for these plants.
The Harrington data do not correlate well with the in-depth filtration
model from Phase 1, but show a strong correlation with E in Figure 77. How-
ever, removal efficiency for Harrington is higher than the plants in Figure
77 at equivalent E values. This result cannot be explained on the basis of
data obtained in this study. Perhaps in-depth filtration and straining are
of nearly equal importance in this case. This postulation may be supported
by considering the slope of the trend line for the Harrington data and the
plants tn Figure 76. The steeper slope in Figure 76 indicates a stronger
dependence of performance on E if straining is the apparent dominant particle
collection mode.
Dimensional Straining Group
In order to segment data for better correlation using the straining model
(Equation (24)) a semi-empirical dimensional grouping to characterize
straining was developed. The approach used is shown in Appendix C. The
grouping selected was:
NSTR ' C^~T^- <27)
158
-------
,2
,1
,08
^_ ,06
C3
^v.
/^-^
O
,04
,02
,01
L\ .
:\ A
\
1,4
\
\
TREND LINE
BA
• AN
n AS
A DP
O RO
TREND LINE
a
V °
^v .\
v
\
1,5
1,6
1,7
1,8
1,9
Figure 76. Normalized data plots, Addison North, Addison South,
Des Plaines River, and Romeoville.
159
-------
,3
,2
CM
£
,1
,08
,06
,04
,02
,01
1,3
\
\
V^TREND LINE
\ AN, AS,DP, RO
\
\
• BA
D LZ
All
OMB
1,4
TREND
\
\
\
1,5
1,6
1,7
1,8
Figure 77. Normalized data plots, Barrington, Lake Zurich, Lisle, and
Marionbrook.
160
-------
The filter medium dimension used was the effective size (d^) of the anthracite
for dual media filters and sand for single medium filters. Values of Ng-j^ cal-
culated from the data cover several orders of magnitude and were used to segment
all Phase 4 data into two equal groups for high and low NSTR. Data are analyzed
on this basis in a later section.
Refiltration Parameter E and C0
Data are analyzed in a later section for segmenting data for values of E
less than or equal to and greater than 1.65, and for Co less than or equal to
and greater than 18 mg/£. These values were selected to divide the 38 data
where refiltration data were obtained in Phase 4 into two equal segments. The
distribution of data for each of the plants in the various groups is shown in
Table 27.
DEVELOPMENT OF OPERATIONAL MATHEMATICAL MODELS
Data for Phases 2, 3, and 4 were analyzed using the three models shown in
Equations 18, 19, and 20 using multiple regressions and log transformations.
A viscosity term was not added to the models because data collection covered
a short time frame where variations were insignificant. Selected models
including suspension properties were then tested using their own data.
Regression Coefficients for Filter Media Depth and Grain Size
The effect of filter media depth and media grain size on performance was
tested for all Phase 4 data using an equation of the form:
C = a <£
-------
Table 27. Segmented data for high and low Nc_, C and E.
o IK O
Addison North
Addison S.-#2
Addison S.-M-7
Harrington
Des Plaines R.
Lake Zurich
Lisle
Marionbrook
Romeoville
Total
No. of
Data
6
4
18
7
6
9
6
5
5
66
N
No.
NSTR
Low
2
2
11
3
2
6
6
0
1
33
High
4
2
7
4
4
3
0
5
4
33
of Segmented Data
c
o
Low
4
0
0
5
1
7
6
0
4
27
High
2
4
18
2
5
2
0
5
1
39
E
Low
1
0
0
4
3
9
6
5
4
32
High
5
4
18
3
3
0
0
0
1
34
162
-------
r>
Table 28. Regression coefficients for Log — = f(Q,C ,E,d,_-,,d ,d /d_-)
G o 3U nm nm j(J
- Phase 4.
r abc d e fg
Log C/C = a QC Ed d 6
o nm
.68 -.17 - -.058 2.55 .031
(Signif.) (.20) (0) (.77)
Log C/C = a C bQCEd d e
6 o o x nm
.68 -.17 -.014 -.058 2.60 .024
(Signif.) (.87) (.21) (0) (.83)
Log C/C = a QCEdd & d_nf
o nm 50
.70 -.15 - -.097 2.22 .098 .19
(Signif.) (.05) (0) (.37) (.08)
Log C/C = a C bQCEdd e dc_f
o o x nm 50
•70 -.16 -.032 -.097 2.31 .085 .19
(Signif.) (.71) (.06) (0) (.46) (.07)
.68 -.20 - -.074 2.50 - - -.048
(Signif.) (.13) (0) (.44)
163
-------
has an effect on filter performance. A weak dependence occurs for analysis of
data between plants. Apparently the effect of particle size on filter per-
formance may be negated by variables with stronger dependencies for analysis
of all Phase 4 data. In addition, the particle size distributions vary widely
between plants, as shown in Figures 44 to 48 , where individual plant data
exhibit log normal, bimodal, and random distributions. Thus, a single para-
meter such as djyjj may not adequately characterize the suspension for com-
parisons between plants.
Development of Models for C = f (Co> Q) and C = f(CQ, Q, E)
The method discussed in Chapter VIII, where regression coefficients and
the slopes and intercepts of plots of calculated vs. measured C are deter-
mined, was used to evaluate models for individual plant data and for group-
ings of data, if a model adequately describes the filtration process, then
the intercept of the line of best fit of the plot of calculated vs. measured
C should be close to zero and the slope close to one. Slopes less than
about 0.6 generally indicate a poor fit of the model. Regression coefficients
and slopes and intercepts for the plots of calculated vs. measured C are given
in Table 29. Except for variable flow plants, the coefficient for Q was
fixed at 0.26, the value obtained for the Addison South Phase 4 data.
