Dynamic simulation of chemical industry wastewater

Water Science and Technology Vol 45 No 4–5 pp 355–363 © 2002 IWA Publishing and the authors
treatment plants
S.J. Bury*, C.K. Groot**, C. Huth** and N. Hardt***
* Engineering Sciences, The Dow Chemical Company, Freeport, Texas, 77541 USA.
(E-mail: sbury@dow.com).
** Dow Benelux, Hertbert H. Dowweg 5 – 4542 NM Hoek Postbus 48 – 4530 AA Terneuzen The
Netherlands
*** Dow Deutschland GmbH, Dow, Werk Stade Postfach 11 20 21677 Stade, Germany

Abstract High variability, stringent effluent permits, and often extreme operating conditions define the
practice of wastewater treatment in the chemical industry. This paper reviews the benefits and challenges of
applying dynamic simulation to chemical-industry wastewater treatment plants by describing case studies at
full-scale wastewater treatment plants (WWTP). The applications range from process troubleshooting to
optimization and control. The applications have been valuable and useful in developing a deeper
understanding of the plants as integrated systems. However there still remains substantial work to
implement the dynamic simulations for daily real-time use by plant engineers and operators. This opportunity
to improve plant operations is still largely untapped and will remain so until dynamic state estimation and data
reconciliation are incorporated into simulation packages for use in developing the on-line simulations.
Keywords Dynamic simulation; industrial activated sludge; modelling; on-line simulation; optimization

Introduction
During the past decade the dynamic simulation of activated sludge plants has grown in
popularity and scope of practical application. Advances in computational power and the
publication of unifying biokinetic frameworks like the ASM1, ASM2 and ASM3 have
driven this increase in use. Over the past four years we have applied dynamic simulation to
Dow Chemical’s eight major production sites, a total of twelve different treatment systems.
The applications have ranged from process troubleshooting and process improvement, with
forays into process control and process optimization, to process design and process
economics. The road to success has been at times rocky but overall the journey has been
worthwhile. This paper presents case studies that highlight some of the ICA conference
themes, namely the use of simulation for control and operability.

Industrial activated sludge treatment

High variability, stringent effluent permits, and compared to municipal wastewater
treatment extreme operating conditions define the practice of wastewater treatment in the
chemical industry. Two strong constraints delineate the successful operating envelope of
the industrial wastewater treatment plant. The first is the protection of the environment by
complete effluent permit compliance and the second is to provide 100% reliability and
availability for the sender plants. As environmental protection always takes priority over
production, the robust operation of the central WWTP is a must for success in a competitive
industry. Environmental protection and permit compliance require the consistent
production of high quality effluent from the wastewater plant even in the face of rapidly
changing influent. Reliability and availability mean that the WWTP operation must
adapt rapidly to the inherent variability. The variability is manifested in both flows and
composition. Flow rate fluctuations of more than 50% due to production changes are 355

Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf

by UNIVERSITY OF EDINBURGH user
on 20 December 2018
 not uncommon, but unlike diurnal loads, do not follow a regular pattern. Rainfall runoff
can triple the influent flow at some plants. The influent composition of organics, nitrogen
and trace nutrients fluctuates as does the alkalinity, pH, temperature and dissolved salts.
A challenging control problem at a number of our plants is the addition of ammonia (NH3)
and phosphate as nutrients to ensure proper chemical oxygen demand (COD) removal
while meeting the effluent permit levels for these nutrients. Two plants have excess
influent nitrogen as both NH3 and organic nitrogen. One was designed for
S.J. Bury et al.

