@@@@@ Biotechnology For The Environment

FOCUS ON BIOTECHNOLOGY
Biotechnology for the

Environment:
Wastewater Treatment and
Modeling, Waste Gas
Handling
Edited by
SpirosN. Agatlios and Walter Reineke
Series Editors: Marcel I lofman and Jozef Anné
Springer-Science+Business Media, B.v.

BIOTECHNOLOGY FOR THE ENVIRONMENT:
WASTEWATER TREATMENT AND MODELING, WASTE GAS HANDLING
VOLUME 3C
FOCUS ON BIOTECHNOLOGY
Volume 3C
Series Editors
MARCEL HOEMAN
Centre for Veterinary and Agrochemical Research, Tervuren, Belgium
JOZEFANNE
Reg a Institute, University of Leuven, Belgium
Volume Editors
SPIROS N. AGATHOS
Universite Catholique de Louvain,
Louvain-la-Neuve, Belgium
WALTER REINEKE
Bergische Univ er sit at,
Wuppertal, Germany
COLOPHON
Focus on Biotechnology is an open-ended series of reference volumes produced by Kluwer
Academic Publishers BY in co-operation with the Branche Beige de la Société de Chimie
Industrielle a.s.b.l.
The initiative has been taken in conjunction with the Ninth European Congress on
Biotechnology. ECB9 has been supported by the Commission of the European Communities,
the General Directorate for Technology, Research and Energy of the Wallonia Region,
Belgium and J. Chabert, Minister for Economy of the Brussels Capital Region.
Biotechnology for the Environment:
Wastewater Treatment and Modeling,
Waste Gas Handling
Volume 3C
Edited by
SPIROS N. AGATHOS
ưniversité Cathoỉique de Louvain,
Louvain-ỉa-Neuve, Belgium
and
WALTER REINEKE
Bergische Universitat,
Wuppertal. Germany
SPRINGER-SCIENCE+BUSINESS MEDIA, B.v.

A C.I.P. Catalogue record for tills book is available from the Library of Congress.
ISBN 978-90-481 -6224-6 ISBN 978-94-017-0932-3 (eBook) DOI 10.1007/978-94-017-0932-3
Printed on acid-free paper
All Rights Reserved

© 2003 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2003
Softcover reprint of the hardcover 1 st edition 2003
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
EDITORS PREFACE
At the dawn of the 21st century we are witnessing an expanding human population in
quest of survival and continued well-being in harmony with the environment. Many
segments of society are increasingly preoccupied with the battle against both diffuse and
concentrated pollution, the remediation of contaminated sites, the restoration of damaged
areas due to anthropogenic activities and the re-establishment of functioning
biogeochemical cycles in vulnerable ecosystems. There is an enhanced awareness of the
value of pollution prevention and waste minimization in industrial, urban and agricultural
activities, as well as an increased emphasis on recycling. Faced with these major
contemporary challenges, biotechnology is emerging as a key enabling technology, and,
frequently, as the best available technology for sustainable environmental protection and
stewardship.
Although the activities of microorganisms and thefr subcellular agents have been
recognized, studied and harnessed already for many years in the envfronmental arena,
there is a new dynamism in the in-depth understanding of the molecular mechanisms
underlying the functioning of microorganisms and thefr communal interactions in natural
and polluted ecosystems, as well as an undeniable expansion of practical applications in
the form of the new industry of bioremediation. A number of distinct but increasingly
overlapping disciplines, including molecular genetics, microbial physiology, microbial
ecology, biochemistry, enzymology, physical and analytical chemistry, toxicology, civil,
chemical and bioprocess engineering, are contributing to major insights into fundamental
problems and are being translated into practical environmental solutions and novel
economic opportunities.
The book set «Biotechnology for the Environment^ based on a compilation of some
of the outstanding presentations made at the 9 th European Congress on Biotechnology
(Brussels, Belgium, July 11-15, 1999) and enriched with newly updated thematic chapters,
captures the vitality and promise of current advances in the field of environmental
biotechnology and is charting emerging developments in the beginning of the new
millennium. This third volume, subtitled ‘Waste Water Treatment and Modeling, Waste
Gas Handling’ presents current technological applications of microorganisms in
wastewater treatment and in the control of waste gas emissions, illustrating the importance
of multidisciplinary methodologies for years to come.
In the first section of the book special emphasis is placed on the use of rigorous
mathematical and conceptual models for an in-depth understanding of the complex
biology and engineering aspects underlying the operation of modem wastewater treatment
installations. Biological treatment plants today are without a doubt the major man-driven
microbiological process in terms of sheer volume and societal impact. Although the design
of wastewater treatment installations has evolved over many decades, it is only recently
that detailed biological knowledge has started being translated into reliable dynamic
models ensuring the predictable operation of such plants and their robustness in the face of
external shocks. A first chapter addresses the theory behind the reduction of complex
mathematical models such as the one describing the activated sludge process, to make
them more suitable for process control. This is followed by three contributions
demonsttating, respectively, the determination of parameter sensitivity in the dynamic
model of a relatively simple wastewater treatment plant, the development of a general
protocol to assess the performance of activated sludge processes from a respfromefry-
based confrol perspective and, finally, the comparative merits of model-, fuzzy logic- or
artificial neural network-based approaches for diagnostic purposes in wastewater
treatment plants, including those employing anaerobic treatment. The next chapter offers
an extensive and structured state-of-the-art review of numerous aspects underlying the
evolving versions of comprehensive activated sludge models, aiming at model calibration
for any scale or practical situation. This represents a major confribution in this field, since,
until now, the requisite information for model calibration and process characterization was
scattered in a vast amount of literature. Given the enormous growth in recent years of
biological nitrogen removal from wastewater, a final chapter in this section critically
reviews emerging developments in process design, operational optimization and confrol of
nifrogen removal.
The second section of the book addresses waste gas biofihration, an expanding
biotechnological application of microbial metabolism for afr quality assurance through
processes ranging from the abatement of hazardous volatile pollutants to the elimination
of nuisance odors. The opening chapter here provides an extensive up-to-the-minute
review of the diversity of biological waste gas treatment methodologies and applications
with a balanced attention on both the microbiological fundamentals and the engineering
principles of bioreactor design. In the next chapter an illustration of biofihration
technology covers current developments in the design and installation of prototype and
full-scale bioreactors for the destruction of toxic or otherwise undesfrable volatile organic
chemicals in industry. Finally, the power of an experimental biofilter in efficiently
biodegrading low-concenfrations of organic volatile contaminants in indoor afr is
presented in view of space travel applications.
The Editors hope that the integration of the depth of scientific fundamentals with the
breadth of current and future envfronmental applications of biotechnology so evident in
these selected contributions will be of value to microbiologists, chemists, toxicologists,
envfronmental scientists and engineers who are involved in the development, evaluation
or implementation of biological treatment systems. Ultimately, a new generation of
envfronmental scientists should take these lessons to heart so that new catalysts inspfred
from the biosphere can be designed for safe, eco- and energyefficient manufacturing and
envfronmental protection.
Spiros N. Agathos Walter Reineke

TABLE OF CONTENTS
Colophon......................................................................................................................II
EDITORS PREFACE...................................................................................................1
Table of contents...........................................................................................................3
PART 1 Waste water treatment.....................................................................................9
Model reduction of activated sludge model no. 1 and bioprocess models for
identification and control........................................................................................11
S.R. Weijers.............................................................................................................11
Summary.............................................................................................................11
1. Introduction..................................................................................................11
1.1. Need for reduction of rigorous, mechanistic models.............................11
1.2. Problem statement and methodology....................................................13
1.3. Reduction approaches for process engineeringsystems........................15
1.3.1. New model building from ‘scratch’................................................15
1.3.2. Simplifying assumptions.................................................................15
1.3.3. Neglect of dynamics by quasi steady state assumptions and singular
perturbation.................................................................................................16
1.3.4. Order reduction methods................................................................16
1.3.5. Black-box identification.................................................................17
2. Reduction approaches for ASM1.................................................................17
2.1. New model building from ‘scratch’......................................................17
2.2. Simplifying assumptions.......................................................................18
2.2.1. Simplification with respect to components....................................18
2.2.2. Simplification due to separation of aerobic and anoxic conditions ..19
2.2.3. Simplifying assumptions with respect to kinetics..........................19
2.2.4. Simplification with respect to dynamics........................................19
2.2.5 Reduced order models........................................................................19
2.3. Dynamics of variables...........................................................................23
2.4. Order reduction methods.......................................................................25
2.5. Black-box identification........................................................................25
2.6. Discussion and conclusions...................................................................25
3. Singular perturbation of bioprocess systems: theory and review.................26
3.1. Singular perturbation theory..................................................................26
3.2. Review of model reduction of (bio)process systems using singular
perturbational approach..................................................................................28
3.2.1. General rule for order reduction.....................................................28
3.2.2. Other reduction approaches............................................................30
3.3. scaling in model reduction....................................................................32
3.3.1. A method for state partitioning based on scaling...........................32
3.3.2. Relationship of scaling with regime analysis and dimensional
analysis........................................................................................................36
3.4. Other methods for timescale analysis....................................................36
3.5. Conclusions...........................................................................................37
4. Scaling for singular perturbation in a simple bioprocess system.................37
4.1. Introduction and methodology..............................................................37
4.1.1. Procedure 1: Direct scaling............................................................38
4.1.2. Procedure 2: Timescale estimation for variables...........................38
4.1.3. Procedure 3: Analytical scaling procedure................................... 42
4.2. Model system: chemostat with biomass and substrate..........................42
4.2.1. Model system definitions and analysis..........................................42
4.2.2. Cases 1 and 2: Ratio s/x very low, T* very large..........................46
4.2.3. Case 3: T* intermediate, Mo«l.......................................................51
4.2.4 Case 4: s and X comparable, T* moderate, Mo » 1...........................53
4.2.5. Case 5: T* close to critical, Mo « 1, Te~ Mo................................54
4.2.6. Case 6: T* close to critical, Mo « 1, Te« Mo................................54
4.3. Other results..........................................................................................57
4.4. Conclusions...........................................................................................57
5. Conclusions and perspectives.......................................................................58
References...........................................................................................................60
Parametric sensitivity of a dynamic model for a pilot single-stage wastewater
treatment plant........................................................................................................65
Igor Plazl, Goran Pipus, and Tine Koloini.............................................................65
Summary.............................................................................................................65
1. Introduction...................................................................................................65
2. Materials and methods..................................................................................66
3. Mathematical model.....................................................................................67
4. Results and discussion..................................................................................69
5. Conclusions..................................................................................................72
Acknowledgment................................................................................................72
References...........................................................................................................72
Applicability of a simulation benchmark to respừometry-based control strategies73
John B. Copp and Henri Spanjers...........................................................................73
Abstract............................................................................................................. 73
1. Introduction...................................................................................................73
2. General benchmark description....................................................................74
3. Plant layout................................................................................................. 75
4. Process models.............................................................................................75
5. Test protocol.................................................................................................76
6. Performance assessment...............................................................................77
7. Application to respirometry-based control strategies...................................78
7.1. Joyce et al. (1974)........... . ...................................................................79
12. Stenstrom and Andrews (1979).................................................................79
7.3. Sprensen (1980)................................................................................. 79
7.4. Sekine et al. (1988)...............................................................................80
7.5. Kim etal. (1996)....................................................................................80
7.6. Larose et al. (1997)...............................................................................80
8. Overview and discussion of results..............................................................81
9. Conclusions..................................................................................................84
Acknowledgements.............................................................................................84
References........................................................................................................... 84
Fault detection and isolation in waste water treatment plants..................................87
Jean-Philippe Steyer and Jerome Harmand..............................................................87
Abstract................................................................................................................87
1. Introduction....................................................................................................87
2. Examples of fault detection and isolation methods for anaerobic digestion
processes..............................................................................................................89
2.1. The experimental process.......................................................................89
2.1.1. Characteristics of the wastewater....................................................89
2.1.2. The anaerobic digestion process.....................................................89
2.2. The model-based FDI approach.............................................................91
2.3. The ANN-based FDI approach...............................................................92
3. Fuzzy supervision of an industrial equalization process...............................94
3.1. The equalization process........................................................................94
3.2. The fuzzy supervisor..............................................................................95
4. Conclusions...................................................................................................98
References............................................................................................................99
Calibration of activated sludge models: a critical review of experimental designs . . .
..........................101
B. Petersen , K. Gemaey , M.Henze , P.A. Vanrolleghem ....................................101
Abstract............................................................... .7..........................................101
1. Introduction..................................................................................................101
2. Description of the state-of-the-art activated sludge models........................102
2.1. Activated Sludge Model N°.l (ASM1)............102
2.1.1. COD components in ASM1..........................................................102
2.1.2. Nitrogen components in ASM1....................................................103
2.1.3. Processes in ASM1.......................................................................106
2.1.4. Restrictions of ASM1....................................................................108
2.2. Activated Sludge Model N°. 3 (ASM3)...............................................109
2.2.1........................................................................................................... COD
components in ASM3...................................................................................................Ill
2.2.2. Nitrogen components in ASM3.......................................................Ill
2.2.3. Processes in ASM3.........................................................................114
2.2.4. Restrictions of ASM3....................................................................115
3. Model Calibration........................................................................................115
3.1. Information set for model calibration...................................................117
3.2. Model calibration levels.......................................................................119
3.2.1. Steady state model calibration......................................................119
3.2.2. Dynamic model calibration...........................................................120
4. Characterisation of waste water and sludge kinetics...................................124
4.1. Wastewater characterisation.................................................................124
4.1.1. Physical-chemical characterisation...............................................124
4.1.2. Summary and discussion of physical-chemical wastewater
characterisation..........................................................................................128
4.1.3. Biological characterisation............................................................129
4.1.4. Summary and discussion of biological wastewater characterisation
............... ...... 143
4.1.5. Discussion on physical-chemical vs. biological wastewater
characterisation...........................................................................................144
4.2. Characterisation of sludge composition................................................148
4.3. Characterisation of stoichiometric and kinetic parameters...................150
4.3.1. Respfrometry.................................................................................150
4.3.2. Nitrate utilisation rates...................................................................160
4.3.3. Titrimetry.......................................................................................161
4.3.4. Summary and discussion of biological characterisation of
stoichiometric and kinetic parameters........................................................161
4.4. Is characterisation via lab-scale experiments relevant?........................163
4.4.1. Kinetic and stoichiometric parameters..........................................166
4.4.2. Relevant kinetic and stoichiometric parameters for lab-scale
characterisation...........................................................................................170
4.4.3. Relevant wastewater components for lab-scale characterisation ....171
5. Biological experimental constraints.............................................................172
5.1. Transferability between model concepts: an example........ . ................173
5.2. Review and discussion of S(0)ZX(0) ratio...........................................174
5.2.1. Effect of S(0)ZX(0) on stoichiometry...........................................175
5 2.2 Effect of S(0)/X(0) on kinetics..........................................................178
5.2.3. Discussion on S(0)/X(0) ratio........................................................179
6. Summary......................................................................................................180
Acknowledgement..............................................................................................181
References..........................................................................................................181
Optimization and control of nitrogen removal activated sludge processes: a review of
recent developments...............................................................................................187
Zhiguo Yuan, Jiirg Keller and Paul Lant................................................................187
Abstract...........7.................................................................................................187
1. Introduction..................................................................................................187
2. Elementary analysis of biological nitrogen removal systems......................189
2.1. System analysis................................................................................... 189
2.1.1. SRT and volume requfrement........................................................189
2.1.2. Anoxic fraction and volume requfrement......................................190
2.1.3. Anoxic fraction and nitrification capacity................................... 190
2.1.4. Anoxic fraction and influent COD utilization efficiency for nitrate
reduction.....................................................................................................191
2.1.5. Alkalinity and pH...........................................................................193
2.2. Optimization opportunities...................................................................193
2.2.1. Optimisation by on-line process control........................................193
2.2.2. Optimisation by improved process design.....................................194
2.3. Conclusions...........................................................................................195
3. Aeration control...........................................................................................195
3.1. Aeration phase length control based on ORP and pH measurement....195
3.1.1. Aeration control based on absolute values of ORP and pH...........197
3.1.2. Aeration control based on bending points of ORP and pH............198
3.2. Aeration control based on respfrometry................................................199
3.3. Aeration control based on ammonia and nitrate measurement.............200
3.3.1. Objective functions........................................................................200
3.3.2. Aeration control of alternating systems.........................................201
3.3.3. Aeration control of pre-denitrification systems.............................202
3.3.4. Model-based control......................................................................203
3.4. Conclusions...........................................................................................204
4. External COD dosage optimisation and control..........................................204
4.1. External carbon sources........................................................................205
4.2. Utilization efficiency of external COD................................................206
4.3. External carbon dosage control systems...............................................207
4.3.1. Control of external carbon dosage to recừculating biological nitrogen
removal systems.........................................................................................208
4.3.2. Control of external carbon dosage to alternating biodenipho®
systems.......................................................................................................210
4.3.3. Control of external carbon to two-sludge post-denitrification systems
........... . .................................................................................................... 210
4.4. Conclusions..........................................................................................210
5. SRT optimisation and control.....................................................................211
5.1. Impact of SRT on a biological nitrogen removal plant........................211
5.2. SRT minimization via surplus sludge waste flow control...................212
5.3. Conclusions...........................................................................................213
6. Side-stream nitrifier supplies.......................................................................213
6.1. Shortening the RTs of inert solids via sludge storage...........................214
6.2. Side stream nitrification of reject water.......................... . .......................215
6.3. Conclusions...........................................................................................216
7. Novel multi-sludge systems.........................................................................217
7.1. Attached growth processes...................................................................217
7.1.1. SRT and volume requừement........................................................218
7.1.2. Treatment capacity.........................................................................218
7.1.3. Influent COD utilization efficiency for nitrate reduction..............218
7.1.4. Comparison with other multi-sludge systems................................219
7.1.5. Aeration in an attached growth system..........................................219
7.2. COD preservation for denitrification....................................................219
7.3. Conclusions...........................................................................................221
8. Conclusions............... . ................................................................................221
Acknowledgment................................................................................................222
References...........................................................................................................222
PART 2 Waste gas biofiltration.................................................................................229
Performance and characterisation of a membrane biological aừ filter for space
applications.......................................................................................................... 231
Jaap Van Der Waarde, Arjan Van Der Werf, Maurice Henssen, Bert Geurkink, Klaas
Van Der Marel, Piet Paul and Marc Gent..............................................................231
Summary........................................................................................................... 231
1. Introduction..................................................................................................231
2. Materials and methods.................................................................................232
3. Results and discussion..................................................................................233
3.1. Physiological characterisation...............................................................233
3.2. Performance of the biological aừfilter (BAF)......................................235
3.3. Molecular characterisation...................................................................235
3.4. Performance of the BAF at extremely low concentrations of volatile
contaminants................................................................................................. 236
4. Conclusion...................................................................................................236
References..........................................................................................................237
Biofilttation for waste gas handling.......................................................................239
Frederic Thalasso , Maria c. Veiga andChristianKennes.............................239
1. Introduction..................................................................................................239
2. Bioscrubbing................................................................................................239
3. Trickling filters............................................................................................240
4. Biofiltration with organic packing materials...............................................242
5. Applications.................................................................................................246
5...............................................................................................L General aspects
....................................................................................................................................246
5.2. Industrial scale applications..................................................................248
6. Other biological gas treatment technologies...............................................250
7. Conclusions.................................................................................................251
Acknowledgements............................................................................................251
References..........................................................................................................252
Bioreactors for the treatment of industrial waste gases containing formaldehyde and
other aliphatic compounds......................................................................................259
Óscar J. Prado, Marta Eừoa, Marfa c. Veiga and Christian Kennes......................259
1. Introduction..................................................................................................259
1.1. Gas phase bioreactors...........................................................................259
1.2. Formaldehyde as industrial aừ pollutant...............................................260
2. Biodegradation of formaldehyde.................................................................261
3. Case studies on biological abatement of formaldehyde in industrial waste
gases...................................................................................................................264
3.1. Background...........................................................................................265
3.2. Optimization of the removal of formaldehyde in biofilters and
biotrickling filters...........................................................................................270
Acknowledgements............................................................................................272
References..........................................................................................................273
INDEX.......................................................................................................................275
PARTI
WASTEWATER TREATMENT
MODEL REDUCTION OF ACTIVATED SLUDGE MODEL NO. 1 AND
BIOPROCESS MODELS FOR IDENTIFICATION AND CONTROL
S.R. WEUERS
Systems and Control Group, Faculty of Applied Physics
Eindhoven University of Technology, Eindhoven, The Netherlands email:
S.R.Weijers@tue.nl
Summary
This paper treats model reduction of the important Activated Sludge Model No. 1
(ASM1) for dynamic modelling of waste water treatment plants. Section 1 motivates the
need for model reduction and gives an overview of reduction approaches in the literature.
Section 2 reviews model reduction of activated sludge models, especially ASM1. The
review shows that model reduction is often performed quite heuristically and that more
insight in the timescale properties of ASM1 is desired. Singular perturbation is a
systematic technique for order reduction. Section 3 reviews application of singular
perturbation to bioprocess systems to reveal if methods exist to obtain or detect the so-
called standard form, which is the difficult part of reduction by singular perturbation.
Because existing methods appear insufficiently straightforward, Section 4 develops a
method for this task. Three procedures are proposed and tested on a simple continuous
general bioprocess model, which yields valuable insight into properties of bioprocess
models. A proposed timescale estimation procedure appears a helpful tool in model
reduction through timescale separation and provides a good basis for further reduction
studies.
1. Introduction
1.1. NEED FOR REDUCTION OF RIGOROUS, MECHANISTIC MODELS
Dynamic simulation based on rigorous, physics-based mechanistic modelling has become

a standard tool in many engineering fields. Examples are finite element models in
structural dynamics, computational fluid dynamics, the SPICE program for circuit
11
S.N. Agathos and w. Reineke (eds.), Biotechnology for the Environm Wastewater Treatment and Modeling, Waste Gas
Handling, 11-63. © 2003 Kluwer Academic Publishers.
S.R.
Weijers
analysis in electrical engineering and dynamic flowsheeting in chemical engineering. In

biotechnology, process modelling is well established (Roels, 1983).
Rigorous models are applied for a variety of tasks. They are useful in system design,
especially to check system behaviour under extreme dynamic loading conditions. They
are very helpful in understanding system behaviour in science and engineering. As they
typically contain much prior knowledge from the relevant application domain, thefr
prediction accuracy and range of validity can be high in principle, if the model building
and model tuning tasks are designed to achieve this accuracy. However, the costs of
rigorous models are generally high because of the effort requfred to construct them.
For engineering tasks described above, typically high-dimensional models result.
Despite the usefulness of rigorous, large models for the tasks described above, for the
following important tasks simple models are better suitable:
• Process control: Many control theory concepts are only applicable to low- order
models. The high dimensionality of large models results in enormous computational
requirements, ill-conditioned problems and often stiff numerical problems due to
interaction of slow and fast dynamics (Kokotovic et al., 1986). Relatively low-order
reduced models are therefore requfred for confroller design and as internal models in
model-based control.
• Model identification: Rigorous mechanistic models typically requfre high
investments in model tuning and validation. Moreover, problems that are more
fundamental exist because large models typically exhibit lack of parameter
identifiability. In addition, mechanistic models contain internal states whose
behaviour is difficult to verify (or falsify), which is referred to as verifiability
(Jeppsson, 1996). The need for simple, well-identifiable models holds true for off-
line identification and even stronger for on-line identification for process monitoring
and adaptive control.
• Understanding of model behaviour: While rigorous models are helpful in system
understanding, at the same time this understanding is hampered by thefr complexity.
Much understanding can be acqufred from reduced models describing only the most
important phenomena.
• System design through rigorous optimisation: Many design problems might be
solved more straightforwardly by applying an analytic design procedure, using
(mathematical) multi-criteria optimisation. Due to thefr size, however, rigorous
models are not suitable for dfrect system design using optimisation; rather, they are
most often used to check designs. Straightforward, systematic system design
employing rigorous multi-objective optimisation would be facilitated if simple
models containing the most important phenomena would be available.
Concluding, one may state that model reduction is requfred for several important tasks. In
wastewater engineering, rigorous modelling and dynamic simulation have become well
accepted during the past decade since the publication of Activated Sludge Model No.l
(ASM1, Henze et al., 1987). This model is now becoming routinely used in waste water
engineering. Model tuning of ASM1 has become an important task for process analysis,
process optimisation and process control. Furthermore, advanced, model based control of
wastewater treatment plants is expected to be important in meeting more stringent
1
2
Model reduction of activated sludge model no. 1 and bioprocess models for identification and control
treatment requfrements that are imposed on wastewater treatment, especially with respect
to nitrogen removal, in an economical way. Mechanistic models based on ASM1 are
however not dfrectly suitable for control. If these mechanistic models accurately describe
the process dynamics, however, reduced models can be extracted for confroller design
thus re-using the knowledge in the mechanistic models. Consequently, model reduction
of ASM1 has received considerable attention in the literature.
In bioprocess engineering, reduced models may be fruitfully applied in process
optimisation and process control, for example for dynamic optimisation of individual
(fed-) batch processes, batch process scheduling and model based process monitoring.
1.2. PROBLEM STATEMENT AND METHODOLOGY
This paper wants to contribute to advancing the field of model reduction of ASM1 and
bioprocess models, with emphasis on application in (model based) control and
identification. The ultimate goal is to provide a methodology to derive nonlinear reduced
models for different timescales in a straightforward, systematic manner. In this paper, the
goal is to develop a method that can provide the starting point for such a straightforward
nonlinear reduction procedure. This section outlines the methodology that is followed in
the sequence to develop such a method.
To select a suitable reduction method, it is useful to first list desfred properties of a
candidate reduction method. The following properties are considered as desired in this
paper.
Nonlinear reduction methods are preferred, because nonlinear models can have a
larger validity range than linear models.
The stiffness of the activated sludge process requires the development of models that
are suitable for different timescales. For example, control of dissolved oxygen (DO)
requires a different timescale than control of ammonia (and nitrate) or control of sludge.
Models on these different timescales are therefore requfred for each of these control
tasks, either for controller design or as internal model. Moreover, this fits well in a
hierarchical control approach. Therefore, especially model reduction based on timescale
separation is a logical approach.
Moreover, it is desfred that the reduction method is systematic and straightforward in
order to avoid time-consuming trial-and-error and iterations, and to be independent of
(too much) application-domain dependent knowledge. Ideally, the reduction method
would also supply an estimate of the error induced by the reduction.
Another desfred property of reduction methods is that the states of the model retain
thefr physical interpretation after reduction. This will enable a more dfrect interpretation
of the controller design results and of the control actions. Moreover, if adaptive control is
applied, identified parameters have a dfrect physical interpretation.
Finally, as this paper focuses on the pre-denitrification plants and carrousels, the
method should be applicable to derive reduced methods for these systems. Several
reduced models have been derived for pre-denitrification plants, or, more exactly, for
aerobic and anoxic conditions separately. However, reduced models for systems where
simultaneous denitrification takes place, such as typically is the case in carrousels, are
lacking. Methods that are helpful in reduction for these systems are therefore desired.
The need for reduced models in wastewater engineering has resulted in a variety of
1
3
S.R.
Weijers
reduction approaches and reduced order models of ASM1. Section 1.3 discusses the most
important reduction approaches in the literature. Section 2 gives a state-of-art overview of
ASM1 model reductions that are classified along the discussed reduction approaches.
Aim of this overview is to obtain clarity into the limitations and possibilities of different
reduced models and reduction methods. This should support selection of a suitable
reduced model or reduction method.
The review in Section 2 shows that reduction through timescale separation is applied
by several authors, but often in a heuristic fashion. Moreover, the timescale properties of
ASM1 are not well understood.
The technique of singular perturbation is therefore studied in depth to better
understand the timescale properties of ASM1 and thus provide a basis for developing a
reduction methodology. This technique has the desừable properties of a candidate
reduction technique listed above. It provides the mathematical basis for reduction by
timescale separation, provides an error estimate, it is a systematic technique, and may
therefore lead to a more straightforward reduction procedure. Moreover, it is applicable
to nonlinear systems and, under certain conditions, the physical interpretation of states is
retained.
In section 3.1, the theory of singular perturbations (as used by Kokotovic et al., 1986)
is summarised. For application of the singular perturbation technique for order reduction,
the model must be in the so-called standard form. This means that the states of the model
can be partitioned into ‘fast’ and ‘slow’ scales. (Also partitioning into more timescales is
possible (e.g., fast, medium and slow)). In that case, the reduction boils down to either
eliminating the fast states to obtain the slow model or to eliminating the slow states to
obtain the fast model. Upon reduction, the physical interpretation of states is then
preserved.
If the model is not in standard form, a state transformation may be applied to bring it
into standard form. In that case, the physical interpretation of the states is not retained. In
fact, the fast and slow modes cannot be assigned to disjunct sets of states. This situation
is not studied here.
It also happens that the model can be partitioned into slow and fast states, but that it is
not easy to recognise this from the model equations. In that case, writing the model
equations in a different way, e.g. by scaling, may show immediately from the scaled
equations that the partitioning is possible indeed. Finding such a scaling is difficult
however. In fact, recognising whether the model is in standard form and detection of the
fast and slow timescales is the difficult part of the reduction technique.
Therefore, the study in this paper focuses on recognising the state partitioning into fast
and slow states and on recognising or obtaining the standard form.
One application of singular perturbation theory to an activated sludge plant model has
been reported, which is discussed in section 2.3. The authors applied a method that is
based on eigenvalues for the state partitioning. The reduction was only partly successful,
section 3.2 reviews application of singular perturbation to reduction of - more general and
closely related - bioprocess systems into more detail to obtain a more fundamental insight
into timescale properties of bioprocess models in general and of ASM1 in particular. The
review shows that singular perturbation of bioprocess models is not sufficiently well
understood and it does not yield a method or rule for the state partitioning. Therefore, a
method for this task is developed. A scaling procedure summarised in section 3.3
1
4
provides the starting point for this method.

Three procedures are proposed in section 4 as candidates for the state-partitioning task.
They are tested on a simple bioprocess model in different operating conditions.
1.3. REDUCTION APPROACHES FOR PROCESS ENGINEERING SYSTEMS
Starting point in model reduction is the definition of the goal for which the model is
intended. It can be argued that, for biological systems even stronger than in other basic
sciences, the goal of the model determines its formulation (Vansteenkiste and Spriet,
1982). The goal determines selection of model inputs, including possible reference
trajectories and disturbances, selection of model outputs, and determines which model
accuracy is required over which time horizon.
For the actual model reduction in this review, several approaches are distinguished
and briefly explained below. The models obtained with these approaches differ in thefr
degree of ‘greyness’. Models that are obtained via systematic reduction of a white model
whilst preserving the physical interpretation of the system states can be considered light
grey. Simple, mechanistic input-output models are considered grey (Carstensen et al.,
1995). The last category is that of "black box models", such as polynomial models or
artificial neural networks. It is noted that also mixed forms can be applied, e.g. models
contain a mechanistic part and an artificial neural network (e.g. van Can, 1997).
Alternatively, artificial neural networks can be configured to contain prior knowledge,
e.g. of model structure.
The models are expected to allow further extrapolation beyond the experimental
domain with increasing ‘lightness’.
1.3.1. New model building from "scratch"

One way to construct simple models is to disregard existing rigorous models and to build
simple models from scratch. Of course, strictly speaking this is no model reduction.
Nevertheless, the prior knowledge contained in rigorous models is often either implicitly
or explicitly used in this approach.
1.3.2. Simplifying assumptions

Application of simplifying assumptions is a very frequently applied approach in model
reduction. Here, we distinguish simplifying assumptions with respect to:
• Components; e.g. aggregation of variables (for example COD and BOD as total
measures of pollutant concentrations);
• Processes; e.g. aggregation of reactions (for example modelling nitrification as a
one-step process whilst it is a two-step process);
• Lumping of space distribution: neglect or simplification of gradients in one or more
dfrections;
• Kinetics, e.g. simplification of complicated kinetic schemes;
• Dynamics; these are treated in the next paragraph.
Often, simplifying assumptions are applied in a rather heuristic fashion, especially with
respect to components. For reduction of distributed systems, several systematic methods
can be applied (Dochain, 1994), which are outside the scope of this paper.
1
5
S.R.
Weijers
13.3. Neglect of dynamics by quasi steady state assumptions and singular perturbation
The neglect of dynamics that are fast or slow compared to the timescale of interest is
discussed in its own right here, as it is a central reduction approach. For example, it
provides the basis for application of hierarchical control. In hierarchical control, a layered
control approach is applied to control large, composite systems, for example in plant wide
control. The control problem is decomposed into a hierarchical set of several levels of
smaller sub-problems. On each level, dynamics of lower levels are assumed very fast and
considered to be in (pseudo) steady state and dynamics of higher levels are assumed very
slow and considered as constant.
In many cases, neglect of dynamics is performed heuristically. From field specific,
physical understanding and process knowledge, fast states are omitted or slow states
considered constant, without firm motivation. Although heuristically reduced models may
(seem to) perform satisfactory, there is a need for a more thorough understanding of
criteria for performing this reduction, preferably supported by formal proof or error
analysis.
A well-known heuristic reduction approach 3 is the quasi-steady state approximation
(QSSA) for reactive intermediate species. This approach, first introduced by Bodenstein
and Lutkemeyer (1924) (Bowen et al., 1962), has been extensively used in kinetic
modelling. Criteria for its application were originally that concenfration and timescale of
the intermediate species are small. Later, the nature and consequence of the QSSA have
been studied more thoroughly.
Singular perturbation is a mathematical technique to analyse timescale multiplicity
and to perform a systematic order reduction and error analysis. It is the appropriate tool to
provide the mathematical basis for application of quasi steady state assumptions. Its
application to the QSSA in Michaelis-Menten kinetics is well studied, as will be
summarised in section 3.3.
13.4. Order reduction methods

In modern control engineering, order reduction of linear models is a very important task
in control system design. Models for control are often obtained from linearization of
high-dimensional rigorous models, resulting in very high dimensional models. Modern
controller design methods, especially robust confrol design methods, yield even higher
dimensional controllers. Order reduction of model or controller has become a necessity if
they are to be implemented, as high-order controllers are usually not accepted in industry.
Moreover, they may exhibit extremely poor robustness against controller parameter
errors.
Consequently, order reduction has rapidly progressed and since the milestone publication
by Moore (1981), a variety of reduction methods have been developed and are relatively
well understood. Examples of reduction methods are Hankel norm reduction, balanced
reduction and modal reduction. More recently, closed loop model reduction has received
attention; see for example Wortelboer (1994), who also gives a thorough discussion on
linear reduction methods.
For nonlinear system, balanced reduction methods have been developed (see
Scherpen, 1994). With these methods, however, it may be more difficult to preserve the
physical interpretation of states, as a state transformation is typically involved. Moreover,
3 With ‘model reduction’, we will also denote ‘model order reduction’.
1
6
these methods do not provide a timescale separation. Instead, they eliminate states that
are poorly jointly observable and controllable and thus contribute least to input-output
behaviour. Because we want to preserve the physical interpretation of states and focus on
timescale separation, these methods are not considered further.
7.3.5. Black-box identification

Reduction through black box identification is an approach that requừes little prior
knowledge. Linear black box identification can be applied for weakly nonlinear systems
within a limited domain. To construct reduced models of nonlinear systems with validity
over a larger domain, nonlinear black box modelling techniques may be applied,
especially artificial neural networks (ANN’s). The design of test signals in reduction
through identification is very important. By selecting appropriate frequency ranges for
the excitation signals, black box models may be obtained for different timescales.
2. Reduction approaches for ASM1
This section reviews approaches applied to reduce ASM1. The emphasis is put on
systems with enhanced nitrogen removal. In most of the cases presented, the purpose of
model reduction is application for identification or control. The subsections follow the
arrangement of reduction approaches as presented in section 1.3. For each case discussed,
the treatment system, goal for model reduction, motivation for selected approach will be
indicated together with (a description of) the reduced models. For a description of ASM1
please consult Henze et al. (1987) or other sources (e.g. Dold and Marais, 1986).
2.1. NEW MODEL BUILDING FROM ‘SCRATCH’
Marsili-Libelli (1989) developed a low-order model for conventional activated sludge

systems with BOD removal and nitrification. Motivation was that literature models are
not suitable for control, due to theứ complexity and poor identifiability. The model was
developed to describe a) biodegradation carbonaceous COD b) nitrification c) DO
utilisation d) sludge sedimentation. As a growth model a predator/prey modified
Volterra-Leslie logistic equation was used instead of the usually applied, poorly
identifiable Monod model. Theoretical and practical identifiability of the model were
investigated and an observer was constructed based on this model.
Isaacs (1996) formulated several simple models for control of the Biodenipho® system,
which is an alternating, sequential semi-batch type system with nitrogen (and
phosphorus) removal. Four (or six) phases are applied in each cycle. The control and
model horizon is one phase, with a timescale in the order of several minutes. Three
model-based control strategies were tested, external carbon addition control (ECAC),
dissolved oxygen setpoint control (DOSPC) and cycle length control (CLC). All
confrollers employ a relational model and a predictive model. The predictive model is
used to compute required denitrification rate, nitrification rate or cycle length during one
phase. The predictive model was roughly the same in all control strategies, assuming
zero-order kinetics both in nitrification and denitrification phase. Changes in biomass
amount and composition need not be predicted, as actually measured denitrification and
denitrification rates from the last preceding cycle are used. Different relational models
1
7
S.R.
Weijers
were applied in the different strategies, as explained below. The relational model in CLC
is trivial.
In ECAC, the prediction model computed the required denitrification rate r d; the
requfred external carbon addition rate qcoD to achieve this r d is computed from the
following relational model with the constants rdj b and rdi max:
=R , _ QCOD
R
r
d “ rd,b + rd,max • —------------------------ (2.1)
K
COD +<1COD
In the model, the half-rate constant KCOD was fixed, so the model is linear in parameters.
The model parameters were estimated on-line from measured carbon addition and
denitrification rates in preceding cycles using recursive least squares.
In DOSPC, the requfred nitrification rate r n is computed with the prediction model.
The requfred DO setpoint to achieve this nitrification rate is computed from the relational
model and is held constant during one phase (using DO control).
so
r
n rn0 "*■ rn,max ■ v c Ko +5O (2.2)
This relational model is estimated from measured DO and nitrification rates in

preceding cycles using recursive least squares.
2.2. SIMPLIFYING ASSUMPTIONS
Several simplifying assumptions are applied in ASM1 reduction, which can be classified
as follows (for an explanation of the symbols, see the symbols list at the end):
2.2.1. Simplification with respect to components
COD components:
• No hydrolysis, no distinction between soluble and suspended biodegradable COD
• Do not consider inert products as a separate component (XỊ includes Xp)
• Do not consider inert components at all (no Si, XI, Xp)
N components:
• No suspended organic bound entrapped N (No XND)

• No ammonia production from organic bound nittogen (No SND)
Alkalinity:
• Do not consider alkalinity
2.2.2. Simplification due to separation of aerobic and anoxic conditions
1
8
• No denitrification under aerobic conditions

• No nitrification under anoxic conditions
2.2.3. Simplifying assumptions with respect to kinetics
• Replace terms with Monod kinetics by first-order or zero-order kinetics (the last
corresponds to neglect of dependency on one or more limiting substrates)
2.2.4. Simplification with respect to dynamics
• Neglect DO dynamics
• Neglect dynamics of solute components
• Neglect dynamics of suspended components
2.2.5 Reduced order models

If one considers the huge number of possible combinations of simplifying assumptions, it
is clear that a large number of reduced order models can be formulated. Up till now, a
limited number of reduced order models have been published, which are presented in this
section.
A reduced and modified ASM1 model to be used in adaptive control was formulated
by van Impe et al. (1991) and by Vanrolleghem (1994). This is not discussed further, as it
was developed for conventional activated sludge systems.
Carstensen et al. (1995) formulated grey models to have operational dynamic models
capable of giving on-line information on the present state of the wastewater treatment
plant. The system was the Biodenipho process, which has an alternating, sequential semi-
batch type of operation, where nitrification and denitrification are separated in time.
Different models are applied for both conditions. No biomass growth is modelled.
Nitrification:
• The rate equation for ammonia removal contains both the maximal autotrophic
growth rate and yield, which are not identifiable if the rate equation is not coupled to
growth model for autotrophic biomass. Therefore, one overall maximal nitrification
rate parameter rnitmax is used
1
9
S.R.
Weijers
• No method for measuring the concentration of the autotrophic biomass exists, (2.3)
so the measured total solids concentration x ss is used. This gives the
following rate expressions during one batch phase.
dS
NH _ _r . SNH__________________So
, ‘nit.max ■ „
K
„ „ „ Ass (2.4)
dt NH+SNH KQ+SQ
dS
NO . SNH____________So
J nit,max TT___________ss
dt K
NH+SNH KO+SO
Denitrification:
• One overall maximal denitrification rate rđenit,max is used.

• No method for measuring the concentration of heterotrophic biomass exists, so the
measured total solids concentration is used.
• The ammonium load to the plant is used as this proved a correlated measure of the
readily biodegradable substrate concentration, similar to Isaacs (1996) (Eq. 2.1).
d$NO _ --r. . . SNO_________________rload V /Ọ

, - rdenit,max • ____■ q____' K, ,+ n , ss ’
K
dt NO + SNO Kload + rload
Lukasse et al. (1997a,b, 1998a,b, 1999) developed reduced models for control of nittogen
removal in an activated sludge process consisting of one CSTR; clarifier dynamics were
neglected. Two levels of model reduction were applied:
• For nonlinear (open-loop) optimisation (optimisation horizon 1 day, Lukasse et al.,

1998a). To investigate optimal aeration strategies, a nonlinear model was developed.
Dynamics of all suspended components, including biomass, were neglected. With
respect to solute components, inert COD and alkalinity were not considered; DO
dynamics were neglected by considering DO as an input instead of a state variable
(assuming tight control). Ammonia production due to hydrolysis and ammonification
was simplified by using an (constant) influent ammonia correction factor (in fact, this
last simplification implies a quasi-steady state assumption on hydrolysis and
ammonification). These assumptions result in a reduction of ASM1 to a reduced
thừd-order nonlinear model with the state variables Ss, SNH, SNO.
• In receding horizon optimal control (Lukasse et al., 1997a,b, 1998a,b, 1999). To
cừcumvent the disadvantages associated with the open-loop strategy, especially
associated with uncertainty of the affinity constants, feedback was introduced by
applying RHOC (Receding Horizon Optimal Control), a Model Predictive Control
law. Here, the requừement was to obtain an identifiable model, suitable for on-line
optimisation of an alternating process, with a
2
0
• timescale of interest of hours. Assumptions

were stated in Lukasse et al. (1997a and
1998a,b) (slightly different models were used,
however): the system was forced to be either
aerobic or anoxic by restricting So e {0, 3},
transients between aerobic and anoxic phase
were neglected, Monod kinetics was simplified
to Blackman kinetics, arguing that the
corresponding affinity constants were small and
the substrate dependency only serves as a
switching function. Additional assumptions were
1) If SNH=0 and the system is aerobic, then the
nitrification rate is such that all ammonia
entering the reactor is nitrified. 2) ammonia
and nitrate concentrations in the sludge
recycle are equal to the concentrations in the
reactor. This leads to the following equations:
d I SNH j _ I -rNH , I DinSNH,in

— 1=1 II (2.6)
dt^SNOJ VNH+rNO7 I -rNO
S
_ I rNH,max NH >0
r
NH Q S (2.7a)
NH =0
— rNO,max S
NH >0
rNO -|o (2.7b)
S
NH =0
with UG {0,1}. (1 for aerobic, 0 for anoxic) and D in is the dilution rate. For use in
adaptive RHOC, SNH,in, rNH,max and rN0,max were estimated based on
measurement of SNH> SNO and So.
Jeppsson (1996) developed a reduced model for ASM1 for application in control
including adaptive control and requfring identifiability and ‘verifiability’ of the
parameters and states (Jeppsson and Olsson, 1993). This last property expresses the
specification that the states of the model are verifiable with information obtained from
measurements. Reactors are assumed to be either aerobic or anoxic (separated in space)
and ideally mixed (pre-denitrification system). The model has been used in studies by
other authors (Ayesa et al, 1995) and is presented here into more detail.
The following assumptions and simplifications were made:
2
1
S.R.
Weijers
• Do not consider alkalinity

• Do not consider inert components (biologically not relevant, Xj is slow)
• No ammonia production from organic bound nitrogen (No XND, SND)
• No hydrolysis, no distinction between soluble and suspended biodegradable COD
(hydrolysis is not well understood; difficult to measure Ss and xs; neglect fast
dynamics of Ss).
• Neglect DO dynamics: assumed to be controlled
2
2
• Approximate Monod expressions by Blackman kinetics (it is assumed that substrate

concentration is low).
Under these assumptions, five model components remain, namely X BH, XBA, SNH, SNO
and XCOD- Two models were derived, one for anoxic and one for aerobic conditions.
Under anoxic conditions, there is no nitrification. Three processes take place: growth
of heterotrophs, decay of heterotrophs and decay of autotrophs. The resulting model
contains four parameters: rH, YH, bH and bA.
r
XCOD - ” VrH ' XCODXBH + bHXBH + bAXBA (2-8)
Y
H
SNH =-iXB(rH * XCODXBH -bHXBH —ÂXBA)

r
(2-9)
r
sN0 = ~2 86Y^rH XcodXbh (21
°)
r
XBH = rH XCODXBH - bHXBH (2.11)
r
XBA = - ÂXBA (2.12)
Under aerobic conditions, there is no denitrification. Four processes take place: growth of
heterotrophs, decay of heterottophs, growth of autotrophs and decay of autotrophs. The
model contains six parameters: rH, YH, bn, rA, YA and bA.
XC0D = -•77“rH XCODXBH +bHXBH +bAXBA

r
(2.13)
ỵ
H
1 Ì (2.14)
r
SNH _-iXB(rH XCODXBH - ^HXBH
+ —— r S X
ÂXBA)— ixB YAJA NH BA
2
3
r
SN0 “ V-rA 'SnhXba (2.15)
*A
r
XBH “rH • XCODXBH “bHXBH
(2.16)
r
XBA -rASNHXBA “bAXBA
(2.17)
Further reduction can be achieved by assuming all decay constants equal.
Motivations for this reduction are that mainly net growth rate of importance and that the
decay parameter is difficult to estimate.
Julien et al. (1998) developed reduced-order models for identification and control for
aerobic and anoxic conditions, with ammonia, nitrate and DO as state variables.
DO dynamics are often decoupled from the other ASM 1 equations, for use in
(adaptive) control of DO. An example is the following (Lindberg, 1997). The decoupled
balance equation for DO over a CSTR by assuming quasi-steady-state for the other
components is written as
= D(So,in - So) + kLa(So - So) - rs0 dt u
(2.18)
The conversion term, rSo, (which is the Oxygen Uptake Rate, OUR), is a time-
varying parameter into which other states and parameters are lumped. It is
modelled with a simple, discrete-time black-box model2:
1
rs (k) = OUR(k) = ------—e(k) (2.19)
0
(l-fq-’xi-q"1)
2.3. DYNAMICS OF VARIABLES
It is generally known that activated sludge systems exhibit stiff dynamics, with timescales
ranging from seconds to weeks. This also holds true for activated sludge models based on
ASM1. This section first summarises published material on timescale properties of
ASM1 and subsequently summarises reductions based on simplifying assumptions with
respect to dynamics.
IAWQ Report No. 1 (Henze et al., 1987) distinguishes three groups of variables:
dissolved oxygen So, dissolved components and particulate components. This was
concluded from a timescale estimation employing only the output (no input) transport
There may however be a danger in doing so, as the OUR is also affected by So itself
and by Ss, which can both vary relatively quickly.
terms and consumption (no production) terms of the balance equations. This yielded a
time constant for So in the order of 1 s., for Ss, SND SNH SALK in the order of 1 min, for
XBH, XBA, XS, Xp, and XND in the order of 10 min. These were estimates for time steps
in Euler integration, and then this poses no problems because the resulting estimate is
rather conservative. This analysis however may give misleading results as it may grossly
underestimate the time constants.
Weijers et al. (1995) carried out an analysis based on the Jacobian matrix of a pre-
denitrification system, consisting of one anoxic and one aerobic reactor. The time
constants, computed in a steady-state under typical operating conditions, ranged form 30
s., associated to a very low DO concentration (0.0114 mg/1) in the anoxic reactor, up to
13 days, associated to inert particulate COD. It is noted that the fastest time constant of
30 s. is much larger than 1 s., reported in (Henze et al., 1987). It was observed that at
elevated DO concentrations, DO dynamics were much slower (approximately 10 times as
slow) than at the very low concentration of 0.0014 mg/1. This was observed both in the
aerobic reactor (So =1.2 mg/1) and in the anoxic reactor (if for computation of the
Jacobian matrix So was artificially elevated to 2 mg/1 to be in the same order of
magnitude as the other solute components). These results indicate that stiffness of DO
dynamics is introduced if the DO concentration is much lower than the concentrations of
other components that are involved in the same reactions.
Based on qualitative reasoning, Olsson and Jeppsson (1994) classified cause-effect
relationships between available manipulated variables and measurable variables into
different timescales. The motivation for classification was to achieve decomposition
based on timescales for plant-wide control. Jeppsson (1996) summarised these in an
incidence matrix, distinguishing fast (minutes), medium (hours), and slow (> days)
dynamic influence. The matrix displayed that most outputs are effected by several inputs,
where different variables may act on different timescales, which is caused by the strong
internal couplings in the system. Interactions occur also within a timescale.
Consequently, the authors expect that control of the activated sludge process requires a
multivariable approach.
In a quantitative study of dynamics of ASM1, Steffens et al. (1997) used a procedure
developed by Robertson and Cameron (1997a,b; see section 3.4), to make a state
partitioning. This required association of states to eigenvalues. They classified as fast S s,
Xs, SNỊỊ, SỊ\JD and XND, as medium SỊ, SALK and SNO and as slow XBA? XBH» XỊ and
Xp
Subsequently, they applied singular perturbation to obtain a systematic reduction of
ASM1 by removing fast and slow states. In addition to the eigenvalues, to select
reducible states they applied additional criteria on the relative error introduced by the
reduction (Robertson and Cameron, 1997a). For fast mode reduction, Ss, SNH, SND and
XND were obtained using a relative error of 5% and with a lower bound on the timescale
of interest of 12 h. The fast mode reduced model without these state variables showed a
large error (100%) in SNH however as a result of nonlinearity. For slow mode reduction,
only XBA was obtained as reducible state using a (large) relative error of 10% and with
an upper bound on the timescale of interest of 20 minutes. The slow mode reduced model
(constant XBA) however showed a large error of 40% in SNH after one day.
Consequently, slow mode reduction with XBA in this case was possible only for relatively
short time horizons of approximately 6 h.
2.4. ORDER REDUCTION METHODS
Van Schagen et al. (1995) applied algebraic reduction techniques to reduce a linearized
ASM1 model of a carrousel system for application in LQG control of ammonia and
nitrate with aeration in a carrousel system. No details on the method applied are given
however.
2.5. BLACK-BOX IDENTIFICATION
Lindberg (1997) applied subspace state space identification to identify a fourth-order

linear model from a nonlinear model including 5 reactors based on ASM1 and a settler
model. The linear model was used in multivariable setpoint control of ammonia and
nitrate in a pre-denitrification plant, with external carbon dosage, internal recfrculation
flow rate and the DO setpoint as manipulated variables. The study was to illustrate
application of black-box identification, the nonlinear model being used to generate
artificial data, rather than to perform model reduction. Nevertheless, it also illustrates
how identification may be used to reduce a calibrated ASM1 model when this is
available.
Some case studies of neural network application to model activated sludge systems
have been reported (Su et al., 1992; Côte et al., 1995). Neural networks may also be used
to obtain nonlinear reduced models. By incorporating prior knowledge into the structure
of neural net, a larger validity range may be obtained. Identification of ANN’S for
modelling of a phosphorous removal activated sludge plant model was unsuccessful, even
for interpolation only (Vanrolleghem, personal communication). No other cases for
wastewater treatment plant modelling however are known to the author.
2.6. DISCUSSION AND CONCLUSIONS
The literature review shows that several approaches are applied in ASM1 model
reduction. In almost all reported cases, the reduction was applied to lumped systems
without concentration gradients, assuming the system to be either aerobic or anoxic.
Thus, the reduction concentrated on reduction of the reaction kinetics, rather than
reduction of transport and mixing in the reactor.
Of the approaches applied to model reduction of ASM1, simplifying assumptions
have been applied most frequently. These assumptions lead to a variety of reduced
models, depending upon the goals and system for which the model is intended. Proposed
models range from simple black-box zero-order kinetic models via grey-box models with
different Monod terms both neglecting biomass dynamics, to the more complicated
model of Jeppsson, which includes biomass growth and decay. The first category is only
valid over a short time horizon and within the domain of experimental conditions. The
latter category may be used for process optimisation over a longer time horizon, and
might be to some extent valid beyond the domain of experimental conditions.
The selection of simplifying assumptions is generally guided by the type of the
treatment system, the input/output relations to be modelled and the time horizon of the
model prediction, in a heuristic fashion. In several cases, the accuracy of the reduced
models is not tested. From the set of reduced models, no single best reduced model can
be selected. A more systematic investigation of the validity and implications of the
various simplifying assumptions under different conditions (system, goals, input/outputs)
would facilitate a more straightforward selection of simplifying assumptions and
resulting reduced models.
Assumptions with respect to dynamics are usually applied to dynamically decouple (some
of) the system equations, especially for application in hierarchical control. A thorough
understanding of the timescale properties of the model is requừed to value validity and
implications of reduced models based on these assumptions. Still however, little
systematic efforts have been done to analyse and understand these properties. The
biokinetic model itself is complicated, describing several biological and chemical
processes that are strongly coupled. Furthermore, it is used to describe relatively
complicated process configurations. Consequently, the dynamics of the model are not
fully understood.
A disadvantage of the order reduction techniques and linear identification techniques
is that the validity of the linear model thus obtained is limited. It is concluded that there is
a need for more systematic order reduction methodology for activated sludge models.
Moreover, more insight in dynamics of ASM1 is desừed. In section 3, for these reasons,
singular perturbation is studied in further detail.
3. Singular perturbation of bioprocess systems: theory and review
Singular perturbation application to bioprocess systems is reviewed in this section. This

method is a promising candidate to be part of a systematic methodology to derive reduced
nonlinear models for different timescales, e.g. for application in hierarchical control, as
was discussed in section 1.2.
In 3.1, that part of the theory is presented that is requừed to understand the material in
sections 3 and 4. Section 3.2 reviews application of singular perturbation to bioprocess
models. Section 3.3 presents a procedure to obtain the standard form that is requừed for
application of singular perturbation for timescale separation. This procedure provides the
basis for developing the procedures for state partitioning in section 4. Section 3.4
discusses some methods that can be considered complementary to scaling in detecting
fast and slow states.
3.1. SINGULAR PERTURBATION THEORY
Let the system be described by n+m equations in state-space notation
X = f(x,z,u,t,e), x(t0) = x0,xe Rn , ueRp (3.1)
ez = g(x,z,u,t,e), z(t0) = z0,ze Rm. (3.2)
with £>0 a small scalar. Then for £—>0 the order reduces to n, because substituting a
root Zi =ội(x,ũ,t) of the equation 0 = g(x,z,u,t,0) into (3.1) yields a reduced model:
X = f (x, ội (x, u, t),u, t,0) - f (x, u, t), x(t 0) = x0, which describes the slow (3.3)
dynamics of the system, also referred to as the outer system or outer layer, or
quasi steady state (x refers to the quasi-steady-state). Model (3.1), (3.2) is said to be in
the so-called standard form if and only if the following crucial condition is satisfied.
Condition 3.1:
In a domain of interest, the equation 0 = g(x, z, u, t,0) has k>l distinct real roots Zj
=ội(x,ũ, t),i = l,2,..,k.
The quasi steady state x(t) can be prescribed to start from Xo and thus be a uniform
approximation of x(t), that is, x(t) =x(t) + 0(e) holds for all te [to, te], including to. The
quasi steady state z(t) however is not free to start from a prescribed initial condition, and
the approximation z(t)=z(t) + 0(e) can be expected to hold only on an interval excluding
to, that is, for te [8, te], with 8>to. During an initial interval [to,8] (the so-called boundary
layer), the original variable z approachesz. The substitution t f =t/e (“stretching” the initial
time) converts (3.1), (3.2) to a set of equations describing the fast dynamics of the
system, the so-called boundary layer or inner system or inner layer (3.4).
dz __
= g(x0, z(tf +z(to),O, to),z(O) = zo - z(t0) (3.4)
dtf
The solution to this problem provides a boundary layer correction term z = z - z which is
used in a possible approximation z= z(t) + z(tf)+0(8), valid for te[to, tj, The important
Tikhonov’s theorem with respect to the boundary layer system states that (3.3) is a valid
approximation of (3.1), (3.2) for all te [to, te] if the following two strong stability
conditions on the boundary layer system are satisfied.
Condition 3.2: The equilibrium z(tf) = 0 of the boundary system (3.4) is asymptotically
stable uniformly in Xo and to, and z(0) = z 0-z(t0) belongs to its domain of attraction, so
z(tf) exists for tf>0.
Condition 3.3: The eigenvalues of 3g / 3z evaluated, for 8 = 0, along x(t),z(t), have

strictly negative real parts, i.e. Re X{3g/3z}< c <0.
Thus, singular perturbation theory allows US to treat slow and fast dynamics
separately. Equations (3.3), (3.4) provide a zero-order approximation of the system
behaviour in the slow and fast timescales respectively, which is exact for 8=0. Higher
order approximations are required to perform a formal error analysis and to obtain more
accurate reduced models as 8>0. They are obtained through series expansion of the state
variables in powers of the perturbation parameter. Two possible procedures are to apply
the matching procedure (Kokotovic et al., 1986) or the boundary function method
(Vasil’eva et al., 1995). A further treatment of approximations is beyond the scope of this
paper.
3.2. REVIEW OF MODEL REDUCTION OF (BIO)PROCESS SYSTEMS USING

SINGULAR PERTURBATIONAL APPROACH
In this section, first a rule proposed by Bastin and Dochain (1990) is discussed. It is
shown that this rule is not generally valid. Other references are therefore investigated to
find a general rule for reduction of bioprocess systems through singular perturbation.
3.2.1. General rule for order reduction
Bastin and Dochain (1990) propose a simple, general rule for order reduction in theừ
book on estimation and control of bioreactors. Given the balance for component
Ặ = z (±)kij<pj - Dî - Qi + F1 ’ (3.5)

at j-ti
(with kij the stoichiometric coefficients, <Pj the reaction rates, D the dilution rate, Fj the
inflow and Qi the gaseous outflow (if applicable) of component i) for a continuous,
ideally mixed bioreactor, the simplification is achieved by setting Jji and dî/dt to zero
which yields the algebraic equation:
22(±)kijcPj =Qi -Fi . (3.6)
This rule is not general however for several reasons. Fkstly, it is not indicated in a general
sense in which cases the dynamics of a component can be neglected, that is, which
component i can be assumed fast, and when and why can be assumed zero. The general
rule is motivated with two specific situations only, which are briefly discussed below,
namely 1. neglect of product dynamics for volatile products with low solubility, and 2.
neglect of substrate dynamics in a model with biomass and substrate. These examples do
however not cover nor explain all possible situations where multiple timescales allow
reduction.
Secondly, the rule is defined for a single, isolated equation. In the more general case
however, several equations are coupled, e.g. via the reaction network, in which case the
proposed general rule cannot be applied in its simple form. In those cases, the reduction
procedure is much more involved, as will be discussed in section 3.2.2.
Singular perturbation technique for products. This case considers product formation in
the following reaction scheme involving one substtate and one product:
s>p
When the product is volatile and has low solubility, its concenừation remains relatively
low. In this case, the saturation constant can be chosen as the perturbation parameter.
Writing the product concentration p as npsat, 0<n<l, with psat the saturation constant, with
£=psat, gives the standard form:
-7- = (p-DS + DSin dt (3.7)
e— = k© - eD n - Q dt
(3.8)
with Q the rate of mass removal in gaseous form. For £—>0, a reduced model is
obtained by substituting the resulting algebraic equation, k<p=Q into (3.7):
= k'Q-DS + DSin- dt (3.9)
In this case, the rule holds (for psat sufficiently small).
Singular perturbation technique for substrate. For the following simple microbial
(autocatalytic) growth process (See also section 4.2):
s —Ox
the component balance equations are written as:
— (3.10)
DX dt
= - ki |1X - D s + D Si, (3.11)
Multiplying by V and using V as a perturbation parameter gives (with VD Sin= Fg. and
XT=VX):
dXT _ _ _____
^r=pXT-DXT (3.12)
e-y- = - ki |1XT - eD s + FSjn which for eô, with kỉ |UXT = Fj , reduces to: (3.13)
dXT _ 1_
-^ = -DXT+ -=-FC . Sta
dt k]
In this derivation, the authors state that it should be understood that the volume (3.14)
is not assumed to be zero, but small enough to neglect e(dS/dt) and eDS.
Therefore, it is considered legitimate to divide the reduced equation by V again
to obtain:
^- = -DX+ T-DSin.in
dt ki (3.15)
However, it can be argued the volume is not the adequate perturbation

parameter in this case because:
• It is not the volume that makes the term EDS go to zero; in fact, the dilution rate
D=Fv/V (with Fv the volumetric flow rate) becomes very large for small e and the
term reads FVS.
• Without any additional information, there is no reason to select the substtate as
the fast state; in fact, the same argumentation can be applied to neglect the
biomass dynamics.
• The physical reasoning is not sound, as smallness of the volume is not the cause
of multiple timescales. Consequently, the derivation is mathematically not
consistent.
In section 4, it will be shown that in some cases the ratio of the biomass and substrate
concentration can be used as a perturbation parameter; procedures that are more
systematic are developed there.
3.2.2. Other reduction approaches

Van Breusegem and Bastin (1991) investigated model reduction of reaction systems with
more general reaction networks, in the case of one (ideally mixed) reactor. The reduction
problem was addressed under the assumption that some reactions are faster than others.
As a perturbation parameter, the average of the fast rate constants was used. In natural
coordinates, the problem is not always in standard form however, as Condition 3.1,
existence of distinct roots, is not guaranteed. A change of coordinates was suggested to
bring the problem into standard form. As the state transformation is non-singular, the
reduced model can be transformed back into natural coordinates. In natural coordinates,
however, there is generally no dừect partition in fast and slow variables and the singular
perturbation cannot be applied to the original, untransformed system, which we however
like to preserve the interpretation of states. The procedure was successfully tested on a
two-step enzyme reaction (Van Breusegem and Bastin, 1991) and on a two-step model of
a methanogenesis process (Van Breusegem and Bastin, 1992)
Vasil’eva et al. (1995) discern as the critical case that class of applied problems in which
the condition of isolated, distinct roots of the reduced equation 0 = g(x,z,u,t,0)is not
satisfied. In fact, in many problems in chemical kinetics the condition is not satisfied. A
procedure is presented using the boundary function method for constructing approximate
solutions based on asymptotic expansions for the standard form. A modification of this
procedure is presented to deal with this critical case and applied to chemical reaction
networks (in batch reaction), where small and fast reactions are involved. As a
perturbation parameter, the smallest of the large rate constants was selected. It is
interesting to note that this modified procedure does not start from nor produces the
standard form. It is noted that this case again shows that a separation in fast and slow
reactions does not correspond to a clear partitioning into slow and fast variables. Instead,
it leads to a set of simplified systems for the fast and slow timescales.
Weiss and Preisig (1997) studied simplifying assumptions in the process of modelling
composite process systems. Very large transfer coefficients between subsystems or very
small capacities of subsystems for example can lead to lumping of the transferred
intensive variables in the concerning subsystems. The difficulties and accuracy in this
type of simplifications were studied on a relatively simple yet generic example of n tanks
with various levels and temperatures, assuming that some of the transfers are very quick.
The modified procedure of Vasil’eva et al. was applied (in fact, the systems studied are
very similar, with fast and slow transfers instead of fast and slow reactions) to reduce the
system and to estimate the resulting error. Reduction of heat balances to describe the
temperatures in the tanks was much more involved and less accurate than reduction of
total mass balances to describe the heights. The singular perturbation parameter in this
case is the inverse magnitude of the fast transfers relative to the slow ttansfers.
The references reviewed above all apply to lumped systems. Dochain (1994) applied
singular perturbation to reduce an (infinite-dimensional) distributed parameter bioprocess
system model to a (finite-dimensional) lumped parameter model. The general distributed
parameter system studied is a plug-flow reactor with dispersion. For the reduction, the
system was rewritten using dimensionless numbers, employing as perturbation parameter
the mass Peclet number, which expresses the ratio of the residence time and characteristic
time for dispersion. A first-order approximation using a series expansion of the solution
and using matching to determine initial conditions of inner and outer solutions yielded a
lumped reduced model.
To apply singular perturbation for order reduction into fast or slow states of a given
model, the model must be in standard form and a state partitioning must be made.
A general rule for reduction of continuous bioprocess systems was given by Bastin
and Dochain (1990). However, this rule is not generally valid, and is not helpful in
general in deciding which parameters can be considered fast (only products with very low
solubility).
Other procedures have been derived for reduction of models in non-standard form.
For reduction of models in natural (original) coordinates, a procedure exists for the so-
called critical case (no distinct roots of the quasi-steady-state). However, the discussed
references do not help to recognise or obtain the standard form. A procedure that does
this is summarised below.
3.3. SCALING IN MODEL REDUCTION

3.3.1. A method for state partitioning based on scaling
Aim of this section is to illustrate the usefulness of a systematic scaling procedure for
state partitioning, to recognise or obtain the standard form. Frequently, the model to be
reduced using is in non-standard form and must be brought into the standard form. One
way to do so is to select or find a suitable, small perturbation parameter. This is often
nontrivial and requừes considerable physical knowledge; smallness of a parameter alone
is not a sufficient condition. Selection of a perturbation parameter is associated with the
issue of scaling, as smallness is a relative notion, and scaling is helpful to bring the model
in the standard form and to select a suitable perturbation parameter. Kokotovic et al.
(1986) presents several approaches in scaling, including the use of dimensionless
parameters, parameter scaling and state scaling. However, no systematic, straightforward
procedure for scaling is given.
Heineken et al. (1967) applied singular perturbation theory to provide a mathemati-
cally sound analysis of the quasi-steady-state assumption (QSSA) in derivation of
Michaelis-Menten enzyme kinetics. Segel and Slemrod (1989) used a systematic
approach based on scaling to obtain nondimensionalised equations in which it is easier to
reveal the relative magnitude of parameters. With the applied scaling procedure, they
were able to refine the conditions for application of the QSSA. The procedure they
applied is outlined as follows and how they applied it to Michaelis-Menten kinetics is
summarised in section 3.3.2:
• Estimate slow and fast timescales Tf and Ts.

• Test on the necessary condition for the QSSA on the timescales.
• Test on the necessary condition on the smallness of the error in the slow state during
the pre-steady-state.
• For both timescales, choose state scaling and subsequently derive scaled equations;
this step should yield the perturbation parameter.
• Reduce the model.
• Test the reduced model, e.g. through numerical simulations.
The tests on the necessary conditions may also provide perturbation parameter
candidates. In theừ paper, in addition to these steps a formal analysis was performed
using approximate solutions of the scaled equations. Both the derivations of the
approximate solutions and a formal error analysis are beyond the scope of this paper.
Application of an analytic scaling procedure to Michaelis-Menten kinetics. In the

reaction model employed in derivation of Michaelis-Menten kinetics, an enzyme E
combines with substrate to form an ES complex, which dissociates to either E and s or to
E and product P:
k k
l 2
E + s c E + p ki
With the initial conditions E(O)=E o, S(O)=So, C(0)=0, P(0)=0, and using the fact that E(t)
+C(t)=E0, the following basic mathematical problem results:
= -ki (Eo - c)s + k-iC, S(O)=So, (3.16)
dt
= ki (Eo - c)s - (k_i + k2 )C ,C(0)=0. dt

(3.17)
Application of the QSSA to simplify this system is standard and can be found in
any elementary textbook on biochemistry. The QSSA implies that after a short pre-
steadystate, the complex concenttation c is approximately constant and the decrease of
the concentration s due to complex formation can be neglected as Co is small. The
derivation is not repeated here. Only the main points are indicated.
We now focus on the procedure of Segel and Slemrod (1989). The steps to apply the
procedure to Michaelis-Menten kinetics were as follows.
(a) Estimation of timescales:
• The fast timescale is associated with the complex formation. To estimate the fast
timescale Tf, the approximate analytical solution of (3.17) for the fast timescale is
used by supposing s is slow and can be approximated by so:
C(t) = c[l - exp(-kt)], (3.18)
with:
k
= MSo + Km) and KmS(k-1+k2). (3.19)
K
m+50 KỊ
and we obtain
Tf=k-1. (3.20)
The slow timescale is associated with the substrate after the pre-steady-state and is
estimated using the following characterisation of a timescale (Here, Segel and
Slemrod (1989) refer to Segel (1984), p.56):
dS
~ (Smax Smin
Here, the maximum and minimum concentration of s are S max=S0 and Smin=0- The
maximum rate after the pre-steady-state is estimated by substituting s=s o and (3.19) into
(3.16), which gives:
Km +so s= k2E0
(3.21)
Test on the necessary condition for the QSSA on the timescales.
A necessary condition for the QSSA to hold is that the duration of the pre-steady state is
much shorter than the characteristic time for substrate change, Tf « Ts, which gives:
T| « (1 4- K)(l + Ơ)2 , (3.22)
with the dimensionless parameters
G= KS JSZL. (3.23)
Km Km k2
Test on the necessary condition on the smallness of the error in the slow state during the
pre-steady-state.
For the change in the substrate concentration to be negligible, the relative concentration
change AS/So must be very small during the pre-steady state, which is estimated by:
Using (3.16) with c=o to determine |dS/dt|max yields the following additional condition in
dimensional variables, which is stronger than (3.22):
ĩ|« (1+ơ). (3.25)
The result also indicates a perturbation parameter candidate, namely 77/(1 +

ơ).
For both timescales, derive scaled equations.
In the pre-steady-state, the time is scaled by Tf. The substrate concentration is scaled by
So and the complex concentration c by the maximal complex concentration
C.With the introduction of the following
dimensionless variables into (3.16) and (3.17),
(3.
(Z5
III
III -p?
III O|IO
26)
we obtain the scaled equations for the fast timescale:
ds Ơ K(K + 1) 1 , (3.
-7— = e - s d-------cs d-------—c , s(0)=l
dtf ơ d-1 Ơ d-1 27)
de Ơ 1 „
-— = s----------cs----------c , c(0)=0. (3.
dtf ơ+1 Ơ+1 28)
After the pre-steady-state, the time is scaled by Ts. The same state scaling then yields:
ds . , ... .. Ơ . K(Kd-l) 1
-7— = (K + l)(ơ + l) - s + OS + ———-—c (3.
dts Ơ +1 Ơ +1
29)
de , . ... .. Ơ 1
= (Kd-l)(ơd-l) s-------—— cs--- c . (3.
dts -
Ơ d1 Ơ d-1 30)
Thus, we now have the problem in standard form, and have obtained a perturbation
parameter:
c= n _ EQ (3.31)
1 + Ơ Km+So
This result is more accurate than the traditional perturbation parameter, Eo/Km. Segel
and Slemrod (1989) obtained this more accurate result through applying this
systematic physical scaling procedure. Besides physical scaling, Segel (1972) also
applied mathematical scaling procedures with the aim to test whether this result can
also be obtained with less prior knowledge. Through one such method, minimal
simplification, the same results were obtained; it is not clear however whether this
method will always work and the method is not further studied here.
Reduce model and test the reduced model
From the standard form, the model was reduced. In numerical simulations, the
conditions on the timescales and errors were verified and confirmed to be correct.
3.3.2. Relationship of scaling with regime analysis and dimensional analysis
In (bio)process engineering, regime-analysis is a tool to detect rate-limiting mechanisms
by comparing the timescales for different transport and reaction processes. For example,
Oosterhuis (1983) applied regime analysis to the scale-up of bioreactors. Evidently, there
is a relationship with the detection of occurrence of multiple timescales as described
above. An important difference is that regime analysis compares the relative size of the
individual terms in the balance equations, while the timescale estimation, as applied by
Segel and Slemrod above, considers the total rate expression of extensive variables. Both
methods can be used in a complementary fashion. Dimensionless groups, obtained by the
ratios of characteristic scales of different phenomena, also provide insight into the
system. The insight obtained by using scaled equations is a strong argument for the use of
scaling to convert the equations to dimensionless form, as advocated by Segel (1972),
even if this scaling requừes some physical reasoning.
3.4. OTHER METHODS FOR TIMESCALE ANALYSIS
In section 3.3, a procedure was described for detection of multiple timescales in models
and transformation of the problem to the standard reducible form. Robertson and
Cameron (1997a) presented another approach to detect timescale multiplicity and for
state partitioning for use with singular perturbation reduction. Their procedure might be
suitable for large systems if the scaling procedure described above is too laborious, or
could be used as a complementary technique. The essential issues of the procedure are
briefly summarised here.
Starting from the definition of a timescale of interest, slow and fast modes are
detected using a linearized model. The timescale of interest is determined by the intended
application, e.g. control of ammonia and nitrate in activated sludge plants or control of
the sludge concentration, which requfre different timescales. The first requfres medium
timescales, the latter slow timescales. Fast mode reduction is performed based on an
eigenvalue-to-state association. This is done using a so-called homotopy parameter. The
procedure starts with a system in which the eigenvalue-to- state association is known (in
thefr study, a matrix with only the diagonal of the Jacobian matrix was used). The
eigenvalues are traced when going from this system to the system for which the
eigenvalue-to-state association must be determined (the full Jacobian matrix). A so-called
homotopy parameter is used to gradually change from the known system (homotopy
parameter is zero) to the unknown system (homotopy parameter equal to one).
If a group of fast time constants exists, then the states associated with these time
constants are candidates for reduction. Slow mode reduction proceeds via Taylor
expansion of the free response. From linear systems, randomly generated by Monte-
Carlo sampling, empfrical relationships for the maximum relative error both for fast and
slow mode reducibility were derived. After detection of slow and fast states, these are
removed from the nonlinear model. The procedure provides no guarantee that the system
is in the standard form and no formal error analysis is given. Tests on an evaporator
system and compressor start-up yielded good results; results on ASM1 were less
successful as discussed in section 2.3.
We indicate two disadvantages of the slow mode detection method. Error estimation can
be too optimistic if in reality a forced response is dominant, because only the free
response is considered in the selection criterion. Furthermore, relatively small changes of
a slow mode variable can be associated with relatively large changes in other variables,
especially if these have small absolute value (such as dissolved oxygen or ammonia in the
case of ASM1). Consequently, on the other hand, the method can be too conservative.
Wasynczuk and Decarlo (1981) suggested an approach for reduction of composite
models. This method is useful for order reduction of large-scale systems such as in plant-
wide control, and is therefore briefly described in here. The procedure indicates whether
model reduction on a component level is possible or not. This must be checked, because
interaction may inttoduce dynamics in the timescale of subsystem dynamics. It this is the
case, then the model reduction cannot be performed simply by reduction of the individual
subsystems only.
In the analysis, the so-called component connection model is employed, which
separately defines the subsystem equations and the interconnection matrix. The procedure
consists of eigenvalue tracking applying a homotopy parameter, which is multiplied by
the interconnection matrix, and is used to vary from completely decoupled subsystems to
the fully interconnected system. This indicates whether reduction on a subsystem level is
sufficient to reduce the complete system.
3.5. CONCLUSIONS
Application of singular perturbation for reduction of (bio)process models was reviewed.