For Hsiung's data, the straining model fits the data much better than
the in-depth model with a slope of 0.89. The magnitude of the correlation
coefficients for Hsiung's data is greatly influenced by two data points at
56 and 64 mg/£, which leads to some question as to the validity of the model.
It will be shown later that the model for straining has high validity since
it predicts performance adequately for some of the Phase 4 straining data and
some of the Phase 4 models provide a good fit for the two data points.
For the Addison South phase 2 data, the slopes are less than 0.5 and no
significant improvement occurs for inclusion of E in the correlation. During
Phase 2, Co was very high and averaged 143 mg/^,. At the high solids levels,
it is unlikely that the filter media was adequately cleaned daring backwash-
ing and solids breakthrough probably occurred. This may be the reason for the
relatively poor correlation and weak dependence on E. The Phase 3 data in
Table 29 show a strong correlation for the in-depth filtration model and a
slight improvement when E is included although the slope is low. As dis-
cussed in Chapter IX, in-depth filtration is considered to be the dominant
mode for these data.
For the Addison South Phase 4 data, no improvement in correlation is
apparent for inclusion of E in the model although the regression coefficient
for E is very high (-10.6). For the straining model, the significance levels
for the coefficients for Co and Q are 0.60 and 0.47, respectively. These
results are caused by the very strong correlation between Co and E, shown in
Figure 56. Thus it does not matter which of the variables Co and E is
included in the model. The validity of the straining model is discussed later.
The straining model for Addison North, Des Plaines River, and Romeoville
shows a very strong correlation with a slope of 0.83. A very poor correlation
with r of 0.10 and slope of 0.07 is obtained for the in-depth filtration model
164
-------
CO
4-1
O
r-l
a.
0)
a
o
r-l
ca
T)
C
CO
4J
a
0)
o
h
ID
4J
c
a
CO
0)
T)
i
r-l
O
W 0)
C M
0) 3
•H co
O CO
•H 0)
m g
m
0) •
o co
o >
d T)
O V
•i-l -U
ca CD
CO i-l
CO
H
*T3
W
o
O'
o
a
H
CJ
0
O"
U
to
II
U
r-l
01
•o
o
*C3
U
»fi
S
°
CO
U
o
eu
CJ
ui
3-
O CO
-* r-l
-» ON
-3- in
00
to m
i— i
in co
tf> 00
CM CO
> >
< <
~*
06
CO
1
*
1— 1
0
r-l
O
CM
r-l
in
to
r-4
CM
r*
r~
*
ON
in
CM
r*.
00
CO
in
oo oo
r*-
CM *
*
00 r-i
oo r-
CO ON
CM .3-
ON r-.
tO O
O ON
CM CO
CM O
00 r*
O
a:
* S
-? 3
r* r^
J- in
CO rJ
i
00
* CM
tO -*
O in
r-J
CM to
00 tO
r- m
in r-.
r- o
CM CM
ON «M
r~ ON
CM
* CO
oo m
r^ to
r~ o
^- 00
m ON
m to
OO CM
CM CM
00 0
P^ ON
s
H- 1 03
J lie
S 3
< CM"
ca -J
ON
CM
r-l
1
*
CO
CO
in
o
CO
— 1
CO
ON
CO
P-*
-*
*
ON
r-l
P^
in
CM
CO
r-l
00
-3-
r-4
.*
r4
iJ
01
Z
!
ON
CO
r-t
tO
CO
ON
m
r-l
tO
0
0
S
•~J>
r-l
X
O CO
co in
CM .*
1 1
* *
r*- to
O CO
r-l *-*
•* rJ
m m
r-4 CM
ON 00
to o
—1
CO P-
oo r-
* *
CM to
00 00
to 00
CO CM
CO CM
in in
oo in
O to
i
00 to
in i->
U
[d
X
-J X
o
r-l
CL
c
a)
<=
X O
r4 U
O V4
UJ O
c «
01 U
•* C8
O 00
U *
* #
165
-------
for these plants.
A good fit is apparent for the in-depth model for Lake Zurich, Lisle,
and Marionbrook. The exponent for Q of 0.32 is nearly identical to the 0.34
value obtained for the Phase 1 Marionbrock dry weather flow model. The ex-
ponent for E in Table 29 is 1.57, the only positive value obtained. This is
not surprising since the normalized data plot in Figure 77 show an increase for
C/GO
Log with increasing E for those individual plants. The Barrington
C04 Q-26
plant data were grouped with these plant data.
For segmenting of the data, the best fits were obtained for the high
NSTR and ni§h E models with slopes of 1.03 and 0.82, respectively. Very poor
fits are apparent for the low NSTR and l°w Co models. The regression coeffi-
cients are strikingly similar for Hsiung's data and for the high E data which
cover different C0 ranges:
b> v Range of C
a *• o^ d(E) mgAf,
Hsiung .57 1.34 -4.14 10-64
High E .51 1.36 -4.53 1-12
This result obtained for different ranges of C and Co greatly enhances the
validity of both models since they can predict outside of the range in which
they were developed.
TESTING OF MODELS WITH OTHER DATA
Selected straining models developed from Phase 4 data were tested, using
independent data. Statistics are given in Table 30. Plots of calculated vs.
measured C are shown in the following figures for Hsiung's data:
Figures Models
78 High and low NSTR
79 High and low E
80 AV4 and AN, DP, RO
As discussed previously, the two data points for C of 56 and 64
have a large effect on the correlations for Hsiung's data. The high
model overpredicts C for the two data points by 20 to 25 percent, as con-
trasted with the high E model which underpredicts by approximately the same
amount and the AV4 and AN, DP, RO models which overpredict by about 10 to 45
percent. These results strengthen the validity of the two data points. The
low NgijR model provides a very poor fit. For other models, the lower C data
show wide scatter with a tendency to overpredict for four of the data points.