nitrification–denitrification while the second was not. Fortunately the second plant also has
a nitrogen-poor train. The solution is to use the effluent from the nitrogen-rich train to
supplement the nitrogen-poor train. Simulations provide insight into the dynamics and
control issues for this integrated system.
The bulk COD removal is typically very good and not a major operational issue. The
amount of particulate substrate in the feed is practically zero. The major organics are small
molecules like volatile fatty acids and glycols. Many individual compounds are monitored
at the effluent and require maintenance of specific plant operating conditions to ensure
adequate treatment. Understanding the dynamics of trace organic removal is an important
aspect of plant operation and we have used dynamic simulation to provide guidance for this
control issue. The compositional variability is also a modelling challenge. In short time
spans (hours to days), ratios and other model parameters can change. Compounding this are
the uncertainties in the lab analyses and on-line measurements like flows, dissolved oxygen
(DO), NH3 etc.
Other variable features of our wastewater treatment are the differences in the reactor
configurations (none of the eight plants are the same), the use of high purity oxygen and air
systems, the high operating temperatures (greater than 35°C) and high dissolved salts at
some of the plants (up to 70 kg/m3). These conditions often require the use of alternative
analytical techniques like total organic carbon (TOC) as a surrogate for COD, which is hin-
dered by chloride interference. The organic variability results in a variable influent TOC to
COD ratio which then becomes a parameter that needs to be estimated. Of course our plants
are also subject to the wide list of challenges list by Olsson (1998).

General use of dynamic simulations

The value of dynamic process simulation is often hard to capture. We have generally used
the following broad categories to describe the value of the modeling exercise.
• Consistent descriptions of the individual plants. The simulations provide material and
energy balances, benchmarking and capacity estimates.
• Operator and engineer training. Simulations are effective tools to demonstrate how the
plant should operate as well as demonstrate the application of operating discipline prin-
ciples. Operators and engineers gain a deeper understanding of the plant as an integrated
system rather than just as separate unit operations. It is also possible to quantify or reject
“Rules of Thumb”.
• Improved R&D efforts. The use of the same model framework allows for more effective
lab and pilot experimentation. Feedback from the success and failure of the simulations
demonstrates where we need more understanding of the process. The use of ASM based
models also emphasizes the use of respirometric measurements.
• Risk free experimentation. The use of dynamic simulations allows for the design and
testing of new control schemes and operating conditions. We can “what if” everything.
• Optimization and planning. The use of simulations provides insights into ways of
maximizing capacity, quantifies process bottlenecks. It also provides a way of com-
paring different expansion scenarios.
356 The above themes are complementary to the experience of others in applying ASM

Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf

by UNIVERSITY OF EDINBURGH user
on 20 December 2018
 based simulations to a full-scale municipal WWTP (Brdjanovic 2000). The following table
(Table 1) shows some of the specific applications of dynamic simulations. The evolution
has been from process troubleshooting to use for control and optimization.

Specific use of dynamic simulations

Successful large-scale simulation is a combination of factors; good biokinetic models,
good hydraulic, reactor and clarifier models, good control models, good ancillary models,

S.J. Bury et al.

data analysis tools, good numerics and customization. The leading wastewater simulation
packages available today combine these factors (complete with the latest GUI) and allow
the user to successfully use simulation in a wide range of applications. Olsson and Newell
(1999) review this in more detail. However it is incumbent on the simulation developer to
understand the structures and assumptions of the underlying models. The following
paragraphs discuss some of our model structures.
Despite the previously mentioned features of our wastewater, the application of ASM1
as a base kinetic framework has been remarkably successful. At the risk of being
Procrustean and contrary to some industrial practitioners we believe that the ASM1 is an
excellent framework for modelling industrial wastewater treatment plants as ASM1’s main
phenomena, oxygen consumption sludge production, nitrification and denitrification
(Gujer et al., 1999) are also major issues for chemical industry WWTPs. It is reported that
ASM1 has a number of defects (Gujer et al., 1999). We have addressed some of these with
extensions, but the essential emphasis on the electron balance has been preserved. This pro-
vides the well established robustness of the framework. Also by keeping the electron bal-
ance, we can effectively exploit direct measurements like respirometry for bulk and
individual kinetic determinations and on-line oxygen transfer measurements in the full-
scale plants. The most significant extensions are the use of a two-step nitrification model,
the temperature dependence of kinetic parameters, aerobic denitrification, preservation of
ammonification, accounting for nitrogen and alkalinity limitations, use of a state variable
for particulate inorganics and the use of a set of extended state variables for individual com-
ponents. These extensions are implemented in the CN2IP process library of GPS-X™ from
Hydromantis, Inc.. Incorporation of toxicity and inhibition kinetics is also a deviation away
from the original ASM1. The selection and use of these extensions is based on our specific
experience with the plants. The use of a two-step nitrification model introduces two more
state variables, Snoi and Xbai, the nitrite and the NH3 oxidizing autotrophs. This is
required to simulate operating conditions that result in nitrite accumulation. Figure 1 shows
process data from one of our plants during an NH3 and organic nitrogen overload with the