Bastin and Dochain (1990) suggested a simple rule for order reduction, which is not
generally applicable, however. In reduction of a system with biomass and substrate, the
suggested selection of the perturbation parameter was not satisfactory. Segel and Slemrod
(1989) applied a methodology based on scaling to formally derive Michaelis- Menten
kinetics. The methodology can support detection of multiple timescales, which are
necessary for applying quasi-steady state assumptions and to order reduction with
singular perturbation. Furthermore, it may help to bring the model in the standard form
and to select a suitable perturbation parameter that is requfred for application of the
singular perturbation method. The methodology forms the basis to develop a method to
recognise or obtain the standard form and for state partitioning. This is applied to a
simple yet basic and important bioprocess with biomass and substrate, which is similar to
the model studied by Bastin and Dochain (1990), in the next section.
4. Scaling for singular perturbation in a simple bioprocess system
4.1. INTRODUCTION AND METHODOLOGY
This section investigates application of singular perturbation to reduction of bioprocess

models. The following observations motivate this further investigation. Section 2 made
clear that more insight into the timescale properties of ASM1 is required. Section 3
showed that only a few occasions have been described in which there is a clear
perturbation parameter and associated timescale separation, namely the case of low
solubility of volatile products and the occuưence of fast and slow reactions. In ASM1
however, these are not the predominating causes of multiple timescales, as no volatile
products are applied in the model and the reaction rates, although different, do not differ
orders of magnitude. Therefore, we want to find additional causes of multiple timescales
and associated perturbation parameters. Moreover, these findings are also of relevance for
bioprocess systems in general.
The following strategy is adopted. Three procedures for state partitioning are
proposed. They are applied to a very simple but basic and important model system,
closely resembling the system used by Bastin and Dochain (1990) (Section 3.2). This
should reveal to reveal timescale multiplicity and the physical conditions that cause
multiple timescales. These conditions are related to a singular perturbation parameter, if
possible, and it is checked whether the model can be brought into standard form.
The three procedures to obtain or recognise the standard form are evaluated with
respect to straightforwardness and requừed prior knowledge. The following procedures
are applied:
• Dừect scaling
• Timescale estimation for variables
• Analytical scaling procedure
These procedures are now described and are applied to the model system described in
section 4.2.
4.1.1. Procedure 1: Direct scaling

This procedure attempts to bring the model into standard form dừectly, without a
thorough scaling procedure and without introducing dimensionless parameters. The
background to apply this procedure here is the conjecture that stiffness is associated with
large concentration ratios. For example, in analysis of DO dynamics, it appeared that a
fast timescale occurred only if the DO concentration was very low. Moreover, in
wastewater treatment systems the suspended components are usually regarded as slow,
and (most of) these variables have a much larger concentration than the suspended
components. From these observations, it was hoped that a simple yet physically
meaningful rule for order reduction could be derived.
4.1.2. Procedure 2: Timescale estimation for variables

Procedure 2 is a systematic procedure for detection of timescale multiplicity through
estimation of timescales of variables. This procedure is proposed as an alternative and
complementary procedure where Procedure 3 fails.
In the example in section 3.3, it was already known that multiple timescales occurred,
and to which variables the fast respectively the slow timescales were associated.
Generally, this is not the case; in fact, the eigenvalue to state association procedure
presented in section 3.4 was developed for this purpose.
Here we will propose a different procedure to estimate timescales of variables. This
procedure is a modification and extension of the procedure of Segel and Slemrod (1989)
in section 3.3, and does not requừe analytical timescale or error estimates. Instead, a
mixed numerical I analytic approach is applied
to detect timescale multiplicity and to estimate
the error in slow states during the initial layer,
thus checking whether the quasi steady state
approximation applies.
We start with the system (4.1)
C = f(C,u,t,e), C(t0) = C0,CeRn+m, ue Rp (4.1)
and try to find a partitioning into fast and slow variables C T = [Cf c J ] so that we can
write (4.1) in standard form as
Cs = fs (Cs, cf, u, t, 8), Cs (t0) = Cs0,c G Rn , u G Rp , (4.2)
8Cf = ff (Cs,Cf ,u,t,8), Cf(t0) = Cf0,CGRm. (4.3)
The procedure consists of the following steps that are subsequently discussed below:
• Step 1. Estimate initial timescales of all variables and select the fast variables.
• Step 2. Estimate timescales of slow variables in the outer layer (in quasi steady
state).
• Step 3. Check the multiplicity of timescales.
• Step 4. Estimate the error of the slow variables.
• Step 5. Reduce model if preceding steps indicate that the QSSA applies.
Stepl: The initial timescale Tf i for variable i in Step 1. is estimated with:
| -C°|
C
i0
d (4.4)
max,
Cj
in
dt cP the quasi steady state value of variable i with the other variables Cj, jî at thek
with
T
fi =
initial values Cjo and |dCi/dt| maXjin the maximal rate during the inner layer, which in this
paper is evaluated with all state variables at thek initial conditions. We would like to
avoid the need to compute cP. Therefore, one can try to rewrite (4.3) to eliminate I Cị (0)
~ cP I as follows. In the quasi steady state for cP we have:
O = fi(Cjo,>i,Ci°,u,t,£) . (4.5)
Subtracting equation (4.5) from the rate equation for Ci (this is allowed, because the
quasi steady state equation is zero) gives (4.6) (Note: |.| denotes the absolute value).
dCj —
|fi(CjO,j?H’CiO>u>l>e) ,u,t,£)| (4.6)
dt max,in
We can try to write this as a product with a term (Ci0 - Cj°):

todCj
obtain: = |gi (C jo, jyi, Ci0, Ci°, u, t, e) • (Ci0 - Ci°
max,in )|
|ci0 -c?|
Tfi = I-----------------------L—---------!Ĩ
|gi (Cj0,jî ’ Cio > Cj >u> t>£) ■ (CjQ — Cj )|
- I gi (C jo,jî, cio, Cị°, u, t, £) I (4.7)
In some cases, it is thus possible to eliminate Cị° completely and to obtain an

expression for the time constant as a function of the parameters and initial conditions
only. The variables that have a (much) faster associated timescale than other variables are
selected as the fast variables, and thus the partitioning is made.
Step 2. The slow timescale Tsi for the slow variables i can be estimated with:
Cjo ” Ci
(4.8)
dCj
max,out
with cio the initial value of variable i, Cj its steady state value and |dCi/dt| maX5oUt the
maximal rate during the outer layer. To avoid the necessity of computing Cj, it is
assumed that the difference between initial value and steady state are in the same order of
magnitude as the initial value and that (4.8) can be approximated by (4.9).
(4.9)
dCj
dt max,out
The maximal rate during the outer layer is evaluated with the slow state variables at theừ
initial values Cjo and the fast states at theừ quasi steady state values.
Step 3. is straightforward, unless the difference between fast and slow timescales is not
very large. Here, we will pragmatically consider a factor of about 10 between largest fast
timescale and smallest slow timescale sufficient for timescale separation.
Step 4. The error in the slow variables during the initial, inner layer is estimated to check
whether the error is small enough to qualify the slow variables as slow. If this is not the
case, this does not mean that there are no multiple timescales, but rather that they cannot
be assigned to disjunct fast and slow variables, i.e. the problem is not in standard form. A
first, conservative approximation employs the maximal rate during the inner layer, with
the advantage of computational simplicity as all requừed quantities are known already.
The condition then is:

A the largest
with Tf 1 d of the small time constants and |dC/dt| max>in the maximal rate during
C C •Tf (4.10)
the inner
C layer.
"Co The estimation
d max,in is more accurate and less conservative if an average rate is
applied:
o t
AC 1 dC
(4.11)
Co co dt avin
In the sequel, the average rate computed as the mean of the initial rate and the rate after
the inner layer at t=Tf will be used to compute the error, unless stated otherwise.
Step 5. Reduce model if preceding steps indicate that the QSSA applies. If the conditions
checked in steps 3. and 4. are satisfied, then this indicates that the quasi steady state
approximation applies and the model may be reduced. One can try to formulate scaled
equations and select a perturbation parameter and thus bring the problem into standard
form. If this is not successful, then we can proceed as follows. From the partitioning into
slow and fast variables, it can be simply concluded that the rate for the fast variable is
much higher than the rate for the slow variable and therefore can be written as:
ff =—hf (Cs,Cf ,u,t,e),

£
where hf (Cs,Cf ,u,t,£) is in the same order of magnitude as f s(Cs,Cf ,u,t,£), because ff
(Cs,Cf ,u,t,£) is much higher than f s(Cs,Cf ,u,t,£). Therefore, (4.1) can be written as (4.2),
(4.12):
Cs=fs(Cs,Cf,u,t,£), Cs(t0) = Cs0,CeRn, UGRP, (4.2)
Cf =7gf(Cs,Cf,u,t,e), Cf(t0) = Cf0,CeRm, (4.12)

£ and it is seen that the system is almost in standard form. Although we do not
have £ in analytic form, we know it is small enough (through the preceding steps) and we
may apply the QSSA to obtain the quasi steady state value for Cf by equation (4.5). An
error estimate is also provided through step 4.
4.1.3. Procedure 3: Analytical scaling procedure

In Procedure 3, the methodology proposed by Segel and Slemrod (1989) (Section 3.3) is
applied. This procedure proceeds via scaling and selection of a perturbation parameter.
For a complete scaling procedure, analytic timescale estimates are requừed for scaling the
fast and slow timescale and analytic error estimates are derived.
4.2. MODEL SYSTEM: CHEMOSTAT WITH BIOMASS AND SUBSTRATE
4.2.1. Model system definitions and analysis

A simple chemostat with one biomass and one substrate will be used to test the proposed
procedure. In this subsection, the system is described, some analytical relationships are
presented and the cases that are discussed in the next subsections are indicated.
In a completely stirred tank reactor, biomass X grows on a single substrate s (the
arrow in the formula denotes an autocatalytic reaction). Motivation for choosing this
system is that this is a very simple yet basic and important continuous biotechnological
process that can exhibit multiple timescales. This simple system is well understood and is
a good study object as better understanding its timescale behaviour is helpful to
understand systems that are more complicated as well.
s -Cx
The reactor is schematically shown in figure 1.

F, Sin
V, s,x
Fig. 1: Chemostat with biomass growth on one substrate
For this system, assuming Monod kinetics, model equations (4.13), (4.14) can be written.
s
x=p—X-DX (4.13)
K+s
s= -ki p—X -DS + DSin K + S (4.14)
with s the substrate concentration,, Sin the substrate concentration in the feed, X
the biomass concentration, D the dilution rate =V/F, ki the reciprocal of the yield, p the
maximum specific growth rate, K the Monod constant, The dimensionless parameters
defined for this system are the dimensionless residence time and the Monod number:
* JU
T = £- (4.15)
D
K Mo = -— (4.16)
Sin
For the system to be viable, no washout must occur so there is a lower limit to the
dimensionless residence time:
Train = —-H—= 1 + Mo (4.17)

Dmax
The steady state concentrations are given by (4.18) and (4.19).
1 DK 1 K
Xoo ^{Sin = (4.18)
kj (JI-D) ki X -1
s - DK - K
00
(JI-D) T*-1 (4.19)
Upon linearization in a state x={X, SJwith U=D, the following linearized system is
obtained:
J^--D
K+S
-kljS- (4.20)
K+S
from which the eigenvalues are computed:
„ ~ S(S + K)-kjXK
\2 = -D, -D + n v 1------------------------------------------- (4.21)
(S + K)2
Starting point in the subsequent timescale analysis is the conjecture that occurrence of
multiple timescales is associated with a large concentration difference between the states.
The ratio s/x in steady state is written as a function of the dimensionless parameters T
and Mo:
Soo 1 Mo kj T*-1-
(4.22)
MO J
From (4.22), we see that s/x is low when Mo is very small or T* very large and six cases
are distinguished as indicated in table 1 and briefly explained below.
Case 1 and 2: When T* is very large (high residence time), substrate conversion is
almost complete and biomass produced is only slowly withdrawn. At high feed substrate
concentration Sin (small Monod, Case 1), the concentration difference is larger at a
given residence time than at lower Sin (Monod ~ 1), because the biomass concentration
is higher whilst the substrate concentration remains unchanged (Case 2). This is easily
seen from (4.19).
Table 1: Cases distinguished in timescale analysis; for explanation, see text
Case Ratio s„/x„ Condition for T* Other condition Section

Mo « 1 4.2.2
2 s«x T*»l Mo~ 1 4.2.2
3 s <x 1* > Mo « 1 4.2.3

< 0(2/
4 s~x T > 0(3/ Mo~ 1 4.2.4
5 s=x *
T TE~ MO, MO «1 4.2.5
=1+Mo
6 s >x +TE *
TE« MO 4.2.6
> =1+Mo
+TE
Table 2: Ratio So/XooOnd ratio of eigenvalues for different cases

1 s«x T*»l
À4/X
Case *
Mo Soo/Xoo Xi X2
2
1 20 0.04 0.0032 0.0022 0.2 90
2 20 0.4 0.0323 0.0226 0.2 8.8
3 3 0.04 0.0306 0.0306 1.3 43.5
3
4 3 0.4 0.375 0.375 1.3 3.56
3
5 1.244 0.04 0.294 1 3.2 3.21
1
6 1.044 0.04 15 236 3.8 0.016
3
7
: Uppercase o denotes “order of magnitude
Case 3: At moderate dilution rate, but far from washout, which is expressed by the
condition T* > 0(2), the ratio s/x can still be small when the feed concentration is very
high (this is the case when the Monod number is very small).
Case 4: If Sin is moderate at moderate dilution rate, then the ratio s/x will be 0(1).
Case 5: At high dilution rate, relatively close to washout, we write T* as T* =1+Mo+T e.
This is an interesting situation, as biomass productivity is optimal close to washout,
namely for Te= A/MO + MO2 .
Case 6: At still higher dilution rates, when T e« Mo, the situation is so close to washout
that substrate conversion is almost zero and high ratio s/x results.
For typical values representing the different cases, the ratio Soo/Xoo and the ratio of the
eigenvalues are given in table 2. Equations (4.18), (4.19) and (4.21) were used, with the
following parameter values as default parameters: |I=4, K=20, ki=1.5.
The results indicate that the supposed relationship between the ratio Soo/Xoo and the
ratio of eigenvalues holds indeed. In the subsequent subsections, the cases are analysed
more thoroughly according to table 1.
4.2.2. Cases 1 and 2: Ratio s/x very low, Ỉ very large
Direct. With the dfrect approach, we introduce scaled variables dfrectly. In this case,
from the supposed association of multiple timescales with a large concentration
difference between the states, both states are scaled with theừ steady state, assuming that
Soo/Xoo is very small.
. s/s~
X/Xoo00= p__ ..X/Xoo - DX/Xoo (4.23)
^K/Soo+Soo 00
s/s~ = - kj u s °° d-X-DS/Soo + DSin /Soo

/s
(4.24)
K/Soo + ^00 ôo
In dimensionless variables we obtain with k=K/Soo:
X = p —-—X -D X (4.25)
k+s
s=-kt n s -Ds + DSin/Soo (4.26)

k + s XQO SQO
With £ = Soo/Xoo (4.26) becomes:
£S= -ki p;.........-. X -eDs + D ^!2_£ = -ki |X .....................X -ED s + fj (4.27)

k+s Xoo Soo k+s
Now the problem is in standard form, and (4.25), (4.27) are the equations for the slow
timescale. With the substitution tf = £t the fast time equations are obtained:
dx , 0B s _ . dtf k + s
(4.28)
—— = - ki |1—— X -£ D s + fin (4.29)

dtf k+s
Note: To arrive at (4.27), it must be shown that fin is not very small, O(E), but that it is of
the same order of magnitude as the other non-negligible terms and thus cannot be
neglected. The result is given here as is without proof, but is can be shown that this
condition holds.
With the dừect approach, we arrived at the equations in standard form. If the ratio
Soo/Xoo is very small, this ratio can be used as a perturbation parameter to obtain the
problem in standard form. In that case, biomass is the slow variable, substrate the fast
variable.
The requừed physical knowledge to obtain the standard form in this case is the
association of low Soo/Xoo with timescale multiplicity. However, the standard form
obtained presupposes that Soo/Xoo be very small, but does not indicate whether this
assumption indeed is valid. In the next two subsections, an alternative formulation of the
standard form will be derived, which states the condition for timescale multiplicity
dừectly in terms of (dimensionless) system parameters.
Timescale estimation for variables. The procedure described in Section 4.2 is applied to
system (4.13), (4.14).
Step 1. Estimate initial timescales and select fast variables.

For estimation of the initial timescale for biomass X, the quasi steady state equation
(4.30) for X with s=so
0= u s° x°-DX° (4.30)
K + So
and the estimate (4.4) are used to obtain an approximation in the form of (4.7):
— = p—-----------D. (4.31)
Tfx FK + SO 1 7
For substtate s, we similarly obtain the estimate for the initial timescale:
Case * Mo A4/X2 Tfs^fx TfX frs
1 20 0.04 0.002 0.0022 10.3 0.023

2
2 20 0.4 0.022 0.0213 10.3 0.22
6
Table 3 shows that the timescale estimation indicates multiplicity of timescales, s is
selected as the fast state. The results obtained with the timescale estimation correspond
well with the eigenvalue results. The advantage with timescale estimation is that the
timescales are dfrectly associated with variables.
Step 2. Estimate timescales of slow variables in the outer layer (quasi steady state).
As s is selected as the fast variable, X is the slow variable. Applying (4.9), the slow
timescale for X is estimated with by substituting s° obtained from (4.33).
1s
1 °
-—“I
T ”;—^77
sX K + S° (4.34)
Table 4: Slow timescales and error for Case 1 and 2 (x(0)=0.5-x( °°))
Case T* Mo Tfs/TsX TsX Tfs

AXQ/XQ
1 20 0.04 0.0042 5.0 0.021 0.0012
2 20 0.4 0.038 5.2 0.20 0.010
Table 4 shows the results of the slow timescale estimates for Case 1 and Case 2.
Step 3. Check the multiplicity of timescales.

From table 4, it is seen that for both Case 1 and Case 2 the condition Tfs/Tsx«l holds.
Step 4. Estimate the error of the slow variables.

Applying (4.11), the error in the slow state X during the initial layer is estimated by
(4.35) (Xo can be eliminated).
AX li So —- ~ p„„
XQ 2(jK + S0
(4.35)
The errors are given in table 4 and indeed the condition AXo/Xo«l holds. In
Case 2, the error is approximately 1 %. It is concluded that timescale separation can be
applied.
Step 5. Reduce model if preceding steps indicate that the QSSA applies.
From the preceding steps, we know that the QSSA applies because timescale multiplicity
occurs and there is a separation into fast and slow states (see Step 5 in section 4.1.2).
Therefore, the system can be slow-mode reduced straightforwardly to:
s° __ _ __ _____
x = g ............—.X -DX (4.36)
K+S°
with s° the positive root of the quadratic quasi steady state equation for s (4.33).
Analytical scaling procedure. In this subsection, scaling is performed employing

estimates based on the analytical solutions of the eigenvalues of the system, which leads
to the standard form.
Estimate timescales
In Case 1 and 2, the large eigenvalue of (4.21) can be approximated as follows, because
x»s, S«K, P»D (because T*»l) and x~sin/ki:
. -kiXK + S(S + K) _ kiXK _ 11 11

Ầ2 = - D + A ~ -D - p 1 _ = -D - = -■£- (4.37)
(S + K)2 (S+K)2 Mo Mo
As the time constant is reciprocal to the real part of the eigenvalue, the fast time and slow
timescales can be scaled as (4.38) and (4.39) respectively.
tf=t-p/Mo (4.38); ts=t-D (4.39)
Test for QSSA necessary conditions

For the QSSA to be valid, Tf«Ts must hold
Mo/p« 1/D or T*/MO » 1 which is the same condition as was found before (4.40)
and which holds under the assumptions made.
Perform error analysis on initial condition for slow state.

For the fast timescale we have Tf=Mo/p. Assuming that So and Xo are in the same
order of magnitude as Soo and Xoo respectively and applying the (conservative)
estimate (4.10) this is approximated by the estimate (£ is a small number O(MO/T)):

For timescales, choose a state scaling and subsequently derive scaled equations and (fry
AXa perturbation Mo Sn _ Mo So Mo
to) find - p—----------D
parameter.« p-M_-D — p = O(e) (4.41)
The states
xo K + S K p state values, but insteadT of symbols Soo and Xoo,
are scaled with their steady
o Sin
now expressions (4.18) and (4.19) are used to find the approximate steady-state values
Xoo~Sin/ki and SOO~K/T*. Substitution of scaled variables x=X k1/Sin and S=S T7K and
for the slow timescale ts= t-D yields:
dx SK/T*
D-£^= p................ 37...; -Dx (4.42)
dts K + SK/T
^dsK/T* , SK/T* sin _ K

D " kiP T J, J - Ds-^ + DSin (4.43)
dts K + SK/T ki T
Dividing (4.42) and (4.43) by D, multiplying numerator and denominator of the

Monod term by K, using Mo=K/Sin and introducing 1/T* as a perturbation parameter
yield the standard form with the outer equations (4.44) and (4.45).
dx
The substitution s tp/Mo gives the inner equations (4.46) and (4.47).
tf=
-------X -X (4.44)
dts 1 + ES
~ ds 1_1
£—— — X — £s + —- (4.45)
= dts 1 + es Mo Mo
dx z s x
-7— = EMO(——— X - x) dtf 1 + £S (4.46)
-7—=--------—-X-£MO s + 1
dtf 1 + £S (4.47)
Thus the timescale estimation and knowledge of scaling of the variables enables a scaling
of the variables which in turn has led to successful selection of a perturbation parameter
1/T . This selection is in accordance with the cases studies, as here it is assumed that T »1,
so that indeed £«1. Compared to the dimensionless equations obtained with the dừect
approach, equations (4.44)-(4.47) have the advantage of dừectly showing the physical
prerequisite under which the model reduction with singular perturbation is allowed.
4.2.3. Case 3: T intermediate, Mo«l
Direct. The result obtained in 4.2.2.1 applies.
Timescale estimation for variables

Step 1. Estimate initial timescales and select fast variables.
The timescale estimation results (Table 5) indicate multiplicity of timescales, s is selected
as the fast state.
Table 5: Fast timescales for Case 3 (x(0)=0.5x( 00))
Case T* Mo Xi/%2 W'Cfx Tfx Tfs
3 3 0.04 0.036 0.034 1.88 0.063
Step 2. Estimate timescales of slow variables in the outer layer (quasi steady state).
X is the slow variable; its timescale is estimated with (4.33). Table 6 shows the results of
the slow timescale estimates.
Table 6: Slow timescales and error for Case 3 (x(0)=0.5x(o°))
Case T Mo Tfs/TsX TsX TfS AXfl/Xo
3 3 0.04 0.076 0 83 0 063 0.021
Step 3. Check the multiplicity of timescales.

From table 6, it is seen that for Case 3 the condition TfS/Tsx«l holds.
Step 4. Estimate the error of the slow variables.
The condition AXo/Xo«l holds as the error is approximately 2 % (Table 6). It is concluded
that timescale separation can be applied.
Step 5. Reduce model if preceding steps indicate that the QSSA applies.
The preceding results indicate that the QSSA is valid. The same comments as in subsection
4.2.2.2 apply.
Analytical scaling procedure. In this subsection, scaling is performed employing estimates

based on the analytical solutions of the eigenvalues of the system, which leads to the
standard form.
Estimate timescales
In Case 3, S«K no longer holds, because s= O(K). The approximation proceeds as follows,
s is in the order of magnitude as Soo, and with S=K/(T*-1), the term (S+K) is written as a-
K, with a=T* /(T*-1). Then, with x»s and X-Sin/ki we obtain (4.48).
= —(4.48) (S + K)2 a2Mo a2Mo
which is in the same order of magnitude as the estimate (4.38) for T* moderate (T*>2).
Perform error analysis on initial condition for slow state
The same results as in subsection 4.2.2.3 apply.
For timescales, choose state scaling and subsequently derive scaled equations and (try to)
find perturbation parameter.
For s, now another scaling is used as s in O(K). Substitution of scaled variables x=X-ki/S in
and S=s/K and for the slow timescale ts= t-D yields transformed equations. Dividing the
transformed equations by D, multiplying numerator and denominator of the Monod term
by K, using Mo=K/Sin yield the standard form with the outer equations (4.49 and (4.50).
dx * s
T ———X —X (4.49)
dts 1+s
_ _ CIS _* s _ _
Mo-—= — T --------x-Mos + 1 (4.50)
dts 1 + s with Mo* as a perturbation parameter. The substitution tf= t p/Mo gives
the inner equations (4.51) and (4.52).
dx M , z s 1
37" = Mo-—X—(4.51) dtf 1 + s T*
-7— =-------- x-Mos + 1 (4.52)
dtf 1 + s
Again, scaling has led to straightforward, successful selection of a perturbation parameter

and reformulation of the problem in standard form. This selection is in accordance with the
case studies, as here it is assumed that Mo«l, so that indeed 8«1.
4.2.4 Case 4: s and X comparable, T moderate, Mo ~1

In the subsequent cases, only the relevant steps of the respective procedures will be
discussed. In Case 4, both the timescales estimated by the eigenvalues and the estimated
timescales according to Procedure 2 indicate that the timescales are relatively close and no
timescale multiplicity occurs (Tables 7 and 8). This was also confirmed in simulation.
Thus, the conjecture that timescale multiplicity is associated with large concentration
differences is confirmed.
Table 7: Fast timescales for Case 4, 5 and 6 (x(0)=0.5x( 00))
Case T* Mo X1/À2 W'tfx TfX TfS
4 3 0.4 0.375 0.182 1.88 0.34

5 .244 0.04 1.00 0.123 1.90 0.24
6 1.044 0.04 0.0042 0.040 6.45 0.26
Table 8: Slow timescales and error for Case 4, 5 and, 6

=0.5x(°°))
(x(0)
Ca * Mo 'W'CsX ^sX
Tf AXo/Xo
se S
4 3 0.4 0.227 1.50 0.34 0.022
5 1.24 0.04 0.118 2.01 .024 0.0036
4
1.04
6 .04 0.002 129 0.26 0.019
4
4.2.5. Case 5: T close to critical, Mo « 1, Te~Mo
Case 5 is very interesting, as this is the situation of optimal biomass productivity in a
chemostat. The eigenvalue ratio (evaluated in steady state) indicates no difference in
timescales (Table 7). Also the eigenvalue ratio in the initial state was computed, which was
also close to 1 (1.56). This seems to be in agreement with the conjecture, because the ratio
Soo/Xoo is close to 1 (Table 2).
However, the fast timescale estimates obtained with Procedure 2 indicate occurrence of
timescale multiplicity, s being the faster variable (Table 7). This is confirmed by the ratio
Tfs/Tsx in table 8, which is much smaller than 1.
Figure 2 shows a simulation of Case 5, which shows that indeed a boundary layer in s
occurs. Apparently, the eigenvalues in the steady state and in the initial state do not reveal
this timescale multiplicity, whereas the timescale estimation by Procedure 2 correctly
indicated timescale multiplicity in this case.
The eigenvalues evaluated in the steady state do not correctly reflect the timescale
behaviour of the nonlinear model. The eigenvalue ratio in the quasi steady state was in
better agreement with the timescale estimation (X1/Ầ2=O.O88; computation not shown)
than the eigenvalue ratio in steady state (X|/X2= l)- An eigenvalue trace might thus better
reveal the timescale properties of the model, as was observed by Steffens et al. (1997).
However, this is less straightforward than the timescale estimation procedure and it is
concluded that the proposed timescale estimation procedure is to be preferred.
The conjecture that stiffness is associated with large concenừation differences is
falsified in this case, as timescale multiplicity occurs when the ratio Soo/Xoo is close to 1
(see Fig. 2).
From the results the timescale and error estimates by Procedure 2, it is concluded that
the QSSA applies in this case and s is the fast state. The model can thus be reduced. In the
next subsection, for the situation that T is close to critical, scaling will be applied to check
whether in this case the model can also be written in standard form and whether a
perturbation parameter can be found.
4.2.6. Case 6: T close to critical, Mo « 1, Te« Mo
Direct. From the conjecture that a low ratio Xoo/Soo is associated with timescale
multiplicity, both states are scaled with theừ steady state, assuming that £=X OO/SOO is very
small. With the scaled variables x=X/X oo and s=S/Soo, the scaled equations (4.53) and
(4.54) are obtained.
s„
X = 11———X -D X k + s (4.53)
sX
s - - Eki u.. —-----Ds + DSjn (4.54)
k + sX„
with Sin= Sự Soo. This system is not in standard form however, so the dữect scaling is not
successful in this case.
fast scale
350
300
250
200
150.
100
50
300
250
200
w
co
150
100
50
Fig. 2: Timescale multiplicity at optimal T
Timescale estimation for variables. From table 7 and table 8, it is seen that the fast
timescale for s (0.26) is much smaller than the fast timescale for X (6.45). The ratio of the
fast timescale for s and the slow timescale for X (129) is even much lower (0.002). Also
the ratio of the eigenvalues is very low. Clearly, multiple timescales are present. This is
confirmed by numerical simulation, which shows a very short boundary layer for s (result
not shown). Thus the QSSA holds and the system is in the form (4.2), (4.12) with £ very
small and can be reduced.
Analytical scaling procedure. Finally, the analytical scaling procedure will be employed.
Estimate timescales
In Case 6, the nontrivial eigenvalue of (4.21) is approximated as follows. We have, with
T*=l+Mo+Te and Te«l,
s Mo . Tp (4.5
—— -- - -——----«1 —
sin 1 + Mo + Tg -1 Mo 5)
ỵ _ (Sịn — S) 'tẹ Sịn (4.5

kq Mo ki 6)
which expresses the fact that conversion is very low and consequently the substtate
concentration is close to the feed concentration Sin. With x«s, and T e«Mo, after some
manipulation and approximation we obtain (the result is given without proof):
, -kiXK + S(S + K)
%2=- D + // 1
,-----= -D Tg (4.5
(S + K)2 7)
The fast and slow timescales can be scaled as (4.58) and (4.59) respectively.
(4.5
tf-t-D
8)
(4.5
ts=tDTe
9)
For the QSSA to be valid, Tf«Ts must hold
(4.6
1/D « l/D-Te, or -Te « 1
0)
which holds in the case studied.
Perform error analysis on initial condition for slow state: this is omitted here.
For timescales, choose state scaling and subsequently derive scaled equations and (try to)
find perturbation parameter.
Substitution of scaled variables x=X k1-Mo/SillTe and s=S/Sin and for the slow timescale ts=
t-D-T£ after some manipulation yields: (4.61), (4.62), which is not in standard form.
_ dx * s
T ——- = X —_x_x (4.61)
e
dts 1+s
ds *s 1 1
Tp -—= — T X— Es + —— dts 1 + s Mo Mo (4.62)
From this result and the scaling result in section 4.2.6.1, it is concluded that for low ratio
Xoo/Soo no simple perturbation parameter can be found. The results obtained with
Procedure 2 clearly indicated timescale multiplicity for low ratio Xoo/Soo, the fast
timescale being associated with the substrate concentration s. From the conjecture,
however, it was expected that for low ratio Xoo/Socthe fast timescale would be associated
with the lower concentration, namely the biomass concentration X. Consequently, for low
ratio Xoo/Soo, the conjecture is falsified.
4.3. OTHER RESULTS
In sections 4.2.2.1 and 4.2.3.1, dừect scaling was applied to the system with Monod
kinetics, which showed that the (small) ratio Soo/Xoo can be applied as perturbation
parameter. It can be shown that is also possible for zero-order and first-order kinetics.
In addition to the simple system presented in section 4.2, the same system but extended
with biomass retention was investigated, as biomass retention is usually applied in
activated sludge systems. Under typical operating conditions, the ratio Soo/Xoo is very
small, which causes an even more pronounced timescale multiplicity between substrate and
biomass in these systems (results not shown).
In addition to the simple system studied above, a slightly more complicated system was
studied which included dissolved oxygen as an additional state variable. Here too
occurrence of timescale multiplicity depends upon the operating conditions, especially the
dilution rate and the oxygen mass ttansfer rate. The timescale estimation procedure was
successfully tested to this system under operating conditions causing two timescales
(Weijers and Weiss, 1999).
4.4. CONCLUSIONS
A simple bioreactor model of a chemostat with one biomass species and one substrate
species was studied to obtain insight into timescale properties of bioprocess models and to
test different procedures for model reduction. Starting point in the analysis was the
conjecture that occurrence of multiple timescales is associated with a large concenttation
difference between the states. Three procedures to bring the problem into standard form
were tested, namely a direct scaling procedure, a systematic, analytical scaling procedure
and a procedure based on timescale estimation of variables.
At low substrate/biomass ratios, which occur at low dilution rate, the conjecture was
valid. The direct scaling procedure showed that the problem can be brought into standard
form and that the ratio substrate/biomass can use as perturbation parameter in this case.
The analytical scaling procedure enabled a more detailed analysis and showed that,
depending upon the operating conditions, the Monod number or the reciprocal of the
dimensionless residence time are suitable perturbation parameters. The substrate
concenttation was the fast variable in this case. The timescale estimation procedure
correctly indicated the validity of the quasi-steady-state assumption for substtate.
For intermediate substrate/biomass ratios, the conjecture was falsified. At dilution rates
with optimal biomass productivity, the ratio of eigenvalues evaluated in steady state did
not reveal timescale multiplicity, whereas the timescale estimation procedure correctly
indicated timescale multiplicity in this case. The substtate concentration was the fast
variable in this case, while the substrate concentration and biomass concentration were in
the same order of magnitude. The timescale estimation procedure correctly indicated the
validity of the quasi-steady-state assumption for substrate.
For high substrate/biomass ratios, the conjecture was also falsified. At high dilution
rates close to washout, the ratio of eigenvalues evaluated in steady state and the timescale
estimation procedure correctly indicated timescale multiplicity. The substrate concentration
was the fast variable in this case, while from the conjecture it was expected the biomass
concentration would be the fast variable. Neither the dfrect scaling procedure, nor the
analytical scaling procedure led to the standard form. The timescale estimation procedure
correctly indicated the validity of the quasi steady state assumption for substrate, as
confirmed by the clear boundary layer for this variable observed in simulations.
With respect to the procedures tested, it is concluded that the timescale estimation
procedure is a helpful tool to detect timescale multiplicity and check validity of the quasi-
steady-state approximation. The use of eigenvalues for this purpose can be misleading in
nonlinear systems. The analytical scaling procedure can be helpful to bring the problem
into standard form and to obtain a perturbation parameter, thus providing insight into the
cause of timescale multiplicity. Dừect scaling is not a generally applicable procedure.
An important motivation to derive reduced nonlinear models is to obtain models that
have a larger validity range than linear models. It is observed, however, that also nonlinear
models obtained by reduction based on timescale separation have a limited validity range,
namely for that operational range in state space or parameter space in which the
assumptions for reduction are valid. A change of operating point may requừe a different
reduced model, as states that were partitioned as slow may become fast or vice versa. This
is probably the cause of the large error induced in the reduction of ASM1 studied by
Steffens and Lant (1997), discussed in section 2.3. Thus, together with the reduced models,
also a validity range should be indicated.
5. Conclusions and perspectives
Rigorous models often must be reduced to low-order models as these are better suitable for
control, identification and are helpful to obtain a better understanding. The aim in this
paper was to develop a systematic reduction procedure to obtain nonlinear reduced models
for controller design. Ffrst, reported reduction approaches and reduced models of ASM1
were reviewed. Several approaches are applied in ASM1 model reduction. Simplifying
assumptions have been applied most frequently. Proposed models range from simple
black-box zero-order kinetic models neglecting biomass dynamics, to the more
complicated model of Jeppsson (1996), which includes biomass growth and decay and
consequently has a larger validity range.
Assumptions with respect to dynamics are also frequently usually applied for model
order reduction. This is logical, as the stiffness of activated sludge process argues to
develop models that are suitable for different timescales. It appears that the reduction often
is done heuristically. Relatively little systematic efforts have been done to analyse and
understand timescale properties of ASM1.
It was concluded that singular perturbation is a promising candidate as a method to develop
a systematic reduction procedure. It is a systematic reduction procedure and it can give
more insight in dynamics of ASM1. Moreover, under certain conditions, it retains the
physical interpretation of the states in the order reduction, it provides an error estimate and
it is suitable for nonlinear model reduction, which is desừed as it is an aim to obtain
reduced models.
To apply the method, the model must be in the so-called standard form, which means that
the states can be partitioned into fast and slow states. Obtaining this form or recognising
that the model is in this form is the difficult part in the method. Therefore, it was decided
to concentrate on this task.
First, methods in the literature were reviewed. Bastin and Dochain (1990) suggested a
simple rule for order reduction of (bio)process models through singular perturbation. This
simple rule is not generally applicable, however, and can be applied in specific cases only,
e.g. in the case of very low product solubility. Moreover, it does not yield the state
partitioning or tell when the reduction can be made. Segel and Slemrod (1989) applied a
methodology based on scaling to formally derive Michaelis-Menten kinetics via singular
perturbation. This methodology was successful in detection of multiplicity of timescales, to
bring the model in the standard form and provided a suitable perturbation parameter that
gave insight in the physical conditions that lead to timescale multiplicity. This procedure
was therefore used as a starting point for the state partitioning (or detection of timescale
multiplicity).
Three procedures were proposed, namely a dfrect scaling procedure, a procedure based
on a mixed numerical / analytic timescale estimation and the analytical scaling procedure
of Segel and Slemrod. The procedures were tested on a simple bioreactor model of a
chemostat with one biomass species and one substtate species. Starting point in the
analysis was the conjecture that occurrence of multiple timescales is associated with a large
concenttation difference between the states.
At low substrate/biomass ratios, occurring at low dilution rates, the conjecture is valid. The
ratio substrate/biomass, the Monod number and the reciprocal of the dimensionless
residence time are suitable perturbation parameters. For intermediate substrate/biomass
ratios, the conjecture was falsified. The substrate concentration was the fast variable in this
case, while the substtate concentration and biomass concenttation were in the same order
of magnitude. For high substrate/biomass ratios at high dilution rates close to washout, the
conjecture was also falsified. The substrate concenttation was the fast variable in this case,
while from the conjecture it was expected the biomass concentration would be the fast
variable.
With respect to the procedures tested, the timescale estimation procedure is a helpful
tool in model reduction, which is concluded from the fact that it in all cases correctly
indicated whether the quasi-steady-state assumption was valid. The use of eigenvalues
evaluated in the steady state for timescale multiplicity however can be misleading in
nonlinear systems. The analytical scaling procedure is helpful to bring the problem into
standard form and to obtain a perturbation parameter in some cases, thus providing insight
into the cause of timescale multiplicity. The direct procedure is not generally applicable.
It is expected that model reduction will remain an active research area in the next years,
for application in confrol, identification and process optimisation. Singular perturbation
provides the formal basis for quasi-steady-state assumptions and timescale separation and
can be an important tool in future reduction studies. The procedure for timescale estimation
of variables is a helpful tool for model reduction, as it provides a timescale to state
association as well as an error estimate of the reduction. Scaling can provide additional
insight into causes of timescale multiplicity.
In almost all reported cases of ASM1 reduction, the reduction was applied to lumped
systems without concentration gradients, assuming the system to be either aerobic or
anoxic. Thus reduction concentrated on reduction of the reaction kinetics, rather than
reduction of transport and mixing in the reactor. However, many activated sludge systems
are of the plug-flow type and exhibit gradients especially with respect to dissolved oxygen.
This also holds true for full-scale bioreactors. Therefore, reduction of distributed systems
may become an important topic in reduction of ASM1 and bioprocess models.
References
Ayesa E., Oyarbide G., Larrea L. and J. L. Garcia-Heras (1995) Observability of reduced order models -
application to a model for control of alpha process, Wat Sci. Tech. 31 (2), 161-170.
Bastin, G. and D. Dochain (1990) On-line estimation and adaptive control of bioreactors, Process measurement
and control, 1, Elsevier, Amsterdam.
Bodenstein, M. and H. Lutkemeyer (1924) z. Phys. Chem. 114 208.
Bowen, J. R. , A. Acrivos and A. K. Oppenheim in (1962) Singular perturbation refinement to quasi-steady state
approximation in chemical kinetics, Chem. Eng. Sci. 18, 177-188.
Côte, M. , B. p. A. Grandjean, p. Lessard and J. Thibault (1995) Dynamic modeling of activated sludge processes:
improving prediction using neural networks, Wat. Res. 29, 995-1004.
Carstensen, J. , p. Harremoes and EL Madsen (1995) Statistical identification of Monod kinetic parameters from
on-line measurements, Wat. Sci. Tech. 31 (2), 125-133.
Dochain, D. (1994) Contribution to the analysis and control of distributed parameter systems with application to
(bio)chemical processes and robotics, Aggregate thesis, Universite Catholique de Louvain, Belgium.
Dold, p. L. and G. V. R. Maris (1986) Evaluation of the general activated sludge model proposed by the IAWPRC
Task group, Wat Sci. Tech 18 (6), 63-89.
Heineken, F. G. , H. M. Tsuchiya and R. Aris (1967) On the mathematical status of the pseudo-steady state
assumption of biochemical kinetics, Mathematical biosciences 1, 95-113
Henze M. et al. (1987). Activated Sludge Model No. 1, Scientific and technical reports No. 1, IAWPRC, London.
Isaacs, s. (1996) Short horizon control strategies for automating activated sludge process. Wat Sci. Tech. 34 (1-2),
203-212.
Isaacs, s. and M. Kummel (1993) Dissolved oxygen setpoint control of an alternating activated sludge process,
Proc. 7th Forum for Applied Biotechnology, Med. Fac. Landbouww. Univ. Gent, 2083-2086.
Isaacs, s. H. , M. Henze and M. Kummel (1995) An adaptive algorithm for external carbon addition to alternating
activated sludge process for nutrient removal from wastewater, Chem. Engng. Sci. 50 (4), 617-629.
Jeppsson, Ư. and G. Olsson (1993) Reduced order models for on-line parameter identification of the activated
sludge process, Wat Sci. Tech. 28 (11), 173-183.
Jeppsson, u. (1996) Modelling Aspects of Wastewater Treatment Plants, PhD Thesis, Lund Institute of
Technology.
Julien, s. , J. p. Babary and p. Lessard (1998) Theoretical and practical identifiability of a reduced order model in
an activated sludge process doing nitrification and denitrification, Wat Sci. Tech. 37 (12), 309316.
Kokotovic, p, H. Khalil and J. O'Reilly (1986) Singular perturbation methods in control: Analysis and design,
Academic Press, London.
Lin, c. c. and L. A. Segel (1974) Mathematics applied to deterministic problems in the natural sciences, McMillan
Publishing Co. Inc. , New York
Lindberg, c. -F. (1997) Control and estimation strategies applied to the activated sludge process, PhD Thesis,
Uppsala University.
Lukasse, L J. s. (1999) Control and Identification of Activated Sludge Processes, PhD Thesis, Wageningen
Agricultural University, Wageningen, The Netherlands.
Lukasse, L J. s. , K. J. Keesman, A. Klapwijk and G. van Straten (1997a) Adaptive receding horizon optimal
control of N-removing activated sludge processes, Proc. 11 th Forum for Applied Biotechnology, Med. Fac.
Landbouww. Univ. Gent 62 (4B), 1665-1672.
Lukasse, L J. s. , K. J. Keesman, A. Klapwijk and G. van Straten (1997b) Identification for models predictive
control of biotechnological processes, case study: nitrogen removal in an activated sludge process, Proc. 11 th
IFAC symp. on Syst. Ident. 3, Fukuoka, Japan, 1525-1530.
Lukasse, L J. s. , K. J. Keesman, A. Klapwijk and G. van Straten (1998a) Optimal control of N-removal in ASPs,
Wat Sci. Tech. 38 (3), 255-262.
Lukasse, L J. s. , K J. Keesman and G. van Straten (1998b) A recursively identified model for short term
predictions of NH4 and NO3 concentrations in alternating activated sludge processes, Journal of Process
Control 9, 87-100.
Lukasse, L J. s. , K. J. Keesman and G. van Straten (1999) A comparison of NH4/NO3 control strategies for
alternating activated sludge processes, Wat Sci. Tech. 39 (4), 93-102.
Marsili-Libelli, s. (1989) Modeling, identification and control of the activated sludge process, In: A. Fiechter
(Ed. ), Advances in Biochemical Engineering/Biotechnology 38, Springer Verlag, Berlin, Heidelberg, 89-
148.
Moore, B. c. (1981) Principal component analysis in linear systems: controllability, observability and model
reduction, IEEE Transactions on Automatic Control, AC-26,17-32.
Oosterhuis, N. M. G. (1984) Scale-up of bioreactors, a scale-down approach, PhD Thesis, Delft University of
Technology.
Olsson, G. and u. Jeppsson (1994) Establishing cause-effect relationships in activated sludge plants - What can be
controlled. Proc. 8th FAB, Med. Fac. Landbouww. Univ. Gent 59(4a), 2057-2071, Gent.
Robertson, G.A. and Cameron, LT. (1997) Analysis of dynamic process models for structural insight and model
reduction-Part 1: Structural identification measures, Computers chem. Engng. 21 (5), 455-473.
Robertson, G.A. and Cameron, LT. (1997) Analysis of dynamic process models for structural insight and model
reduction-Part 2: A multi-stage compressor shut-down case study, Computers Chem. Engng. 21 (5), 475-488.
Roels, J. A. (1983) Energetics and Kinetics in Biotechnology, Elsevier Science Publishers, Amsterdam.
Scherpen, J. M. A. (1994) Balancing for nonlinear systems, PhD Thesis, University of Twente, Enschede, The
Netherlands.
Segel, L. A. (1972) Simplification and scaling, SIAM Review 14 (4), 547-571.
Segel, L. A. and M. Slemrod (1989) The quasi-steady state assumption: a case study in perturbation. SIAM
Review. 31 (3), 446-477.
Segel, L A. (1984) Dynamic phenomena in Molecular and Cellular Biology, Cambridge University Press,
Cambridge.
Steffens, M. A. , p. A. Lant and R. B. Newell (1997) A Systematic Approach for reducing biological wastewater
tteatment models, Wat Res. 31 (3), 590-606.
STOWA (1996). Methods for influent characterization: Inventory and guidelines, Report STOWA 96-8, Utrecht,
The Netherlands (in Dutch).
Su, H. -T. , T. J. Me. Avoy and p. Werbos (1992) Long-Term Predictions of Chemical Processes Using Recurrent
Neural Networks: A Parallel Training Approach. Ind. Eng. Chem. Res. 31 (5), 1338-1352.
Van Breusegem, V. and G. Bastin (1991) Reduced order dynamic modelling of reaction systems: a singular
perturbations approach, Proc. 30th conference on decision and control, Brighton, England, 1049-1054.
Van Breusegem, V. and G. Bastin (1992) Order reduction in bioprocess modelling: a singular perturbation
solution, Proc. IFAC modeling and control of biotechnical processes, Colorado, u. s. A., 347-350.
van Can, H. J. L. (1997) Efficient mathematical modelling for bioprocesses based on macroscopic balances and
neural networks, PhD. Thesis, Delft University of Technology.
Van Impe, J. F. , p. A. Vanrolleghem, w. Verstraete, B. De Moor and J. Vandewalle (1991) Model based
monitoring and control of activated sludge wastewater treatment processes, Part n: nonlinear control of the
biotransformation and sedimentation process, Proc, of the 1991 European Simulation Symposium, Nov. 6-8,
Gent (Belgium), 221-226.
Vanrolleghem, p. (1994) On-line modelling of activated sludge processes: development of an adaptive sensor,
PhD Thesis, University of Gent.
van Schagen, K M., Veersma, A. M. J. , Meinema, K and Roest, H. F. van der (1995) Multivariable: the new
generation?, H2O 28 (16), 480-483 (in Dutch).
Vansteenkiste, G. c. and J. A. spriet (1982) Modelling ill-defined systems, In: Progress in Modelling and
Simulation, Ed. Cellier F. E. , Academic Press, London, 11-38.
Vasil’eva, A. B. , V. F. Butuzov and L. V. Kalachev (1995) The boundary function method for singular
perturbation problems, SIAM Studies in Applied Mathematics, Vol. 14.
Wasynczuk, o.and Decarlo, R.o. (1981) The components connection model and structure preserving model
reduction, Automatica 17 (4), 619-626.
Weijers, s. R. , J. J. Kok and H. A. Preisig (1995) Analysis of dissolved oxygen dynamics in the IAWQ Model
No. 1, Proc. 9th Forum for Applied Biotechnology, Med. Fac. Landbouww. Univ. Gent 60(4B), 2455-2458.
Weijers, s. R. and M. Weiss (1999) Quasi-steady state assumptions in the modeling of bioreactors: model
reduction and validation, Proc. American Conttol Conference, 1999.
Weiss, M. and H. A. Preisig (1997) Simplifying hypotheses in computer-aided modelling: a singular perturbation
approach, Computers Chem. Engng, 21, Suppl., S721-726.
Wortelboer, p. (1994) Frequency-weighted Balanced Reduction of Closed-loop Mechanical Servo-systems:
Theory and Tools, PhD Thesis, Delft University of Technology.
Symbols
ASM1 symbols ______________________________ ____________________
Stoichiometric parameters:
YH :: Heterotrophic yield (-)
YA:: Autotrophic yield (-)
fp:: Fraction biomass yielding inert (-)
products
Ixb •• Fraction N in biomass (kg N/kg COD)
ixp::____________________ Fraction N in inert products (kg N/kg COD)
Kinetic parameters:
|1H:: Heterotrophic growth rate (d-1)
constant
bH:: Heterotrophic decay rate constant (d-1)
Ks:: Affinity constant for Ss (kg tn'3)
KOH •• Heterotrophic affinity constant (kg nA
for So
KNHH " Heterotrophic affinity constant (kg nA
for SNH
KALKH Heterotrophic affinity constant (mol nf3)
for SALK
ng:: Correction factor for anoxic (-)
growth
|1A:: Autotrophic growth rate constant (d1)
bA:: Autotrophic decay rate constant (d1)
KOA:: Autotrophic affinity constant for (kg nA
So
KNHA :: Autotrophic affinity constant for (kg nA
SNH
KNO :: Affinity constant for SNO (kg nA
KALKA •• Autotrophic affinity constant for (Mol nA
SALK
Symbols
kh:: Hydrolysis rate (d1)
Kx:: Hydrolysis affinity constant (kg m3)
ka:: Ammonification rate (d1)
■Hh:: Correction factor for anoxic (-)
hydrolysis
Components:
Ss :: Readily biodegradable COD (kg m3)
Si:: Soluble inert COD (kgm3)
SNH •• Ammonia and ammonium (kgNm3)
SNO:: Nitrite and nitrate (kgNm-3)
SND •• Soluble biodegradable organic N (kgNnf3)
SALK •• Alkalinity (Mol m'3)
So:: Dissolved oxygen (kgm-3)
XBH •• Active heterotrophic biomass (kg m'3)
XBA :: Active autottophic biomass (kg m3)
Xs:: Slowly biodegradable COD (kg m’3)
Xj:: Particulate inert COD (kg m-3)
XND :: Particulate biodegradable org. N (kgNnf3)
Xp:: Particulate COD from decay (kg m-3)
PARAMETRIC SENSITIVITY OF A DYNAMIC MODEL FOR A PILOT
SINGLE-STAGE WASTEWATER TREATMENT PLANT
IGOR PLAZL, GORAN PlPUằ, AND TINE KOLOINI

Department of Chemical Engineering, University of Ljubljana
Askerceva 5, 1001 Ljubljana, Slovenia igor.plazl@ Uni-Lj.si
Summary
The application of models in most treatment plants is limited due to a lack of advanced
input parameter values requừed by the models. Although the numbers of parameters are
presented in the literature, the range of some parameters is too wide considering the
parametric sensitivity of the dynamic model. On the other hand, some parameters depend
on the nature of a specific wastewater treatment plant. The present paper is concerned
with a dynamic model of the activated sludge process in a single stage wastewater
treatment plant with parametric sensitivity and evaluation. Model calibration was
successfully experimentally confirmed for the steady-state operational conditions.
1. Introduction
The model parameters and state estimation associated with modern control studies based
on the available noisy process measurements. An alternative parameter approach is via
laboratory analysis [1]. These procedures of the IAWPRC Activated Sludge Model I [2]
are presented in Ekama et al. [3,4]. However, an important aspect of insttumentation
theory is that measurements are never deterministic variables, since they always involve
random noise as well as experimental random errors [5]. On the other hand, many authors
discuss various nonlinear numerical estimation methods (Bayesian estimation method -
[6]; Maximum Likelihood methods - [7]). Kabouris and Georgakakos [1] presented a
continuous process model formulation in a state-space form. They presented the
measurement model, including the relationship between the measured quantities and
model state variables, followed by the development of the Linearized Maximum
Likelihood (LML) algorithm. In this study attempts were made to present a dynamic
model for activated sludge process. Efforts were made to simplify the parameter
evaluation procedure. The calibration of the dynamic model was successfully
experimentally confirmed.
65
S.N. Agathos and w. Reineke (eds.), Biotechnology for the Environment: Wastewater Treatment and Modeling, Waste
Gas Handling, 65-72.
© 2003 Kluwer Academic Publishers.
Igor Plazl, Goran Pipus, and Tine Koloini
2. Materials and methods
The laboratory at Vodovod-Kanalizacija Ltd., Ljubljana, monitors daily the activated

sludge process of a pilot single-stage wastewater treatment plant with reactor volume, Vi
= 1.585 m3, and settler volume, v2 = 0.709 m3 (Fig. 1). For the purpose of this work, some
additional analyses were made. The concenttations of autotrophic nitrifying biomass, X A,
heterotrophic biomass, XH, activated sludge concenttation, inlet and outlet ammonium, A,
dissolved oxygen, So, and the concentration of inlet and outlet substtate (in g COD m' 3),
SCOD, were determined by standard laboratory methods. The average measured values
for the period from 4 to 20 July 1998 are presented in table 1.
Fig. 1. Sketch of the single-stage wastewater treatment plant. ( Ỉ - reactor, 2 - settler)
Table 1. The average measured values of process parameters for the period from 4 to 20 July 1998
(source: JP Vodovod Kanalizacija Ltd., Ljubljana).
Qo (inlet flow) = 0.263 m3 h’1 = 3.6

So g 02 m’3 = 416 g
SCOD,O (inlet) COD m’3 = 16.1 g N-
Ao (inlet) NH/ m‘3 = 42 g COD
SCOD,2 = SCOD,I (outlet) m3 = 0.5 g N-NH4+
A2 = Al (outlet) m’3 = 3104 gm3 = 96
XH,1 g in3 = 4920 g m'3 =
XA.1 . 65%
Activated sludge concentration
Biomass fraction
66
Parametric sensitivity of a dynamic model for a pilot single-stage wastewater treatment plant
3. Mathematical model
The dynamic model is obtained by mass balances of substrate, ammonium, autotrophic

and heterotrophic biomass for the single-stage wastewater treatment plant, separately for
the reactor and the settler [1,2,8], considering aerobic growth of heterotrophs and
autotrophs and equations of time dependent flow rates. The set of linear differential
equations can be solved by an appropriate numerical method, using literature data on the
kinetic and stoichiometric parameters. However, in order to get an acceptable agreement
between the measured (Table 1) and predicted sludge process, numerical estimation of
some parameters is necessary. To simplify this procedure, considering that the
experiments were performed at steady-state conditions, the dynamic model was rebuilt
for the case of steady-state operation. In this way, the set of differential equations
becomes a set of ordinary equations and the equations for flow rates become time
independent. On the basis of the experimental values of sludge process variables (Table
1) and by using some of the parameters from literature (Y H, YA- heterotrophic and
autotrophic yield coefficient; K0,H, K0,A- oxygen half saturation for heterotrophs and
autotrophs; KA“ ammonia half saturation; Ks- substrate half saturation) the estimation of
other parameters (ỊLLmax,í-b Umax,A- maximum growth rate for heterotrophs and
autotrophs; bH, bA- decay rate for heterotrophs and nitrifiers; i x-N content of biomass) can
be obtained by steady-state model calibration, using an appropriate numerical method.
However, it was found again that the parametric sensitivity of the simplified model is still
very high and we are unable to simply calibrate the parameters with ordinary numerical
methods, built in a powerful mathematical package (Mathematica 3.0). The following
iterative procedures were eventually found to be the most effective and successful
method for model calibration:
Step a - Some parameters (YH, YA, Ko,H5 Ko,A, KA) were adopted from literature.
Step b - The assumed value for Ks can be found in the literature as the initial value for
calibration. The study of parametric sensitivity of the dynamic model has shown that the
influence of Ks (in range 4-60 gCOD m'3) on system stability is negligible in our case. On
the other hand, the linear dependence of Ks on evaluation of the maximum growth rate of
heterotrophic organisms can be obtained (Eq. 2: JWH[KS=4 g COD nf3] = 0.0146 h1;
Pmax,H [Ks=60 g COD m3] = 0.0324 if1).
Step c - The assumed value for ix can be found in the literature (0.07 g N (g COD) x) as
the initial value for calibration.
Step d - The assumed value for bA can also be found in the literature (0.0063-0.002 h' 1) as
the initial value for calibration (0.002 h'1).
Step e - The value for heterotrophic decay rate, b H, is calculated from the following
67
equation (Eq. 1):
68
Parametric sensitivity of a dynamic model for a pilot single-stage wastewater treatment plant
Qo (ScQD,0 - ^COD,l yYtỉ

bH = bA (1)
which can be obtained by equalising equations of mass balances of substrate and

heterotrophic biomass arranged for steady-state, considering a substrate concentration
equality, SCOD,I=SCOD,2> and the condition XỊỊJ2=2XỊỊJ.
Step f - The value for the maximum growth rate of heterotrophic organisms, Pmaxjb can
now be calculated from the equation (Eq. 2):
I Vo I Y \
"T *Ô,H °COD,\ TKS
z^max,2f (2)
$0 Ạ $COD,l J
which is
derived
from the mass balance of heterotrophic organisms arranged for steadystate operation,
considering again the substrate concentration equality, SCOD,I=SCOD,2, and the condition
XH,2=2XH,1. Finally, equation 2 is obtained after simple mathematical procedures
considering the expressions for bn (Eq. 1).
Step g - The maximum value of N content of biomass, ix,max, is determined from the
condition of positive growth rate at defined operational conditions (see Eq. 5):
l
Ì <i = 0o(A)“A) x - X,max “ D T7
ỉ
(3)
If the initial value of ix is bigger than ix,max, the procedure is then repeated from
step c) until reaching a satisfying system solution (Eq. 6). The study of parametric
sensitivity of the dynamic model has shown that the influence of i x on system stability is
very high. After a number of iterations the final value, i x = 0.001 g N (g COD)1, was
found for our experimental conditions at which the system error is negligible.
Step h - The value for the maximum growth rate of autotrophic organisms, jXmaX’A,
can be calculated from the equation (Eq. 4):
feo(A? ~ A-Ỉ — ixQp(SCOD,O — Scop,ii^HÂXA + KẠ)(s<> + KQ,A)

= (4)
A
■ + which is derived from the mass balance of ammonium arranged for steady-
state operation, considering the ammonium concentration equality, A1=A 2, and condition
XA,2=2XA,I. After some simple arrangements, RA can be expressed:
69
R _Q0(A<)-AJYA ÌXRHVỊYẠ
A
V^ + ixYA) V^ + ixYA)- (5)
Step i - The system error considering the initial value of b A can be determined from the
following equation (Eq. 6):
0 = Ô2Â,2 -ÔiX +RAVỵ-bAXAAVx, which is derived from the mass balance of

Ajl (6)
autotrophic organisms arranged for steadystate operation. From Equation 6 the new
value for bA can be obtained. The procedure is then repeated from step d) to i) until
reaching a system solution.
4. Results and discussion
On the basis of the described method the calibrated parameters defined for the steadystate
activated sludge process taking place in a pilot single-stage wastewater treatment plant
(Table 1) can now be used for successful process simulation based on the dynamic
model. The calibrated parameters used in the dynamic model are presented in table 2.
Table 2. Estimated and typical literature values for some kinetic and stoichiometric coefficients [8]
Parameter Steady-state model Literature data Units
calibration
bH 0.0079 0.008-0.016 h-1
bA 0.0018 0.002-0.006 h1
ix 0.001 (0.066) 0.07 gN(gCOD)
’1
lhnax,H 0.0146 (0.0324) 0.125-0.25 h1
Mmax,A 0.0218 0.015-0.042 h1
Substantial deviation can be seen for ix and Jimax,H when the system calibration error is
practically zero. However, for practical purposes, which still allowed stability of the
system, the value 0.066 g N (g COD)1 for ix and the value 0.0324 h1 for Pmax,H[Ks=60
g COD m'3] were calculated. An example of dynamic model experimental confirmation
for the single-stage wastewater treatment plant (JP Vodovod Kanalizacija Ltd.,
Ljubljana) is presented in figure 2.
70
Fig. 2. Dynamic model prediction of substrate concentration for the steady-state activated
sludge process taking place in a single-stage wastewater treatment plant.
2.0Ị ' I ' I .■■■ I .................................................. —
QP(t) = 0.263 + [ts30Exp(-0.2 ậJ/8- !(? II
______ Ạ - reactor \-
settler
°-°0 50 100 150 200 250 300

too
Fig. 3. Dynamic response of ammonium concentration in the reactor and the settler during the
simulation of “real” two-day disturbance of inlet wastewater flow.
Fig. 4. Dynamic response of substrate concentration in the reactor and the settler during the
simulation of sinuous fluctuation of inlet wastewater flow.
Dynamic simulation of the activated sludge process is presented in figure 3 during the
simulation of “real” disturbances of inlet wastewater flow. Dynamic responses of
substrate and heterotrophs concentration during the sinuous simulation of inlet waste
water flow are presented in figures 4 and 5.

t(h)
■ 1 • 1 --------------1--------------1--------------
Fig. 5. Dynamic response of heterotrophs concentration in the reactor and the settler during the
simulation7000
of sinuous fluctuation of inlet wastewater flow.
5. Conclusions
6000
A mathematical model of the activated sludge process taking place in a pilot singlestage
wastewater ừeatment plant was built. A relatively simple procedure is proposed for the
5000
evaluation of the kinetic and stoichiometric coefficients. The dynamic model calibration
’s
4000
--
3000 -
-
was successfully confirmed experimentally for steady-state operational conditions.
Acknowledgment
The authors thank the representatives of JP Vodovod-Kanalizacija Ltd., Ljubljana, in

particular Maj a Drolka, Jurij Kus and Alojz Hojs for theứ encouragement, cooperation,
and financial support. We also express our appreciation to Mitja Lakner for useful
mathematical discussions.
References
[1] Kabouris, J.C., Georgakakos, A.P.: Wat. Res. 30 (1996), 2853-2865.
[2] Henze, M., Grady, C.P.L., Gujer, w., Marais, G.R., Matsuo, T.: Activated Sludge Model No. 1, IAWQ
Scientific and Technical Reports No. 1, IAWPRC, London, ISSN 1010-707X, 1987.
[3] Ekama, G.A., Marais, G.V., Pitman, A.R., Keay, G.F.P., Buchan, L., Gerber, A., Smollen, M.: South
African Water Research Commission, Pretoria, South Africa, 1984.
[4] Ekarna, G.A., Dolt, P.L., Marais, G.R.: Wat. Sci. Tech., 18 (1986), 91-114.
[5] Willard, H.H., Merritt, Jr.L.L., Dean, J. A., Seattle, Jr.EA.: Instrumental Methods of Analysis; Wadsworth,
Belmont, Calif., 7th edition, 1988.
[6] Beck, M.B.: System Simulation in Water Resouces; North-Holland Publishing Co., Amsterdam, 1976.
[7] Mybeck, P.S.: Stochastic Models, Estimation and Control; Volume 2, Academic Press, New York, 1982
[8] Henze, M., Grady, C.P.L., Gujer, w., Marais, G.R., Matsuo, T.: Activated Sludge Model No. 2, IAWQ
Scientific and Technical Reports No. 3, Preprint for IAWQ Specialised Seminar on Modelling and Control
of Activated Sludge Processes, Copenhagen, (1994), 22-24.
APPLICABILITY OF A SIMULATION BENCHMARK TO RESPIROMETRY-
BASED CONTROL STRATEGIES
JOHN B. COPP AND HENRI SPANJERS

AEST, Environmental Technology, Wageningen University and Research
Centre, Wageningen, The Netherlands. Fax: +31 317 482108;
e-mail: John. Copp @ algemeen. mt. wau. nl
Abstract
A simulation benchmark is applied to a number of respừomeữy-based control sfrategies.

The benchmark is a platform-independent simulation envừonment defining a plant
layout, a simulation model, influent loads, test procedures and evaluation criteria. The
purpose for defining a benchmark is to create a general simulation envừonment for
evaluating and comparing different conttol sttategies. In the paper we describe the
definition of a general benchmark for the evaluation of conttol sttategies in wastewater
ữeatment plants and identify the features of the benchmark which need adjustment to
make it applicable to a particular family of control sttategies, namely those based on the
measurement of respừation rate.
1. Introduction
The activated sludge process aims to achieve, at minimum costs, a sufficiently low
concenttation of biodegradable matter in the effluent together with minimal sludge
production. To do this, the process has to be conttolled. Many conttol strategies have
been proposed in the literature. However, the literature does not provide a clear basis for
comparison because of the many confounded influences, which have an impact on the
system. Many of these influences are easily recognised. For instance, the influence of a
conttol sttategy on process performance is expected to vary with different disturbances,
thus the disturbances used to test the conttol sttategy become important. Additionally, the
objective of reported strategies is not always consistent which may result in the omission
of data necessary to make a faừ and unbiased comparison. As well, physical
characteristics of the process can have an impact on process performance, which makes
the comparison of sttategies applied to different reactor layouts difficult. Also
complicating the evaluation is the lack of standard evaluation criteria. That is, effluent
requừements and tteatment costs (i.e. labour costs) are often location specific. This
73
John B. Copp and Henri Spanjers
makes it difficult to judge the particular influence of an applied conttol sttategy from a
reported performance increase.
Two common aerobic conttol strategies involve the maintenance of biomass levels
and dissolved oxygen concentrations in the aeration tanks by manipulating waste sludge
flow, return sludge flow or aeration capacity. Such strategies are based on measurements
of mixed liquor suspended solids and dissolved oxygen; however, alone these
measurements give limited information as to process stability and performance.
Alternatively, the respfration rate of activated sludge provides valuable information for
modelling and control of the activated sludge process because respfration is linked to two
important biochemical processes in a wastewater treatment plant: biomass growth and
substrate consumption.
The respfration rate is an important variable in the activated sludge process as it
provides information on activity and concenttation of the biomass, influent waste
concenttation and composition, toxicity and concenttation of biodegradable matter in the
effluent. Highly sophisticated and reliable respirometers have been developed in recent
years, but the application of these insttuments in control of full-scale activated sludge
plants lags behind the recognised potential of respừomeừy. This is due, at least in part, to
a lack of understanding with respect to the information content of respừomeừic data and
confusion arising from inconsistency in implementation of control sttategy methods.
The literature related to respirometty is substantial and numerous control strategies
involving respfrometry have been proposed and tested. However, from a practical
standpoint, it is not reasonable to experimentally test and verify the effectiveness of all
reported control strategies. Alternatively, given a standardised procedure, it is possible to
efficiently evaluate numerous strategies through computer simulations. Validation of the
computer simulations is difficult without supporting experimental or full-scale data, but
the value of the work is enhanced through the use of accepted activated sludge models.
Because appropriate simulation tools for the activated sludge process are available this
approach has numerous advantages, but still there is a need for a standardised evaluation
procedure.
In an attempt to standardise control sttategy evaluation by simulation, a defined
methodology, called a benchmark, has been developed. The simulation benchmark is a
comprehensive simulation envfronment, which defines a plant layout, a simulation
model, influent loads, test procedures and evaluation criteria. In this instance, the
evaluation criteria relate the model output to performance indicators such as costs and
effluent quality. This paper briefly describes the features of a generalised simulation
benchmark for evaluating and comparing different control sttategies and then specifically
identifies features of that benchmark that need adjustment to make it applicable to
respirometry-based control strategics.
2. General benchmark description
The assessment of activated sludge confrol strategies is confounded in most cases by the
multifaceted nature of the process under study. That is, a true comparison of reported
strategies is difficult because of the many variables that have an impact on process
performance. To combat these problems and decrease the influence of these many
variables, there has been a recent effort to develop a standardised testing procedure - a
74
Applicability of a simulation benchmark to respừometry-based control strategies
benchmark.
Once defined, the benchmark forms the building block from which the various control
strategies are evaluated. Several benchmarks have been published or are under
development [1,8,11]. Spanjers et al. (1998) defined a simulation benchmark and used it
for the evaluation of a respirometry-based control strategy. The European Co-operation in
the field of Scientific and Technical Research (COST) 624 Action is currently working
on the development of a general benchmark. The COST benchmark (in this paper referred
to as the general benchmark), although based on the work of Spanjers et al., is an attempt
to create a generally accepted benchmark for the evaluation of all types of control
strategies. Cross-platform testing of the COST benchmark has been successfully
demonstrated such that similar results can be attained using GPS-X™ (Hydromantis Inc),
Simba™ (Ifak System GmbH), WEST^ ™ (Hemrnis n.v.) and a user defined Fortran
code [1,8]. A brief description of the COST generalised benchmark is outlined below.
3. Plant layout
The COST benchmark plant was designed to treat an average flow of 20 000 m 3 d"1 with
an average influent biodegradable COD concentration of 300 g m -3. The plant was
designed for pre-denitrification using a 5 tanks-in-series layout (5999 m3) with a
secondary settler (6000 m3). The first two tanks are fully mixed, but unaerated and the
last three are fully mixed and aerated. For modelling purposes, the secondary settler is
comprised of 10 equal volume layers. The plant has two internal recycles including one
from the underflow of the settler to the head of the plant, and a second recycle directly
from the fifth to the first tank, figure 1 shows a schematic representation of the proposed
configuration.
Fig. 1. General benchmark plant layout
4. Process models
The IAWQ Activated Sludge Model No 1 [4] was chosen to simulate the biological
processes. The double-exponential settling velocity model proposed by Takács et al. [14]
was selected to describe the behaviour of the settler. Within the framework of the general
benchmark, several manipulated variables are defined. In particular, oxygen supply, the
internal and external recycle flow rates, and the waste flow rate are defined as potential
manipulated variables.
5. Test protocol
The test protocol begins with a series of procedures meant to verify the simulation
75
software being used. The first procedure involves simulating to steady state (or, if
preferred, for lOx the sludge age) the described layout with all process variables constant
including all flows and influent constituents. The characteristics of the prescribed influent
are defined and represent the flow-weighted average concentrations of a 14-day dry
weather influent data file [2,15]. Following verification of the steady state output data, the
dynamic response of the simulation is checked by first using a dynamic dry weather
influent file as input to the system.
Time (days)
Fig. 2. A typical dry weather day as depicted in the dry weather datafile.
The dry weather file is one of three weather files (each two-weeks in duration) generated
to test the dynamic response of the system [2,15]. This dry weather file exhibits
characteristic diurnal variations in flow and component concentrations. Also incorporated
in the file is a substantial (20%) decrease in flow and load during the ‘weekend’. Figure 2
shows a sampling from a typical day of data. The second and third files are based on the
dry weather data with an added rain event during the second week. The first of these rain
files has, during the second week, a simulated rain event that results in a constant increase
in influent flow and lasts for approximately 2 days. In particular, this file depicts a
constant hydraulic load increase without any increase in COD or nittogen load as
compared to the dry weather file. The second of these rain files has two storm events
during the second week. These storm events are shorter in duration than the rain event,
but are more intense. Also, in addition to increasing the hydraulic load, these storm
events have an associated increase in particulate COD load as compared to the dry
weather data.
Verification of the dynamic response is done using all three of the weather files
following simulation with three weeks of dry weather data. That is, from the steady state
results the system is simulated using the dry weather file as input and the state of the
76
entire system is saved. Then, in turn, each of the three weather files is used from that
saved state. As the first week of each file is the same, this provides three weeks of dry
weather data response followed by one week of alternative weather. Analysis of the
system response is calculated based on the output data from the fourth week. At each
stage, the output data are verified with published data given in the benchmark definition
[3]. It is assumed that the model implementation is correct once there is agreement
between the published data and the user’s output data.
To further validate the user’s software, a basic control strategy is proposed. That is,
prior to defining and testing a new control strategy users must validate thefr software by
implementing a defined control strategy. Again, the generated output can be compared to
a standardised output as defined in the benchmark description. The basic control strategy
has two parts and consists of:
• dissolved oxygen (DO) control in the 5 th tank (set-point 2 mg L 1) by manipulation of

the aeration rate via the oxygen transfer coefficient, KLa (in the 5* tank).
• nitrate control in the 2nd unaerated tank (set-point of 1 mg L 1) by manipulation of the
internal recycle flow rate (5th -> 1st tank).
When performing the control test, the system is again simulated to steady state (or, for
lOx the sludge age if preferred), but this time with active controllers. Following this, the
dry weather file is applied and then each of the weather files is used as described
previously, also with active controllers. The results are then compared to the available
data. By following these steps and validating the results, it is assumed that the simulator
is correctly tuned for further control strategy evaluations.
6. Performance assessment
Performance assessment of the controlled system is made at two levels. The first level
concerns the local control loops and is assessed by:
• IAE (Integral of the Absolute Error)

• ISE (Integral of the Squared Error)
• maximum deviation from set-points
• error standard deviation
Basically, this serves as an indication that the proposed control strategy has been applied
properly.
The second level quantifies the effect of the control strategy on plant performance and it
can be divided into three sub-levels:
• effluent quality
• operational costs
• potential increase in capacity
Within the general benchmark, effluent quality is considered through an effluent quality
77
index, which is meant to quantify into a single term the effluent pollution load to a
receiving water body. Constraints with respect to the effluent quality are defined and the
percentage of time that the constraints are not met is reported. As well, the number of
violations also is reported based on a 15-min sampling time. The operational costs are
considered through three terms: sludge treatment/disposal, pumping and aeration energy,
and an empirical estimation of the wear and tear on the pumps due to the control strategy.
Finally, a potential increase in capacity (PIC) is calculated for the controlled case and is
defined as the potential increase in influent flow that could be obtained using the
proposed control strategy. The PIC is an empirical attempt to quantify the savings
resulting from a potential increase in capacity of the plant as opposed to a plant
expansion.
By defining the benchmark in these specific terms and setting up the simulations in
this way, the user ensures a consistent and unbiased methodology for the evaluation of
future control strategies.
7. Application to respirometry-based control strategies
Control strategies can be evaluated by model simulation and respfrometry-based control

strategies are no exception. However, as already stated, the methodology used in the
evaluation is a critical step and is the most efficient way to ensure unbiased comparisons.
This implies that to make unbiased comparisons, each control strategy must be evaluated
under the same conditions. It also implies that the effect of the control strategy must be
compared to a fully defined and suitable reference output. Only then is it possible to truly
evaluate a control strategy and compare it with another strategy.
In an attempt to standardise control strategy evaluation, the general benchmark
simulation concept of the COST 624 Working Group on Operation [8] was evaluated as a
tool for evaluating a number of respfrometry-based control strategies. Although the
general benchmark is based on a benchmark developed for the evaluation of
respirometry-based strategies [11], it has since undergone several changes. Hence, it was
of interest to investigate the applicability of the general benchmark to this specific family
of confrol strategies.
The main characteristic of respirometry-based control strategies is that the measured
variable, and thus the controller input, is either a respfration rate or a variable deduced
from a respfrometric experiment. Because of this characteristic, respirometry-based
control strategies are a particular family of control strategies that need a dedicated
simulation tool for thefr evaluation and comparison. In short, the general benchmark
provides an excellent starting point, but it does not provide sufficient flexibility for the
evaluation of respkometry-based control strategies. The following section outlines a
number of respirometry-based strategies and the changes in the benchmark that are
needed for a complete evaluation.
7.1. JOYCE ETAL. (1974)
• Strategic objective - to decrease the variability in the effluent (as defined by the
BOD5) and to improve the effluent quality
• Measured variables - respkation rate, DO, 5 min settling volume, influent total
78
organic carbon (TOC), return sludge suspended organic carbon (SOC)

• Controlled variables - F/M, DO, respkation rate and 5 min settling volume
• Manipulated variables - influent flow distribution (i.e. step-feed), waste sludge,
recycle sludge, and ak flow rates
Joyce et al. devised a complicated algorithm to control a full-scale fully aerobic plant
receiving a significant fraction of its influent flow (up to 75%) from several food
processing plants. Necessary changes to the general benchmark include aeration to all
tanks, TOC, soc and sludge settling test measurements as well as step-feed capability for
the influent flow.
7.2. STENSTROM AND ANDREWS (1979)
• Strategic objective - to maintain a constant biological growth rate in a reactor

subjected to a diurnal variation in influent load
• Measured variables - respkation rate, mixed liquor volatile suspended solids (MLSS),
influent flow rate
• Confrolled variables - SRT, specific respkation rate
• Manipulated variables - waste and recycle sludge flow rates, influent flow
distribution (i.e. step-feed)
To evaluate the Stenstrom and Andrews control strategies using the benchmark several
changes would be needed including a fully aerobic nitrifying plant layout, with stepfeed
capability for the influent flow. Another variation relates to the evaluation of the
effectiveness of the control strategy which Stenstrom and Andrews related to the variance
in the controlled variable, specific respkation rate.
7.3. S0RENSEN (1980)
• Strategic objective - to decrease the variability in the effluent quality (that is, the
author’s aim was to devise a strategy which decreased the variance in the effluent
quality, and not necessarily improve the effluent quality)
• Measured variable - blower speed (to maintain constant dissolved oxygen, DO) in the
last tank of a fully aerobic tanks-in series plant layout
• Controlled variable - (derived) respkation rate in the last tank
• Manipulated variable - influent flow distribution (i.e. step-feed)
S0rensen outlined a control strategy based on a derived oxygen utilisation rate in the last
tank of a fully aerobic three-tanks in series layout. As applied to the general benchmark,
several changes are required including a flow proportional recycle flow rate, aeration in
all tanks, and step-feed capability for the influent flow.
7.4. SEKINE ETAL. (1988)
• Strategic objective - to maintain optimal conditions for microbial activity (i.e. to

diminish DO limitations in the aeration tank) by controlling the DO setpoint at a
adequate level
• Measured variables - influent flow rate, endogenous and aeration tank outlet
79
respfration rate, DO, mixed liquor suspended solids (MLSS)

• Controlled variables - DO, critical specific respiration rate
• Manipulated variable - air flow rate
Sekine et al. developed a hybrid DO conttol strategy that used respiration rate
measurements to optimally determine an appropriate DO set-point. The system studied
was different from the general benchmark in several ways. The system was fully aerobic
and nitrifying. It was operated at a long (15 day) SRT and had a feed-forward control
loop based on influent flow measurements for DO in the aeration tanks. Unlike the
general benchmark, waste sludge was not continuously removed from the system, but
rather was removed once per day.
7.5. KIM ETAL. (1996)
• Strategic objective - to maintain a constant organic loading rate