The other data points are relatively close to the line with a slope of one.
The data point at a measured C of 12 mg/^ in Figure 80, where C is greatly
166
-------
Table 30. Intercepts and slopes for plots of calculated vs. measured C for
testing models with independent data.
Data
Set
Hsiung
High E
AV 4
AN, DP, RO
AV 4
AN, DP, RO
Model
High N
Q f* T*
Low N .
str
High E
Low E
AV 4
AN, DP, RO
Hsiung
AN, DP, RO
AV 4
r*
.93
.91
.93
.87
.72
.82
.77
.84
.67
.83
.81
Intercept
mg/1
-6.8
7.1
1.0
8.9
-7.6
-5.6
2.7
1.4
1.7
1.9
0.2
Slope
1.24
.07
.72
.49
1.24
1.12
.75
1.01
.49
.79
1.01
*Bivariate correlation coefficient for plot.
167
-------
6
o>
80
70 -
60 -
50
40
J> 30
a
O
20
10
0
Model
• High NSTR
a Low NSTR
High NSTR ~7
Low NSTR
10 20 30 40 50
Measured C-mg/l
60 70
Figure 78. Calculated vs. measured C, models for high and
low NorTO, Hsiung's data.
3 IK
168
-------
70
60
50
e
6 40
"O
ZJ
o
30
o 20
10
0
Model
• High E
n Low E
Slope =
n
»•
„
0 10 20 30 40 50 60 70
Measured C-mg/
Figure 79. Calculated vs. measured C, models for high and low E,
Hsiung's data.
169
-------
90
80
70
60
O 50
-a
-------
overpredicted, results from a very low value of E and the high negative coef-
ficients for E in the AV4 and AN, DP, RO models.
The straining model developed from Hsiung's data was tested using selected
data from Phase 4. A plot of calculated vs. measured C is shown in Figure 81
for the high E data. Except for the Addison South data, a good fit is appar-
ent. Three of the AS data points at a measured C of 5 to 7 mg/£, where C is
overpredicted, are for Filter No. 2. The Filter No. 2 data were not used in
development of the AV4 model because only four data were obtained and may be
suspect because of malfunctions in the air scour system. C is predicted
within ± 50 percent for about 75 percent of the data in Figure 81.
The plot for the AV4, AN, DP, and RO data for the model obtained from
Hsiung's data is shown in Figure 82 . The line of best fit for the AV4 data
has a slope of 1.0 but overpredicts C slightly. A better prediction is ob-
tained for most of the AN, DP, and RO data. For all four plants, C is pre-
dicted within ± 30 and 40 percent for about 70 and 80 percent of the data,
respectively.
The AN, BA, DP model is tested using the AV4 data in Figure 83. The
data are not as well correlated as in the plot for the AV4 model and AN, DP,
RO data shown in Figure 84. If the data points for E ^ 1.65 are excluded, a
near perfect fit of the data is obtained in Figure 84. The Addison North
data point that falls on the line with slope of one has an E of 1.64. In
Figure 84, C is predicted within ± 30 percent for about 80 percent of the
data. The AV4 model gives an excellent fit for the data with an intercept
of 0.2 mgAf, and a slope of 1.0 for the line of best fit, although there are
several outlying points. C is predicted within ± 20 percent for about 60
percent of the data and within ± 40 percent for 80 percent of the data.
The AV4 model provides an excellent fit for the AN, DP, RO data, even
though the coefficients for Co and E have low significance, as discussed pre-
viously. The regression coefficients for E for the AV4 and AN, DP, RO models
have high negative values of -10.6 and -8.28, respectively. The high value
for the AV4 model may have resulted from the strong correlation between Co and
E. The high negative coefficients may be the reason why some of the calcu-
lated C data in Figure 80, for testing with Hsiung's data, are low.
MODEL SELECTION AND APPLICATION
The validity and utility of a model is greatly enhanced if it can ade-
quately predict outside the range of data from which it was developed. Two
of the models developed from Phase 4 data for C less than 12 mgAf, were able
to predict performance at C = 56 and 64 mgAf, with ± 20 to 25 percent. Testing
of the model obtained from Hsiung's data with high C gave predictions within
± 30 percent for 70 percent of the data where straining was considered to be
the dominant particle collection mode.
As in Phase 1, filter media size and depth did not correlate with filter
performance. The AV, DP and RO plants represent the extremes in media size
and depth encountered in the study. The fact that the data from these plants
correlate well further supports the conclusion from Phase 1 that filter per-
171
-------
25
_ 20
E
i
o
15
^
"o
O
O
10
0
Slope =
a
High E
a
Legend
AN -LZ
AS —LI
BA —MB
DP T RO
I I
0
5 10 15 20
Measured C-mg/
25
Figure 81. Calculated vs. measured C»i model for Hsiung's data,
high E data.
172
-------
E
i
o
16
14
12
10
0)
6
O
O
0
Data
o AV4
• AN
A DP
ARO
AN,DP,RO
0 2 4 6 8 10 12
Measured C-mg/l
Figure 82. Calculated vs. measured C, model for Hsiung's data,
Addison South, Addison North, Des Plaines River, and
Romeoville data.
173
-------
10
01
ft
8
£
I
O
TJ 6
o>
_o
o
O
0
0
Addison South
o Low Flow
o Med. Flow
A High Flow
D
-------
10
8
CJ3
C.J
J.65
AN 0 Q
DP • Q
RO A A
2468
MEASURED C-IWL
10
Figure 84. Calculated vs. measured C, Addison South straining
model, Addison North, Des Plaines River, and
Romeoville data.
175
-------
formance is relatively independent of media size and depth over the ranges
encountered in this study.