Table 1 Examples of specific applications of dynamic WWTP simulations within Dow

Task Example Treatment goal Plant configuration

Process trouble- Nitrification failure BNR 2-stage CSTR*

shooting Solids overload COD removal 1 stage, multiple CSTR
Denitrification COD removal 2-stage CSTR
Capacity rating Nitrification BNR 2-stage CSTR
Train separation COD and BNR Carousel
Design New high volume COD WWTP COD removal Extended basins with
virtual mixed cells
Economics Selection of best sludge COD removal 2 train, multiple CSTR plant
destruction technology
Control and Thickener operation COD and BNR Carousel
optimization Response to sender plant COD removal UNOX™
shutdown and restart
Sensor location BNR 2-stage CSTR
* CSTR – continuous stirred reactor 357

Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf

by UNIVERSITY OF EDINBURGH user
on 20 December 2018
 50

1st Stage NO2

40 1st Stage Predicted NO2
2nd Stage NO2
Oxidized Nitrogen gN/m3

2nd Stage Predicted NO2

2nd Stage NO3
30
S.J. Bury et al.

2nd Stage Predicted NO3

20

10

0 10 20 30 40
Days
Figure 1 This data is from a two-stage CSTR system (total volume = 40,000m3) that was designed for COD
removal only. The accumulation of the nitrite in the system is clear. This can only be effectively modeled using
a two step nitrification process with two autotrophic biomass fractions. There was no appreciable amount of
nitrite or nitrate in the influent. The solid lines are simulation result discussed in case study 3 in the text

subsequent very high levels of nitrite and nitrate in the system. The presence of nitrogen
containing organics in our influent is the driver behind keeping the ammonification process
as the biomass is nearly 100% efficient in freeing the amine groups. At the above men-
tioned plant, NH3 typically represents only 25–45% of the nitrogen load on the plant; the
rest is from organic nitrogen. Although these particular organics are only a small fraction of
the total COD load, the use of a constant COD:N ratio is not appropriate due to the daily
variations in their concentration. This plant is discussed in case study 3. At this time we
have not investigated internal storage polymers although this would be a structural change
to our current implementation of ASM1 rather the adoption of ASM3.
Other keys to successful simulations include having good hydraulic models for the
reactors and clarifiers including the use of residence time distributions (RTD) to quantify
non-ideality. The benefits of advanced hydraulic modeling have been discussed by Lee et
al. (1999). In conjunction with Hydromantis we have extended the Takács layered clarifier
model (Takács et al., 1991) to handle deep cones and the circular wedge clarifiers found on
Hoechst BIOHOCH® reactors. The appreciable use of settling and pressing aids also needs
to be incorporated into the simulation. The incorporation of ancillary equipment and
processes, in particular solids handling, can often significantly change the quality of the
simulation. The need to model inorganic reactions is determined on a plant specific basis.
Many reactions including dissolution and precipitation can effect the solids material
balances. The addition of more detail may lead to slow simulations times and the intro-
duction of more uncertainty due to incomplete or poor measurements.

Simulation strategy
General approaches to calibration and validation of activated sludge models and
simulations have been reported (Koch et al., 2000, Brdjanovic et al., 2000, Yuan et al.,
358 1993). The value of respirometric tests is widely known (Brouwer et al., 1998, Spanjers et

Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf

by UNIVERSITY OF EDINBURGH user
on 20 December 2018
 al., 1998, Vanrolleghem and Verstraete, 1992). The evolution of our simulation strategy
has followed our increasingly sophisticated uses of the simulations and incorporates many
of the ideas found in the above works. The basic strategy is as follows; 1) formulate a base
case simulation, 2) collection and analysis of historical operating data, 3) tuning of con-
trollers and control schemes, 4) wastewater and biomass characterization, 5) first pass cali-
bration, 6) collection of missing or incomplete data, 7) final calibration and validation.
These are not done in sequence but usually in parallel and quite often with iteration.

