DYNAMIC PROCESS MODELLING OF ACTIVATED

SLUDGE PLANTS

Mark Smith and Jeremy Dudley

WRc, Frankland Road, Blagrove, Swindon SNF 8YF, UK

CIWEM International Conference in “Activated Sludge into the 21st Century”

Manchester, 17 - 18 September 1997

ABSTRACT

This paper reviews the developments in process modelling, outlining the benefits

brought with the use of dynamic modelling compared to steady-state modelling. The

alternative uses of modelling, for design, operation and training are included.

The main models used today are reviewed. The models covered are the COD-based

models from the IAWQ, Models #1 and #2 and WRc’s BOD-based model. The

relevance of BOD modelling for practitioners, rather than academics, is discussed

with ideas on using each model. Likely routes for future developments of models are

presented.

The authors’ practical experience with using these models, principally under WRc’s

STOAT software, are also presented. This is in the form of case studies, bringing out

the experience of sampling programmes, calibration, validation and the continuing

importance of the engineers role.

KEYWORDS

Activated sludge; Process modelling; IAWQ COD models; WRc BOD models;

Sampling Programmes; Calibration; Validation; Case studies.

INTRODUCTION

Activated sludge modelling began in the early 1970s with the use of steady-state

models. Dynamic modelling, where the effect of time variations in flow, composition

or operation is included, began in the late 1970s. Within the UK the Jones activated
(1)
sludge model , ASMODEL, was available from WRc in the late 1970’s for

mainframes, which was funded by the Water Authorities and in the mid-1980s for
(5)
PCs, while the IAWQ activated sludge model became widely available through the
(2)
Clemson University , SSSP, around 1989. Using ASMODEL most of the UK water

authorities made use of dynamic modelling in some form and by 1987 companies

such as Severn Trent Water and Thames Water routinely used dynamic models as

part of their process design methods. The ASMODEL is now part of the WRc

STOAT package (Sewage Treatment Operation and Analysis over Time). SSSP

became a standard model within the USA, and other parts of the world, where its

free availability helped to ensure that dynamic process modelling became part of the

engineer’s routine design activity.

The IAWQ activated sludge model helped to turn modelling from a university-based

research field, where there where probably more models than there were

researchers, into a more focused field with a consensus model that was then

extended to incorporate phenomena of interest, such as phosphorus removal and

bulking. Many of these developments have been incorporated into the IAWQ
(16)
activated sludge model #2 , which is another consensus model intended to control

the proliferation of biological phosphorus removal models.

REVIEW OF EXISTING ACTIVATED SLUDGE MODELS

Most recent models have used COD to define the organic fraction of sewage. This is

2
not the only choice. Prior to about 1977 most models used BOD rather than COD,

while TOC is a possible choice that future models may adopt. Of the currently used

models only the Jones activated sludge model continues to use BOD. The choice of

COD or BOD brings various advantages and disadvantages and these are compared

below.

The comparison of different activated sludge models is commonly made mainly on

the choice of BOD or COD. There are other structural differences that can be used in

comparing models, of which one is the representation of bacterial activity. The most

common approach has been to use the Monod model, as in the IAWQ activated

sludge models. Another approach has been to extend the Monod model to include

the effect of non-viable bacteria, by adding a Michaelis-Menten term to the models.

Some of the current research models are abandoning the macroscopic approach

implied by the Monod model and adopting a more detailed representation of bacterial

behaviour by using structured models.

Process modelling has commonly focused on the aeration tank. However, the

effluent quality from activated sludge systems frequently is mainly affected by the

performance of the final settling tanks. Until recently settling tank modelling has been

the poor relation in the activated sludge process, and this has been reinforced by the
(3)
highly successful development of the Takacs model . Settling tank modelling has

become of greater interest in recent years, as the limits to the Takacs model have

been explored.

AERATION TANK MODELS

The main COD based models are the IAWQ models and the BOD based model

developed by Jones. Each model has its advantages and disadvantages and these

3
are compared below under the following headings:

• Wastewater Characterisation

• Organic Carbon

• Nitrogenous material

• Biomass Viability

The IAWQ models are COD based models and are seen by the academic world as

the choice of preference over BOD because of its better properties in conserving the

oxygen mass balance.

Wastewater Characterisation

Organic Carbon - COD or BOD?