• Measured variable - respfration rate
• Controlled variable - loading rate
• Manipulated variable - influent flow rate
Kim et al. linearly related measured mixed liquor respfration rate to organic loading rate
and devised a control strategy for maintaining a constant loading rate through the
manipulation of the influent flow rate. However, to evaluate such a strategy using the
benchmark, the benchmark would have to allow for manipulation of the influent flow
what it doesn’t. Further, the system studied by Kim et al. was fully aerobic and nitrifying
with a solids retention time of 17 days.
7.6. LAROSE ET AL. (1997)
• Sttategic objective - to optimise the cycle time in a biological phosphorus removing

sequencing batch reactor
• Measured variable - respiration rate
• Controlled variable - anaerobic rapidly biodegradable COD concentration
• Manipulated variable - anaerobic cycle time
Larose et al. applied a respfrometric technique to a sample of sludge from the anaerobic
stage of a cyclical SBR being used for biological phosphorus removal. Several problems
become immediately apparent when an attempt is made to apply the general benchmark
to this strategy. For instance, biological phosphorus removal is not included in the
biological process model ASM1. Also, the SBR has intermittent feed and this strategy has
potential variable cycle lengths (or variable volumes, if translated) that are not defined in
the general benchmark description [3].
8. Overview and discussion of results
From the vast literature of respfrometry-based control [12], a selection of six

respfrometry-based control sfrategies was examined for the applicability of the general
benchmark as developed by COST. Table 1 lists some of the necessary changes needed to
80
make the general benchmark applicable. The general benchmark can be used to evaluate
respfrometry-based control strategies. However, for each strategy a number of benchmark
features must be modified especially those related to the controller and plant layout. In
particular, several features, not included in the general benchmark, are commonly
associated with respirometry-based strategies. For instance, often these strategies are
applied to fully aerobic systems including short SRT, carbon removing only plants and
longer SRT, nitrifying systems. Also, manipulation of the influent flow (i.e. step-feed) is
a popular mechanism.
Table 1. Overview of necessary changes to the general COST benchmark for the evaluation of a selection of
respirometry-based control strategies.
Strategy Fully Need for Step-feed External Variable Need for
aerobic process and respừometer influent evaluation
process flow criteria
model changes
changes
Joyce et al.
Stensttom and ự z ự
ự ự
ự
ự
Andrews
S0rensen ự ự ự
Sekine et al. ự ự ự
Kimetữ/. ự ự ự ự
Laiose etal. ự ự ự
In an attempt to make the general benchmark applicable to this family of confrol

strategies, several necessary additions are suggested. Figure 3 shows a schematic
representation of the slightly modified layouts. In this case, two new fully aerobic designs
were developed including one for carbon removal and one for nitrification. Also, step-
feed capability was added to all three designs. The adopted layouts are all 5 tanks-in-
series designs, but of course not all proposed sừategies are applied to this type of
configuration. So, for those strategies, assumptions need to be made to fit the strategies to
the 5 tanks-in-series benchmark layout and may include approximating the source of
samples. Changes to the general benchmark should be limited to those that are requked,
and those that have a dkect and significant impact on the control sttategy performance.
Performance assessment remains the most important part of any strategy evaluation.
The general benchmark lays out a performance assessment procedure for evaluating
control skategies, but the applicability of such a procedure is highly dependent on the
strategy objective. The general benchmark assessment assumes that the objective of all
strategies is to improve effluent quality or decrease the cost of treatment. As shown in
this paper, that is not always the case. Table 1 shows that the general benchmark
evaluation criteria are not totally applicable to respkometty-based skategies. That is, in
many respkomeky-based skategies the objective is to decrease the variability in a given
parameter and not necessarily improve it [6,10,13]. In these instances, the performance
assessment defined in the general benchmark is unsuitable because it does not provide a
81
means to evaluate whether or not the strategy was successful in achieving its stated
objective. Hence, the evaluation assessment must be modified to be applicable in these
cases. Further work to develop a comprehensive and applicable assessment technique
clearly is needed. In particular, an evaluation criteria for variability assessment is
requked.
Fig. 3. Schematic representation of the altered plant layouts. Layout A is consistent with the carbon-only and
nitrifying layouts showing 5 completely mixed and aerated tanks in series. Layout B is a schematic of the
denitrifying layout with tanks 1 and 2 mixed and unaerated, and tanks 3, 4 and 5 aerated.
Strategies based on respkomeky include the need to measure a respkation rate. In many
instances, as in the Larose et al. strategy outlined above, there may be a need to measure
respkation rate in a separated mini-reactor, i.e. a respkometer. These measurements often
are complicated and need to be accounted for in any evaluation procedure because they
may have a dfrect impact on the performance of the conttol strategy. For instance, time
delays, which would be inherent in any measurement of respfration rate, need to be taken
into account. Further, respfrometric measurements are often obtained through
sophisticated experiments, which may combine substrate and biomass from various
sources within the process. As such, these experiments often need to occur outside the
process itself, in a separated envfronment. Also, when variables are deduced from
respiration rate data, the practical methodology of deducing the variables needs to be
considered.
Take for example a strategy, which aims to maintain a constant active biomass
concentration. As it is not possible to measure active biomass in practice, an alternative
measuring approach must be devised. For instance, it might be proposed that the
endogenous respfration rate be used as a measurable indicator of the active biomass. In
practice, the endogenous respiration rate may be measured by sampling sludge from a
completely mixed reactor during low loading conditions or from the end of a plug-flow
reactor. Alternatively, the active biomass concentration may be estimated from the
maximum respfration rate measured in an external apparatus using an excess of a defined
readily biodegradable substrate. Such a procedure would requfre a separate experimental
set-up. There are several ways to measure indicators of active biomass, but for
benchmarking, the problem remains as to how practice is represented in the modelling
82
envfronment. That is, any model simulation study which aims to evaluate a given control
strategy must mimic (as closely as possible) what can be done in practice.
In a simulation envfronment, using ASM1, it is possible to simply calculate the
concentration of active biomass from the state variables; however, as already stated, in
practice this is not possible. So, to more closely approximate what can be measured in
reality, the endogenous respfration rate might be proposed as a modelled indicator of the
active biomass. Unfortunately, in ASM1, the endogenous respfration rate is not well
defined; and therefore, a dfrect correlation between respfration rate and active biomass
concenttation may not be an accurate assumption in the modelling envfronment. An
alternative approach for simulating the measurement of the endogenous respiration rate is
to model an external un-fed recfrculating plug-flow reactor to approximate a practical
method. The maximum respiration rate might also be used. In either of these cases, the
respfrometry experiment would have to be modelled because the procedure can have a
dfrect impact on the effectiveness of the control strategy. However, such capabilities are
not included in the general benchmark.
The benchmark is a comprehensive description of a simulation environment and
evaluation procedure including the definition of a plant layout, a simulation model,
influent loads, test procedures and evaluation criteria. A benchmark, by definition, must
be independent of the simulation software being used. That is, the simulation software
should have no effect on the modelling output such that different simulators modelling
the same system should give the same result. Because of the many simulator specific
options, this is not always the case, nor is it a trivial task to ensure similar results using
different simulators. A substantial effort has gone into this aspect of the benchmark
development. However, by stipulating specific model equations, modelling procedures
and simulator specific options, similar results can be achieved. It should be noted that
synchronising the output is not limited to commercially available simulation software.
That is, similar results should be attainable irrespective of the wastewater simulation tool
that is used including user defined computer code. Demonstrating similar results in this
way is the first step in the evaluation procedure, and ensures that the simulators are tuned
in the same way and also ensures the consistent comparison of control strategies. Four
different simulation platforms have been used (BioWin™, GPS-X™, Simba™ and
WESTF™) in order to verify that the results were independent of the software.
9. Conclusions
In the paper we describe a number of respừometry-based control strategies and identify

the features of the COST general benchmark, which need adjustment in order to make it
applicable to these strategies. The results of this investigation clearly demonstrate that
development of a universally applicable benchmark is not possible because of the many
variables and permutations that can occur in the activated sludge process. However, the
general benchmark does provide a reasonable starting point for evaluations. That is, the
procedures described in the benchmark definition for tuning the simulation software are
universally applicable. These procedures are valid and provide a standardised method for
software/model verification even if several changes to the benchmark are required for the
evaluation of a particular control strategy.
83
Acknowledgements
The authors are pleased to acknowledge the paid corporate sponsors who have made this
work possible: Aquafin (Belgium), Biotim n.v. (Belgium), Dow Chemical (USA), Eco
Process and Equipment Inti. Inc. (Canada), ENEA (Italy), Hemmis n.v. (Belgium),
Hydromantis Inc. (Canada), International Water Association (United Kingdom), CIRSEE
- Lyonnaise des Eaux-Durnez (France), Severn Trent Water (United Kingdom), STOWA
- Dutch Foundation for Applied Water Research (The Netherlands), WRc (United
Kingdom).
References
[1] Alex J., Beteau J.F., Copp J.B., Hellinga c., Jeppsson u., Marsili-Libelli s., Pons M.N., Spanjers H. and.
Vanhooren H. (1999) Benchmark for evaluating control strategies in wastewater treatment plants.
Proceedings of the European Control Conference, ECC ’99, Karlsruhe, Germany, August 31-September
3,1999.
[2] Copp J. B. (1999) Development of standardised influent files for the evaluation of activated sludge control
strategies. IAWQ Scientific and Technical Report Task Group: Respirometry in Control of the Activated
Sludge Process - internal report.
[3] COST 624 Action website (http://www.ensic.u-nancy.fr/COSTWWTP) - The European Co-operation in the
field of Scientific and Technical Research.
[4] Henze M., Grady Jr C.P.L., Gujer w., Marais G.V.R. and Matsuo T. (1986) Activated sludge model No.l,
IAWQ Scientific and Technical Report No.l, IAWQ, London.
[5] Joyce, R.J., Ortman, c. and Zickefoose, c. (1974) How to optimise an activated sludge plant. Wat. Sewage
Wks 121, 96-99 .
[6] Kim C.W., Choi E.H., Kim B.G., Lee T.H. and Park T.J. (1996) Stable loading control in a field-scale
activated sludge plant using on-line respiration meter. Proceedings of the IAWQ’S 18th Biennial
International Conference and Exposition - ‘Abstracts - Poster Presentations’, June 23-28,1996.
[7] Larose, A., Perrier, M. and Comeau, Y. (1997) Respừometric control of the anaerobic period duration of an
SBR Bio-P process. Wat. Sci. Technol. 36 (5), 292-300.
[8] Pons M. N., Spanjers H. and Jeppsson Ư. (1999) Towards a benchmark for evaluating control strategies in
wastewater treatment plants by simulation. Proceedings of 9th European Symposium on Computer Aided
Process Engineering, Budapest, Hungary, May 31-June 2,1999.
[9] Sekine, T., Sato, s., Furuya, N. and Sunuhara, H. (1988) Supervision and control of the activated sludge
process utilising the respiration rate activity. Envừon. Technol. Lett. 9,1317-1326.
[10]Sprensen, P.E. (1980) Evaluation of operational benefits to the activated sludge process using step feed
control strategies. Prog. Wat. Technol. 12 (6), 109-125.
[11]Spanjers H., Vanrolleghem p., Nguyen K. Vanhooren H. and Patry G.G. (1998) Towards a benchmark for
evaluating control strategies in wastewater treatment plants by simulation. Wat. Sci. Technol., 37(12), 219-
226.
[12]Spanjers H., Vanrolleghem p., Olsson G. and Dold P.L. (1998) ) Respừometry in Control of the Activated
Sludge Process: Principles, IAWQ Scientific and Technical Report No.7, LAWQ, London.
[13]Stenstrom, M.K. and Andrews, J.F. (1979) Real-time control of the activated sludge process. J.
Environ.Engng 105, 245-260.
[14]Takács I., Patry G.G. and Nolasco D. (1991) A dynamic model of the clarification thickening process, Wat.
Res., 25,10,1263-1271.
[15]Vanhooren H. and Nguyen K. (1996) Development of a simulation protocol for evaluation of respừometry-
based control strategies, Technical Report, University of Gent, Gent, Belgium.
84
FAULT DETECTION AND ISOLATION IN WASTEWATER TREATMENT
PLANTS
Comparison of different approaches and experimental results
JEAN-PHILIPPE STEYER AND JEROME HARMAND

Laboratoire de Biotechnologie de VEnvironnement - INRA, Avenue des
étangs, 11100 Narbonne, France email: steyer@ensam.inra.fr
Abstract
This paper presents the problem of detecting and isolating faults in biological waste
water treatment plants. Depending on the specific problems to be solved and on the
knowledge available on the process, model-based, fuzzy logic-based and artificial neural
network-based approaches are investigated and tested on different processes. Finally,
some conclusions and perspectives are drawn.
1. Introduction
Due to the increasing complexity and necessity of safety in industrial processes, efficient
diagnosis systems are becoming more and more important. Indeed, even in normal
operation conditions, several types of disturbances can be present and they can largely
affect the operating conditions of the process. A clear need for advanced control and
diagnosis systems is thus expressed in order to keep the system performances as close as
possible to the optimal conditions. This fact is particularly true in wastewater treatment
processes (WWTPs) where the normative constraints are becoming more and more
stringent.
Anaerobic digestion processes are commonly used for the treatment of industrial
wastewaters and theừ use presents a number of advantages. In particular, anaerobic
sludge production is much lower compared to aerobic activated sludge processes. The
organic matter is also anaerobically degraded into a gas mixture of carbon dioxide and
methane, thus producing valuable energy. In addition, this process is able to operate
under severe conditions : high strength effluents and short hydraulic retention time. For
all these reasons, anaerobic digestion processes are particularly well adapted for
concentrated wastes such as agricultural (e.g. plant residues, animal wastes) and food
industry wastewater. However, biological processes in general - and anaerobic
87
Jean-Philippe Steyer and Jerome Harmand
digestion processes in particular - exhibit some very specific behaviours and the conttol
needed is not the same as it would be usually understood by a control engineer : see for
instance the survey paper by Andrews (1994) in which it is stressed that, in biological
processes, things seem to work fafrly well until some failures or faults occur. Hence,
there is an increasing need for Fault Detection and Isolation systems (also known as FDI
systems).
Automatic FDI systems use analytical or heuristic knowledge in order to detect as
early as possible deviations from the normal operation that tend to degrade the overall
system performance. Such malfunctions may occur in the sensors, the actuators, the
components of the process (e.g., inhibition of the biomass in a biological reactor) or in
the control system, if the process is running in closed loop.
Model-based FDI methods have been first initiated by Willsky (1976). These
methods can be divided into two categories. The first method is based on state estimation.
It includes detection filters presented by White and Speyer (1987), parity space
approaches developed by Chow et al. (1986), Patton and Chen (1991), and Delmafre et
al. (1994), as well as works of Frank (1993) on diagnostic observer based methods.
Parameter estimation techniques, proposed in particular by Isermann (1984) belong to the
second category.
In parallel to these model-based methods, there has been rapidly a growing interest
for using techniques such as artificial neural networks or fuzzy logic for fault detection
purposes (e.g., Frank, 1994; Isermann and Balle, 1997) and few demonstrations of the
advantages of fuzzy-based FDI systems can be found in the literature (see for example
Ulieru and Isermann, 1993; Montmain and Gentil 1993; Kiupel and Frank 1996; Gfraud
and Aubrun, 1996; Genovesi et al. 2000). Other recent studies use this technique for
sensor validation (Boudaoud and Masson, 1998) or combine it with artificial neural
networks (Steyer et al. 1997).
However, a good FDI system for a biological process also requfres inclusion of
specific biological knowledge. Artificial Intelligence (Al) methods such as knowledge-
based systems or fuzzy logic allow one to bring new insights into biological process
control and to introduce a "biological dimension" (See for example Konstantinov and
Yoshida, 1992; Aynsley et al. 1993; Steyer et al. 1993; Slimes et al. 1995, Roca et al.
1996; Steyer et al. 1996).
This paper presents the advantages and the drawbacks relevant to three design
methods respectively based on explicit mathematical models, Fuzzy Logic and Artificial
Neural Networks (ANNs). In section 2, the model-based and ANN-based approaches are
applied to an anaerobic biological process. Then, in an other context, a fuzzy-based
approach is used to adapt functioning conditions for the supervision of an industrial
equalization process. Experimental results are presented and analysed before some
conclusions are given.
2. Examples of fault detection and isolation methods for anaerobic digestion
processes
2.1. THE EXPERIMENTAL PROCESS
2.1.1. Characteristics of the wastewater
88
Fault detection and isolation in wastewater treatment plants
The wastewater to be treated is an industrial wine distillery wastewater. The wine

industries have changing operating conditions characterised in particular by an increase
of activity in spring and autumn months while decreasing the rest of the year. Theừ
rejected wastewaters contain the following elements: sugars, alcohol, esters, organic
acids, yeast and bacteria. Theừ pH is about 4 and, to improve the treatment efficiency,
these wastewaters have to be neutralised before entering the reactor. In addition, theừ
Chemical Oxygen Demand (COD) is about 10 to 45 times more important than in urban
waste waters. The general characteristics are summarised in table 1.
Table 1: Characteristics of wine distillery wastewaters
Component Range Mean value
Volatile Fatty Acids (g/1) [5.00 - 6.00] 5.50
Total Organic Carbon (g/1) [2.50 - 6.00] 4.25
Phenol (mg/1) [90.0 - 275] 182.5
COD (g/1) [9.00 -17.4] 13.2

Soluble COD (g/1) [7.60 -16.0] 11.8
Total Suspended Solids (g/1) [2.40 - 5.00] 3.7
Volatile Suspended Solids (g/1) [1.20-2.70] 1.95
Alkalinity (meq/1) [30.8 - 62.4] 46.6
pH [4.00 - 5.40] 4.7
2.1.2. The anaerobic digestion process

The experiments were performed using an anaerobic up-flow 120 litre fluidised bed
reactor at the "Laboratoire de Biotechnologie de I'Environnement" of INRA in Narbonne,
France. The process is associated with a 520 litres cừcular settler. The feeding tank
consists in 3 tanks (27 m 3 each) that are connected to an intermediate 20-litre buffer tank
by pipes of about 0.4 m3. Figure 1 describes the pilot plant.
The reactor is made of a 3.65 meter high cừcular column. Support for bacteria
attachment is made of small volcanic particles of 300 pm mean diameter. Fluidisation of
the bed is ensured by a high liquid recừculation flow rate (i.e., 630 1/h using a centrifugal
pump). The 20 litre buffer system has a floater in order to regulate the volume and a
degassing system which eliminates the gas contained in the liquid phase
of the effluent to be treated. A remote controllable peristaltic pump ensures the

connection between this tank and the process.
89
Fig. 1: Synthetic synoptic view of the 120 litre anaerobic digestion fluidised bed reactor
The input liquid is mixed into the recycled flow just before entering a heat exchanger.
This heat exchanger is used to regulate the temperature of the liquid phase into the
reactor (nominal set-point is 35°C). The heated liquid is then introduced at the bottom of
the reactor. The gas is collected at the top of both the column and the settler. The liquid
from the top of the reactor is collected by overflow and rejected in the public sewage
system.
Four different local control loops are present within this process:
• the pH of the influent is regulated at 6.3 in the 20 litre buffer tank using a local
PID confroller,
• the temperature of the liquid phase is regulated in the recycled loop by a local
PID controller at the set-point 35 °C,
• the recirculation flow rate is controlled through a controllable valve using a thfrd
local PID controller,
• finally, the output gas flow rate is regulated around 110 1/h by adjustment of the
input liquid flow rate. The control input is saturated at 30 1/h for security
reasons.
The sensors and actuators are connected to an input/output device that allows the
acquisition, the treatment and the storage of data on a PC. For this purpose the ’’Modular
SPC" software developed by the French company S.E.R.I. Envfronnement is used. This
software performs advanced control law calculations as well as process supervision.
2.2. THE MODEL-BASED FDI APPROACH
The first example is concerned with the closed-loop detection and isolation of an actuator
problem; see Harmand (1997) and Harmand and Steyer (1998) for further details. This
study was in fact initiated since the experiments run for several years reveal either
90
continuous or abrupt deviations between the expected values and the measurements of
the Input Liquid Flow Rate (ILFR). This problem is very important since it can lead to
wrong actions that can degrade nominal performances of the process (i.e., under or over
feeding of the reactor according to the sign of the disturbance). Several reasons can
explain these observed biases:
• the practical experience has clearly established that, most of the time, slow- acting
deviations are the consequence of a clogging of the pipes used to feed the reactor
with the wastewater.
• the abrupt deviations can be explained, depending on the magnitude of the bias,
either by a leak in the input pipes or by the appearance of a gas bubble into the
sensor.
In fact, the present FDI model-based approach is integrated with a control approach
called "Disturbance Accommodating Control" or DAC (Johnson 1976). Mathematically
speaking, the DAC is the association of the "Disturbance Modelling Principle" (DMP)
with a classical regulator, the DMP assuming that the disturbance signal (here the bias on
the ILFR) is a linear combination of four different basic functions: constants, ramps, sine
or polynoms. When the disturbances are modelled by these mathematical functions, the
model of the process is augmented by the model of the disturbances. Notice however that
the higher the degree of the disturbance model, the higher the degree of the controller.
Then, under appropriate observability hypothesis, the DAC allows US to efficiently
control the process while the use of a state estimator (e.g., a Kalman filter) allows us to
on-line estimate the process state together with the expected disturbances without
actually measuring them. If disturbances vary with time, this strategy holds if the
dynamics of the estimator are faster than those of the disturbances.
There are a number of advantages to use this approach:
• this FDI method can be used without the DAC, that is either in open-loop or in
closed-loop configurations,
• in case of a closed-loop configuration, the method makes explicit use of the structure
of the system to estimate and attenuate the negative effects of a disturbance on the
controlled variables.
A number of experiments have been performed in order to test this integrated FDI/control
methodology in closed-loop. In particular, the accommodation of an actuator constant
bias has been tested while controlling the output gas flow rate at a set point of 110 1/h
using the input liquid flow rate (See figure 2). At t = 100 h, an artificial actuator bias of
-2 1/h was applied to the ILFR until t = 165 h. To this end, the calibration parameter of
the input pump confrolling the ILFR was voluntarily and
91
manually biased. It means that when the output of

the controller is u (what is called the free-
disturbance-expected-value of the input), then
the real input (called the measured or real value
of the input) of the process equals (w-2) instead
of u (see figure 2.a). In such a case, the
controller reacts almost immediately in
estimating the bias (see the artificially applied
Ỗ ]0J j; Time (h)

$5 115 135 155 175
Fig. 2.a: Free-disturbance-expected-value (black-

bias and its estimation
line) and the real value (grey line) of the ILFR (in in figure 3) and
IZh) corrects the ILFR signal
in order to maintain the output gas flow rate
around the functioning point (see the
accommodation results in figure 2.b). Finally, it
is to be noticed that the exact same FDI/control
integrated approach could be used for other WWTPs
(for an application to a nitrification process,
see Harmand et al. 1996 and 1997).
Fig. 2,b: Regulation of the OGFR at 110 1/h in the
presence of an actuator bias of -2 l/h magnitude
92
Fig. 2: Accommodation of the control law and OGFR regulation

2.3. THE ANN-BASED FDI APPROACH
The previous method is perfectly adapted when the process is run in closed-loop and
when the identified model is valid. However, if these conditions are not fulfilled, the
DMP principle cannot be applied and other techniques are to be used.
95 115 135 155 175
Fig. 3: Actuator bias (bold line) and its estimation (thin line) in Uh
93
In the following, a hybrid approach that uses both fuzzy logic and artificial neural
networks is presented for on-line detection and analysis of problems occuưing when the
process is run either in open or closed-loop (for further details, see Steyer et al. 1997).
94
g pS 1.0
I (b)
0.8
o 0.6
0.4
0.2
Time (h)
0.0
0 10 20 30 40 50 60 70
Fig. 4: Comparison of fault detection using fuzzy qualification and artificial neural network
when the organic loading rate is changed (measurements are centred between 0 and 1). ("Fuzzy
DiagJSin" is the signal indicating the presence of a fault on the input of the process using the
fuzzy logic approach and "ANN DiagJSin" has the same meaning but it is generated from an
artificial neural network. When these signals are close to 0, there is no fault and their increase
indicates the presence of a fault)
The raw data available on the process (i.e., pH, temperature, recừculation flow rate, input
flow rate and gas flow rate) and theừ signal characteristics (i.e., mean value, variance,
etc.) are pre-processed using fuzzy logic to build a vector of features (i.e., a pattern
vector). This feature vector is classified into a pre-specified category (i.e., a class) that is
a state of the system, according to discrimination fuzzy rules. An artificial neural network
is then used to classify the process states and to identify the faulty or dangerous ones: if
there is no fault, the signal provided by the FDI system is equal to zero. On the other
hand, if a dangerous fault is detected and isolated, the signal is set to 100 %. In between
these two extremes, the higher the value of the signal, the more dangerous a fault is.
This approach was developed to handle in real time problems such as, for example,
foam forming, sudden changes in the effluent to be treated (due to a change in
concentration), pipe clogging (due to struvite formation) or bad temperature regulation
95
(due to improper setting of the control parameters).

In the example presented in figure 4, the influent liquid flow rate (ILFR) was disturbed
on purpose (see Fig. 4.a) without indicating it to the FDI procedure (i.e., the ILFR
measurements provided to the FDI were constant and equal to the ILFR value at t = 0).
Effects of these changes can be seen on the output gas flow rate (Fig. 4.b). Only
analysing the gas flow rate measurements, the fuzzy FDI procedure and the ANN FDI
procedure provided respectively the results in figure 4.C and in figure 4.d. It can be seen
that both approaches are able to detect efficiently the changes of the ILFR with a better
detection though using the ANN FDI procedure.
3. Fuzzy supervision of an industrial equalization process
3.1. THE EQUALIZATION PROCESS
Any industrial WWTP is subject to large magnitude variations in both the influent flow
rate and in the influent pollutant concentration and it is well known that either flow or
concentration shockwaves can have catastrophic consequences on the downstream
biological processes. To deal with this problem, a solution consists in implementing an
equalization system at the primary stage of the WWTP to minimise the influence of the
input disturbances on the downstream biological treatment processes. Such systems
consist in a number of buffer tanks interconnected through pipes, pumps and on/off
valves that can be either manually or automatically controlled.
Equalization systems are thus used:
• to overcome the operational problems caused by flow rate and load variations,
• to improve the performance of the downstream processes,
• to reduce the size and cost of the plant.
The equalization process under interest in the study is represented in figure 5 where
EQOi (i = 1 to 3) represents the three interconnected buffer tanks, Q represents the liquid
flow rates and c represents waste water concentration measured as Total Organic Carbon
concentration or TOC. The control problem can then be stated as developing algorithms
that automatically operate pumps and eventually on/off valves so that constant - or nearly
constant - flow rates and output concentrations are achieved before being introduced in
the treatment plant itself. In terms of control, this problem is very challenging since the
control system must take into account hard constraints. These consttaints are related to
the physical capability of the equalization system to face increases in the influent flow
rate reflected by the total capacity of the interconnected tanks as well as the pump
characteristics.
Qinl> Cini
Qout - <21 +
ổ2 CM =
TOC
Fig. 5: Configuration of the equalization system
96
3.2. THE FUZZY SUPERVISOR
Very important problems can arise even though the equalization process is carefully
controlled (see Devisscher et al., 1999, De Clercq et al., 1999 and Harmand et al., 1999).
In fact, the control schemes give good performance but only when the buffer tanks are
neither full nor empty. In these two extreme situations, the control system misses the
necessary degree of freedom to buffer the flow rate and the TOC output concentration at
the same time. In fact, this comes from the inability of the control algorithms to fulfil the
control objectives.
In order to better understand what happens, let US consider the two following
extreme cases. On one hand, if it is assumed that the input flow rate is significantly
higher than its expected mean value for a long period of time, then, without any adaptive
ability of the control system, it is obvious that, after a given time, tanks will overflow. On
the other hand, if it is now assumed that the input flow rate is significantly smaller than
its mean value for a long period of time, then it is quite obvious that the mean values
have to be decreased to avoid emptying the tanks. The ’’adaptation ability” of the control
system has also to deal with different time constants. In other words, the "corrections" to
be applied to the flow rates computed by the controllers have to deal with both short and
long time disturbances. In order to overcome this difficulty, a fuzzy- based supervisory
system and a long-term filter are added to a classical control strategy.
This supervisor operates both short time and long time corrections on the flow rates
computed by the controllers in order to avoid saturation of the volumes over short and
large periods of time. The supervisor is tested using a data set obtained from the real
process over a two-month period. The simulation results are provided in figure 6. From
these simulations, different statistical characteristics were obtained and summarised in
table 2. The general structure of the overall supervisor system is presented in figure 7.
Notice that these statistical results have only a relative value from a control
engineering perspective. Indeed, they only have a real pertinence when concerned with a
classical regulation problem evaluation. However, this is not exactly the case in this
fuzzy model-based approach since the set-point (due to the corrections computed by the
fuzzy system) is adjusted depending on the input disturbances. As a consequence, the set-
point is time-varying and the variance loses its absolute physical meaning.
97
Fig.6 a: Input (thin line) and output (bold line) flow rates with the supervisor
Fig. 6b: Model-based control: input (thin line) and output (bold line) TOC concentrations with
the supervisor
Time (h)
Fig. 6c: Input (thin line) and output (bold line) loading with the supervisor
Fig. 6d: Volumes (VỊ to V3from thinner to thicker lines) with the supervisor
The following conclusions can be drawn from these results:
• The control objectives are quite well fulfilled. Indeed, Ổout is almost constant and the
output waste concentration is very well attenuated compared to the input.
• At the same time, using the fuzzy adaptation system, it is possible to manage the
overall time period without any saturation in the volumes while variations of the
both input flow rates and input TOC concentration are quite well attenuated.
Table 2: Statistical evaluation of the equalization performance
Variances
Flow rate Waste Concentration Loading
Influent 209 1.5 106 2.8 10lu
Effluent 55 3.9 ỈÕ4 5 6.6'ĨÕ9
Fig. 7: Configuration of the global control system of the equalization process (Q are the mean values of the flow
rates)
mathematical model of the process can be obtained, then it can be expected that a model-
based method will give satisfactory results. On the other hand, as it often happens in the
case of a WWTP, if only qualitative expert knowledge is available, fuzzy-based methods
will rather be used. However, another context is common when dealing with the V/WTP:
a large amount of data is available but the structure for a general mathematical model is
very difficult to define. In this case, an approach based on the ANN usually will give
efficient and pertinent results for FDI purposes.
4 Conclusions
5Basically, the choice for an FDI approach is determined by the problem to be solved and
by the knowledge that is available on the process to be diagnosed. On one hand, if a
References
Andrews J.F. (1994) Dynamic control of wastewater treatment plants, Envữon. Sci. Technol. 28(9), 434A- 439A.
Aynsley M., Hofland A.J., Morris A.J., Montague G.A., Di Massimo c. (1993) Artificial intelligence and the
supervision of bioprocesses (Real-time knowledge-based systems and neural networks) in A. Fiechter ed.,
Bioprocess Design and Control, Springer-Verlag, Advances in Biochemical Engineering/Biotechnology
48,1-27.
Boudaoud, A.N. and Masson, M.H. (1998) An adaptive fuzzy diagnosis system for on-line sensor data
validation. 3th IFAC Workshop on On-line Fault Detection and Supervision in the Chemical Process
Industries, Lyon, France.
Chow E.Y., x.c. Lou, G.c. Verghese and A.s. Willsky (1986) Redundancy relations and robust failure detection,
in Basseville M. and A. Beneveniste Eds., springer Verlag, Berlin, Detection of Abrupt Changes in Signals
and Dynamic Systems, Lecture Notes in Control and Information Sciences 77, 275- 294.
De Clercq B., B. Vanderhaegen, J. Harmand, and p. A. Vanrolleghem (1999) Evaluation of a rule-based control
strategy for an equalization facility with technical/physical constraints, European Control Conference
ECC'99, 8 pages (on CD-ROM), Kalsruhe, Germany, 31st August-3rd September 1999.
Delmaire G., J.p Cassar and M. Staroswiecki (1994) Identification and parity space techniques for failure
detection in SISO systems including modelling errors, 33rd IEEE CDC Conf., 2279-2285, Lake Buena Vista
(USA).
Devisscher M., J. Harmand, J. p. Steyer, B. Vanderhaegen and p. A. Vanrolleghem (1999) A combination of
fuzzy and linear control techniques for the equalization of an industrial wastewater treatment plant, 18 th
Benelux Meeting on Systems and Control, Houthalen, Belgium, March 3-5.
Frank P.M. (1993) Advances in observer-based fault diagnosis, TOOLDIAG'93 Conference, 3, 817-836,
Toulouse (France).
Frank, P.M. (1994) Application of fuzzy logic to process supervision and fault diagnosis. IF AC Symp. on Fault
Detection Supervision and Safety for Technical Processes, SAFEPROCESS'94, Espoo, Finland, 2, 531
538,13-16 June 1994.
Giraud, D. and Aubrun, c. (1996) A fuzzy fault diagnosis method applied to a steam circuit. IEEE International
Conference on Fuzzy Systems FUZZIEEE'96, 3, 1944-1950, New Orleans, Louisiana, USA.
Genovesi, A., Harmand, J. and Steyer, J-P. (2000) Integrated fault detection and isolation : Application to a
winery's wastewater treatment plant. Applied Intelligence Journal (APIN), 13,207-224.
Harmand J., R.E. Skelton, J-Ph. Steyer (1996) Disturbance accomodation control of a nitrification process with
structural uncertainties, IEEE-IMACS Conference CESA'96, 70-75, Lille, France, 9-12 July 1996.
Harmand J., J-Ph. Steyer, R.E. Skelton (1997) On disturbance accommodating control of biological processes,
Forum for Applied Biotechnology FAB'97, Gent, Belgium, 25-26 September 1997.
Harmand J. (1997) Identification et commande avancée de precedes biologiques de dépollution, Ph.D. Thesis (in
French), University of Perpignan, France, September, 18th 1997, 192 pages
Harmand J., J-Ph. Steyer (1998) The disturbance accommodating control of biological processes, IFAC-
EurAgEng International Workshop on "Decision and Control in Waste Bio-Processing", WASTE-
DECISION'98, 7 pages (on CD-ROM), Narbonne, France, 25-27 February 1998.
Harmand J., J. p. Steyer, M. Devisscher, B. De Clercq, B. Vanderhaegen and p. A. Vanrolleghem (1999)
Advanced control of an industrial equalization system, European Control Conference ECC'99, 8 pages (on
CD-ROM), Kalsruhe, Germany, 31st August-3rd September 1999.
Isermann R. (1984) Process fault detection based on modelling and estimation methods - A survey, Automatica
20, 387-404.
Isermann, R. and Ballé, p. (1997) Trends in the application of model-based fault detection and diagnosis of
technical processes, Contr. Eng. Pract. 5(5), 709-719.
Johnson c. D. (1976) Theory of disturbance-accommodating controllers, Control and Dynamic Systems :
Advances in Theory and Applications, 18, Editor c. Leondes, London.
Kiupel, N. and Frank, P.M. (1996) Fuzzy supervision for an anaerobic wastewater plant. IEEE-IMACS Int. Conf
CES A'96, Line, France, 1, 362-367.
Konstantinov, K.B. and Yoshida, T. (1992) The way to adequate control of microbial processes passes via real-
time knowledge-based supervision. J. Biotechnol. 24, 33-51.
Montmain, J. and Gentil, s. (1993) Decision-making in fault detection : A fiizzy approach. Int. Conf.
TOOLDIAG'93, Toulouse, France.
Patton R.J. and J. Chen. (1991) A review of parity space approaches to fault diagnosis, Proc. IFAC/IMACS
Symp. SAFEPROCESS'91,1, 239-255, Baden-Baden (Germany).
Roca, E., Flores, J., Rodríguez, I., Cameselle, c., Nunez, M.J. and Lema, J.M. (1996) Knowledge-based control
applied to fixed bed pulsed bioreactor, Bioproc. Eng. 14,113-118.
Siimes, T., Linko, p., von Numers, c., Nakajima, M. and Endo, I. (1995) Real-time fiizzy-knowledge-based
control of baker's yeast production. Biotechnol. Bioeng., 45, 135-143.
Steyer, J-P., Queinnec, I. and Simoes, D. (1993) Biotech : a real-time application of artificial intelligence for
fermentation processes. Cont. Eng. Pract. 1(2), 315-321.
Steyer, J-P., Queinnec, I., Capit, F. and Pourciel, J-B. (1996) Qualitative rules as a way to handle the biological
state of a fermentation process: an industrial application. Journal Européen des Systèmes Automatisés
RAIRO-APII, 30(2/3), 381 -398.
Steyer J.Ph., Rolland D., Bouvier J.c. and Moletta R. (1997) Hybrid fuzzy neural network for diagnosis :
Application to the anaerobic treatment of wine distillery wastewater in fluidised bed reactor. Wat. Sci.
Technol. 36, 209-217.
Ulieru M. and Isermann R. (1993) Design of a fiizzy-logic based diagnostic model for technical processes. Fuzzy
Sets and Systems 58, 249-27 L
White J.E. and J.L. Speyer. (1987) Detection filters design : spectral theory and algorithms, IEEE Transaction on
Automatic Control, 32, 593-603.
Willsky A.S. (1976) A survey of design methods for failure detection in dynamic systems, Automatica 12, 601-
611.
CALIBRATION OF ACTIVATED SLUDGE MODELS: A CRITICAL REVIEW
OF EXPERIMENTAL DESIGNS
B. PETERSEN V K. GERNAEY s M.HENZE3, P.A.

VANROLLEGHEM 1
1
BIOMATH Department, Ghent University, Coupure Links 653, B-9000
Gent, Belgium. E-mail: Peter.Vanrolleghem@rug.ac.be
2
EP AS n.v., Technologiepark2, B-9051 Zwijnaarde, Belgium.
3
Department of Environmental Science and Engineering, Technical
University of Denmark, Building 115, DK-2800 Lyngby, Denmark.
Abstract
This review begins with an overview of literature data on methodologies that have been
applied in other studies to calibrate Activated Sludge Model No. 1 (ASM1). An attempt
was made to gather and summarise the information needed to achieve a successful model
calibration, and based on this a general model calibration procedure is proposed. The
main part of the literature review is devoted to the different methods that have been
developed and applied for the characterisation of wastewater and reaction kinetics in
relation to ASM1. The methodologies are critically discussed and it is attempted to
illustrate the power of the different methods for characterisation, all within the frame of
ASM1 calibration. Finally, it is discussed which waste water components and parameters
are most relevant to be characterised via lab-scale experiments. This discussion also
includes the problem of transferability between lab-scale and full-scale observations, and
potentially different model concepts. One of the most discussed experimental factors
determining the experimental response is the ratio between initial substrate and biomass
concentration (S(0)/X(0)). A separate section is focusing upon this factor.
1. Introduction
One of the most widespread biological wastewater treatment techniques is the activated
sludge process. In this process, a bacterial biomass suspension is responsible for the
removal of pollutants. Depending on the design and the specific application, an activated
sludge wastewater treatment plant can achieve biological nitrogen removal and biological
phosphorus removal, besides removal of organic carbon substances. The increased
knowledge about the mechanisms of different biological processes taking
101
Petersen B., Gemaey K., Henze M., Vanrolleghem
P.A.
place in an activated sludge plant was translated into dynamic models that were developed
to describe the degradation processes in the activated sludge plant. This review will focus
on the Activated Sludge Model No. 1 (ASM1) (Henze et al., 1987), which through the
years has been the state-of-the-art model for activated sludge plants with biological
nitrogen removal.
2. Description of the state-of-the-art activated sludge models
In the following the model concepts of ASM1 (Henze et al., 1987) and the recent
modifications leading to ASM3 (Gujer et al., 1999) are described. A description of
ASM2/ASM2d (Henze et al., 1995, 1999) is, however, not included since phosphorus
removal is not dealt with in this review.
2.1. ACTIVATED SLUDGE MODEL No.l (ASM1)
ASM1 is presented in a matrix format in table 1 according to Henze et al. (1987). Many of
the basic concepts of ASM1 were adapted from the activated sludge model defined by
Dold (1980). Some of the central concepts (the different model components and
processes) of ASM1 are summarised below. For further details the reader is referred to the
IAWQ Task group reports.
2.1.1. COD components in ASM1

COD is selected as the most suitable parameter for defining the carbon substrates as it
provides a link between electron equivalents in the organic substrate, the biomass and
oxygen utilised. In ASM1 the COD is subdivided based on (1) solubility, (2)
biodegradability (3) biodegradation rate and (4) viability (biomass):
• The total COD is divided into soluble (S) and particulate (X) components.
• The COD is further subdivided into non-biodegradable organic matter and
biodegradable matter. The non-biodegradable matter is biologically inert and passes
through an activated sludge system in unchanged form. The inert soluble organic
matter (Si) leaves the system at the same concentration as it enters. Inert suspended
organic matter in the waste water influent (Xi) or produced via decay (Xp) becomes
enmeshed in the activated sludge and is removed from the system via the sludge
wastage.
• The biodegradable matter is divided into soluble readily biodegradable (S s) and
slowly biodegradable (Xs) substrate. Already here it should be stressed that some
slowly biodegradable matter may actually be soluble. The readily biodegradable
substrate is assumed to consist of relatively simple molecules that may be taken in
dfrectly by heterotrophic organisms and used for growth of new biomass. On the
contrary, the slowly biodegradable substrate consists of relatively complex molecules
that requfre enzymatic breakdown prior to utilisation.
102
Calibration of activated sludge models: a critical review of experimental designs
• Finally, heterotrophic biomass (XBH) and autotrophic biomass (XBA) are generated by
growth on the readily biodegradable substrate (Ss) or by growth on ammonia nitrogen
(SNH)- The biomass is lost via the decay process where it is converted to Xp and x s
(death regeneration, see below).
Summarising, the total COD balance of ASM1 is defined by Eq. 1 and further illustrated
in figure 1.
CODtot — Sj + s<5 + Xj + x<5 +Xgjj "J"XBA + XP (1)
Fig.l. COD components ỉn ASM1 and ASM3 (figure modified from Jeppsson, 1996), components
specifically related to ASM3 are given in bold and the ones only related to ASM1 are given in
italics
2.1.2. Nitrogen components in ASM 1
Similar to the organic matter, total nitrogen can be subdivided based on (1) solubility, (2)
biodegradability and (3) biodegradation rate:
• Total nitrogen can be subdivided into soluble (S) and particulate (X) components.
• The nitrogen is divided into non-biodegradable matter and biodegradable matter. The
non-biodegradable particulate organic nitrogen (XNI) is associated with the non-
biodegradable particulate COD (XỊ or Xp), whereas the soluble non-biodegradable
organic nitrogen (SNi) is assumed to be negligible and therefore not incorporated into
the model.
103
P.A.
Table 1. The ASM1 process matrix (Henze et al., 1987) (conf on next page)
5 6 9
3 7 10
Component (i) 1 2 4 X X 8 S 11
X B B X N SN
—> ị Process (j) Si Ss i Xs H A p So O H SND
1 Aerobic growth of
heterotrophic
biomass
1 1-YH ~b
1 YH CB
YH
2 Anoxic growth of
heterotrophic 1
biomass Y 1 "ixB
H
3 Aerobic growth of
autotrophic 1
4.57-YA 1
biomass 1 Y -ĨXB
YA — YA
A
4 Decay of
heterotrophic
biomass 1- - f
fp 1 p
5 Decay of
autotrophic
biomass 1- - f
fp 1 p
6 Ammonification OÍ -1
soluble organic
nitrogen 1
7 Hydrolysis of
slowly
biodegradable 1 -1
substrate
8 Hydrolysis of 1
organic nitrogen
104
Table 1. The ASM1 process matrix (Henze et al., 1987) (conf from previous page)
12 13
Process rate (Pj)
XND SALK
1
SJJ So Y
ixB 14 AnaxH Y V Q XBH
Ks + K0H + 0Q
2
1~YH *XB n „
rmaxH „ „ „
S
s__________KOH___________SNO
_ „ BH
. Y_._
14-2.86-YH 14 Ks + òs K0H + so KNQ + SN0
3
2 .. $NH____________So
ixB “max A' 1Q Y c ^BA
14YA 14 K
NH +Ò
NH OA + ÒO
K
4
ixB “fp
• XBH
ixp
5
ixB - fp • ’ XBA
ixp
6
2
ka ■ SND • XBH
14
7 Xs/
v , / XBH_____________So K
OH___________SN0 v
K K
Kv + sAr OH + ỒO OH + 50 KNO + SNO
x
/ XBH
-1 ^•(XND/XS)
• The biodegradable nitrogen is subdivided into ammonia nitrogen (SNH), nitrate + nitrite
nitrogen (SN0), soluble organic nitrogen (SND) and particulate organic nitrogen (XND).
The particulate organic nitrogen is hydrolysed to soluble organic nitrogen in parallel
with hydrolysis of the slowly biodegradable organic matter (X s) (either present in the
wastewater or produced via the decay process). The soluble organic nitrogen is
converted to ammonia nitrogen via ammonification. Ammonia nitrogen serves as the
105
P.A.
nitrogen source for biomass growth (the parameter ỈXB indicates the amount of
nitrogen incorporated per COD unit). Finally, the autotrophic conversion of ammonia
results in nitrate nitrogen (SNo), which is considered to be a single step process in
ASM1.
Summarising, the total nitrogen balance for the components in ASM1 is defined by Eq. 2
and further illustrated in figure 2.
Ntot = SNH H-SNP +SNO 4-XNP +XNI +ỈXB (XBH +XBA)+iXp Xp (2)
Fig. 2. Nitrogen components in ASM1 (modified from Jeppsson, 1996); components specifically related to ASM3
are given in bold and the ones only related to ASM! in italics
2.1.3. Processes inASMl

Basically there are four different main processes defined in ASM1 (Henze et al., 1987):
• Growth of biomass
• Decay of biomass
• Ammonification of organic nitrogen
• Hydrolysis of particulate organic matter
The substrate flows in ASM1 are illustrated in figure 3.
106
Fig.3. Substrate flows inASMl andASM3 (modified from Gujeretal., 1999)
2.1.3.1. Aerobic growth of heterotrophic biomass Growth takes place by degradation of

soluble readily biodegradable substrate (S s) under the consumption of oxygen (So).
Ammonia nitrogen (SNH) is incorporated into cell mass, as described above. Both the
concentrations of Ss and So may be rate limiting for the growth process. The Monod
relationship is used to describe the growth of heterotrophic and autotrophic organisms.
2.1.3.2. Anoxic growth of heterotrophic biomass (denitrification) In the absence of

oxygen the heterotrophic organisms are capable of using nitrate as the terminal electron
acceptor with Ss as substtate resulting in biomass growth and nitrogen gas. The same
Monod kinetics as used for aerobic growth is applied except that the kinetic rate
expression is multiplied by a correction factor T|g (<1). This factor is accounting for the
fact that the anoxic substrate removal rate is slower compared to aerobic conditions. This
can either be caused by a lower maximum growth rate or because only a fraction of the
heterotrophic biomass is able to denitrify. Furthermore, anoxic growth is inhibited when
oxygen is present which is described by the switching function KOH/(KOH+SO)- The
coefficient KOH has the same value as in the expression for aerobic growth. Thus, as
aerobic growth declines, the capacity for anoxic growth increases.
2.1.3.3. Aerobic growth of autotrophic biomass (nitrification) Ammonia nitrogen (SNH)

is oxidised to nitrate resulting in production of autotrophic biomass. Furthermore, a part of
the SNH is also incorporated in the autotrophic cell mass. As for heterotrophic growth the
concentrations of SNH and So can be rate limiting for the process. Nitrification has a
considerable effect on the alkalinity (SALK)-
2.1.3.4. Decay of heterotrophic biomass The death regeneration concept of Dold (1980)
is applied to describe the different reactions that take place when organisms die. The
traditional endogenous respừation concept describes how a fraction of the organism mass
disappears to provide energy for maintenance. However, in the death regeneration concept
oxygen is not directly associated with microbial decay. Decay is assumed to result in the
release of slowly biodegradable substrate that is recycled back to soluble substrate and
used for more cell growth. Thus, the oxygen utilisation normally associated directly with
decay is calculated as if it occurs indừectly from growth of new biomass on released
107
P.A.
substrate. A parallel conversion of organic nitrogen to ammonia nitrogen occurs. It should

be noted that the magnitude of the decay coefficient used in this approach is different
from that of the endogenous respừation. In endogenous respfration the loss of one unit of
biomass COD leads to the utilisation of one unit of oxygen minus the COD of the inert
particulate products that are formed. However, in the death regeneration model the loss of
one biomass COD unit results in the ultimate formation of one unit of COD due to the
formed readily biodegradable substrate minus the formed inert particulate products. When
the readily biodegradable COD is used for cell synthesis, only a fraction of a unit of
oxygen (determined by the yield) will be requừed because of the energy incorporated into
the cell mass. That cell mass undergoes in turn decay etc. before the unit of oxygen is
finally removed.
Summarising, to give the same amount of oxygen utilisation per time due to the decay
process, the decay rate coefficient must be larger for the death regeneration concept than
if a more traditional endogenous decay process was adopted. This has the effect that the
cell mass turnover rate increases, resulting in a higher microbial growth rate in the death
regeneration model.
2.1.3.5. Decay of autotrophic biomass The decay of autotrophs is described similar to the
heterotrophic decay process.
2.1.3.6. Ammonification of soluble organic nitrogen (S ND) Biodegradable soluble organic

nitrogen (SND) is converted to ammonia nitrogen (SNH) in a first order process. Hydrogen
ions consumed in this conversion process result in an alkalinity change.
2.1.3.7. Hydrolysis Slowly biodegradable substrate (X s) enmeshed in the sludge is broken

down producing readily biodegradable substrate (S s). The degradation of slowly
biodegradable matter has appeared rather important to realistic modelling of activated
sludge systems because it is primarily responsible for realistic electton acceptor profiles
(Dold, 1980). This process is modelled on the basis of surface reaction kinetics and occurs
only under aerobic and anoxic conditions. The hydrolysis rate is reduced under anoxic
conditions in the same way as anoxic growth, by applying a correction factor T|h (<1).
The rate is also first order with respect to the heterotrophic biomass concentration present
but saturates, as the amount of entrapped substrate becomes large in proportion to the
biomass.
2.1.4. Restrictions ofASMl

A number of restrictions concerning ASM1 are summarised below (Henze et al., 1987):
• The system must operate at constant temperature.
• The pH is constant and near neuttality. It is known that the pH has an influence on
many of the parameters, however only limited knowledge is available to be able to
express these possible influences. Consequently, a constant pH has been assumed.
The inclusion of alkalinity in the model, however, does allow for detection of pH
problems.
• No considerations have been given to changes in the nature of the organic matter
within any given wastewater fractions (e.g. the readily biodegradable substrate).
108
Therefore, the parameters in the rate expressions have been assumed to have constant
values. This means that only concentration changes of the wastewater components
can be handled whereas changes in the wastewater character can not.
• The effects of nutrient limitations (e.g. N and P) on the cell growth have not been
considered. It is, however, easy to add limitation terms in the model if needed.
• The coưection factors for denitrification (r| g and T|h) are fixed and constant for a
given wastewater, even though it is possible that theừ values are depending on the
system configuration.
• The parameters for nitrification are assumed to be constant and to incorporate any
inhibitory effects that wastewater constituents may have on them.
• The heterotrophic biomass is homogeneous and does not undergo changes in species
diversity with time. This assumption is inherent to the assumption of constant kinetic
parameters. This means that any changes in substrate concentration gradients, reactor
configuration, etc. on sludge settleability are not considered.
• The entrapment of particulate organic matter in the biomass is assumed to be
instantaneous.
• The hydrolysis of organic matter and organic nitrogen are coupled and occur
simultaneously with equal rates.
• The type of electron acceptor present does not affect the loss of biomass by decay.
• The type of electron acceptor does not affect the heterotrophic yield coefficient.
• ASM1 is developed for simulation of treatment of municipal waste water, and it is
therefore not advised to apply the model to systems where industrial contributions
dominate the characteristics of the waste water.
• ASM1 does not include processes that describe behaviours under anaerobic
conditions. Simulations of systems with large fractions of anaerobic reactor volume
may therefore lead to eưors.
• ASM1 can not deal with elevated nitrite concentrations.
• ASM1 is not designed to deal with activated sludge systems with very high load or
small sludge retention time (SRT) (<1 day).
2.2. ACTIVATED SLUDGE MODEL NO. 3 (ASM3)
ASM3 is presented in matrix form in table 2. In the development of ASM3 some

limitations of ASM1 were evaluated, and combined with the experiences gained with the
application of ASM1 the following list of “defects” of ASM1 was defined (Gujer et al.,
1999):
• ASM1 does not include expressions to deal with nitrogen and alkalinity limitations.
• ASM1 considers biodegradable soluble and particulate organic nitrogen as model
components. These can, however, not easily be measured and may in most cases
unnecessarily complicate the use of ASM1.
• The ammonification kinetics can not be easily quantified, and moreover this process
is typically rather fast and does therefore not affect model predictions significantly.
• ASM1 differentiates between inert suspended organic matter present in the influent
wastewater and produced within the activated sludge process. In reality, however, it is
impossible to distinguish between these two components.
• Hydrolysis has a rather dominating effect upon the predictions of the oxygen
109
P.A.
consumption and denitrification by heterotrophic organisms. In reality this process

includes different coupled processes such as hydrolysis, lysis and storage of
substrates. Therefore, the identification of the kinetic parameters of this combined
process is difficult.
• The death regeneration concept is covering lysis combined with hydrolysis of
released substrate and subsequently growth on this subsfrate. In reality it is difficult
to determine the decay coefficient related to the death regeneration concept.
• Elevated concentrations of readily biodegradable organic substrates can lead to
storage of poly-hydroxy-alkanoates, lipids or glycogen. This process is not included
in ASM1.
• ASM1 does not include the possibility to differentiate between decay rates of
nitrifiers under aerobic and anoxic conditions. This may lead to problems with the
predictions of the maximum nitrification rates in cases of high SRT and high fractions
of anoxic reactor volumes.
The main difference between ASM1 and ASM3 is the recognition of the importance of
storage polymers in the heterotrophic conversions in the activated sludge processes in
ASM3. The aerobic storage process in ASM3 describes the storage of the readily
biodegradable substrate (Ss) into a cell internal component (XST0). This approach requfres
that the biomass is modelled with cell internal structure similar to ASM2. The energy
requfred for this process is obtained via aerobic respfration. This internal component is
then subsequently used for growth. In ASM3 it is assumed that all Ss is first taken up and
stored prior to growth. Thus, a division of the storage and growth process, allowing
growth to take place on external substrate dfrectly, is not considered.
Furthermore, the death regeneration concept is replaced by endogenous respfration, which
is closer to the phenomena observed in reality. Endogenous respiration can readily be
obtained from a simple batch test (see below, section 4.1.3.1). Also, ASM3 allows a
differentiation between aerobic and anoxic decay.
Figure 3 illustrates the difference in COD flows between ASM1 and ASM3. The first
thing to notice is that the conversion processes of both groups of organisms (autotrophs
and heterotrophs) are clearly separated in ASM3, whereas the decay regeneration cycles
of the autotrophs and heterotrophs are strongly interrelated in ASM1. This change of
decay concept (and introduction of the storage step) means that there exist more “entry”
points for oxygen utilisation resulting in, at some points, easier separation and
characterisation of the processes. Second, there is a shift of emphasis from hydrolysis to
storage of organic matters. This gives a change in how wastewater characterisation should
be defined since the separation between Ss and x s now should be based on the storage
process rather than on the growth process. Still, the separation remains somewhat based
on biodegradation rates. In ASM3 hydrolysis is obviously of a less dominating importance
for the rates of oxygen consumption since only hydrolysis of x s in the influent is
considered.
Below the components and processes of AMS3 are summarised focusing on the
differences between ASM1 and ASM3.
2.2.7. COD components in ASM3

The COD components in ASM3 are basically defined in the same way as in ASM1. Only
110
the separation between inert suspended organic matter in the wastewater influent (Xi) and
produced via the decay process (Xp) is no longer maintained, and, second, the component
XSTO is introduced, as described above. The substrate Ss goes through the storage process
but is basically still biodegradable. Thus, the total COD balance is defined by Eq. 3 and
further illustrated in figure 1, where the components specifically related to ASM3 are
given in bold and the ones only related to ASM1 are given in italics.
CODtot = SỊ + Sg + Xj +x<s + X +XẬ +X<5JQ (3)

2.2.2. Nitrogen components inASM3
The nitrogen balance in ASM3 is simplified compared to ASM1, since the soluble and
particulate organic nitrogen components are no longer considered. Furthermore, a nitrogen
gas component (SN2) is included allowing for a closed nitrogen mass balance. The nitrogen
incorporated in Si, Ss, Xi, Xs, and the biomass is defined in ASM3 as a fraction of these
components. This fraction is consumed or produced when the corresponding COD
fraction is formed or degraded respectively. Summarising, the total nitrogen balance for
the components in ASM3 is defined by Eq. 4, and further illustrated in figure 2. Again, the
components specifically related to ASM3 are shown in bold and the ones related to ASM1
in italics.
Ntot = SNH + SNO + SN2 + ỈNSI ■ SI + ỈNSS * Ss + ÌNXS ■ xs + ifciBM • (XH + XA) + ÌNXI • XI (4)
111
P.A.
Table 2A. Rate expressions and stoichiometry ofASM3 (Gujer et al., 1999)
j Process Process rate equation pj., all pj > 0
X /X
k- S H X-
1 Hydrolysis H
KX + XS/XH H
Heterotrophic organisms, aerobic and denitrifying activity

Aerobic storage of S
V . O2__________Ss . Y_
2 STO IT 4-Q Ko+So H
Ss K
O2 + SO2 Ks + Ss
V_____„_____________ KO2__________$NQX___________$s . Y__

Anoxic storage of STO 7NOX Y■q Y
3 q Y1c
Ss K
O2 + SO2 K
NOX + SNOX K +s
s s
_ $02____________$NH4_____________$ALK____________XST0/XH
4 Aerobic growth ' • XH
K
O2 + $02 KNH4 + $NH4 KALK + $ALK KST0 + XSTO IXH
„ . JJ . K
O2_____________$N0X_____________$NH4 .
/NOX
KO2+SO2 KNOX+SNOX KNH4+SNH4
Anoxic growth
5 $ALK_____________XST0/XH
(denitrification)
K
ALK + $ALK KSTO + XSTO IXH
Aerobic u . $02 Y„
6 endogenous b
H,02 • ~ • XH
02 + S02
K
respừation
Anoxic u_____________K02_____________$N0X . Y-
7 endogenous DH.NOX • ~ AH
K
02 + S02 XNOX + SNOX
respữation
Aerobic u______________$02 . Y_____
8 respừation of tJ
ST0,02 • , c • ASTO
K
02 + 5O2
XsTO
Anoxic respừation t,_______________K02_____________$N0X . Y_____
9 b
ST0,N0X • ~ • ASTO
of XsTO K
02 + s02 KN0X + sN0X
Autotrophic organisms, nitrifying activity

1 Aerobic growth of ... . $02_______________$NH4_______________$ALK .Y.
0 XA, Nitrification K
A,02 + $02 KA,NH4 + $NH4 KA,ALK + $ALK
Aerobic
1 u____________$02________ Y .
endogenous DA,02 ■ -AA
1 A A, 02 + s02
respiration
Anoxic
1 u _________ KA,02__________________$N0X_________Y .
endogenous D
A,N0X • ,q Y 1c ■ Aa
2 K
A,02 + s02 KA,N0X + sN0X
respứation
112
Table 2B. Rate expressions and stoichiometry of ASM3 (Gujer et al., 1999)
compound
1 2 3 4 5 6 7 8 9 10 11 12 13
i>
SN SN SAL Xs
j Process S02 Si Ss H4
SN2
OX K
Xi Xs XH
TO
XA Xss
expressed as CO CO Mo CO CO CO CO CO
V
> o2 D D
N N N
le D D D D D ss
0.0 -
1 Hydrolysis fsi 1 i
01
-1
0.75
Heterotrophic organisms, aerobic and denitrifying activity

1-
Aerobic YS - 0.0 0.0 0.8
2 1 3 02 5
0.51
storage ofSs TO,
2
-
Anoxic - 0.0 0.0 0.0 0.8
3 1 3 7
0.0
07 0
0.48
storage ofSs 7
- - - -
Aerobic -
4 0.6 0.0 0.0 1 1.6
0.06
growth 0 7 05 0
Anoxic - - -
0.3 0.0 -
5 growth 0.0
0
0.3
16
1 1.8
0.21
(denitrific.) 7 0 5
Aerobic -
0.0 0.0 0.2 -
6 endog. 0.8
66 05 0
-1
0.75
respừation 0
Anoxic -
0.0 0.2 0.0 0.2 -
7 endog. 66 8
0.2
25 0
-1
0.75
respừation 8
Aerobic -
8 respừation of -1 -1
0.60
XsTO
Anoxic 0.3
-
0.0 -
9 respừation of 5
0.3
25
-1
0.60
5
XsTO
Autotrophic organisms, nitrifying activity
Aerobic - - -
1 4.1
18. 4.2 0.6 1 0.90
0 growth 0fXA 7
04 4 00
Aerobic -
0.0 0.0 0.2 -
11 endog. 0.8
66 05 0
-1
0.75
respừation 0
Anoxic -
1 0.0 0.2 0.0 0.2 -
endog. 66 8
0.2
25 0
-1
0.75
2 8
respiration
2.2.3. Processes inASM3
In ASM3 there are also four basic processes, however, slightly different from ASM1
(Gujer et al., 1999):
113
P.A.
• Storage of readily biodegradable substrate

• Growth of biomass
• Decay of biomass
• Hydrolysis of particulate organic matter
2.2.3.1. Aerobic storage of readily biodegradable substrate This process describes the
storage of readily biodegradable substrate (S s) in the form of XSTO with the consumption
of oxygen. As stated above, it is assumed that all S s first becomes stored material before
use for cell growth. It is realised that this is not in accordance with reality. However, no
model is currently available to predict the separation of Ss into dfrect growth and storage.
Gujer et al. (1999) therefore suggested to apply a low storage yield (YSTO) and a higher
growth yield (YH) to approximate dfrect growth.
2.2.3.2. Anoxic storage of readily biodegradable substrate This process is identical to the
aerobic storage, only is nitrate used as terminal electron acceptor instead of oxygen.
Furthermore, a correction factor (Ĩ|NO) is applied to indicate that only a fraction of the
heterotrophic biomass may be capable of denitrifying.
2.2.3.3. Aerobic growth of heterotrophs Aerobic heterotrophic growth takes place by

degradation of XSTO with the consumption of oxygen (So). Ammonia nitrogen (SNH) is
incorporated into cell mass, as described above for ASM1.
2.2.3.4. Anoxic growth of heterotrophs (denitrification) Anoxic growth is similar to

aerobic growth but respfration is based on denitrification. Again, a correction factor
CRNO) is applied to account for the observation of reduced anoxic respfration rates
compared to aerobic respfration.
2.2.3.5. Aerobic growth of autotrophs (nitrification) This process is described similarly to

ASM1.
2.2.3.6. Aerobic decay of heterotrophs The energy requfrements not associated with
growth but including maintenance, lysis, etc. are described by endogenous respiration in
ASM3 according to a simple first order reaction kinetics.
2.2.3.7. Anoxic decay of heterotrophs ASM3 allows for a description of anoxic decay in a
similar way as the aerobic decay process.
2.2.3.8. Aerobic and anoxic decay of autotrophs The decay of autotrophs is described in
the same way as the heterotrophic decay process.
2.2.3.9. Aerobic and anoxic respiration of storage products These processes are
analogous to endogenous respữation and ensure that the storage product X STO decays
together with the biomass.
2.2.3.10. Hydrolysis Just as in ASM1 hydrolysis is responsible for the breakdown of

slowly biodegradable substrate (X s) to readily biodegradable substrate (S s). However, in
ASM3 hydrolysis is assumed to be electton donor independent, and as stressed above the
hydrolysis does not play the same dominating role as in ASM1.
114
2.2.4. Restrictions of ASM3

The number of restrictions listed for ASM1 above (see 2.1.4) basically still holds for
ASM3, except for the restriction stating that the type of electron acceptor does not affect
the biomass decay.
3. Model Calibration
In this review model calibration is understood as the adaptation of a model to fit a certain
set of information obtained from the full-scale WWTP under study. This task is often
rather time-consuming, and typically the time needed for a model calibration is
underestimated. Even though more than a decade has passed since the publication of
ASM1, a fully developed model calibration procedure has not been defined yet. We have
not been able to find a complete model calibration report in literature. There may be many
reasons for this. Important to realise is that the purpose of a model being built is very
much determining on how to approach the calibration, making it difficult to generalise
(Henze et al., 1995). Still, considering the wide application of the activated sludge models
there are surprisingly few references that contain details on the applied model calibration
procedure. Most often it is not specified in detail how the model was calibrated but the
focus is more on the applications, e.g. for process scenarios and optimisations etc. Thus,
to obtain information on model calibration procedures one often has to collect bits and
pieces from various sources to obtain an overview.
Before going on with a discussion on how to approach a model calibration of ASM1,
it is relevant to define how parameter estimation is understood in this review and what the
difference is between parameter estimation and model calibration. Furthermore, the term
identifiability will be defined and the problem of identifiability with respect to ASM in
general will be addressed.
Parameter estimation consists of determining the “optimal” values of the parameters
of a given model with the aid of measured data. Here, the numerical techniques for
estimation will not be discussed, but reference is made to the literature (Robinson, 1985;
Vanrolleghem and Dochain, 1998). Only the basic idea behind parameter estimation is
schematised in figure 4. Initially, the model structures, of which selected parameters need
to be estimated, and the experimental data need to be defined. Moreover, first guesses of
the initial conditions, i.e. concentrations, and parameters, have to be given. The parameter
estimation routine then basically consists of minimising an objective function, which for
example can be defined as the weighted sum of squared errors between the model output
and the data. When the objective function reaches a minimum with a certain given
accuracy the optimal parameter values are obtained.
Thus, parameter estimation is carried out via specific mathematical search algorithms.
However, due to the high complexity caused by the numerous parameters and the
unidentifiable nature of the ASM models, it will be rather cumbersome to apply
mathematical calibration techniques.
Indeed, a major problem encountered in calibration of ASM is the (lack of)
identifiability of the model parameters. Identifiability is the ability to obtain a unique
combination of parameters describing a system behaviour. A distinction should be made
between theoretical and practical identifiability. Theoretical identifiability is a property of
115
P.A.
the model structure, and relates to the question whether it is at all possible to obtain
unique parameter values for a given model structure considering certain selected outputs,
and assuming ideal measurements. Practical identifiability, on the other hand, includes the
quality of the data. Thus, theoretically identifiable parameters may be practically
unidentifiable if the data are too noise corrupted (Holmberg, 1982; Jeppsson, 1996). This
subject is dealt with in great detail in Petersen (2000).
Fig.4. Illustration of parameter estimation routine (modified from Wanner et al., 1992)
Here, it should only be stressed that a typical problem related to the model calibration of
ASM is that more than one combination of influent characteristics and model parameters
can give the same good description of the collected data (Dupont and Sinkjaer, 1994,
Kristensen et al., 1998). Indeed, this indicates identifiability problems of either theoretical
or practical origin.
The model calibration of ASM is typically based on a step-wise procedure, and by
changing just a few of the many parameters instead of applying an automatic
mathematical optimisation routine. Based on the above statements concerning
identifiability problems it is, however, obvious that a calibration procedure where the
model parameters are changed by trial and error until a good description of the measured
data is reached is not advisable (Dupont and Sinkjaer, 1994, Kristensen et al., 1998).
Thus, it becomes important to gather as much information as possible that can help the
framing of realistic parameter combinations. In this review it was attempted to gather and
summarise the type of information needed for successful model calibration.
3.1. INFORMATION SET FOR MODEL CALIBRATION
116
The set of information that should be collected for successful model calibration was
extracted and combined from different sources (Henze et al., 1987; Henze, 1992; Lesouef
et al., 1992; Pedersen and Sinkjaer, 1992; Siegrist and Tschui, 1992; Stokes et al., 1993;
de la Sota et al., 1994; Dupont and Sinkjaer, 1994; Funamizu and Takakuwa, 1994;
Weijers et al., 1996; Xu and Hultman, 1996; Coen et al., 1997; Mino et al., 1997;
Kristensen et al., 1998), and is summarised below:
• Design data: reactor volumes, pump flows and aeration capacities.

• Operational data:
> Flow rates, as averages or dynamic trajectories, of influent, effluent, recycle and
waste flows.
> pH, aeration and temperatures.
• Characterisation for the hydraulic model, e.g. the results of tracer tests.
• Characterisation for the settler model: e.g. zone settling velocities at different mixed
liquor suspended solids concentrations.
• Characterisation for the biological model, ASM, of:
> Wastewater concentrations of full-scale WWTP influent and effluent (as well
as some intermediate streams between the WWTP’s unit processes), as
averages or as dynamic trajectories: e.g. ss, COD, TKN, NH4-N, NO3- N,
PO4-P etc.
> Sludge composition: e.g. ss, vss, COD, N and/or p content.
> Reaction kinetics: e.g. growth and decay rates.
> Reaction stoichiometry: e.g. biomass yields
The list does not discuss on how the particular information can be collected in practice,
since this will be discussed more in detail in the sections below.
As mentioned above, the required quality and quantity of information will depend
very much on the purpose of the modelling exercise. In case the model is to be used for
educational purposes (e.g. to increase basic understanding of the processes), for
comparison of design alternatives for non-existing plants or in other situations where
qualitative comparisons are sufficient, the default parameter values defined by Henze et
al. (1987) can be applied. A reasonably good description can most often be obtained with
this default parameter set for typical municipal cases without significant industrial
influences (Henze et al., 1997). However, if the calibrated model is going to be used for
process performance evaluation and optimisation, it may be necessary to have a more
accurate description of the actual processes under study. Some processes may need a more
adequate description than others depending on the purpose of the model calibration. This
may especially apply for models that are supposed to describe the processes in an
industrial or combined municipal and industrial treatment plant (Coen et al., 1997, 1998).
In such cases the wastewater characterisation, and thereby the activated sludge, may differ
significantly from standard municipal wastewater. In addition, special attention often has
to be paid to the characterisation of nitrification kinetics (e.g. Dupont and Sinkjaer, 1994),
since nitrification typically is the determining process for the process designs. Also, the
availability of readily biodegradable carbon substances is important for the successful
achievement of both denitrification and biological p removal, and may need to be
characterised in more detail (Coen et al., 1997).
In this review the focus will mainly be on the information needed for the biological
117
P.A.
model. Although not considered in detail, it should be stressed that the information listed
in the first 4 points is also very essential, and should not be neglected for a successful
model calibration. Major calibration problems can, for example, be related to rather
simple errors in the recording of operational data, e.g. eưoneous data of the waste sludge
measurements might result in an incoưect sludge balance (Melcer, 1999). Moreover good
characterisation of hydraulics and settling can be of great importance since e.g. poor or
erroneous hydraulic modelling may result in hydraulic effects being lumped into the
biological parameters of ASM1.
Fig.5. Schematic overview of the different general steps in an activated sludge model calibration procedure
The information needed for the characterisation of the biological model can basically be
gathered from three sources:
• Default values from literature (e.g. Henze et al., 1987).