S.J. Bury et al.

The first three steps are a significant amount of front-end loading in development of the
simulation, but done well they provide valuable insight that saves time in the long run. Even
for emergency troubleshooting these steps should be done to guard against errors intro-
duced from hastily put together simulations. The formulation of the base case simulation
involves developing the flowsheet description from the process and instrument diagrams
(PIDS) with adjustments made for equipment and piping not reflected in the PIDS but pres-
ent in the plant. The major control schemes and operating discipline are also translated into
the simulation. The collection and analysis of historical operating data can be quite time
consuming depending on the level of data collected and how it is stored. We have identified
the lack of automation in the conversion of plant data into a form accessible by the simula-
tor as a key barrier to the real-time implementation. The statistical analysis (traditional and
multivariate), gross error identification and data reconciliation done in this step will reveal
useful information about the general behavior of the plant, missing and poor data and sys-
tematic biases. This data conditioning also aids in the development of the simulation.
Techniques for these types of analyses are available elsewhere (Romagnoli and Sáchez,
2000, Olsson and Newell, 1999). Readers interested in multivariate techniques as applied
to processes are referred to Chiang et al. (2001). Our plants all operate under automatic
control. To ensure fidelity, it is important that the captured major controllers and schemes
are tuned properly, provided of course that the plant controllers are also tuned and not set in
manual. Aeration and DO control are often difficult to simulate correctly in the beginning.
One trick is to use the plant DO measurement as the setpoint of a tuned aeration controller.
This removes the confounding effects of different DO levels compared to the plant on the
reaction rates and simplifies the later calibration. It can also reveal DO and aeration meas-
urement issues.
Respirometry and supporting analytical techniques are used to characterize the waste-
water and the biomass. Some of the required data are: COD/TOC ratio, dominant organics,
the nitrogen profile, growth rates, saturation constants, yields, nitrification rates,
heterotrophic decay rates and the nitrogen content of the biomass. Examples of the last
three are 1.6 gN m–3h–1, bh = 0.13 d–1 and a range of 0.03–0.089 gN/gCOD, well outside the
typical default values. We also characterize the variation in this data.
An iterative process that includes parameter estimation; analysis of plant-model
mismatches; further experimentation; and adjustments in the model structure as required is
used to carryout the final three steps. The ready availability of fast computers and numeri-
cal routines to estimate parameters makes it tempting to estimate many model parameters.
Overfitting is a well known danger. This can be avoided by adhering to defined strategy of
simulation development that provides a good model structure and good quality data.

Case studies
Nitrification Failure
The first case study is a process control and operability study of an overloaded nitrogen
removal plant treating the water from a single source. This study started out as a
troubleshooting exercise. After a successful 3 month commissioning using a synthetic
influent, nitrification failed after two months of running on real influent. The plant was 359

Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf

by UNIVERSITY OF EDINBURGH user
on 20 December 2018
 designed to remove between 18 and 30 kgN/day with denitrification using a two stage
CSTR (720 m3) system with acetic acid as an external carbon source. Toxicity was the
suspected cause. Extensive operational data was available from the commissioning period
as well as the plant operation. Characterization of the wastewater beyond the minimum of
COD, NH3, nitrite and nitrate was not available nor was the biomass characterized as
described earlier. The simulation was calibrated using a portion of the synthetic feed period
followed by a validation check using the remaining data from the synthetic period. The
S.J. Bury et al.