Both COD and BOD are lumped components. Should TOC be used in future this

also would be a lumped component. These ‘lumps’ can be partitioned into different

fractions, providing a better approximation to the underlying distribution of

compounds that COD and BOD are measuring. Partitioning COD and BOD by itself

is not sufficient. The reaction models using these fractions must also reflect the

differences implied by the different fractions, principally in the rate at which

degradation of each fraction takes place. The values for the kinetic parameters may

be markedly different from the values found with tests on pure components. As an

example, the Monod half-saturation constant may be orders of magnitude higher

than ever seen with pure component tests. In the standard chemical engineering

literature it is recognised that if every true component reacts with first-order kinetics

the apparent behaviour of the lumped component will be to react with an order

greater than one. How much greater will depend on the composition of the mixture,

although upper and lower bounds on the order of reaction can be established.

4
 (4)
Vavilin has demonstrated that the Monod equation can be seen as the lumped

component representation when the true components all have zeroth-order

reactions.

In COD based models the organic material in the wastewater is split into two

fractions, biodegradable and non-biodegradable COD. Each of these fractions is

further subdivided into soluble and particulate. The soluble degradable COD is

available for direct use by bacteria while the particulate degradable COD must first

be hydrolysed into soluble degradable COD before being utilized.

One of the main problems with COD is the estimation of the biodegradable fraction

of the total COD. The technical report that describes the IAWQ activated sludge
(5)
model No. 1 recommends extensive laboratory trials to estimate the heterotrophic

yield and measure the oxygen utilization rate (OUR) which then allows the

biodegradable fraction of COD to be derived. Under normal modelling conditions,

engineers will not have the time, facilities or funds to carry out these trials and are

therefore forced to look at other methods of estimation. Alternative approaches

include

• Carrying out 20 day BOD tests and using this figure as the readily biodegradable

COD concentration. However, it has been our experience that at five NAMAS-

accredited laboratories problems occurred such as BOD5 concentrations being

greater than BOD20 concentrations and BOD20 concentrations being greater than

total COD concentrations. The differences in concentration were larger than could

be readily accounted for by the usual accuracy associated with these tests. No

satisfactory explanations have been offered for these results but one laboratory

commented that they had seen this behaviour with other samples. Problems such

as these can lead to difficulties calibrating the models and the modeller should

5
 always design sampling programmes containing BOD20 measurements with

simultaneous BOD5 measurements.

• Measuring the BOD5 concentration and apply a multiplier. If this method is to be

used then the multiplier should be estimated from laboratory trials and not default

values.

• Setting the fraction in the wastewater to that value which allows them to match the

effluent COD. While this can be seen as a calibration exercise, it should not be

encouraged as it turns a complex mechanistic model into a glorified regression

model.

In the BOD based model the organic material in the wastewater is specified in terms

of particulate and soluble BOD. As in the COD models, the soluble BOD is available

for use by the bacteria with the particulate BOD having to be hydrolysed before it can

be utilized. BOD5 is the most common form of BOD used.

The simplicity of data acquisition with BOD models therefore gives the BOD

approach a practical advantage, in that the modellers are using fractions that are

easily measurable and therefore cannot be changed to allow the model to fit the

data.

COD has an advantage in that it is possible to carry out a COD balance

Mass of COD in = COD out + biomass COD + Oxygen consumed.

The BOD test is equivalent to the respired fraction of COD, and therefore

underestimates the total amount of organic material. There is no BOD balance

equivalent to a COD balance. There are BOD models that accommodate this by

assigning a BOD to the sludge biomass produced, so that the liberation of BOD upon

6
cell death can be included in the model. Most BOD models do not include such a

feature.

Nitrogenous Material

The IAWQ models handle nitrogenous material in five fractions: soluble and

particulate, degradable and nondegradable organic nitrogen and ammoniacal

nitrogen. Each of these organic nitrogen fractions can be derived from laboratory

tests and ammoniacal nitrogen is also easily measurable. Only ammoniacal nitrogen

can be utilized directly by the biomass, so a conversion train of particulate

degradable to soluble degradable to ammoniacal nitrogen exists within the model.

(1)
The Jones BOD model allows the user to input only ammoniacal nitrogen which is

available for use by the bacteria. The onus is therefore on the modeller to estimate

the fraction of total Kjeldahl nitrogen in the wastewater that is non-biodegradable.

BOD models can be extended to include the nitrogen cycle as used by the IAWQ

model, but with current research only using COD-based models this development is

unlikely to happen.

Biomass Viability

Aside from the difference in the utilization of COD against BOD, the main difference

between the IAWQ models and the Jones model is in the handling of biomass

viability. In the IAWQ models, bacteria exist as either viable, those which utilize COD

for growth or nonviable, those that are dead and have no further function in

consumption of biomass. COD removal is therefore a growth-related process.