• Full-scale plant data
> Average or dynamic data from grab or time/flow proportional samples.
> Conventional mass balances of the full-scale data.
> On-line data.
> Measurements in reactors to characterise process dynamics (mainly relevant for
SBR’s and alternating systems).
Information obtained from different kinds of lab-scale experiments with wastewater and
activated sludge from the full-scale plant under study.
Again, the intended use of the model will determine which information source to choose
for the characterisation of the different biological processes in the model. In addition, the
118
purpose will decide to which level the model has to be calibrated, since the quality of the
desừed model predictions will depend strongly on the quality of the model calibration.
Figure 5 illustrates the different general steps in a model calibration exercise. It should be
stressed that not all steps may have to be taken, depending on the purpose. This will be
discussed further with examples below, and the procedure has been concretised for a
municipal-industrial case study in Petersen (2000).
3.2. MODEL CALIBRATION LEVELS
Steps 1-5 in figure 5 indicate the collection of information. Design (1) and operational (2)
data are in general always needed for a model calibration. E.g. the flow and load
variations are important in the design of measuring campaigns for hydraulic, sludge
settling and biological characterisation of the full-scale WWTP. The hydraulics (3) are
typically characterised via tracer tests at the full-scale installation (De Clercq et al., 1999).
The settling properties (4) can be characterised via on-line or lab-scale settling tests
(Vanderhasselt et al., 1999). Finally, the biology can be characterised via different
information sources (see below).
In figure 5 steps 6-10 illustrate different calibration levels. The calibration of the hydraulic
model via tracer test results, and the settler model calibration via results from sludge
settling tests are indicated in steps 6 and 7 respectively. A first ASM calibration level is
typically a simple steady state model calibration (8).
3.2.1. Steady state model calibration

In this step data obtained from the full-scale WWTP are averaged, thereby assuming that
this average represents a steady state, and a simple model not including hydraulic detail is
calibrated to average effluent and sludge waste data. Typically, the calibrations of the
ASM and the settler are linked together, since the aim is most often to describe the final
effluent quality. Moreover, the recycle from the settler has an influence on the activated
sludge system. Thus, at this stage, there may be an interaction between the steady state
calibration and the settler model calibration, indicated in figure 5 with the double arrow.
Finally, the characterisation of wastewater components may be adjusted according to the
calibration of the full-scale model, indicated with the double arrow between (8) and (5) in
figure 5.
The next step in the calibration procedure is a steady state model calibration that includes
the hydraulic model (9). In general, with a steady state model calibration, only parameters
responsible for long-term behaviour of the WWTP can be determined, i.e. Y H, fp, bH and
Xi in the influent (Henze et al., 1999; Nowak et al., 1999). These parameters are
correlated to a certain degree, meaning that a modification of one parameter value can be
compensated by a modification of another parameter value. In the study of Nowak et al.
(1999) on mass balances of full-scale data, it was therefore chosen to fix Y H and fp,
leaving Xi in the influent and bH to be determined from the steady state data. In the study
of Lesouef et al. (1992), two WWTP models were calibrated via steady state calibration
only, and this calibrated model was applied to simulate dynamic process scenarios.
However, if one relies entirely on a steady state calibration some problems may be
encountered since the real input variations are usually faster than the slow process
dynamics that were focused upon during the steady state calibration. In other words, the
119
P.A.
process does not operate in steady state but one still attempts to fit a steady state
simplification of the model to an unsteady situation. A steady state calibration is,
however, very useful for the determination of initial conditions prior to a dynamic model
calibration and for the initiation of first parameter iteration (e.g. Pedersen and Sinkjaer,
1992; Stokes et al., 1993; Dupont and Sinkjaer, 1994; Xu and Hultman, 1996; Kristensen
et al., 1998).
3.2.2. Dynamic model calibration

If it is the aim to describe and predict more short-term and dynamic situations, a model
calibration to dynamic data will be needed since such data contain more information than
steady state data, especially on fast dynamic behaviour. The important point in model
calibration based on dynamic data is to obtain a more reliable estimation of the maximum
specific growth rates PmaxH and PmaxA (Henze et al., 1999), which are the most
important parameters in predicting dynamic situations.
At WWTP’s data are most often collected routinely with a daily or weekly sampling
frequency. This sampling frequency may, however, not be high enough, and for more
accurate modelling it may therefore be requừed to run special measuring campaigns (e.g.
Pedersen and Sinkjaer 1992; Dupont and Sinkjaer, 1994; de la Sota et al., 1994; Xu and
Hultman, 1996). The sampling frequencies should be chosen in relation to the time
constants of the process and influent variations. One of the important time constants of the
process is the hydraulic retention time (HRT). Ideally, one should choose to sample about
five times faster than the hydraulic retention time and have a test duration of 3-4 times this
key time constant (Ljung, 1987). However, since measurements on full-scale WWTP’s are
relatively expensive these recommendations may not always be completely fulfilled.
Furthermore, data from the full-scale installation alone may be insufficient for a
dynamic model calibration since the reaction kinetics can not be readily obtained from
such data, except for specific designs like SBR's and alternating systems (Vanrolleghem
and Coen, 1995). For a dynamic model calibration on a full-scale WWTP the modeller is
therefore typically aiming at combining more information rich results derived from lab-
scale experiments (carried out with sludge and wastewater from the full-scale installation)
with data obtained from measuring campaigns on the WWTP under study (Dupont and
Sinkjaer, 1994; Xu and Hultman, 1996; Kristensen et al., 1998).
In table 3 an attempt is made to gather and summarise the available literature
examples on model calibrations where detailed information is given on the model
calibration procedures. The table should not be regarded as a complete list of possibilities
but can serve as a starting point. The purpose of the different model calibrations is given
together with the applied calibration strategy. Furthermore, the information sources for the
characterisation of (1) waste water, (2) sludge, (3) kinetics and (4) stoichiometry, are
listed. Table 3 does not indicate the kind of experiments that may have been carried out to
gather the information, since this will be discussed in one of the next sections of this
review. The model parameters that are not mentioned in table 3 have either been taken
from literature or their origin may not have been clearly indicated in the references.
Considering waste water characterisation it is not always specified how the wastewater
information was converted into the wastewater components according to ASM1. In these
cases only the type of measurement (e.g. COD, TKN etc.) is listed in the table.
Based on table 3, it is obvious that the choice of information needed for the model
120
calibration is governed by the purpose. E.g. in the studies of Pedersen and Sinkjaer (1992)
and Dupont and Sinkjaer (1994) the emphasis was to have a description of the nitrification
and denitrification, and the model calibrations therefore focused on adjustment of the
parameters related to these processes. In contrast, other studies aimed at a description of
both COD and N removal, and as a result more parameters had to be considered for
adjustment in the model calibration (Siegrist and Tschui, 1992; de la Sota et al., 1994; Xu
and Hultman, 1996; Kristensen et al., 1998).
The wastewater characterisation has both been carried out via full-scale data combined
with mass balances and via lab-scale experiments, e.g. for the inert components Si and Xi
(Lesouef et al., 1992) and the Ss component (Xu and Hultman, 1996; Kristensen et al.,
1998). In one study all wastewater components were determined via calibration on the
full-scale data (de la Sota et al., 1994). The determination of the stoichiometric and
kinetic parameters is often carried out via calibration of the model to the full-scale data
only. However, some studies have also included the effort of characterising some
parameters in lab-scale experiments, e.g. for the determination of the specific growth rate
of the autotrophic biomass (e.g. Lesouef et al., 1992; Dupont and Sinkjaer, 1994) or to
collect further information on the half-saturation coefficients (Kristensen et al., 1998).
In addition, table 3 indicates that if the purpose of the model calibration was more than
“just” a description of the processes, more emphasis was put on the characterisation of the
relevant parameters via lab-scale experiments. For example in the study of Dupont and
Sinkjaer (1994) the aim was to apply the model for optimisation of nitrogen removal.
121
P.A.
Table 3. Information sources for model calibration of ASM 1 (conf on right hand page)
Reference Purpose Calibration Characterisation
strategy Wastewater
Full-scale Model components
data
Mass Lab- Model
balances scale calibration
ST92 Description: Steady state * 3,7,8,9 Si Xi
Nitrification, Dynamic
COD removal
L92 Optimisation: Steady state 3,4,8,9 Ss, Xs, XBH SI,

N-removal XI
PS92 Description: N- Steady state 3,4,5,7,8, Ss, Si, XI

removal Dynamic 9
DS94 Optimisation: Steady state 3,4,5,7,8, *

N-removal Dynamic 9
S93 Description: Steady state 1,3,5,6,8 *

Nitrification, Dynamic
COD removal
dS94 Optimisation: Steady state 3,5,7,8,9, all

All processes Dynamic 10
XH96 Description: Steady state 3,4,6,8,9 Si, SS Ss, Xi

COD removal, Dynamic XBH,
N removal Xs
K98 Description: Steady state 1,2,3,4,7, Ss

COD removal, Dynamic 8,9
N removal
*: procedure not described in detail, but probably carried out.
1. ss: Suspended Solids 6. TN: Total Nitrogen

2. vss: Volatile Suspended Solids 7. TKN: Kjeldahl Nitrogen
3. CODtot: total COD 8. NH4-N: Ammonium Nitrogen
4. CODsol: soluble COD 9 NOX-N: Nitrate + Nitrite Nitrogen
5. BOD5: Biological OxygenDemand (5days) 10. PO4-P : Ortho-phosphate
122
Table 3. Information sources for model calibration ofASMl (cont’ from left hand page)
Reference Purpose Calibration Characterisation
strategy Sludge
Lab-scale Kinetic and Lab-scale
analyses stoichiometric experiments
Model
calibration
ST92 Description: Steady state * YH, PH,

Nitrification, Dynamic KS, kh, Kx,
COD removal bn, PA, KNH
L92 Optimisation: Steady state 1,2,3,7 Kx,kh PA, bA

N-removal
PS92 Description: Steady state PA, KS, T|g

N-removal Dynamic
DS94 Optimisation: Steady state * Ks, Hg PA, bA, KNH,

N-removal Dynamic KOA
S93 Description: Steady state 1,2,3 PH, KS,

Nitrification, Dynamic PA
COD removal
dS94 Optimisation: Steady state 1,2,3,7 PA, bA, PH,

All processes Dynamic KS, kh, Kx,
KNO, fig, fih,
XH96 Description: Steady state 1 PA, KOH,
COD removal, Dynamic KNH, fig
N removal
K98 Description: Steady state 1,2 kh, Kx, bn, fig, bn, PH, PA,
COD removal, Dynamic KOH KOA
N removal
References
ST92 Siegrist
Finally, tableand4 Tshui
aims(1992) dS94 the
at summarising de lamost
Sota etrelevant
al. (1994)parameters to adjust in the steady
L92 Lesouef et al. (1992) S93 Stokes etal. (1993)
state and dynamic model calibration. The parameters related to the hydrolysis process are
K98 Kristensen etal. (1998) DS94 Dupont and Sinkjaer (1994)
not included in table
XH96 Xu and Hultman (1996)4. This was
PS92 Pedersen and Sinkjaer it
done on purpose since was not clear from the literature
(1992)
whether the parameters of this process are most influential to short- or longterm treatment
plant behaviour.
Table 4. Most relevant parameters in steady state and dynamic model calibration.
Steady state calibration Dynamic calibration
Predictions Long-term Short-term
123
P.A.
Main relevant parameters YH, fp, bn, Kj,influent PmaxHj gmaxA, T|g, T|h, K$, KNH,
KOH, KQA
4. Characterisation of wastewater and sludge kinetics
Different methods may be proposed to structure the wealth of methods that have been
developed and applied for the characterisation of wastewater and reaction kinetics in
relation to ASM1. At this point it is assumed that the reader is familiar with the ASM1
terminology. In this review it has been chosen to focus on the methodologies, i.e. what
can be achieved with different methods, their advantages and disadvantages, rather than
focus on the different wastewater components and processes separately. This choice was
motivated by the fact that some methods typically can yield information on more than one
component or process. In the end of the review it is attempted to illustrate the power of the
different methods for wastewater and sludge kinetics characterisation in the frame of
ASM1. Finally, the relevance of characterising the different components and processes in
the frame of ASM1 model calibration is critically evaluated.
4.1. WASTEWATER CHARACTERISATION
Waste water can be characterised either with physical-chemical methods or with

biological methods. In practice one typically ends up with a combined approach to obtain
an estimate of the concentrations of all components. In the following physicalchemical
and biological methods will first be described separately to obtain an overview of what
can be achieved with the different methods. Finally, an overview of what can be achieved
by combining both approaches is illustrated and discussed. In ASM1 the CODtot of the
wastewater is considered to consist of inert soluble organic matter (Sj), readily and slowly
biodegradable substrate (Ss and xs respectively) and inert suspended organic matter (Xi),
whereas biomass in the wastewater is considered to be insignificant:
CODtot = SỊ+s§+XỊ+xs (5)
4.1.1. Physical-chemical characterisation A wastewater can be separated into different

components in a relatively simple manner via physical-chemical separation methods. The
difference in molecular size can give an indication on biodegradability because small
molecules can be taken up directly over the cell membranes whereas bigger molecules
need to be broken down prior to uptake. Enzymatic hydrolysis is primarily a surface
phenomenon, which means that the hydrolysis rate is dfrectly related to the surface area.
Thus, smaller molecules are readily degraded whereas degradation of larger material can
be kinetically limited.
In early studies the wastewater components were separated physically into four size
depending fractions by successive sedimentation, centrifugation, and filtration. The
fractions were classified as settleable, supracolloidal, colloidal, and soluble (Rickert and
Hunter, 1971), and were analysed for chemical oxygen demand (COD). An important
conclusion from these studies was that the particles smaller than 1.0 Jim were
approximated to be the true soluble fraction. Moreover, the particles smaller than 1.0 Jim
124
were observed to be more rapidly degradable than particles larger than 1.0 pm. In a more
recent study Levine et al. (1985) studied the size distribution of the organic matter in
wastewater and the relationship to different waste water treatment processes. In this study
it was concluded that separation over a membrane with a pore size of 0.1 Jim was valid
for a differentiation between the true soluble and particulate organic fractions. The
organic particles smaller than 0.1 Jim are typically cell fragments, vfruses,
macromolecules and miscellaneous debris. The major groups of macromolecules in
wastewater are polysaccharides, proteins, lipids and nucleic acids. The fraction measured
by the standard test for suspended solids (1.2 pm) includes protozoa, algae, bacterial
flocks and single cells. However some bacterial cells, cell fragments, vfruses and
inorganic particles have a size from 0.1 to 1.2 pm and will thus also pass through the more
typically applied filter size of 0.45 pm for separation between soluble and particulate
matter (Levine et al., 1985). The size of colloidal matter is typically in the range 0.1-50
pm whereas material with a size larger than 50 pm usually settles (Levine etal., 1985).
The ASM models do not differentiate between filtered, colloidal and settleable
wastewater fractions. It is therefore necessary to convert the fractions resulting from a
physical-chemical characterisation to the ASM components. The possibilities and
limitations of physical-chemical methods to accomplish this task are summarised and
discussed below.
4.1.1.1. Inert soluble organic matter 5/Soluble inert organic matter SỊ is present in the
influent, but, importantly, is also produced during the activated sludge process (Chudoba,
1985; Orhon et al., 1989, Boero et al., 1991; Genmrli et al., 1991; Sollfrank et al., 1992).
Most of the evidence for the production of soluble organics by microorganisms is
collected from experiments with simple known substrates, e.g. glucose (Chudoba, 1985;
Boero et al., 1991). However, the production has also been proven to take place with
wastewater (Orhon et al., 1989; Germfrli et al., 1991; Sollfrank et al., 1992). The Sj
production seems to depend on the initial substrate concentration and on cultivation
conditions (Chudoba, 1985). A model has been proposed relating the SỊ formation to the
hydrolysis of non-viable cellular materials in the system, thereby linking the SỊ production
to the initial substrate concentration and the decay of the produced biomass (Orhon et al.,
1989). This model was verified in a Study with different industrial wastewaters and,
although the data were not of very high quality, some evidence was given that the Si
production depends very much on the waste water type (Germirli et al., 1991). The
hypothesis that the Si production originates from the decay process was, however,
contradicted in a study on municipal waste water (Sollfrank et al., 1992) where it was
concluded that the Si production was related to the hydrolysis of slowly biodegradable
COD of the incoming wastewater.
Thus, although the origin of the Si production may remain unexplained, it seems clear
that it does take place to various extents depending on different factors as mentioned
above, resulting in a Si concentration in the effluent that may be higher than the influent.
Such Si production is, however, not included in the ASM models, where Si is considered a
conservative component. To deal with this discrepancy between model concept and reality
a simplified approach is typically applied by the definition of a fictive model influent
concentration Si that includes the produced Si together with the real Si influent
concentration (Henze, 1992).
It is not possible to measure Si directly and different approximations are therefore
125
P.A.
usually applied. Most often Si is determined by the soluble effluent COD, which has
appeared to be a good estimate for Si in case of a low loaded activated sludge process
(Ekama et al., 1986). On the other hand Siegrist and Tschui (1992) suggested that the
influent Si could be estimated as 90% of the effluent COD. These approximations may
hold in most cases, but a more correct approach would be to consider it as the soluble
effluent COD minus the soluble effluent Biochemical Oxygen Demand (BOD) multiplied
with a BOD/COD conversion factor (Henze, 1992). Furthermore, Si can be determined as
the soluble COD remaining after a long-term BOD test with the influent (Henze et al.,
1987; Lesouef et al., 1992). The latter approach is in fact a combination of physical-
chemical and biological methods. However, in case of significant Si production during the
test the influent Si may be overestimated (Sollfrank et al., 1992), which may lead to an
underestimation of influent Ss eventually. Finally, a procedure was developed to
distinguish between Si of the influent wastewater and Si produced during degradation
(Germirli et al., 1991). However, in order to achieve significant response glucose was
added in these tests assuming that the wastewater under study resembled glucose, an
assumption that may not hold in practice.
Summarising, it will be case depending whether it is needed to characterise the
produced Si or whether the model component can be approximated as described above.
4.1.1.2. Readily biodegradable substrate Ss The soluble COD fraction excluding the
soluble inert organic matter (Si) is mostly considered to be the readily biodegradable
substrate Ss. The correctness of this approach does however evidently depend on the pore
size of the filters used for the separation. As described above the “true” soluble fraction
passes through a 0.1 pm filtration step according to Levine et al. (1985). However, in
practice larger filter sizes are most often used, which may result in an overestimation of
the soluble readily biodegradable substrate concentration, assuming that the definition of
Levine et al. (1985) holds.
Another study confirmed that the fraction passing a 0.1 pm filter gave a good
representation of the soluble readily biodegradable substrate (Torrijos et al., 1994). It was
confirmed biologically (via respirometry, see below for a detailed description) that the
studied wastewater did not contain any particulate readily biodegradable matter. In
contrast with this, Spanjers and Vanrolleghem (1995) found, also via respkometry, that
filtered wastewater (0.45 pm) had a lower biological response than unfiltered wastewater,
indicating that parts of the readily biodegradable wastewater fraction was retained on the
filter. Similarly, for an industrial wastewater it was found that the filtrate fraction
produced via ulfrafiltration (pore size < 0.001 pm) had a lower biodegradability (13% of
CODtot) than the fraction determined with a respừometric characterisation method (20%
of CODtot) (Bortone et al., 1994). Further it was also found that part of the soluble COD
can be slowly biodegradable (Sollfrank and Gujer, 1991).
Finally, a method based on flocculation with Zn(OH) 2 has been developed to remove
colloidal matter of 0.1-10 pm that normally passes through 0.45 pm filter membranes, and
was successfully applied to a phosphorus removal activated sludge system (Mamais et
al.,1993). However, the flocculation has appeared to be rather sensitive to interference and
appears highly depending on the pH value during the flocculation (Haider, 2000).
4.1.1.3. Inert suspended organic matter X/The test proposed for the determination of Si,
126
as the residual soluble COD remaining after a long-term BOD test, by Lesouef et al.
(1992) can also be applied to determine Xj. The Xi concentration is then determined as the
residual particulate COD, assuming that XỊ is not produced during the test. This
assumption may, however, be questionable since XỊ will be produced due to decay during
the long-term BOD test and corrections for this will have to be considered.
4.1.1.4. Slowly biodegradable substrate xs As mentioned earlier, a physical

characterisation based on different molecular sizes can be used to distinguish between
readily biodegradable substrate Ss and slowly biodegradable substrate x s. In one study it
has been proposed that xs may be determined as the colloidal fraction defined by 0.1 - 50
pm (Torrijos et al., 1994). However, this hypothesis could not be supported since the
results indicated that the colloids mainly disappeared according to a physical removal
mechanism without any related biological oxidation. In another study of contact
stabilisation, a multiple filfration procedure was used to isolate and monitor the variation
in concenttation of the colloidal fraction between 0.03 - 1.5 pm (Bunch and Griffin, 1987).
Here it was further confirmed that colloidal matter was removed physically, probably by
adsorption. However, the subsequent increase in soluble organic matter, and
corresponding oxygen uptake resulting from breakdown of colloidal substrate, were not
observed. Thus, based on these two studies it is not clear whether colloids can be
considered equal to xs. Part of the colloidal substtate may be inert, as was probably the
case in the example of Bunch and Griffin (1987), but this was not considered in these
studies.
In addition, parts of the soluble subsfrate (Sollfrank and Gujer, 1991) and the
settleable matters may belong to the xs fraction making it rather problematic to
characterise xs entirely by a physical-chemical method. Finally, if the components Ss, Si
and Xi are known and if it is assumed that the biomass concentration is negligible, x s can
be determined via a simple mass COD balance.
4.1.1.5. Biomass XBH and XBA It is not possible to distinguish biomass concenữations via
a physical-chemical method.
127
P.A.
4.1.1.6. Nitrogen components SNH, SND, SN0, XND The nitrogen components can rather
easily be detected by physical-chemical analysis via a combination of standard analyses of
ammonium, nitrite and nitrate and Kjeldahl nitrogen (TKN) on filtered and nonfiltered
samples (Henze et al., 1987).
4.1.2. Summary and discussion of physical-chemical wastewater characterisation

Based on the descriptions and discussions above it can be concluded that a wastewater
characterisation entirely based on physical-chemical characterisation alone will not be
sufficient to obtain an accurate distribution of the organic substrate over the different
ASM1 components (Fig. 6A). However, physical-chemical methods alone may be
adequate for the estimation of the nitrogen components (Fig. 6B). In figure 6 the dashed
line indicates the range of uncertainty with respect to the determination of the organic
components.
physical - chemical biological ASM1 ASM1 physical-chemical
inert
total settleable ' biomass
COD total N Particulate TKN
|xs
X,
slowly XBH
colloidal
biodegradable XB Soluble TKN
S
c
ND
A
inert
readily Ammonium
soluble biodegradable SNH

inert
J'Jitrite+Nitrate _2>N0
128
Fig.6. Characterisation of ASM 1 wastewater components by physical-chemical methods (A: modified

from STOWA, 1995; B: modified from Henze et al., 1995).
Summarising, the two main problems with respect to determination of the organic
components entirely by physical-chemical means are:
• The reliability of Ss determination based on soluble COD depends very much on the
applied filter size but, even more, on the kind of wastewater under study since it is
possible that part of the particulate substrate is also readily biodegradable.
• Defining Xs as being the colloids can induce eưors because the colloidal fraction may
also contain inert matter. Moreover, parts of the soluble and settleable fractions may
belong to Xs- Thus, it is not possible to separate the particulate Xs, Xi and XBH
components adequately.
Table 5 summarises the characterisation of wastewater components via physicalchemical

methods, and the assumptions needed, as described in the literature review above.
According to this table it can be seen that with some assumptions and a combination of a
physical-chemical and a biological method for assessment of XỊ (longterm BOD test)
(Lesouef et al., 1992), it is possible to determine all COD components (Ss, s b xs and Xi).
Knowledge of XỊ allows a determination of xs via a mass balance of particulate COD,
assuming that XBH is zero. However, it should be kept in mind that the determination of
XỊ via a long-term BOD test may not be accurate, as discussed above. Moreover, the
assumption that particular COD is not readily biodegradable may be incorrect.
4.1.3. Biological characterisation

The ASM models are in general biologically defined models. Thus, it is not surprising that
biological wastewater characterisation methods have found wider application and
acceptance than physical-chemical characterisation tests. In the biological methods the
fractionation of organic matter is based on its rate of degradation (Henze, 1992), which
makes the relation to the ASM concepts more direct. It is obvious that mainly the
biodegradable components and the microbial biomass in the wastewater (Ss, Xs, SNH,
SND, XND and XBH) can be characterised directly by these methods, whereas the inert
components Sj and Xi may be determined by a combination of physical-chemical and
biological tests, as already mentioned above (Lesouef et al., 1992). Typically, a biological
characterisation is based on measurements of the biomass response during substrate
degradation in either a continuous flow-through system or batch type experiment. This
means that the concentration determination of the biodegradable components is indfrect,
since the biomass activity has to be interpreted in terms of a substrate concentration. In
principle the consumption of substrate can be measured dfrectly by measurements of e.g.
COD. However, this is typically not very practical due to problems of sampling and
filtration of sludge samples etc. Instead, the biomass response can be monitored by
recording the utilisation of electron acceptors (such as oxygen or nitrate), or the
production of components during substrate degradation (such as protons, nitrate or carbon
dioxide).
A main part of the review on biological characterisation will deal with respfromefry.
Respirometry is defined as the measurement and interpretation of the oxygen uptake rate
129
P.A.
of activated sludge (Spanjers et al., 1998). In fact the main goal of a WWTP is to reduce
the biochemical oxygen demand of the waste water, and ASM1 was primarily developed
to yield a good description of the sludge production and consumption patterns of electron
acceptors, as described above. Thus, it is not surprisingly that respfrometry has turned into
one of the most popular biological characterisation methods, since the total respiration
rate is affected by the concentration of all aerobically biodegradable components, to
which the majority of wastewater components usually belong. However, nitrate utilisation
rates can also be applied for characterisation of the denitrification potential of a
wastewater. Finally, a titrimetric technique, especially applicable for determination of the
ammonium concenfration available for nitrification, will be reviewed.
130
Table 5. Overview of physical-chemical methods for determination of wastewater components (Fields with grey
background indicate that a physical-chemical method is not applicable)
Component Method Additional Assumptions Reference
informatio s
Si 0.45 pm filttation of effluent n. low loaded system (no E86
biodegradable substtate in
effluent)
90% of effluent COD ST92
7-8 pm filttation after long- no Si production during L92; S92
term aeration test degradation
COD profiles in batch tests wastewater similar to G91
glucose
Ss 0.1 pmfiftration Si Particulates do not L85; T94
7-8 pm fiftration SI contain readily L92
Zn(OH>2 flocculation SI biodegradable matters M93
Xi 7-8 pm fiftration after long- no Xi production during L92
term aeration test degradation
Xs mass balance Ss, Si, XI XBH, XBA negligible H87
XBH
XBA
Xp
So Standard analysis of oxygen
concenttation
SNO Standard analysis H87
SNH Standard analysis H87
SND Standard analysis of soluble SNH H87
TKN
XND Standard analysis of H87
particulate TKN
SALK Standard analysis of
alkalinity
E Ekamaer al., 1986 L Lesouef et al., 1992 S Siegrist and Tschui,
G Germirli et al., 1991 L Levine etal., 1985 S92 Sollfrank et al., 1992
9
H Henze et al., 1987 85 M Mamais et al., 1993 S96 STOWA, 1996
8 93 T94 Torrijos etal., 1994
7
Before the description and discussion on the application of respừometry, nitrate utilisation
rates and titrimetry for wastewater characterisation, the methodology of each method is
described in more detail. Thus, the readers already familiar with these methodologies can
skip these intermediate sections and dừectly continue reading about theừ applications for
wastewater characterisation.
4.13.1. Respirometry Historically, the determination of the Biochemical Oxygen Demand

(BOD) during an incubation period of 5 to 7 days (BOD 5 or BOD7) has been widely
applied to quantify the effects of pollutants on the oxygen demand of receiving waters,
and was further applied for the characterisation of wastewater. However, due to the rather
arbitrary choice of 5 or 7 days the test result represents a varying part of the ultimate BOD
of different wastewaters, depending on the wastewater composition. For a more complete
analysis of the ultimate oxygen demand of a wastewater the BOD test
131
P.A.
can be expanded to 20-30 days, typically 28 days.

In the BOD tests the oxygen content is typically
only recorded at the start and end of the test
without information on the evolution of the oxygen
consumption over time. This means that the test
can not give any information on the different
biodegradable fractions.
The test length of 5-7 days or even longer is not very suitable in the frame of
wastewater treatment plant operation. As a consequence the concept of short-term
biochemical oxygen demand (BODst) was infroduced (Vernimmen et al., 1967). The
concenttation of BODst can be determined via respkomefry. As defined above,
respirometry deals with the measurement and interpretation of the oxygen uptake rate, r 0,
of activated sludge. In general, the r0 may be considered to consist of two components
(Spanjers, 1993): The exogenous oxygen uptake rate (r0,ex)> which is the immediate
oxygen uptake needed to degrade a subsfrate, and the endogenous oxygen uptake rate
(r0,end)- Different definitions of r0,end appear in literature. The definition applied by
Spanjers (1993) is that the r0,end is the oxygen uptake rate in absence of readily
biodegradable subsfrate. In the context of ASM1 it is assumed that r 0,end is associated
with the oxidation of readily biodegradable matter produced by (1) hydrolysis of the
slowly biodegradable matter that results from lysis of decayed biomass and, (2) the use of
substrate for maintenance. The integral of the r 0,ex profile is a measure of BOD st (Spanjers
etal., 1998).
Contrary to the BOD5 method, the BODst test is carried out with the same biomass as
in the activated sludge plant under study and may therefore be a more representative
measure of the effect of the wastewater on the particular activated sludge plant under
study. Several attempts have been made to correlate BOD5 to BOD st (Vernimmen et al.,
1967; Farkas, 1981; Suschka and Ferreka, 1986; Vandebroek, 1986; Ciaccio, 1992;
Vanrolleghem and Spanjers, 1994). However, the success of such a correlation seems to
depend sfrongly on the type of wastewater, since the wastewater may contain varying
proportions of readily and slowly biodegradable fractions.
Figure 7 illusfrates the conceptual idea of respkomefry. The degradation of substrate
Si and s2 (Fig. 7A) results in a total exogenous uptake rate r 0,ex (Fig- 7B). Figure 7B
illusfrates a rather typical respkogram (i.e. a time course of respkation rates) with an
initial peak in r0,ex caused by oxidation of the most readily biodegradable matter, here Si,
followed by, in this case, one “shoulder” in the r 0,ex profile where component s2 continues
to be degraded. Thus, in this example the confribution of Si and s 2 to the total r0,ex> and
related total BODst, can easily be distinguished.
However, it will become clear from the “wheel-work” described in table 6 (Vanrolleghem
et al., 1999) that most of the processes in ASM1 eventually act on the oxygen mass
balance and may result in more complicated r0,ex profiles.
According to ASM1 the total r 0,ex of the activated sludge in contact with waste water is
given in Eq. 6.
-/1 V \ ^BH’AmaxH Ss ro,ex -U-YH)....Ụ7 77

YH KS+3S
(6)
132
+ (4.57 - YA ) ■ X
BA-AmaxA . NỊL
S
YA KNH+SNH
The concenttation of Ss and SNH depend on the influent wastewater and also on the rates
at which xs, SND and XND are degraded. As an example we will follow the arrows from
XBH to So (Table 6): in the mass balance of the heterotrophic biomass XBH (column, c.,
5) the production of XBH by aerobic growth (row, r., 1) is counteracted by the loss of XBH
by heterotrophic decay (r. 4). In this decay process component XBH (C. 5) is converted to
component xs (c. 4). This production of xs is counteracted by the loss of xs by hydrolysis
(r. 7), leading to production of component S s (c. 2). Ss is subsequently used for
heteroừophic growth (r. 1) where it is converted to component XBH (C. 5) with
concomitant consumption of oxygen So (c. 8), i.e. respfration. A similar reasoning can be
made for the processes involving the nitrogen components (SNH, SND and XND) and
autotrophic (nitrifying) organisms (XBA)-
Fig.7. Conceptual respirogram resulting from degradation of substrate Sỉ and S2
Figure 8 shows different examples of respfrograms collected’in batch experiments where

synthetic substrate (Fig. 8A) or different wastewaters (Fig. 8B-D) were added to
endogenous sludge. Note that in figure 8C-D only the exogenous oxygen consumption
due to substrate oxidation is given, r 0.ex, whereas the total r0 is given in figure 8A-B. It
now becomes clear that the respirograms can differ significantly in shape depending on
the substrate added and may not be as straightforward to interpret as the conceptual
example given in figure 7. Thus, the challenge is to interpret and perhaps divide the
133
P.A.
respirogram according to the contribution of ro,ex by different wastewater components.
134
There are two approaches for the determination of model parameters and components:
direct methods focus on specific parameters and components which can dừectly be
evaluated from the measured respiration rates (Ekama et aL, 1986; Spanjers et al., 1999),
whereas optimisation methods use a (more or less simplified) model that is fitted to the
measured data (Kappeler and Gujer, 1992; Larrea et al., 1992; Wanner et al., 1992;
Spanjers and Vanrolleghem, 1995; Brouwer et al., 1998; Coen et al., 1998). In the latter,
numerical techniques are used to estimate parameter values that lead to the smallest
deviation between model predicted and measured respiration rates (see figure 4).
Below, examples of respfrometric experiments to assess the different wastewater
components will be reviewed and important experimental factors with respect to
wastewater characterisation will be discussed. The overview does not attempt to review
and evaluate different respirometric principles, since a review of these is included in
Spanjers et al. (1998) and Petersen (2000). Different methods may only be included here
to illustrate points that are specifically related to wastewater characterisation.
Readily biodegradable substrate S5

The readily biodegradable substrate is presumably composed of simple and low molecular
soluble compounds, such as volatile fatty acids, alcohols, etc. (Henze, 1992). The
characteristic of these compounds is that they are degraded rapidly and hence result in a
fast respirometric response, e.g. figure 8A.
135
P.A.
40
~ 30
E
"QJ 20
I
10
0
0 1 2 3 4 5
0 5 10 15 20 25 30 35 6
Time (min) Time (h)
Fig.8. A: Typical acetate profile B: Municipal wastewater (after Kappeler and Gujer, 1992)
136
Table 6. Kinetic and stoichiometric relationships for COD removal, nitrification and denitrification
(Vanrolleghem etal., 1999) (conf on next page)
Component 1 2 3 4 5 6 7
i j Process Si Ss Xi Xs XBH XBA Xp
1_
1 Aerobic hetero- 1
trophic growth YH
2 Anoxic hetero- ___. ....
1_ 1
trophic growth
Y
H
a t . <57 :
Aerobic auto- 1
ttophic growth
4 Het. decay 1-fp^ B -1 f
p
5 Aut. Decay 1-fp -1 fp
6 Ammonification
7 TT,*■ 1 -1
Hydrolysis
8 < 1 .
/'■KT
Hydrolysis of
Observed conversion
rates ML'3T4
j j
Stoichiometric Nomenclature, see text All units in ML’3 (COD or N, depending on
parameters (see text) variable)
137
P.A.
Table 6. Kinetic and stoichiometric relationships for COD removal, nitrification and denitrification
(Vanrolleghem et al., 1999) (conf from previous page)
9 10 11 12 13 Process rate pj
ML’3'!”1
SNO SNH SND X
ND SALK
“ixB *XB 14 AH • X
BH
1-YH 1 ~ YỊỊ *XB

2.86 YH
-*XB
2.86 -14 YH 14
AH ' X
BH
1 . 77"^ 2 *XB
1 YA B A A • XBA
14-YA 14
ỵ
A1
ỈXB -fp ixp ’ XBH

*XB -fp i ^
t>A ■ XBA
X
P
1 * -1 ka ' $ND * XBH

14
k’h-Xs
___ -1 kh -XND
j j
Kinetic parameters
Nomenclature, see text All units in ML’3 (COD or N, depending on variable)
(see text)
138
Time (min)
Tlme(min)
Fig. 8. C: Municipal wastewater (Spanjers and Vanrolleghem, 1995), D: Industrial wastewater

(Coen etaL, 1998)
The most typical batch test for determination of Ss involves the addition of a wastewater
sample to endogenous sludge, and the monitoring of the respfration rate until it returns
back to the endogenous level (Ekama et al., 1986 among others). The examples shown in
figure 8 are all obtained with such an approach. The respirometric methods may vary
from a very simple lab-scale batch test to more complex methods that may even be
applied on-line. The concentration of readily biodegradable substrate initially present in
the mixture of biomass and wastewater in the experiment is generally calculated
according to Eq. 7.
SS(0)=—kexdt
(7)
1- ITJ ị <0 ,
The concentration of Ss in the wastewater is then easily calculated by taking the dilution
into account. The end point tfin of the integration interval is the time instant where Ss is
completely oxidised and where the exogenous respfration rate for Ss becomes zero. The
integral can directly and easily be obtained by determining the area under the r 0,ex
profile, e.g. by using a spreadsheet program. An alternative consists of solving the mass
balance equations with a numerical integrator to predict the exogenous respfration rates
for Ss and a given initial value Ss(0). It may be a bit overdone to apply numerical
integration for the profile illustrated in figure 8A, however for more complex profiles
(Fig. 8B-D), the approach may become necessary and more straightforward than direct
calculation, as will be discussed further below.
Notice that knowledge of the heterotrophic yield coefficient YH is needed for the
calculation of Ss from respfration rates (Eq. 7). The yield indicates the COD fraction that
is converted to cell mass. The rest of the COD is used to provide the energy that is
requfred to drive different synthesis reactions. This energy is made available by oxidative
phosphorylation, which requfres a terminal electron acceptor, in this case oxygen. The
produced energy is proportional to the mass of electron acceptor utilised, which in turn is
proportional to the COD consumed. As a consequence (1-YH)C0D is equal to the integral
under the r0,ex curve. Evidently, the parameter YH is always involved when oxygen
consumption is converted to substrate equivalents.
The batch test described above is also used to assess other ASM1 components and,
likewise, kinetic and stoichiometric parameters. This will be explained further in the next
section on characterisation of sludge kinetics, but this indicates already the popularity of
this test in assessing wastewater components and reaction kinetics.
Apart from the typical batch test as described above, other experimental designs have
also been tried out for the determination of S s. One example consists of monitoring the
respfration rate of unsettled sewage without inoculum for a relatively long period,
approximately 20 hours (Wentzel et al., 1995). A respfrogram similar to the one depicted
in figure 9 is obtained. The Ss concentration is calculated from the respfration rates
observed between the start of the test up to the time with the precipitous drop (due to
depletion of Ss), with correction for the increasing endogenous respfration due to the
increase of biomass during the test. In addition to Y H, knowledge of the maximum
specific growth rate is requừed, information that can be obtained from the same test (see
Figure 10).
An often-referred continuous flow-through method was developed by Ekama et al.
(1986), see figure 10. This method involves the monitoring of respfration rate in a
completely mixed reactor operated under a daily cyclic square-wave feeding pattern. The
experiment is designed in such a way that the supply of S s from hydrolysis of xs remains
constant for a period after the feed is stopped and gives rise to a second r 0 plateau. It is
hypothesised that the difference in r0 plateau values corresponds uniquely to the S s that
has entered via the influent. Hence, the concentration of readily biodegradable substrate
in the waste water can be calculated as given in Eq. 8.
c -v
OQ ——-—__ (8)
Q (1-YH)
An obvious disadvantage of this method is the length of the experiment (24 h, which is
not including the stabilisation of the continuous reactor used for the test), and the fact that
sufficient xs is needed in the feed to achieve a constant hydrolysis rate and to create as
such the step change in r 0. In addition, the method is rather difficult to carry out in
practice (Sollfrank and Gujer, 1991; Wentzel et al., 1995).
A final method for the evaluation of Ss was based on the evolution of the respừation
rates obtained in a continuously fed respfrometer during transients between two modes of
operation; a mode of endogenous respfration and wastewater addition respectively
(Spanjers et al., 1994). In the work of Lukasse et al. (1997) the estimation technique
developed for the determination of S s in the respfrometer of Spanjers et al. (1994) was
further evaluated and improved. In the work of Witteborg et al. (1996) the same
continuously fed respfrometer was used but a different estimation of S s was proposed as
now the measurement of respừation rate was performed under three different waste water
loading conditions. The waste water S s was calculated by numerically solving a set of
mass balances pertaining to different loading conditions of the respirometer.
Time (h) Time (h)
Fig. 10. Respiration rates obtained with

the experimental set-up of Ekama etal.
(1986)
Fig.9. Respiration rates measured in a
batch experiment for estimation of
FmaxH and Ks (after Kappeler and
Gujer, 1992.
Slowly biodegradable substrate Xs

It is assumed that slowly biodegradable substrate x s is composed of (high-molecular)
compounds ranging from soluble to colloidal and particulate (Henze, 1992). The common
feature of these compounds is that they cannot pass the cell membrane as such, but have
to undergo hydrolysis to low-molecular compounds (Ss), which are subsequently
assimilated and oxidised. The respfrometric response on x s is slower because the
hydrolysis rate is lower than the oxidation rate of Ss.
In a batch test an exponentially decreasing “tail” can frequently be observed in
respfrograms (Fig. 8B-C). In figure 8B, this tailing starts after approximately 0.75 hours.
The wastewater concentration of xs can be assessed in a similar way as above, Eq. 7
(Sollfrank and Gujer, 1991; Kappeler and Gujer, 1992). Simultaneously occurring
oxidation processes such as nitrification might interfere and complicate the separation of
the respfration rate due to hydrolysis in the total respfration rate. In that case a
nitrification inhibitor may be used to facilitate the assessment of x s (Spanjers and
Vanrolleghem, 1995). Alternatively, if the data of such respfrometric batch tests are used
in combination with mathematical curve fitting techniques to match the response of the
model to the data, the nitrification part can rather easily be extracted from the
respfrogram (Spanjers and Vanrolleghem, 1995).
It has also been proposed to estimate x s based on a long-term BOD test where x s is
obtained by subtracting Ss from BOD/(1-YH) (STOWA, 1996). Note that the value of YH
here should be lower than the one applied in Eq. 7, due to internal turnover of substrate
from decayed biomass in long-term tests.
Heterotrophic biomass XRH

In the ASM1 report the influent concentration of heterotrophic biomass, XBH, is assumed
to be negligible, as mentioned earlier. However, some wastewaters can contain a
signtficant concentration of heterotrophic biomass (Henze, 1992), and there may
therefore be a need to quantify this component. A batch test has been proposed where
XBH is assessed from the respfrometric response of raw wastewater without inoculum
(Kappeler and Gujer, 1992; Wentzel et al., 1995). The calculation requfres knowledge of
YH together with two parameters (PmaxH and b H) that can be obtained from the same
data. Respfrograms look similar to the one presented in figure 9. The procedure basically
backtracks the amount of heterottophic biomass originally present in the wastewater by
comparing the original respfration rate with the respfration rate after significant (hence,
well quantifiable) growth of XBH-
Autofrophic biomass XRA

So far, no procedures were found by which the autottophic biomass concenttation in
wastewater is determined. However, it could be imagined that a similar procedure as the
one developed for XBH is applicable. Thus, by evaluation of the respfration rate for
nittification, FQ ex , of the autottophs present in the waste water and by comparison to the
respfration rate of a culture with known autottophic biomass concenttation XBA> e.g.
after significant growth, the originally present XBA could be determined.
Ammonium SNH
The concenttation of ammonium in wastewater can be determined by using conventional
analytical techniques, as mentioned earlier. However, respfrometty also offers the
possibility to deduce SNH from batch measurements in a similar way as S s and xs,
provided the test is done with nitrifying activated sludge and the oxygen consumption for
nitrification can be separated from the other oxygen consuming processes. As follows
from table 6, the autottophic yield coefficient Y A is needed to convert the oxygen
consumption for nittification to a nittogen concenttation by division by (4.57-YA), where
4.57 indicates the amount of oxygen needed to oxidise one unit of ammonium nittogen.
The value of YA is typically 0.24 g COD(biomass)/g N, which means that the
determination of SNH is not very sensitivity to Y A since its value is small compared to
4.57.
Notice that part of the available ammonium may be assimilated into new
heterottophic biomass, which may be a considerable fraction of the nittogen in case a
large amount of COD is biodegraded (CODDegraded) simultaneously with the nitrification.
The actual nittified ammonium nitrogen, denoted N Nltr, can be approximated by Eq. 9 in
which ỈXB is the nittogen content of newly formed biomass:
NNitr = SNH -ĨXB YH CODDegraded (9)
From this equation one can easily deduce the original nittogen concenttation when
CODDegraded, and the stoichiometric parameters ỈXB and Y H are given. Note, however that
fitting a model in which carbon and nittogen oxidation are included to the respfromettic
data will automatically take this correction into account (Vanrolleghem and Verstraete,
1993; Spanjers and Vanrolleghem, 1995; Brouwer et al., 1998).
Organic nifrogen Sfrjp and slowly biodegradable organic nitrogen XND

Probably because the ammonification and hydrolysis rates of organic nittogen
compounds are relatively fast, little attention has been devoted so far to the establishment
of respfromettic techniques for SND and XND quantification. In batch tests these
compounds are typically converted to SNH before the SNH that was originally present in
the wastewater is removed by nitrification. Therefore, SND and XND are not directly
observable in such tests but may be lumped into the fraction of nitrified ammonium. Still,
for some industrial wastewaters the ammonification and hydrolysis steps may be
considerably slower and quantification of these component concentrations may be
required. In such cases, one can imagine a procedure in which the nitrification respiration
rate rQ ex is monitored and interpreted in terms of ammonification and hydrolysis, similar
to the way the respiration resulting from COD degradation is interpreted in terms of the
biodegradation of readily biodegradable substrate and the hydrolysis process.
Subsequently, the amounts of nitrogen containing substrates could be assessed by taking
the integral of TQ ex for the corresponding fractions and dividing these by (4.57-YA). In
case simultaneous COD-removal is taking place, correction should again be made for
nitrogen assimilated into new heterotrophic biomass (see above).
4.1.3.2. Nitrate utilisation rates
Readily or slowly biodegradable substrate Ss and Xs

The basis for wastewater characterisation via monitoring of nitrate utilisation rates
(rNO3) for the determination of the denitrification potential is rather similar to that of
respfrometry (Nichols et al., 1985; Ekama et al., 1986; Kristensen et al., 1992; Naidoo et
al., 1998; Spérandio, 1998; Urbain et al., 1998; Kujawa and Klapwijk, 1999). The
application of nitrate utilisation rates for wastewater characterisation within the frame of
ASM1 is however not as widespread as respirometry.
The readily biodegradable component Ss (or xs) is determined by Eq. 10 (similar to
Eq. 7). A typical rN03 profile is given in figure 11, indicating two biodegradable
wastewater fractions.
o 2.86
S
s(°) - ■ IrNO3,exdt (10)
1— JLR Ỉ <0 ,
The factor 2.86 g o2/g NO3-N originates from the fact that the theoretical electron
acceptor capacity of nitrate (as N) is 2.86 times that of oxygen (as O), assuming that
NO3-N is converted completely to nitrogen gas N 2 (Payne, 1981; van Haandel et al.,
1981). The factor has been verified experimentally by Copp and Dold (1998).
In Eq. 10 it is assumed that the Y H of aerobic and anoxic substrate degradation is
equal, as also assumed in ASM1. In a study on a pure denitrifying culture it has however
been reported since long that aerobic yields are larger than anoxic yields (Koike and
Hattori, 1975). It has been theoretically proven, based on the energetics of the metabolic
processes, that anoxic yields indeed are consistently lower than aerobic ones (Orhon et
al., 1996). Indeed similar differences between aerobic and anoxic yield were obtained
experimentally with activated sludge (McClintock et al., 1998; Spérandio et al., 1999).
Thus, to apply nitrate utilisation rates for waste water characterisation it is important to
correct for this difference in aerobic and anoxic yield since application of aerobic yield
values in Eq. 10 will lead to overestimation of the readily biodegradable wastewater
components.
4.13.3. Titrimetry The buffer capacity of water samples can be measured accurately by
advanced titration techniques (Van Vooren et al., 1995), and has recently been
successfully applied for the determination of ammonium and phosphorus in low
concentrations (0 - 100 mg/1) in effluents, surface waters and manure (Van Vooren,
2000).
Some efforts have been done to characterise VFA concentrations related to anaerobic
processes based on titration procedures and pH measurements (e.g. Munch and
Greenfield, 1998). These techniques may also be applicable for waste water
characterisation in the frame of ASM2 where one component is defined as the
concentration of fermentation products. This will however not be dealt with any further in
this presentation.
Fig.ll. Typical profile of rN03 as function of time for determination of Ss and Xs (Urbain et al., 1998)
Alternative to the classical titration methods (up and down titrations) Ramadori et al.
(1980) proposed to monitor the acid and/or base consumption rate that was needed to
keep the pH constant in an activated sludge sample where pH-affecting biological
reactions occur. This titrimetric method has been successfully applied for the monitoring
of nitrification, which has a clearly defined effect on the pH, and concentrations of StfH
(Massone et al., 1995; Gernaey et al., 1997). Recently, it has also been attempted to apply
the method for the determination of the total nitrifiable nitrogen concentration of a waste
water (Yuan et al., 1999).
Ammonium, SNH
A typical cumulative base addition curve and a pH profile collected during a titration
experiment with nitrifying sludge sampled on-line from a pilot plant are shown in figure
12 (Gemaey et al., 1998). In a first phase, the pH of the sludge sample is increased to the
pH setpoint, and base is added at a maximum rate. This phase took about 2 minutes for
the example of figure 12 For the experiments described here, a pH setpoint ± ApH
interval value of 8.2 ± 0.03 was used. Every time the pH of the sludge sample becomes
lower than 8.17 (= pH setpoint minus ApH interval), base is added to the sludge. Dosage
of base is repeated until the pH has returned within the pH setpoint ± ApH interval range.
Here, the nitrification phase is finished after about 25 minutes.
Fig. 12. Typical cumulative base addition curve (expressed as amount of base dosed per liter of
activated sludge sample) and pH profile obtained during an on-line titration experiment with a
mixed liquor sample. For this example, the nitrification phase is finished after about 25 minutes
(Gemaey et al., 1998).
The analysis of the data can either be via a simple manual interpretation or model-based
(Gernaey et al., 1998). The simple procedure is based on the detection of the two slopes
(SI and S2) in the cumulative base addition curve, followed by an extrapolation of the
different lines to the Y-axis (Fig. 13). The SNH concentration (mg N/l) and the
nitrification rate rN (mg N/l.min) can be calculated according to Eq. 11 and 12, where the
intercepts Bl and B2 are expressed in meq/1 units. The factor 0.143 meq/mg N (i.e., 2
mole H+ per mole N), is the stoichiometric coefficient relating the amount of acid (meq)
produced per mg of nitrogen nitrified. The slopes SI and S2 are expressed in meq/l.min
units.
(B2-B1)
OMU — ................. -...... . ..... (11)
0.143
_ (S1-S2)
ĨN (12)
0.143
Fig.13. Simple manual interpretation of a typical cumulative base addition curve (Gemaey et al., 1998).
In the application of Gernaey et al. (1998) the sludge was sampled at the last
compartment of an activated sludge pilot plant thereby reducing the likelihood of
presence of organic subsừates. In case ammonification is slower than nitrification it may
be relevant to determine SND, as described above in the section on respừomeừy. Thus, the
titrimetric method may also be applicable for SND determination. It may be foreseen,
however, that degradation of organic subsfrates may cause acid or base consumption
effects that may interfere with the determination of SNH according to the described
methodology.
Readily biodegradable subsfrate Ss.

The tiừimeừic methodology has also been applied for the determination of readily
biodegradable COD available for denitrification, and within confrol sừategies for
additional carbon dosage (Bogaert et al., 1997). A complicating factor is that depending
on the carbon source denitrification will either produce or consume acid (Bogaert et al.,
1997). Preliminary results (Dhaene, 1996; Rozzi et al., 1997) have indicated that the
method may be used to evaluate Ss in concenữated wastewaters.
4.1.4. Summary and discussion of biological wastewater characterisation

The capabilities of the different biological methods presented above to dừectly determine
the ASM1 waste water components are illusfrated in figure 14 (the dashed lines indicate
areas of uncertainties) and summarised in table 7. According to figure 14 it is obvious
that the readily biodegradable organic wastewater components, i.e. S s and parts of xs (Fig.
14A), and the niừogen components SNH and parts of SND and XND (Fig. 14B), can be
determined dừectly via the biological methods. The determination of the slower
biodegradable component Xs can be carried out indừectly via a long-term BOD test and
knowledge of Ss (STOWA, 1996). However, uncertainties may be inừoduced
by long-term BOD tests since significant
interference from product formation may occur
during the lengthy test.
Table 7. Overview of biological methods to estimate wastewater component concentrations. (Fields with grey
background indicate that a respirometric method is either not applicable or not relevant. Foran explanation of the
references, see table 8)
Component Metho Assumption References
d Type of Additional s
experiment Information
Si R BODoo, H87; L92
YH
Ss R WW YH E86
B, ww add. YH, M-mlb Ks We95
B, ww c c YH Wi96
(on/off) YH E86; SG91; We95
E86; K92; N98;
N B, ww add. YH
U98; KK99
T B, ww,s C/N C/N B97; R97; D96
constant
Xi
xs SG91;KG92; SV95
R B, ww YH
BODoo, ww S96
B, ww add. YH N98; U98; KK99
XBH R B, ww YH KG92; We95; B95
XBA R B, ww YA This paper
Xp
So
SNO T B, s C/N C/N B97
SNH constant
R B, ww YA, ÍXB, Y„, COD1*8 W93, SV95; Br98
T B, ww M95, G97; G98
SND R B, ww YA. ÍXB, YH.COD1*8 This paper
XND R B, ww YA, ÍXB, YH COD1*8 This paper
SALK
For the determination of SNH it should be remembered that it is in fact the nitrifiable
nitrogen that is determined via the biological methods (as indicated with dashed lines into
the regions of organic nitrogen, since parts of the organic nitrogen may be hydrolysed
making it readily available for nitrification). This is in contrast to the physical-chemical
method where the SNH component is determined via a chemical analysis of ammonia.
4.1.5. Discussion on phy steal-chemical vs. biological wastewater characterisation

By definition the total COD in ASM1 is sub-divided based on (1) solubility, (2)
biodegradability, (3) biodegradation rate and (4) viability (biomass), as described earlier.
Summarising, the COD components to consider in a wastewater are:
CODtot = SỊ + s§ +XỊ + x§ + (Xpj|) (13)
In previous sections it has been thoroughly reviewed how to determine these

components by either physical-chemical or biological methods, and different limitations
of the methodologies have been underlined and discussed. Furthermore, it is obvious that
the division of the wastewater into model components is to some extent artificial. For
example, a division is made between soluble and readily biodegradable substrate (S s) and
particulate slowly biodegradable matter (Xs), although it is, for example, known that
some slowly biodegradable substrate may be soluble etc.
Total COD ASM1 Respirometry Nitrate Utilisation Total N ASM1 Respirometry Titrimetry
Rate
X|
XBH;XBAT XND
SND
Xs
SNH
Ss
___?NO
Fig. 14. Characterisation of ASM! wastewater components by different biological methods (the
dashed lines indicate areas of uncertainties). A: COD components; B: Nitrogen components
It became clear that an application of physical-chemical methods alone is not sufficient

for characterisation of the wastewater into model COD components. These methods
basically only allow to distinguish between soluble and particulate COD and do not
differentiate with respect to biodegradability (non-biodegradable versus biodegradable
matters) and biodegradation rate (readily versus slowly biodegradable substrates).
However, by application of biological characterisation methods it is possible to obtain
knowledge of the biodegradability and biodegradation rate of the wastewater.
Thus, it is obvious that a combination of physical-chemical and biological
characterisation methods is advantageous for the translation of the wastewater
characteristics into the ASM1 model components. A suggestion for such a combined
approach, based on the literature review above, is presented in figure 15. Here it is
suggested to determine the readily biodegradable substrate (Ss) dừectly via respừation
tests (respừometry or nitrate utilisation rates). The presence of biomass in the wastewater
may also be determined by respửation tests. The slowly biodegradable matter (X s) can be
determined via the results of a long-term BOD test. The same kind of test may provide
information on the soluble and particulate inert (Sj and XỊ) matters. Here, however, the
reservation should be repeated that long-term BOD tests may not be
very accurate due to possible product formation (Si) and decay, which results in XỊ.
Therefore, the determination of Xi via a long-term BOD test may be questionable.
Indeed, it is proposed by Henze et al. (1987) to determine the influent XỊ via the
complete model during the calibration of the sludge balance. Subsequently, Xs may be
determined via a COD mass balance as the difference between total COD and the other
components. If it is chosen to determine Si by a long-term BOD test, it may be advisable
to combine it with analyses of the effluent, as proposed in the section about physical-
chemical methods. It is again clear from figure 15 that the borderline especially between
particulate and soluble COD, the differentiation between model components (Ss and Xs)
and the results from short-term respiration and long-term BOD tests may not be
completely consistent.
The nitrogen ASM1 components are somewhat easier to determine since they can
basically all be determined via mass balances based on standard chemical analyses of
total nitrogen, Kjeldahl nitrogen, ammonium nitrogen and nitrate nitrogen (see figure 6).
It can, however, be advantageous to combine these chemical analyses with biological
methods (respirometry or titrimetry) to obtain the nitrifiable nitrogen as a measure of SNH
(see figure 14) for studies where the focus is specifically on nitrification capacities.
In a study of STOWA (STOWA, 1996) a similar, but less extensive, study of
physical-chemical versus biological (only respfrometric) influent wastewater
characterisation was carried out. In this study guidelines for the COD components were
finally defined based on a more fraditional choice of physical-chemical methods
combined with long-term BOD measurements to allow for an easy implementation in
already existing routine analysis programs. It was concluded that respirometry is not yet
at a state where it can easily be applied for routine wastewater characterisation. The
STOWA guidelines for determination of the COD components are summarised in figure
16. Here the concentration of inert soluble matters (SỊ) is determined as 90% of the
effluent COD for low loaded systems, according to Siegrist and Tschui (1992). For high
loaded systems Si is also determined as 90% of the effluent COD but the effluent BOD
(multiplied by a COD/BOD factor) is subtracted. Ss is determined as the difference
between soluble COD and Si. Furthermore, the concentration of x s is based on a longterm
BOD test as the difference between BOD/(1-Y H) and Ss, as described above. The yield
coefficient in this long-term test is set to 0.20. Finally, Xi is defined as the difference
between particulate COD and the determined x s. Obviously, in this approach the division
of the waste water into ASM1 components is based on solubility and to some extent on
biodegradability according to physical-chemical methods supplemented by measurements
of the ultimate BODoo or BOD 5. The problem with this approach is that the
biodegradation rate of the wastewater is not really considered. This means that the
division of the biodegradable substrate into readily and slowly biodegradable substrates
may not be correct. It should be stressed though that the approach chosen by STOWA is
simple to implement into existing standard measuring routines at full-scale WWTP’s,
which is a factor not to be underestimated.
Combine
d
Fig.15. Suggested wastewater Fig. 16. The STOWA (1996)

characterisation by combined physical- guidelines for determination of COD
chemical and biological methods components
The STOWA guidelines for nitrogen components are also rather simple and based on
physical-chemical analyses. The SNH component is obtained based on standard analyses
of soluble ammonium nitrogen, and the determination of the organic nitrogen fractions
(SND and XND) is based on certain fixed fractions of N in organic components. It is
advised that these organic nitrogen fractions are checked regularly based on
measurements of total nitrogen, Kjeldahl nitrogen etc. according to figure 6B.
In this literature review the focus has been on characterisation of the ASM1
wastewater components. However, with the introduction of ASM3 (see table 2), that also
focuses on a description of oxygen consumption, sludge production and N removal, it is
interesting to discuss whether the approaches for wastewater characterisation applied for
ASM1 holds for ASM3 as well.
As described above, there is a shift of emphasis from hydrolysis to storage of organic
matter in ASM3. Furthermore, all S s is supposed to go through the storage process
(conversion to XSTO) before being used for growth. This means a change in how
wastewater characterisation should be viewed, since the separation between S s and xs
should now be based on the storage process rather than on the growth process. In ASM3
(Gujer et al., 1999) it is supposed that the soluble (Ss) and particulate (Xs) biodegradable
components can be differentiated with filtration over 0.45 pm membrane filters, whereas
a significant fraction of xs in ASM1 may be contained in the filtrate of the influent
wastewater. In ASM3 the latter is assumed to be caused by the conversion of soluble
biodegradable COD to storage polymers in the respiration tests. Whether this may hold in
any case seems yet rather unclear. In Gujer et al. (1999) it was recognised that the model
concept of converting all Ss into a storage component is not in accordance with reality.
Indeed, it was illusttated by Krishna and van Loosdrecht (1999) that the difference
between feast and famine phases could not be described accurately. This was caused by
the fact that ASM3 does not allow growth on the substrate S s alone. Therefore, a new
model sttucture was proposed where growth on external substrate is allowed in parallel
with the storage process. It remains however uncertain how to differentiate between the
amount of Ss that is dứected to storage and growth respectively. Furthermore, the yield
coefficient (which is needed to convert respirometric responses to COD components) in
ASM3 is composed of two factors: Ynet=YsTo ¥H, where YSTO is the storage yield and
YH the heterotrophic yield for the growth process. Also, here it does not seem clear how
to differentiate between the two yields. Basically, concerning the characterisation of COD
wastewater components, more experience will be needed before a wastewater
characterisation of the COD components related to the new storage concept of ASM3 can
be proposed.
The characterisation of the nitrogen components in ASM3 is however simplified by
the fact that organic nittogen components are included in the model as a fraction of the
corresponding COD components. Degradation of the corresponding COD component
results in immediate release of the organic nitrogen as ammonium. The latter was based
on the assumption that the ammonification is fast and the conversion of organic nifrogen
into ammonium therefore hardly affects the model predictions (Gujer et al., 1999). Thus,
the nifrogen balance includes on the one hand ammonium nitrogen (SNH) and nitrate
nifrogen (SN0), which both can be measured easily via standard chemical analyses, and on
the other hand organic nifrogen components. However, typically the fractions of organic
nitrogen in the COD components can be considered to be constant.
4.2. CHARACTERISATION OF SLUDGE COMPOSITION
In this section special attention is only paid to the assessment of the slower varying
sludge characteristics. Knowing the initial value of the concenfrations of soluble
components (e.g. ammonia) is not really essential because it has little impact on typical
simulation results with a calibrated model. Hence, the concenfrations of the following
particulate, slowly varying components must be assessed: XBII, XBA and XỊ (+Xp),
assuming that the system is in balance with no accumulation of x s. Only two
concenttations must be assessed since the sum of the concenfrations is equal to the
particulate COD (X) of the sludge that can easily be measured by using traditional COD
analysis (Eq. 14)
X = Xj +(Xp)+XBH +XBA (14)
Below some fast and direct methods for assessing sludge components are summarised.
Notice that the particulate nifrogen components are not considered here as thefr
concenfrations are assumed to be low.
Heterotrophic biomass XBH

One can show that the concenfration of heterotrophs in a continuous system in steady
state is equal to:
0X COPDegraded ỚH‘ l+bH
X
BH (15)
ỚX
where 0X is the sludge age, 0H is the hydraulic retention time, CODDegraded the total amount
of COD removed (taken over a sufficiently long period, e.g. one sludge age), b H the decay
rate coefficient and YH the yield coefficient. Respừometric methods to determine the
parameters bH and YH are discussed below, while a respfrometric evaluation of
CODDegraded can be performed with the respirometric measurements of biodegradable COD
fractions (Ss, Xs) that was already presented above.
As an alternative, Bjerre et al. (1995) used the method of Kappeler and Gujer (1992)
to determine the concentration of heterotrophs in the mixed liquor. Recently, this method
was thoroughly evaluated by Ubisi et al. (1997).
Autotrophic organisms XRA

In much the same way, the concentration of nitrifying organisms in the activated sludge
can be evaluated by means of a mass balance for the autotrophs (over a sufficiently long
time) (Dupont and Sinkjaer, 1994):
n f Aerobic XT Nitr
Y_ V.. ______________________
BA A (16)
" 'ỚH 1 + bA-ớx
where fAerobic is the aerobic fraction of the reactor; N Nitrthe amount of nitrified nitrogen ; b A
the autotrophic decay rate coefficient and YA the autotrophic yield coefficient. The
methods to determine the parameters bA and YA are discussed in the next paragraph,
while Nmtr can be quantified using the respfrometry-based nitrifiable nitrogen evaluation
methods that were given above.
Produced inert suspended organic matter Xp

To determine the produced inert matters, Xp, an evaluation of the mass balance of Xp in
steady state can be made. Assuming that the autotrophic biomass can be neglected, Eq. 17
is obtained:
X
P -fp bH XBH ớx (17)
The total concentration of inert matters, including the often significant

contribution of suspended inert material from the influent, is given in Eq. 18.
Xp --^Xị +fp bH XBH -ớx (18)
Respirometry can be involved in calculating this fraction via fp and bn (see

below).
4.3. CHARACTERISATION OF STOICHIOMETRIC AND KINETIC PARAMETERS

Similar to the overview of wastewater characterisation the overview on characterisation
of stoichiometric and kinetic parameters will be clarified according to the applied
methodology. The focus will, however, only be on different biological methods since
physical-chemical characterisation is not very relevant when it comes to characterisation
of reactions. As highlighted in the previous section the majority of the processes involves
oxygen consumption, which means that respirometry will again be the dominating
method in the review. However, also other methods such as nitrate utilisation rates,
titrimetry and ammonium uptake rate are powerful to assess some of the kinetic and
stoichiometric parameters.
4.3.1. Respirometry
4.3.1.1. Stoichiometric parameters By definition, determination of stoichiometric

parameters requfres the measurement of two factors that are related to the substrate
uptake. One of these factors may be the respiration rate. Theoretically, for ASM1 the
following stoichiometric parameters can be evaluated using respừometry: Y H, YA, ỈXB
and fp, though attempts are reported only for the first two.
Heterotrophic yield coefficient YH

This parameter not only influences the estimation of sludge production and oxygen
demand but also has an impact on the value of other parameters whose determination
requires a value for YH (see table 6) An example is the determination of S s from
respirometric data as described above (Eq. 7). Hence, an accurate value for Y H is of great
importance. YH can be determined using respfrometry by addition of an amount of
wastewater COD and measurements of the substrate oxidation r0,ex (Sollfrank and Gujer,
1991; Brands et al., 1994). Eq. 19 is then applied to evaluate YH .
t
COD degradable ” iro,ex (t)dt
COD degradable
The amount of degradable COD (CODdegradabie) is given by the COD concentration in the
filtered wastewater minus the inert fraction (Si). In the study of Sollfrank and Gujer
(1991) Si was determined as the soluble COD concentration in the effluent.
Brands et al. (1994) and Liebeskind et al. (1996) circumvent the problem of
determining Si by using a completely biodegradable substrate (acetate) instead of
wastewater. Hence, CODdegradabie is known exactly. This approach is, however, doubtful.
First, the choice of acetate is rather arbitrary and there is quite some evidence that the
yield coefficient for acetate differs from the influent wastewaters (Dfrcks et al., 1999).
Hence, acetate is not really representative for waste water COD. Moreover, due to the
experimental conditions in the batch reactor, it can be expected that part of the acetate is
Stored in the cell (Maj one et al., 1999). In
this case the observed oxygen demand only
represents the needs for transport of the
substrate and incorporation in storage material of
the cell, and not for the complete conversion into
new biomass. Conclusively, these procedures for
estimation of the heterotrophic yield do not seem
without problems.
Autotrophic yield coefficient YA

A value of 0.24 g biomass COD per g nitrified nittogen is generally assumed to be a good
theoretical value for YA. If requfred it is possible however to determine the actual YA
from a respừomeừic batch experiment in which a known pulse of ammonium (SNH(0)) is
added to a nitrifying activated sludge sample (Eq. 20).
4.57 SNH (0) - JfQex (t)dt

_____________Ố________
(20)
S
NH(0)
In this approach care has to be taken that no significant net growth of heterotrophs take
place as they would incorporate part of the added ammonium. In the model based data
interpretation applied by Spanjers and Vanrolleghem (1995) correction for incorporation
of SNH into biomass is taken into account directly via the model.
Nitrogen content of the biomass ixn

Obviously, the most likely method for evaluation of ixB would consist of a nitrogen
analysis of the biomass. However, one can imagine (albeit maybe not very realistically)
that nitrogen incorporation into biomass can be assessed using two respfrometric
experiments with nitrifying sludge in which different amounts of COD are degraded, the
difference being denoted as ACODDegraded. The reduction in the oxygen consumption for
nitrification A Jro,exơ)dt that can be observed for the higher COD loading then allows a
calculation of ixB (development of Eq. 9).
Degraded
; , YH ACOD
(21)
XB 4.57-YA r <t)
4 0.«
dt
Inert particulate fraction of the biomass fp
Decay of biomass results in a fraction being transformed into inert particulate products.
Typically 20 % of the biomass consists of inert material (Henze et al., 1987). This inert
biological fraction is called f’p. The model fp can be calculated starting from the
biological f’p with the following implicit equation:
fn =——fp, —
p
l-YH (l-fp) (22)
If the studied activated sludge has a yield coefficient (estimated for instance by using
respfrometry) deviating from the one reported in literature, the fp-value must be adapted
for this. Keesman et al. (1998) theoretically showed that the value of fp can be estimated
dfrectly from a batch test in which only the evolution of the respiration rate and sludge
concentration are monitored over sufficiently long time.
4.3.1.2. Kinetic parameters Basically the kinetic parameters that can be determined via
respfrometry are related to aerobic growth, decay and nitrification.
Heteroựophỉc decay coefficient bpr

The classical respfrometric method for determination of b H described by Henze et al.
(1987) is the protocol proposed by Marais and Ekama (1976) and is the most typical
method applied for the determination of the decay coefficient (e.g. Sollfrank and Gujer,
1991; Kappeler and Gujer, 1992). Sludge is inhibited for nitrification and is aerated in a
non-fed batch reactor. The (endogenous) respfration rate is measured at certain time
instants over a period of several days. Since the endogenous respiration is proportional
with the active biomass concenfration, a plot of the logarithm of the endogenous
respiration rate r0,end as function of time describes the exponential biomass decrease as a
straight line with slope b’H.
The death regeneration concept implies that the classical methods for determination
of the decay of biomass based on endogenous decay can not be applied directly. The
parameter based on the endogenous decay concept has to be translated to the death
regeneration concept, similarly to fp (Eq. 22), leading to the ASM1 decay coefficient b H
(Eq. 23).
(23)
Hence, the stoichiometric parameters YH and fp are necessary for calculation of bH.
Vanrolleghem et al. (1992) describe a fast method for estimation of b’ H using only
one measurement of the endogenous respfration (in absence of nitrification) in a batch
reactor. By means of Eq. 24 describing endogenous respfration, b H can be calculated on
condition that fp and XBH are known.
r
o,end = (1 “fp ) • bH ’ XBH (24)
The estimation of b’H can also be based on the fact that the respiration rate for substrate
oxidation is proportional to the heterotrophic biomass concentration (Spanjers and
Vanrolleghem, 1995). If a sufficiently high amount of oxygen So and substrate S s are
present, r0,ex is not substrate limited and will only be proportional to X BH. Consequently,
the decay of the heterotrophic biomass can be determined by (i) taking a sludge sample
from the aerated and non-fed batch reactor at certain time instants (tk), (ii) adding a
sufficient amount of substrate and (iii) measuring the maximum respiration rate.
Assuming that YH and M-maxH remain constant during incubation, plotting the logarithm
of r0,ex(tk) as function of time again allows to determine b’H as the slope of the curve
obtained via linear regression. In the study of Spanjers and Vanrolleghem (1995) a
model-based interpretation was applied to obtain accurate values of the maximum
respừation rates. However, only two data points were used for the semilog regression,
which does not make the estimated decay coefficients in this study very reliable.
■
Ệ
I
0 16 30 46 60 76
Tlme(mln) Tlme(mln)
Fig. 17. Respirograms obtained after injection of a C/N mixture for the simultaneous
determination of bn and bA according to the procedure of Spanjers and Vanrolleghem (1995).
Left: after 1 day incubation, Right: after 7 days
Ệ
i
I'
In the study of Avcioglu et al. (1998) a similar procedure was developed, where the
decay rate b’ir was assessed by monitoring the decrease in maximum respừation rate.
Avcioglu et al. (1998) included more data points compared to the study of Spanjers and
Vanrolleghem (1995). It was proposed that this method of determining the decay rate
should be more reliable, since interference of slowly biodegradable substrate, especially
in the initial phase of the traditional test of Marais and Ekama (1976), and inaccuracy of
low endogenous respừation rate measurements were avoided. The latter will, however,
evidently depend on the sensitivity of the applied respừometric method.
Furthermore, in the work of Avcioglu et al., (1998) it was experimentally verified that the
anoxic heterotrophic decay rate was reduced with about 40-50% compared to aerobic
conditions. Other studies confirm the observation that the heterotrophic decay is slower
under anoxic conditions (McClintock et al., 1988; Siegrist et al., 1999).
Autotrophic decay rate coefficient bA

The death regeneration concept is not applied for the autotrophic biomass in ASM1.
However, the approach of monitoring the decrease in r0,end as function of time can not be
applied for the determination of bA since that would requừe for instance an inhibition of
the heterotrophic biomass. Instead, the method based on the maximum substrate (here
SNH) degradation rate as function of time can be applied similar to the procedure for the
heterotrophic decay coefficient. In fact, in the procedure described by Spanjers and
Vanrolleghem (1995) the heterotrophic and autotrophic decay rate coefficients were
determined simultaneously by addition of a mixture of acetate and ammonium. Figure 17
shows the r0,ex data for the two respirometric tests performed after one and seven days of
sludge incubation, clearly illustrating the decreasing activity.
Nowak et al. (1994) pointed to the fact that the release of nitrogen due to decay of
heterotrophic biomass may result in some growth of nitrifying organisms. Hence, an
underestimation of bA would result. To correct this, they proposed the incubation of the
sludge under anoxic conditions to prevent growth of nitrifiers. Daily a sludge sample was
removed from the anoxic reactor and (after aeration) the maximum respừation rate was
determined. It was however observed that the reduction in maximum respừation rate was
significantly smaller (about 50%) under anoxic than aerobic conditions. This was further
confirmed by work on immobilized Nitrosomonas (Leenen et al., 1997) and by the
findings of Siegrist et al. (1999).
Maximum specific heterotrophic growth rate UmaxH and half-saturation concentration t

Ks
The maximum heterotrophic growth rate PmaxH can easily be determined from the
maximum r0,ex (Eq. 25) (Ekama et al., 1986), assuming that the substrate concentration is
in excess and the yield coefficient and heterotrophic biomass concentration (see previous
section) are known.
r
O,exYH
AmaxH ”■
(1- YH) XBH
However, the methodology proposed by Ekama et al. (1986) does not provide
information on Ks.
s (mg/l)
Fig. 18. A plot of substrate uptake rate versus substrate concentration for estimation of the parameters
for growth, example with valeric acid (Cech et al., 1984)
The increase of the substrate uptake rate with increasing Ss concentration is depicted in
figure 18. From such Monod type evolution the maximum specific growth rate Lt maxH
and the half-saturation constant Ks can be determined. In Cech et al. (1984) a
respứometric method is described in which a number
of measurements are performed, each of which add
one point to figure 18. In this procedure
experiments are carried out with addition of
different amounts of wastewater (substrate) to
endogenous sludge, allowing to achieve various
substrate uptake rates, i.e. exogenous respiration
rates (rOex), up to a maximum rate.
The parameters pmaxH and Ks can, for instance, be found by Lineweaver-Burk
linearization of Eq. 26 that describes the curve in figure 18 (Cech et al., 1984), although
the statistical quality of this procedure is not optimal (Robinson, 1985).
dss _ ^maxH XBH Ss

(26)
dt YH KS+SS
The method of Cech et al. (1984), which was also applied by e.g. Volskay and Grady
(1990), is rather time consuming and the experimental effort is high. As an alternative a
more efficient approach was presented, using a continuously aerated respirometer to
which a single substrate pulse is added (Vanrolleghem et al., 1990; Kong et al., 1994). In
this method r0,ex is recorded frequently as the experiment progresses and one experiment
is sufficient for the determination of both UmaxH and K s provided that the concentration
of added substtate is sufficiently high. In this approach a model (Eq. 27 - 28) is fitted to
the r0,ex profile for the determination of UrnaxH and K s. An example of an acetate
addition is illusttated in figure 19 (obtained from Kong et al., 1994) where r0,ex is
illustrated together with the corresponding cumulative oxygen consumption and substrate
concenttations as function of time.
dSs _ _ AmaxH XBH . Ss dt YH KS+SS

(27)
„ _/i V x.^maxH-XBH Ss
r
o,ex = (f - YH )- - - - -V-------
ỵ Ks +Ss (28)
H
The heterotrophic kinetic parameters can also be determined based on the cumulative
oxygen uptake profiles rather than oxygen uptake rate data. In the methodology described
by Ellis et al. (1996) and Smets et al. (1996) the kinetics are determined for specific
organic chemicals. However, the procedure is dừectly applicable for waste waters as well.
0.80 3.00
£
0.50 2.50 Ệ
0.40 2.00 g
0.30 1.50 Í2.
0.20 1.00
0.10 0.50 -Ị
0.00 J
0.00 £
-5 0 5 10 15 20
Time (mln)
Fig. 19. ro,ex (symbols), cumulative oxygen uptake (increasing line) and substrate concentration
(decreasing line) in batch experiment (Kong et al., 1994).
A batch experiment with high initial substrate (wastewater) to sludge ratio (called the
S(0)/X(0) ratio) was proposed by Kappeler and Gujer (1992). This procedure also enables
estimation of PmaxH and Ks from a single experiment. An alternative to the method of
Kappeler and Gujer (1992) is to plot the oxygen uptake rate versus the cumulative
oxygen consumption (Smets et al., 1996). Figure 9 shows a respirogram obtained with
such an experiment (Kappeler and Gujer, 1992). Contrary to the procedures of e.g.
Vanrolleghem and Verstraete (1993) biomass growth is significant and PmaxH can be
assessed dừectly without knowledge of Y H. A plot of the logarithm of the r 0
measurements versus time has the slope (pmaxH - bn). If b H is known, a calculation of P-
maxH is possible (in the work presented by Kappeler and Gujer (1992), it is assumed that
the decay rate is 5% of the growth rate). Attention has to be paid to the fact that the high
S(0)/X(0) ratio in this experimental set-up (about 4/1) gives rise to significant growth of
the biomass during the experiment. This means that the observed kinetic characteristics
may no longer be representative for the original sludge, due to the risk that the
experimental conditions may have favoured fast growing organisms that become
dominant during the experiment. Novák et al. (1994) gave practical evidence for this
hypothesis by evaluating results from experiments with different S(0)/X(0) ratios. A 2.5
times higher spectfic growth rate was obtained at high S(0)/X(0) ratio, compared to an
experiment with a low S(0)/X(0) ratio.
In the work of Grady et al. (1996) the terminology of inttinsic and extant kinetics was
introduced. Intrinsic kinetics refer to the ultimate capacity of the biomass, whereas extant
kinetics refer to the biomass activity prior to the lab-scale experiments, e.g. in the full-
scale plant. This will be discussed further in later sections.
Maximum specific autotrophic growth rate UmaxA and half-saturation concentration

KNH In the studies by Drtil et al. (1993) and Nowak et al. (1994) the above mentioned
methodology of Cech et al. (1984) was applied to evaluate the maximum specific
autotrophic growth rate and half-saturation
concentration KNH- TO assess the r0 for autotrophic
activity only, the heterotrophic endogenous
respfration was determined by a separate
experiment, where ATU was added, and was
subtracted from the total r0 obtained from an
ammonium addition. Here too knowledge of Y A and XBA
is needed for the calculation of pmaxA. In the work
by Nowak et al. (1994) the concentration of XBA was
determined based on full-scale data.
Alternatively, ỊimaxA and KNH can be obtained dfrectly from experimental data of a
simple ammonium addition as presented in figure 20. In a study by Spanjers and
Vanrolleghem (1995) a model-based interpretation was applied for the determination of
the nitrification kinetic parameters (Eq. 29), similar to the approach described above for
the kinetic parameters of heterotrophic growth.
r= (A. S7 Y v^maxA’XBA ro,ex - t4-^ ' - Y S
NH
A )---------------------------------- V--------------
Y
A K
NH +SNH
(29)
Hydrolysis constants k^, Kx

As far as known the only experimental protocol that enables a determination of both
parameters of the hydrolysis process is the “cyclic square wave feed” experiment
proposed by Ekama et al. (1986). This method has already been described earlier for the
determination of Ss with a typically obtained profile shown in figure 10. To determine the
hydrolysis parameters the data obtained after the drop in respfration rate are important. If
r0 remains constant on a plateau value (as is noticed in figure 10 between t = 12 and t = 15
h), this is related to the hydrolysis that proceeds at maximum rate and the biomass that is
saturated with hydrolysable products (Xs /XBH » Kx). As such, these data contain the
information to assess the value of kh on condition that the heterotrophic biomass
concentration XBH and the yield coefficient YH are known. With decreasing xs also the
rate of hydrolysis decreases and the respfration rate is depending on the value for K x,
allowing its estimation. Estimation of the parameters is best by means of model
optimisation (Henze et al., 1987).
In many cases the dependency of the rate of hydrolysis on the heterotrophic biomass
concentration may be neglected and first order hydrolysis process dynamics are then
obtained (Sollfrank and Gujer, 1991). This assumes that XS/XBH « Kx. Sollfrank and
Gujer (1991) proposed a method to determine the first order hydrolysis constant, i.e.
kh/Kx, using respfration rates measured by dosage of wastewater to a continuous flow
pilot reactor. To simplify the estimation, they suggested to present the respfration rate as
function of the residual amount of substrate. In this plot one is able to isolate a linear part
from which the hydrolysis constant kj/Kx is deduced (provided YH is known).
For estimation of the first order hydrolysis constant k h/Kx Kappeler and Gujer (1992)
performed a batch experiment with an initial COD based S(0)/X(0) biomass ratio which
was 10 times higher than thefr experiment for determination of the maximum specific
growth rate (S(0)/X(0) = 1/2). Figure 8B shows the respfration rate data of such an
experiment, from which the slowly biodegradable substrate, x s, can also be determined, as
described above. Once the readily biodegradable substrate Ss is removed (in Figure 8B
after 0.75 h) the further decrease of the respiration rate is determined by hydrolysis of x s.
As a consequence, the r0 measurements enable to estimate the hydrolysis rate constant.
The authors advise to do this exercise at different biomass concenfrations to check for a
possible dependency of the hydrolysis rate to the biomass concentration.
Tlme(mln)
Fig.2O. Respirogram obtained after injection of 3.31 mg NH4-N in 1.4 I activated sludge (Spanjers and
Vanrolleghem, 1995)
Parameters of “switching functions” KOH, KQA

Kappeler and Gujer (1992) determined the respfration rate as function of different oxygen
concentrations in the respfration chamber of theừ respfrometer. According to these
authors the concenfration of readily biodegradable substrate Ss needs to exceed a
minimal concentration in order to have an accurate determination of KOH. The same
technique can be used for KOA with ammonia as substrate.
Ammonification rate constant ka

So far, no respirometric method has been reported for the determination of the
ammonification rate. However, it is theoretically possible (see table 6) to assess this
parameter from the evolution of the oxygen consumption for nitrification resulting from
ammonified nitrogen, provided ammonification is the rate limiting step.
Simultaneous determination of heterotrophic and autofrophic kinetic parameters.