In order to develop empirical models of filter performance, it is neces-
sary to vary the flow to individual filters receiving the same influent sus-
pension. In flow variation tests, filter performance varied as a function of
flow to the 0.27 to 0.45 power if refiltration parameters were not included.
With inclusion of the parameter E, the regression coefficient for flow was 0.26
for one plant as compared to 0.34 for the Marionbrook in-depth filtration
model developed in Phase 1. The difference in Q° for these two exponents is
relatively small (5 percent at 2 gpm/sf).
For plants where in-depth filtration was believed to be the dominant
particle collection mode, inclusion of E in models did not result in a sig-
nificant improvement in correlations. Prediction of filter performance can
be greatly improved using empirical refiltration parameters as opposed to
solids loading if straining is the dominant particle collection mode. In
testing of models with data from other plants, both the straining and in-depth
models tend to overpredict effluent suspended solids if the other particle
collection mode dominates for most data.
From this study, the model considered to best represent in-depth fil-
tration is the Marionbrook dry weather flow model:
C = 0.40 C086 Q-34 (22)
The Marionbrook model is considered to be best because it provided the best
fit for Phase 1 data and for data in later phases where in-depth filtration
was considered to dominate.
For straining the choice is between the model developed from high E data,
the AN, DP, RO model, and the Addison South Phase 4 model:
High E (> 1.65)
C = .51 Co'36 Q-26 E'4-53 (29)
AN, DP, RO
C = 4.24 C^-40 Q-26 E-8-28 (30)
AV4
C = 10.1 C0^49 Q-26 E'10-6 (31)
The model for the high E data has a much lower negative exponent for E than
the values of 10.6 and 8.28 obtained for the Addison South and AN, DP, RO
Phase 4 data. A lower negative exponent may be more desirable because the
model would tend to overpredict C less if in-depth filtration is the dominant
particle collection mode. However, the model for high E data tends to over-
predict C for the Addison South data.
176
-------
The high E model provides the best fit for Hsiung's data and has nearly
the same constant and exponents. However, the near perfect fit obtained in
Figure 84 for the AV4 model tested with AN, DP, RO data, if the data for
E ^ 1.65 are excluded, strongly favors its use for high E values. Since the
AN, RO, DP model has a lower negative exponent for E, it is the model of
choice for high E.
A method does not now exist to determine a priori which particle collec-
tion mode will be dominant over a range of operating conditions. Even so, it
is believed that these models can now be applied with caution in predicting
filter performance. For plants that have tertiary ponds ahead of the filters,
the in-depth model can be applied since particle size and E will likely be
small, as shown for the Lake Zurich, Lisle, and Marionbrook plants. In any
event, particle size and refiltration tests should be conducted before the
model is applied.
For sand filters, the nearly perfect fit for high E strongly suggests
that the AN, DP, RO model can be applied directly (c.f., Figure 84). Until
further work is done, the model should not be used for suspensions with low E.
In the interim, the Marionbrook model should be used for low E data. It pro-
vided a relatively good fit for the sand filtration data reported by EPA
(1975) (c.f., Figure 43).
For dual media filters, the situation is not as clear since the AN, DP,
RO model tested with the Addison South data does not provide a significant
improvement over an in-depth filtration model for Addison South tested with
its own data. As discussed previously, this may be caused by the high corre-
lation between Co and E. A further uncertainty is that the Harrington data
does not correlate well with the AN, DP, RO model, as shown below for E £ 1.65.
Calculated C Measured C
11.4 6.5
6.2 3.8
3.8 3.0
For dual media filters, it is believed that the model should not be used
unless djuji and E are greater than about 15 \o& and 1.7, respectively.
177
-------
LITERATURE CITED
Aitchison, J. and J. A. Brown, The Lognormal Distribution, Cambridge University
Press, 1957.
Baumann, E. R., and J. Y. C. Huang, "Granular Filters for Tertiary Wastewater
Treatment," J. Water Pollution Control Federation, 46 (8), 1968-1973,
1974.
Cleasby, J. L., and E. R. Baumann, "Wastewater Filtration-Design Considera-
tions," Environmental Protection Agency, EPA-624/4-74-DOT, July 1974.
Cookson, J. T., "Removal of Submicron Particles in Packed Beds," Environ-
mental Science and Technology, 4 (2), 128 (1970).
Craft, T. F., and G. G. Eicholz, "Mechanism of Rapid Filtration in a Uniform
Filter Bed," Water Resources Research, 6 (2), 527 (1970).
Crandall, C. J. "Evaulation of Tertiary Solids Removal Techniques, 49th Annual
Meeting Central States Water Pollution Control Association, Arlington
Heights, Illinois, May 1976.
Gulp, R. L., and G. L. Gulp, Advanced Wastewater Treatment, Van Nostrand
Reinhold Co., New York, 1971.
Dallavalle, J. M., C. Orr, Jr. and H. G. Blocker, "Fitting Biomodal Particle
Size Distribution Curves," Ind. Engr. Chem. 43 (6):1377 (1951).
Dean, R. B. and L. F. Stanley, "The Reliability of Advanced Waste Treatment,"
Preprint, U.S. Environmental Protection Agency, Municipal Environmental
Research Laboratory, Cincinnati, Ohio, December 1975.
Dahab, M. F., and J. C. Young, "Unstratified Bed Filtration of Wastewater,"
J. Environmental Engineering Division, American Society of Civil
Engineers, 103 (EE1), 21-36 (1977).
Eckenfelder, W. W. and J. A. Roth, "Effluent Variability as Related to
Developing Performance Standards," Conference on Effluent Variability
from Wastewater Treatment Processes and its Control, New Orleans,
Louisiana, December 1974.