Biomass breakdown is modelled by decay. Decay assumes that the biomass is split

into two fractions - nondegradable particulate COD and degradable particulate COD

which can then be utilized for biomass growth. This allows the IAWQ model to

handle endogenous respiration and the effect of COD accumulation through cell

7
death when there is insufficient oxygen available to consume the death products. It

also allows the re-release of nitrogen from cell lysis to be modelled. This is known as

the ‘death-regeneration cycle’.

In the development of the Jones BOD model, account was taken of earlier
(6)
research which indicated that only a small proportion of respiring bacteria are

actually viable. Biomass therefore exists in three states. Viable bacteria utilize BOD

for growth-related processes which is modelled using the Monod equation. Nonviable

bacteria are unable to take part in growth, but continue to have a residual enzyme

activity which is lower than that of viable bacteria because the dominant means of

substrate transport through the cell to the enzyme sites is passive rather than active.

The Michaelis-Menten equation is used to model this behaviour. The third state is

dead biomass which have no biological activity. The model decouples BOD removal

from growth and was developed as a means of reconciling observations on biological

activity and counts of viable bacteria. The model does not include BOD leakage

through cell decay, using instead a term for endogenous respiration. This approach

was common until the advent of the IAWQ model.

Despite the different approaches of each model to sludge viability, in practice the

difference is less clear cut. Unpublished WRc comparisons between the two models,

using data from UK sewage works, indicated that the two models will give similar

predictions for effluent quality and oxygen demand, providing that both models have

been appropriately calibrated. The Jones model in WRc usage is rarely re-calibrated,

the defaults having been set in 1974. The IAWQ model is usually calibrated for each

application. It should be noted that this feature is not because of the use of BOD or

COD but as a result of the different approach taken to the bacterial activity.

In summary, both the COD and BOD based models have the ability to accurately

8
model the complex reactions that occur in the aeration tank of an activated sludge

plant. Each model uses a lumped component to describe the organic carbon in the

wastewater and uses Monod-type kinetics to describe the reactions to occur. The

COD models suffer from difficulties in specifying accurately the readily biodegradable

COD fractions but can model endogenous respiration whereas the BOD model it is

easier to characterise the influent wastewater but cannot model endogenous

respiration. It is our opinion that given sufficient data a modeller can accurately

model an activated sludge process in either COD or BOD.

FINAL SETTLEMENT TANK

In order to accurately model an activated sludge plant, it is necessary to model the

final settlement tanks accurately. A lot of plants fail because of the final settlement

stage of the process. There are three main approaches to modelling final settlement

tanks:

• Some of the earlier activated sludge models

(7)
ignored the effects of

sedimentation and treated settling tanks as perfect splitters. This is obviously a

limitation and would be unacceptable for accurate modelling because of the

problems experienced at a large number of activated sludge plants with achieving

good separation.

• The most common approach is to use the Takacs model

(3)
which models the

effects of thickening and clarification but assumes the internal hydraulics are

those of a perfect settlement tank. This model is based on the limiting flux theory

which was originally developed for steady-state operation, with simple heuristics

to extend the model for dynamic modelling. The basis of the limiting flux method is

that if the rate at which solids enters the sludge blanket exceeds the rate at which

9
 solids leaves then the sludge blanket must grow. Solids are partitioned into two

fractions, settleable and non-settleable and only the settleable fraction is used in
(8)
calculating the settling velocity. The model has been recently reviewed who

have demonstrated that it fails to correctly resolve rapid changes in the sludge

blanket. Despite this theoretical weakness, the Takacs model is widely used

because it does provide predictions that match with measured data.

• Another approach that has been used is to attempt to model the internal tank

hydraulics using computational fluid dynamics (CFD) combined with settlement

(9)
equations . This method is theoretically more accurate because of the ability of

the model to simulate the effects of deadspots, short-circuiting and internal flows

on the suspended solids in the effluent. The user should be aware that when

using a 2-D model, the grid must be sufficiently fine to allow accurate simulation

of the internal flow paths. This can cause the model to run slowly but with the

ongoing improvement in computing power, it is likely that the use of CFD models

with fine grid resolutions will increase.

It is our opinion that although a 2-D final settlement tank model is theoretically more

accurate, it is not currently at a stage where it can be used for routine dynamic

modelling with a sufficiently fine grid. It can be used to design a new final settlement

tank and ensure that the flow path has no obvious problems. However, with accurate

calibration, the more popular Takacs model is adequate for most users’

requirements.