In the previous sections on determination of heterotrophic and autotrophic growth
kinetics the focus was put on how to determine the kinetic parameters for the
heterotrophic and autotrophic processes separately. However, except for the examples of
Sollfrank and Gujer (1991) and Kappeler and Gujer (1992) the presented examples
mainly dealt with additions of known substrates (acetate as carbon source and
ammonium). The fact is that when dealing with real wastewater and activated sludge both
heterotrophic and autotrophic processes will take place simultaneously, and a detailed
data interpretation of the respfrograms and good experimental design will be needed to
“separate” and as such determine the kinetic parameters for the different processes.
Vanrolleghem and Versttaete (1993) proposed an experimental design that enables to

simultaneously measure both heterotrophic and autotrophic maximum respfration rates.
In their approach a mixture of ammonium and acetate was added to endogenous sludge.
The maximum respừation rate for carbon oxidation and nitrification can be derived from
the respừograms on the condition that the two aerobic processes can be clearly
distinguished from each other. The problem with this approach is however that the kinetic
parameters are highly dependent on the nature of the substrate. Thus, the use of a single
compound like acetate to represent a complex substrate like wastewater is difficult to
justify scientifically.
In the study on wastewater by Spanjers and Vanrolleghem (1995) experiments with
municipal wastewater were presented with much lower substrate to biomass ratios
(S(0)/X(0)) compared to Kappeler and Gujer (1992). Figure 21 shows a typical
respfrogram from an experiment with a S(0)/X(0) of 1/200. This respfrogram is much
more complicated to interpret than the ones shown so far. Ffrst, simultaneous carbon
oxidation and nitrification take place. The only seven minutes lasting initial peak in r 0,ex
is assumed to be due to the oxidation of Ss. After some time only nittification and,
assumingly, oxidation of substtates released by hydrolysis occurs. In this work the
respfrograms of the waste water were interpreted with a more complex ASM1 based
model including degradation of two readily biodegradable substtates Ssi and S S2, first
order hydrolysis and nitrification. Thus, kinetic parameters for all these processes were
obtained simultaneously. Experiments in the presence of a nitrification inhibitor ATU
were performed to check the conttibution of nitrification to the respfration rate. This is
shown in the insert of figure 21, where the r 0,ex related to the degradation of Ss and xs can
be observed.
An approach cừcumventing ATU addition, suggested by Spanjers and Vanrolleghem
(1995), consisted of the following two-step procedure. Ffrst, the nittification process is
characterised separately via an experiment where only ammonium was added, as
described above and illustrated in figure 20. In a second step, the full model is applied to
fit to the data of figure 21. However, during this step the nitrification parameters are kept
at theừ values obtained from the separate nitrification experiment, and are thereby used to
“eliminate” the nitrification oxygen consumption in an experiment with addition of
wastewater. The amount of nitrogen in the wastewater sample can be estimated
simultaneously as it determines the length of the nitrification shoulder. Spanjers and
Vanrolleghem (1995) demonstrated that the ATU and model-based elimination of the
nitrification respừation rate lead to similar values for the kinetic parameters and waste
water characteristics.
Another example of a detailed interpretation of a respừomettic test with municipal
wastewater addition is given by Brouwer et al. (1998). Here a model including
degradation of two readily biodegradable substrates, hydrolysis and two step nitrification
is applied to interpret wastewater respừograms. The problem encountered in this study
was, however, that not all processes were clearly identifiable from the respfrograms. It
was thus suggested that the number of unknown model parameters should be reduced for
this example by including experiments with separate additions of synthetic substrates, for
example ammonium and nitrite. In this way it would be possible to fix these kinetic
parameters in the characterisation of the complex wastewater, similar to the approach of
Spanjers and Vanrolleghem (1995).
Tlme(mln)
Fig.21. Respiration rate after injection of 70 ml raw waste water to 1.5 I activated sludge. Insertion:
similar experiment but after addition ofATU (Spanjers and Vanrolleghem, 1995)
Finally, an application with industrial wastewater (no nitrification) was presented by

Coen et al. (1998), where a model-based interpretation approach was applied for the
determination of kinetic parameters and substrate concentrations of simultaneous
degradation of three COD wastewater fractions.
4.3.2. Nitrate utilisation rates

Characterisation of reaction kinetics via analysis of the nitrate utilisation rate is basically
very similar to the methodology based on oxygen respừation rates, and different studies
have dealt with the comparison of r ,ex and r 03,ex (e.g. McClintock et al., 1988; Kristensen
0 N
et al., 1992; Orhon et al., 1996; Sozen et al., 1998). In ASM1, the same kinetic
expressions are applied for nitrate utilisation processes as for oxygen, with the only
difference that a correction factor T| is incorporated in the equations for anoxic processes.
This factor allows to describe that only a fraction of the total biomass is capable of
respừing with nitrate and/or that the anoxic rate is lower than the aerobic one. Typically,
one applies the relationship given in Eq. 30 in order to relate r0,ex with ^NO3,ex’
77 = 2.86-rNO3,ex (30)
r
o,ex
Correction factors for anoxic growth and hydrolysis n
It has been shown that the value of T| can vary significantly for different activated sludge
systems. In different studies values have been recorded in the range 0 - 0.95 (Van
Haandel et al., 1981; Henze, 1986; Henze et al., 1987; 1995; McClintock et al., 1988;
Kristensen et al., 1992; Sozen et al., 1998; Spérandio et al., 1999). Some theories were
developed based on general mass balances that allowed for an estimation of T| from
wastewater characteristics, treatment plant layout and operation (Henze, 1986). It was
shown that the dominating factor for T| is the potential inlet fraction of denitrifiers, which
includes the denitrifying fraction of the influent biomass plus the primary produced
anoxic biomass. Based on some practical constraints concerning e.g. minimum anoxic
sludge age and minimum aerobic sludge age to keep both nitrification and denitrification
in the system, it was estimated that in practice T| might be in the order of 0.4 - 0.9
(Henze, 1986).
An underlying assumption behind Eq. 30 is that the aerobic and anoxic yields are
equal. As discussed above significant evidence exists that the anoxic yield may be lower
than the aerobic one. In the studies by Orhon et al. (1996) and Sozen et al. (1998) very
high values (>1) for the conversion factor T| were related to possible lower anoxic yields
for which correction will be needed. The occurrence of lower anoxic biomass yields was
already discussed in the section about application of nitrate utilisation rates for the
determination of readily or slowly biodegradable substrate Ss and xs.
4.3.3. Titrimetry
Maximum specific autotrophic growth rate Umax A and hạự-saturation concentration
KNH So far the titrimetric technique, based on pH control and monitoring of the
cumulative amount of base or acid added to keep the pH set-point, proposed by Ramadori
et al. (1980), and introduced in more detail above, has only been applied to the
determination of the nitrification kinetic parameters jUmaxA and K NH- AS illustrated in
figures 12 and 13 and by Eq. 12, the cumulative amount of base added can be used to
calculate the nitrification rate and thereby provide kinetic information. In the work of
Gernaey et al. (1998) a model-based data interpretation was applied for the estimation of
gmaxA and KNH. The model is similar to the one applied for the description of
respirometric and nitrate utilisation rate data. The only difference is the stoichiometric
coefficient relating the ammonium degradation to proton production Hp (Eq. 31).
_ 2 + YA ixB AmaxA XBA SNH

r (31)
Hp - 4 TT
14 IA KNH+SNH
4.3.4. Summary and discussion of biological characterisation of stoichiometric and

kinetic parameters
The above review on biological characterisation has illustrated that, theoretically, nearly
all parameters can be determined with biological methods. Especially respừometry stands
as a powerful characterisation method but other methods too are useful for the
characterisation of specific processes, e.g. titrimetry for the characterisation of
nitrification and application of nitrate utilisation rates for the determination of the
correction factor for denitrification.
One of the challenges in the application of the biological methods is how to interpret
and relate the experimental data to the different processes that may take place
simultaneously. It is obvious that experiments with addition of known and simple
substrate such as ammonium or acetate are easier to interpret in terms of determination of
stoichiometric and kinetic parameters than experiments with real wastewater. For
example, it is difficult to assess the heterotrophic yield Y H by experiments with real
wastewater, and in some cases it was therefore suggested to determine it from an
experiment with known substrate in the form of acetate (Brands et al., 1994; Liebeskind
et al., 1996). It has also been suggested to determine the maximum specific growth rate
PmaxH based on experiments with acetate in respừometric experiments (e.g.
Vanrolleghem and Verstraete, 1993). However, acetate does not represent the actual
wastewater very well. As already stressed above it is generally questionable to use a
single substrate to represent complex waste waters. Furthermore, it is a known
phenomenon that acetate easily gets dừected towards the storage process instead of
dừectly being consumed for growth (Majone et al., 1999). This means that if such data
are only interpreted in terms of the growth process, the estimated parameters related to
growth will be erroneous. E.g. the stoichiometric growth yield (YH) will be overestimated
(Dfrcks et al., 1999). On the other hand, characterisation of the stoichiometric and kinetic
parameters for nitrification can be done by respirometric or ti tri metric experiments with
single additions of pure ammonium.
It is of course advantageous if several parameters (kinetic or stoichiomefric) and
some wastewater components can be obtained from the same experiment. This was
illustrated in studies with municipal wastewater by e.g. Spanjers and Vanrolleghem
(1995) and Brouwer et al. (1998), and also for an indusfrial COD removal case (Coen et
al., 1998).
In table 8 (adopted and modified from Vanrolleghem et al., 1999) the experiments
described above for characterisation of stoichiometry and kinetics are concisely
represented. Attention is drawn to
• The method (respừometry, nitrate utilisation rates, titrimetry)

• The type of reactor set-up (continuous or batch experiment) and the additions
performed;
• The requừement for other information collected from other experiments (or
assumed);
• Major assumptions made during the interpretation of the data;
From table 8, it can for example be seen that in the work of Spanjers and Vanrolleghem
(1995) with waste water (reference SV95 and experiment type “B, ww add.”) the
parameters Pmaxib Ks, PmaxA, KNH and kh and the substrate components Ss, xs and SNH
could be retrieved from a single experiment.
It will now be attempted to evaluate whether the characterisation approaches of the
kinetic and stoichiometric parameters as reviewed for ASM1 can hold for ASM3 too.
As reviewed above, it should be theoretically possible to assess the ammonification
rate from respừomeưic data, provided that ammonification is the rate limiting step.
However, in most applications this is not the case making it difficult to quantify the
kinetics of ammonification. Furthermore, ammonification does not affect the model
predictions significantly, since it is usually a fast process. Thus, with this in mind the
ammonification process was not included in ASM3, thereby also eliminating the need to
determine its kinetic rate.
Another simplification in ASM3 is the way the decay process is described. Instead of
the more complex death regeneration concept it was chosen to describe decay with a
more traditional and simple endogenous decay process. This means that the results from a
simple long-term aeration test (Marais and Ekama, 1976), where the endogenous
respfration rate is monitored over a period of several days, can be applied more dfrectly.
In this way a transformation of the data from the endogenous test to the death
regeneration concept is no longer needed. Furthermore, the exclusion of the death
regeneration concept also resulted in a simplification of the hydrolysis process, since this
process is now only involved in hydrolysis of slowly biodegradable substrate (Xs)
contained in the influent.
However, with the introduction of the storage model concept it becomes difficult to
separate between the kinetics of storage and growth. Already in the discussion of
wastewater characterisation it was pointed out that the yield obtained from a
respfrometric test is composed of two factors Y net=YsTO*YH- Furthermore, it does not
seem clear how to differentiate between the storage rate and growth rate from e.g. a
respfromefric test
4.4. IS CHARACTERISATION VIA LAB-SCALE EXPERIMENTS RELEVANT?
In previous sections the sources of information that can be used for calibration of ASM1
were reviewed and attention was especially focused on how to characterise the different
waste water components, stoichiometric and kinetic parameters. Different problems were
already highlighted.
The focus is now turned back to calibration of ASM1 and the aim of describing a full-
scale WWTP. It should be remembered that the purpose of the model calibration
determines the degree of detail of the information that is needed, e.g. which wastewater
components and parameters need a more accurate determination than others. Even though
it may be possible to characterise some components or parameters, it may not always be
relevant for the actual purpose.
Meth Type of Additional Assumptions References
od experiment information
Stoichiomet parameters
ric
YH R B, ww add. Si SG91
B, Ac add. Ac representative of Ss Br94
YA R B, NH4 add. ixB=0 SV95
B, NH4+AC ÌXB, YH SV95
fp R add.
B MLVSS K98
ixB R B, COD add. YH, YA, this paper
ixp
Kinetic parameters
Pkn R B, n S,Ac adds. YH, XBH C84; VG90
ax B, s, Ac add. YH, XBH bH jirnaxH represent
K94; E96;
H B, ww add. YH, XBH original XBH
Sm96
B, ww add. KG92
SV95, B98
Ks R B, n S,Ac adds. C84; VG90
B, s, Ac add. YH, XBH K94; E96;
B, ww add. YH, XBH bH Ks represent original Sm96
B, ww add. YH, XBH XBH KG92
KO R B, n So So Ss sufficiently high SV95,
KG92 B98
KN
bH R B, no add. fp, YH ME76; SG91
B, n Ac add. YH , PmaxH constant SV95
B, no add. fp, YH V92
Pm R B, n NH4 adds. YA, XBA C84
ax B, NH4 add. YA, XBA D93; SV95
A T B, NH4 add. YA, XBA G98
A B, NH4 add. YA, XBA K92
KN R B, n NH4 adds. YA N94
H B, NH4 add. YA D93; SV95
B, ww add. YA SV95, B98
KO R B, n So SNH SNH sufficiently high KG92
bA R B, NH4 add.
B, n NH4+AC H87; N94 SV95
R+N add.
B, ww add. K92; S99
ka R B, s add. This paper
kh R c (on/off) c, ww YH, XBH YH E86
add. max. hydrolysis rate Kx SG91
B, ww add. very large Kx very large KG92; SV95;
Kx R c (on/off) YH, XBH B98
E86
T|h R+N B, ww add K92
Table 8. Overview of biological methods for estimation of ASM1 parameters. (Fields with grey
background indicate that respirometry is either not applicable or not relevant)
Table 8bis references cited in Table 8
B95: Bjerre et al., 1995 L92 Lesouef et al., 1992

ME76 Marais and Ekama, 1976
B97 Bogaert et al., 1997
B94 Brands et al., 1994 M95 Massone et al., 1995
N98 Naidoo etal., 1998
B98 Brouwer et al., 1998
N94 Nowak etal., 1994
C84 Cech et al., 1984
R97 Rozzi et al., 1997
D96 Dhaene, 1996
D93 Drtil et al., 1993 Sm96 Smets et al., 1996
SG91 Sollfrank and Gujer, 1991
DS94 Dupont and Sinkjaer, 1994
E86 Ekamaeia/., 1986 SV95 Spanjers and Vanrolleghem, 1995
E96 Ellis et al., 1996 S99 spérandio etal., 1999
S96 STOWA, 1996
G97 Gemaey et al., 1997
U97 Ubisi et al., 1997
G98 Gemaey et al., 1998
H87 Henze et al., 1987 U98 Urbain etal., 1998
V92 Vanrolleghem et al., 1992
KG92 Kappeler and Gujer, 1992
K98 Keesman et al., 1998 W93 Vanrolleghem and Verstraete, 1993
VG Volskay and Grady, 1990
K94 Kong et al., 1994
We95 Wentzel et al., 1995
K92 Kristensen et al., 1992
Kujawa and Klapwijk 1999 Wi96 Witteborg et al., 1996
KK99
However, the problem is not only whether it is possible to carry out a characterisation of
the different model components or parameters in lab-scale. A probably more relevant
question is whether it is possible to transfer the lab-scale observations to the full-scale
system. Or, to apply the terminology suggested by Grady etal. (1996), do the lab-scale
experiments provide extant kinetic parameters, i.e. parameters representative for the
biomass prior to the experiments? Furthermore it will be discussed how the relations are
between lab- and full-scale observations, and how the biological processes are presented
in ASM1:
• Transferability between lab-scale and full-scale observations: Are the different

components and parameters that may be determined via lab-scale experiments
representative, i.e. transferable to the full-scale system? That is, do the experiments
provide extant kinetic parameters?
• Transferability between full-scale observations and modelled processes: Are the full-
scale processes described in a biologically realistic way in the model or are the
model processes lumping different biological processes? If so, it may be impossible
to characterise them by any experiment.
• Transferability between model processes and lab-scale observations: Are the
processes defined in the model reflected by the lab-scale experiments?
Fig.22. Schematic representation of discussion on transferability
These conflicts of transferability are illustrated in figure 22, and the discussion is taken
below considering the different wastewater components, kinetic and stoichiometric
parameters. The aim of this discussion is to decide which information source is most
relevant for the different components and parameters. Of course, in principle all
components and parameters can be obtained from the model, e.g. via the default
parameter set or via adjustment of the values during the model calibration exercise.
However, some model processes do not reflect reality completely, although they
enable a mathematical description of the biological observations. The model components
and parameters related to such processes can not be characterised reliably via either lab-
scale or full-scale data and should preferably be tuned during the model calibration with
the full ASM1. Then there are some components and parameters that readily and reliably
can be transferred from a lab-scale experiment. For others the labscale results are difficult
to transfer to the model of the full-scale system, and for instance a mass balance with full-
scale data may be more appropriate as information source. Whether a certain component
or parameter should be obtained via lab-scale or full-scale data or should be tuned
directly via the model will depend on what the component or parameter in question is
depending on. In this discussion it is assumed that the values of the components and
parameters can depend on either the actual biomass in the activated sludge system or the
actual WWTP operation. It should be stressed that only the actual state of the system is
considered in this discussion, since this is what the calibrated model is aimed at
describing. Obviously, the biomass character (e.g. maximum specific growth rate, decay
rate etc.) of the WWTP is determined by both the incoming wastewater and WWTP
operation. However, changes in biomass characteristics caused by changing WWTP
operation or waste water character are more long-term effects. Description of these
effects is not within the scope of the ASM models. Thus, the actual wastewater
considered for the model calibration is assumed to be representative for the general
wastewater composition to which the biomass has adapted and by which the biomass
character is determined.
4,4.1. Kinetic and stoichiometric parameters

Below, the information sources for the most relevant kinetic and stoichiometric
parameters will be discussed in relation to figure 22. Furthermore, the discussion on
whether a parameter is depending on the waste water, biomass and/or WWTP operation
is summarised in table 9. Finally, the most relevant information source is indicated in
table 9. Brackets in table 9, i.e. (X), indicate that a lab-scale experiment is possible for
determination but the ttansferability of the obtained parameters to the full-scale situation
is uncertain for different reasons, as explained below. Finally, as mentioned above all
parameters can in principle be determined based on the model alone without additional
supporting information.
The maximum specific heterotrophic growth rate, UmaxH
The observed actual specific growth rate in the full-scale system, |i’ maxH, depends on the
sludge age and therefore depends both on the actual wastewater and the WWTP
operation. If the wastewater contains a significant amount of biomass, Il’maxH will
depend primarily on the wastewater, whereas it will depend on the operation if the
biomass is primarily produced within the plant. On the conttary, the maximum specific
growth rate, Pmaxib is the maximum possible specific growth rate of the actual sludge,
and is only influenced by the actual kind of bacteria present. It may be important to
distinguish here between PmaxH and the growth rate, |1, which is influenced by the
mixed liquor substrate concenttation. Thus, HmaxH is not depending on the wastewater
whereas |1 is.
This means that the problem of ttansferability between the lab-scale and the fullscale
observations will be insignificant (conflict a in figure 22) if the lab-scale experiment is
carried out under conditions that are comparable to the full-scale system (e.g. with respect
to pH, temperature, ratio between substtate and biomass concentration etc.). In other
words, if the lab-scale experiments are performed in a way that allows measurement of
extant parameter values, then little or no conflict will arise. As described earlier, the death
regeneration concept in the model has the effect that the cell mass turnover rate increases,
resulting in a higher growth rate than if a more traditional concept of endogenous decay
was applied. Thus, this should be taken into account in the interpretation of lab-scale
experiments and in the transferability of results to the fullscale model (conflict c in figure
22). Similarly the death regeneration model concept and the way it influences the
maximum specific growth rate may not reflect the fullscale process completely (conflict
B in figure 22), but may allow for an adequate description of observations.
Summarising, the PmaxH is one of the most relevant parameters to study in lab-scale
experiments and can be considered to be a biological parameter, which is only
determined by the actual bacteria present (see table 9).
The maximum specific autotrophic growth rate, Umax A

Although specific bacterial groups undertake nitrification, they adapt to the actual
envừonment and the bacterial species can therefore vary. Therefore, the discussion on the
maximum specific autottophic growth rate PmaxA is rather similar to the one of fimaxH-
Thus, it is possible to determine the value of PmaxA from lab-scale experiments, and
transfer the value to the model of the full-scale system.
Haự-saturatìon coefficients: Ks, KNH, KNQ, KQA and Kfffl

In pure cultures the half-saturation coefficients can be regarded as pure biological
parameters that give measures of the affinity of the biomass for substtates. However, in
cultures where the bacteria grow in flocks (as in activated sludge), the flock size and
structure play a role in the diffusion of substtate to the cell and thereby on the apparent
value of saturation coefficients. Especially in full-scale systems mixing characteristics
will further influence the apparent values. Even in lab-scale tests under simpler mixing
characteristics, mixing may play a role and influence the obtained values of the half -
saturation coefficients. Thus, the different mixing characteristics of the lab-scale and full-
scale system make it difficult to transfer the lab-scale observation to the full-scale system
(conflict A in figure 22). If the flock size decreases due to e.g. a more intensive mixing in
the small batch-scale experiments, the obtained coefficients will be smaller than required
to describe the full-scale behaviour (Henze et al., 1999). This makes it difficult to obtain
a model relevant value of the half-saturation coefficients from labscale experiments
(conflict c in figure 22). The saturation coefficients in ASM1 describing a full-scale
situation may therefore be regarded more as model parameters with the purpose of
preventing unrealistically high substrate uptake and growth rates. Thus, the biological
meaning of the model half-saturation coefficients is mixed with the hydraulics of the
system (conflict B in figure 22). Obviously, if a very detailed model is available to
describe the hydraulics of a system accurately, it may be possible to separate the effects
of biomass affinity for a substrate and the hydraulic effects from mixing. However,
usually the hydraulic pattern is approximated by a simple tanks-in- series model that may
be sufficient for a mathematical description but not accurate enough for a complete
elimination of hydraulic effects on the biological parameters.
Table 9. Discussion on relevant information sources for kinetic and stoichiometric parameters. A bracketed X
indicates that a lab-scale experiment is possible for determination but the transferability of the obtained
parameters to the full-scale situation is uncertain (see text for further explanation)
Dependency Relevant information source

Sludge/biomass Plant operation Lab-scale Full-scale data Model
experiment Mass balances calibration
gmax H X X X
gmaxA X X X
KS.KN X X (X) (X) X
O
KNH X X (X) (X) X
KOH, X X (X) (X) X
K QAA
bii,b X X X
YjnaxH X (X) X
YjnaxA X (X) X
kh X X
Kx X X
T|g X X X X
T|h X X
Thus, all half-saturation coefficients of the full-scale system will depend on both the
WWTP operation (mixing) and the actual kind of biomass present. The coefficients can
be determined by lab-scale experiments but the values obtained may not be very
representative. It may therefore be better to estimate these parameters from full-scale
data, via the operational rate of COD removal found by mass balances as function of the
operational range of COD concentrations. The question is of course whether the fullscale
data is informative enough for such determinations. Thus, in practice these values may
have to be tuned during the model calibration.
Decay rate of heterotrophs bn and autotrophs b A

The decay rate in a full-scale WWTP is in principle a characteristic of the actual biomass,
and can, similarly to the maximum specific growth, rate be considered as a biological
parameter. However, it may be difficult to obtain a representative value of the decay rates
of a full-scale system from the lab-scale tests presented above (conflict A in figure 22),
since decay and growth due to substrate inflow (and internal production) take place
simultaneously in the full-scale WWTP. On the contrary, decay is typically investigated
under starving conditions (endogenous respừation) in lab-scale experiments.
Furthermore, the decay rate in the full-scale plant is typically influenced by grazing, i.e.
presence of protozoa, which may not be present or may not be able to survive in the lab-
scale experiment.
In ASM1, the death regeneration concept includes both lysis combined with
hydrolysis of released substrate and, subsequently, growth on this substrate. As discussed
earlier, this interaction of different processes makes it difficult to determine the decay
coefficient related to the death regeneration concept (conflict B in figure 22). However,
according to ASM1 it is possible to transfer the decay rate obtained from a lab-scale
experiment with decreasing endogenous respừation as function of time for determination
of the endogenous decay rate to the death regeneration model concept (via Eq. 23,
conflict c in figure 22). Obviously, the change in ASM3 to the endogenous respừation
decay concept makes it more straightforward to determine the decay rate of the model by
a lab-experiment.
In conclusion, it is possible to determine the decay rate via lab-scale experiments, and
to convert the obtained value to the death regeneration concept of ASM1, but the value
may to some extent have to be adjusted during the model calibration procedure.
Maximum heterotrophic and autotrophic yield, YH and YA

The observed yields in a full-scale WWTP, Y’H and Y’A, are depending on the process
operation, i.e. the actual wastewater load and the sludge age. On the other hand, the
actual maximum yields (YH and YA) are depending on the kind of biomass present. For
municipal WWTP’s the parameters YH and YA are typically assumed to be rather
constant, indicating that the biomass character is rather similar among different municipal
WWTP’s. However, it may still be needed in some cases to determine the biomass yields.
This can be carried out in lab-scale experiments, but there may be some experimental
difficulties, e.g. caused by the possible influence of storage which may be induced by the
conditions in the lab-scale experiment (Majone et al., 1999), as earlier described (conflict
A in figure 22).
In fact the typical maximum heterotrophic yield of 0.67 for municipal wastewater
(Henze et al., 1987) is higher than the yields observed with pure cultures (Heijnen et al.,
1992). The reason for this may be that the model yield covers different processes as
storage, death regeneration etc. and may thereby be considered more as a model yield
(van Loosdrecht and Henze, 1999) (conflict B and c in figure 22). Although the
heterotrophic yield seems influenced by the available electron acceptors (the anoxic yield
is reported to be lower than the aerobic one, Koike and Hattori, 1975; Orhon et al., 1996;
McClintock et al., 1998; Spérandio et al., 1999), the yield may be more influenced by
storage than by the electron acceptor.
Hydrolysis rate kh and half-saturation coefficient Ky

Although only limited knowledge is available about hydrolysis, the process is needed in
ASM1 to describe the degradation of slowly biodegradable organic matter originating
from the influent COD and from internal turnover of substrate in the death regeneration
cycle.
As described above attempts have been made to analyse hydrolysis in lab-scale
experiments. It may be possible to compare the real enzymatic hydrolysis as it takes place
in lab-scale with the full-scale hydrolysis process. However, the real enzymatic
hydrolysis is not the same as the hydrolysis process in the model, as it might also cover
consumption of storage polymers, hydrolysis of decayed biomass (death regeneration),
protozoan activity etc. (conflict B and c in figure 22) (van Loosdrecht and Henze, 1999).
Thus, it remains problematic to design an experiment that is representative for both the
model concept and the hydrolysis process as it takes place in full-scale. If this is
compared to the determination of e.g. the maximum specific growth rate, we note that
this parameter also covers many details but still only describes one process, i.e. growth.
In conclusion, the real hydrolysis process is probably determined by the actual
biomass, which produces the enzymes, but for the model calibration of ASM1 it does not
seem relevant to attempt to characterise this process via lab-scale tests. Hence, the
hydrolysis as it is described in ASM1 should be considered as a model process that has to
be adjusted during the model calibration procedure. It should be remembered that the
definition of hydrolysis has changed in ASM3 and is closer to the real biological
hydrolysis. Thus, a characterisation of the hydrolysis parameters from a lab-scale
experiment will be more relevant for ASM3. The problem remains, however, to design a
good experiment for characterising the real biological hydrolysis.
Correction factors for denitrification n£ and rih

The correction factors for denitrification can be found via a combination of respirometric
and nitrate utilisation rate experiments for the determination of the growth and hydrolysis
process, although some problems may be encountered in the case where the aerobic and
anoxic yields can not be considered equal. It was also referred above that the correction
factors can be determined based on some general mass balances of the full-scale system
(Henze, 1986). Both correction factors will depend on the actual biomass character.
However, no particular conflicts, as indicated in figure 22, are apparent concerning the
correction factor for growth, T|g. Determination of the correction factor for hydrolysis
will suffer from the same problems as indicated above for the hydrolysis itself, and may
therefore also be considered more as a model parameter.
4.4.2. Relevant kinetic and stoichiometric parameters for lab-scale characterisation

In the discussion on the relevance of characterising the stoichiometric and kinetic
parameters of ASM1 via lab-scale experiments, one has to remember that none of the
ASM model processes are pure or microbiologically correct. To some extent they are all
bulk processes. It has clearly been illustrated above that experiments oriented in
identifying mechanisms inttoduced in the model might easily lead to conflict with the
actual model coefficients (van Loosdrecht and Henze, 1999). Thus, although possible, it
may not always be relevant to retrieve the model parameters from lab-scale tests.
Above the discussion was taken on these conflicts between lab- and full-scale
observations and the links to the model processes. Table 9 summarised the dependency of
the parameters on the biomass and WWTP operation, and it was attempted to indicate the
most relevant information source based on these discussions. Notice the difference to
table 8 that listed how the different parameters could be estimated from lab-scale tests,
whereas table 9 indicates whether this is relevant or not, considering that the parameter
should correspond reasonably well both with the full-scale behaviour i.e. extant
parameters are sought, and with the model concepts.
From table 9 it is deduced that it may be relevant to determine the following list of
stoichiometric and kinetic parameters from lab-scale experiments. It is not judged
whether it is always needed to characterise these parameters since that will depend on the
purpose of the model calibration. For the same reason it is not attempted to make an
indicative order of parameter importance.
• fimaxH
• PmaxA
• fig
• bA
• bH
• (YH)
• (YA)
The yields are included in the list, knowing that they are not easy to determine in labscale
tests and that they are usually assumed to be rather constant. However, it should also be
realised that the yield coefficients have an important influence on nearly all the processes
(see table 6), and therefore it would be rather relevant to have a more accurate
determination of these.
The remaining parameters can be determined via either full-scale data or dfrectly via
the model calibration, as indicated in table 9. It is important to notice that the above
parameter list is significantly reduced compared to the list of parameters rettieved from
experiments based on table 8, basically due to the fact that the half-saturation coefficients
and hydrolysis parameters are left out.
4.4.3. Relevant wastewater components for lab-scale characterisation

Only the side of the triangle dealing with the conflict between lab-scale observations and
model concepts (conflict C) outlined in figure 22 is relevant when it comes to
characterisation of wastewater components. Also, the discussion summarised in table 9 is
not relevant here, since the wastewater components do not depend on the biomass or the
WWTP operation. Therefore, the discussion on wastewater components is less extensive
here (see also the earlier discussion and summary of wastewater characterisation
methods) and is not divided according to the different components.
As discussed above in the review of wastewater characterisation a conflict may
indeed exist between the need for quantification of some of the ASM1 wastewater
components and what is practically obtainable from lab-scale experiments. The origin of
this problem mainly lays in the way the components are defined in ASM1. The death
regeneration cycle and the hydrolysis processes of ASM1 are model processes that are
not dừectly measurable in lab-scale experiments, as discussed above. Thus, the slowly
biodegradable substrate and inert particulate matter components, x s and XỊ respectively,
that are related to these processes, may then also be regarded as model components that
should rather be quantified during the model calibration exercise than through dedicated
experiments. Indeed, it was proposed by Henze et al, (1995) to estimate XỊ in the influent
via the complete model during the calibration of the sludge balance and, subsequently,
estimate xs from the difference between total COD and the other COD components, as
discussed earlier. A determination of the heterotrophic biomass (XBH) in the wastewater
is possible via lab-scale experiments, as described above. However, in most cases the
XBH present in waste water is not of great importance, since the growth rates are so high
that wash-out of XBH never occurs in practice. Thus, an inclusion of X BH in the xs
component does not affect the modelling significantly, although it will affect the value of
the heterotrophic yield coefficient (a slightly smaller yield may need to be chosen)
(Henze et al., 1999). On the confrary, the presence of autotrophic biomass (XBA) in the
wastewater may be of importance to prevent wash out of the nitrifiers. The concenfration
of XBA can in principle be determined via lab-scale experiments, but in practice the
procedure may not be straightforward and XBA may rather be adjusted during the model
calibration.
In general there is no need for a detailed characterisation of the nitrogen components
since the main part of nittogen in wastewater is ammonium without any coupling to the
organic matter (Henze et al., 1999). An exception to this may, however, exist for some
industrial wastewaters. Thus, the wastewater components relevant to be characterised
separately via lab-scale experiments are listed below. Again, an indicative order of
importance is not aimed for, since this will depend on the actual case.
• SNH
• Ss
• Sj
• (SND, XND)
The relevance of determining the inert soluble matter (Sĩ) is linked to the determination
of the soluble readily biodegradable substrate (S s) since Si may be needed for the mass
balance of soluble COD.
5. Biological experimental constraints
In the previous section the wastewater components and the stoichiomefric and kinetic
parameters that are considered most relevant to be determined in lab-scale experiments
were listed. This list was compiled on the basis of considerations that the component or
parameter resulting from the lab-scale experiment should be relevant to full-scale
behaviour and fit within the model concepts.
In this last section, we will further zoom in on the problem of transferability between lab-
scale results and full-scale behaviour, i.e. the problem of obtaining extant kinetic
parameters. As discussed above care should be taken in the transfer of results derived
from lab-scale experiments to a model of the full-scale system. Summarising, the reason
for problems with transferability are on the one hand differences in biological
experimental conditions between lab-scale and full-scale experiments (conflict A in
figure 22) and, on the other hand, differences in the models used (conflict c in figure 22).
At the experimental level the lab-scale behaviour may not equal the full-scale
behaviour due to, for instance, differences in feeding pattern resulting in other
concentration profiles, differences in envfronmental conditions such as pH, temperature
or mixing behaviour, or differences in sludge history. One of the most discussed
biological experimental factors is the ratio between initial substrate concentration (So)
and initial biomass concentration (Xo). This S(0)/X(0) ratio is considered to be one of the
important factors determining (1) the response of the sludge with a certain wastewater or
substrate and (2) whether the experimental response is sufficiently informative for
adequate interpretation. The first point is of a more basic nature since it has been
observed that the S(0)/X(0) ratio dfrectly influences the behaviour of the sludge, leading
to different characteristics (Chudoba et al., 1992; Grady et al., 1996, Pollard et al., 1998).
The second point is more related to the practical identifiability of model parameters, i.e. it
affects the quality of the experimental data (Spanjers and Vanrolleghem, 1995; Spérandio
and Paul, 2000). For instance, if S(0)/X(0) is very high the measured response, e.g.
respfration rate, may be too small and the experiment may take too long. On the other
hand, if S(0)/X(0) is very low the respfrometric response may be too short for a reliable
measurement, or it may be swamped into the endogenous respừation rate. Below, special
attention is paid to the first point, where the S(0)/X(0) problem will be discussed in more
detail.
At the modelling level the results from lab-scale experiments may be described with a
model different from the model used to describe the full-scale behaviour. Although not
obvious at first sight, the use of a simple model for interpretation of the lab-scale data
increases calculation speeds significantly, resulting in, for instance, a faster and more
straightforward parameter estimation. Problems arise when the model uses different
concepts that may not allow to transfer the estimated parameters from one model to the
other, e.g. the death regeneration versus endogenous respừation concept (Yuan and
Stenstrom, 1996).
5.1. TRANSFERABILITY BETWEEN MODEL CONCEPTS: AN EXAMPLE
In ASM1 the death regeneration concept is applied, whereas the model to describe the
lab-scale results may only include the degradation of substrates, i.e. decay and death
regeneration are omitted, because they are considered insignificant in relation to the time
scale used in the experiment (Spanjers and Vanrolleghem, 1995). In ASM1 oxygen is
consumed for growth on incoming substrate plus growth on substrate produced due to
death regeneration, whereas in a lab-scale model one may only consider that oxygen is
consumed for growth on incoming substrate. This is illustrated in figure 23.
In this figure the line illustrates substrate uptake rate, r s, as function of time and the
values at the left hand side of the y-axes indicate the corresponding substrate
concentrations Ss. The left figure illustrates how the substrate uptake rate is interpreted in
the lab-scale (batch) experiments whereas the right figure gives the ASM1 interpretation.
In both cases the total oxygen consumption rate is the same, but it is interpreted
differently in the lab and full-scale model. In the lab-scale model oxygen is consumed to
degrade the incoming substrate and the substrate concentration will eventually go to zero.
Apart from oxygen for substrate degradation (r0,ex)» oxygen is also used for endogenous
respftation (r0,end)« In ASM1 substrate will also be degraded. However the concentration
will not reach zero since there will be some production of substrate from the death
regeneration process. Thus, according to the ASM1 model concept oxygen will be
consumed for degradation of both the incoming substrate and the produced substrate. In
figure 23 it is assumed for clarity that the concentration of the produced substrate is 10
mg COD/1. This slightly higher substrate availability in ASM1 means that the
contribution of the observed total r0 to degradation of incoming substrate is lower in
ASM1 than in the lab-scale model. As a consequence the estimated maximum growth
rate, which is proportional to the maximum r0, will be lower in the batch system. This is
illusfrated with the size of the double arrow at the right hand side of both graphs in figure
23. Also, the value of the half-saturation coefficient K s will be underestimated in the
batch model compared to ASM1. In the batch model this is illustrated by a K s value of 50
whereas it may be 55 in ASM1.
Fig.23. Illustration of difference in interpretation of substrate uptake rate in lab-scale (endogenous

respiration, left) model versus ASM1 (death regeneration, right).
As discussed earlier, it is possible to derive analytical transformations between both

model concepts for the decay and growth rates, the yield and the fraction of inert material
produced (Henze et al., 1987). However, a transformation for Ks is more complicated.
5.2. REVIEW AND DISCUSSION OF S(0)/X(0) RATIO
Depending on the experimental conditions the organic substrate (COD) uptake rate in
both lab- and full-scale may consist of different responses. This is illustrated in figure 24.
In this concept COD is produced from decay (flow 1). Maintenance (flow 2) is defined as
the external substrate requftements to maintain the organisms in theft current State. Note
the difference here to endogenous respfration, which can be defined as the respfration in
absence of external substrate (for a detailed review see van Loosdrecht and Henze, 1999).
However, here it is assumed that external substrate is present. Growth (flow 3) is divided
in two; (i) increase in biomass due to production of cell constituents (e.g. proteins etc.)
but without cell multiplication, (ii) increase in biomass caused by cell multiplication.
Storage (flow 4) is defined as the accumulation of polymers, e.g. poly-hydroxy-
alkanoates and glycogen. Energy spilling (flow 5) (Zeng et al., 1995) is defined as
substrate waste that may take place when the organisms are exposed to very high
substrate concentrations. In such cases the organisms may not be able to regulate the
catabolism rate to the needs for anabolism, resulting in inefficient use of substrate and
possible excretion of metabolites. Figure 24 illusfrates the possible COD flows in the
single organisms. Depending on the experimental conditions one of the flows may
dominate in a single organism (Fig. 24). The same experimental conditions also provoke
a particular distribution of COD over the different organisms. Competition may
eventually lead to a shift in the population (Novak et al., 1994).
Fig.24. Different flows of external COD in the organisms
As mentioned above the S(0)/X(0) ratio is considered to be one of the determining factors
for the way the organisms respond in a system. However, even though the importance of
this ratio has been recognised, only few references that deal with the subject in more
detail can be found (Chudoba et al., 1992; Novak et al. 1994; Zeng et al., 1995; Grady et
al., 1996; Liu, 1996). They all deal with the subject from a more theoretical point of view
without much experimental support, and there is still a lack of both qualitative and
especially quantitative explanation of the exact role of the S(0)/X(0) ratio. The discussion
on the effect of S(0)/X(0) can be considered from a reaction stoichiometry or reaction
kinetics point of view.
5.2.1. Effect ofS(0)/X(0) on stoichiometry

Both Chudoba et al. (1992) and Liu (1996) explain the importance of the S(0)/X(0) ratio
from a thermodynamic point of view based on the observations that the observed yield
(Y’H) decreased with increasing S(0)/X(0) ratio (Fig. 25).
In the work of Chudoba et al. (1992) substrate (COD) profiles versus time were
measured. Here it was assumed that autocatalytic growth would cause substrate uptake at
an increasing rate whereas substrate uptake at a constant rate was assumed to be an
indirect evidence of storage. It was hypothesised that at low S(0)/X(0) ratio the main
response is storage (flow 4 in figure 24) since the energy level in the cell will be too low
to trigger cell multiplication, resulting in less subskate being oxidised (Daigger and
Grady, 1982) and thereby a higher Y’H.
At high S(0)/X(0) ratios on the contrary, the growth response where cell multiplication
(flow 3 in figure 24) is dominating results in lower observed yields (Chudoba et al.,
1992). However, the lower Y’H at higher S(0)/X(0) ratios may as well be explained by
less energy being requừed for growth without associated cell multiplication (flow 3) and
without the involvement of storage (flow 4). A second possible explanation of the data of
Chudoba et al. (1992) is that the contribution of endogenous respừation to the total
amount of oxygen consumed is higher at high S(0)/X(0) ratio. This could also result in
lower Y’H, since the experiments at high S(0)/X(0) ratios take longer time and therefore
the amount of decayed biomass (flow 1) is higher.
so/xở (mgCOD/mgMLSS)
Fig.25A. Literature data of Yobs as function of S(O)/X(O) ratio Rao and Gaudy, 1966; Data digitised from Liu
(1996).
0.45 -
0.35
I
I 0.25
0.05
10 20 30 40 50 60
SQ/XO (mgCOD/mgMLSS)
Fig.25B. Literature data of Yobs as function of S(0)/X(0) ratio.Chudoba et al., 1969; Data digitised from
Liu (1996).
A Still different explanation of the decreasing observed yield with increasing S(0)/X(0) is
found in the work of Liu (1996), who presented an attempt to quantify the importance of
S(0)/X(0). Here the decrease in Y’H is explained by an increase in energy spilling (flow
5) with increasing S(0)/X(0) (Liu, 1996). However, the problem in verifying this
approach is to define at which S(0)/X(0) energy spilling will start to take place. In the
study of Liu (1996) the ratio is assumed to be 1. The proposed model was tested on
literature data, but the S(0)/X(0) ratios of all the literature data used in the study were
higher than 1, making the evidence for the model incomplete. It should be noted that none
of these studies attempted to explain the observed behaviour with a more complex model,
such as ASM1.
so/xo (mgCOD/mgMLSS)
Fig.25C. Literature data of Yobs as function of S(0)/X(0) ratio. Chang et al., 1993; Data digitised from Liu (1996).
SQ/XQ (mgCOD/mgMLSS)
Fig.25D. Literature data of Yobs as function of S(0)/X(0) ratio. Chudoba et al., 1991. Data digitised from
Liu (1996).
5.2.2. Effect ofS(0)/X(0) on kinetics
Another way of looking at the influence of S(0)/X(0) is from a kinetic point of view
focusing on the physiological, i.e. enzymatic, state and adaptation. In order to describe
these phenomena, the concept of the machinery necessary for protein synthesis (PSS) has
been introduced (Grady et al., 1996). This should basically be understood as follows. If
the organisms are adapted to grow under substrate limited conditions, its PSS will not be
sufficient to quickly increase the growth rate if the substrate limitation is removed. Thus,
the PSS and eventually the specific growth rate will gradually increase during time, until
the maximum possible value according to the new conditions, i.e. physical adaptation has
taken place. It has been stated that the synthesis of storage polymers requires less
physiological adaptation than the growth response (Daigger and Grady, 1982). Thus, this
would mean that if a substtate limitation is removed, as described above, a storage
response may be triggered as a fast response and as an alternative mechanism when the
growth response is too slow.
A simple example of physiological adaptation is illustrated in figure 26 where three
pulses of acetate were added consecutively to a sludge sample (Vanrolleghem et al.,
1998). Each of the three responses is characterised by a fast start-up of about two
minutes. These two minutes are assumed to be the time needed by a cell to take up fresh
substrate and oxidise it (Vanrolleghem et al., 1998). In the first two responses a more
gradual increase of r0 is observed for about 10 minutes, presumably due to an increased
conversion capacity (e.g. enzyme activation or synthesis). In the third response (after 40
minutes) this capacity has become constitutive. Starvation of the culture for one night
turned the capacity down (the organisms “forgot”) and a similar behaviour could be
observed when acetate was added again (results not shown).
Fig.26. ro,ex profiles obtained by 3 additions of acetate to an activated sludge sample (Vanrolleghem et al.,
1998).
In both Chudoba et al. (1992) and Liu (1996) the applied S(0)/X(0) ratios are very high
(above 1), whereas in the example of figure 26 the S(0)/X(0) ratio was very low (below
1/20). It is commonly assumed that it is necessary to work under low S(0)/X(0) ratios
(Chudoba et al., 1992; Novak et al, 1994; Spanjers and Vanrolleghem, 1995; Grady et
al., 1996). Indeed, if the S(0)/X(0) ratio is high this may result in a change of maximum
specific growth and substtate removal rate due to physiological adaptation, which
eventually may result in changes of the proportions among slow-growers and fastgrowers
leading to population shifts (Novák et al., 1994). The kinetics measured under such
conditions will more represent the ultimate capabilities of the organisms (intrinsic
kinetics), whereas kinetics measured in experiments performed under low S(0)/X(0) ratio
may be more representative of the physiological state of the cells prior to the experiments
(extant kinetics) (Grady et al., 1996). In the example of Kappeler and Gujer (1992) a very
high S(0)/X(0) ratio was applied resulting in overestimation of the growth rates due to
shift in biomass composition towards fast-growers. In addition, population shifts will also
take place ư the subsfrate source is changed.
5.2.3. Discussion on S(0)/X(0) ratio

As illustrated above the discussion on the effect of the S(0)/X(0) ratio is looked at from
many angles and it seems difficult to draw a coherent picture. However, instead of
focusing on a threshold value for the S(0)/X(0) ratio it may be more relevant to consider
the following factors in the discussion of what kind of response can be expected in a lab-
scale experiment:
• AS: how big is the change of substrate concentration in the lab-scale system
compared to the full-scale system, i.e. to what extent are organisms subjected to a
drastic change in thefr environmental conditions.
• Time: for how long is AS maintained, i.e. what is the time frame of the experiments.
• History: how strong is the history of the sludge, e.g. starvation periods prior to the
experiment
These three factors should be understood as follows. If AS is low and the experiment is
performed over short-term, the risk for changing the response of the sludge compared to
the full-scale system is probably low and extant parameters can be obtained. If AS is high
and the time is short the risk for excess substrate uptake not resulting in immediate
growth increases (maybe induction of storage or spilling). Finally, if AS is high and the
experiment is performed over long-term the risk for physiological adaptation due to
enzymatic changes is increasing, eventually leading to a population shift. The specific
growth rate may increase during the experiment resulting in an increase in growth
response and a decrease in excess substrate uptake response, i.e. the initial stress reaction
such as storage or energy spilling will decrease as the organisms get adapted to the new
environment. Thus, somehow a compromise between AS and time is needed.
Furthermore, the history of the sludge will play a role in the experimental designs,
since for example starvation periods prior to the experiment will result in an initial slower
response of the sludge. It is, however, not really clear if for example starvation periods
can lead to an initial different response.
The above discussion on S(0)/X(0) focused on heterotrophic organisms and their
response to a carbon substrate. However, the discussion can easily be extended to
autotrophic organisms where the substrate is ammonium. In this case a too high AS may
result in inhibition of the nittification process. However, the risk for a population shift
may be lower since the nitrifying group of organisms is supposed to be rather uniform in
character. Still, adaptations to new envừonments will take place and the bacterial species
can vary.
6. Summary
In this extensive review numerous aspects of activated sludge model calibration have
been touched upon. As an introduction the industry-standard Activated Sludge Model No.
1 was introduced to set the scene and it was compared to the more recent update ASM3.
The waste water and sludge fractions considered in these models were described and the
processes taking place among them were given. All these items are focused upon when
calibrating such model.
In a next section an overview was given on the descriptions of calibration procedures
that were found in literature. Surprisingly, it is not possible to find a single paper where a
comprehensive overview is given. The information is only available as “bits and pieces”
and is scattered in a vast amount of literature. The information sets that are typically
requừed were presented and a 10-step calibration procedure was proposed.
The multitude of methods for model calibration was structured along three lines: (1)
wastewater characterisation, (2) sludge composition analysis and (3) stoichiometric and
kinetic parameters.
The wastewater characterisation is typically done either by physical-chemical or
biological characterisation methods. Whereas the former appear the easiest to apply, even
in routine lab analysis, theừ results are not directly related to the model concepts and,
moreover, the results need to be augmented with specific characteristics obtained from
biological characterisation methods. Among these biological methods attention was
particularly given to the respừometric tests as they form the core technique, but nittate
utilisation tests and the upcoming titrimetric tests were presented as well. For the
exttaction of the model-related information, either dừect or model-based analysis is
needed. Whereas the former is really simple, the latter allows exttacting multiple
characteristics from a single experiment.
For the sludge composition analysis, mainly in-out mass balancing methods are being
used. The estimation of stoichiometric and kinetic parameters is typically based on
dedicated batch experiments using respừometers. Special attention was drawn to the
simultaneous estimation of parameters from well-designed single experiments. Especially
for this, model-based analysis is requừed. It is also noteworthy that these more complex
approaches not only lead to stoichiomettic and kinetic parameter estimates, but typically
also lead to estimates on wastewater composition.
In the last section of this review attention was focused upon the problem of
ttansferring the results of the specific tests to a model apt to describe the full-scale
behaviour. It was indeed argued that quite some estimation results give a near-perfect
description of what happened in the batch test. However this result could not be applied
in the practical situation because, for instance, the insufficiently modelled mixing
characteristics have to be lumped into the biological parameters of the full-scale model.
Still, it was attempted to point towards the parameters whose values can most likely be
assessed realistically from lab-scale tests and transferred to the full-scale model.
All in all, this review has led to the belief that a considerable potential exists for
efficient characterisation of Activated Sludge Models, provided that precautions are taken
with respect to constraining the experimental conditions. The PhD thesis of Petersen
(2000) was entirely devoted to this question. The thesis focused on the design of optimal
experiments that not only lead to high-information content data sets with good
identifiability properties, but that also take into account the biological constraints to
guarantee transferability of calibration results to the full-scale model.
Acknowledgement
The work reported in this paper was supported (in part) by FWO-project G.0286.96 of the
Fund for Scientific Research (Belgium), and by the Flemish Institute for the Promotion of
Scientific-Technological Research in Industry (IWT, Brussels, Belgium).
References
Avcioglu E., Orhon D. and Sozen s. (1998) A new method for the assessment of heterotrophic endogenous
respiration rate under aerobic and anoxic conditions. Water Sci. Technol., 38(8-9), 95-103.
Bjerre H.L., Hvitved-Jacobsen T., Teichgrăber B. and te Heesen D. (1995) Experimental procedures
characterizing transformations of wastewater organic matter in the Emscher river, Germany. Water Sci.
Technol., 31(7), 201 - 212.
Boero V.J., Eckenfelder w.w. Jr. and Bowers A.R. (1991) Soluble microbial product formation in biological
systems. Water Sci. Technol., 23,1067 - 1076.
Bogaert H., Vanderhasselt A., Gemaey K., Yuan z., Thoeye c. and Verstraete w. (1997) New sensor based on
pH effects of denitrification process. J. Environ. Engineering., 123, 884 - 891.
Bortone G., Cech J.S., Germirli F., Bianchi R, and Tilche A. (1994) Experimental approaches for the
characterisation of a nitrification/denitrification process on industrial wastewater. Water. Sci. Technol.,
29(7), 129-136.
Brands E., Liebeskind M. and Dohmann M. (1994) Parameters for dynamic simulation of wastewater treatment
plants with high-rate and low-rate activated sludge tanks. Water Sci. Technol., 30(4), 211 - 214.
Brouwer H., Klapwijk A. and Keesman J. (1998) Identification of activated sludge and wastewater
characteristics using respirometric batch-experiments. Water Res., 32,1240 - 1254.
Bunch B. and Griffin D.M. Jr. (1987) Rapid removal of colloidal substrate from domestic wastewater. J. Water
Pollut. Confrol Fed., 59, 957 - 963.
Cech J.S., Chudoba J. and Grau p. (1984) Determination of kinetic constants of activated sludge
microorganisms. Water Sci. Technol., 17, 259 - 272.
Chang J., Chudoba p. and Capdeville B. (1993) Determination of the maintenance requirements of activated
sludge. Water Sci. Technol., 28,139 - 142.
Chudoba J. (1969) Residual organic matter in activated sludge processes effluents V-effluent of the initial food-
to-microorganisms ratio. Sci. Paper, Inst. Chem. Technol. Prague F1-F5, 23 - 34.
Chudoba J. (1985) Quantitative estimation in COD units of refractory organic compounds produced by
activated sludge microorganisms. Water Res., 19, 37 - 43.
Chudoba p., Chevalier J.J., Chang J. and Capdeville B. (1991) Effect of anaerobic stabilisation of activated
sludge on its production under batch conditions at various So/Xo. Water Sci. Technol., 23, 917 - 926.
Chudoba p., Capdeville B. and Chudoba J. (1992) Explanation of biological meaning of the So/Xo ratio in lab-
scale cultivation. Water Sci. Technol., 26(3-4), 743 - 751.
Ciaccio L.L. (1992) Instrumental determination of energy oxygen and BOD5. Water Sci. Technol., 26(5-6),
1345 -1353.
Coen F., Petersen B., Vanrolleghem P.A., Vanderhaegen B. and Henze M. (1998) Model-based characterisation
of hydraulic, kinetic and influent properties of an industrial WWTP. Water Sci. Technol., 39(1), 195-214.
Coen F., Vanderhaegen B., Boonen I., Vanrolleghem P.A. and Van Meenen p. (1997) Improved design and
control of industrial and municipal nutrient removal plants using dynamic models. Water Sci. TechnoL,
35(10), 53-61.
Copp J.B. and Dold P.L. (1998) Confirming the nitrate-to-oxygen conversion factor for denitrification. Water
Res., 32, 1296-1304.
Daigger G.T. and Grady C.P.L. (1982) The dynamics of microbial growth on soluble substrates. A unifying
theory. Water Res., 16, 365 - 382.
De Clercq B., Coen F., Vanderhaegen B. and Vanrolleghem P.A. (1999) Calibrating simple models for mixing
and flow propagation in waste water treatment plants. Water Sci. Technol., 39(4), 61 - 69.
de la Sota A., Larrea L., Novak L., Grau p. and Henze M. (1994) Performance and model calibration of R-D- N
processes in pilot plant. Water Sci. Technol., 30(6), 355 - 364.
Dhaene R. (1996) Een nieuwe biosensor voor het denitrificatieproces gebaseerd op een pH-regelaar. M.Sc.
thesis at Hogeschool West-Vlaanderen, Kortrijk, Belgium, (in Dutch)
Dircks K., Pind P.F., Mosbaek H. and Henze M. (1999) Yield determination by respirometry - The possible
influence of storage under aerobic conditions in activated sludge. Water SA, 25, 69 - 74.
Dold p. (1980) A general model for the activated sludge process. Prog. Wat. Tech., 12(6), 47 - 77.
Dupont R. and Sinkjaer o. (1994) Optimisation of wastewater treatment plants by means of computer models.
Water Sci. Technol., 30(4), 181 - 190.
Drtil M., Németh p. and Bodík I. (1993) Kinetic constants of nitrification. Water Res., 27, 35 - 39.
Dupont R. and Sinkjaer o. (1994) Optimisation of wastewater treatment plants by means of computer models.
Ekama G.A., Dold P.L. and Marais G.V.R. (1986) Procedures for determining influent COD fractions and the
maximum specific growth rate of heterotrophs in activated sludge systems. Water Sci. Technol., 18(6), 91
-114.
Ellis T.G., Barbeau D.S., Smets B.F. and Grady C.P.L. Jr. (1996) Respirometric techniques for determination of
extant kinetic parameters describing biodegradation. Water Environ. Res., 38, 917 - 926.
Farkas p. (1981) The use of respirography in biological treatment plant control. Water Sci. Technol., 13, 125
-131.
Funamizu N. and Takakuwa T. (1994) Simulation of the operating conditions of the municipal wastewater
treatment plant at low temperatures using a model that includes the IAWPRC activated sludge model.
Water Sci. Technol., 30(4), 150-113. - - -
Germirli F., Orhon D. and Artan N. (1991) Assessment of the initial inert soluble COD in industrial
wastewaters. Water Sci. Technol., 23(4-6), 1077 - 1086.
Gemaey K., Bogaert H., Massone A., Vanrolleghem p. and Verstraete w. (1997) On-line nitrification
monitoring in activated sludge with a titrimetric sensor. Environ. Sci. Technol., 31, 2350 - 2355.
Gemaey K., Vanrolleghem P.A. and Verstraete w. (1998) On-line estimation of Nitrosomonas kinetic
parameters in activated sludge samples using titration in-sensor-experiments. Water Res., 32, 71 - 80.
Grady C.P.L., Smets B.F and Barbeau D.s. (1996) Variability in kinetic parameter estimates: a review of
possible causes and a proposed terminology. Water Res., 30,742 - 748.
Gujer w., Henze M., Mino T. and van Loosdrecht M.C.M. (1999) Activated sludge model No. 3. Water Sci.
Technol., 39(1), 183 - 193.
Haider s. (2000) CSB-Elimination in A-stufen und ihre Auswirkung auf die Stickstoffelimination von AB-
Anlagen unter dem Gesichtspunkt der mathematischen Modellierung. PhD. Thesis. TU Wien, Austria, (in
preparation)
Heijnen J.J., van Loosdrecht M.C.M. and Tijhuis L. (1992) A black box mathematical model to calculate auto-
and heterotrophic biomass yields based on Gibbs energy dissipation. Biotechnol. Bioeng., 40,1139 -1154
Henze M. (1986) Nitrate versus oxygen utilisation rates in wastewater and activated sludge systems. Water Sci.
Technol., 18(6), 115 - 122.
Henze M. (1992) Characterization of wastewater for modelling of activated sludge processes. Water Sci.
Technol. 25(6), 1 - 15.
Henze M., Grady C.P.L. Jr., Gujer w., Marais G.V.R. and Matsuo T. (1987) Activated Sludge Model No. 1.
IAWQ Scientific and Technical Report No. 1, London, UK.
Henze M., Gujer w., Mino T., Matsuo T., Wentzel M.C.M. and Marais G.V.R. (1995) Activated Sludge Model
No. 2. IAWQ Scientific and Technical Report No. 3, London, UK.
Henze M., Gujer w., Mino T., Matsuo T., Wentzel M.C., Marais G.V.R. and van Loosdrecht M.C.M. (1999)
Activated sludge model No. 2D, ASM2D. Water Sci. Technol., 39(1), 165 - 182.
Henze M., Harremoẽs p., la Cour Janssen J. and Arvin E. (1997) Biological and chemical wastewater treatment,
2.edition, springer, Berlin.
Holmberg A. (1982) On the practical identifiability of microbial growth models incorporating Michaelis-
Menten type nonlinearities. Mathematical Biosciences, 62, 23-43.
Jeppsson u. (1996). Modelling aspects of wastewater treatment processes. Ph.D. thesis: Department of Industrial
Electrical Engineering and Automation, Lund Institute of Technology, Sweden, pp. 428.
Kappeler J. and Gujer w. (1992) Estimation of kinetic parameters of heterotrophic biomass under aerobic
conditions and characterization of wastewater for activated sludge modelling. Water Sci. Technol., 25(6),
125-139.
Keesman K.J., Spanjers H. and van Straten G. (1998) Analysis of endogenous process behaviour in activated
sludge. Biotechnol. Bioeng., 57,155 - 163.
Koike I. and Hattori A. (1975) Growth yield of a denitrifying bacterium, Pseudomonas denitrificans, under
aerobic and denitrifying conditions. J. Gen. Microbiol., 88,1-10.
Kong z., Vanrolleghem P.A. and Verstraete w. (1994) Automated respừation inhibition kinetics analysis
(ARIKA) with a respừographic biosensor. Water Sci. Technol., 30(4), 275 - 284.
Krishna c. and van Loosdrecht M.C.M. (1999) Substrate flux into storage and growth in relation to activated
sludge modelling. Water Res., 33, 3149 - 3161.
Kristensen H.G., Elberg J0rgensen p. and Henze M. (1992) Characterisation of functional micro-organism
groups and substrate in activated sludge and wastewater by AUR, NUR and OUR. Water Sci. Technol.,
25(6), 43 - 57.
Kristensen H.G., la Cour Janssen J. and Elberg Jprgensen (1998) Batch test procedures as tools for calibration
of the activated sludge model - A pilot scale demonstration. Water Sci. Technol., 37(4-5), 235 - 242.
Kujawa K. and Klapwijk B. (1999) A method to estimate denitrification potential for predenitrificaiton systems
using NUR batch test. Water Res., 33(10), 2291 - 2300.
Larrea L., Garcia-Heras J.L., Ayesa E. and Florez J. (1992) Designing experiments to determine the coefficients
of activated sludge models by identification algorithms. Water. Sci. Technol., 25(6), 149 - 165.
Leenen E.J., Boogert A.A. van Lammeren A.A., Tramper J. and Wijffels R.H. (1997) Dynamics of artificially
immobilized Nitrosomonas europaea: Effect of biomass decay. Biotechnol. Bioeng., 55,630 - 641.
Lesouef A., Payraudeau M., Rogalla F. and Kleiber B. (1992) optimising niữogen removal reactor
configurations by on-site calibration of the IAWPRC activated sludge model. Water Sci. Technol., 25(6),
105 - 123.
Levine A.D., Tchobanoglous G. and Asano T. (1985) Characterisation of the size distribution of contaminants
in wastewater: treatment and reuse implications. J. Water Pollut. Control Fed., 57(7), 805 - 816.
Liebeskind M., Schapers D., Bomemann c., Brands E., Freund M. and Rolfs T. (1996) Parameter determination
and model fitting - two approaches for modelling processes in wastewater treatment plants. Water Sci.
Technol., 34(5-6), 27 - 33.
Liu Y. (1996) Bioenergetic interpretation on the So/Xo ratio in substrate-sufficient lab-scale culture. Water
Res', 30,2766-2770.
Ljung L. (1987) System Identification - Theory for the User. Prentice-Hall, Englewood Cliffs, New Jersey.
Lukasse L.J.S., Keesman K.J. and van Straten G. (1997) Estimation of BODst, respừation rate and kinetics of
activated sludge. Water Res., 31, 2278 - 2286.
Majone M., Dừcks K. and Beun J.J. (1999) Aerobic storage under dynamic conditions in activated sludge
processes. The state of the art. Water Sci. Technol., 39(1), 61 - 73.
Mamais D., Jenkins D. and Pitt p. (1993) A rapid physical-chemical method for the determination of readily
biodegradable soluble COD in municipal wastewater. Water Res., 27,195 - 197.
Marais G.V.R. and Ekama G.A. (1976) The activated sludge process. Part 1 - Steady state behaviour. Water SA,
2,163 - 199.
Massone A., Gemaey K., Rozzi A., Willems p. and Verstraete w. (1995) Ammonium concentration
measurements using a titrimetric biosensor. Med. Fac. Landbouww. Univ. Gent, 60, 2361 - 2368.
McClintock S.A., Sherrard J.H., Novak J.T. and Randall c.w. (1988) Nitrate versus oxygen respừation in the
activated sludge process. J. Water Pollut. Control Fed., 60, 342 - 350.
Melcer H. (1999) Full scale experience with biological process models - calibration issues. Water Sci. Technol.,
39(1), 245 - 252.
Mino T., San Pedro D.C., Yamamoto s. and Matsuo T. (1997) Application of the IAWQ activated sludge model
to nutrient removal process. Water Sci. Technol., 35(8), 111-118.
Miinch E.v. and Greenfield P.F. (1998) Estimating VFA concenttations in prefermenters by measuring pH.
Water Res., 32, 2431 - 2441.
Naidoo V., Urbain V. and Buckley C.A. (1998) Characterisation of wastewater and activated sludge from
European municipal wastewater tteatment plants using the NUR test. Water Sci. Techno!, 38(1), 303 - 310.
2
Nichols H.A., Pitman A.R. and Osborn D.w. (1985) The readily biodegradable fraction of sewage: its influence
on phosphorus removal and measurement. Water Sci. Technol., 17, 73 - 87.
Novák L., Larrea L. and Wanner J. (1994) Estimation of maximum specific growth rate of heterottophic and
autottophic biomass: A combined technique of mathematical modelling and batch cultivations. Water Sci.
Techno!, 30(11), 171-180.
Nowak o., Franz A., Svardal K., Muller V. and Kuhn. (1999) Parameter estimation for activated sludge models
with help of mass balances. Water Sci. Technol., 39(4), 113-120.
Nowak o., Schweighofer p. and Svardal K. (1994) Nitrification inhibition - A method for the estimation of
actual maximum growth rates in activated sludge systems. Water Sci. Technol., 30(6), 9-19.
Orhon D., Artan N. and Cimsit Y. (1989) The concept of soluble residual product formation in the modelling of
activated sludge. Water Sci. Technol., 21(4-5), 339 - 350.
Orhon D., Sozen s. and Artan N. (1996) The effect of heterottophic yield on assessment of the correction factor
for the anoxic growth. Water Sci. Technol., 34 (5-6), 67 - 74.
Payne W.J. (1991) Denitrification. Wiley-Interscience, New York.
Pedersen J. and Sinkjaer o. (1992) Test of the activated sludge models capabilities as a prognostic tool on a pilot
scale wastewater treatment plant. Water Sci. Technol., 25(6), 185 - 194.
Petersen B. (2000) Calibration, identifiability and optimal experimental design of activated sludge models. PhD
Thesis. BIOMATH Department, Ghent University, Belgium.
Pollard P.C., Steffens M.A., Biggs C.A. and Lant P.A. (1998) Bacterial growth dynamics in activated sludge
batch assays. Water Res., 32, 587 - 596.
Ramadori R., Rozzi A. and Tandoi V. (1980) An automated system for monitoring the kinetics of biological
oxidation of ammonia. Water Res., 14,1555 - 1557.
Rao B.s. and Gaudy A.F.Jr (1966) Effect of sludge concentration on various aspects of biological activity in
activated sludge. J. Water. Pollut. Conttol. Fed., 38,794 - 812.
Rickert D.A. and Hunter J.v. (1971) General nature of soluble and particulate organics in sewage and secondary
effluent. Water Res., 5,421 - 436.
Robinson J.A. (1985) Determining microbial parameters using nonlinear regression analysis: Advantages and
limitations in microbial ecology. Adv. Microb. Ecol., 8, 61 - 114.
Rozzi A., Massone A. and Alessandrini A. (1997) Measurement of rbCOD as biological nittate demand using a
biosensor: Preliminary results. In: Proceedings EERO/EFB International Symposium Environmental
Biotechnology ISEB3. April 21-24, Ostend, Belgium.
Siegrist H. and Tschui M. (1992) Interpretation of experimental data with regard to the activated sludge model
no. 1 and calibration of the model for municipal wastewater tteatment plants. Water Sci. Technol., 25(6),
167-183.
Siegrist H., Brunner I., Koch G., Linh Con Phan and Van Chieu Le (1999) Reduction of biomass decay rate
under anoxic and anaerobic conditions. Water Sci. Technol., 39(1), 129 - 137.
Smets B.F., Jobbagy A., Cowan R.M. and Grady C.P.L. Jr. (1996) Evaluation of respfrometric data:
Identification of features that preclude data fitting with existing kinetic expressions. Ecotoxicology and
Environmental Safety, 33, 88 - 99.
Sollfrank u. and Gujer w. (1991) Characterisation of domestic wastewater for mathematical modelling of the
activated sludge process. Water Sci. Technol., 23,1057 - 1066.
Sollfrank u., Kappeler J. and Gujer w. (1992) Temperature effects on wastewater characterization and the
release of soluble inert organic material. Water Sci. Technol., 25(6), 33-41.
Sozen A., Ubay Cokgốr E., Orhon D. and Henze M. (1998) Respirometric analysis of activated sludge
behaviour -n. Heterottophic growth under aerobic and anoxic conditions. Water Sci. Techno!, 32(2), 476
-488.
Spanjers H. (1993) Respirometry in activated sludge. Ph.D. thesis, Landbouwuniversiteit Wageningen, the
Netherlands, 199 p.
Spanjers H. and Vanrolleghem (1995) Respirometry as a tool for rapid characterisation of wastewater and
activated sludge. Water Sci. Techno!, 31(2), 105 - 114.
Spanjers H., Olsson G. and Klapwijk A. (1994) Determining influent short-term biochemical oxygen demand
and respiration rate in an aeration tank by using respirometry and estimation. Water Res., 28, 1571 - 1583.
Spanjers H., Takacs I. and Brouwer H. (1999) Direct parameter extraction from respirograms for wastewater
and biomass characterisation. Water Sci.Technol., 39(4), 137 - 145.
Spanjers H., Vanrolleghem P.A., Olsson G. and Dold p. (1998) Respirometry in control of the activated sludge
process. International Association on Water Quality, London, UK.
spérandio M. (1998) Developpment d’une procédure de compartimentation d’une eau résiduaire urbaine et
application à la modélisation dynamique de procédés boues activees. PhD thesis, Institut National des
Sciences Appliquées de Toulouse, France, 221 p.
spếrandio M., Urbain V., Audic J.M. and Paul E. (1999) Use of carbon dioxide evolution rate for determining
heterotrophic yield and characterising denitrifying biomass. Water Sci. Technol., 39(1), 139 - 146.
spérandio M. and Paul E. (2000) Estimation of wastewater biodegradable COD fractions by combining
respirometric experiments in various So/Xo ratios. Water Res., 34,1233 - 1246.
Stokes L., Takács I., Watson B. and Watts J.B. (1993) Dynamic modelling of an A.s.p. sewage works - A case
study. Water Sci. Technol., 28(11-12), 151 - 161.
STOW A (1996) Methoden voor influentkarakterisering - Inventarisatie en richtlijnen. STOWA Report 80-96.
STOWA, Utrecht, The Netherlands.
Suschka J. and Ferreira E. (1986) Activated sludge respirometric measurements. Water Res., 20,137 - 144.
Torrijos M., Cerro R.M., Capdeville B., Zeghal s., Payraudeau M. and Lesouef A. (1994) Sequencing batch
reactor: A tool for wastewater characterisation for the IAWPRC model. Water Sci. Technol., 29(7), 81 - 90.
Ubisi M.F., Jood T.W., Wentzel M.c. and Ekama G.A. (1997) Activated sludge mixed liquor heterotrophic
active biomass. Water SA, 23, 239 - 248.
Urbain V., Naidoo V., Ginestet p. and Buckley C.A. (1998) Characterisation of wastewater biodegradable
organic fraction: accuracy of the nitrate utilisation rate test. In Proceedings of the Water Environmental
Federation 71st Annual Conference and Exposition, October 3-7, Orlando, Florida (USA), 247 - 255.
Vandebroek R. (1986) Study and development of a microcomputer controlled sensor for the determination of
the biodegradability and the toxicity of wastewaters: The RODTOX. PhD. Thesis. Faculty of Agricultural
Sciences. University of Gent, Belgium, pp. 171.
Vanderhasselt A., Aspegren H., Vanrolleghem P.A. and Verstraete w. (1999) Settling characterisation using on-
line sensors at a full-scale waste water treatment plant. Water SA, 25,453 - 458.
van Haandel A.C., Ekama G.A., Marais G.V.R. (1981) Thẹ activated sludge process - part 3. Single sludge
denitrification. Water Res., 15,1135 - 1152.
van Loosdrecht M.C.M. and Henze M. (1999) Maintenance, endogenous respfration, lysis, decay and starvation.
Vanrolleghem P.A. and Coen F. (1995) optimal design of in-sensor-experiments for on-line modelling of
nitrogen removal processes. Water. Sci. Technol., 31(2), 149-160.
Vanrolleghem P.A. and Dochain D. (1998) Model Identification. In: Advanced Instrumentation, Data
Interpretation and Control of Biotechnological Processes, Kluwer Academic Publishers, Dordrecht, The
Netherlands, 251-318.
Vanrolleghem P.A. and Spanjers H. (1994) Comparison of two respirometric principles for the determination of
short-term biochemical oxygen demand. In: Proceedings 49th Purdue Industrial Waste Conference. Lewis
Publ., Chelsea, Michigan, 177-188.
Vanrolleghem P.A. and Verstraete w. (1993) Simultaneous biokinetic characterization of heterotrophic and
nitrifying populations of activated sludge with an on-line respirographic biosensor. Water Sci. Technol.,
28(11-12), 377 - 387.
Vanrolleghem P.A., Dries D. and Verstraete w. (1990) RODTOX: biosensor for rapid determination of the
biochemical oxygen demand and the on-line monitoring of the toxicity of wastewaters. In: Proceedings 5 th
European Congress on Biotechnology. Copenhagen, Denmark, July 8- 13 1990. Vol. 1,161-164.
Vanrolleghem P.A., Gemaey K., Coen F., Petersen B., De Clercq B. & Ottoy J.-P. (1998) limitations of short-
term experiments designed for identification of activated sludge biodegradation models by fast dynamic
phenomena. In: Proceedings 7th IFAC Conference on Computer Applications in Biotechnology CAB7.
Osaka, Japan, May 31 - June 4 1998.
Vanrolleghem P.A., Spanjers H., Petersen B., Ginestet p. and Takács I. (1999) Estimating (combinations of)
Activated Sludge Model No.l parameters and components by respirometry. Water Sci. Technol., 39(1),
195-215.
Vanrolleghem P.A., Van Impe J.F., Vandewalle J. and Verstraete w. (1992) Advanced monitoring and control
of the activated sludge process: On-line estimation of crucial biological variables in a structured model
with the RODTOX biosensor. In: Modeling and Control of Biotechnical Processes. Eds. Karim M.N. and
Stephanopoulos G., Pergamon Press, Oxford. 355-358.
Van Vooren L. (2000), Buffer capacity based multipurpose hard- and software sensor for envừonmental
applications. PhD thesis, BIOMATH department, Ghent University, Belgium.
Van Vooren L., Willems p., Ottoy J.P., Vansteenkiste G.c. and Verstraete w. (1995) Automatic buffer capacity
based sensor for effluent quality monitoring. In: Proceedings IAWQ Conference on Sensors in Wastewater
Technology, October 25 - 27 , Copenhagen, Denmark.
Vemimmen A.P., Henken E.R. and Lamb J.c. (1967) A short-term biochemical oxygen demand test. J. Water
Pollut. Control Fed., 39,1006 - 1020.
Volskay V.T.Jr., Grady C.P.L.Jr. and Tabak H.H. (1990) Effect of selected RCRA compounds on activated
sludge activity. Res. J. Water Pollut. Control Fed., 62, 654 - 664.
Wanner o., Kappeler J. and Gujer w. (1992) Calibration of an activated sludge model based on human expertise
and on a mathematical optimisation technique - A comparison. Water Sci. Techno!, 25(6), 141 - 148.
Weijers S.R., Kok J.J., Preisig H.A., Buunen A. and Wouda T.W.M. (1996) Parameter identifiability in the
IAWQ model no. 1 for modelling activated sludge plants for enhanced nitrogen removal. In: Proceedings
6th European Symposium on Computer Aided Process Engineering ESCAPE-6, Rhodes, May 1996. pp. 6.
Wentzel M.C., Mbewe A. and Ekama G.A. (1995) Batch test for measurement of readily biodegradable COD
and active organism concentrations in municipal waste waters. Water SA, 21,117 - 124.
Witteborg A., van der Last A., Hamming R. and Hemmers I. (1996) Respừometry for determination of the
influent Ss-concentration. Water Sci. Technol., 33(1), 311 - 323.
Xu s. and Hultman B. (1996) Experiences in wastewater characterisation and model calibration for the activated
sludge process. Water Sci. Technol. 33(12), 89 - 98.
Yuan w. and Stenstrốm M.K. (1996) The modelling of biomass decay in aerobic activated sludge systems:
death-regeneration versus endogenous respữation. In: Proceedings 69th Annual WEF Conference and
Exposition. 73 - 82.
Yuan z., Bogaert H. and Verstraete w. (1999) A titrimetric respừometer measuring the total nitrifiable nitrogen.
In: Proceedings 13th Forum on Applied Biotechnology, September 23 - 24, Gent, Belgium. Med. Fac.
Landbouww. Univ. Gent, 64(5a), 73 - 80.
Zeng A.p. and Deckwer W.D. (1995) A kinetic model for substrate and energy consumption of microbial
growth under substtate-sufficient conditions. Biotechnol. Prog., 11,71 - 79.
OPTIMIZATION AND CONTROL OF NITROGEN REMOVAL ACTIVATED
SLUDGE PROCESSES: A REVIEW OF RECENT DEVELOPMENTS
ZHIGUO YUAN, JURG KELLER AND PAUL LANT

The Advanced Wastewater Management Centre, The University of
Queensland, St Lucia, QLD 4072, Australia, Fax: +61 7 33654726; email:
zhiguo @ cheque, uq. edu. au
Abstract
The optimisation of biological nitrogen removal processes has attracted a lot of research
in the past few years. Considerable achievements in not only optimised process operation
and control but also improved process designs have resulted. In this paper, we review
these new developments in light of the progress they represent towards the solution of the
fundamental problems with biological nitrogen removal. It is emphasized that, while
being able to find the optimal or sub-optimal tradeoffs between different objectives, on-
line process control optimises a process within the constraints imposed by the process
design. The integration of innovative process design and optimised process control
represents the solution to the fundamental problems with biological nitrogen removal.
1. Introduction
We are witnessing an enormous growth in biological nitrogen removal from wastewater.