Fair, G. M., J. C. Geyer, and D. A. Okun, Water and Wastewater Engineering,
Vol. 2, John Wiley and Sons, New York, 1968.
178
-------
Feuerstein, D. L., "In-Depth Filtration for Wastewater Treatment," Report
to Environmental Protection Agency, Contract 14-12-852, February 1976.
FitzPatrick, J. A., "Mechanisms of Particle Capture in Water Filtration,"
Ph.D. Dissertation, Harvard University, 1972.
FitzPatrick, J. A., and L. A. Spielman, "Filtration of Aqueous Latex Suspen-
sions Through Beds of Glass Spheres," J. Colloid and Interface Sci.,
43, 350-369 (1973).
Ghosh, M. M., T. A. Jordan, and E. L. Porter, "Physiochemical Approach to
Water and Wastewater Filtration," J. Environmental Engineering Division,
101 (EE1), 71-86 (1975).
Guiver, K., and R. Huntingdon, "A Scheme for Providing Industrial Water
Supplies by the Re-use of Sewage Effluent," Water Pollution Control
(British), 70, 75, (1971).
Hall, W. A., "An Analysis of Sand Filtration," J. Sanitary Engineering
Division, American Society of Civil Engineers, 83 (SA3), 1276 (1957).
Heertjes, P. M., "Formation of Filter Cakes and Precoats," Chapter 4,
Scientific Basis of Filtration, Noordhoff Int., Leyden, Netherlands,
1974.
Herzig, J. P., D. M. Le Clerc, and P. Le Goff, "Flow of Suspensions Through
Porous Media—Application to Deep Filtration," Ind. Engr. Chem., 62
(5), 8 (1970).
Hoogerhyde, T. C., "Evaluation of Tertiary Filters for Secondary Treated
Wastewater," short course on "Wastewater Treatment," University of
Michigan, January 1973.
Hsiung, K.-Y., "Filterability Study of Secondary Effluent Filtration,"
J. Sanitary Engineering Division, American Society of Civil Engineers,
98 (SA3), 505-513 (1972).
Ison, C. R., "Dilute Suspensions in Filtration," Ph.D. Dissertation,
University of London, 1967.
Ives, K. J., "Theory of Filtration—Special Subject No. 7," International
Water Supply Congress and Exhibition at Vienna, International Water
Supply Association, Vienna, 1969.
Ives, K. J., "Filtration of Water and Wastewater," Critical Reviews in
Environmental Control, 2 (2), (1971).
Iwaski, T., "Some Notes on Sand Filtration," J. American Water Works
Association, 29 (10), 1591 (1937).
179
-------
Kavanaugh, M. C., "Mechanisms and Kinetic Parameters in Granular Media
Filtration," Ph.D. Dissertation, University of California, Berkeley,
1974.
Kreissl, J. F., "Granular Media Filtration of Secondary Effluent," News of
Environmental Research in Cincinnati, Environmental Protection Agency,
December 1974.
La Mer, V. K., and R. H. Smellie, "Flocculation, Subsidence, and Filtration
of Phosphate Slimes II," J. Colloid Science, 11, 710-719 (December 1956).
Lykins, B. W., and J. M. Smith, "Interim Report on the Impact of Public Law
92-500 on Municipal Pollution Control Technology," Environmental Pro-
tection Agency, EPA-600/2-76-018, January 1976.
Michaelson, A. P., "Under the Solids Limit at Ashton-Under-Lyne," Water
Pollution Control (British), 70, 533 (1971).
Minix, R. J., "Performance Assessment of Wastewater Filtration Using Pilot
Studies," M.S. Thesis, Northwestern University, 1979.
Novak, J. T., Harry Becker, and Andrew Zurow, "Factors Influencing the Thick-
ening and Dewatering Properties of Activated Sludge," J. Environmental
Engineering Division, American Society of Civil Engineers, 103 (EE5),
815-828, (1977).
Oakley, H. R., and T. Cripps, "British Practice in the Tertiary Treatment of
Wastewater," J. Water Pollution Control Federation, 41, 36 (January
1969).
O'Melia, C. R., and W. Stumm, "Theory of Water Filtration," J, American Water
Works Association, 59, 1392-1410 (1967).
Payatakes, A. C., and D. H. Brown, "On the Transient Behavior of Deep Bed
Filtration," 83rd National AIchE Meeting, Houston, Texas, March 1977.
Shea, T. G., and R. M. Males, "Investigation of the Response Surfaces of the
Microscreen Process," Report to Environmental Protection Agency,
Contract 14-12-819, December 1971.
Spielman, L. A., and S. L. Goren, "Capture of Small Particles by London Forces
from Low Speed Liquid Flows," Environmental Science and Technology, 4
(2), 135 (1970).
Springer, S. C., "Influence of Hydraulic Variables on the Performance of a
Full Scale Granular Bed Tertiary Filter," M.S. Thesis Northwestern
University, 1975.
Standard Methods for the Examination of Water and Wastewater, American
Public Works Association, 1971.
180
-------
Tchobanoglous, George, "Filtration Techniques in Tertiary Treatment,"
J. Water Pollution Control Federation, 42, 604-623 (1970).
Tchobanoglous, George, and Rolf Eliasson, "Filtration of Treated Sewage
Effluent," J. Sanitary Engineering Division, American Society of
Civil Engineers, 96 (SA2), 243 (1970).
Tebbut, T. H. Y., "An Investigation into Tertiary Treatment by Rapid Filtra-
tion," Water Research (British),5, 81 (January 1971).
Thomas, H. A., Woodward R. L., and Kabler, P. W., "Use of Molecular Filter
Membranes for Water Potability Control," J. American Water Works
Association, 48, (11), 1391-1402, 1956.
Tiller, F. M., "Theory and Practice of Solid-Liquid Separation," University
of Houston, 1975.