FUTURE DEVELOPMENTS IN MODELLING

Growth models

The Monod equation was originally developed for modelling a pure culture, grown on

10
a single substrate, operating under balanced growth (effectively, steady-state)
(10)
conditions . The extension of the Monod model to activated sludge systems, where

the cultures are not pure, the substrate (whether measured as COD or BOD)

contains many different compounds, and where balanced growth does not apply, is a

statement of faith. It is a faith that has been well rewarded. The use of the Monod

equation would not have survived unless the results were adequate for engineering

use.

(11)
The current trends in modelling include

• Dividing biodegradable soluble and particulate COD into several fractions, to allow

the lumped parameter nature of COD to provide a closer match to the underlying

spread of organic matter. This was begun with phosphorus removal modelling,

where the behaviour of the systems is markedly affected by the presence of

volatile fatty acids separately from other biodegradable soluble COD material.

• Using more than one biomass component for the heterotrophs. The phosphorus

removal models divide the heterotrophs into two fractions, phosphorus

accumulating organisms (PAOs) and non-PAOs (which are called heterotrophs,

for want of a better name). There are also bulking models that divide the

heterotrophs into floc-formers and filament-formers, and there are some models

that would use several biomasses:

• PAOs

• G-bacteria, competing with PAOs for volatile fatty acids but not supporting

biological phosphorus removal

• Floc-formers, as ‘other heterotrophs’

• Filament formers, with growth parameters designed to make them the

abundant species under low dissolved oxygen or COD conditions

11
 • Filament formers, with growth parameters designed to make them the

abundant species under high available COD conditions

• Nitrosomonas

• Nitrobacter

• Higher species that graze on the above bacteria

Progress in activated sludge modelling is splitting into two camps. The first approach

is exemplified by the IAWQ Activated Sludge Model #2. This takes the basis of the

IAWQ Activated Sludge Model #1 and adds additional processes, with switching

functions to decide when a given process will be activated by the presence or

absence of oxygen, nitrate, phosphorus, or whatever other parameter is felt

appropriate. The result is a parameter-rich model, but where the parameters have

been added to force the model to fit available data. The switching parameters have

been added for numerical reasons and may take on any value that permits the model

to match the data. The rate equations describing the new processes (excluding the

switching functions) are either first-order or Monod equations. Some of the

developers of these models would regard the Monod part of the equation as another

switching function, so that the models all become first-order equations modified by

switching functions.

The alternative approach has been to make use of ‘structured’ models, and this
(12)
approach is exemplified in the work of Smolder’s . Instead of starting with the

Monod equation and adding switching functions until the model is forced to fit, the

approach instead is to develop a mental view of the internal bacterial processes, and

develop models to represent these processes. A common requirement for these

models is that they should simplify under steady-state conditions to the Monod

equation. These models, like the previous approach, have a large number of

parameters. But unlike the previous approach the parameters are generally justified

12
as having been introduced for a mechanistic purpose, and further that they may not

take on any values that forces the model to fit the data. Instead the parameter values

are constrained by physical limits (perhaps wide limits, but limits) behind the reason

for introducing the parameter.

The ‘switching function’ approach is fundamentally limited compared to structured

models, but in terms of pragmatic use it is likely to remain the dominant modelling

method. This is because of its strengths: computationally fast, relatively easy to

understand, relatively easy to calibrate, providing predictions that are adequate for

engineering use, and easy for most people to extend to accommodate new

situations. The more mechanistic approach of structured models make them more

robust when applied outside the calibration limits, but at present there is not much

experience with the use of these models to know how widely applicable they will be.

Settling tank models

Settling tank modelling divides into three areas

• Appropriate models for sludge settling rates, as a function of solids concentration

and possibly other sludge characterisation parameters. These models need to

accommodate the clarification behaviour of activated sludge at low

concentrations, the zone settling behaviour at high concentrations and the

compressibility and thickening behaviour near the tank floor. There have been

three such models published:

• the Takacs model , that uses one equation to cover clarification and zone
(3)

settling, and ignores compressibility effects;

• the Dupont model

(13)
, that divides the sludge into two fractions. One

fraction is assumed to behave as discrete particles and defined the

13
 clarification limits on the settling tank. The other fraction behaves as

sludge in the zone thickening regime. Compressibility effects are again

ignored.

• The Otterpohl model

(14)
. Like the Dupont model the sludge is divided into

two fractions, the first fraction behaving as discrete particles and the

second as activated sludge in the zone settling regime. This second

fraction has its settling velocity modified by a complicated function that is

intended to represent the effects of compressibility.