Nitrogen removal presents specific challenges beyond traditional COD (carbon) removal.
The optimisation of biological nitrogen removal processes has attracted a lot of research
in the past few years. Considerable achievements in not only optimised process operation
and control but also improved process designs have resulted. This paper aims at
reviewing these new developments in light of the progress they represent towards the
solution of the fundamental problems with biological nitrogen removal, and, on the basis
of this, identifying new directions for the future research.
Biological nitrogen removal takes place via the following two steps:
Biological nitrification, by which the ammonium nitrogen, either directly contained

in the wastewater or ammonified from the incoming organic
187
Gas Handling, 187-227. © 2003 Kluwer Academic Publishers.
Zhiguo Yuan, Jiirg Keller and Paul Lant
nitrogen by heterotrophs is oxidized to nitrate nitrogen by autotrophs (nittifiers)

under aerobic conditions; and
• Biological denitrification, by which the nitrate nitrogen is reduced to molecular
nitrogen by heterotrophs using COD as the electton donor, in the absence of
dissolved oxygen.
• Compared to the traditional COD removal activated sludge process, biological
nitrogen removal activated sludge systems have the following fundamental
differences:
• Unlike a COD removal process, which requữes only the heterotrophic bacteria
functioning as long as sufficient oxygen is supplied, the nitrogen removal process
requừes two types of bacteria: autotrophs and heterotrophs, which function under
conflicting conditions. Autotrophs function aerobically, denitrifiers, the portion of the
heterotrophs that denitrify, function anoxically (dissolved oxygen is absent but
nitrate or nitrite is present). A biological nitrogen removal plant has to be operated so
that both aerobic and anoxic conditions are present in the plant and under proper
control.
• Autotrophs grow slowly and therefore requừe a long sludge retention time. This
causes over-growth of heterotrophs and over-accumulation of inert solids, incurring
large capital cost.
• In addition to the disturbance that an ordinary COD removal plant receives, for
instance the fluctuation of influent flow rate and substrate concentrations, the
wastewater composition imposes a severe disturbance to the operation and control of
nitrogen removal plants. The removal of nitrogen depends on the availability of
COD. Unfavourable COD to nitrogen ratio in the influent often limits the nitrogen
removal efficiency of a biological nitrogen removal plant.
• An inherent problem with biological nitrogen removal is that denitrification should
naturally be preceded by nitrification, while the latter is always accompanied by
aerobic COD oxidation. A large fraction of the influent COD, which is fed or carried
over to the aerobic zone, is oxidized aerobically and is therefore not available for
denitrification. The situation becomes even worse with simultaneous biological
phosphorus removal, due to the added competition for COD by phosphorus
accumulating organisms.
The above features make the operation of a biological nitrogen removal plant difficult.
However, they have also offered greater possibilities for performance improvement by
means of optimised process design and on-line process monitoring and control, as has
been witnessed by the achievements made in the past few years.
The organization of the paper is shown in figure 1. In section 2, the single-sludge
systems, which have been overwhelmingly used for biological nitrogen removal, are
analysed, to identify theừ shortcomings and the opportunities for performance
improvement using on-line process control and improved process designs. The three most
important control problems identified, namely aeration control, external carbon dosage
control and SRT control, are then addressed in sections 3 to 5, respectively. Performance
improvement by means of improved process designs is discussed in sections 6 and 7.
Some novel process designs are presented and analysed. Conclusions are given in section
8.
188
Optimization and conttol of nitrogen removal activated sludge processes: a review of recent developments
Fig.l. Structure of the paper
2. Elementary analysis of biological nitrogen removal systems
Single-sludge systems, which have been predominantly used for biological nitrogen
removal, are analysed in this section. Theừ volume requừement, treatment capacity and
influent COD utilization efficiency for nitrate reduction are discussed. The analysis aims
at identifying opportunities for performance improvement of biological nitrogen removal
systems by means of on-line process control and improved process designs.
2.1. SYSTEM ANALYSIS
2.1.1. SRT and volume requirement

A characteristic of a single sludge system is that all solids, including all types of bacteria
as well as inert solids, are mixed together. Therefore they all have the same retention time
(RT) in the system. This identical time is termed as the sludge retention time (SRT) or
sludge age.
Nitrifiers grow slowly, and additionally, are sensitive to envừonmental changes (pH,
temperature, toxic and inhibitory compounds, etc.). Therefore, the SRT of a biological
nitrogen removal plant should be designed sufficiently long in order to secure the
nitrification process. This results in equally long retention times of heterotrophs and inert
solids, causing a large increase in the amount of MLSS (mixed liquor suspended solids)
and hence of the size of the plant. The dependency of the MLSS concentration in a
biological nitrogen removal plant on SRT is shown in figure 2, which is obtained using
the stoichiometric and kinetic parameters and the typical domestic wastewater
composition given in Henze et al. (1987). As shown, the amount of MLSS increases
almost linearly with SRT. The declining fraction of active biomass in MLSS indicates
that the increase of MLSS is mainly caused by inert solids.
189
Fig. 2. Autotrophic biomass and MLSS concentrations and the fraction of active biomass in MLSS, as a function
ofSRT of a single-sludge system
2.1.2. Anoxic fraction and volume requirement

In a single-sludge system, each type of solids goes through all of the existing conditions
in the plant (aerobic, anoxic and anaerobic in case of simultaneous phosphorus removal).
At any moment, only a fraction of nitrifiers and denitrifiers are functional. In terms of
SRT design, this implies an even longer SRT than requked by a fully aerobic nitrification
plant. Denoting the latter as 0X,N, the SRT of a nitrification-denitrification single-sludge
system (abbreviated as an N and DN plant below), denoted as 0x,NDN here, is generally
designed as (Henze et al., 1995),
A &X,N m
U
X,NDN ~ '
where a is the fraction of the anoxic and anaerobic volume. Therefore, the SRT and hence
the size of the plant is further increased by a factor of l/(l-a) due to the existence of
anoxic/anaerobic volume. ỡỵ,N = Oỵ,NDN (1-ỡặ is often termed as the aerobic sludge age
of an N and DN plant.
2.1.3. Anoxic fraction and nitrification capacity

Using equation (1) to design the SRT of an N and DN plant, it is expected that the N and
DN plant has the same nitrification capacity as an aerobic nitrification plant with an SRT
of ỡỵ,N. However, this is not true. Based on the fact that the two plants have the same
nitrification capacity only when the amount of nitrifiers present in the aerobic volume of
the N and DN plant equals the total amount of nitrifrers in the aerobic nitrification plant,
the following relationship is obtained (by making mass balances for nitrifiers in the two
plants),
190
x Ní
' “ í+abA0XMM ỉ-a+ũđ>Aỡx N where bA is the autotrophic decay rate; ỡỵ,N,eq is
the SRT of an aerobic nitrification plant that has the equivalent nitrification (2)
capacity to an N and DN plant with an SRT of Ox,NDN and an aerobic SRT of &X,N- For
simplicity of derivation and expression, it was assumed that nittifiers decay at the same
rate under aerobic and anoxic conditions. Also assumed was that the amount of
assimilated nitrogen was independent of sludge age. A few examples of the relationships
among ỡXtNDN, 0x,N and ỡx,N,eq^ calculated from equation (2), are shown in table 1 {bA =
o.ld1 was used in the calculation).
Table 1. Reduction of the nitrification capacity of a single-sludge biological nitrogen removal plant as a function of
the anoxic fraction CL ỠX.NDN is the sludge age, ỞX,N is the aerobic sludge age and &x,N,eq is the sludge age of
an aerobic nitrification plant with an equivalent nitrification capacity.
a 0.1 0.3 0.5
0X.NDN 10 20 30 10 20 30 10 20 30
(days)
&JV (days) 9 18 27 7 14 21 5 10 15
(days) 8.2 15 20.8 5.4 8.8 11 3.3 5 6
Obviously, the nitrification capacity of a single-sludge nitrification-denitrification plant is

dramatically reduced by the presence of the anoxic conditions in the system. The fact that
the equivalent sludge age is significantly smaller than the aerobic sludge age implies that
the real nitrification capacity of the plant is significantly smaller than what it is designed
for. Furthermore, for any given a, there exists a maximally achievable ỂỊ&N,eq-> which
can be obtained with equation (2) with NDN = oo,
(3)
For example, an N and DN plant with a - 0.5 can never achieve the same nitrification
capacity of a fully aerobic plant with an SRT>10 days (assuming bA = o.ld1). This
seriously limits the applicability of single-sludge systems at low temperatures.
2.1.4. Anoxic fraction and influent COD utilization efficiency for nitrate reduction
A single-sludge biological nitrogen removal plant is able to use the influent COD for
denitrification. Its influent COD utilization efficiency for nitrate reduction is analysed
below using a pre-denitrification system as an example.
Influent biodegradable COD (bCOD) consists of two parts: soluble bCOD and particulate
bCOD with fractions of p and 1-ff respectively. When contacting the sludge, the latter is
normally entrapped on sludge flocks. As the particulate bCOD has to be hydrolysed
before being degraded, the degradation of this part of bCOD proceeds slowly. It is
reasonable to assume that the particulate bCOD is equally available for both anoxic and
aerobic reactors. The fractions that are taken anoxically or aerobically
191
Zhiguo Yuan, Jũrg Keller and Paul Lant
depend on the volume fractions, provided that the electron acceptors are readily available
in the respective reactors. In contrast, soluble bCOD is usually more available for the
anoxic reactor than for the aerobic reactor in a pre-denitrification system. Nevertheless, a
significant part of the soluble bCOD is still washed to the aerobic reactor due to its
relatively large affinity constant (half-saturation coefficient). It is assumed here that a 1-7/
fraction of the soluble influent COD is leaked to the aerobic reactor.
Therefore, of the incoming bCOD, a fraction of is initially ’removed’
with nitrate as the electton acceptor and the rest (a fraction of (l-T7)yft-(l-6Ộ(7-y3)) is
initially ’removed’ with oxygen as the electron acceptor.
Part (with a fraction of i-Yff) of the initially ’removed’ bCOD is oxidized to carbon
dioxide. The rest (fraction YH) is assimilated into biomass cells or built as cell storage
products (see e.g. Majone et al., 1998), part of which is oxidized later via endogenous
respfration. YH is the short-term yield factor, which can be rather high. In IAWQ ASM1
(Henze et al., 1987), where cell assimilation is assumed, a value of 0.67 is recommended.
In IAWQ ASM3 (Gujer et al., 1998), where COD storage is used as the mechanism for
instant COD removal, a value of 0.8 is recommended. With the same reasoning as done
for particulate COD, it can be assumed that the cell COD is equally available for aerobic
and anoxic oxidation. The fraction of influent bCOD that is oxidized via endogenous
respfration is YH - YH)Obs, where YH>obs is the observed yield factor of the plant, which is,
Y — + fpbỉi@ X Y
H,obs H
b 0 +l where bH, Ox and fp are, respectively, the heterotrophic biomass
H x (4)
decay rate, SRT in the plant and the fraction of inert COD contained in biomass
cells (Henze et al., 1987).
Based on the above analysis, the fraction of influent bCOD used for nitrate reduction
(called the utilization efficiency) in a pre-denittification system is obtained as,
r = (1 - Y ) 77/?+ a(l - Al - Y„ ) - YHfibs)

H (5)
A graphical representation of equation (5) is shown in figure 3. In the calculation, 0x=i5

days, YH = 0.75, bH =Q.2,fp=Q.2 and ft=Q.3 were used.
The following observations are made:
• The influent COD utilization efficiency for nitrate reduction is generally low.
For example, when the anoxic fraction is 0.3, the efficiency is about 20%.
• The efficiency increases linearly with the anoxic fraction a.
• Factor T| does not have much influence on the COD utilization efficiency.
192
Fig.3. Influent bCOD utilization efficiency for nitrate reduction as a function of anoxic fraction (a) and the fraction
of the soluble influent bCOD consumed in the anoxic zone (rỤ
2.1.5. Alkalinity and pH

Nitrification and denitrification have an opposite pH effect. The pH in a single-sludge
biological nitrogen removal system is thus self-regulating. The alkalinity produced by
denitrification also partially compensates its consumption by nitrification. This is
considered a nice feature of a single-sludge biological nitrogen removal system.
2.2. OPTIMIZATION OPPORTUNITIES
2.2.1. Optimisation by on-line process control

The anoxic fraction a is obviously an important operating parameter for a biological
nitrogen removal plant. A larger a increases the availability of COD for nitrate reduction
(Fig. 3) and hence improves nitrate removal, while a smaller value increases the
nitrification capacity (Table 1), improving ammonia removal. Therefore, manipulating
parameter a on-line is important for achieving the highest degree of total nitrogen
removal.
SRT is another important operating parameter of a biological nitrogen removal plant.
Minimizing SRT on-line results in a significant reduction of the amount of suspended
solids in the system, allowing the plant to take higher loads. The SRT control system,
when integrated into the plant design, allows a less conservative design of the plant and
therefore reduces the capital cost of a biological nitrogen removal plant.
Nitrate recừculation flow in a pre-denitrification biological nitrogen removal plant is
designed to recừculate nitrate from the aerobic zone to the anoxic zone. However, this
parameter on its own is not an effective on-line confrol handle for nitrate removal as it
influences the ’COD utilization for nitrate reduction’ via its limited impact on parameter
7/, which has been shown above to be not very influential to the COD utilization
efficiency. As an alternative, external carbon addition has been developed for a number of
years as an effective control handle for the denitrification process in a biological nitrogen
removal system. The availability of external carbon sources has also improved the control
authority of nitrate recừculation because it guarantees that all the recfrculated nitrate is
removed in the anoxic zone, unlike in the case of using influent COD as the sole carbon
193
source for denitrification.

As will be discussed in sections 3 to 5, a lot of research has been devoted in the past
few years to the above control problems. Considerable achievement has been obtained.
2.2.2. Optimisation by improved process design

While on-line process control is able to find the optimal or sub-optimal tradeoffs among
different objectives, it plays a limited role in solving the fundamental problems associated
with a single-sludge biological nitrogen removal system, namely its large volume
demand, low nitrification/denitrification capacity and low utilization efficiency of
influent COD for nitrate reduction.
The analysis given in section 2.1 indicates that a multi-sludge system may be superior
to a single-sludge one, as the separation of nitrifiers and denitrifiers allows different
conditions being provided to each type of bacteria.
Indeed, some multi-sludge systems have already been in use for nitrogen removal for
quite a long time (Henze et al., 1995). The most popular configuration is to have a
separate denitrification reactor treating the effluent from the aerobic nitrifying plant.
While solving the problems caused by the co-existence of anoxic and aerobic conditions
in a single reactor, the plant does not use influent COD for denitrification. The addition of
external carbon to the denitrification reactor increases the operational cost dramatically.
In addition, just like an ordinary single-sludge biological nitrogen removal plant,
autotrophs, heterotrophs and inert solids are mixed together in the nitrifying plant. A large
volume is thus still demanded. Low pH and alkalinity in the nitrifying plant may
represent another problem of the design.
There also exist plants where COD removal and nitrification are separated, for
instance an activated sludge system with an SRT that does not allow nitrification to take
place, followed by a nitrification reactor (usually a biofilter). While this type of design
greatly reduces the size of the plant and hence the capital cost, nitrate removal has again
to be done with external carbon source (Marsman et al., 1997). Although the nitrate-rich
effluent from the nitrification reactor can theoretically be recycled to the COD removal
reactor for denitrification (Balmer et al., 1998), the settlers, which are in the recừculation
loop, have to be expanded in order to accommodate the largely increased hydraulic load,
offsetting the savings gained by using smaller reactors.
Some novel multi-sludge systems have been developed in recent years. In these
systems, autotrophs and heterotrophs grow at different locations, making it possible to
provide each type of bacteria the requfred working conditions. Superior to the
conventional multi-sludge systems as described above, they are able to make efficient use
of influent COD for denitrification. In addition, the pH and alkalinity in these systems are
self-regulating just like in a single-sludge system.
Another novel process design consists of altering the retention times of different
particulate components in sludge, by means of providing nitrifiers from a side-stream
unit, such that nitrifiers have a significantly longer retention time than inert solids. The
system may therefore contain the same amount of nitrifiers as a conventional system, but
significantly less inert solids. The volume requừement is thus significantly reduced.
As will be analysed in sections 6 and 7, these systems have great potential to solve
some of the fundamental problems with single-sludge systems.
194
2.3. CONCLUSIONS
Mass balance analysis has revealed some inherent constraints of single-sludge biological
nitrogen removal systems. These include its large volume demand and hence large capital
cost, limited influent COD utilization efficiency for nitrate removal and low nitrification
capacity. Aeration, SRT and external carbon source addition have been identified as
effective control handles for the on-line optimisation of these systems. It is pointed out
that on-line process control optimises the processes within the constraints imposed by
process designs. The elimination of these consfraints by applying novel process designs
allows further optimisation of biological nitrogen removal processes.
3. Aeration Control
Optimising the anoxic fraction a on-line by means of aeration control has been one of the
major research areas. Different from the conventional work on aeration control, which
was mainly focussed on designing appropriate conttol loops to control the DO at the pre-
selected set-points (see e.g. Olsson, 1976; Ko et al., 1982; Olsson et al., 1985; Holmberg
et al., 1989; Marsili-Libelli, 1989), the recent work has been devoted to the on-line
determination of the optimal aeration phase length/aerobic volume and the optimal DO
set-points for the local DO control loops (see Olsson and Newell, 1999).
A wide variety of on-line measured signals have been used in designing the control
systems. Substantially different control strategies have thus resulted. A detailed
discussion of these strategies is given in this section.
3.1. AERATION PHASE LENGTH CONTROL BASED ON ORP AND PH

MEASUREMENT
Aeration control in Sequencing Batch Reactors (SBR) and in intermittently aerated

continuous systems using Oxidation Reduction Potential (ORP) and/or pH as the
measured signals has been studied by many researchers in the past decade (see e.g.
Charpentier et al., 1987,1989; Wareham et al., 1993, 1994; Lo et al., 1994; Demuynck et
al., 1994; Sasaki et al., 1996; Bertanza, 1997; Plisson-Saune et al., 1996; Wouters-
Wasiak et al., 1994; Zipper et al., 1998; Al-Ghusain et al., 1994; 1995; Hao and Huang,
1996; Hamamoto et al., 1997; Yu et al., 1997, 1998). The control systems designed are
inferential ones, due to the fact that ORP and pH are indirect measures of the nitrification
and denitrification processes.
Figure 4 depicts typical ORP and pH profiles in an alternating aerobic-anoxic nitrogen
removal bioreactor with excess aerobic and anoxic periods. The ammonia nitrogen,
nitrate nitrogen and DO profiles are also shown in the figure.
195
time (hour)
Fỉg.4. ORP, NH3-N, NO3-N, DO and pH profiles in a nitrogen removal bioreactor (illustrative, adapted
from Wareham et al., 1994; Hao and Huang, 1996)
As illustrated, the ORP in the reactor rises when aeration is switched on and drops when
it is switched off. Two bending points may occur on the ORP curve: the ’ammonia break
point’, caused by a sharp DO rise due to the depletion of ammonia nitrogen in the mixed
liquor; and the 'nitrate break point’ or the nitrate knee’, caused by the depletion of nitrate
in the reactor. Therefore, the two bending points correspond to the ends of nitrification
and denitrification, respectively.
The pH in the bioreactor also varies periodically. Unlike ORP, whose variation is
caused by the presence/absence of oxygen and nitrate, the variation of pH is caused by
the biochemical reactions of nitrification and denitrification. When aeration is switched
on, nitrification takes place, resulting in a decrease in the mixed liquor pH until the end of
nitrification. Aeration then brings pH up to a higher value, resulting in a bending point,
called the ’ammonia valley’ (see figure 4) on the pH curve at the end of nitrification.
When aeration is switched off, denitrification takes place, resulting in an increase in pH
until the end of denitrification. This is then followed by a decrease of pH caused by the
anaerobic process. Another bending point, called the ’nitrate apex’ is thus formed on the
pH curve at the end point of denitrification. A more detailed analysis of the pH curve can
be found in Hao and Huang (1996).
For both ORP- and pH-based aeration control systems, two types of control strategies
have been studied: namely the absolute value based control strategy and the bending point
based conffol strategy.
3.1.1. Aeration control based on absolute values of ORP and pH

The absolute value based control strategy was initiated by the fact that ORP and pH vary
in a certain range during an aerobic-anoxic cycle. The basic idea is to pre-select two
limits and to switch on/off the aeration when the limits are exceeded.
196
For the ORP-based controller, aeration is switched off when the measured ORP
exceeds the upper limit, and switched on when the lower limit is exceeded (see e.g.
Charpentier et al., 1987, 1989; Wouters-Wasiak et al., 1994; Zipper et al., 1998). Ideally,
the two limits should be chosen as the two ORP values of the two bending points, which
enables complete nitrification and denitrification with minimum cycle time. A too
low/high upper/lower limit results in incomplete nitrification/denitrification, while a too
high/low upper/lower limit results in unnecessarily long phase lengths, reducing the
treatment capacity of the plant. Charpentier et al. (1987, 1989) reported full-scale
applications of this strategy. Satisfactory nitrogen removal was achieved.
By merging the two limits into one, Lo et al. (1994) and Bertanza (1997) studied an
ORP set-point control strategy. Aeration is controlled such that the ORP is maintained at
the set-point. Unlike the two limits strategy, which results in aerobic-anoxic cycle with
typically a period of a few hours, the set-point strategy results in simultaneous
nitrification and denitrification (Bertanza, 1997). They reported a nitrogen removal rate of
81-89% on a full-scale plant.
pH-based controllers work in a similar way. Al-Ghusain et al. (1994, 1995) applied
such a controller to an aerobic-anoxic sludge digestion process, which achieved a
nitrogen removal rate of 50%.
A major problem with the absolute value (either ORP or pH) based control strategy is
the determination of the limits or set-point. It is commonly known that ORP as measured
is not a true thermodynamic parameter. It is merely an indication of the overall oxidative-
reductive state of the system, the absolute ORP value per se does not impart any process
significance (Al-Ghusain et al., 1994). In addition, the measured ORP values depend on
the initial treatment of platinum probes, and thus depend on the surface characteristics of
the metal. Therefore, the limits and set-point requừed by the controller are site and probe
specific and must be determined individually. Another problem with ORP measurement
is the drifting of the signal. Significant drift of the signal may occur within a relatively
short period. Hao and Huang (1996) reported a drift in the ORP value from about 120mV
to -50mv in 40 hours. This makes it necessary to calibrate the limits on-line. However, no
work on this aspect has been reported. An idea may be to include calibration cycles into
the operation, during which excess aerobic and anoxic periods are applied. The ’ammonia
break point’ and the ’nitrate knee’ detected suggest the new limits.
The pH value does not suffer the same problem. A pH value has the same meaning
regardless the system measured or the probe used. However, the choice of appropriate
limits is not at all straightforward, given the complicated variation of pH during an
aerobic-anoxic cycle (Fig. 4). pH values of 6 and 8 were chosen in Al-Ghusain et al.
(1994, 1995) as the limits for the aerobic-anoxic sludge digestion processes studied
therein. Such a large range does not seem to be applicable to an intermittently aerated
wastewater tteatment process. The pH in such a process likely varies in a much narrower
range during one operation cycle, which takes typically a few hours (see Fig. 4). Indeed,
guidelines for the determination of the two limits are still to be established.
3.1.2. Aeration control based on bending points of ORP and pH
The bending point based control strategies are designed based on the detection of the
bending points on the ORP (the ’ammonia break point’ and the 'nitrate knee’) or the pH
(the ’ammonia valley’ and the ’nifrate apex’) curves. Aeration is switched off when the
’ammonia break point’ or the ’ammonia valley’ is detected, and is switched on when the
197
’nitrate knee’ or the 'nitrate apex’ is detected. In this way, the lengths of the aerobic and
anoxic phases are controlled to be just sufficient for complete nitrification and
denitrification, respectively. Plisson-Saune et al. (1996) used the two ORP bending points
to control a lab-scale plant treating domestic wastewater. 89% of the influent nitrogen
was removed. Al-Ghusain et al. (1994; 1995) used the two pH bending points to conttol a
lab-scale sludge digestion process. Nearly complete nitrogen removal was achieved.
The ’ammonia break point’ on the ORP curve, which appears only when the DO is
subject to a sharp rise from a low level to a significantly higher one at the end of
nitrification (Wouters-Wasiak et al., 1994), is usually difficult to identify. Wareham et al.
(1993, 1994) used an aeration strategy based on the ’nitrate knee’ alone: aeration is
switched on when the ’nitrate knee’ is detected and the aerobic phase is set to be equal to
the previous anoxic phase. For an alternating aerobic-anoxic-anaerobic biological
nitrogen and phosphorus reactor, Sasaki et al. (1996) proposed to contiol the aeration
phase such that the ’nitrate knee’ appears at the specified time. This was achieved by
using a feedback controller: the present aerobic period = the previous aerobic period +
K\set_knee_tỉme - actual_knee_timé), where K, a positive constant, was the feedback
gain. A two-month pilot plant study achieved 93% of total N removal and 90% of total p
removal.
The concerns about the reliability of the bending-point detection have initiated the
idea of combining ORP and pH signals for the aeration control (Yu et al., 1997, 1998;
Hamamoto et al., 1997). In Yu et al. (1997, 1998), the derivatives of ORP and pH with
respect to time were calculated simultaneously. The combination of the two derivative
signals, together with an ANN (artificial neural network) predictor, which predicts the
time and magnitude of the bending points, significantly improved the reliability of the
bending point detection and hence the performance of the contioller. Both the treatment
capacity and the nitrogen removal rate of the controlled plant were significantly improved
compared to a contiol plant with fixed phase lengths. Hamamoto et al. (1997) adopted a
different approach. The on-line measured ORP and pH signals, together with DO and
water level data, were fed to a fuzzy contioller, which inferred the aeration lengths based
on linguistic rules.
With a bending point based control strategy, nitrification and denitrification come to theừ
ends in the aerobic and anoxic phase, respectively. This is not necessarily an optimal
strategy. For an intermittently aerated continuous system, high effluent ammonia and
nitrate peaks may appear alternatively, resulting in high effluent nitrogen concentration,
when the plant is over-loaded with nitrogen. It would be more desirable to have the
phases switched before ammonia/nitrate reaches a too low concenttation that limits the
nitrification/denitrification rate. For an SBR, the cycle time is often not at the disposal of
the controller, as the plant has to take what it receives. Given the cycle time, a complete
nitrification phase may lead to a too short denitrification phase and vice versa, resulting in
effluent nitrate or ammonia peaks. Again, it is more desfrable to have the phases switched
before ammonia/nitrate reaches a too low concentration so that a better compromise
between nitrification and denitrification is found.
3.2. AERATION CONTROL BASED ON RESPIROMETRY
Notwithstanding that respfrometers and respfrometry-based control of activated sludge

have attracted a large amount of research in the past few decades with considerable
198
progress (Spanjers et al., 1998; Vanrolleghem et al., 1998), few respirometry-based

aeration control strategies have been developed in recent years for nitrogen removal
processes.
An OUR bending point based aeration control strategy has been suggested by several
researchers (Olsson and Andrews, 1978; Demuynck et al., 1994; Johansen et al., 1997;
Klapwijk et al., 1998). The idea is to switch off the aeration when the bending point
corresponding to the end of nitrification is detected on the OUR profile.
An obvious problem of this strategy is the reliability of the bending point detection. The
nitrification bending point may not be readily detectable on the OUR profile. Another
problem is that the OUR profile can hardly tell when to terminate the anoxic phase. For
an intermittently aerated continuous plant that receives influent only during anoxic phase,
Klapwijk et al. (1998) suggested that the anoxic phase be terminated when a sharp rise of
OUR is measured using an on-line continuous-flow respừometer. The underlying
assumption is that the consumption rate of the readily biodegradable COD by
denitrification is higher than its feeding rate as long as nitrate is present. In this case, a
sharp rise of OUR during the anoxic phase indicates an accumulation of readily
biodegradable COD caused by the cease of denitrification due to depletion of nitrate. The
applicability of the strategy is restricted by the feeding regime requừed.
Brouwer et al. (1998a) proposed a simple but more promising feedforward control
strategy to control the aerobic volume of a continuous activated sludge plant. A batch
experiment based respừometer was used to characterize the wastewater composition and
the sludge kinetics. The respfrometer allowed the estimation of, among other variables
and parameters, the concentration of the influent nitrogen that was to be nitrified in the
treatment plant, and the maximum ammonia oxidation rate of the sludge. The aerobic
volume in the plant requfred to completely nitrify the incoming nitrogen was then
calculated as: Aerobic volume = influent flow rate* concentration of the influent nitrogen
to be nitrified/maximum ammonia oxidation rate.
Compared to nutrient sensors (which will be discussed below), respfrometers, when used
to measure ammonia concentration, have the disadvantage of providing discrete and
considerably delayed signals. The measurement delay imposed is proportional to the
concentration measured, and typically ranges from one to a few hours. In addition, a
respfrometer is not able to measure nitrate concentration. These probably explain the
limited achievement made so far using respừometer in controlling the aeration to a
nitrogen removal plant. However, respirometers have the advantage of being able to
providing sludge kinetic parameters (see e.g. Vanrolleghem et al. 1995; Brouwer et al.,
1998b), which is obviously valuable for the control system design. More research in this
direction is demanded.
199
3.3. AERATION CONTROL BASED ON AMMONIA AND NITRATE

MEASUREMENT
With the continuous improvement of reliability, accuracy and ease of maintenance of

ammonia and nitrate sensors (Thomsen and Kenneth, 1996; Londong and Wachtl, 1996),
aeration control using nutrient sensors has been studied by many researchers and full
scale applications have been reported.
R1
—
)----------- ■-
R2
ĨT'""
Fig. 5. Flow path, aeration status and typical variations of ammonia and nitrate in a six- phase BioDenipho®
process cycle (after Bundgaard et al., 1989). N: nitrification; DN: denitrification; solids curve: ammonia nitrogen
concentration; dashed curve: nitrate nitrogen concentration; circled points: switching points
3.3.1. Objective functions

Two types of objective functions have been used in the control system design. One
(called Type I Objective in the sequel) is to control the measured ammonia nitrogen
concentration (sensor normally located at the outlet of the bioreactor) at a pre-selected
set-point or within two predefined boundaries (Balslev et al., 1996; Hoen et al., 1996;
Husmann et al., 1998). The underlying idea is to control the effluent ammonia at a
satisfactory level and at the same time to minimize aeration thus reducing the effluent
nitrate concentration by increased denitrification on the one hand, and reducing the
aeration cost on the other hand.
Another (called Type II Objective in the sequel) is to control the aeration such that the
effluent total nitrogen is minimized (Thornberg et al., 1993; Sorensen et al., 1994;
Nielsen and Onnerth, 1995; Potter et al., 1996; Sorensen, 1996; Onnerth et al., 1996;
Leeuw and van ’t Oever, 1996; Lukasse et al., 1998; Isaacs and Thornberg, 1998;
200
Steffens and Lant, 1999). With such an objective, the effluent ammonia and nitrate
concentrations are comprised. Either nitrification or denitrification can be favoured in
defining the objective function. However, nitrification should usually be favoured, given
the fact that the amount of nitrifiers present in the system is determined by the amount of
ammonia oxidized, while the amount of heterotrophs is independent of the amount of
nitrate removed. An elevated effluent ammonia content reduces the amount of nitrifiers,
and hence the nitrification capacity of the plant. Compromising ammonia removal for
nitrate removal may eventually be detrimental to nitrate removal, as the system may
requừe a smaller anoxic fraction. Another reason for favouring ammonia removal is that
ammonia is toxic to water lives.
3.3.2. Aeration control of alternating systems

Aeration control to the alternating BioDenipho® (or BioDenitro™) plants has been well
studied using conventional feedback control systems based on ammonia and nittate
measurements. The BioDenipho® is a rather complicated activated sludge nittogen and
phosphorus removal process, with nitrification and denitrification accomplished in a
semi-batch manner by periodically changing the path of flow through two parallel
aeration tanks that are periodically aerated (Bundgaard et al., 1989). The aeration control
of this system is addressed in some details here because of the pioneering role it played in
the application of nutrient sensors to the control of wastewater treatment systems, and its
general applicability to other types of intermittently aerated systems. In addition, this
section is also intended to clarify the relationships among the various aeration control
systems reported in literature for this type of systems.
A typical BioDenipho® cycle comprises six phases, which are shown in figure 5
(Bundgaard et al., 1989). The variation of nitrate and ammonia nitrogen concentrations in
the two aeration tanks is also indicated in the figure. The aeration control laws typically
consist of the following ’switch-point’ rules (Potter et al., 1996):
• Rule 1: Transition from Phase A to Phase B takes place when the ammonia
concentration in reactor R1 reaches NHmax.
• Rule 2: Transition from Phase B to Phase c takes place when the nitrate
concentration in reactor R1 reaches NOmin.
• Rule 3: Transition from Phase c to Phase D takes place when the ammonia
concenfration in reactor R2 reaches NHmin.
The transitions from D to E, E to F and F to A mirror the above rules. Obviously, Type II
Objective as discussed in the previous section has been adopted in the design.
To improve the robustness to external disturbance (transient loading, temperature and pH
variations, etc.), the concept of ’criteria function’ has been proposed (Thornberg et al.,
1993). The idea is to determine the switching points NHmỉn and NOmỉn on-line, so that they
are adapted according to the reactor status. An example of the criteria functions is shown
in equation (6) (Potter et al., 1996),
NOâNHfN+P
NHmịa=1NOi-N + Ổ
201
where a, p, /and Ỗ are predefined parameters, and NH4-N and NO3-N are the measured
ammonia and nitrate nitrogen concentrations, respectively. The importance of the criteria
function can be illustrated with the following example. When reactor R1 receives a high
nitrogen load in Phase B, a high NOmin results so that Phase B is terminated earlier,
leaving more time for R1 to nitrify the accumulated ammonia.
Some phases in a BioDenipho® may be dropped out, resulting in slightly different
phase length control problems. Isaacs and Thornberg (1998) studied the phase length
control of a four-phase BioDenipho® process, with Phase c and Phase F (see figure 5)
left out. Due to the absence of Phase c and Phase F, Rule 2 and Rule 3, as discussed
earlier, apply simultaneously, resulting in a conflicting situation. This was resolved by
merging the two rules: transition from Phase B to Phase D takes place when the
conditions in both Rule 2 and Rule 3 are satisfied. This implies that the reactor, which
first completes its task, is made to wait for the other reactor before the roles of the two
reactors are switched. Thornberg et al. (1993) studied another type of four-phase process,
where Phase A and Phase D were left out. In this process, Rule 1 was no longer
applicable. The phase lengths were controlled by Rule 2 and Rule 3. Several full-scale
applications of the control system were reported in Thomberg et al. (1993). The total
effluent nitrogen concentrations were significantly reduced, accompanied by considerable
savings of energy consumption.
In parallel to confrolling the aerobic-anoxic phase lengths, the DO set-points during
the aerobic phases were also studied. A typical control law is to control DO at an on line
determined set-point, with lower and upper boundaries (Thornberg et al., 1993; Sorensen
et al., 1994; Nielsen and Onnerth, 1995; Sorensen, 1996). The DO set-point is calculated
proportionally to the measured ammonia concentration.
3.3.3. Aeration control of pre-denitrification systems

The above control systems have also been successfully applied to pre-denitrification
systems. By applying intermittent aeration to the aerobic zone and controlling the aerobic
and anoxic phase lengths and the DO set-points using the control system discussed above,
30% energy consumption and 100% external carbon source were saved on a full-scale
plant with 17,000PE, accompanied by slightly reduced effluent total nitrogen content
(Nielsen and Onnerth, 1995; Onnerth et al., 1996). Similar aeration control strategies
were also studied by Balslev et al. (1996) on a pilot-scale predenitrification plant.
Husmann et al. (1998) studied the aeration control of a step-feed full-scale biological
nitrogen removal plant (60,000 PE) using a simple feedback control system. The plant
consisted of two anoxic and aerobic zones, with a configuration of anoxic-aerobic-
anoxic-aerobic. The influent waste water was fed to the two anoxic zones with a ratio that
was determined on-line. Type I Objective (see the previous section) was employed in the
control system design. A very simple control law was used to control the aeration to the
aerobic zones:
• the DO set-points in the aerobic zones are reduced in steps by 0.5 mg/L starting from
the first aerobic zone until a concentration of 0.5 mg/L is reached in both aerobic
reactors, when the measured ammonia concentration in the second aerobic zone is
lower than the lower limit of the targeted ammonia range (0.8 mg N/L in the reported
practice).
• the DO set-points in the aerobic zones are increased in steps by 0.5 mg/L starting
202
from the second aerobic zone until a concentration of 2.0 mg/L is reached in both
aerobic reactors, when the measured ammonia concentration in the second aerobic
zone is higher than the upper limit of the targeted ammonia range (1.3 mg N/L in the
reported practice).
• the second anoxic zone is aerated and all wastewater is dừected to the first anoxic
zone when the measured ammonia concentration exceeds an extreme value (2.0 mg
N/L in the reported practice).
The control action output intervals are obviously important to the stability of the control
system. Sufficient time must be given before a new control action is taken.
Compared to the reference lane, where DO in the two aerobic zones were constantly
controlled at 2.5 mg/L, and two-thfrd of the wastewater were fed to the first anoxic zone,
the effluent total nitrogen was reduced by 50% in summer and 33% in winter. The
aeration cost was also saved by 16%.
3.3.4. Model-based control

Some researchers have also studied model-based aeration control in the past few years
(see e.g. Hoen et al., 1996; Lukasse et al., 1998; Steffens and Lant, 1999). The designs
are typically based on a simplified model consisting of ammonia and nitrate dynamics.
Model predictive control was employed by Hoen et al. (1996) to control the aerobic
volume of a single-sludge post-denitrification plant. The aerobic volume was controlled
such that the effluent ammonia concentration was maintained within the target range
{Type Ĩ Objective). Instead of directly using the measured ammonia concentration in the
control action calculation, as used by Husmann et al. (1998) (see the previous section),
the predicted effluent ammonia concentration was used. This approach obviously had the
advantage of more prompt control, provided that the predictions were sufficiently
accurate. The model used in the prediction was a semi-mechanistic, non-linear one
consisting of ammonia and nitrate dynamics, which was obtained by simplifying a
mechanistic model for nitrification and denitrification. The model involved three time-
varying rate coefficients that were identified on-line with an observer.
The approach was only illustrated with two simulation case studies. Its applicability to
real plants remains unclear. The critical part of the approach is the estimation of the three
rate coefficients, which are lumped factors and may therefore vary significantly. The
estimation results in the simulation studies were not reported.
Lukasse et al. (1998) applied receding horizon optimal control design to the aeration
control of a completely mixed system. By choosing the intermittent aeration regime, the
control problem was simplified as to minimize, within the predicting horizon, the
deviations of the predicted effluent ammonia and nitrate nitrogen concentrations from
thefr targeted values {Type IỈ Objective).
The model used in the prediction was a linear, semi-mechanistic one consisting of
ammonia and nitrate dynamics, obtained by neglecting other biological processes than
nitrification and denitrification, and assuming zero order nitrification and denitrification
rates.
The control system was demonstrated on a pilot plant study with the predicting
horizon being one measurement interval (20 minutes) (Lukasse et a/., 1998). In a related
study, the control system was compared to a few other controllers using simulation
203
(Lukasse, 1999). This controller failed to outperform one with a conventional feedback
control based on ammonia measurement alone.
Steffens and Lant (1999) evaluated, by means of a simulation study, a few model-
based control designs by applying them to the control of the DO set-points in the two
aerobic reactors of a pre-denittification biological nitrogen removal plant. The designs
evaluated included linear quadratic control (LQC), dynamic matrix control (DMC) and
non-linear optimal control (NOC). They used three criteria to compare the controllers:
operating costs, discharge costs and a process performance indicator. The last indicated
the extent of capacity creep that the process could withstand, which dừectly relates to
savings in deferred capital expenditure. They concluded that for all disturbance scenarios
examined, the model-based controllers outperformed the base case and PI controllers. The
major factor being that all the investigated model based controllers provided scope for
increased throughput, whereas the base case controller failed to meet the license specs
and the PI controller was operating at the consttaint.
3.4. CONCLUSIONS
Aeration control has been proven to be an effective means for optimising the nitrogen
removal efficiency in a biological nitrogen removal system.
Compared to other types of sensors, nutrient sensors support the dừect control of the
ammonia and nitrate nitrogen concentrations in the system. The control systems designed
based on these sensors therefore exhibit more flexibility in making comprise between
nitrification and denitrification. With the continued improvement of the reliability and
ease of maintenance of nutrient sensors, it can be expected that the nutrient sensor based
aeration control systems will be more widely used, especially in large biological nitrogen
removal plants.
While several researchers have studied model-based aeration control systems, limited
achievement has been made. No applications have been reported so far. The bottleneck is
to obtain simple process models that are identifiable, applicable to control system design
and yet characterizing the processes reasonably well.
Aeration control does not change the fact that only a fraction of nittifiers and denittifiers
are functional at any given moment in a single-sludge biological nitrogen removal plant.
4. External COD dosage optimisation and control
To reject the disturbance of low influent COD to N ratio, external COD addition has been
developed as an effective means of controlling denitrification processes (see e.g. Isaacs et
al. 1994). Addition of external COD to the anoxic zone/phase significantly increases the
denitrification rate, and therefore enhances nitrate removal. This section aims to review
the research on external carbon sources and the control of theừ addition to a biological
nitrogen removal plant.
4.1. EXTERNAL CARBON SOURCES
Methanol, ethanol, and hydrolysate from fermentation of primary sludge have been the
main external carbon sources used for denitrification. They are either added to the anoxic
zone of a single-sludge system (pre- or post-denitrification) as supplement to the influent
204
COD, or to the denitrification tank of a two-sludge post-denitrification system.

Investigating the effectiveness of different carbon sources is obviously important for the
choice of the most appropriate carbon source.
Several researchers have made comparative studies on the effects of methanol and
ethanol as external carbon sources for denitrification (Christensson et al., 1994; Hallin et
al., 1996; Nyberg et al., 1996; Hallin and Pell, 1998). The properties compared included
the specific denittification rates they support, the time the sludge needs to adapt to the
carbon source, the response time of the effluent nitrate to the addition of the carbon
source, and the ability of the adapted sludge to denitrify with other types of carbon
sources. The results are summarized in table 2.
The comparison shown in table 2 suggests that ethanol should be a better external
carbon source than methanol, especially when added to the denitrification zone of a
single-sludge biological nitrogen removal system. In the latter case, the influent and
external carbon is used concomitantly for denitrification and the external one is provided
only when the influent COD to N ratio is low.
Hydrolysate from fermentation of primary organic solids has also been used as
external carbon source for denitrification. Compared to using methanol and ethanol,
where ’clean’ carbon is used to remove waste (nitrate) and, as a side effect, to generate
new waste (sludge production), using hydrolysate is more envừonmentally friendly. It
may also be more cost-effective if the operational cost for the fermentation can be
maintained low.
The hydrolysate is most often generated in a separate fermentor where desirable
conditions are provided (e.g. Aesoy and Odegaard, 1994; Brinch et al., 1994; Charlton,
1994). A yield factor (unit mass soluble organics generated per unit mass volatile
suspended solids added) of 0.06-0.25 has been reported in literature (Aesoy and
Odegaard, 1994; Brinch et al., 1994; Skalsky and Daigger, 1995; Rabinowitz and
Barnard, 1997). A large fraction of the soluble COD generated is volatile fatty acid, of
which acetic acid forms a big part, making the hydrolysate a deskable carbon source for
denitrification or for phosphorus removal. The denitrification rate using hydrolysate has
been found to be similar to that of using acetate (Kristensen and Jorgensen, 1990; Isaacs
and Henze, 1995) or ethanol (Aesoy et al., 1998). Full-scale applications of using
hydrolysate for improved denitrification have been reported (see e.g. Brinch et al., 1994;
Chaiton, 1994; Rabinowitz and Barnard, 1997).
Hydrolysate may also be generated in the primary clarifier with fermentation under
the sludge blanket that is obtained by extending the sludge retention time in the clarifier
(Barnard, 1984). Christensson et al. (1998) reported an increase of 10 mg/L of readily
biodegradable COD in the primary effluent, however at the price of a significantly
increased suspended solids concentration (40%) in the primary effluent. The latter was
caused by the high sludge blanket maintained in the clarifier (estimated to be half the
height of the clarifier). Obviously, this approach is useful only when a small amount of
extra readily biodegradable COD is needed.
Table 2: Comparison of methanol and ethanol as carbon sources for denitrification
Specific Adaptation time Response time Denitrification capacity
denitrification rate with other carbon
Methanol relatively low one sludge age days lower capacity with
carbon other than
ethanol*
205
Ethanol relatively high (2- one sludge age hours higher capacity with a
3 times higher large variety of carbon
than methanol) sources
Comments denitrification adaptation time is methanol is not methanol adapted sludge
with ethanol caused by change desừable for has a lower capacity of
requừes a much of microbiology in intermittent using influent COD for
smaller volume sludge dosage denitrification
References Christensson et al. Hallin et al. Christensson et al. Hahin and Peh (1998)
(1994); Nyberg et (1996); Nyberg et (1994); Nyberg et
al. (1996) al. (1996) al. (1996); Hahin
and Pell (1998)
*compared with the reference sludge to which no external carbon was added
Primary sludge may also be used as external carbon source (see e.g. Kurata et al., 1996).
Due to its high inert solid content, adding primary sludge inevitably results in a
significant increase of MLSS in the system, which may not be allowed in all cases.
4.2. UTILIZATION EFFICIENCY OF EXTERNAL COD
The supplementary external carbon may be added directly to the anoxic zone of a single-
sludge biological nitrogen removal system (with either pre- or postdenitrification
configuration), or to a separate post-denitrification reactor designed to further denitrify
the effluent from the mainstream biological nitrogen removal system. In the latter case, a
two-sludge system results. As will be analysed below, the utilization efficiency of the
dosed COD for nitrate reduction is significantly different in the two cases.
Due to the small affinity constant (half-saturation coefficient) of the external carbon
sources, it is reasonable to assume that the carbon is 'removed' instantly after the addition.
This is especially true when the dosage rate is properly controlled (see the next section).
For a pre-denitrification system, the instant removal implies negligible leakage of
external carbon in soluble form.
In analogy to the reasoning made in section 2, the fractions of the totally added
external COD that is used for nitrate reduction in the single-sludge and two-sludge
systems, denoted as Ỵone and Ytwo, respectively, are calculated as follows,
rone = (7}
V = 1 —K, , t two 1
ti,obs
where a is the anoxic fraction, YH is the short-term yield and yH)Ohsis the observed yield as
defined in equation (4). Equation (7) indicates that the two-sludge post-denitrification
system has a significantly higher COD utilization efficiency. With 6k = 15 days, Ftf =
0.75, bH =0.2 d'1 andyp=0.2, ỵone (for <2=0.1, 0.3 and 0.5) and Ytwo are calculated and
shown in table 3. ratios are also shown in the table.
Table 3: COD utilization efficiencies for nitrate reduction in a two-sludge post denitrification system, and a single-
sludge pre-denitrification system with different anoxic fractions (0.1, 0.3 and 0.5)
a 0.1 0.3 0.5
Tone 0.25 0.39 0.48
Ỵtwo 0.7
ỴonelỴtwo
0.36 1 1 0.55 1 0.68
206
The above analysis indicates that significantly more external carbon is needed for the
removal of the same amount of nittate in the single-sludge system than in the two- sludge
system. Note that the extta amount of carbon demanded by a single-sludge system is
oxidized aerobically, incurring added aeration cost.
Adding the supplementary COD to the mainstream system has the advantage of not
requừing a separate reactor. However, when the supplementary COD is requfred in a
large amount, the great savings of carbon source and aeration cost may justify the
construction of a separate denittification tank. Denitrification using external carbon in a
separate tank also avoids influencing the microbiology in the mainstream system. The
latter may reduce the capability of the heterotrophs to denitrify using influent COD (see
the previous section).
When the external carbon is added to the main-stteam reactor, an option is to add the
carbon into a second anoxic zone that is near the end of the reactor (the volume of the
first one can be reduced as less nittate is to be removed there) (Nyberg et al., 1996). This
option allows the reduction of the nittate recừculation flow, and therefore the reduction of
oxygen that is ttansferred into the anoxic zone. This improves the carbon utilization
efficiency. Adding a second anoxic zone allows reducing the effluent nittate
concentration to levels that is difficult to achieve with pre-denitrification only.
4.3. EXTERNAL CARBON DOSAGE CONTROL SYSTEMS
The dosing rate of external carbon to a biological nitrogen removal plant is important.
The dosage should guarantee a satisfactory nittate removal, and in the mean time, should
be at a minimum level (except when excess hydrolysate is available). Dosing too much
increases the operational cost considerably due to higher carbon source consumption,
higher sludge production and increased oxygen demand. Several control strategies have
been proposed. Control systems have been designed for both singlesludge pre-
denhrification systems (Londong, 1992; Isaacs et al., 1995; Hoen et al., 1996; Yuan et
al., 1996, 1997; Lindberg and Carlsson, 1996; Zeghal et al., 1997; Steffens and Lant,
1999) and two-sludge post-denittification systems (Puznava et al., 1998).
4.3.1. Control of external carbon dosage to recirculating biological nitrogen removal

systems
Depending on the conttolled variable chosen in the design, two sttategies have been
proposed for the conttol of external carbon dosage to a pre-denitrification system. One is
to conttol the effluent nitrate nittogen concenttation below a pre-specified limit (Londong,
1992; Hoen et al., 1996) or at a set-point (e.g. Steffens and Lant, 1999) (Strategy I). The
other is to control the nitrate nitrogen concentration in the denitrification zone at a low
set-point (Yuan et al., 1996, 1997; Lindberg and Carlsson, 1996; Zeghal et al., 1997)
(Strategy II).
Strategy I
Londong (1992) and Hoen et al. (1996) proposed to add external carbon to the anoxic
zone when the measured/predicted effluent nittate nittogen concenttation exceeded its
limit. The dosage rate in these periods was conttolled such that the COD to N ratio to the
anoxic zone was at a pre-selected value. The conttoller is obviously a feedforward one. In
207
addition to the measurement of effluent nittate concenttation, the controller also requừes
measuring the influent readily biodegradable COD concenttation. This sttategy has also
been studied and evaluated using model-based and PID feedback conttollers (Lindberg
1998; Steffens and Lant,1999), with some simulation results reported.
While some improvement to the nitrate removal can generally be expected with this
strategy, it does not guarantee a minimum dosage of the external carbon source. The
dosage rate is determined on the basis of effluent nittate concenttation, regardless whether
or not the addition improves the nittate removal. Addition of external carbon obviously
does not increase the denitrification rate when the nitrate concenttation in the anoxic zone
is zero. The full-scale experiment reported in Regan et al. (1998) showed that the effluent
nittate nittogen was reduced by only 0.05-0.01 mg per mg methanol dosed due to a too
low nittate nittogen concenttation in the denitrification zone.
Strategy II
The sttategy of conttolling the nittate nittogen concenttation (SNO,AN) at a low set-point by
manipulating the external carbon dosage rate represents a solution to the above problem
(Yuan et al., 1996,1997; Lindberg and Carlsson, 1996; Zeghal et al., 1997, 1998).
Conttolling SNO,AN at a low but non-zero set-point guarantees the effectiveness of
external carbon on the one hand, and an (almost) complete removal of the recừculated
nitrate on the other hand, thus preventing insufficient denittification in the anoxic zone.
Obviously, the sttategy is not able to conttol the effluent nittate concenttation at a set-
point. This concenttation varies with the influent nittogen loading. However, as will be
shown below, the average effluent nitrate nittogen concenttation can be controlled at a
given level using this conttol sttategy and by applying an appropriately determined
constant nittate recứculation flow rate.
The choice of the set-point for SNO,AN> denoted as SNo,AN,sp, is obviously important for
the minimization of carbon addition. The denittification rate depends on both COD and
niữate concenữations. A lower SNO,AN,SP requừes a higher COD concenttation in the anoxic
zone in order to maintain a sufficient denitrification rate for the removal of the
recừculated niữate, resulting in a larger leakage of COD (both influent and external) to
the aerobic zone. A higher SNo,AN,sp results in a lower removal rate of the recừculated
nitrate. A higher niữate recừculation flow has to be used in order to keep the average
effluent nitrate at the requừed level, causing more COD leakage to the aerobic zone. The
arguments above suggest that the amount of carbon that leaks to the aerobic zone, and
hence is no longer available for denittification, is minimal with some intermediate value
of SNO,AN- Yuan et al. (1997) investigated the determination of the optimal SNo,AN,sp using
the IAWQ ASM1 (Henze et al., 1987). By investigating the dependency of the requừed
carbon dosage rate on SNo,AN,sp, it was suggested that 1 mg N/L be chosen as the set-
point. It was also shown that the optimal set-point is rather insensitive to model
parameters and loading conditions.
The response of SNO,AN, to the carbon dosage rate was also analysed in Yuan et al.
(1997) by means of linear zing ASM1 (Henze et al., 1987). It was found that the conữol
channel is approximately a first order system, which indicates that a proportional
feedback conữoller with a high gain could be used to conữol the carbon dosage without
causing oscillations in the controlled variables or loosing stability of the closed-loop
208
system, provided that perfect measurement of SNO,AN is available. The conclusion was
validated by simulation studies using ASM1 (Henze et al., 1987). SNO,AN was tightly
conữolled at its set-point.
Taking into consideration the noise and delay that are normally associated with the
niữate measurement, another two controllers were designed in Yuan et al. (1997). Both
employed a feedforward component to release the feedback gain. One requừed the
measurement of the nitrate concentration in the aerobic zone to provide the feedforward
information, the other used a constant feedforward based on the average COD and
niữogen loading to the plant. In the latter case, a non-linear feedback gain was designed
to improve the conữol accuracy. Both conữollers were validated by simulation studies.
The latter was further evaluated by a full-scale experiment with satisfactory results.
Zeghal et al. (1997) developed a similar conữoller for a pre-denitrification Biostyr®
up-flow floating biofilter system. Analysis showed that the response of SNO,AN (at the
end of the anoxic zone) to the carbon dosage can be characterized by a first order transfer
function plus a delay. A PI controller was thus designed to manipulate the carbon dosage
rate so that SNO,AN was controlled at Img N/L.
A model-based carbon dosage conữol system to control SNO,AN at a set-point was studied
by Lindberg and Caisson (1996). The conữoller was designed based on a linear model
using the generalized minimum-variance approach (Clarke and Gawthrop, 1975). The
parameters involved in the conữoller were estimated on-line using a recursive least square
algorithm. In addition to SNO,AN,, the controller also requừed the measurement of the
influent COD and the niữate concenữation in the nitrate recừculation flow. The latter two
provided information not only to the feedforward component of the controller, but also to
the parameter estimation algorithm. The conữoller was validated by both simulation and
pilot plant studies.
As mentioned, Strategy II does not dfrectly control the effluent nitrate nitrogen
concentration. Although it is able to keep average effluent nitrate nitrogen at the requfred
level, it by no means guarantees that the instant effluent nitrate will be lower than its
limit. The problem can be solved by adding a nitrate recirculation control loop, which
increases recirculation flow when the effluent nitrate limit is exceeded (Londong, 1992).
The multiple-loop controller guarantees that the right amount of nitrate is recirculated and
that all the recfrculated nitrate is removed.
4.3.2. Control of external carbon dosage to alternating biodenipho® systems

Isaacs et al. (1995) studied the control of external carbon dosage to an alternating
BioDenipho® system. A typical cycle of the BioDenipho® process has been shown in
Fig. 5. The system studied by Isaacs et al. (1995) consisted of only four phases (Phase c
and Phase F in figure 5 were not present). External carbon was dosed during the anoxic
phase of a reactor.
The objective was to remove all nitrate initially present in a reactor when transiting
from an aerobic phase to an anoxic phase with minimum carbon dosage. The latter was
equivalent to maximize the usage of influent COD for denitrification. Therefore the
control strategy used was to control the dosage rate such that the denitrification was just
completed at the end of the anoxic phase (Isaacs et al., 1995). A feedforward controller
was designed, and demonstrated by simulation and pilot plant studies.
209
4.3.3. Control of external carbon to two-sludge post-denitrification systems

The control of carbon dosage to post-denitrification system presents a simpler problem,
compared to the case of single-sludge pre-denitrification systems, due to the absence of
the disturbance from influent COD. A simple solution is to use a feedforward controller,
which keeps the COD to nitrate ratio to the reactor at an appropriately chosen constant
value. A feedback component may also be included to adjust the dosage rate based on the
nitrate concentration in the effluent of the denitrification reactor.
Puznava et al. (1998) evaluated three controllers on an upflow floating biofilter
denitrification process. A feedback controller was designed to control the effluent nitrate
nitrogen concentration at a set-point (2 mg N/L). A feedforward controller was designed
to control the inlet COD to nitrate ratio. The combination of the two resulted in the third
controller. The latter two performed equally well, while the performance of the feedback
controller was less satisfactory owing to a long process delay.
4.4. CONCLUSIONS
External carbon addition has been proven to be an effective means for improving nitrate
removal in a biological nitrogen removal system. The type of the carbon source, the
location where it is added and the addition rate are important for the efficient use of the
carbon sources:
• Ethanol has been shown to be a better (but more expensive) external carbon source than
methanol, especially when added to the anoxic zone of a singlesludge biological
nitrogen removal system.
• When a large amount of external carbon is needed, it may be necessary to build a

separate denitrification reactor, where the external carbon is supplied to further
reduce the nitrate content in the effluent from the mainstream system. Significant
savings of the carbon source can be expected, compared to the case where the carbon
is dfrectly added to the mainstream system.
• The carbon addition rate needs to be controlled on-line. A good strategy is to control
the outlet nitrate concentration from the denitrification zone/reactor at a low set-
point.
However, the control system does not increase the utilization efficiency of the influent
COD for nitrate removal.
5. SRT optimisation and control
Notwithstanding the importance of SRT to a nitrogen removal process, the on-line control
of this parameter has not attracted much research. In this section, the impact of SRT on a
biological nitrogen removal plant will be analysed. This will be followed by a brief
summary of an on-line SRT controller, which was designed to minimize the SRT without
risking the nitrification process (Aquafin and Severn Trent Water, 1998).
5.1. IMPACT OF SRT ON A BIOLOGICAL NITROGEN REMOVAL PLANT
210
The sludge retention time (SRT) is an important design and operating parameter for a
biological nitrogen removal plant. As mentioned in a previous section, the SRT of a
biological nitrogen removal plant should be designed and operated sufficiently long in
order to secure the nitrification process. Applying long SRT also has the advantage of
reducing the sludge production. On the other hand, a short SRT offers several advantages.
A short SRT reduces the total amount of MLSS in a treatment plant. As has been
shown in figure 2, the amount of MLSS, and hence the size of the plant, increase almost
linearly with SRT. Obviously, significant capital cost can be saved if a shorter SRT can
be used in the design stage of a biological nitrogen removal plant. At the level of
operation, a smaller MLSS concentration implies a smaller loading to the secondary
settler, which is beneficial to the sludge and water separation in the settler. This further
implies that the plant may be able to take higher hydraulic loading, reducing the number
of bypassing the wastewater dfrectly to receiving waters during wet weather periods. A
smaller MLSS concentration also reduces aeration cost due to decreased endogenous
respfration.
Moreover, a short SRT is beneficial for phosphorus removal, when it is accomplished
simultaneously with nitrogen removal (van Loosdrecht et al., 1998). The optimal SRT for
phosphorus removal reported in literature falls in the range of 5 to 12 days (see e.g. Choi
et al., 1996; Chuang et al., 1997; Nolasco et al., 1998).
In addition, a shorter SRT results in ’younger’ sludge. As shown in figure 2, the
active biomass to MLSS ratio increases significantly with the decrease of SRT. There has
also been evidence that a smaller SRT may result in the washout of nitrite oxidizers so
that nitrification ends up with nitrite (see e.g. Hellinga et al., 1998; van Loosdrecht
211
and Jetten, 1998). Denitrification from nitrite

requfres less COD and less oxygen to oxidize
ammonia.
However, a shorter SRT leads to a higher sludge production rate.
5.2. SRT MINIMIZATION VIA SURPLUS SLUDGE WASTE FLOW CONTROL
A control system to optimise SRT on-line by means of manipulating the surplus sludge
waste flow has been reported (Aquafin and Severn Trent Water, 1998). The strategy
employed was to minimize the SRT without risking the nitrification process.
Fig.6. Structure of a patent pending SRT control system (after Aquafin and Severn Trent Water, 1998). SNO'AN,
SNO,AE are the nitrate nitrogen concentrations at the ends of anoxic and aerobic reactors, respectively; SNH and
SNH,sp are the effluent ammonia nitrogen concentration and its set-point, respectively; CN is the nitrification
capacity of the sludge; K is a feedback gain.
The following four loops were used to optimise SRT (see figure 6):
• The feedforward loop a calculates the nominal surplus sludge waste flow based on the
maximum specific growth rate of the nitrifiers (|iA,max), such that a certain amount
of spare nitrification capacity is provided to the plant. The amount of the spare
capacity is determined according to the variation of the influent nitrogen load. In
general, this loop generates a waste flow rate that is significantly higher than what
would be used without the control system. |iA,max is estimated on-line using the on-
line measured nitrate data (Nowak et al., 1994; Yuan et al., 1999).
• The ON/OFF feedback loop b switches off the surplus sludge waste flow when the
212
effluent ammonia concenttation (flow proportional daily average) goes exceeds a

certain level. This loop protects against estimation errors made in the feedforward
loop on the one hand, and responds to an abnormally high nitrogen load to the plant
on the other hand.
• The ON/OFF feedback loop c switches off the surplus sludge waste flow when the
nitrification capacity of the plant decreases below a certain percentage of the moving
average of this variable. This loop protects the system against a sharp drop of the
nifrification capacity due to toxicity incidents or abnormally low nifrogen load.
• Finally, the outer proportional feedback loop d corrects conttol errors of the inner
loops. It adjusts the waste flow rate generated by the feedforward loop using a
proportional feedback loop with a low gain.
The control system has been evaluated using simulation and pilot plant studies and
implemented into a Nitrogen Removal Control Kit (NRC-Kit) (Aquafin and Severn Trent
Water, 1998).
5.3. CONCLUSIONS
The sludge retention time of a biological nitrogen removal plant can be minimized online
by means of monitoring the nitrification process. The minimization of SRT results in less
MLSS in the system, allowing the plant to take a load that is higher than designed.
However, on-line SRT confrol does not change the fact that all types of solids in the
biological nitrogen removal plant have the same retention time. The resulting SRT,
though minimized, still leads to a large accumulation of inert solids in the plant.
6. Side-stream nitrifier supplies
The following two sections review the optimisation of nitrogen removal processes by
means of innovative process designs. Different from the on-line process confrol
technology, which optimises a process within the constraints imposed by the process
design, this approach favours changing, or eliminating the constraints of fraditional
designs, and thus presents more fundamental solutions to the problems.
A large fraction of the capital cost of a biological nitrogen removal plant is caused by the
fact that all particulate components in activated sludge (autotrophs, heterotrophs and inert
solids) have the same retention time. Providing nitrifiers the required retention time (RT)
results in equally long RTs of heterotrophs and inert solids. The accumulation of the inert
solids in the system is in fact responsible for the large volume demand.
Schemes have been developed to alter the retention times of different components in
the system such that the RTs of active biomass are extended to the desfred level without
raising the RTs of inert solids to the same level (RT decoupling). The volume
requfrement is thus significantly reduced.
213
Zhiguo Yuan, Jurg Keller and Paul Lant
6.1. SHORTENING THE RTS OF INERT SOLIDS VIA SLUDGE STORAGE
Yuan et al. (1998, 2000) investigated the properties of the plant as shown in figure 7.
Different from an ordinary biological nitrogen removal plant, the plant contains a surplus
sludge storage tank. The idea is to design the main stream plant with an SRT that allows
the plant to treat the ordinary loads (including diurnal variations), while keeping the
sludge that is needed for treating shock nitrogen load and/or inhibitory/toxic influent in
the storage tank. During ordinary load periods, the surplus sludge is wasted to the surplus
sludge storage tank, which is properly aerated. The ’overflow’ of that tank then goes to
the sludge treatment. During periods of nitrogen shock loads and/or inhibitory/toxic
influent, the sludge is pumped back into the main stream to temporarily enhance the
nitrification capacity.
Fig.7. An activated sludge wastewater treatment plant with a surplus sludge storage tank
(after Yuan etal., 1998)
The surplus sludge storage tank is designed as follows. In order to give the plant shown in
figure 7 (called the new plant), with a main stream SRT = Oxmaini the same capability to
treat nitrogen shocks as an ordinary plant with a SRT = Ox,trad>Ọx,main> the SRT in
the storage tank Ox,st, which is defined as Ox,st = VsJQw where vst is the volume of the
storage tank and Qw is the waste flow rate, should be designed as (Yuan et al., 1998),
_ 0Y^'-0Y.
/3 __ A ,traa X ,main
ơ
x,st ” ~r~ (8)
bAOY +1
A A ,main
where bA is the decay coefficient of the autotrophic biomass. For instance, in order to
provide a plant with Ox,main =10 days the same capability to counter nitrogen shocks as
a traditional plant with Ox,trad =15 days, Ox,st is requừed to be 2.5 days (bA = 0.1 d1
assumed).
Yuan et al. (1998) mathematically proved that the overall RTs of the four particulate
components, namely autotrophs, heterotrophs, inert solids from influent and inert solids
214
produced by biomass decay, in the whole plant,

denoted as OH,overall’ 0A)OVeraib 0[,overall and
Op,overall’ respectively, satisfy,
n @ x,frad
V A,overall > @x,trad
@ H,overall
=
@x,main + @x,st <
@x,trad (9)
<
@x,main + @x,st <
@x,trad
@ I,overall
Op,overall Equation (9) clearly indicates that the active biomass has longer RTs than
the inert solids. For example, when Ox,main = 10 days and Ox,st = 2.5
days, 0A,overall’ OH,overall’ 0Itoveraii and Opoverall are 15, 17.5, 12.5 and
11 days, respectively. This implies that, with such a design, one may extend the retention
time of the active biomass to the desfred level without rising the retention times of the
inert solids to the same level. It was estimated that the savings on the reactor volume are
typically around 20% (Yuan et al., 1998).
The concept has been fully verified on a pilot plant by Yuan et al. (2000). Furthermore, it
was observed that the decay rate of the nitrifiers in the storage tank could be maintained
at an extremely low level by controlling the DO at a low level.
6.2. SIDE STREAM NITRIFICATION OF REJECT WATER
The reject water produced by sludge thickening and dewatering contains high ammonia
content. When recycled to the secondary treatment, the reject water typically represents
10-30% of the total nitrogen load to the plant (Hellinga et al., 1998; Jeavons et al., 1998;
Rosen et al., 1998). Recycling the reject water in an unbalanced manner may cause
significant fluctuations in effluent ammonia concentration (Jeavons et al., 1998). The
added oxygen demand may also cause an oxygen limitation situation in the aeration tank
(Hellinga et al., 1998), which demands expansions of the tank. To solve the problems,
processes with side-stream nitrification of the reject water have been developed and
applied to full-scale biological nitrogen removal plants (Hellinga et al., 1998; Jeavons et
al., 1998; Mossakowska et al., 1997; Rosen et al., 1998; Wett et al., 1998). The surplus
sludge waste from the side stream nitrification reactor, which is nitrifier rich, is often
recycled to the mainstream reactor (Hellinga et al., 1998; Jeavons et al., 1998).
Kos (1998) studied the process using dynamic simulation. He concluded that, due to the
supply of nitrifiers from the side-stream reactor, the mainstream system demands a
significantly smaller SRT than what would otherwise be needed to achieve the same
degree of nitrification. A more theoretical analysis of the process is developed below.
Assuming that the mainstream plant has an SRT of Ox,main’ while that of the side
stream nitrification reactor is Ox,ssn- Further assuming that:
215
Mass balance analysis shows that, by wasting the surplus sludge of the side stream
system to the main stream one, the amount of nitrifiers contained in the latter is
equivalent to that of a plant receiving the same influent but operated with an SRT of,
0 + 1 + ^)
Xe
" bAỡXjsn-ồbA0ỵ^+ỉ (10)
A graphical representation of equation (10) is shown in figure 8. Obviously, the side-

stream process is more beneficial when Õ is higher and when the side stream SRT
(^x,ssn) is lower. While parameter ốis generally not at the disposal of the designer, 0x>ssn
should be designed as small as possible. Fortunately, the reject water usually has high
temperature, making short SRT in the side stream possible. The SHARON (Single reactor
High Activity Ammonia Removal Over Nitrite) developed in Hellinga et al. (1998) to
treat reject water has an SRT of 1.5 days. The aerobic sludge age is only 1 day.
As the wasted sludge from the side stream reactor contains negligible inert solids, feeding
the sludge does not significantly alter the retention time of inert solids in the mainstream
reactor. Therefore, a regime of different retention times for different components is
formed in the mainstream system.
6.3. CONCLUSIONS
Supplying nitrifiers from a side-stream system results in shorter retention times of inert
Fig.8. The equivalent retention time of nitrifiers in a plant with main stream SRT Gx,main =10
solids than nitrifiers in a biological nitrogen removal system, allowing a significant
days and with a side stream nitrification reactor treating reject water, as a function of the side
reduction of the volume of a biological nitrogen removal plant, and hence the capital cost.
stream SRT (0K,ssn) and the nitrogen strength of the reject water (Ổ)
The techniques provide a low-cost option for upgrading a COD removal plant to nitrogen
removal.
Control systems are requừed for the operation of these plants. For example, the SRT
control system presented in section 5 can be used to minimize the SRT of the side stream
nitrification reactor treating reject water.
216
7. Novel multi-sludge systems
7.1. ATTACHED GROWTH PROCESSES
Providing a surface area inside reactors for the biomass to grow on has been employed in
many different ways and for several decades (e.g. trickling filters, fixed bed filters etc).
However, new processes have been emerging in recent years that tty to overcome some of
the disadvantages of the existing systems, such as blocking or channelling. These
processes use biomass-growth-supporting media, either fixed or as suspended carriers, in
the reactors. The systems have been developed as high-rate COD removal processes or as
an economic means for upgrading COD removal plants to nutrient removal. The latter is
of particular interest since they provide an alternative to the single-sludge activated
sludge systems whereby a two-sludge process can be established without additional
reactors and clarifiers. As such, they offer a possible solution of the fundamental
difficulties associated with single-sludge BNR systems. The attached growth processes
(AGP) are analysed here in terms of theừ volume requirement, treatment capacity, and
influent COD utilization efficiency for nitrate reduction as well as other properties.
An AGP typically has a pre-denittification configuration, with the media added to the
aerobic reactor (Emori et al., 1994; Rusten et al., 1994; Sen et al., 1994; Deguchi and
Kashiwaya, 1994; Morper, 1994; Takizawa et al., 1996; Mishima et al., 1996; Randall
and Sen, 1996; Chuang et al., 1997; Matsumura et al., 1997; Aravinthan et al., 1998; van
Benthum et al., 1998b), or to both the aerobic and anoxic reactors (Deguchi and
Kashiwaya, 1994; Rusten et al., 1995a, 1995b; Takizawa et al., 1996; Su and Ouyang,
1996; Kim et al., 1997; Welander et al., 1997; Welander et al., 1998; Aravinthan et al.,
1998; Zhang et al., 1998). This results in two different types of attached growth
processes, one with autotrophs growing on the media but heterotrophs growing in
suspension (sometimes also called hybrid systems), and the other with both autotrophs
and heterotrophs growing on the media.
To simplify the discussion, the latter is used as an example for the analysis, a basic
structure of which is shown in figure 9. Growth media is used in all three zones. In the
anoxic (AN) zone, heterotrophs oxidize influent COD using nitrate as electron acceptor.
In the aerobic/anoxic (AA) zone, heterotrophs oxidize COD using either nitrate or oxygen
as electron acceptor. Oxygen is supplied when nitrate is not present. In the aerobic (AE)
zone, autotrophs oxidize ammonia nitrogen.
An important feature of the system is that autotrophs and heterotrophs are physically
separated. Autotrophs grow neither in the AN zone nor in the AA zone due to either the
absence of oxygen or failure to compete with heterotrophs for oxygen (Hem et al., 1994;
Boiler et al., 1994; van Benthum et al., 1998a). Similarly, heterotrophs could hardly grow
in the AE zone because of the lack of COD. Zhang et al. (1998) reported that the number
of nittifiers in the AE zone is about 102 to 103 times higher than that of
heterotrophs. Bacteria normally grow in the suspended phase as well. However, the
amount of bacteria in suspension is small compared with that on biofilm. Many systems
use growth support media in all tanks also in order not to have any sludge recycle, and the
clarifier is only used to separate the waste sludge from the effluent.
The system shown in figure 9 is analysed below in terms of its volume requừement,
treatment capacity and influent COD utilization for nitrate reduction.
217
sludge recycle (optional) sludge waste
Fig.9. An attached growth process with polypropylene pellets used in both aerobic and anoxic
zones as bacteria-growth-supporting media
7.1.1. SRT and volume requirement

Growing on the supporting media, bacteria typically have a longer retention time than in a
conventional activated sludge system. Therefore more bacteria are maintained in the
system. However, they are accommodated in a more compact space due to the higher
solids density in a biofilm system compared to the suspended flock processes. This has in
fact been the most appealing feature of AGP’s. The system also requừes a much smaller
settler, particularly if there is no sludge recycle.
7.1.2. Treatment capacity

The treatment capacity of an AGP was found to be significantly higher than that of a
conventional activated sludge system. The separation of heterotrophs and autotrophs is
likely responsible for the increased treatment capacity. Both nitrifiers and denitrifiers are
only present in the zones of the reactor where they are requừed. This leads to a situation
where the full nitrification and denitrification capacity of the plant can be used
simultaneously, e.g. all nitrifiers in the system are functional at all times. Comparison has
shown that the maximum nitrification rate in the AE zone of an AGP can be a few times
higher than a conventional activated sludge system receiving the same nitrogen load (see
e.g. Deguchi and Kashiwaya, 1994; Randall and Sen, 1996). A similar increase of the
denitrification rate was also observed by Deguchi and Kashiwaya (1994).
7.1.3. Influent COD utilization efficiency for nitrate reduction
As analysed in section 2, COD is leaked to the aerobic zones of a conventional activated

sludge system in three ways, particulate COD which is adhered to sludge flocks, cell
COD and soluble COD. Since most of the particulate and cell COD is kept in the biofilm
in an AGP, the COD leakage in the first two ways is dramatically reduced. The leakage of
soluble COD is also reduced because of the increased denitrification rate in the AN and
AA zones. Kim et al. (1997) reported that 95% of the influent suspended solids and 80%
of influent BOD are removed in an anoxic biofilter. Liu et al. (1998) reported a similar
BOD removal rate (85%).
The better availability of influent COD for denitrification implies an AGP system
requừes a lower influent COD to nitrogen ratio than does a conventional activated sludge
218
system.
7.1.4. Comparison with other multi-sludge systems

Attached growth systems are often preferred to many other multi-sludge systems because
of the significantly lower complexity, particularly compared to systems with multiple
clarifiers. Furthermore, the capability of using influent COD for denitrification as
outlined above and the self-regulation of pH and alkalinity by the aerobic-anoxic recycle
are additional advantages over systems where nitrification and denitrification are
performed separately.
7.1.5. Aeration in an attached growth system