Treweek, G. P., and J. J. Morgan, "Size Distributions of Flocculated Parti-
cles: Application of Electronic Particle Counters," Environmental
Seienee and TeoTmology, 11, 707 (1977a).
Treweek, G. P., and J. J. Morgan, "Comparison, of Coagulation Effectiveness,"
Preprint, Environmental Engineering Science, California Institute of
Technology, 1977b.
181
-------
APPENDIX A
BIVARIATE CORRELATION COEFFICIENTS
Table A-l. Bivariate correlation coefficients for secondary
plant operating data and C , C, and C/C -
Phase 1 - variable flow plants.
Secondary
Plant
Variable
Addiaon South
F/M
Rear. SS-mg/i
Contact SS-mg/i
SVl-mi'tng
D.O.- ing/^
Lake Zurich
F/M
Reaar. SS-mg/^
Contact SS-mg/i
SVI-mi/mg
D.O.-m-i/i
Lisle
F/M
Reaer. SS-mg/^
Contact SS-mg/Z
SVT-m^/mg
D.O.-mg/i
Marionbrook
F/M
Rear. SS-mg/i
Contact SS"mg/i
SVI-mZ/mg
D.O.-mg/2
o .
f
ata
17
35
35
35
35
5
8
12
12
31
12
12
12
12
12
34
34
34
34
31
Mean
.11
6900
4600
70
0.9
.03
3600
2700
100
2.4
.12
5700
4800
140
1.0
.10
5200
4100
130
1.1
r—
Co
-.20
-22
-.11
-.26
.44
.
-
.17
-.24
-.22
.23
.01
-.03
-.04
.57
.21
-.03
.08
.06
.12
r - C
Low
Q
-.47
.59
.63
.51
-.23
.
-
-
-
.51
.56
.16
.09
.28
-.21
-.31
.28
.47
.52
-.26
Med.
Q
-.37
-.22
.10
-.30
.33
.
-
.61
-.36
-.10
X
.42
.62
.29
-.05
-.10
-.26
.21
.38
.29
-.31
High
Q
-.36
.30
.16
-.07
.43
.
-
-
-
-
.75
.34
.08
.10
-.19
-.21
.31
.24
-.19
.08
r - C/C0
Low
Q
-.43
.47
.58
.46
-.38
.
-
-
-
-
.30
-.14
.10
.21
-.19
.24
.12
.16
.13
-.02
Med.
Q
-.35
.09
.19
-.11
.10
—
-
.03
-.02
.22
.24
.46
.27
-.04
-.41
-.45
.22
.28
.09
-.26
High
Q
-.34
.28
.27
-.02
.18
—
-
-
-
-
.43
.08
.19
.10
-.42
-.63
.22
-.06
.42
-.55
182
-------
CD
4J
CO
-a
60
c
CO
H
X)
•^ c
PQ CO
CN
x-v
%w
JJ
1
C
§
•U
5
c
S
TJ
^
o
0
o
u
o
a
CM CM •-< r- >O
•* 0 -1 0 .-1
t i
a
at
& o -^ *O CM
•* CM CM O O »
a)
1 1 c
vj
a
bJ
»"4 O O iH i-* GO
. . . . „
s § §s=;
O i-l CM CO
ON t-* c^ en M
r-4 *J C^ n O
III 1
en r* a> en ^t o
CM r-l O O O «
, I C
f4
00 ON tN ^O O^ CO
o f* irt m o (3
1
oo O O O fcrt
O fn .3* f-* m
• oo ^3-
CM
i-l
1
C CO
« ^ •*
•3 3
,,.,.Jj ,,.9
3 -0
C CO
•H
vn o o oo
O O Q 1 •
in ao oo o CM
•* m i-i i-i
^ iH i-l i-l 1
i
o i-i
i-l CM O CM 1 i-l
• • • . -f*
1 II >
i
CO CM i-l CO O
m o CM CM i etf
1 1 1
CO O O O
i-l O O 00 1
l>- CO
O tN ca eg CM enr^r*.r*o
(si 10 m in in m in m *n
fbO
•» ** g
"M ? f i "So P P ff
S CO HI • S CO M 1
1 CO > O 1 CO > O
co ca • ca co •
ca u o ca u o
• o • o
aw o u
Sis Sac
2 c? 5 * & S
iH i-t CO CO 1
1 1 1
rH i-l O O 1
o O
ca ca •
ca u a
• 0
b
-------
O CO
O O
1-1
§4J
CO
i-l i-(
4J 4->
CO CO
r-< 4J
cu m
p CD
O N
O i-l
CO
CM
i
rH
rH
00
CM
1
00
rH
1
rH
rH
O
1
S
'
O
rH
en
i
e
•o
oo m oo cn CM VO CM rH CM
'
en vo vo oo CM cn rH
-H o o in O o en
lit i
ov o OD vo m cn vo
cn -a- cn r» cn rH CM
iii ii
O J- 00 rH 00 VO O
O O O O rH rH
Svo rH O oo r^ r*-
•^ ^ oo cn cn ^
iii ii
S S i '* ^
S 8 B H CM — oj
•O 'O *O 2 G b **
*
1
rH
CM
.»
cn
i
r^.
CM
VO
rH
O
1
o
VO
in
'
VO
cn
i
S
1
vO
I
CM
en
i
CM
O
CM
1
rH
O
en
CM
i
00
CM
cn
CM
1
in
i
VO
rH
1
CM
1
rH
O
rH
O
rH
in
i
en
CO
CM
in
CM
1
O
in
en
en
1
rH
1
O
1
in
CM
1
o
in
in
CM
cn
1
§
a
i
s
184
-------
TABLE A-3 (cont'd.)