dynamic simulation showed that the operational issues were the main culprits in the loss of
nitrification. During the commissioning period the COD:N ratio was about 9. However
during the real operation the COD:N ratio often exceed 40. The simulation indicated a
significant loss of nitrifying biomass occurred prior to a spike in the influent NH3. The low
nitrifying biomass prevented the system from handling the spike load. The simulation
results for the nitrifying second tank during the spike are shown in Figure 2. The predictions
are quite good for the NH3 and nitrate, but the predicted nitrite level was low. We did not
attempt to improve this prediction. We used a two-step nitrification kinetic model in this
simulation. The simulation was then used to develop a control scheme based on the incom-
ing NH3 and COD loads to adjust the feed rates of supplemental NH3 and acetic acid
addition to maintain a stable nitrifying population during outages at the sender plant. A
second simulation was developed to investigate the feasibility of bypassing this nitrifying
plant and sending the feed directly to another WWTP used for COD removal. This simula-
tion incorporated a dynamic enthalpy balance (Sedory, 1995) to allow the operability
testing of the proposed system in case of the loss of a major heat source in winter. The risk
was assessed using Monte Carlo techniques to capture the sender plant reliability and the
expected reliability of the WWTP. This revealed that there was a significant operational
risk due to the rapid transition from the normal warm system at 35°C to a cold system below
15°C during the loss of the heat source. The proposed solution was to utilize a cooling pond
for the hot stream and operate the WWTP at a cooler temperature at the onset of winter.

500

NH3 (Data)
NH3 (simulation)
400 –
NO2 (Data)
–
NO2 (simulation)
–
NO3 (Data)
–
300 NO3 (simulation)
gN/m3

200

100

2/11/98 2/16/98 2/21/98 2/26/98 3/3/98 3/8/98

date
Figure 2 Data and simulation of the nitrogen spike described in the text. The system had not seen adequate
360 NH3 prior to the spike to maintain an adequate nitrifying population

Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf

by UNIVERSITY OF EDINBURGH user
on 20 December 2018
 Train segregation and thickener operation
The second case study is the use of dynamic simulation to evaluate the issues associated
with the segregation of two treatment trains of a high internal recycle carousel into a sweet
water train and a saltwater train. At this same site, the simulation was used to optimize the
operation of the sludge thickener. The model development here was done in cooperation
with the consulting firm Vertis in The Netherlands. This model development more fully uti-
lized wastewater and biomass characterization via respirometry. Particularly interesting

S.J. Bury et al.

were the slow decay rate and the low nitrification rate mentioned earlier as well as the
development of kinetic expression for the bio-oxidation of thiosulfate from the respirome-
try data. The low nitrification rate is of concern because, if split, one train will be nitrogen
rich and require an active nitrifying biomass. The low decay rate was also unexpected based
on our experience at other plants. Calibration of this model has been hindered by poor pre-
dictions of dissolved oxygen in the carousel. The carousel was modeled using 10 CSTRS;
however, DO measurement showed that substantial vertical gradients existed in addition to
the one along the length of the reactor. Subsequent fluid velocity measurements confirmed
the non-ideal flow pattern in the basin. We have plans to revisit this simulation with a
coupled CFD model to improve the simulation of reactor. The optimization of the biosolids
thickener was more successful. The objective here was to maintain a high concentration of
suspended solids in the thickener underflow being sent to the dewatering unit. Using the
simulation, an operating and control discipline was developed that maintained the solids
concentration at nearly 4% using a series of 4 or 5 batch filling and pump-outs per shift.
This was a significant change from the nearly continuous operation previously used
that often resulted in clear water going to the dewatering unit. In addition to savings in the
dewatering costs, it also reduced labor cost by allowing for a reduction of shifts over
the weekend.