The main drawback of AGP systems has been identified as its high aeration cost (see e.g.
Welander et al., 1998). The reason for this is that a high bulk DO concentration is needed
to drive the diffusion of oxygen into the biofilm. It has been reported that bulk DO
concentrations below 3-4mg/L start limiting the nitrification rate (Takizawa et al., 1996;
Aravinthan et al., 1998; Welander et al., 1998).
However, the added aeration cost due to the increased bulk DO concentration is
partially compensated by a lower oxygen uptake rate (OUR). The OUR in the AE zone of
the system is significantly smaller than that in a conventional activated sludge reactor,
due to the lack of or reduced heterotrophic activities (COD oxidation and endogenous
respừation). On-line aeration control may further reduce the aeration cost. The bulk DO
concentration should be controlled at the minimum level achieving complete nitrification
under the specific loading. The fact that the nitrification rate is linearly dependent on the
bulk DO concenttation in a rather large range (Rusten et al., 1994; Mishima et al., 1996;
Aravinthan et al., 1998; Welander et al., 1998) makes the DO set-point an effective
control handle for the improvement of nitrogen removal.
7.2. COD PRESERVATION FOR DENITRIFICATION
Similar to the attached growth systems discussed above, the DEPHANOX and similar
processes (Wanner et al., 1992; Kuba et al., 1993, 1996; Bortone et al., 1994, 1996; Sorm
et al., 1996, 1997; Jun et al., 1997) are other types of multi-sludge systems developed in
recent years. Furthermore, these systems allow for simultaneous phosphorus removal
using the same COD as for nitrate removal.
The basic structure of a DEPHANOX system is shown in figure 10 (Wanner et al., 1992;
Bortone et al., 1994, 1996; Sorm et al., 1996, 1997). In the anaerobic reactor (1), the
influent particulate COD is enttapped on the sludge flocks. A large fraction of the soluble
COD is also ’taken' into the sludge via different mechanisms: adsorption, absorption or
anaerobic storage. In the presence of phosphorus accumulating organisms (PAO), the
short chain fatty acids are taken up by PAO and stored as intracellular products (PHA),
accomplished by the release of phosphate from poly-phosphate. Jun et al. (1997) reported
that, after 30 minutes anaerobic contact, more than 90% of the total COD and soluble
COD were separated from the liquid phase. The intermediate settler (2) following the
anaerobic reactor separates the organic substrate-rich activated sludge from the ammonia-
rich supernatant. The supernatant then goes to an aerobic biofllm reactor (3) where
nitrification takes place, while the settled sludge bypasses the nitrification phase, entering
the anoxic reactor (4) together with the effluent from the biofilm reactor. In the anoxic
219
reactor (4), denitrification takes place. The organic substrate contained in the sludge is
oxidized by heterotrophs, including a large portion of PAO, using nitrate as the electron
acceptor. Denitrification by PAO is accompanied by the simultaneous phosphorus uptake
(see e.g. Mino et al., 1998; Meinhold et al., 1998). Note that, in this case, PHA stored by
PAO is used for both P-uptake and nitrate reduction. The aerobic reactor (5) allows
nitrogen gas stripping from the sludge before the latter enters the final settler (6). It also
further improves p uptake and removes any residual COD. Kuba et al. (1993, 1996) and
Jun et al. (1997) implemented a similar process using SBRs.
Similar to the attached growth systems, the DEPHANOX process is also a two-
sludge system and in fact uses attached growth media for the nitrification reactor. With
nitrification accomplished in a separate reactor, the process has similar features in terms
of volume requữement and treatment capacity. Through stimulating PAO to denitrify, the
DEPHANOX process further improves the influent COD utilization efficiency.
A reported problem of the DEPHANOX process is the high effluent ammonia
concentration (see e.g. Kuba et al., 1996). When bypassing the settled sludge from settler
(2) to reactor (4), considerable nitrogen in both soluble and particulate form is bypassing
the nitrification phase. Although part of the nitrogen is assimilated into biomass cells
during heterotrophic growth, a large fraction is dừectly discharged to the effluent,
resulting in high nitrogen content in the effluent.
Fig. 10: The DEPHANOX process: 1. Anaerobic reactor; 2. Intermediate settler; 3. Fixed- film nitrification
reactor; 4. Anoxic reactor; 5. Re-aeration reactor; 6. Final settler (after Wanner et al., 1992)
Kuba et al, (1996) suggested solving the problem by reducing the ratio between the
bypassing flow and the supernatant flow (Settler 2). However, the approach has limited
effect. Reducing the ratio has no impact on the amount of the particulate nitrogen that is
fed to reactor (4). A better approach may be to add attached-growth media into reactor
(5), which will support the growth of nitrifiers, despite the small aerobic sludge age of the
suspended sludge.
7.3. CONCLUSIONS
The novel multi-sludge systems presented in this section have provided promising
solutions to the fundamental problems of single-sludge BNR systems. More research is
requừed to further verify these concepts and to refine the designs. On-line control of these
systems should also be studied as it offers likely significant advantages in managing the
highly variable influent loads of typical domestic treatment plants.
220
8. Conclusions
Biological nitrogen removal is a complex process because it involves two conflicting

processes, each process being performed by different bacteria. In most cases, these
bacteria co-exist in 'one sludge'. Elementary mass balance analysis has revealed the
following fundamental problems of single-sludge biological nitrogen removal systems:
• The co-existence of aerobic and anoxic conditions in the system reduces the
treatment capacity of the system, as only a fraction of nittifiers and denittifiers are
functional at any given moment. This is particularly a problem for nitrification as the
number of functioning nittifiers is usually the limiting factor.
• The co-existence of aerobic and anoxic conditions also results in low utilization
efficiency of influent COD for nitrate reduction, as a large fraction of influent COD
is oxidized aerobically, making it difficult to have a high degree of nitrogen removal
from waste water with low COD to nitrogen ratio.
• The long SRT requữed by nitrifiers results in over-growth of heterotrophs and over-
accumulation of inert solids. Large reactor and settler volumes are thus requừed, with
a large increase in capital cost.
In this paper, we have shown that there are two major schools of thought for addressing
the problems, which adopt either incremental or revolutionary solutions. The 'incremental'
approach is to use process control technology to attempt to get the process to perform to
its capability, within the consttaints imposed by the process designs. The revolutionary
approach favours changing, or removing, the problem by eliminating the design
constraints. This is being achieved by innovative process designs.
On-line process control has been extensively studied. Put simply, the control problem
is how to determine the optimum between nitrification and denitrification online, given
continuous variations in loading. We have investigated control in terms of the available
and effective 'control handles', namely aeration, COD dosage and SRT. The control
strategies have been classified in terms of the measured variables used.
These variables may be considered as either inferential variables, such as ORP, pH and
respfrometty, which are indừect measurements used to infer the key variables, or dfrect
measurements such as ammonia, nitrite and nitrate nitrogen concentrations.
Advances over the last fifteen years in the control of biological nitrogen removal have
been shaped by two major advances: analyser technology and models. The availability of
robust on-line nutrient analyser technology has influenced the control work significantly,
with more and more workers focussing on using these direct measurements for control.
The second major influence has been the development and uptake of models of nitrogen
removal. This has resulted in a large growth in the application of models to the control
system design. The models have been extensively used in the analysis of the processes,
which provides valuable information for the control system designs. Simulation using the
models has provided an economic means to the preliminary verification of the designs.
However, model-based control of biological nitrogen removal is still in its infancy. The
difficulty has been and still is to obtain simple yet accurate models applicable to the
design of control systems. Optimisation of the design of biological nitrogen removal
systems has been shown a critical issue for the performance improvement of biological
221
nitrogen removal systems. We have shown that there is a growing amount of

contemporary work looking at innovative process designs such as attached growth,
multiple sludge and side-stream nitrifier supplying systems. The common link with these
concepts is the desfre to exploit the biomass behaviour, rather than treat it as a consttaint
as is the case with the control work discussed above. We believe that this area is where
the next major advances in biological nitrogen removal operation will occur.
The integration of the innovative process designs with on-line process control will
result in much more efficient biological nitrogen removal systems in the future.
Acknowledgment
The authors would like to thank Prof. Peter Vanrolleghem from the BIOMATH
Department, University of Gent, Belgium, Tekn. Lie. Christian Rosen from the IEA,
Lund University, Sweden and Mr. James Lennox from the AWMC, the University of
Queensland for the fruitful discussions. They also would like to thank ir. Herwig Bogaert
from Aquafin N.V., Belgium for his permission of using unpublished materials in this
paper. The first two authors thank CRC for Waste Management and Pollution Control
Ltd., Australia, for the financial support provided.
References
Aesoy, A., Odegaard, H., Bach, K., Pujol, R. and Hamon, M. (1998). Denitrification in a Packed Bed Biofilm
Reactor (Biofor)- Experiments with Different Carbon Sources. Wat. Res 32(5): 1463-1470.
Al-Ghusain, I., Huang, J., Hao, o. and Lim, B. (1994). Using pH as Real-Time Control Parameter for
Wastewater Treatment and Sludge Digestion Processes. Wat. Sci. Tech. 30(4): 159-168.
Al-Ghusain, I., and Hao, J. (1995). Use of pH as Control Parameter for Aerobic/Anoxic Sludge Digestion. J.
Envừ. Engrg. 121(3): 225-235.
Aquafin, N. V. and Severn Trent Water (1998). An Automatic Controller for the Surplus Sludge Waste Flow in
Nitrifying Activated Sludge Wastewater Treatment Plants. Patent Proposal BE0980377.
Aravinthan, V., Takizawa, s., Fujita, K. and Komatsu, K. (1998). Factors Affecting Nittogen Removal from
Domestic Wastewater Using Immobilized Bacteria. Wat. Sci. Tech. 38(1): 193-202.
Balmer, p., Ekfjorden, L., Lumley, D. and Mattsson, A. (1998). upgrading for Nitrogen Removal Under Severe
Site Restrictions. Wat. Sci. Tech. 37(9): 185-192.
Balslev, p., Lynggaard-Jensen, A. and Nickelsen, c. (1996). Nutrient Sensor Based Real-Time On-Line Process
Control of a Wastewater Treatment Plant Using Recirculation. Wat. Sci. Tech. 33(1): 183-192.
Barnard, J. (1984). Activated Primary Tanks for Phosphate Removal. Water s. A. 10(3): 121-126.
Bertanza, G. (1997). Simultaneous Nitrification-Denitrification Process in Extended Aeration Plants: Pilot and
Real Scale Experiences. Wat. Sci. Tech. 35(6): 53-61.
Boiler, M., Gujer, w. and Tschui, M. (1994). Parameters Affecting Nitrifying Biofilm Reactors. Wat. Sci. Tech.
29(10-11): 1-11.
Bortone, G., Malaspina, F., Stante, L. and Tilche, A. (1994). Biological Nitrogen and Phosphorus Removal in an
Anaerobic/Anoxic Sequencing Batch Reactor with Separated Biofilm Nitrification. Wat. Sci. Tech. 30(6):
303-313.
Bortone, G., Saltarelli, R., Alonso, V., Sorm, R., Wanner, J. and Tilche, A. (1996). Biological Anoxic
Phosphorus Removal - the DEPHANOX Process. Wat. Sci. Tech. 34(1-2): 119-128.
Brinch, p., Rindel, K. and Kalb, K. (1994). upgrading to Nutrient Removal by Means of Internal Carbon from
Sludge Hydrolysis. Wat. Sci. Tech. 29(12)7 31-40.
Brouwer, H., Bloemen, M., Klapwijk, B. and Spanjers, H. (1998a). Feedforward Conttol of Nitrification by
Manipulating the Aerobic Volume in Activated Sludge Plants. Wat. Sci. Tech. 38(3): 245-254.
Brouwer, H., Klapwijk, A. and Keesman, K. (1998b). Identification of Activated Sludge and Wastewater
Characteristics Using Respirometric Batch-Experiments. Wat. Res. 32(4): 1240-1254.
222
Bundgaard, E., Andersen, K. and Petersen, G. (1989). Bio-Denitro and Bio-Denipho Systems-Experiences and
Advanced Model Development: The Danish Systems for Biological N and p Removal. Wat. Sci. Tech.
21:1727 1730.
Charlton, J. (1994). Biological Nutrient Removal Applied to Weak Sewage. Wat. Sci. Tech. 29(12): 41-48.
Charpentier, J., Florenz, M. and David, G. (1987). Oxidation-Reduction Potential (ORP) Regulation: A Way to
Optimise Pollution Removal and Energy Savings in Low Load Activated Sludge Process. Wat. Sci. Tech.
19: 645-655.
Charpentier, J., Godart, H., Martin, G. and Mogno, Y. (1989). Oxidation-Reduction Potential (ORP) Regulation
as a Way to optimise Aeration and c, N and p Removal: Experimental Basis and Various Full-scale
Examples. Wat. Sci. Tech. 21(10): 1209-1223.
Choi, Y., Shin, E. and Lee, Y. (1996). Biological Phosphorus Removal from Wastewater in a Single Reactor
Combining Anaerobic and Aerobic Conditions. Wat. Sci. Tech. 34(1-2): 179-186.
Christensson, M., Lie, E. and Welander, T. (1994). A Comparison Between Ethanol and Methanol as Carbon
Sources for Denitrification. Wat. Sci. Tech. 30(6): 83-90.
Christensson, M., Lie, E., Johansson, p. and Welander, T. (1998). Increasing Substrate for Polyphosphate-
Accumulating Bacteria in Municipal Wastewater Through Hydrolysis and Fermentation of Sludge in
Primary Clarifiers. Water Envfron. Res. 70(2): 138-145.
Chuang, s., Ouyang, c., Yuang, H. and You, s. (1997). Effects of SRT and DO on Nutrient Removal in a
Combined AS-Biofilm Process. Wat. Sci. Tech. 36(12): 19-27.
Clarke, D., and Gawthrop, p. (1975). Self-tuning Conttoller. Proc. IEE 122: 929-934.
Deguchi, H., and Kashiwaya, M. (1994). Study on Nitrified Liquor Recycling Process Operations Using
Polyurethane Foam sponge Cubes as a Biomass Support Medium. Wat. Sci. Tech. 30(6): 143-149.
Demuynck, c., Vanrolleghem, p., Mingneau, c., Liessens, J. and Verstraete, w. (1994). NDBEPR Process
Optimisation in SBRs: Reduction of External Carbon-Source and Oxygen Supply. Wat. Sci. Tech. 30(4):
169-179.
Ekama, G., and Marais, G.V.R. (1984). Biological Nittogen Removal. Pretoria, Water Research Commission.
Emori, H., Nakamura, H., Sumino, T., Takeshima, T., Motegi, K. and Tanaka, K. (1994). High Rate and
Compact Single Sludge Pre-Denitrification Process for Retrofit. Wat. Sci. Tech. 30(6): 31-40.
Gujer, w., Henze, H., Mino., T. and van Loosdrecht, M. (1998). Activated Sludge Model No.3. Wat. Sci. Tech.
39(1): 183-193.
Hallin, s., Lindberg, C.-F., Pell, M., Plaza, E. and Carlsson, B. (1996). Microbial Adaptation, Process
Performance and a suggested Control Sttategy in a Pre-denitrification System with Ethanol Dosage. Wat.
Sci. Tech. 34(1-2): 91-99.
Hallin, s., and Pell, M. (1998). Metabolic Properties of Denitrifying Bacteria Adapting to Methanol and Ethnol
in Activated Sludge. Wat. Res. 32(1): 13-18.
Hamamoto, Y., Tabata, s. and Okubo, Y. (1997). Development of the Intermittent Cyclic Process for
Simultaneous Nittogen and Phosphorus Removal. Wat. Sci. Tech. 35(1): 145-152.
Hao, o., and Huang, J. (1996). Alternating Aerobic-Anoxic Process for Nitrogen Removal: Process Evaluation.
Water Environ. Res. 68(1): 83-93.
Hellinga, c., Schellen, A., Mulder, J., van Loosdrecht, M. and Heijnen J. (1998). The SHARON Process: an
Innovative Method for Nitrogen Removal from Ammonium-rich Waste Water. Wat. Sci. Tech. 37(9): 135-
142.
Hem, L., Rusten, B. and Odegaard, H. (1994). Nitrification in a Moving Bed Biofilm Reactor. Wat. Res. 28(6):
1425-1433.
Henze, M., Grady Jr, c., Gujer, w., Marais, G., Matsuo, T. (1987). Activated Sludge Model No.l. London,
IAWQ.
Henze, M., Harremoes, p., la Cour Jansen, J. and Arvin, E. (1995). Wastewater Treatment. London, springer.
Hoen, K., Schuhen, M. and Kohne, M. (1996). Conttol of Nitrogen Removal in Wastewater Treatment Plants
with Predenitrification, Depending on the Actual Purification Capacity. Wat. Sci. Tech. 33(1): 223-236.
Holmberg, u., Olsson, G. and Andersson, B. (1989). Simultaneous DO control and Respiration Estimation. Wat.
Sci. Tech. 21: 1185-1195.
Husmann, M., Orth, H., Schlegel, s. and Teichgraber, B. (1998). Application of Process Conttol for Improved
Nitrogen Removal. Wat. Sci. Tech. 38(3): 263-269.
Isaacs, s., Henze, M., Soeberg, H. and Kummel, M. (1994). External Carbon Source Addition as a Means to
Conttol an Activated Sludge Nutrient Removal Process. Wat. Res. 28(3): 511-520.
Isaacs, s., and Henze, M. (1995). Controlled Carbon Source Addition to an Alternating Nitrification-
Denitrification Wastewater Treatment Process Including Biological p Removal. Wat. Res. 29(1): 77-89.
Isaacs, s., Henze, M. and Kummel, M. (1995). An Adaptive Algorithm for External Carbon Addition to an
223
Alternating Activated Sludge Process for Nutrient Removal from Wastewater. Chem. Engng. Science
50(4): 617-629.
Isaacs, s., and Thomberg, D. (1998). Rule Based Conttol of a Periodic Activated Sludge Process. Wat. Sci.
Tech. 38(3): 281-289.
Jeavons, J., Stokes, J., Upton, J. and Bingley, M. (1998). Successful Sidestteam Nitrification of Digested Sludge
Liquors. Wat Sci. Tech. 38(3): 111-118.
Johansen, N., Andersen, J. and la Cour Jansen, J. (1997). optimum Operation of a Small Sequencing Batch
Reactor for BOD and Nitrogen Removal Based on On-line OUR-Calculation. Wat. Sci. Tech. 35(6): 29-
36.
Jun, H. B., Lee, s. H., Seo, I. s. and Moon, M. J. (1997). Denitrification by the Preserved Influent COD in a
Separated Sequencing Batch Reactors System. Biological Nutrient Removal 3, Brisbane, Australia.
Kim, Y., Mikawa, T., Tanaka, K. and Emori, H. (1997). Development of Novel Anaerobic/Aerobic Filter
Process for Nitrogen Removal Using Immobilized Nitrifier Pellets. Wat. Sci. Tech. 36(12): 151-158.
Klapwijk, A., Brouwer, H., Vrolijk, E. and Kujawa, K. (1998). Control of Intermittently Aerated Nitrogen
Removal Plants by Detection Endpoints of Nitrification and Denitrification Using Respirometer Only. Wat
Res. 32(5): 1700-1703.
Ko, K., Mcinnis, B. and Goodwin, G. (1982). Adaptive Control and Identification of the Dissolved Oxygen
Process. Automatica 18(6): 727-730.
Kos, p. (1998). Short SRT (Solids Retention Time) Nitrification Process/Flowsheet. Wat. Sci. Tech. 38(1): 23-
29.
Kuba, T., Smolders, G., van Loosdrecht, M. and Heijnen, J. (1993). Biological Phosphorus Removal from
Wastewater by Anaerobic-Anoxic Sequencing Batch Reactor. Wat. Sci. Tech. 27(5-6): 241-152.
Kuba, T., van Loosdrecht, M. and Heijnen, J. (1996). Phosphorus and Nitrogen Removal with Minimal COD
Requirement by Integration of Denitrifying Dephosphatation and Nitrification in a Two-Sludge System.
Wat. Res. 30(7): 1702-1710.
Kurata, G., Tsumura, K., Nakamura, s., Kuwahara, w., Sato, A. and Kanaya, T. (1996). Rettofit of Biological
Nutrient Removal Process Assisted by Numerical Simulation with Activated Sludge Model No.2. Wat. Sci.
Tech. 34(1-2): 221-228.
Larose, A., Perrier, M. and Comeau, Y. (1997). Respirometric Control of the Anaerobic Duration of an SBR
Bio-P Process. Wat. Sci. Tech. 36(5): 293-300.
Leeuw, E., and van't Oever (1996). Process Selection, Design and Operation of the EDE WWTP. Wat. Sci.
Tech. 33(12): 57-63.
Lindberg, C.-F, and Carlsson, B. (1996). Adaptive Conttol of External Carbon Flow Rate in an Activated
Sludge Process. Wat. Sci. Tech. 34(3-4): 173-180.
Lindberg, C.-F. (1998). Multivariable Modelling and Control of an Activated Sludge Process. Wat. Sci. Tech.
37:149-156.
Liu, J., Li, w., Wang, X., Liu, H. and Wang, B. (1998). Removal of Nitrogen from Coal Gasification by
Nitrosofication and Denitrosofication. Wat. Sci. Tech. 38(1): 39-46.
Lo, c., Yu, c., Tam, N. and Traynor, s. (1994). Enhanced Nutrient Removal by Oxidation-Reduction Potential
(ORP) Controlled Aeration in a Laboratory Scale Extended Aeration Treatment System. Wat. Res. 28(10):
2087-2094.
Londong, J. (1992). Strategies for optimised Nitrate Reduction with Primary Denitrification. Wat. Sci. Tech.
26(5-6): 1087-1096.
Londong, J., and Wachtl, p. (1996). Six Years of Practical Experience with the Operation of On-Line Analysers.
Wat. Sci. Tech. 33(1): 159-164.
Lukasse, L., Keesman, K., Klapwijk, A. and van Straten G. (1998). optimal Control of N-Removal in ASPs.
Wat. Sci. Tech. 38(3): 255-262.
Lukasse, L. (1999). Control and Identification in Activated Sludge Processes. Wageningen, Ph.D. Thesis,
Wageningen Agricultural University: 155p.
Majone, M., Dircks, K. and Beun, J. J. (1998). Aerobic Storage Under Dynamic Conditions in Activated Sludge
Process. The State of the Art. Wat. Sci. Tech. 39(1): 61-73.
Marsili-Libelli, s. (1989). Modelling, Identification and Control of the Activated Sludge Process. Advances in
Biochemical Engineering/Biotechnology. A. Fiechter (Ed.). Berlin Heidelberg, Springer-Verlag. 38: 89-
148.
Marsman, E., Roeleveld, p. and Rensink, J. (1997). High Nutrient Removal in the Three-Sludge Sewage
Treatment System: Results and Economic Evaluation. Wat. Sci. Tech. 35(10): 129-136.
Matsumura, M., Yamamoto, T., Wang, p., Shinabe, K. and Yasuda, K. (1997). Rapid Nitrification with
Immobilized Cell Using Macro-Porous Cellulose Carrier. Wat. Res. 31(5): 1027-1034.
224
Meinhold, J., Filipe, c. D. M., Daigger, G. T. and Isaacs, s. (1998). Characterisation of the Denitrifying Fraction
of Phosphate Accumulating Organisms in Biological Phosphate Removal. Wat. Sci. Tech. 39(1): 31-42.
Mino, T., van Loosdrecht, M. and Heijnen, J. (1998). Microbiology and Biochemistry of the Enhanced
Biological Phosphate Removal Process. Wat. Res. 32(11): 3193-3207.
Mishima, K., Nishimura, T., Goi, M. and Katsukura, N. (1996). Characteristics of Nitrification and
Denitrification of the Media-Anaerobic-Anoxic-Oxic Process. Wat. Sci. Tech. 34(1-2): 137-143.
Morper, M. (1994). upgrading of Activated Sludge Systems for Nitrogen Removal by Application of the
LINPOR-CN Process. Wat. Sci. Tech. 29(12): 167-176.
Mossakowska, A., Reinius, L. and Hultman, B. (1997). Nitrification Reactions in Treatment of Supernatant
from Dewatering of Digested Sludge. Wat. Environ. Res. 69(6): 1128-1133.
Nielsen, M., and T. Onnerth (1995). Improvement of a Recirculating Plant by Introducing STAR Control. Wat.
Sei. Tech. 31(2): 171-180.
Nolasco, D., Daigger, G., Stafford, D., Kaupp, D. and Stephenson, J. (1998). The Use of Mathematical
Modelling and Pilot Plant Testing to Develop a New Biological Phosphorus and Nitrogen Removal
Process. Water Environ. Res. 70(6): 1205-1215.
Nowak, o., Schweighofer, s. and Svardal, K. (1994). Nitrification Inhibition-A Method for the Estimation of
Actual Maximum Autotrophic Growth Rates in Activated Sludge Systems. Wat. Sci. Tech. 30(6): 9-19.
Nyberg, u., Andersson, B. and Aspegren H. (1996). Long-term Experiences with External Carbon Sources for
Nitrogen Removal. Wat. Sci. Tech. 33(12): 109-116.
Olsson, G. (1976). State of the Art in Sewage Treatment Plant Control. AIChE Symposium Series 72(159): 52-
76.
Olsson, G. and Andrews, J. F. (1978) The Dissolved Oxygen Profile - A Valuable Tool for Control of the
Activated Sludge Process. Wat. Res., 12, 985-1004.
Olsson, G. and Newell, B. (1999) Wastewater Treatment Systems: Modelling, Diagnosis and Control. IWA
Publishing, London.
Olsson, G., Rundqwist, L., Eriksson, L. and Hall, L. (1985) Self-tuning Control of the Dissolved Oxygen
Concentration in Activated Sludge Systems, in Advances in Water Pollution Control (Drake, R. A. R. ed.),
IAWPRC, Houston, Texas, USA, Pergamon Press, 473-480.
Onnerth, T., Nielsen, M. and Stamer, c. (1996). Advanced Computer Control Based on Real and Software
Sensors. Wat. Sci. Tech. 33(1): 237-245.
Plisson-Saune, s., Capdeville, B., Mauret, M., Deguin, A. and Baptiste, p. (1996). Real-Time Control of
Nitrogen Removal Using Three ORP Bending-Points: Signification, Control Strategy and Results. Wat.
Sci. Tech. 33(1): 275-280.
Potter, T., Koopman, B. and Svoronos, A. (1996). Optimisation of a Periodic Biological Process for Nitrogen
Removal from Wastewater. Wat. Res. 30(1): 142-152.
Puznava, N., Zeghal, s. and Reddet, E. (1998). Simple Control Strategies of Methanol Dosing for Post -
Denitrification. Wat. Sci. Tech. 38(3): 291-297.
Rabinowitz, B. a. B., J. L. (1997). The Use of Primary Sludge Fermentation in Biological Nutrient Removal
Processes. Biological Nutrient Removal 3, Brisbane, Australia.
Randall, c., Barnard, J. and Stensel, H. (1992). Design and Retrofit of Wastewater Treatment Plants for
Biological Nutrient Removal. Western Hemisphere, Technomic Publishing Company.
Randall, c., and Sen, D. (1996). Full-Scale Evaluation of an Integrated Fixed-Film Activated Sludge (IFAS)
Process for Enhanced Nitrogen Removal. Wat. Sci. Tech. 33(12): 155-162.
Regan, J., Koopman, B., Svoronos, s. and Lee, B. (1998). Full-Scale Test of Methanol Addition for Enhanced
Nitrogen Removal in a Ludzack-Ettinger Process. Water Environ. Res. 70(3): 376-381.
Rosen, B., Ullman, A. and Ragnarsson, N. (1998). upgrading for Nitrogen Removal, Using a Combination of
SBR (Sequencing Batch Reactor) Technique and Unloading of Existing Biological Stage. Wat. Sci. Tech.
37(9): 17-24.
Rusten, B., Siljudalen, J. and Nordeidet, B. (1994). upgrading to Nittogen Removal with the KMT Moving Bed
Biofilm Process. Wat. Sci. Tech. 29(12): 185-195.
Rusten, B., Hem, L. and Odegaard, H. (1995a). Nitrification of municipal wastewater in moving-bed biofilm
reactors. Water Envfron. Res. 67(1): 75-86.
Rusten, B., Hem, L. and Odegaard, H. (1995b). Nitrogen removal from dilute wastewater in cold climate using
moving-bed biofilm reactors. Water Environ. Res. 67(1): 65-74.
Sasaki, K., Yamamoto, Y., Tsumura, K., Chichi, s. and Mori, Y. (1996). Development of 2-Reacor Intermittent-
Aeration Activated Sludge Process for Simultaneous Removal of Nitrogen and Phosphorus. Wat. Sci. Tech.
36(1-2): 111-118.
Sen, D., Mitta, p. and Randall, c. (1994). Performance of Fixed Film Media Integrated in Activated Sludge
225
Reactors to Enhance Nitrogen Removal. Wat. Sci. Tech. 30(11): 13-24.

Skalsky, D., and Daigger, G. (1995). Wastewater Solids Fermentation for Volatile Acid Production and
Enhanced Biological Phosphorus Removal. Water Environ. Res. 67: 230.
Sorensen, J., Thomberg, D. and Nielsen, M. (1994). Optimisation of a Nitrogen-Removing Biological
Wastewater Treatment Plant Using On-Line Measurements. Water Environ. Res. 66(3): 236-242.
Sorensen, J. (1996). Optimization of a Nutrient-Removing Wastewater Treatment Plant Using On-Line
Monitors. Wat. Sci. Tech. 33(1): 265-273.
Sorm, R., Bortone, G., Saltarelli, R., Jenicek, p., Wanner, J. and Tilche, A. (1996). Phosphate uptake under
Anoxic Conditions and Fixed Film Nitrification in Nutrient Removal Activated Sludge System. Wat. Res.
30(7): 1573-1584.
Sorm, R., Wanner, J., Saltarelli, R., Bortone, G. and Tilche, A. (1997). Verification of Anoxic Phosphate
Uptake as the Main Biochemical Mechanism of the "DEPHANOX" Process. Wat. Sci. Tech. 35(10): 87-
94.
Spanjers, H., Vanrolleghem, p., Olsson, G. and Dold, p. (1998). Respfrometry in Control of the Activated
Sludge Process: Principles. London, IAWQ.
Steffens, M., and Lant, p. (1999). Multivariable Conttol of Nutrient-Removing Activated Sludge Systems. Wat.
Res. (in press).
Su, J., and Ouyang c. (1996). Nutrient Removal Using a Combined Process with Activated Sludge and Fixed
Biofilm. Wat. Sci. Tech. 34(1-2): 477-486.
Takizawa, s., Aravinthan, V. and Fujita, K. (1996). Nitrogen Removal from Domestic Wastewater Using
Immobilized Bacteria. Wat. Sci. Tech. 24(1-2): 431-440.
Thomsen, H. a. K., K. (1996). N and p On-Line Meters: Requirement, Maintenance and Stability. Wat. Sci.
Tech. 33(1): 147-157.
Thomberg, D., Nielsen, M. and Andersen, K. (1993). Nutrient Removal: On-Line Measurements and Control
Strategies. Wat. Sci. Tech. 28(11-12): 549-560.
van Benthum, w., Garrido, J., Mathijssen, J.,Sunde, J., van Looscrecht, M. and Heijnen, J. (1998a). Nitrogen
Removal in Intermittently Aerated Biofilm Airlift Reactors. J. Envừ. Engrg., ASCE 124(3): 239-248.
van Benthum, w., Derissen, B., van Loosdrecht, M. and Heijnen, J. (1998b). Nitrogen Removal Using Nitrifying
Biofilm Growth and Denitrifying Suspended Growth in a Biofilm Airlift Suspension Reactor Coupled with
a Chemostat. Wat. Res. 32(7): 2009-2018.
van Loosdrecht, M., and Jetten, M. (1998). Microbiological Conversion in Nitrogen Removal. Wat. Sci. Tech.
38(1): 1-7.
van Loosdrecht, M., Brandse, F. and Vries, A. (1998). upgrading of Waste Water Treatment Processes for
Integrated Nutrient Removal-the BCFS Process. Wat. Sci. Tech. 37(9): 209-217.
Vanrolleghem, p., Van Daele, M. and Dochain, D. (1995). Practical Identifiability of a Biokinetic Model of
Activated Sludge Respừation. Wat. Res. 29(11): 2561-2970.
Vanrolleghem, p., Spanjers, H., Petersen, B., Ginestet, p. and Takacs, I. (1999). Estimating (Combination of)
Activated Sludge Model No.l Parameters and Components by Respừometry. Wat. Sci. Tech., 39(1), 195-
214.
Wanner, J., Cech, J. and Kos, M. (1992). New Process Design for Biological Nutrient Removal. Wat. Sci. Tech.
25(4-5): 445-448.
Wareham, D., Hall, K. and Mavinic, D. (1993). Real-Time Control of Wastewater Treatment Systems Using
ORP. Wat. Sci. Tech. 28(11-12): 273-282.
Wareham, D., Mavinic, D. and Hall, K. (1994). Sludge Digestion Using ORP-Regulated Aerobic-Anoxic
Cycles. Wat. Res. 28(2): 373-384.
Welander, u., Henrysson, T. and Welander, T. (1997). Nitrification of Landfill Leachate Using Suspended-
Carrier Biofilm Technology. Wat. Res. 31(9): 2351-2355.
Welander, u., Henrysson, T. and Welander, T. (1998). Biological Nitrogen Removal from Municipal Landfill
Leachate in a Pilot Scale Suspended Carrier Biofilm Process. Wat. Res. 32(5): 1564-1570.
Wett, B., Rostek, R., Rauch, w. and Ingerle, K. (1998). pH-Controlled Reject Water Treatment. Wat. Sci. Tech.
37(12): 165-172.
Wouters-Wasiak, K., Heduit, A., Audic, J. and Lefevre, F. (1994). Real-Time Control of Nitrogen Removal at
Full-Scale Using Oxidation Reduction Potential. Wat. Sci. Tech. 30(4): 207-210.
Yu, R., Liaw, s., Chang, c., Lu, H. and Cheng, w. (1997). Monitoring and Control Using On-line ORP on the
Continuous-Flow Activated Sludge Batch Reactor System. Wat. Sci. Tech. 35(1): 57-66.
Yu, R., Liaw, s., Chang, c. and Cheng, w. (1998). Applying Real-Time Control to Enhance the Performance of
Nitrogen Removal in the Continuous-Flow SBR System. Wat. Sci. Tech. 38(3): 271-280.
Yuan, z., Bogaert, H., Vanrolleghem, p., Thoeye, c., Vansteenkiste, G. and Verstraete, w. (1996). Carbon
226
Dosage Control for Predenitrification Processes. Proceedings Workshop Modelling, Monitoring and
Control of Wastewater Treatment Plants. Med. Fac. Landbouww. Univ. Gent, 61/4a, 1733-1743.
Yuan, z., Bogaert, H., Vanrolleghem, p., Thoeye, c., Vansteenkiste, G. and Verstraete, w. (1997). Control of
External Carbon Addition to Predenitrifying Systems. J. Envừ. Engrg. 123(11): 1080-1086.
Yuan, z., Bogaert, H., Vansteenkiste, G. and Verstraete w. (1998). Sludge Storage for Countering Ammonia
Shock Loads and Toxicity Incidents. Wat. Sci. Tech. 37(12): 173-180.
Yuan, z., Bogaert, H., Devisscher, M., Vanrolleghem, p. and Verstraete, w. (1999). On-line Estimation of the
Maximum specific Growth Rate of Nitrifiers in an Activated Sludge System. Biotechnology and
Bioengineering 65: 265-273.
Yuan, z., Bogaert, H., Leten, J. and Verstraete, w. (2000). Reducing the Size of a Nitrogen Removal Activated
Sludge Plant by Shortening the Retention Time of Inert Solids via Sludge Storage. Wat. Res. 34(2): 539-
549.
Zeghal, s., Puznava, N. Subra, J. p. Sauvegrain, p. and Vignoles, c. (1997). Methanol Dosing Feedback Control
for Denitrification. Biological Nutrient Removal 3, Brisbane, Austalia.
Zeghal, s., and Puznava, N., Subra, J. and Sauvegrain, p. (1998). Process Control for Nutrients Removal Using
Lamella Sedimentation and Floating Media Filtration. Wat, Sci. Tech. 38(3): 227-235.
Zhang, M., Tay, J., Qian, Y. and Gu, X. (1998). Coke Plank Wastewater Treatment by Fixed Biofilm System for
COD and NH3-N Removal. Wat. Res. 32(2): 519-527.
Zhao, H., Isaacs, s., Soeberg, H. and Kummel, M. (1994). A Novel Control Strategy for Improved Nitrogen
Removal in an Alternating Activated Sludge Process - PART I. Process Analysis. Wat. Res. 28(3): 521534.
Zhao, H., Isaacs, s., Soeberg, H. and Kummel, M. (1994). A Novel Control Strategy for Improved Nitrogen
Removal in an Alternating Activated Sludge Process - PART II. Control Development. Wat. Res. 28(3):
535-542.
Zipper, T., Fleischmann, N. and Haberl, R. (1998). Development of a New System for Control and Optimisation
of Small Wastewater Plants Using Oxidation-Reduction-Potential. Wat. Sci. Tech. 38(3): 307-314.
227
PART 2
WASTE GAS BIOFILTRATION
PERFORMANCE AND CHARACTERISATION OF A MEMBRANE
BIOLOGICAL AIR FILTER FOR SPACE APPLICATIONS
JAAP VAN DER WAARDE1, ARJAN VAN DER WERF1, MAURICE

HENSSEN1, BERT GEURKINK1, KLAAS VAN DER MAREL1, PIET
PAUL2 AND MARC GENT2
‘Bioclear Environmental Biotechnology, Groningen, The Netherlands, Tel:
+31 505718455, FAX: +31 505717920, email: waarde@bioclear.nl
2
STORK Engineers and Contractors, Amsterdam, The Netherlands
Summary
A membrane Biological Ak Filter (BAF) is designed for the degradation of low

concenfrations of various organic contaminants in indoor ak. The BAF showed stable
performance during a 15 month test in which a near complete removal of most organic
volatile contaminants was observed. Even at exfremely low concentrations (unto a few
pg/m3) good biodegradation efficiencies are obtained. Molecular and physiological
methods to detect and identify bacteria showed that the biodegradation process is
concentrated in the biofilm and that a mixed bacterial population is present growing on
all added organic components.
1. Introduction
Environmental quality assurance and recycling of raw materials are essential elements in
manned spacecraft missions. Indoor ak quality plays an important role, since the limited
amount of ak in the spacecraft is continuously recycled. Both astronaut activities and
materials onboard the spacecraft are sources of ak contamination. A wide range of
volatile organic contaminants has been observed in these closed envkonments, including
aliphatic hydrocarbons, alcohols, aldehydes, aromatic hydrocarbons and chlorinated
aliphatic components. Most of these components are biodegradable but the
concentrations are usually low (mg/m3). A biological system to remove these components
needs to meet several criteria: active at low concentrations; combined removal of a wide
range of organic contaminants; removal to levels below the space maximum allowable
concentration (SMAC), usually below 1 mg/m 3; stable activity over long periods
(months); small volume and low weight; no contact between bacteria and astronauts. A
biological ak filter (BAF) (Fig. 1) has been designed to purify ak in manned spacecraft
and meets these criteria (Binot and Paul, 1989).
231
Jaap Van Der Waarde et al
contaminant.
nwibrw ự C02 ;
With pons .. .■ . water
waste products / \ nutnents

liquid loop
Fig. 1: Principle of the Biological Air Filter (BAF)
This system has shown efficient removal of aromatic components like toluene, xylenes
and chlorobenzene. Improved biodegradation of chlorobenzene was observed under
mixed substrate conditions (Keuning et al, 1991). Air from a model spacecraft training
unit during a training session was efficiently purified using the BAF. Removal of toluene,
isopropanol and acetone to levels well below the SMAC values was demonstrated (Binot
et al, 1994).
Finally, a model BAF system has been designed to test the effects of space conditions
(p-gravity) on biodegradation. Efficient removal of 1,2-dichloroethane has been
demonstrated under space conditions with this system (van der Waarde et al, 1997). In
this report the results are described from long term biodegradation studies in the BAF
with low concentrations of mixtures of contaminants.
2. Materials and methods
The following bacterial sfrains were used (growth substrate in brackets): Pseudomonas
marginalis GJ8 (p-xylene), Xanthobacter autotrophicus GJ10 (1,2-dichloroethane),
Pseudomonas fluorescens GJ31 (toluene), Pseudomonas putida GJ40 (chlorobenzene),
Ancylobacter aquaticus AD20 (1,2-dichloroethane), Pseudomonas fluorescens BC20
(hexane) and Pseudomonas putida BCG2 (p-xylene). All strains are envfronmental
isolates. Growth was maintained on mineral medium MMY supplemented with 5 IĨ1M
substrate (Oldenhuis et al, 1991). Maximum growth rates (p max) of the selected strains are
determined for various subsừates by adding 4 mM subsữate to a 1% inoculate and
232
Performance and characterisation of a membrane biological air filter for space applications
following the growth by measuring the optical density (at 450 nm). The BAF system
consists of a multilayer membrane cassette (12 Accurel membranes), with a volume of 50
ml (25 ml gas phase and 25 ml liquid phase) and a membrane area of 0.086 m 2. The BAF
was inoculated with a mixture of the selected bacterial strains and operated during 400
days. The BAF system was fed with a gas flow of 65 ml/min (HRT of 23 sec.). The liquid
loop (total volume 0.5 1) was recycled with a flow of 400 ml/h. Biofilter removal
efficiency was monitored by GC analysis of influent and effluent gas flow once to twice
per week. Low concentrations contaminants (< 2 mg/m 6) were determined by leading the
gas flow through activated carbon, followed by desorption with cs 2 and GC analysis.
Water samples for FISH (Fluorescent In Situ Hybridisation) analysis were fixed in 36%
formaldehyde (final concentration 5%), diluted to 50% ethanol and stored at -20°C until
analysis. Sample (20 pl) was air dried on Vectabond coated microscope slides and
washed in 50, 80 and 96% ethanol for 3 min each. FISH analysis was performed
according to a protocol adapted from Stahl et al. (1991). The probes EƯB338 (a general
probe directed against all Eubacteria) (Amann et al. 1990b), ALF (dfrected against the
oc-subclass of bacteria) (Manz et aZ.1992), BET (dừected against the p-subclass of
bacteria) (Manz et aZ.1992), GAM (directed against the y-subclass of bacteria) (Manz et
aZ.1992) and DEL (dừected against the S-subclass of bacteria) (Amann et al. 1990a)
were hybridised to separate samples in a final concenfration of 8 ng/pl each with
formamide at 45°c overnight. Slides were viewed with an Olympus fluorescence
microscope with Fluorplan too* objective and Olympus U-MNB filter set.
6 Results and discussion
The BAF research was focused on biodegradation of a mixture of volatile organic

components that were model components for contaminated afr in closed manned
spacecraft. The following components were selected: hexane (poorly water-soluble
aliphatic), benzene, toluene, ethylbenzene and xylenes (common, non-chlorinated
aromatic compound), chlorobenzene (chlorinated aromatic compound) and 1,2-
dichloroethane (1,2-DCE, chlorinated aliphatic compound).
These components were selected since they can form problems in air biofiltration and
chlorinated components may have an effect on pH levels in the system.
3.1. PHYSIOLOGICAL CHARACTERISATION
Several bacterial strains that had been isolated from envfronmental sources were tested
for thefr capacity to biodegrade these components and thefr maximum growth rates were
determined (Table 1). The specific maximum growth rates of the pure cultures on the
tested substrates ranged from 0.05 h1 (BC20 with hexane) to 0.56 h1 (GJ31 with toluene).
It is clear from these data that BTEX biodegradation can be performed by a range of
bacterial strains, but biodegradation of hexane or chlorobenzene is dependent on the
presence of a single strain, sfrain BC20 or GJ31 respectively.
233
Table 1 Physiological characterisation and maximum growth rates (h1) of bacterial strains used
Benzen Toluen Ethyl- p- Chlor 1,2- n-

e e benzen Xylene obenzene DCE Hexane
e
GJ8 - + - 0.266’ - - -
1
GJ10 - + + - - 0.087 -
1
GJ31 + 0.555 + - 0.224 - -
GJ40 0.308 0.440' 0.365 - - - -
1
AD20 - - - - - 0.077 -
BC20 - - - - - - 0.050’
1
BCG2 - + - 0.085 - - -
+: growth detected, no growth rate determined -: no growth detected

*: originally isolated on this substrate.
Table 2 Biofilter removal efficiency and elimination capacity (EC)

Component Ingoin B AF B AF BAF performance
g performance (gas performance (gas after 6 weeks
Conce flow = 65 ml/min) flow = 195 starvation (gas flow =
n- ml/min) 65 ml/min)
tration
[mg/m3 Eff. EC Eff. EC Eff. EC
] [%] [g/m3.h] [%] [g/m3.h] [%] [g/m3.h]
1,2-DCE 70-200 30-40 2-5 50-60 22-26 100 4-7

Benzene 30-80 75- 2-6 60-80 22-27 100 2-4
100
Toluene 100- 90- 6-11 70- 19-22 100 2-14
190 100 100
Ethylbenzene 30-120 90- 3-9 80- 19-23 100 1-5
100 100
p-Xylene 40-80 90- 2-7 90- 13-15 100 1-5
100 100
Chlorobenzene 90-130 90- 4-10 90- 24-26 100 6-10
100 100
Hexane 40-200 0-10 0.5-3 4-10 10-
0.5-1.5 0.5-1.5
20
Eff. : Removal efficiency = (1-(effluent concentration / influent concentration))* 100% EC : Elimination
capacity in g component/(m3 reactor*hour)
3.2. PERFORMANCE OF THE BIOLOGICAL AIR FILTER (BAF)
After inoculation of the BAF at bacterial densities of 10 4/ml for each bacterial sfrain, an
adaptation period was observed in which the elimination capacity improved from 50% to
90-100% for most components. Biofilter performance was stable after 130 days of
operation, hexane was poorly removed by the biofilter (Table 2). In this period the
ingoing concentrations fluctuated between 20 and 200 mg/m3. Biofilter capacities are
relatively low due to the low influent concentrations and low gas flow.
Raising the load of the BAF for two weeks by a threefold increase of the gas flow
234
results in an increase in the elimination capacity. The effect of starvation on BAF

performance was determined by switching the gas flow off for 6 weeks. After feeding the
system again the removal efficiency for most components is near 100%, hexane is poorly
removed (Table 2). This indicates that the biofilter retains its activity and biodegradation
capacity throughout the 6 weeks starvation period and completely removes most of the
organic contaminants upon re-establishing the feed to the system.
3.3. MOLECULAR CHARACTERISATION
A substrate depletion test was performed with a sample from the liquid loop of the BAF,
using chlorobenzene and benzene as substrate. It was found that biodegradation occurs
after a lag phase of 8 hours for chlorobenzene and 5 hours for benzene (data not shown),
clearly indicating that the bacterial cells in the liquid phase are not adapted to active
biodegradation. Samples from both the substrate depletion test and the liquid loop of the
BAF were analysed for the presence of bacteria using FISH analyses. Group specific
probes were used to discriminate between the used bacterial sfrains. Cell activity was
based on the brightness of the FISH signal. The sample from the liquid loop does contain
bacterial cells but most cells are inactive as evidenced by a low EUB signal (Table 3).
Table 3. FISH analyses of bacteria in samples from the liquid loop and a substrate depletion test
EƯB338 ALF BET GAM DEL
Liquid loop + (20%) - ± (10%) + (5%) -

Substrate depletion ++ (70%) - + (20%) ++ (50%) -
test
no signal;
+
+
poor signal clear signal strong signal
between the brackets percentage of the cells that responds to FISH analysis
Bacteria from the AD20 type are not present or active, since no ALF positive cells were
detected. Sfrain GJ10 is probably present, since the probe BET shows a positive
response. A sample from the substrate depletion test shows much higher numbers of cells
and more cells are active than a sample from the liquid loop of the BAF. This indicates
that biodegradation in the B AF module is caused by the bacteria in the biofilm on the
membrane, and that bacteria in the liquid phase are poorly active and do not significantly
contribute to the biodegradation process.
Six of the seven inoculated strains hybridise to the EUB probe, only strain GJ8 could
not be detected. ALF hybridises with strain AD20, BET with strain GJ 10 and GAM with
strains GJ31, GJ40, BC20 and BCG2. The DEL probe hybridises with none of the
inoculated strains (data not shown).
3.4. PERFORMANCE OF THE BAF AT EXTREMELY LOW CONCENTRATIONS

OF VOLATILE CONTAMINANTS
Extremely low concentrations of volatile contaminants were dosed to the system to

determine the aừ purification efficiency at concentrations that are representative for
indoor air in manned spacecraft. The removal of these components was determined once
235
a week during 16 weeks.

Table 4. Biofilter performance at extremely low concentrations (range in 16 measurements)
Component Influent concenttation range [|xg/m3] Removal efficiency [%]

Toluene 2-46 40-100
Xylenes 5-57 16-100
Chlorobenzene 5-90 90-100
1-Octene <100 14-100

n-Butylacetate <70 96-100
Limonene <114 80-100
From these experiments it is clear that extremely low concentrations (pg/m 3) of volatile
organic components can efficiently be removed by the BAF. Good removal was found
during prolonged periods for limonene, toluene, 1-octene, butylacetate, chlorobenzene
and xylenes. The large variation in removal efficiencies is caused by inaccuracies in the
dosing system and in the analysing method at these low concentrations. Limonene, 1-
octene and butylacetate were dosed to the system without addition of new bacterial
strains. This implies that the biofilm in the BAF is able to quickly adapt to new
contaminants in the air.
4. Conclusion
The biological aừ filter (BAF) has shown to be effective in removing low concentrations
of volatile contaminants from air. Removal capacities range between 2 and 26 g/m 3.h at
ingoing concentrations of 20 to 200 mg/m 3 and at retention times of 823 seconds. Even at
extremely low concentrations (up to concentrations of a few Lig/m 3) good biodegradation
efficiencies are obtained.
Molecular and physiological methods to detect and identify bacteria show that the
biodegradation process is concentrated in the biofilm and that a mixed bacterial
population is present growing on all added organic components.
References
Amann, R.L, B.J. Binder, R.J. Olson, s.w. Chrisholm, R. Devereux and D.A. Stahl. (1990a) Combination of 16S
rRNA-targeted oligonucleotide probes with flow cytometry for analysing mixed microbial populations.
Appl. Env. Microbiol. 56, 1919-1925.
Amann, R.I., L. Krumholz and D.A. Stahl. (1990b) Fluorescent-oligonucleotide probing of whole cells for
determinative, phylogenetic, and envừonmental studies in microbiology. J. Bacteriol. 172,762-770.
Binot, R.A. and P.G. Paul. (1989) BAF - an advanced ecological concept for aứ quality conttol. SAE technical
paper series no. 891535.
Binot, R.A., R.J. Breukers, P.G. Paul and D. Jager. (1994) BAF-EXEMSF92: Testing of the biological air filter
for aứ quality control during a manned space mission simulation. SAE technical paper series no. 941343.
Keuning, s., D. Jager, P.G. Paul, and R.A. Binot. (1991) Biodegradation study with space cabin contaminants to
determine the feasibility of Biological Air Filtration (BAF) in space cabins. ESA SP-324.
Manz, w., R. Amann, w. Ludwig, M. Wagner and K.-H. Schleifer. (1992) Phylogenetic oligodeoxynucleotide
236
probes for the major subclasses of proteobacteria: problems and solutions. Syst. Appl. Microbiol. 15,593-
600.
Oldenhuis, R., J.Y. Oedzes, J.J. van der Waarde and D.B. Janssen. (1991) Kinetics of chlorinated hydrocarbon
degradation by Methylosinus trichosporium OB3b and toxicity of trichloroethylene. Appl. Envữon.
Microbiol. 57, 7-14.
Stahl, D.A. and R. Amann. (1991) Development and application of nucleic acid probes. Nucleic acid techniques
in bacterial systematics. Eds. E. Stackebrandt and M. Goodfellow. Wiley Ltd, New York, 205248.
Waarde, J.J. van der, H. Doienbos, E. Dijkhuis, s. Keuning, P.G. Paul, c. Klabbers and R.A. Binot. (1997)
Determination of the space influence on the kinetics for biodegradation of organic volatile contaminants.
ESA SP-400.
Wagner, M., R. Amann, H. Lemmer, and K.H. Schleifer. (1993) Probing activated sludge with oligonucleotides
specific for proteobacteria: inadequacy of culture-dependent methods for describing microbial community
structure. Appl. Environ. Microbiol. 59, 1520-1525.
237
BIOFILTRATION FOR WASTE GAS HANDLING
FREDERIC THALASSO1, MARIA c. VEIGA2 AND CHRISTIAN

KENNES2
department of Biotechnology, CINVESTAV, Av. IPN 2508,
CP 07000 Mexico City, Mexico 2University of La Coruna, Chemical
Engineering Laboratory, Campus da Zapateira, E-15071-La Coruna,
Spain. Fax: 34981167065. e-mail: Kennes@udc.es
1. Introduction
Facing the increasing concern about the envứonment, biological gas cleaning has been
developed since the 1920's [1, cited by 2]. Nevertheless, it is only since a few decades ago
that biological gas cleaning is being accepted as a competitive alternative to the more
conventional physico-chemical treatment technologies. Nowadays, it is commonly used
for the cleaning of a wide variety of gaseous pollutants. Since the beginning, three main
groups of biological gas cleaning technologies were considered: biofiltration,
bioscrubbing and trickling filtration. These technologies differ by the presence or absence
of a carrier material and of a mobile liquid phase (Table 1). Trickling filters and
bioscrubbers are quite similar concerning the presence of a mobile liquid phase serving as
nutrient source for the microorganisms. Conversely, biofiltration is characterized by the
use of an organic carrier ensuring the nutrient supply and by the absence of a mobile
liquid phase or by the use of an inert carrier but with intermittent supply of nutrients. This
chapter presents the main characteristics of these three technologies although it mainly
focuses on biofiltration with organic carriers, which is the most largely used process [3].
Table 1: Characteristics of the three main biological gas-cleaning technologies.

2. Bioscrubbing
Reactor_______ Mobile phases Carrier Active biomass
In bioscrubbing the pollutant
Bioscrubbers is first
Liquid and gas absorbed in a liquid phase, which is then treated in a
Dispersed
Tricklingstage
second filters in a bioreactor
Liquid(Fig.
and gas
1). The mainInert
advantages of this technology
Fixed are: (i)
Biofilters Gas Organic/Inert Fixed
239
Frederic Thalasso, Maria c. Veiga and Christian Kennes
the wash-out of the reaction products, avoiding their potential inhibitory effects, (ii) an
easy conttol of the biological process thanks to the control of the composition of the liquid
medium, and (iii) a good adaptation capacity of the microbial biomass to the composition
of the gas to be cleaned.
Treated air
Fig. 1: Bioscrubber.
The major drawback of this technology is the requirement to dissolve the gaseous
pollutants in an aqueous phase, which generates problems of gas transfer. Bioscrubbing is
therefore of interest for gaseous pollutants having a low air/water partition coefficient.
This is of major importance since most of the target pollutants are volatile and poorly
water-soluble. According to Kok as well as van Groenestijn and Hesselink [4, 5],
bioscrubbing is of interest for the treatment of pollutants having an air/water partition
coefficient lower than 0.01. This is probably why bioscrubbing is less in focus than
biofilttation, although several examples of successful applications have been reported [6,
7] and the interest of bioscrubbing has been pointed out for the treatment of several
Volatile Organic Compounds (VOC) [8]. Furthermore, recent developments [9, 10]
foresee a new interest in this technology, among others because the utilisation of
bioscrubbing for biological desulphurisation of very large gas flow rates (up to 2 10 6 m7
per hour) seems to be feasible.
Treated aừ
7 Trickling filters
The trickling filtration technology consists in the use of an inert carrier on which a biofilm
grows. The polluted gas passes through the carrier material, co- or counter- currently to
the mobile liquid phase, which ensures the nutrient supply to the microorganisms (Fig. 2).
In some cases, the liquid supply is limited and the reactor design and operation are similar
to organic-biofilters. For that reason, the border between the conventional biofilter and the
trickling filter is relatively narrow. Several authors have used different names for
sometimes very similar systems: biofilter [11, 12], trickling biofilter [5], biotrickling filter
[13, 14] or trickle bed aừ biofilter (TBAB) [15, 16].
240
Biofilttation for waste gas handling
Fig. 2: Trickling filter
The main carriers used are plastic or ceramic structured packing [17, 18, 19, 20, 21],
Celite [22, 23], perlite [24], vermiculite [25] activated carbon [26, 27, 28], polyurethane
foam [29, 30], lava stone [31, 32] or mixtures of them [33].
Both co- and counter-current operation have been used. The counter-current operation
has the advantage of a more uniform activity and biomass growth but suffers from higher
pressure drops [34].
This technology presents similar advantages as bioscrubbing: (i) wash-out of the
reaction products, (ii) easy control of the biological process and (iii) good adaptation
capacity of the active biomass. Like bioscrubbing, the major drawback of this technology
is the problem of gas transfer resulting from the need to dissolve the gaseous pollutants in
an aqueous phase. According to Kok and van Groenestijn and Hesselink [4, 5], trickling
filters are of interest for the treatment of compounds having an air/water partition
coefficient lower than 0.1. Concerning that aspect, different publications [20, 26, 35, 36]
have pointed out the interest to restrain the liquid supply to the minimal microbial needs
obtaining better gas treatment efficiencies with a limited liquid supply. On the other hand,
this limitation of liquid supply can lead to a decrease of the wetted area of the filter
carrier, which can be approximated to the active area [18]. This sfresses on the importance
of a careful carrier design and of the homogeneity of the liquid supply.
Another drawback, specific to trickling filters, is the potential excessive biofilm
growth on the carrier material. This microbial growth progressively reduces the empty
volume of the carrier and results in an increase of the pressure drops. This well described
phenomenon [2, 37, 38], can lead to the complete clogging of the filter-bed [39]. Although
not systematically observed [17], it reinforces the importance of careful carrier design.
Methods have been developed to restrain clogging problems, either by limiting microbial
growth [21, 22], by regular filter-bed washing [22, 40, 41, 42, 36] or by reducing the
liquid supply [35]. An extensive review of the characteristics and efficiency of all such
methods has been published very recently [43].
241
Until the late nineties, trickling filters were indubitably less in focus than biofilters.
According to a literature overview, about 70% of the papers, presented results obtained
with filter beds composed mainly of organic material. After 1998, a clear change was
observed and nowadays, inorganic carriers are used in about half of the lab-scale
biofilters.
4. Biofiltration with organic packing materials

In conventional biofilters, the polluted gas stream flows through a filter-bed generally
composed of organic matter (peat, compost, sawdust, etc.) and serving both as a support
for the active biomass and as a reservofr of nutrients for microorganisms (Fig. 3). A very
important characteristic of this process is the absence of a mobile liquid phase. Thanks to
this characteristic, biofilters are very efficient for the treatment of poorly water-soluble
pollutants. According to Kok [4] and van Groenestijn and Hesselink [5], biofiltration is of
interest for the treatment of pollutants having an air/water partition coefficient lower than
1. Several examples of successful applications have appeared in the literature [44, 45, 46,
47] and nowadays some industrial plants can treat up to 200,000 m3 per hour.
The nature of the filter-beds is quite variable. According to Clark and Wnorowski [48],
almost any organic material presenting a “satisfactory structure and composition” could be
used. Thirteen important characteristics of good biofilter media have been listed by Bohn
[49], including physical, chemical and biological parameters. The most important physical
characteristics are: (i) a high surface area, for an optimum microbial development, (ii) a
low bulk density for an easy and cheap carrier operation and (iii) a high void fraction to
limit pressure drop and clogging problems. In addition to these physical characteristics,
the natural presence of a large number of different bacteria and fungi in the carrier as well
as a balanced chemical composition are of major importance in order to favour microbial
adaptation and activity inside the biofilter-bed.
Water /Nutrient
—Q—<— Water
Organic r\ Gas humidifier
Carrier
Polluted air
Treated aừ
Fig. 3: Biofilter, design and control.
Peat, soil and compost are the most commonly used filter-bed materials. Although peat
and soil are often well characterized, they are sometimes relatively complex mixtures. The
word “compost” derives from the Latin “compositum”, meaning mixture and refers to
decaying organic matter. Many raw materials may be used as compost: yard ttash [50],
digested sewage sludge [50, 51], forest sub-products [51, 52], wastewater “biosolids” [52],
242
manure [53], or the organic fraction of municipal waste [54]. Many other carrier materials
are mentioned in the literature, among others, bark, saw dust or dried wastewater sludge as
shown in table 2. A new tendency consists in the use of specially designed nutrient beads
for the immobilization of the active biomass [55].
The carrier materials are ever more frequently prepared by mixing organic and
inorganic matter. Approximately 85 % of the papers published in the early eighties
concerned pure organic matter while over the last five years this percentage declined to
less than fifty percent The main objective of incorporating an inorganic material, such as
perlite [33, 52, 56], glass beads [57], polyurethane foam [33] or polystyrene [58, 59], into
the filter-bed is to reduce the pressure drop and to limit channelling problems. Sometimes,
activated carbon and buffer compounds are also incorporated to minimize the fluctuation
of the pollutant concentration [60, 61] or to stabilize the pH of the filterbed [52, 56, 62,
63].
The filter-beds are simple or multi-stage and may reach up to four stages [33, 50, 51,
52]. According to Ottengraf et al. [64], “the treatment of a waste gas containing different
biodegradable and xenobiotics compounds may advantageously take place in a multi-
staged filter-bed”. This allows the presence of different envfronmental conditions in a
same biofilter and allows to optimise the degradation of each of the compounds to be
removed. Usually, the height of filter beds is between 0.5 and 1 meter but they can reach
up to 2 or 3 meters high in some cases [7, 65, 66].
Since organic carriers provide nutrients necessary to the microorganisms, there is no
need for any external nutrient supply initially. Bohn [67] underlines that the nutrient
supplying power of the filter media are usually high enough for the pollutant loads
generally applied. This is confirmed, for instance by Cardenas-Gonzalez et al. [68] who
found no significant nittogen and total carbon decrease in a compost biofilter after 5 years
operation. Nevertheless, Dalouche et al. [70] underline the lack of phosphorus in peat and
Morgenroth et al. [51] reported a problem of nutrient limitation during hexane degradation
in a compost biofilter. To limit such a risk, some authors adopted the alternative to
discontinuously spray a nutrient solution on top of the filter-bed [12, 57]. Even
temporarily using tap water seems to be suitable to ensure high biodegradation efficiencies
[71]. Although nutrients are available in organic filter beds, the additional intermittent
supply of nufrients to the reactor allows increasing biofilter performance and long-term
stability [72].
243
Table 2: Main materials used as biofiltration carriers (updated from [124]).

Carrier Volume ratio Reference
Compost 1 11, 12,73,74,75,76

Compost - yard trash 1/3, - 61,77
Compost - Wood chips -, 3/7 78,79
Compost - perlite 1/1 52,51,80,81
Compost - diatomaceous earth - 82
Compost - polystyrene 1/1 58, 83
Compost - chaff 1/1 50
Compost - manure - 69
Pellets: compost + Inorganic 2/1 84
polymer
Peat 1 85, 86, 87, 88, 89
Peat balls 1 90
Peat - polyurethane foam 7/3 33
Peat - polyurethane foam - 2/2/1 33
vermiculite
Peat - polyurethane foam - perlite 2/3/5 33
Peat - perlite 2/3 33, 56, 91
Peat - pine bark - 90
Peat - compost 1/1 92
Peat - glass beads 2/1, 4/1 57, 93
Wood bark 1 73, 94
Wood waste - perlite 1/1 95
Soil 1 7,44, 47, 82, 96
Soil - sand - peat - compost 20/2/3/3 97
Seaweeds 1 70
Sawdust - manure - 53
Dry wastewater sludge 1 98
One of the most significant parameters of the filter-bed is its moisture content, which is of
major importance for the microbial activity. Ensuring a high enough moisture level is
important, because of the high sensitivity of the microorganisms to the water activity [99,
100]. However, too high a moisture content may generate serious problems such as the
formation of stagnant zones with diffusion limitation and possible anaerobic conditions
[75] or increased pressure drops [94, 101]. As clearly presented by Wang et al. [12] the
maximum degradation capacity of the biofilter is obtained at a moisture content of about
35% when compost is used as filter-bed and with a moisture content of about 40% when
peat is used. Wang et al. [12] also underline that a drying period could provoke
irreversible effects by modifying the structure of the filter-bed, which was also confirmed
by Bohn [67]. An adequate average value seems to be in the range of 30 to 80% for peat,
compost and wood sub-products [7, 47]. For soil-beds, the optimal moisture content seems
to be in a 10-20 % range [47, 65, 96, 102]. Bohn and Bohn recently published an
extensive review on moisture in biofilters [103].
The control of the filter-bed moisture content is currently done either by means of a
spray system dừectly spraying water on the filter-bed, and/or by indừectly controlling its
moisture content by means of the humidity of the polluted gas fed to the biofilter. In the
case of dừect spraying, Van Lith et al. [101] underlined the importance of a small droplet
size in order to reach a near homogenous liquid distribution and to limit the mechanical
impact of the droplets on the carrier material. Sprayers commercially available can
produce droplets of less than 40 10’6 m diameter [20, 35]. Suchlike small droplets present
also the advantage of an extremely high gas/liquid transfer area (above 150 m 2 per liter of
liquid injected) and therefore of a quick liquid saturation. It is important to underline that
244
some authors do not specify if the humidity of theừ filterbed is given on a dry [12, 54] or
on a wet [50, 51, 52, 57, 58, 104] basis. Although both are used in the literature, it is likely
that the majority of the results are given on a wet basis.
Obviously, temperature is also an important parameter. Although microbial activity is
possible in a 0 - 95°c range, the optimal temperature for the filter-beds corresponds to the
mesophilic temperature, i.e. about 30°C [12, 65, 95, 105]. Some examples indicate that the
temperature may be lower without any significant microbial deactivation in the range of
10 to 20°C [85, 106], or even at 2°c as presented by Ebinger et al. [102] for propane
treatment in a soil-bed system, or as observed by Elsgaard [107] in a biofilter treating
ethylene. In this context, it is important to stress that a temporary biological deactivation
of the filter bed does not necessarily result in the complete loss of the treatment capacity,
because of the existence of adsorption phenomena in some filterbeds. Consequently,
biofilters can cope with temporary and relatively significant temperature decreases, even
to temperatures below o°c [108]. Additionally, some examples of biofiltration at
thermophilic temperatures have recently been presented [109, 110]. Finally, it should be
recalled that biodegradation processes are exothermic and may lead to significant
increases of the filter-bed temperature [111, cited by 74]. Obviously, such temperature
increases result in a decrease of the relative humidity of the gas phase, which can provoke
a drying effect of the filter-bed [112] especially if channels are present.
The pH of the filter-bed is also an important parameter. A good carrier material should
preferably have a near neutral pH [2, 113] and must present a good buffering capacity
[49]. This is of major importance since numerous processes generate acidic, basic or toxic
products such as hydrochloric acid during the biodegradation of chlorinated compounds
[44], sulphuric acid during biodegradation of hydrogen sulphide [7, 85, 78] or methyl
sulphide [54] and basic compounds during ammonia elimination [70]. Devinny and Hodge
[27] present results of acetaldehyde and acetic acid generation during ethanol degradation
when applying overloads. Under non-regulated pH conditions, the elimination capacity
usually drops concomitantly with pH decrease, as observed by Bronnenmeier et al [114]
and by Hartmans and Tramper [115].
According to Bohn [49], soil presents a higher buffering capacity than compost which,
according to Smet et al. [54] is five times more buffered than wood bark. On the other
hand peat is characterized by its naturally low pH, around 3 to 4 [70, 87]. Therefore, a pre-
treatment of the filter-bed is sometimes a prerequisite in order to increase its pH value or
to improve its buffering capacity. Several compounds are used to control the pH of the
filter-bed: lime [75, 105, 113] and powdered oyster shell [51, 52, 80, 81] mixed with the
filter-bed during the pre-treatment or Ca(OH) 2 [88, 116] NaHPO4, NaHCO3 [50] and
NaOH [57] either mixed with the filter-bed or sprayed on top of it during the process.
Peculiar cases of pH resistance have been reported by Windsperger et al. and by Yang
and Allen [77, 117] with an optimum H 2S degradation rate obtained at pH 3.2, and by
Shinabe et al. [19] observing no inhibition of H 2S degradation at pH 1 in a trickling filter.
During batch studies of H2S removal by Thiobacillus thiooxidans Shinabe et al. [19] found
an optimal pH of 2.5. It is important to stress that during hydrogen sulphide degradation,
inorganic sulphate compounds do accumulate in the filter-bed and can reach up to 10% by
weight [112]. Similarly, the formation of up to 100 mole nitrites and nitrates per m 3 carrier
during ammonia degradation has been reported [118, 119].
The pressure drop created by the gas phase passing through the filter-bed can represent a
non-negligible part of the treatment costs. The pressure drop depends on numerous
factors, mainly the nature of the filter-bed and its moisture content. Soil induces the
245
highest pressure drop, followed by compost, peat and finally by wood bark [2, 75, 94, 120,
121]. Obviously, the size of the particles forming the filter-bed is of major importance
since for a same filter-bed material, small particles can provoke a pressure drop more than
10 times higher than observed with large particles [112]. The effect of the moisture
content in organic-biofilters has been shown, among others, by Shoda [7] and by Van
Langenhove et al. [94] and can result in a pressure drop increase of up to 100 %. This
once again indicates the importance of a good moisture control.
5. Applications
5.1. GENERAL ASPECTS
Biofiltration of waste gases was originally applied to a rather limited range of sources and
compounds, mainly odours usually quantified by means of Odour Units (OU) rather than
through the identification of specific compounds.
Over the past two decades, the technology has been extended to a much wider variety
of sources and pollutants. Nowadays, more than thousands gas-phase bioreactors are being
operated in Europe. In Germany and neighbouring countries, odours emitted from
wastewater treatment plants are solved with biofilters in almost two third of the cases [3].
A non-exhaustive list of pollutants biodegradable in conventional and trickling biofilters is
given in table 3. Theoretically, any pollutant biodegradable under aerobic conditions could
be degraded in gas-phase biofilters. Even compounds requiring anaerobic environments
for their biological removal have been degraded in gas-phase lab-scale biofilters.
Perchloroethylene (PCE), which needs strict anaerobic conditions for biodegradation
[122] appeared to be biodegradable in a biofilter fed with PCE- polluted air [123]. The
removal of hardly biodegradable contaminants in biofilters may require the inoculation of
selected specialized microorganisms [124]. For the treatment of waste gases polluted with
non-recalcitrant compounds in biofilters packed with natural organic carriers, inoculation
is often not necessary since indigenous microorganisms will rapidly grow and adapt to the
contaminants. However, seeding the reactor usually allows to shorten the start-up phase
[125], although inoculated strains may finally be overgrown by indigenous strains [124].
246
Table 3. List of some pollutants for which biodegradation in gas-phase biofilters has been shown.
Dioxane Methylmethacrylate
Ethane Methyl-Tertiary-Butyl-Ether
(MTBE)
Ethanol Nitric oxide
Ethene Nitrobenzene
Ethylacetate Phenol
Ethylaldehyde a-pinene
Acetaldehyde Acetone Ethylbenzene Propane
Ethylene Propanol
Acetonitrile Acetylene Formaldehyde Propionaldehyde
Acrylonitrile Ammonia Heptane Styrene
Benzene 1,3-Butadiene Hexane Terpenes
Butanal Butanol Butyl Hydrogen sulphide Tetrachloroethene
acetate Butyric acid Isobutylacetate Thiocyanates
Carbon monoxide Isopentane Toluene
Carbon disulphide Isopropanol Trichloroethane
Chlorobenzenes Cresol Methane Trichloroethene
Cyclohexane Methanethiol Trichloromethane
Dichloroethane Methanol Triethylamine
Dichloromethane Methoxypropylacetate Trimethylamine
Diethylether Methyl-Ethyl-Ketone (MEK) Vinyl acetate
Dimethylsulphide Methyl-Iso-Butyl-Ketone Vinyl chloride
Dimethyldisulphide (MIBK) Xylenes
Methylmercaptan
In the first studies undertaken and published on biofihration, bioreactor performances
were relatively poor and rather low elimination capacities of only a few grams per cubic
meter per hour were usually reported which was, however, acceptable for full-scale
applications at wastewater treatment plants or composting facilities. The improvement of
bioreactor design and a better understanding of the phenomena affecting theứ performance
have allowed reaching much higher elimination capacities together with high removal
efficiencies (Table 4). Thus, nowadays biofiltration may be used in a much wider range of
sectors and applications. Although it has recently been shown to be suitable for the
treatment of high gas flow rates and high loads, the treatment of low pollutant
concentrations released at low flow rates, i.e. low loads, still remains a major application
field of biofitration. High efficiencies have often been obtained in lab-scale experiments
(Table 4). Nevertheless, it is important to point out that in some of the reported lab-scale
studies high elimination capacities were maintained only for short periods of time, while
full-scale systems must operate under stable conditions for long periods of at least several
years. It should also be mentioned that in lab-scale experiments, a few authors report high
maximal elimination capacities corresponding, however, to low removal efficiencies. At
industrial scale, such situation is generally not viable, and whenever high loads are to be
tteated, relatively high removal efficiencies are required in order to avoid or minimize air
pollution.
247
Table 4: Some typical recent examples of relatively high elimination capacities (EC) and removal efficiencies (RE)
reached in biofilters and biotrickling filters for typical Volatile Inorganic and Organic Compounds (VICs and
VOCs) (more examples are available in [124]).
Packing Compound Max. EC (g/m3.h) 1 Ref.
Coưesponding RE (%)
BIOFILTERS:
Fuyolite Ammonia 22.8(*)/>99 [126
Compost/dolomite Dimethylsulphide 70/85 ] [127
Perlite Ethylbenzene 85/99 ] [24]
Compost/Hog fiiel/Perlite Hydrogen sulphide 120/30 [128
Compost/Glass beads Phenol 700/25 ] [57]
Compost/Wood chips a -Pinene + Methanol 40+ 250/>90 (each) [129
Perlite Styrene 80/99 [130
Perlite Toluene 73/99 [24]
Perlite Toluene + Ethylbenzene + 120 (total)/>99 [125
Ơ-Xylene (1:1:1) ]
Perlite Ơ-Xylene 64/99 [24]
BIOTRICKLING FILTERS:
Coal particles Acrylonitrile 465/95 [131
Activated carbon Hydrogen sulphide 142/90 ] [132
Coal particles Pentane + styrene 20 + 55/80 [133
Sintered glass Propionaldehyde 500/>80 [134
Pall Rings Toluene 83/35 [135
*gx/kg dry packing . d (addition of nutrients 4xlh/day) ]
5.2. INDUSTRIAL SCALE APPLICATIONS
Although the use of conventional biofilters with organic filter-beds remains the most
classical and most widely accepted alternative, a few pilot-scale and industrial-scale
bioreactors packed with inert carriers have recently been installed as well. Inert carrier
materials are in most cases packed in biotrickling filters. However, recent studies have
proven that the intermittent (weekly or even monthly) addition of a nutritive aqueous
phase to conventional biofilters packed with inert carriers may result in a very high reactor
performance [136]. Full-scale systems are already being operated under such conditions.
Table 5 shows a list of compositions of some gaseous effluents successfully tteated in
pilot- or full-scale biofilters. Over the past decade the number of applications increased
fast and it would be difficult to cite them all. Most examples published in the literature and
some non-published full-scale applications indicate that bioreactor technologies are
suitable above all when dealing with waste gases containing either single compounds or
simple mixtures of pollutants. However, good results have sometimes also been obtained
in the case of more complex mixtures as, for example, in the biofiltration of petroleum
hydrocarbons released to the atmosphere from soil vapour exttaction processes. The
treatment of mixtures of pollutants usually provokes a decrease of the biodegradation rates
of each compound of the mixture in comparison with the biodegradation rates of
individual compounds [124]. However, sometimes very similar elimination capacities may
be reached for pollutants fed individually to a biofilter or in mixture. This was the case for
a waste gas containing toluene, ethylbenzene and ơ-xylene [24]. Still in other situations,
the presence of a given pollutant may stimulate the removal of another one in case of co-
metabolic processes. The most recalcitrant pollutant will limit the overall removal rate,
meaning that the biofilter will be oversized for the most easily biodegradable pollutants.
248
Regarding the application range of full-scale biofilters, several parameters must be

considered. The flow rates may range from relatively low values of a few hundreds cubic
meters per hour [137] up to reported values of more than 200,000 m 3/h [138]. Some
projects on the treatment of waste gas flow rates greater than 1,000,000 m 3/h have been
studied [124]. Pilot-scale and industrial-scale biofilters treating flow rates below hundred
cubic meters per hour have been described [139, 140]. However, this is less usual.
Volumetric biofilter loading rates are between approximately 5 and 600 m 3/m3.h [124].
Higher values are possible and have been published for bioscrubbers; reaching 1,428
m3/m3.h in a bioscrubber treating a methanol polluted waste gas [5].
In most applications, total pollutant concentrations in the feed are of a few hundreds
mg per cubic meter but full-scale biofilters have also already been used to treat lower
concentrations, below 10 mg/m3, or higher concentrations of a few grams per cubic meter.
For the treatment of waste gases containing more than 3-5 g/m 3, biofiltration is normally
not suitable [124]. Wright et al. [141] describe the treatment of aừ contaminated with up
to 2.7 g/m3 of petroleum hydrocarbons in a pilot-scale compost biofilter and with
relatively high removal efficiencies. Several authors have shown that when feeding high
concentrations of pollutants, the removal efficiency will drastically decrease. Jorio et al.
[142] observed that feeding 6.2 g/m3 toluene or 8.2 g/m3 xylenes resulted in, respectively,
approximately 60% and 23% removal efficiencies at empty bed residence times (EBRT)
of 102 s for toluene and 78 s for the xylene isomers. The best results are reached when
combining low pollutant concentrations and relatively high gas flow rates. Biodegradation
of very low pollutant concentrations, of only a few ppbv, is sometimes also difficult and
may result in low removal efficiencies as well [81].
Although optimal values for the empty bed residence time (EBRT) are most often
between 20 s and one minute, above 90% removal efficiency has been reported in fullscale
applications at EBRT below 20 s and even below 10 s in some cases [2]. Lower loads, i.e.
lower pollutant concentrations, allow using lower EBRT. At lab-scale, EBRT of several
minutes have sometimes been used [143], but at industrial scale such high values are not
recommended. Nevertheless, EBRT between 6 and 22 minutes have been reported in
pilot- and full- scale biofilters at sites where polluted air from soil vapour extraction was
treated [139]. Highly recalcitrant VOCs requừe longer EBRT than easily biodegradable
compounds. The optimal EBRT will also depend on the selected filterbed.
The largest above-ground biofilters built for the treatment of waste gases are of the
order of 3000-4000 m3 [138], while the area necessary for typical open bed soil biofilters
may reach 3000 m2. If necessary, full-scale biofilters may be operated in series or in
parallel.
It has already been mentioned previously in this chapter that pressure drop is an important
parameter since it affects biofilter performance. At industrial scale, it is also very
important because pressure drop and costs will increase simultaneously. High pressure
drops may be reached after several years of operation mainly as a result of compaction
when using organic filter-beds or because of excess biomass accumulation on inert
packing materials. The presence of a liquid phase in trickling biofilters will also lead to
higher head losses and the need of more power of fans or blowers used for feeding the
waste gas to the reactor. Filter-bed replacement is generally requừed after 45 year
operation for organic carriers. Inert carriers may be used for longer periods of time
depending on the operating conditions.
Table 5. Examples of mixtures of pollutants biodegradable in biofilters
249
Source of contamination Pollutants
Coating industry Toluene, VOCs