No. of cases
d4m
dnm
dam
dmm
NT
2
CT
CJ/X
% SI
% S2
% S3
% S4
Phase
G tc
AS
9
.17
.35
.22
.18
-.73
.12
.05
-.47
.32
.79
-.20
3
:
DP
19
.10
.22
.21
.16
-.41
.22
-.10
-.05
-.17
.02
.23
Phase 4
G to:
All
38
-.25
-.19
-.24
-.25
.22
-.25
-.28
.04
.06
.16
-.19
Phase
E to
AS
9
.47
.47
.50
.48
-.70
.44
.47
-.34
.07
.61
.19
3
DP
19
.11
.19
.17
.14
-.19
.17
-.12
-.07
-.14
.08
.16
Phase 4
E to:
All
38
.20
.14
.17
.19
.14
.19
.19
.06
-.23
-.14
.16
185
-------
APPENDIX B
CALCULATION OF EQUIVALENT MEDIA DIAMETER
A modifed form of Kozeny's equation for flow through porous medium
was used to calculate d , the equivalent media diameter to give the same
headloss.
Kozeny Equation:
L = Kg~" W 7*^
where
K = Kozeny constant
h = headloss
L = depth of bed
v = kinematic viscosity
e = porosity, approximately 0.4 for sand and 0.6
for anthracite
S = shape factor, approximately 6.1 for sand and
7 .4 for anthracite
d = geometric mean grain diameter
V = velocity of flow
Rearrangement of Kozeny's equation and including headloss components for
sand and anthracite gives
hgK „
W
d
s
e3
s
S "
s
d
d
s
-V2
ea
a
"s "
d
a_
186
-------
where subscripts denote sand and anthracite media. This equation reduces
to
L L
209 -4r + 40.6 -f = A
d d
s a
The sum 209 L + 40.6 L was determined and d defined as
s a e
209 L 4- 40.6 L
s a
L L
209 -|+ 40.6 -f
d d
s a
1/2
187
-------
APPENDIX C
UNITS FOR REFILTRATION AND STRAINING PARAMETERS
Units for E
K" 1/2
From Chapter VI: R = AL p and G = ( a)
t "«.
From Equation 11, K1 has the same units as
1/3 2/3 ,V/3 or
Lt S V
Tl/3 ,L2N2/3 ,L* , u A T
L (7T) (TT) = (qr) » where L and T
L T •*•
are length and time, respectively.
/ - \ •• J 1711 ^lr'^3 .. i
(t—T2V » K = QK ) so units are —
L b J-
T 3 -I /Q j
Units for G: f TTT*' - '—*' ~ -
\ 2, ~ „,
Units for R = Mass (M)
Units for E
) \ 1/2 3 .
L I _ /ML v 1/6
188
-------
Then E should appear in a grouping to the 6th power for mass to cancel out
evenly. Other variables that should be considered are C , Q, d , d,rt,
o nm 1U
and u. For mass to cancel y must be squared in the denominator. If one
c Q E*>
considers the grouping o the units are
U2
M_ L ML3
L3 T T
M2
L2T2
and d and d,rt would have different exponents by 3 for the grouping to be
nm iu
dimensionless . Suspended particle size would be expected to have a much
stronger influence on straining than media diameter for the range encoun-
tered in this study and should therefore have a larger exponent than media
size and be in the numerator.
Thus, if an exponent of 4 was selected for d , an exponent of 7
nm
would be needed for d1Q. For this reason, an arbitrary dimensional group-
ing, dimensionless except for L, was selected to represent straining:
C Q E6 d1*
° ...— 5m
STR u2 d2
10
The group has units L5.
189
-------
APPENDIX D
SURVEY OF U.S. TERTIARY FILTERS
Prior to the conduct of this award (EPA Grant R803212), a survey of exist-
ing and planned municipal tertiary filters in the U.S. was conducted. Lists
of tertiary filter locations and operating data were requested from the 50
states'water pollution control administrators in September 1973 and are
believed current as of June, 1974.
Five states did not respond and nineteen states had no tertiary filters
following municipal effluents at that time.
The following summary of tertiary filters in fourteen states comprised
all tertiary filters in existence at that time except for small biological
treatment package plants (these were quite numerous). A more detailed
summary of the 50 states response is given on Table D-l.
*
State Number of filters Operating information available
Alaska 2
California 7 x
Colorado 4
Illinois 25 x
Indiana 2
Maryland 2 x
Michigan 7 x
Minnesota 5
Mississippi 2
New Jersey 7
New Mexico 2
New York 3
Oregon 4
Texas 5 x
*
design or operating information including suspended solids and BOD have
been received by us or are available in the literature.