• Appropriate models for sludge settling, all based on the steady-state limiting flux

analysis. The flux analysis model is then modified to include the effects of tank

depth and unsteady-state conditions. Most of these modifications are heuristic,

(15)
going back to work of Vitasovic and beyond. The most common extension has

been that due to Takacs, so that the ‘Takacs model’ is commonly used to mean

both the settling velocity model and the limiting flux extensions. Recent models

have been based on an analysis of shock-wave theory that has demonstrated that

the heuristics used are not theoretically valid. Of the shock-wave models the one
(8)
that has received most exposure has been that due to Diehl . Diehl has

demonstrated that the Takacs model is numerically unstable, and that the

Otterpohl model is numerically inconsistent. The instability of the Takacs model is

reduced by modelling the settling tank with a small number of vertical layers.

Because this is common practice, with the results being adequate for engineering

use, the Takacs model is likely to remain widely used for the foreseeable future.

• Appropriate models for the hydraulic behaviour. Most of the current models use a

one-dimensional flow field, where the flow is either vertically up or down

(clarification and thickening) and therefore is a idealisation of the flow pattern. The

14
 alternative is to use two- and three-dimensional models based on computational

fluid dynamics (CFD), which models the flow field within the tank, including short-

circuiting and recirculation currents. A fine numerical mesh is required to correctly

resolve the flow behaviour; using a coarse mesh will provide numerical answers,

but may not calculate the correct flow field. The CFD approach is becoming more

common but places a large computational burden on the model. As an example, a

typical one-dimensional model will use 10 cells to represent the settling tank; a

typical two-dimensional CFD model will use 10,000 or more cells to correctly

resolve the flow field.

MODELLING PRACTICE

There are many ways to approach modelling of an activated sludge process, none of

which can be called the ‘right’ way. However, it is our experience that the modeller

should approach a new project in a systematic and defined way. The main stages in

this approach are discussed below.

Preliminary Stage

The modeller should be familiar with the plant that is to be modelled. This should

include obtaining all documented information about the tank sizes, configurations

etc. Wherever possible, all available detailed P&I diagrams should be consulted. It is

also recommended that the modeller visit the sewage treatment works and talk to the

operations staff. They will usually have information on how the plant really is

operated and not how it is supposed to operate. You will also discover any local

factors that would affect how the plant should be modelled. This may include flow

splits that are not equal, hydraulic limitations, sludge handling strategy etc. The

modeller should take photographs where required for future reference. At the end of

15
the preliminary stage, the modeller should be able to accurately construct the model

on the computer and move onto the data acquisition stage. Site visits will help put

the subsequent modelling into context, so that the abstract manipulation of process

models is related to what is, or will be, in the ground.

Data Acquisition Stage

All relevant historical flow and compositional data should be collected and analysed

for suitability of use in the modelling exercise. In some situations the historical flow

and compositional data is either inaccurate or inadequate for use in the modelling

exercise. It is our opinion that there is no substitute for a short-term intensive survey

to characterise the diurnal variation in the wastewater coupled with longer term

historical data to assess the seasonal changes occurring. Flows and compositional

data ideally need to be monitored every time they are altered in the environment

encompassed by the model. Monitoring should be carried out at the points where

they enter and leave the modelled environment.

• Flows should be recorded every 2 minutes and averaged over 15 to 120 minutes.

Ensure that the flowmeter used is suitable for the position where it is being used

and that the meters are accurately calibrated. A common cause of inaccurate

measurement is grit build-up in the flume. We are aware of a meter that was

calibrated assuming a square channel, where the true cross-section was square

sides and a semi-circular base. The result was a 50% over-estimation of the true

flow. If storm flows are being measured, ensure the flowmeter is suitable for use

in a channel that may be dry most of the time.

• Composition: Each parameter should be monitored upstream and downstream of

each modelled process. The parameters that need to be measured depend on

whether the model is to be COD or BOD based. The following table sets out the

16
 parameters that need to be measured for each model and suggestions on how to

measure them.

IAWQ Model #1 (COD) Jones Model (BOD)

Soluble BOD5 Filtered BOD5

Particulate BOD5 Total BOD5 - Filtered

BOD5

Soluble biodegradable COD Filtered BOD20 (a)

Soluble non-biodegradable COD Total BOD20 - Filtered

BOD20(b)

Particulate non-biodegradable COD Note #1 (c)

Particulate biodegradable COD Total COD - (a+b+c)

Volatile suspended solids Volatile SS

Non-volatile suspended solids Total SS - volatile SS

Ammoniacal nitrogen Ammoniacal Nitrogen Ammoniacal Nitrogen

Soluble biodegradable organic nitrogen Filtered TKN - Ammonia

Particulate biodegradable organic Total TKN - Filtered

nitrogen TKN

Soluble non-biodegradable organic set to zero

nitrogen

Particulate non-biodegradable organic set to zero

nitrogen

Note #1: Remove an aliquot of reactor contents from a completely mixed reactor treating the wastewater at an SRT

in excess of 10 days and aerate it in a batch reactor. Remove samples periodically and measure for COD until no

change occurs in the analysis.