Unexpected denitrification
The third case study is the application of dynamic simulations at our largest WWTP. This
plant consists of two trains. Train one treats a high volume (160,000 m3/day) low COD
stream in four parallel CSTRS (total vol. 54,000 m3). This stream contains neither nitrogen
nor phosphate. Train two treats a stream that is low volume (16,000 m3/day) but high COD
strength. This stream also has significant NH3 and organic nitrogen compounds. It is
configured with a primary reactor and two secondary reactors of the same type as in train
one. Both trains treat water with a salinity that varies between 40 and 70 kg/m3 as NaCl.
The excess sludge from the trains is treated in a autothermal aerobic digester (ATAD). The
effluent from this digester is sent back to train two.
The initial simulation of train two was done in response to floating sludge in the
clarifiers. It had all the symptoms of a denitrifying clarifier; however, we believed that
nitrification would not occur in a high salt chemical industry WWTP. The first simulations
also indicated that nitrification would not be supported. The reactor data indicated other-
wise, showing high levels of nitrite and nitrate. Further investigation with the simulation
predicted incomplete COD removal in the primary reactor due to nitrogen limitation but
this did not match the plant data either. The simulation failure was due to incomplete
characterization of the nitrogen in the influent stream. The lab measurements were only for
NH3 and did not have the organic nitrogen. Incorporation of the total Kjeldahl nitrogen
(TKN) into the simulation completely changed the results. The COD removal in the
primary reactor matched the plant data and nitrification was predicted as well as the
subsequent denitrification in the clarifier. Predictions of the oxidized nitrogen are shown in
Figure 1. The dynamics of the nitrate in the second stage were not reflected very well by the
simulation and may have been due to inhibition that was not modeled. The nitrification and 361

Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf

by UNIVERSITY OF EDINBURGH user
on 20 December 2018
 denitrification is controlled via dissolved oxygen levels in response to the incoming
COD:N ratio. This improved clarifier performance. At the time of this study the ATAD unit
was not operational but its subsequent operation sends more nitrogen back to train two. The
excess ammonia is used in train one in place of purchased ammonia. Simulations were used
to evaluate the operability of the combined system.

Load swings
S.J. Bury et al.

The fourth case study involves using simulation to improve the operation of an UNOX™
train (total volume 3,180 m3) that is alternatively overloaded and underloaded with COD.
This plant treats the wastewater from a single source and is subject to the production swings
experienced by the sender plant. During heavy loads the first basin can become anoxic and
during the underloaded periods excess NH3 is released in the last basin due to decay. This is
compounded by the need to add nitrogen and phosphate at this plant also. The development
of this simulation required a extensive amount of data reconciliation to correct for errors in
the material balance. The major errors involved the measurements around the clarifier. The
results of the simulation were used to size an upgrade for the mixing and aeration system
that went away from the traditional splash mixer. Computational fluid dynamics and scale
mixing studies showed that the traditional four-blade turbine did not work well due large
support beams present in the basins. The simulations were also used to test various opera-
tional changes like split recycles and split feeds to try and minimize the amount of NH3
released in the last basin. A split-feeding scheme was adopted to take the load off the first
basin. Operations now emphasize maintaining a nitrifying population to act as effluent
polishers.

Trace compounds
The fifth case study used simulations to examine the biodegradation of low level organics
(<500 mg/m3) in large-scale systems. Determination of kinetic parameters from field data
is difficult. However by combining very sensitive respirometry tests, lab-scale reactor runs
and simulations of these it is possible to get reasonable kinetic data that allow for good
prediction of full-scale plants. The kinetic models use a specific biomass for each
component of interest analogous to having heterotrophs and autotrophs. We have found
that this greatly improves the stability and predictions of the simulations. Simulations with
this structure also accurately reflect our field experience in maintaining these low level
biomass populations through proper reactor configuration and operating conditions such as
pH and DO. We also have used this structure for inhibitory compounds like phenols that are
present at higher levels with good success.

Real-time application
The applications of dynamic WWTP simulations have been valuable and useful in develop-
ing a deeper understanding the plants as integrated systems. However there still remains
substantial work to implement the dynamic simulations for daily real-time use by plant
engineers and operators. Even with the latest GUI interfaces the use of dynamic simulators
is still largely an art practiced by a few and not easily translated into a robust and reliable
application. We do not see the application as closed-loop but rather as an open-loop
advisory system. To achieve this application the installed simulation must auto-calibrate
and auto-validate based on the recent and current operating conditions and data from the
WWTP otherwise the simulation will fall out of use or may not even make it to the control
room. This self-maintaining simulation has yet to be developed. This opportunity to
improve plant operations will remain unrealized until dynamic simulations move into daily
362 use.

Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf

by UNIVERSITY OF EDINBURGH user
on 20 December 2018
 Conclusions
The five case studies have shown the broad range of application of dynamic simulation for
chemical industry activated sludge using the ASM1 biokinetic framework. It is clear that
the practical applications of dynamic simulations require substantial amount of effort to
achieve valid and sustainable predictions of large-scale WWTP behavior. This effort goes
beyond the calibration of a few biokinetic parameters and includes selection of the correct
hydraulic and supporting models, data reconciliation and state estimation as well as model

S.J. Bury et al.

maintenance. These provide a significant barrier to the daily and routine use of dynamic
models.

Acknowledgements
SJB would like to thank Mike Stenstrom, Imre Takács and Peter Vanrolleghem for their
candid discussions about WWTP modeling and simulation that have helped shape many of
the author’s views the subject. CKG and SJB would like to thank Henri Spanjers for his
assistance and expertise in performing the respirometry work for the carousel plant.

References
Brdjanovic, D., Van Loosdrecht, M.C.M, Versteeg, P., Hooijmans, C.M.,Alaerts, G.J. and Heijen, J.J.
(2000). Modeling COD, N and P removal in a full-scale WWTP Haarlem Waarderpolder. Wat. Res.
34(3), 846–858.
Brouwer, H., Klapwijk, A. and Keesman, K.J. (1998). Identification of Activated Sludge and Wasterwater
characteristics using respirometric batch-experiments. Wat. Res. 32(4), 1240–1254.
Chiang, L.H., Russell, E.L. and Braatz, R.D. (2001). Fault Detection and Diagnosis in Industrial Systems.
Springer-Verlag, London.
Gujer, W., Henze, M., Mino, T. and Loosdrecht. (1999). Activated Sludge Model No. 3. Wat. Sci.Tech.
39(1), 183–193.
Koch, G. Kühni, Gujer, W. and Siegrist, H. (2000). Calibration and Validation of Activated Sludge Model
No. 3 for Swiss Municipal Wastewater. Wat. Res. 34(14), 3580–3590.
Lee, T.T., Wang, F.Y. Wang and Newell, R.B. (1999). Distributed Parameter Approach to the Dynamics of
Complex Biological Processes. AIChE. J. 45(10), 2245–2268.
Olsson, G. and Newell, B. (1999). Wastewater Treatment Systems: Modelling, Diagnosis and Control. IWA
Publishing. London.
Olsson, G. (1998). Process Control in Biological Wastewater Treatment in Water Quality Management
Library Vol 6. 2nd Ed. Technomic Publishing. Lancaster.
Romagnoli, J.A. and Sáchez, M.C. (2000). Data Processing and Reconciliation for Chemical Process
Operations. Academic Press. London.
Sedory, P.E. and Stenstrom, M.K. (1995). Dynamic prediction of wastewater aeration basin temperature. J.
Environ. Eng. (N.Y.) 121(9), 609–18.
Spanjers, H., Vanrolleghem, P., Olsson, G., and Dold, P. (1998). Respirometry in Control of Activated
Sludge Processes. IAWQ Scientific and Technical Reports No 7, IAWQ, London.
Takács, I., Patry, G.G. and Nolasco, D. (1991). A dynamic model of the clarification/thickening process.
Wat. Res., 25(10), 1263–1271.
Vanrolleghem, P. and Verstraete, W. (1992). Simultaneous biokinetic characterization of heterotrophic and
nitrifying populations of activated sludge with an on-line respirographic biosensor. Wat. Sci. Tech.,
28(11–12), 415–426.
Yuan, Weibo, Okrent, D. and Stenstrom, M.K. (1993). Model calibration for the high-purity oxygen activat-
ed sludge process – algorithm development and evaluation. Wat. Sci. Tech., 28(11–12) 163–171.

363

Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf

by UNIVERSITY OF EDINBURGH user
on 20 December 2018
Downloaded from https://iwaponline.com/wst/article-pdf/45/4-5/355/425267/355.pdf
by UNIVERSITY OF EDINBURGH user
on 20 December 2018