190
-------
•
r-4
1
Q
0)
.-i
Cd
H
CN
to ^ a
4J CO 0 CO
co C7 a N^
fi CU v^ / H rt
g
O
CD t-i
Stt
cd c
PH O
00
•S
4J
cd
CU
Qj
°
4J
cd
n
dl $>*>»
^^ cd cri
in S •— ' m
V^ £N
*
&
}-^ ^^
CN <-i m
'ft
a) a) a) cu ftcucucu
22 22ft ^ c^- B 2 2 2
\H1/ fp-4 V^ O
^. oo -H m co
mmmmmm^mmmmmm
^•^^o^r^o^^mo^^i^^n
^•-l>>»-^ ^i-l-».,-l CN^x' ^H-^
cy\ t--i •— * f—t oo o^ cr^
t— i
CO
4J
4J -H
0) ft
cd co cd ft
w o -H raobowMSClco
cdScotUrHcd-Hcu-Hocdcd
J*2
[3
191
-------
cu
u
•o
cu
d
I
D
,0
TO
Q
^
fj
o
H
CO
CU
CO
CO
cu
H
•a
x^
CM
0)
O /-^
d en
cd v-'
O 00
)-i co
cu O "ft
CU ft ft *H ft
d ft ftx-vx-\ ft /—. v*< x-\
o ">— ' ^-^ ft f" ctj ft in c>«
^ [5 in ft ^* ' ^j ft ^-— ^
CU •« *^ vO v^ *» CT>
|j4CMen ^-^ O CT» }-i *^^ *^^ *•**
CJN *— i CM en ?— i ON cr» t— i
-^ 0 ^
&
(U
•H cd -2
,£3 >> O d C
co cu o -H cd cd
ft co 1-1 Ai H4-I cd >
g^XM 00 S H
cdcocuajo M ^i od>i
^P^^^^i cdd vdoco
cd UQOedoOd
cucucucucu • • ^d 2 ^ a)
S5 53 S3 525 53 53 53 O O O P i
^^
cn
x,^
/-x CU
x-> CO
CU
B ^^** ^^^
Of-* en
CO **-* v-'
XXX X
/— X
ft
ft
V— /
-* CM
<~+,
ft
ft
CM
^ ^-s Q) Q) CU
ft £Xx— . ft O O ft O
ft **^ c>» x«x £23 [25 ft 52J
OO CM Cn rH t-l
CU 0)
CO CO
en dencncnen«cr cen-srcnen
C^ p, d. d C^. CiL C^ oLdClC^i^.
\o co ^ vo vo r*» CT\ co i^ i— i «d* *^
— v,CS|CM*^
i— 1 CT> O> »-l i— I vO CTi
O O
0 cd cd
cd fi c *H
iH *H cd fl) O (3 ti
09 r— ( 4J (y Cd ^J *r^ *rH
M O O CO 4J*r1bOtiQCObO
M ^ co aqqMdc
CD cd cd d) co O*H*H*HO»H
?icJ o O cj cd »^4 3 W) pC ^ o 3
o dXcdMMCo coo
jz • •»
Mc/)C/3HE-iD>>!2SSS
192
-------
•o
ft
00
O
o
o
o"
o
CO
V
CO
cu
•rl
4J
•H
H
•rl
0
Cfl
MH
rH
rH
Cfl
CO
rl
0
4-1
rj
Cfl
H
ft
CU
00
cfl
r^
CJ
cfl
cu
MH
O
M
CU
,0
g
3
C
co
cu
^
iH
00
1
'ft
ft
s-x
rl
CU
.0
B
3
CJ
4-1
O
a
o
co
4-1
C
Cfl
rH
ft
0)
00
,3
CJ
Cfl
1~l
i-H
|
CO
cu
cfl
^
CU
4-1
CU
ft
CU
•S
g
cu
4-1
CU
*"CJ
o
cu
rH
,£>
Cfl
a
3
1
c-»
S^X
^J
. 3
1
CO
**~s
a
00
•rl
CO
CU
T3
8
00
C
o
•H
4-1
Q
O
MH
C
•rl
0)
cu
CJ
cu
M
CU
co
ti
0
co
CU
0
C!
1
^1
o
CO
pi
cd
o
Pi
193
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
EPA-600/2-80-005
2.
4. TITLE AND SUBTITLE
Evaluation of Full-Scale Tertiary Wastewatf
7. AUTHOR(S)
J. A. FitzPatrick and C. L
. Swan son
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Northwestern University
Evanston, Illinois 60201
12. SPONSORING AGENCY NAME AND ADDRESS
Municipal Environmental Research Laborator
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, Ohio 45268
3. RECIPIENT'S ACCESSION- NO.
5. REPORT DATE
T Filters May 198° (Issuing Date)
6. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT NO.
10. PROGRAM ELEMENT NO.
BC611 SOS#3 Task B/05
11. CONTRACT/GRANT NO.
Grant No. R803212
13. TYPE OF REPORT AND PERIOD COVERED
y--Cin..OH Final ("11/9/71 */l*/7*')
14. SPONSCWN'G'AG'ElMCY CODE '
EPA/ 600/1 4
15. SUPPLEMENTARY NOTES
Project Officer - S. A. Hannah (513/684-7651)
is. ABSTRACTConventional methods for treatment of municipal wastewaters frequently produce
effluents that will not meet local discharge requirements. Granular media filters
are being installed to provide tertiary treatment for increased removals of suspended
solids and BOD. This report provides performance data for full-scale tertiary
filters from eight treatment plants and discusses effects of various design and
operating practices. Semi -empirical mathematical models in the report relate filter
clarification efficiency to characteristics of secondary effluent particulate matter
and to filter operating parameters. The models developed in this study may be applied!
with some caution to predict filter suspended solids removal or clarification
efficiency without pilot-scale tests.
An important design consideration for small scale tertiary wastewater filters
was found to be the ability to handle shock loads caused by secondary process
upsets. This consideration generally favors those designs with slow rate of headless
development. Clarification efficiency is only weakly dependent on filter media depth
and media grain size. Better correlations of filter clarification efficiency have
been obtained for grab compared to composite sample data. Composite process vari-
ables generally fit a log normal distribution best and normal distribution second
best.
17.
KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
Sewage Treatment*, Filtration*, Sewage
Filtration*, Waste Treatment, Water
Pollution, Clarification
18. DISTRIBUTION STATEMENT
Release to Public
b.lDENTIFIERS/OPEN ENDED TERMS
Physical -chemical
treatment, tertiary
treatment
19. SECURITY CLASS (This Report)
Unclassified
20. SECURITY CLASS (This page)
Unclassified
c. COSATI Field/Group
13B
21. NO. OF PAGES
210
22. PRICE
-•
EPA Form 2220-1 (9-73)
194
GOVERNMENT PRINTING OFFICE 1980.657-146/5653
------- |