17
We recommend a sampling period of 72 hours duration (or at least three times the
1
sewage retention time, whichever is the greater ), with 4 spot samples taken every

one or two hours. Ideally flow-averaged composites are preferred but time-averaged

are usually the ones available. The sampling regime should be carried out under dry

weather conditions (not dry weather flow conditions) with separate sampling required

for unusual events such as storm flows etc. The 72 hour duration should allow the

modeller sufficient data to derive a typical 24 hour diurnal pattern. The modeller must

also be aware of the cost implications of carrying out such detailed monitoring and

analysis but it is our opinion that it is absolutely crucial to get accurate diurnal

variations before modelling is started. It is our belief that there is no such thing as a

typical diurnal curve for a size of sewage works, as has been cited in some recent

papers. At the end of this stage the modeller should have enough data to create an

accurate influent flow and concentration profile and calibrate the model against

collected data.

Calibration stage

The modeller must now calibrate the model against the actual data collected from

the plant. Any model parameters that have been measured on site should be entered

into the model at this stage. This includes data such as KLa values, settling velocities

1
Modelling is always affected by the unknown initial conditions. The two most common

approaches to this problem are either to start the model with the steady-state results for the

sewag works, given the initial flow and composition, or to continually run the flow and

compositional data through the model until dynamic equilibrium has been attained. The

acceptability of either approach requires that the data that has been collected should not be a

relfection solely of what was in the tanks during the sampling interval. this is what sets the

minimum sampling time to be at least three times the sewage retention time.

18
etc. The model should then be run with all other parameters set to the model

defaults. It is important the model is run for at least three sludge ages or hydraulic

retention times to allow dynamic equilibrium to be achieved before calibration is

attempted, otherwise the model will be predicting start-up conditions. The model

predictions should be compared with the actual results and the modeller must decide

which are the appropriate parameters to change to achieve a better fit. If required

certain measured parameters should be remeasured before adjustment within the

model. This process can quickly become being very expensive and the modeller

must review the cost effectiveness of continuing calibration. There are many

numerical methods of comparing predicted data against actual data e.g. minimising

sum of squares, t-tests etc. which can be used but it is our opinion that a graphical

comparison, although subjective, is the best way of carrying out this process,

because of the relatively limited data available that is normally available.

Validation Stage

The modeller should now run the calibrated model against a new set of independent

data and compare the results as before. If the comparison is unacceptable, the

validation data can now be used to further calibrate the model. If this is done, a new

set of data must be used to validate the newly calibrated model. Again the modeller

must be aware of the cost effectiveness of this process.

After these stages have been followed, the modeller can be confident that there

exists an accurate model for that sewage works. There will be no further need for

calibration of the model unless significant changes occur at the treatment works.

CASE STUDIES

The following three case studies demonstrate the advantages of using dynamic

19
modelling as an integral part of different applications. All the case studies listed here

were modelled using WRc STOAT.

Bellozanne STW, Jersey.

This project involved a review of the process strategies to uprate Bellozanne STW to

treat additional flow. The sewage works at Bellozanne includes a screening plant, a

grit/grease separator, primary treatment and a diffused air activated sludge plant.

This is followed by final settlement tanks and a UV disinfection plant. The WRc

STOAT model was constructed and calibrated using data collected from an on-site

survey. Automatic samplers were installed to collect samples of crude sewage,

settled sewage and activated sludge plant effluent as well as spot samples of the

supernatant liquors returned to the primary sedimentation tanks from the digested

sludge thickening tank. Each sample was analysed for BOD, COD, suspended

solids, total Kjeldahl nitrogen, ammoniacal nitrogen, total oxidised nitrogen and total

phosphorus. Two 24-hour periods of intensive sampling took place and the data from

one survey was used to calibrate the works and from the second survey to validate

the model. Figure 1 shows the calibration curves achieved.

20
Figure 1 - Calibration curves for Bellozanne STW

The curves show a good fit with respect to all three parameters but especially with

respect to ammonia.

The calibrated model was then used to assess the effect of higher flows through the

plant with following results.

• The maximum flow rate of 1200 l/s could be treated by the existing primary tanks

but caused hydraulic overloading of the activated sludge settlement tanks. The

maximum throughput of the activated sludge plant was determined by the

maximum effluent suspended solids which permit adequate disinfection in the

downstream UV plant.

• A new biological treatment plant would be required to operate in parallel to the

activated sludge plant. Some preliminary sizing of a BAF plant was also carried

out.

• Retrofitting anoxic zones into the existing activated sludge plant would

21
 enable better nitrogen removal to be achieved.

This study demonstrates the importance of having valid up-to-date analytical data

available to allow accurate calibration of the computer model and thereby increasing

the client’s confidence in the results.

Daldowie STW, Glasgow.

Daldowie STW is a large activated sludge plant on the East of Glasgow operated by
3
West of Scotland Water (WOSW) and treats a DWF of 85,512 m /d. It consists of a

primary sedimentation plant followed by a surface aeration activated sludge plant.

The Scottish Environmental Protection Agency (SEPA) indicated that the consent

would be changed which would mean Daldowie needed to produce a fully nitrified

effluent while receiving higher flows and loads because of other works being closed.

WOSW, WRc and SEPA carried out a prolonged period of experimentation at the

plant which provided data for a detailed computer model to be constructed which

was then used to predict future performance under extra flow and load conditions.

The modelling of the plant was aided by the large amount of data available from the

experimental programme which was used to calibrate and validate the works. Many

simulations were carried out to assess the future ability of the plant to achieve its

nitrification consent. Figure 2 shows a STOAT output which demonstrates the effect

of a storm event on the plant’s effluent with the existing plant. Typically we would

model two or three storm events of different durations, and we would require that the

efffluent quality at all times meet the 95%-ile values for the effluent consent. (When

we are also working with catchment models we have occasionally used a year’s

worth of predicted sewage, with storm events, and then we have allowed individual

storms to exceed the 95%-ile values, providing that the entire year’s predictions were

within the 95%-ile. This was not the case at Daldowie, which therefore used

22
representative storms.)

Figure 2 - A simulated storm event at Daldowie

As can be seen from the summary statistics in this figure the plant would fail its

consent for all three main parameters. Further simulations were carried out to assess

the effect of the same storm with extra aeration capacity and extra final settlement

tank capacity with the results shown in Figure 3.

23
Figure 3 - Modified plant receiving a storm event

Now it can be seen that the modified plant would be able to accept the storm load

without contravening the consent.

This case study showed the ability of modelling to expand the boundaries of design

parameters and assess the effect of shock loads on the plant. The project concluded

that extra aeration and final settlement capacity would be needed to allow Daldowie

STW to accept and treat all the flow that is due to pass to the works in the near

future.

Phosphorus Removal System

This sewage works used chemical phosphorus removal ahead of the primary

settlement tank, with polishing of the final effluent through the normal phosphorus

uptake of biomass. Unlike the two examples given above here we used a

24
COD-based model. We began with preliminary modelling looking at COD and solids

removal. The COD-based models were chosen because the data was provided as

COD rather than BOD, and choosing an appropriate model to match the data

minimises the number of conversions that would be required. The results from the

calibration work are shown in Figure 4 for COD. Data was available as daily values,

taken during the working week. Missing values were interpolated. This accounts for

some of the discrepancies between the data and the predictions, as the predictions

are made based on imperfect knowledge about the full history of the sewage

entering the works.

100

90

80

70

60

50

40

30

20

10

0
1 29 57 85 113 141 169 197 225

Figure 4 - COD Effluent Predictions

The IAWQ Activated Sludge Model #2 was then used to model the phosphorus.

Because no data about the chemical dosing was available at the start of the

modelling we developed a simple ‘black box’ model to correlate phosphorus in and

out of the primary tank. The results from this correlation are shown in Figure 5, which

demonstrates that a reasonable match was obtained. Chemical processes can often

be reasonably correlated by simple steady-state equations, because of the speed of

25
the reactions.

0
01/05/96 29/05/96 26/06/96 24/07/96 21/08/96 18/09/96 16/10/96 13/11/96 11/12/96

Figure 5 - Correlation Results for Settled Sewage Phosphorus

The settled sewage was then fed to the activated sludge tanks. The aeration tanks

are not configured to promote biological phosphorus removal but the effluent clearly

demonstrates that there is such a removal. This removal is associated with the

normal phosphorus requirements for biological growth, and this behaviour is included

in the IAWQ Activated Sludge Model #2. The results using the default parameter

values for that model are shown in Figure 6, which demonstrates that both the

magnitude and dynamics of the removal have been matched.

26
 1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
02/05/96 30/05/96 27/06/96 25/07/96 22/08/96 19/09/96 17/10/96 14/11/96 12/12/96

Figure 6 - Model Predictions for Final Effluent Phosphorus

The standard activated sludge models, such as the IAWQ Activated Sludge

Model #1, do not include phosphorus removal as a normal growth-related process

and therefore predict that the effluent quality is unchanged from the settled sewage.

This demonstrates that an appropriate model must be chosen for the modelling

requirements. Because of the large number of parameters available in the Model #2,

required for the biological phosphorus removal sub-model, it would not be

appropriate to use this model where phosphorus removal was not a required part of

the study.

This modelling work is still in progress. On completion of calibration the model will be

used to investigate the limits to treatment capacity, and to assess the effect of

operational changes on the effluent phosphorus.

Conclusions

1. Computer modelling of the activated sludge process is a very important part of

both the design of these facilities and the optimisation of existing plants. It will

27
 become even more popular in the immediate future with most plants having a

model.

2. The most important part of modelling is the data collection which will allow the

model to be accurately calibrated and validated for all conditions.

3. Given sufficient data, a modeller can accurately model the activated sludge

process in either COD or BOD.

4. Future developments in models will embrace the more mechanistic approach of

structured models rather than the ‘switching function’ approach. However, the

latter approach will remain popular due to the fact that it is relatively easy to

understand and calibrate the models and it provides predictions that are adequate

for engineering use.

5. Computational fluid dynamics will be used to more accurately model the flow

behaviour of the final settlement tanks, which are often the cause of failure of an

activated sludge plant.

REFERENCES

1. Jones, GL, 1978, “A mathematical model for bacterial growth and substrate

utilisation in the activated sludge process”, in A James (ed) Mathematical Models in

Water Pollution Control, Wiley

2. Bidstrup, SM and CPL Grady, 1988, “SSSP - Simulation of single-sludge

processes”, Journal of the Water Pollution Control federation 60(3), 351-361

3. Takacs, I, GG Patry and D Nolasco, 1991, “A dynamic model of the clarification-

thickening process”, Water Research 25(10)

28
4. Vavilin, VA and VS Vasil’ev, 1981, “Comparative evaluation of mathematical

models used for calculating aeration tanks”, Water Resources 8(4) 420-431

5. Henze, M, CPL Grady, W Gujer, GvR Marais and T Matsuo, 1987, Activated

Sludge Model No. 1, IAWQ

6. Wooldridge, WR and AFB Standfast, 1936, “The role of enzymes in activated

sludge and sewage oxidation”, Biochemical Journal 30(9) 1542-1543

7. Gujer, W and M Henze, 1991, “Activated sludge modelling and simulation”, Water

Science and Technology 23(4/6)

8. Jeppsson, U and S Diehl, 1995, “Validation of a robust dynamic model of

continuous sedimentation”, Med Fac Landbouw Univ Gent 60/4b

9. McCorquodale JA, EM Yuen, Z Vitasovic and R Samstag, 1991, “Numerical

simulation of unsteady conditions in clarifiers”, Water Pollution Research Journal of

Canada 26(2) 201-222

10. Ip SY, JS Bridger and NF Mills, 1983, “An engineering approach to a completely

mixed activated sludge system”, Australian Water and Wastewater Association

Tenth Federal Convention, Sydney , 13.1-13.14

11. Orhon, D and N Artan, 1994, Modelling of Activated Sludge, Technomic Press

12. Smolders, GFJ, J van der Meij, MCM van Loosdrecht and JJ Heijnen, 1995, “A

structured metabolic model for anaerobic and aerobic stoichiometry and kinetics of

the biological phosphorus removal process”, Biotechnology and Bioengineering 47,

277-287

13. Dupont, R and M Henze, 1992, “Modelling of the secondary clarifier combined

with the activated sludge model No. 1”, Water Science and Technology 25(6) 285-

29
300

14. Otterpohl, R and M Freund, 1992, “Dynamic models for clarifiers of activated

sludge plants with dry and wet weather flows”, Water Science and Technology

26(5/6) 1391-400

15. Vitasovic, ZZ, 1990, “Clarifier modelling”, in GG Patry and D Chapman (eds)

Dynamic Modelling and Expert Systems in Wastewater Treatment Engineering,

Lewis Publishers

16. Henze, M, W Gujer, T Mino, T Matsuo, MC Wentzel and GvR Marais, 1995,

Activated Sludge Model No. 2, IAWQ

ACKNOWLEDGMENTS

The authors would like to thank West of Scotland Water and the Public Services

Department, Jersey, for permission to use their data. The authors would also like to

thank WRc plc for permission to publish this paper. The opinions expressed within

this paper are those of the authors and not necessarily those of WRc plc, West of

Scotland Water or the Public Services Department, Jersey.

30