Commercial bakery Ethanol, methane, ethylacetate, aliphatics, VOCs
Composting Odours, ammonia, hydrocarbons
Fibreglass industry Styrene, acetone
Flavour industry Odours, flavours
Flexographic printing Alcohols, acetone
Food industries Odours
Foundry Ethanol, VOCs
Foundry Phenol, ammonia
Lacquering industry Toluene, ethylbenzene, xylenes, butyl acetate
Latex production Styrene, butadiene
Metal foundry Benzene
Oil production Odours
Pharmaceutical industry Alcohols, acetone, dichloroethane
Plastic dashboard manufacturing Styrene, butylacetate
Plastic resins Styrene
Polluted soil Gasoline
Pulp and paper industry Mixture of sulphur compounds
Slaughterhouses Odours
Rayon manufacturing H2S, CS2
Tobacco industry
Wastewater ừeatment NH3, nicotine, odours
Wood industries Odours
Formaldehyde, phenol, methanol
6. Other biological gas treatment technologies
To be complete, other technologies described in the literature should be mentioned. The

use of membranes are for instance presented by Fischer, Hartmans et al., Reij et al. and
more recently by Parvatiyar et al. and Ergas et al. [144, 145, 146, 147, 148], an interesting
external loop airlift bioreactor was described by Ritchie and Hill [149], a new technology
using a double liquid phase was proposed by Cesário et al. [150, 151, 152], a biorotor
reactor was used by Buisman et al. [153] for the removal of hydrogen sulphide and an
horizontal flow biofilter has been proposed by Lee et al. [154]. A combination of
photochemical oxidation and biodegradation seems a promising alternative for some
specific applications [155]. Nevertheless, these technologies have only been tested on a
lab-scale or pilot-scale.
7. Conclusions
Biological gas cleaning is, in some aspects, very different from the main biological
cleaning technologies currently used for water and/or soil treatment. Although the
biological mechanisms involved are quite similar, the gaseous state of the pollutant to be
treated involves new engineering concepts and has led to a new application field of
envfronmental biotechnology.
The new engineering concepts induced by the development of biological gas treatment
processes comprises mainly (i) the possibility of handling quite high flow rates, with (ii)
very short contact times, in (iii) technically simple reactors that (iv) ensures a synergy
between absorption and degradation processes. Nowadays, it is possible to treat up to
200,000 m3 waste gas per hour with a contact time of about 30 seconds in reactors formed
only of an organic filter-bed and a spraying system as major features.
250
The new application field opened by biological gas cleaning is characterized by the
possibility to treat volatile and/or non-water soluble pollutants, even at relatively low
concentrations. Biological gas cleaning is therefore a technology that complements the
more traditional treatment processes. It can be used with pollutants dừectly emitted to the
atmosphere from the production source, and with pollutants, which have a significant
envfronmental impact even at low concentration.
As a first relatively recent and obvious application field of biological gas cleaning one
should mention the treatment of industrial off-gases. Biofilters or other bioreactors are
therefore currently used in, chemical, food and beverages or mechanical industries among
others (Table 5). Secondly, biological gas treatment may be used as a polishing step in
combination with other treatment technologies. For instance, it is employed in wastewater
treatment plants to avoid the emission of volatile pollutants transferred from the liquid to
the gas phase. Thirdly, biological gas treatment can be used for site remediation. Indeed,
combined to soil vapour extraction or stripping, biological gas treatment allows the
elimination of the volatile fraction of soil pollutants. So, the problems of handling large
amounts of pollutants in on-site soil remediation technologies or the problem of the low
microbial kinetics in in-situ soil remediation can partly be overcome. Finally, thanks to the
recent developments in the removal of metallic compounds and radionuclides, this gas
treatment technology can be applied in the remediation of sites polluted by non-organic
aừbome compounds.
In conclusion, after a few decades of development, biological gas treatment is now
seen as an increasingly valuable technology, which can play an important role in aừ
pollution control. It is a promising and significant tool, which assists mankind in cleaning
up our envừonment.
Acknowledgements
Part of this chapter has been prepared thanks to funds from project PPQ2001-0557 to CK
References
[1] Bach, H. (1924). Gesundheits-Ingenieur, 46, 370-376.
[2] Ottengraf, s.p.p. and Diks, R. (1992). Review paper: process technology of biotechniques. In Biotechniques
for air pollution abatement and odour control policies, A. J. Dragt and J. van Ham (Eds.), Elsevier,
Maastricht, The Netherlands, 17-32.
[3] Frechen, F.-B. (1994). Odour emissions of wastewater treatment plants-recent German experiences. Wat.
Sei. Technol., 30, 35-46.
[4] Kok, H., (1992). Bioscrubbing of air contaminated with high concentration of hydrocarbons. In
Biotechniques for air pollution abatement and odour control policies, A. J. Dragt and J. van Ham (Eds),
Elsevier, Maastricht,, 77-82.
[5] van Groenestijn, J.w. and Hesselink, P.G.M. (1993). Biotechniques for air pollution control. Biodegradation,
4, 283-301.
[6] Paul, E. (1987). Experiences with large-scale biological scrubber. In Biological treatment of industrial waste
gases, state of the art and comparison with physico-chemical processes, Heidelberg, Germany,, oral 14.
[7] Shoda, M. (1991). Methods for the biological treatment of exhaust gases. In Biological degradation of
wastes, A. M. Martin (Ed.), Elsevier, London, 31-46.
[8] Neal A.B. and Loehr R.c. (2000). Use of biofilters and suspended-growth reactors to treat VOCs. Waste
Manage., 20, 59-68.
[9] Anonymous (1995). The effective and reliable removal of both sulphur compounds and heavy metals from
water and (flue) gas. Paques b.v., Postbus 52, 8560 AB Balk, NL.
[10]Dijkman, H. (1995). Biological gas desulphurisation, Med. Fac. Landbouww. Univ. Gent, 60/4b, 2677-
2684.
251
[11] Ottengraf, s.p.p. (1987). Biological systems for waste gas elimination, Trends Biotechnol., 5,132-136.
[12]Wang, z., Govind, R. and Bishop, D.F. (1996). Review of biofiltration - Effect of support media on Biofilter
performance. In 89th annual meeting & exhibition Air & Waste Management Assoc., Nashville, Tennessee,
96-wp87A.O5.
[13]Seignez C; Atti A; Adler N and Peringer p (2002). Effect of biotrickling filter operating parameters on
chlorobenzenes degradation. J. Envfron. Eng., 128, 360-366.
[14]Le Cloừec, p.; Humeau, p. and Ramirez-Lopez, E.M. (2001). Biotreatment of odours: control and
performances of a biofilter and a bioscrubber. Water Sci. Technol., 44, 219-226.
[15]Smith, F.L.; Sorial, G.A.; Suidan, M.T.; Biswas, p. and Brenner, R.c. (2002). Development and
demonstration of an explicit lumped-parameter biofilter model and design equation incorporating Monod
kinetics. J. Air Waste Manage. Assoc., 52, 208-219
[16]Chang, K.S.; Lu, C.Y. and Lin, M.R. (2001). Treatment of volatile organic compounds from polyurethane
and epoxy manufacture by a trickle-bed air biofilter. J. Biosci. Bioeng., 92,126-130.
[17]Diks, R.M.M., Ottengraf, s.p.p. and Vrijland, s. (1994). The existence of a biological equilibrium in a
trickling filter for waste gas purification, Biotechnol. Bioeng., 44,1279-1287.
[18]Pedersen, A.R. and Arvin, E. (1995). Removal of toluene in waste gases using a biological trickling filter.
Biodegradation, 6,109-118.
[19]Shinabe, K., Oketani, s., Ochi, T. and Matsumura, M. (1995). Characteristics of hydrogen sulphide removal
by Thiobacillus thiooxidans KS1 isolated from a carrier-packed biological deodorization system. J. Ferment.
Bioeng., 80, 592-598.
[20]Thalasso, F., Ancia, R., Willocx, B., L’Hermite, Ph., Naveau, H. and Nyns, E.-J. (1993). The ’’Mist- Foam”
concept: A concept for biological treatment of gaseous organic compounds. In Characterisation and control
of odours and voc in the process industries, Vigneron, s., Hermia, J. and Chaouki, J. (Eds.), Elsevier,
Amsterdam, The Netherlands, 419-429.
[21]Weber, F.J. and Hartmans, s. (1996). Prevention of clogging in a biological trickle-bed reactor removing
toluene from contaminated air. Biotechnol. Bioeng., 50, 91-97.
[22]Smith, F.L., Sorial, G.A., Suidan, M.T., Breen, A.W., Biswas, p. and Brenner, R.c. (1996). Development of
two biomass control strategy for extended, stable operation of highly efficient biofilters with high toluene
loadings. Envfron. Sci. Technol., 30,1744-1751.
[23]speitel, G.E. and McLay, D.s. (1993). Biofilm reactors for treatment of gas stream containing chlorinated
solvents. J. Env. Eng., 119, 658-678.
[24]Kennes, c., Cox, H.H.J., Doddema, H.J. and Harder, w. (1996). Design and performance of biofilters for the
removal of alkylbenzene vapours. J. Chem. Technol. Biotechnol., 66, 300-304.
[25]Garcia-Pena, E.I.; Hernandez, s.; Favela-Torres, E.; Auria, R. and Revah, s. (2001). Toluene biofiltration by
the fungus Scedosporỉum apiospermum TB1. Biotechnol. Bioeng., 76, 61-69.
[26]De Heyder, B., Overmeừe, A., Van Langenhove, H. and Verstraete, w. (1994). Ethene removal from a
synthetic waste gas using a dry biobed, Biotechnol. Bioeng., 44, 642-648.
[27]Devinny, J.s. and Hodge, D.s. (1995). Formation of acidic and toxic intermediates in overloaded ethanol
biofilters. J. Air Waste Manage. Assoc., 45, 125-131
[28]Kirchner, K., Hauk, G. and Rehm, H.J. (1987). Exhaust gas purification using immobilised monocultures
(biocatalyst), Appl. Microbiol. Biotechnol., 26, 579-587.
[29]Moe, W.M. and Irvine, R.L. (2001). Effect of nitrogen limitation on performance of toluene degrading
biofilters Wat. Res., 35, 1407-1414
[30]Moe WM and Irvine RL (2001). Polyurethane foam based biofilter media for toluene removal. Water Sci.
Technol., 43, 35-42.
[31]Pol A.; van Haren F.J.J.; den Camp J.H.M.O. and van der Drift, c. (1998). Styrene removal from waste gas
with a bacterial biotricking filter. Biotechnol. Lett., 20,407-410.
[32]Chitwood, D.E. and Devinny, J.s. (2001). Treatment of mixed hydrogen sulphide and organic vapours in a
rock medium biofilter. Water Environ. Res., 73,426-435.
[33]Shareefdeen, z., Baltzis, B.C., Oh, Y.-S. and Bartha, R. (1993). Biofiltration of methanol vapours.
Biotechnol. Bioeng., 41, 512-524.
[34]Lu, C.S.; Lin, M.R. and Chu, C.H. (2002). Effects of pH, moisture, and flow pattern on trickle-bed air
biofilter performance for BTEX removal. Adv. Envừon. Res., 6, 99-106.
[35]Thalasso, F., Naveau, H. and Nyns, E.-J. (1996). Effect of dry periods in a “Mist-Foam” bioreactor designed
for gaseous substrate. Envừon. Technol., 17, 909-913.
[36]Zhu, X.Q.; Alonso, c.; Suidan, M.T.; Cao, H.W.; Kim, B.J. and Kim, B.R. (1998). The effect of liquid phase
on voc removal in trickle-bed biofilters. Water Sci. Technol., 38, 315-322.
[37]Le Cloirec, p., Fanlo, J-L. and Degorce-Dumas, J. (1991). Traitement des odeurs et désodorisation
industrielle. Innovation 128, Paris, 1991.
[38]Jorio H; Bibeau L . and Heitz M (2000). Biofiltration of aừ contaminated by styrene: Effect of nitrogen
252
supply, gas flow rate, and inlet concentration. Envừon. Sci. Technol., 34, 1764-1771.
[39]Okkerse, W.J.H.; Ottengraf, S.P.P.; Osinga-Kuipers, B. and Okkerse, M. (1999). Biomass accumulation and
clogging in biotrickling filters for waste gas treatment. Evaluation of a dynamic model using
dichloromethane as a model pollutant. Biotechnol. Bioeng., 63,418-430.
[40]Alonso, c.; Zhu, X.Q.; Suidan, M.T.; Kim, B.R. and Kim, B.J. (2001). Mathematical model of biofiltration
of VOCs: Effect of nitrate concentration and backwashing. J. Envừon. Eng., 127, 655-664.
[41]Serial, G.A.; Smith, F.L.; Suidan, M.T. and Brenner, R.c. (2001). Removal of ammonia from contaminated
air by trickle bed air biofilters. J. Ak Waste Manage. Assoc., 51, 756-765.
[42]Lim, J.S.; Park, S.J.; Koo, J.K. and Park, H (2001). Evaluation of porous ceramic as microbial carrier of
biofilter to remove toluene vapours. Envừon. Technol., 22,47-56.
[43]Kennes, c. and Veiga, M.c. (2002). Inert filter media for the biofiltration of waste gases-Characteristics and
biomass control. Re/Views. Envừon. Sci. Bio/Technol, 1, 201-214.
[44]Bentz, R. (1987). Biological waste treatment: experience of a chemical company in Switzerland. In
Biological treatment of industrial waste gases. State of the art and comparison with physico-chemical
processes, Dechema, Heidelberg, oral 13.
[45]Hereth, H. (1987). Biological waste aừ treatment in a gelatine plant In Biological treatment of industrial
waste gases. State of the art and comparison with physico-chemical processes, Dechema, Heidelberg, oral 11
[46] Lebeault, J.-M. (1993). Le traitement des effluents gazeux. Biofutur, September, 29-31.
[47]Prokop, w. and Bohn, H. (1985). Soil bed system for control of rendering plant odours. In 78th annual
meeting of the Air Pollution Control Association, Detroit, Michigan, 6, 85-79.
[48]Clark, R. and Wnorowski, A. (1992). Biofilters for sewer pump station vents : influence of matrix
formulations on the capacity and efficiency of odorant removal by an experimental biofilter. In
Biotechniques for air pollution abatement and odour control policies, A. J. Dragt and J. van Ham (Eds.),
Elsevier, Maastricht, The Netherlands, 183-186.
[49]Bohn, H. L. (1996). Biofilter media. In 89th annual meeting & exhibition Air & Waste Management Assoc.,
Nashville, Tennessee, 96-WP87A.01.
[50]Tang, H.-M., Hwang, S.-J. and Hwang, S.-C. (1996). Waste gas treatment in biofilters. J. Aừ Waste Manage.
Assoc., 46, 349-354.
[51]Morgenroth, E., Schroeder, E.D., Chang, D.P.Y. and Scow, K.M. (1996). Nutrient limitation in a compost
biofilter degrading hexane. J. Air Waste Manage. Assoc., 46, 300-308.
[52]Veir, J. K., Schroeder, E.D., Chang, D.P.Y. and Scow, K.M. (1996). Interaction between Toluene and
Dichloromethane degrading populations in a compost biofilter. In 89th annual meeting & exhibition Air &
Waste Manage. Assoc., Nashville, Tennessee, 96-WP87A.07.
[53]Elias A; Barona A; Arreguy A; Rios J; Aranguiz I; Penas J (2002). Evaluation of a packing material for the
biodegradation of H2S and product analysis. Process Biochem., 37 (8), 813-820.
[54]Smet, E., Chasaya, G., Van Langenhove, H. and Verstraete, w. (1996). The effect of inoculation and the type
of carrier material used on the biofiltration of methyl sulphides. Appl. Microbiol. Biotechnol., 45, 293-298.
[55]Ibrahim, M.A.; Mizuno, H.; Yasuda, Y.; Fukunaga, K. and Nakao K (2001). Removal of mixtures of
acetaldehyde and propionaldehyde from waste gas in packed column with immobilized activated sludge gel
beads. Biochem. Eng. J., 8, 9-18.
[56]Baltzis, B.c. and Shareefdeen, z. (1994). Biofiltration of voc mixtures: Modeling and pilot Scale
experimental verification. In 87th annual meeting & exhibition Air & Waste Management Assoc.,
Cincinnati, Ohio, A907.
[57]Zilli, M., Fabiano, B., Ferraiolo, A. and Converti, A. (1996). Macro-kinetic investigation on phenol uptake
from air by biofilttation: Influence of superficial gas flow rate and inlet pollutant concentration. Biotechnol.
Bioeng., 49, 391-398.
[58]Deshusses, M.A., Hamer, G. and Dunn, I. J. (1995). Behaviour of biofilters for waste air biotreatment. 2.
Experimental evaluation of a dynamic model. Envfron. Sci. Technol., 29,1059-1068.
[59]Arulneyam, D. and Swaminathan, T. (2000). Biodegradation of ethanol vapour in a biofilter. Bioprocess
Eng., 22, 63-67
60] Weber, F. and Hartmans, s. (1995). Use of activated carbon as a buffer in biofiltration of waste gases with
fluctuating concentration of toluene. Appl. Microb. Biotechnol., 43, 365-369.
[61]Abumaizar, R.J.; Kocher, w. and Smith, E.H. (1998). Biofiltration of BTEX contaminated air streams using
compost-activated carbon filter media. J. Hazard Mat., 60,111-126.
[62]Tholander, p. (1987). Biological processes for odour abatement in industrial processes. In Biological
treatment of industrial waste gases. State of the art and comparison with physico-chemical processes,
Dechema, Heidelberg, poster 10.
[63]Dragt, A. and Ottengraf, s.p.p. (1987). Process engineering aspects and new developments in biological afr
pollution control technology.. In Biological treatment of industrial waste gases, state of the art and
comparison with physico-chemical processes, Heidelberg, Germany, oral 21.
253
[64]Ottengraf, S.P.P., Meesters, J.J.P., van den Oever, A.H.C. and Rozema, H.R. (1986). Biological elimination
of volatile xenobiotic compounds in biofilters. Bioprocess Eng., 1, 61-69.
[65] Bohn, H. (1975). Soil and compost filters for malodorous gases. J. Air Pol. Control. Assoc., 25, 953-955.
[66] Pomeroy, R.D. (1982). Biological treatment of odorous air, J. WPCF, 54,1541-1545.
[67]Bohn, H.L. (1993). Biofiltration: Design principles and pitfalls. In 86th annual meeting & exhibition Air &
Waste Manageme. Assoc., Denver, Colorado, 93-TP-52A.01.
[68]Cardenas-Gonzalez, B.; Ergas, S.J.; Switzenbaum, M.S. and Phillibert, N. (1999). Evaluation of full-scale
biofilter media performance. Envfron. Prog., 18, 205-211.
[69]Cardenas-Gonzalez, B.; Ergas, S.J. and Switzenbaum, M.S. (1999). Characterization of compost biofiltration
media. J. Afr Waste Manage. Assoc., 49,784-793
[70]Dalouche, A., Lemasle, M., Le Cloirec, p., Martin, G. and Besson, G. (1989), Utilisation de biofiltres pour
1’épuration de gaz charges en composés azotés et soufres. In Man and his ecosystem. 8th world clean air
congress, eds. Brasser L.J. and w.c. Mulders, The Hague, The Netherlands, 1989, 379-384.
[71]Kennes, c., Cox, H.H.J., Veiga, M.c. and Doddema, H.J. (1995). Continuous removal of benzene related
compounds from waste gases. Med. Fac. Landbouww. Univ. Gent., 60, 279-284.
[72]Delhoménie, M.C., Bibeau, L., Roy, s., Brzezinski, R. and Heitz, M. (2001). Influence of nitrogen on the
degradation of toluene in a compost-based biofilter. J. Chem. Technol. Biotechnol. 76, 997-1006
73] Van Langenhove, H. and Schamp, N. (1987). Inhibitory effect of sulphur dioxide on biofiltration of
aldehydes. In Biological treatment of industrial waste gases. State of the art and comparison with physico-
chemical processes, Dechema, Heidelberg, poster 8.
[74]Holubar, p. and Braun, R. (1995). Biofilttation - bottlenecks in biological air purification and possible future
solutions, Med. Fac. Landbouww. Univ. Gent, 60/4b, 2303-2312.
[75]Ottengraf, s.p.p. and Van den Oever, A.H.C. (1983). Kinetics of organic compound removal from waste
gases with a biological filter, Biotechnol. Bioeng., 25, 3089-3102.
[76]Rands, M., Cooper, D., Woo, C.-P., Fletcher, G. and Rolfe, K. (1981). Compost filters for H2S removal
from anaerobic digestion and rendering exhausts, J. WPCF, 53,185-189.
[77] Yang, Y., and Allen, E.R. (1994). Biofiltration control of hydrogen sulfide. 1. Design and operational
parameters. J. Air Waste manage. Assoc., 44, 863-868.
[78] Allen, E.R. and Phatak, s. (1993). Control of organo-sulphur compound emissions using biofiltration -
Methyl mercaptan. In 86th annual meeting & exhibition Air & Waste Manage. Assoc., Denver, Colorado,
93-WA-52B.03.
[79] Nicolai, R.E. and Janni, K.A. (2001). Biofilter media mixture ratio of wood chips and compost treating
swine odours. Water Sci. Technol., 44, 261-267.
[80] Ergas, S.J., Kinney, K., Fuller, M.E. and Scow, K.M. (1994) Characterization of a compost biofiltration
system degrading dichloromethane. Biotechnol. Bioeng., 44, 1048-1054.
[81] Ergas, S.J., Schroeder, E.D., Chang, D.P.Y and Morton, R.L. (1995). Control of volatile organic compound
emissions using a compost biofilter. Water Environ. Res., 67, 816-821.
[82] Hodge, D. s., Medina, V.F., Islander, R.L. and Devinny, J.s. (1991). Treatment of Hydrocarbon fuel vapours
in biofilters. Environ. Technol., 12, 656-662.
[83] Deshusses, M.A. and Hamer, G. (1993). The removal of volatile ketone mixtures from air in biofilters.
Bioprocess. Eng., 9,141-146.
[84] Sofia Martin, c. (1995). Evaluación de nuevos materiales de relleno para la depuration de aừe en biofiltros.
B.s. thesis, University of Santiago de Compostela, Spain.
[85] Cho, K., Hirai, M. and Shoda, M. (1992). Enhanced removal efficiency of malodorous gases in a pilotscale
peat biofilter inoculated with Thiobacillus thioparus DW44. J. Ferm. Bioeng., 73, 46-50.
[86] Dalouche, A., Gillet, M., Lemasle, M., Martin, G. and Orain, L. (1981). Biodésodorisation des effluents
gazeux. Pol. Atm., 23, 317-322.
[87] Hirai, M., Ohtake, M. and Shoda, M. (1990). Removal kinetics of hydrogen sulfide, methanethiol and
dimethyl sulfide by peat biofilters, J. Ferm. Bioeng., 70, 334-339.
[88] Morales, M., Pérez, F., Auria, R. and Revah, s. (1994). Toluene removal from air stream by biofiltration,
Adv. Bioproc. Eng., 405-411.
[89] Yoon, I.K. and Park, C.H. (2002). Effects of gas flow rate, inlet concentration and temperature on
biofiltration of volatile organic compounds in a peat-packed biofilter. J. Biosci. Bioeng., 93,165-169.
[90] Rothenbũhler, M., Heitz, M., Beerli, M. and Marcos, B. (1995). Biofiltration of volatile organic emissions in
reference to flexographic printing processes. Water Air Soil Poll., 83, 37-50.
[91] Baltzis, B.c. and Androutsopoulou, H. (1994). A study on the response of biofilters to shock-loading. In
87th annual meeting & exhibition Air & Waste Management Assoc., Cincinnati, Ohio, 1994, A905.
[92] Don, J. and Feenstra, L. (1984). Odour abatement through biofiltration. In Characterisation and control of
odoriferous, SBF (Ed.), SBF, Louvain-la-Neuve, Belgium, 337-349.
[93] Zilli, M.; Palazzi, E.; Sene, L.; Converti, A. and Del Borghi, M. (2001). Toluene and styrene removal from
254
air in biofilters. Proc. Biochem., 37, 423-429.

[94] Van Langenhove, H., Wuyts, E. and Schamp, N. (1986). Elimination of hydrogen sulphide from odorous air
by a wood bark biofilter, Wat. Res., 20,1471-1476.
[95] Lee, B.D., Apel, W.A., Walton, M.R. and Cook, L.L. (1996). Treatment of methanol contaminated air
streams using biofiltration. In 89th annual meeting & exhibition Air & Waste Manage. Assoc., Nashville,
Tennessee, 96-RP87C.03.
[96] Bohn, H. and Bohn, R. (1988). Soil beds weed out pollutants. Chem. Eng., April 25,73-76.
[97] Frye, R.J., Welsh, D., Berry, T.M., Stevenson, B.A. and McCallum, T. (1992). Removal of contaminant
organic gases from air in closed systems by soil. Soil Biol. Biochem., 24, 607-612.
[98] Fanlo, J.-L., Degorce-Dumas, J. R., Kowal, s. and Le Cloirec, p. (1991). Precede biologique de traitement de
gaz, biofiltres et application à la désodorisation de gaz, Patent 91.053346.
[99] Atlas, R. (1989). Microbiology, Fundamentals and Applications. Macmillan, New York.
[100] VanDemark, p. and Batzing, B. (1987). The microbes: An introduction to their nature and importance.
Benjamin/Cummings, Menlo Park, California.
[101] Van Lith, c., David, S.L. and Marsh, R. (1990). Design criteria for biofilters. Trans IChemE, 68, 127-
132.
[102] Ebinger, M.H., Bohn, H.L. and Puls, R.w. (1987). Propane removal from propane-air mixtures by soil
beds. J. Air Poll. Control. Assoc., 37,1486-1489.
[103] Bohn, H.L. and Bohn, K.H. (1999). Moisture in biofilters. Environ. Prog., 18, 156-161
[104] Deshusses, M.A., Hamer. G. and Dunn, I.J. (1996). Transient-state behaviour of a biofilter removing
mixtures of vapours of MEK and MIBK from air. Biotechnol. Bioeng., 49, 587-598.
[105] Joi, A. and Dragt, A. (1988). Biotechnological elimination of volatile organic compounds in waste gases,
J. Dechema Biotech. Conf., 2, 373-389.
[106] Kleis, G., Schelchshom, J. and Vinke, A. (1987). Desulphurisation of H2S containing gases by biological
oxidation. In Biological treatment of industrial waste gases. State of the art and comparison with physico-
chemical processes, Dechema, Heidelberg, poster 3.
[107] Elsgaard, L. (2000). Ethylene removal at low temperatures under biofilter and batch conditions. Appl.
Environ. Microbiol., 66, 3878-3882.
[108] Lehtomẩki, J., Tốrrốnen, M. and Laukkarinen, A. (1992). A feasibility study of biological waste-air
purification in a cold climate. In Biotechniques for air pollution abatement and odour control policies, A. J.
Dragt & J. van Ham (Eds), Elsevier, Maastricht, The Netherlands, 131-134.
[109] Lee, B.D.; Apel, W.A. and Smith, W.A. (2001). Oxygen effects on thermophilic microbial populations in
biofilters treating nitric oxide containing off-gas streams. Environ. Prog., 20,157-166
[110] Matteau, Y. and Ramsay, B. (1999). Thermophilic toluene biofiltration. J. Air Waste Manage. Assoc., 49,
350-354.
[111] Plas, c., Holubar p., Moser, K., Ploder w. and Braun, R. (1994). Die Bilanzierung von wasser und
kohlenstoff bei der biofiltration, VDI-Berichte, 1104, 273-278.
[112] Yang, Y. and Allen, E.R. (1994). Biofiltration control of hydrogen sulfide. 2. Kinetics, biofilter
performance and maintenance. J. Air Waste Manage. Assoc., 44, 1315-1321.
[113] Leson, G. and Winer, A.M. (1991). Biofiltration: An innovative air pollution control technology for voc
emissions. J. Air Waste Manage. Assoc., 41,1045-1054.
[114] Bronnenmeier, R., Fitz, p. and Tautz, H. (1994). Reinigung von Lackiererei-Abhift mit einem
Gittertrager-Biofilter. VDI Berichte, 1104, 203-215.
[115] Hartmans, s. and Tramper, J. (1991). Dichloromethane removal from waste gases with a trickle-bed
bioreactor. Bioproc. Eng., 6, 83-92.
[116] Acuna, E., Auria, R., Pineda, J., Pérez, F., Morales, M. and Revah, s. (1996). Studies on the microbiology
and kinetics of a biofilter used to control toluene emissions. In 89th annual meeting & exhibition Air &
Waste Management Assoc., Nashville, Tennessee, 96-WP87A.03.
[117] Windsperger, A., Buchner, R. and Stefan, K. (1990). Rienigung losungmittelhaitiger abluft mit biofiltem.
Staub-Reihaltung der Luft, 50, 465-470.
[118] Hunik, J.H., Meijer, H.J.C. and Tramper, J. (1992). Kinetics of Nitrosomonas europea at extreme
substrate, product and salt concentration. Appl. Microbiol. Biotechnol., 37, 802-807.
[119] Smits, M.C.J., Hoek, A.P., Osinga, B. Ottengraf, s.p.p and Wijngaard, M.H. (1995). Removal of ammonia
and odour from composting facility waste gas with a bio-trickling filter. Med. Fac. Landbouww. Univ.
Gent., 60, 2313-2320.
[120] Martin, G., Gaid, K., Lemasle, M. and Nogrix, p. (1979). Etude d’un procédé biologique de
desodorisation, Pol. Atm., 21, 31-34.
[121] Smit, F. and Derber, H. (1987). Biofiltration - an economical and efficient waste gas treatment method. In
Biological treatment of industrial waste gases. State of the art and comparison with physico-chemical
processes, Dechema, Heidelberg, Germany, poster 15.
255
[122] Kennes, c., Veiga, M.c. and Bhatnagar, L. (1998). Methanogenic and perchloroethylene-dechlorinating
activity of anaerobic granular sludge. Appl. Microbiol. Biotechnol. 50,484-488.
[123] Kim, J.o. (1997). Gaseous TCE and PCE removal by an activated carbon biofilter. Bioproc. Eng. 16, 331-
337.
[124] Kennes, c. and Thalasso, F. (1998). Waste gas biotreatment technology. J. Chem. Technol. Biotechnol.
72, 303-319.
[125] Veiga, M.c. and Kennes, c. (2001). Parameters affecting performance and modeling of biofilters treating
alkylbenzene-polluted air. Appl. Microbiol. Biotechnol. 55, 254-258.
[126] Kim, N.J., Sugano, Y., Hirai, M. and Shoda, M. (2000). Removal characteristics of high load ammonia
gas by a biofilter seeded with a marine bacterium, Vibrio alginolyticus. Biotechnol. Lett. 22, 1295-1299.
[127] Smet, E., Van Langenhove, H. and Philips, G. (1999). Dolomite limits acidification of a biofilter
degrading dimethyl sulphide. Biodegradation. 10, 399-404.
[128] Wani, A.H., Lau, A.K. and Branion, R.M.R. (1999). Biofiltration control of pulping odors - hydrogen
sulfide: performance, macrokinetics and co-existence of effects of organo-sulphur species. J. Chem.
Technol. Biotechnol. 74, 9-16.
129 Mohseni, M. and Allen, D.G. (2000). Biofiltration of mixtures of hydrophilic and hydrophobic volatile
organic compounds. Chem. Engin. Sci. 55,1545-1558.
[130] Cox, H.H.J., Moerman R.E., van Baalen s., van Heiningen W.N.M., Doddema H.J. and Harder w. (1997).
Performance of a styrene-degrading biofilter containing the yeast Exophiala jeanselmei. Biotechnol. Bioeng.
53, 259-266.
[131] Lu, c., Lin, M.-R. and Lin, J. (2000). Removal of acrylonitrile vapours from waste gases by a trickle bed
air biofilter. Biores. Technol. 75, 35-41.
[132] Guey, c., Degorge-Dumas, J.R. and Le Cloừec, p. (1995). Hydrogen sulfide removal on biological
activated carbon. J. Odours VOC's. 1, 144-145.
[133] Lu, c., Lin M.-R. and Wey I. (2001). Removal of pentane and styrene mixtures from waste gases by a
trickle-bed aừ biofilter. J. Chem. Technol. Biotechnol. 76, 820-826.
[134] Kirchner, K., Wagner, s. and Rehm, H.-J. (1992). Exhaust gas purification using biocatalysts (fixed
bacteria monocultures) - the influence of biofim diffusion rate (02) on the overall reaction rate. Appl.
Microbiol. Biotechnol. 37, 277-279.
[135] Cox, H.H.J. and Deshusses, M.A. (1999). Biomass control in waste aừ biotrickling filters by protozoan
predation. Biotechnol. Bioeng. 62: 216-224.
[136] Prado, O.J., Mendoza J.A., Veiga M.c. and Kennes c. (2002). Optimization of nutrient supply in a
downflow gas-phase biofilter packed with an inert carrier material. Appl. Microbiol. Biotechnol. (in press).
[137] Windsperger, A. (1991). Use of biofilters for the purification of gases containing solvents. Radex
Rundschau. 3-4, 457-464.
[138] Huber, J. (1992). Planting, Durchfuhrung und erste Erfahrungen zum Biofilter
Tierkorperbeseitigungsanlage Plattling. In Dragt and van Ham (eds.), Biotechniques for air pollution
abatement and odour control policies. Elsevier, Amsterdam, 161-165.
[139] Leson, G. and Smith, B.J. (1997). Petroleum envừonmental research forum field study on biofilters for
control of volatile hydrocarbons. J. Envừon. Eng. 123: 556-562.
[140] Swanson, W.J. and Loehr R.c. (1997). Biofiltration: Fundamentals, design and operation principles and
applications of biological APC technology. J. Envừon. Eng. 123, 538-546.
[141] Wright, W.F., Schroeder, E.D., Chang, D.P.Y., and Romstad, K. (1997). Performance of a pilot-scale
compost biofilter treating gasoline vapours. J. Envừon. Eng. 123, 547-555.
[142] Jorio, H., Kiared, K., Brzezinski, R., Leroux, A., Viel, G. and Heitz, M. (1998). Treatment of aừ polluted
with high concentrations of toluene and xylene in a pilot-scale biofilter. J. Chem. Technol. Biotechnol.
73,183-196.
[143] Kraislas, s., Tuan Pham, Q., Amal, R., Jiang, J.K. and Heitz, M. (2000). Effect of inlet mass loading,
water and total bacteria count on methanol elimination using upward flow and downward flow biofilters. J.
Chem. Technol. Biotechnol. 75, 299-305
[144] Fischer, K. (1992). Vergleichende Untersuchungen: Biofilter- Oder Biomembranvefahren zur reinigung
losemittelhaltiger industrieabluft. In Biotechniques for aừ pollution abatement and odour control policies, A.
J. Dragt and J. van Ham (Eds.), Elsevier, Maastricht, The Netherlands, 97-102.
[145] Hartmans, s., Leenen, E.J.T.M and Voskuilen, G.T.H. (1992). Membrane bioreactor with porous
hydrophobic membranes for waste-gas treatment In Biotechniques for air pollution abatement and odour
control policies, A. J. Dragt and J. van Ham (Eds.), Elsevier, Maastricht, The Netherlands, 103-106.
[146] Reij, M.W., de Gooijer, K.D., de Bont, J.A.M. and Hartmans, s. (1995). Membrane bioreactor with a
porous hydrophobic membrane as a gas-liquid contactor for waste gas treatment. Biotechnol. Bioeng., 45,
107-115.
[147] Parvatiyar, M.G., Govind, R. and Bishop, D. F. (1996). Treatment of trichloroethylene (TCE) in a
256
membrane biofilter. Biotechnol. Bioeng., 50, 57-64.

[148] Ergas, S.J., Shumway, L., Fitch, M.w. and Neeman, J.J. (1999). Membrane processes for biological
treatment of contaminated airstreams. Biotechnol. Bioeng. 63, 431-44.
[149] Ritchie, B.J. and Hill, G.A. (1995). Biodegradation of phenol-polluted aừ using and external loop airlift
bioreactor. J. Chem. Technol. Biotechnol., 62, 339-344.
[150] Cesário, M.T., Beeftink, H.H. and Tramper, J. (1992). Biological treatment of waste gases containing
poorly soluble pollutants. In Biotechniques for aừ pollution abatement and odour control policies, A.J. Dragt
and J. van Ham (Eds.), Elsevier, Maastricht, The Netherlands, 135-140.
[151] Cesário, M.T., Beeftink H.H. and Tramper, J. (1994). Removal of poorly water-soluble compounds from
waste gases using water-immiscible organic solvents. In 6th. Eur. Congr. Biotechnol., L. Alberghina, L.
Frontali, and p. Sensi (Eds.), Elsevier, Amsterdam, 1207-1210.
[152] Cesário, M.T., Beeftink, H.H. and Tramper, J. (1995). Feasibility of using organic water-immiscible
solvents in biological waste-gas treatment. Bioproc.Eng. 12, 55-63.
[153] Buisman, C.J., Wit, B. and Lettinga, G. (1990). Biotechnological sulphide removal in three polyurethane
carrier reactors: Stirred reactor, biorotor reactor and upflow reactor. Wat. Res., 24, 245-251.
[154] Lee, D.H.; Lau, A.K. and Pinder, K.L. (2001). Development and performance of an alternative biofilter
system. J. Aừ Waste Manage. Assoc., 51, 78-85.
[155] Van Groenestijn, J.W., Doddema, H., Kok, H. and Koster, T.P.M. (1994). Combined photochemical and
biological treatment of off-gases. VDI Berichte 1104, 313-324.
257
BIOREACTORS FOR THE TREATMENT OF INDUSTRIAL WASTE GASES
CONTAINING FORMALDEHYDE AND OTHER ALIPHATIC COMPOUNDS
ÓSCAR J. PRADO, MARTA EIROA, MARÍA c. VEIGA AND

CHRISTIAN KENNES
University of La Coruna, Chemical Engineering Laboratory, Campus da
Zapateira, E -15071 - La Coruna, Spain. Fax: 34 981167065. e-mail:
kennes@udc.es
1. Introduction
1.1. GAS PHASE BIOREACTORS
Over the past decades, bioreactors have proven to be efficient and cheap systems for the
abatement of a variety of common aừ pollutants. Among theừ main advantages, one
should mention theừ high efficiency, minimal side-effects on health and on the
envừomnent and theừ relatively low cost. Three basic types of bioreactors can be
distinguished [1]:
• Bioscrubber (Fig. la): composed of an absorption column, where the pollutants

are absorbed in a liquid phase, and a stirred tank bioreactor, in which
biodegradation takes place.
• Trickling biofilter (Fig. lb): consists of a fixed film bioreactor, which is
continuously fed a liquid medium.
• Biofilter (Fig. 1c): similar to the previous one, but no continuous supply of liquid
medium is used (liquid supply can be periodical or simply non-existent).
Conventional biofilters based on organic or natural filter beds have been used for several
decades for the treatment of polluted aừ streams mainly at wastewater treatment plants
and composting facilities. Much more recently, the application of biofiltration has been
extended to new sources, among others industrial waste gases. The design of biofilters
has been improved, new filter beds have been tested and new types of bioreactors have
been developed. The present chapter describes the use of biofilters for the treatment of
industrial waste gases containing a mixture of formaldehyde and other aliphatic
compounds as methanol. Very little has been published on this topic and the first papers
dealing with such industrial waste gases were published only a few years ago, in the
nineties.
259
Óscar J. Prado, Marta Eừoa, Maria c. Veiga and Christian Kennes
Clean air Clean air Clean . A?"e°us

A I phase
Ỳv (°ptional)
Water and
Polluted nutrients
air
(b)
Absorption Bioreactor Sludge Water Polluted

column drain drain air
Fig. 1: Bioreactor schemes: (a) Bioscrubber; (b) Trickling biofilter; (c) Biofilter
In this chapter, after reviewing the literature available on formaldehyde biodegradation,

case studies on biofiltration of formaldehyde-containing waste gases will be described,
260
Bioreactors for the treatment of industrial waste gases
before presenting our own preliminary data on the treatment of formaldehyde containing
waste gases in biofilters and biotrickling filters packed with different inert carriers. The
possibility of treating waste gases containing both formaldehyde and methanol is
described.
1.2. FORMALDEHYDE AS INDUSTRIAL AIR POLLUTANT
Formaldehyde (HCHO) is a common compound in the chemical industry used in a wide

variety of processes and frequently found in wastes, causing envfronmental pollution. It is
a colourless gas at normal temperature and pressure. Formaldehyde presents a
characteristic pungent odour and it is irritating to the mucous membranes at
concentrations above 20 mg/L [2]. It is readily soluble in water, alcohols, and other polar
solvents, but has a low degree of solubility in non-polar fluids. Some physical and
chemical properties of formaldehyde are shown in table 1.
Table 1: Physical and chemical properties of formaldehyde (modified from [2])
Relative molecular mass 30.03

Relative gas density (aừ =1) 1.075
Boiling point (°C) -19
Melting point (°C) -118
Inflammation temperature (°C) 430
Henry’s constant (Pa m3/mol) 0.02
Formaldehyde is present in the environment originating from

natural processes and man-made sources. It is formed in large quantities in the
troposphere by the oxidation of hydrocarbons. Minor natural sources include the
decomposition of plant residues. Formaldehyde is produced industrially in large
quantities and used in many applications, as in glue production, wood products,
preservatives, permanent press fabrics, paper product coatings and certain insulation
materials. Building products made with formaldehyde resins can emit formaldehyde gas.
These products include particleboard used as sub-flooring or shelving, fibreboard in
cabinets and furniture, plywood wall panels and foamed insulation. Incomplete
combustion, cigarette smoking and burning wood, kerosene and natural gas also release
formaldehyde. In our laboratory, formaldehyde biodegradation in waste gases is studied
because of its presence in gaseous effluents of industrial resin/glue producing industries.
Such resins/glues are mainly used in wood industries. Formaldehyde and urea are major
raw materials in their production.
Formaldehyde can show adverse effects on humans exposed to high concentrations of
the pollutant. Symptoms of formaldehyde exposure include nausea, vomiting, abdominal
pain or diarrhoea. Formaldehyde is considered to present a carcinogenic risk; it has
caused cancer in laboratory animals. It can react with microbial DNA and RNA
molecules, as well as proteins, resulting in cell damage.
2. Biodegradation of formaldehyde
In spite of its inhibitory effect to microorganisms, formaldehyde is known to be
261
biodegradable under aerobic conditions. Background information on formaldehyde

biodegradation is given below. Most studies deal with batch experiments in aqueous
media, although some authors have also used continuous liquid phase reactors.
Bonastre et al. [3] studied the biological degradation of formaldehyde using the
activated sludge process in order to carry out kinetic studies. In their work, the
experimental system consisted of reaction vessels containing a medium stirred
magnetically and aerated with compressed air. The tests were performed in a thermostatic
bath at three different temperatures of 15, 25, and 35 °C. The initial formaldehyde
concentrations ranged from 100 to 2300 mg/L.
An analysis of the experimental data was performed, observing a satisfactory fit
between the data and the Vanillin’s kinetic model, corresponding to Equation 1.
ids sn
X dt K"-2S2O+Sn (1)
In Equation 1, JI is the specific rate of subsfrate consumption, Ịiniax is the maximal

specific rate of substrate consumption, s is the substrate concentration, So is the initial
substrate concentration, X is the biomass concentration and n and Ks are kinetic constants.
The data were minimised using equation 1 and the following maximal specific rates
of substrate consumption were obtained ((mg/L)s/(mg/L) x-h): between 0.4 and 0.8 at 15
°C, between 1.0 and 2.2 at 25 °C and between 4.0 and 6.0 at 35 °C. The relation between
the maximal specific rate of substrate consumption and temperature followed an
Arrhenius correlation (Equation 2).
In u =-9413.05-^ + 32.13 (2)

v z
* IIloX rji1 where
ỊẤmax is expressed in h and T in °C.
Different authors have shown the biodegradability of formaldehyde by pure microbial
cultures. Adroer et al. [4] reported formaldehyde biodegradation by a strain of
Pseudomonas putida. Formaldehyde-using bacteria were isolated from sludge of an
industrial wastewater treatment plant by successive enrichments in a salt medium
containing formaldehyde. The isolated microorganisms were used in a fluidised bed
bioreactor treating an industrial wastewater containing formaldehyde. From this
bioreactor the microorganism used in this work was selected as a formaldehyde-using
bacterium. The isolated strain was classified as belonging to the species Pseudomonas
putida. Batch assays were carried out in a shaking bath at 30 °C, containing a cell
suspension (30 - 180 mg/L) and formaldehyde (250 - 500 mg/L).
Pseudomonas putida was grown batch wise on formaldehyde in order to study the
degradation of formaldehyde by growing cells. The results indicated that the
biodegradation of formaldehyde led to the simultaneous appearance of formic acid and
methanol. The biodegradation of these metabolites started after exhaustion of
formaldehyde in the medium. In order to study the nature of formaldehyde dismutase
formation, experiments were performed using resting cells obtained from cultures
previously grown in different media. The enzyme activity was measured and the results
are shown in table 2. Dismutase specific activities were different, being higher for the
cells grown in a medium containing formaldehyde as sole carbon source. The results
262
indicate that the presence of formaldehyde stimulates the production of the enzyme.
Table 2: Formaldehyde dismutase specific activities in resting cells of Pseudomonas putida A2 previously grown
in different media [4]
Medium Specific activity (pmol/mg protein
min)
Glucose 3.5
Glucose and formaldehyde 5.2
Formaldehyde 6.0
Tryptone and yeast extract 1.7
Azachi et al. [5] isolated a bacterium from soil collected at a storage site for
formaldehyde near a chemical plant that uses formaldehyde in the production of glue. The
strain, called MA-C, was identified as Halomonas sp. and was found to be a highly
formaldehyde-resistant halotolerant bacterium. Soil samples were used to inoculate 250
mL flasks with 50 mL medium containing 10 % NaCl, other salts, 5.0 g/L sodium
succinate, 0.5 g/L yeast extract and formaldehyde.
Halomonas sp. MA-C was able to grow in the presence of formaldehyde
concentrations of up to 75 - 100 mg/L in a salt medium. At formaldehyde concentrations
of 125 and 150 mg/L, growth was significantly inhibited. During growth, formaldehyde
disappeared from the medium. At a formaldehyde concentration of 150 mg/L, growth was
limited but formaldehyde was still transformed to a significant extent, probably owing to
a high formaldehyde dehydrogenase activity of the cells. In order to follow the fate of
formaldehyde metabolised by strain MA-C, 14C-labelled formaldehyde was added to
cultures to test whether formaldehyde was incorporated into the cells or oxidised to
carbon dioxide or to non-volatile dissolved products. The amount of labelled
formaldehyde incorporated into the cells and that remaining in the culture supernatant
were determined. No significant incorporation of radioactivity into the cells was
observed, and the radioactivity of the culture supernatant slowly declined. The main
product of formaldehyde transformation by Halomonas sp. MA-C is probably carbon
dioxide.
The presence of formaldehyde dehydrogenase in cell exfracts of sfrain MA-C was
tested. Activities measured were higher in cells grown for 24 hours in the presence of 20
mg/L formaldehyde than in cells that had not been exposed to formaldehyde during at
least 10 fransfers. Activities obtained were between 650 and 850 nmol reduced NAD/mg
protein min at room temperature in extracts of cells grown in the presence of 20 mg/L
formaldehyde. In cell extracts prepared from cells grown in similar media, but in the
absence of formaldehyde, the activities obtained were between 280 and 480 nmol reduced
NAD/mg proteinmin.
Kaszycki and Koloczek [6] investigated formaldehyde and methanol biodegradation
by the methylotrophic yeast Hansenula polymorpha. Hansenula polymorpha is a
methylofrophic yeast that can use both formaldehyde and methanol as single carbon
source. The formaldehyde and methanol metabolism found in methylotrophic yeasts
consists of a complex enzymatic pathway comprising both energy-yielding dissimilation
reactions as well as the assimilation of carbon into cell structural components.
The Hansenula polymorpha cells were grown, at 37 °C and pH 5.0, on methanol in
order to induce the enzymes of the methylotrophic pathway. In experiments of
formaldehyde biodegradation, the cell culture was additionally treated with 300 mg/L
263
formaldehyde about 8 hours before the start of each experiment. This step was found to
be necessary to preadapt the cells to the toxic envfronment and to induce the enzymes
directly involved in formaldehyde utilisation. The experiments were performed in a
simplified medium, which resembled the mineral content of many wastewaters originated
from chemical indusfries, and was close to the salt solution serving as a medium for
methylottophic yeasts. The medium contained 200 mg/L (NH 4)2SO4, 300 mg/L KC1, 30
mg/L H3PO4, and 25 mg/L yeast extract.
In a simplified environment of a model wastewater solution, Hansenula polymorpha
cells were able to grow and metabolise formaldehyde at concentrations typical for
wastewaters of chemical indusfries. The yeast culture inoculated at a low cell density of
3.0xl05 cells/mL was able to grow and enter the exponential phase on initial
formaldehyde levels of about 400 mg/L. Above this concenfration the toxic effect of
formaldehyde prevented the cells from proliferation and led to the final loss of survival.
The capability of methylofrophic yeasts to assimilate exogenously supplied formaldehyde
into cell components is in contrast to formaldehyde-resistant bacteria, where
formaldehyde is oxidised directly into carbon dioxide.
In other experiments a culture grown to the late logarithmic phase was used at a high
cell density of l.lxio7 cells/ml to study the maximum biodegradation potential of
formaldehyde. The yeast cells were fully viable and able to degrade formaldehyde present
at initial concenfrations of up to 700 mg/L. At initial concenfrations over 700 mg/L,
formaldehyde or some of its by-products became toxic for the Hansenula polymorpha
culture. Above this concenttation toxicity prevented the cells from proliferation and led to
the final loss of survival, leading to the complete loss of biodegradation capabilities. The
highest rate of formaldehyde biodegradation was approximately 400 mg/L h.
Yamazaki et al. [7] studied the biodegradation of formaldehyde by a formaldehyde-
resistant bacterium isolated from coastal seawater. The bacterium, designated as a DM- 2
strain, was cultured aerobically at 28 °C and pH 6.8. The medium contained 30 g/L NaCl,
5 g/L tryptone, 5 g/L yeast extract, 6 g/L Na2HPO4, 3 g/L KH2PO4, 1 g/L NH4CI, 1.2 g/L
MgSO4, and 0.1 g/L CaCl2. Cells of the DM-2 strain were precultured until the late
logarithmic growth phase in either the absence or the presence of 200 or 400 mg/L
formaldehyde and were used for biodegradation experiments. In these experiments cells
were added to medium containing either 200 or 400 mg/L formaldehyde as the sole
carbon source.
Cells of DM-2 precultured in the presence of 200 or 400 mg/L formaldehyde were
able to completely degrade 200 mg/L formaldehyde within 15 hours. DM-2 cells
precultured in the absence of formaldehyde were also able to degrade formaldehyde;
however, more than 70 % of the substrate remained even after 20 hours. The degradation
of 400 mg/L formaldehyde was also investigated and the ability of formaldehyde
degradation was dependent on the preculture conditions. Production of formaldehyde
dehydrogenase may be induced by formaldehyde during the growth phase; consequently,
cells of DM-2 precultured in the presence of formaldehyde showed high formaldehyde
degradation activity.
The effect of the cell concentration on the formaldehyde degradation rate was also
studied. Higher cell concentrations resulted in increased formaldehyde degradation rates.
The highest formaldehyde degradation rate reached in this study was 45 mg/L h at a
concentration of the DM-2 strain corresponding to an optical density of 1.2 at 660 nm.
264
Hidalgo et al. [8] reported formaldehyde removal in a synthetic medium and in

industrial wastewater by Rhodococcus erythropolis UPV-1. Rhodococcus erythropolis
strain UPV-1 was isolated from a phenol-polluted site in an estuary. Batch experiments
were performed at 30 °C in 250 IĨ1L flasks containing 100 mL medium. Continuous
formaldehyde feed without variation of the culture volume was achieved.
The disappearance of formaldehyde was studied for different initial concentrations.
The specific rate of formaldehyde disappearance remained constant over the selected
concentration range. For different initial biomass concentrations these specific rates
remained almost constant, decaying slightly at the highest biomass concentration tested.
Bacterial growth was negligible in all cases.
When a repeated and discontinuous feed approach was used, Rhodococcus
erythropolis UPV-1 was able to remove several consecutive doses of 20 mg
formaldehyde/L from the culture medium. The last pulses of formaldehyde took longer to
be completely removed than the previous ones, showing some kind of cumulative effect.
A continuous formaldehyde delivery system, which allowed formaldehyde to diffuse into
the medium at a constant rate of 0.41 mg/L h, was used. At this rate, the formaldehyde
was completely removed over long periods of time.
3. Case studies on biological abatement of formaldehyde in industrial waste gases
Over the past decades, the increasing concern about envfronmental contamination has led
to a fast development of technologies suitable for treating recalcitrant compounds both in
gas and in aqueous phases. As described above, in the case of formaldehyde, a number of
studies concerning its (bio) elimination have been published recently, mainly for aqueous
systems. Very little has been published on the treatment of gaseous effluents containing
formaldehyde, but results reported on aqueous phase reactors suggest that its
biodegradation should be possible in biofilters even in presence of other compounds as
methanol as found in waste gases from synthetic resin producing industries. In the next
pages we present background information and a general compendium of experiences
obtained with industrial gas-phase bioreactors used for formaldehyde biodegradation in
different applications, before presenting our own results and experience.
3.1. BACKGROUND
Ferranti [9] has reported data on the performance of formaldehyde-degrading biofilters, at

pilot- and industrial-scale. In his studies he used an inert hydrophilic support called
BioKey™ (Corain Impianti Engineering & Contracting s.r.l, Rome, Italy), consisting of
spherical pellets with a nominal size of 2 cm. Before starting the experiments, this filter
bed was fed appropriate nutrients in aqueous phase in order to allow colonisation of the
packing material by the mixed microbial population inoculated. Start-up of a pilot-scale
reactor, with a total filter-bed volume of 1.33 m 3 and 75 cm high, was performed with a
continuous air flow rate of 100 Nm 3/h, and with a formaldehyde concentration around 30
mg/Nm3. The empty bed residence time was 48 s. The biofilter was equipped with a
prescrubber, in which the gas was humidified up to saturation and cooled at the wet bulb
temperature. Moreover, the bed was periodically sprinkled. All of this was done in order
to keep the adequate temperature and moisture content in the filter bed. The relative
265
humidity of the bed remained between 45 % and 50 % during all the experiment, while
the pH of the water sprinkled was between 7.0 and 8.0. Under these conditions, the
biofilter almost immediately reached a high removal efficiency of around 60 %, which
gradually improved on the subsequent days, reaching, on day 14, the minimum desfred
removal efficiency of 95 %. This pilot-scale biofilter was operated for 20 weeks, with an
inlet formaldehyde concentration of 45 ± 13 mg/Nm 3. The total gas flow rate was
increased from 100 to 300 Nm3/h immediately after start-up, and then stepwise increased
to a value of 550 Nm3/h on day 35 of operation. The gas temperature had a constant value
of 35 °C, and its relative humidity was approximately 100 %. After a short start-up phase,
this reactor worked continuously with a removal efficiency above 97 %, with a peak of
99.98 % on day 49, when the inlet concentration was 40.33 mg/Nm 3. On day 59 the
formaldehyde inlet concentration increased up to 317.5 mg/Nm 3. Such an important
modification of that parameter did not lead to a significant decrease of the reactor’s
efficiency, which dropped only to 88.6 %. This proves the adaptability of the biofilter to
shock loads. After a transition period of two weeks the system recovered its usual
efficiency. With this exception, the reactor worked during all the experiment with a
removal efficiency always above 95 %. The total pressure drop remained below 50 mm
H2O.
The results obtained with this pilot-scale biofilter were used to build an industrial plant
located in Viadana (Italy) [9]. The plant was initially designed for the treatment of 80000
Nm3/h of total exhaust afr with a formaldehyde concentration between 30 and 50
mg/Nm3, but after 9 months operation the flow rate was increased to 100000 Nm 3/h. The
system consisted basically of a prescrubber and four bioreactors connected in parallel,
with two beds of filter material each. Each one of the eight filtering beds, which had an
area of 33 m2 and a height of 80 cm, was fed individually. The specific velocity of the
exhaust gas was 306 Nm3/m2h before the flow rate was increased, with an empty bed
residence time of 7.5 s. After that, the specific velocity was increased to 383 Nm 3/m2h,
with a residence time of 6 s. A removal efficiency over 90 % was expected. In this case,
the start-up of the plant was performed without inoculation of microorganisms in the filter
bed, resulting in the need of an additional week to reach a nearly-constant removal
efficiency of around 91 %, with peaks of 98 %. Hence, approximately three weeks were
needed for the start-up of this reactor. Formaldehyde concentrations at the inlet (Cin) and
at the outlet (Cout) of the bioreactor were measured during 878 days (Fig. 2).
. .♦ Cin ...• Cout —A—Flow rate
266
Fig. 2: Performance of the Viadana bioreactor (modified from [9])
As can be seen in figure 2, both the aừ flow rate and the inlet formaldehyde concentration
were relatively variable. However, the removal efficiency of the system was high during
all the experiment, with a mean value of 90.65 % and even reaching
98.8 % during certain periods, corresponding to mean elimination capacities around 17
g/m3h. These results prove that the conditions employed allowed maintaining an efficient
formaldehyde removal.
Another pilot plant designed for formaldehyde elimination was started-up by the same
group in Pomponesco (Italy) [9]. In this case, the pollutant was not fed continuously to
the reactor, because the production was stopped during the weekends. Hence, the air
supply to the biofilter was also interrupted. The inlet gas flow rate was initially 290
Nm3/h, and was increased to 500 Nm3/h after a few days of operation. The gas residence
time during these periods was, respectively, 16.6 and 9.5 s. The gas temperature was
around 80 °C, and its absolute humidity was about 30 g of water per Kg dry aừ.
Formaldehyde concentration at the inlet was lower than in the previous case, between
7.8 and 15.6 mg/Nm3. Under these conditions, the efficiency of the biofilter was very
high during all the study, which lasted more than two months. Values were constantly
over 98 %, reaching 99.7 % after six and a half weeks operation. Interruptions on
weekends did not negatively affect its performance. Data obtained from this pilot plant
were, once again, used for the development of an industrial-scale bioreactor located in
Pomponesco (Italy). This reactor was designed to treat a total flow rate of 120000 Nm 3/h
of exhaust air polluted with formaldehyde [9]. The system employed presented four lines
of toxic supply, with very variable flow rates. The mean formaldehyde concentration was
about 20 mg/Nm3, and, as shown in figure 3, the efficiency remained high during the 15
months operation, varying between 82 and 98 %.
Fig. 3: Performance of the Pomponesco bioreactor (modified from [9])
Other studies have been published on the removal of formaldehyde but in presence of
other pollutants. Doronina et al. [10] developed a laboratory-scale biofilter for the
treatment of formaldehyde, methanol and methylamine. The pollutants were fed in
267
aqueous solution, and immediately mixed with a large aft flow to produce contaminated
aft. The reactor was filled with a polyacrylamide fibrous carrier and a washed and boiled
porous ceramic carrier (Ceramsite). The biocatalyst, consisting of an Ethylobacterium
extorquens VKM V-1837D culture, was grown in a nutritive medium containing
methanol as carbon source, before its inoculation into the biofilter. The inoculation was
carried out by means of paper strips in which the bacteria were retained. These paper
strips were transferred from flasks to an agar medium or to a liquid medium for
incubation under optimal conditions, and then inoculated directly into the 2 L biofilter.
The selected filter bed allowed the immobilisation of as much as 100 mg dry biomass per
g polyacrylamide fibrous carrier and up to 10 mg dry biomass per g Ceramsite. A fibrous
polyacrylamide layer located on top of the biofilter allowed a uniform distribution of air
and contaminants. As mentioned above, the latter were supplied in the form of a liquid
medium, which contained 420 - 520 mg/L formaldehyde and 10 mg/L methanol and
which was initially fed at a flow rate of 80 mL/h. During this first phase, which lasted
three days, the removal efficiency of formaldehyde exceeded 99 %, while methanol was
completely eliminated. The flow rate was gradually increased during the next days,
resulting in a slight decrease in the formaldehyde removal rate (Fig. 4), although the
methanol removal efficiency remained constant and high at a value of 100 %, proving
that the method chosen for the start-up phase was adequate, although it may not exactly
represent the real industrial-scale situations since the feed was prepared by mixing an
aqueous and a gas phase.
Flow rate (mUh)
Fig. 4: Formaldehyde removal efficiency as a Junction of liquid flow rate [10]
After the start-up phase, the nutritive medium containing 950 mg/L formaldehyde was
supplied to the biofilter at a constant flow rate of 20 mL/h, obtaining an elimination
efficiency of 99 %. In another series of experiments, the formaldehyde concentration was
increased stepwise to 2000 mg/L. Results showed that, during the first two days operation
with a liquid medium flow rate of 150 mL/h, the removal efficiency was high (about 95
%), but it tended to decrease, reaching 60 - 70 % on the fourth day. Data of formaldehyde
268
utilisation as a function of its concentration are represented in figure 5.

Three weeks after that experiment, the composition of the polluted afr was switched
from formaldehyde to a mixture of methanol and methylamine, at a constant liquid flow
rate of 150 mL/h and a gas flow rate of 240 L/h. Inlet methanol concentrations ranged
from 400 to 1890 mg/L, while methylamine concentrations ranged from 400 to 1300
mg/L. In the case of methanol, removal efficiencies higher than 97 % were found even at
the highest inlet methanol concentration. This means that microorganisms rapidly adapt
from formaldehyde to methanol utilisation, which is not very surprising since methanol is
an intermediate metabolite formed during the biodegradation of formaldehyde by some
microorganisms, as already mentioned above. However, adaptation to methylamine
consumption was slower. At an initial methylamine concentration of 400 mg/L, only 47.5
% was degraded on the first day operation. After 3 days, 76.3 % was degraded, and after
5 days the removal efficiency had reached 99.3 %. A subsequent increase in the pollutant
concentration up to 700 mg/L resulted in 99.6 % degradation of methylamine. Another
increase in the methylamine supply up to 1300 mg/L performed two days later did not
affect the removal efficiency. These results prove that certain microorganisms (e.g.,
Methylobacterium extorquens) grown in a medium containing methanol may be suitable
for the elimination of formaldehyde, methanol and methylamine, although in the case of
methylamine an adaptation period may be necessary for a correct performance.
Fig. 5: Formaldehyde removal efficiency as a function of formaldehyde concentration in the liquid phase
[10]
Huckschlag [11] reported the case of a series of biological systems designed for the
treatment of a mixture of formaldehyde and phenol in different industrial reactors. One of
those reactors, used for the treatment of waste gases from a fibreglass-producing indusfry,
was fed an exhaust gas flow rate between 330 and 650 m 3/h. Phenol inlet concentrations
usually varied between 10 and 15 mg/m 3, while concentrations around 0.3 mg/m 3 were
found in the outlet. This means that the removal efficiency for phenol was between 97
269
and 98 %. In the case of formaldehyde, its concentration in the exhaust gas was
approximately 10 mg/m3, while in the outlet gas it was around 2 mg/m3, which
corresponds to a removal efficiency of 80 %. These results clearly show that a mixture of
low concentrations of phenol and formaldehyde may be efficiently treated by means of a
biological system.
The same study also included data obtained from an industrial plant where a highly
fluctuating amount of phenol, formaldehyde and ammonia was produced and treated
biologically. Table 3 shows the concentration range of these substances both in the inlet
and in the outlet gas stream.
Table 3: Phenol, formaldehyde and ammonia concentration in the inlet and in the outlet gas mi
Compound Inlet gas concentration (mg/m3) Outlet gas concentration (mg/m3)
Phenol 33 -160 1-2
Formaldehyde 6-40 0.5-7
Ammonia 4-42 2-4
Tautz and Rutenfranz [12] used a bioscrubber for formaldehyde, methanol and
ethyleneglycol abatement at pilot-scale, comparing the degree of reduction of each
substance between the suspension phase (in the absorption column) and the fixed phase
(in the regeneration unit). The results show that, in the case of formaldehyde, between 9
and 13 % removal took place in the aqueous phase, being the rest degraded in the
bioreactor. The other chemicals were degraded only in the bioreactor. The bioscrubber
was operated for more than 70 days with an elimination efficiency always above 89 %,
reaching 100 % in some cases.
Mackowiak [13] also reported interesting results regarding formaldehyde abatement
in a chipboard production industry. A pilot-scale biofilter packed with compost and wood
chaff was used for the tteatment of variable flow rates of polluted air, between 400 and
1450 m3/h, with an unspecified toxic load. Some experiments were made in order to
determine the effect of different classical organic packing materials as bark, peat and
compost, on the pressure drop registered in the biofilter. Their results showed that, at gas
loads below 350 m3/m2h, bark led to the lowest pressure drop, while at gas loads above
that value, peat gave the lowest pressure drop. Regarding formaldehyde biodegradation,
removal efficiencies between 53 and 93 % were found, with an average value around 80
%.
All these experiments prove that a formaldehyde-degrading biofilter may be started-
up in a short period of time, needing only a basic medium and little work if some
considerations are taking into account. First, an optimum environment is required to
favour the development of the microbial population. This means that all physical and
chemical parameters affecting bioreactor performance (temperature, moisture, pH,
nuttient concenttations, etc.) must be within a range that allows the optimum activity of
the biocatalyst. Moreover, the gradual increase of the pollutant concenttation, when
possible, will avoid poisoning of the microorganisms in the first stages of operation. The
presence of other volatile organic compounds in the waste gas may, sometimes,
significantly reduce the overall removal efficiency.
3.2. OPTIMIZATION OF THE REMOVAL OF FORMALDEHYDE IN BIOFILTERS

AND BIOTRICKLING FILTERS
270
In our own laboratory, the tteatment of waste gases typical from synthetic resin producing
industries is being studied. The actual air stream contains at least four different
compounds, among which formaldehyde has been identified as the dominant one,
followed by methanol. In a first study, three biofilters and one trickling biofilter were
used for the freatment of formaldehyde in mixture with methanol. As described above
(see section 2), methanol is a potential intermediate metabolite formed during the aerobic
biodegradation of formaldehyde by Pseudomonas sp. To start-up the reactors, each one of
them was inoculated with an aerobic sludge obtained from the wastewater treatment plant
of a resin-producing industry. No adaptation step was necessary since the sludge was
obtained from a wastewater already containing formaldehyde and nitrogen compounds as
major pollutants, allowing induction of the enzymes involved in the biodegradation of the
organic compound (see also section 2). By the way, studies undertaken in batch and
continuous liquid phase reactors have proven the ability of that sludge to degrade
mixtures of formaldehyde and methanol [14, 15]. Some of the most relevant
characteristics of the sludge used for the inoculation are shown in table 4.
Table 4: Physical and chemical parameters of the sludge used for the inoculation
Parameter Mean value (± Standard Deviation)
Density (g/L) 1039.4
Optical Density (diluted 50 x) 0.371 (±0.003)
P-PO43 concentration (mg/L) 153.2 (±3.5)
N-NH/ concentration (mg/L) 171.5 (±4.9)
N-NO2’ concenttation (mg/L) 2.99 (±0.04)
N-NO3‘concentration (mg/L) 0.00 (±0.00)
Total Suspended Solids (g/L) 9.95 (±0.21)
Volatile Suspended Solids (g/L) 8.75 (±0.00)
Chemical Oxygen Demand (mg/L) 233.5 (± 13.7)
pH_____________________________ 7.58 (±0.01)
All three biofilters were inoculated the same way: 2.3 L sludge was added to the
bioreactors and, after three hours, the liquid was drained off through the bottom of the
reactor, leaving time enough for the biomass to attach to the filter bed. It is important to
mention that each of these biofilters was packed with a different packing material:
volcanic earth, perlite and activated carbon. In the case of the trickling biofilter, only
volcanic earth was used. For its inoculation 750 mL sludge was continuously recirculated
in a trickling mode through the top of the reactor at a constant flow rate of 3.0 L/h. The
medium was recirculated without any treatment. After the inoculation, the gas flow rate
was set at 0.15 m3/h. Contrary to previously reported studies (see section 3.1.), nutrients
were not added to the reactors, although a very fast start-up was observed. Results
showed that all four reactors presented a high removal efficiency of formaldehyde and
methanol already immediately after inoculation. The initial loads of methanol and
formaldehyde were relatively low. Removal efficiencies of 100 % were found for
methanol in all reactors already on the first day of operation, while the formaldehyde
removal efficiency varied from 49.0 % in the perlite-biofilter to 68.6 % in the trickling
biofilter. During the first eight days operation, the formaldehyde removal efficiency
remained high in all bioreactors, being the trickling biofilter the system that showed the
highest elimination rate (Table 5). After that, the removal efficiency of formaldehyde
decreased in all reactors. 1 L of the nutrient solution described by Kennes et al. [16] was
added to each reactor on day 22 of operation, but it had only little effect on formaldehyde
271
removal.
Table 5: Formaldehyde removal in all four reactors during the first week operation
Volcanic earth- Periite- Activated carbon- Biotrickling
Load (g/m3h) biofilter
1.2 ±0.3 biofilter
1.3 ±0.6 biofilter 1.2 ±0.5 filter2.1 ±0.3
Efficiency 63.2 ±10.9 55.8 ±15.3 57.0 ± 14.0 68.1 ±2.3
(%)
Since methanol was better degraded than formaldehyde, its concentration was increased
first while maintaining the formaldehyde load relatively low. The methanol inlet
concentration was daily modified from the first day of operation, in order to simulate a
real industrial situation. Table 6 shows methanol loads and elimination efficiencies of all
biofilters during the first month operation.
Table 6: Methanol removal in all four reactors during the first month operation
_______________Volcanic earth-biofilter Periite-biofilter Activated carbon-biofilter Biotrickling filter

Max. Methanol
Load 1050.8 610.6 944.2 1634.5
(g/m3h)
Removal
Efficiency at 96.2 98.4 89.5 99.2
Max. Methanol
Load (%)
Mean
Methanol Load 397.8 ±428.5 170.4 ±218.2 300.3 ± 384.9 433.9 ±635.3
(g/m3h)
Mean Removal
Efficiency 94.5 ±7.3 95.2 ±7.2 92.9 ± 14.5 89.3 ± 19.0
(%)
As shown in table 6, all biofilters were able to remove high methanol loads, above 1
Kg/m3h, during the first month operation, even though the toxic load changed
significantly from one day to another. Results suggest that the presence of methanol in the
mixture of pollutants could have a significant influence on formaldehyde removal.
Methanol seems to be a more accessible carbon source for the microorganisms, which
may explain the relatively low formaldehyde removal. The formaldehyde concentration is
presently being increased to study the effect of higher loads of both formaldehyde and
methanol.
Acknowledgements
The research described in this chapter is being funded by projects PR404E 2000/6-0 and
PPQ 2001-0557. The doctoral research of 0JP and ME is financed by Ph.D. fellowships
of the Xunta de Galicia and the Spanish Ministry of Education and Culture, respectively.
References
[1] Kennes, c. and Thalasso, F. (1998). Waste gas biotreatment technology. J. Chem. Technol. Biotechnol.
72:303-319.
272
[2] Walker, J. F. (1964). Formaldehyde. American Chemical Society Monograph Series. Reinhold Publishing
Corporation. New York, Amsterdam, London.
[3] Bonastre, N.; de Mas, c. and Solà, c. (1986). Vavilin equation in kinetic modeling of formaldehyde
biodegradation. Biotechnol. Bioeng. 28:616-619.
[4] Adroer, N.; Casas, c.; de Mas, c. and Sola, c. (1990). Mechanism of formaldehyde biodegradation by
Pseudomonas putida. Appl. Microbiol. Biotechnol. 33:217-220.
[5] Azachi, M.; Henis, Y.; Oren, A.; Gurevich, p. and Sarig, s. (1995). Transformation of formaldehyde by a
Halomonas sp. Can. J. Microbiol. 41:548-553.
[6] Kaszycki, p. and Koloczek, H. (2000). Formaldehyde and methanol biodegradation with the methylotrophic
yeast Hansenula polymorpha in a model wastewater system. Microbiol. Res. 154:289296.
[7] Yamazaki, T.; Tsugawa, w. and Sode, K. (2001). Biodegradation of formaldehyde by a formaldehyde-
resistant bacterium isolated from seawater. Appl. Biochem. Biotechnol. 91-93:213-217.
[8] Hidalgo, A.; Lopategi, A.; Prieto, M.; Serra, J. L. and Llama, M. J. (2002). Formaldehyde removal in
synthetic and industrial wastewater by Rhodococcus erythropolis ƯPV-1. Appl. Microbiol. Biotechnol.
58:260-263.
[9] Ferranti, M. M. (2001). Formaldehyde biological removal from exhaust air in the composite panel board
industry from pilot tests to industrial plant. 35 th International Particleboard Composite Materials
Symposium. Pullmann Washington State, U.S.A. April 2-5.
[10] Doronina, N. V.; Ezhov, V. A. and Trotsenko Y. A. (1996). Aerobic biodegradation of formaldehyde,
methanol and methylamine by immobilized Methylobacterium extorquens cells. Appl. Biochem. Microbiol.
33:138 141.
[11] Huckschlag, w. (1992). Biotechnologische Behandlung Phenol und Formaldehydhaltiger Abluft. In Dragt,
A. J. and van Ham, J. (Eds) Biotechniques for Air Pollution Abatement and Odour Control Policies.
Elsevier Science Publishers BV, Amsterdam, The Netherlands, pp. 279-286.
[12] Tautz, H. and Rutenfranz, c. (1992). Biologischer Abbau toxischer Substanzen - Verfahrensauswahl und
Betriebserfahrungen mit einer Biowascher-pilotanlage. Chem. Ing. Tech. 64:192-194.
[13] Máckowiak, J. (1992). Abscheidung von Formaldehyd aus der Abluft im Biofilter. In Dragt, A. J. and van
Ham, J. (Eds) Biotechniques for Air Pollution Abatement and Odour Control Policies. Elsevier Science
Publishers BV, Amsterdam, The Netherlands, pp. 273-278.
[14] Cantó, M.; Gomez, J.; Kennes, c. and Veiga, M. c. (1998). Integrated anoxic-aerobic treatment of
wastewaters from a synthetic resin producing factory. In Proceedings of the European Conference on New
Advances in Biological Nitrogen and Phosphorus Removal for Municipal or Industrial Wastewaters.
Narbonne, France, October 12-14.
[15] Eiroa, M.; Kennes, c. and Veiga, M. c. (submitted). Simultaneous nitrification and formaldehyde
biodegradation in aerobic assays.
[16] Kennes, c.; Cox, H. H. J.; Doddema, H. J and Harder, w. (1996). Design and performance of biofilters for
the treatment of alkylbenzene vapours. J. Chem. Technol. Biotechnol. 66:300-304.
273
INDEX
activated sludge .11, 13, 14,17, 19, 20, 23, 24, 25, 26, 36, 57, 58, 60, 61, 65, 66, 69, 70,
71,72,73, 74, 84, 85, 87, 101, 102, 108, 109,110, 115,118, 119, 125,126,127, 129,
131,139,140,141, 142, 143, 149,151, 152,158, 160, 161, 166, 167,178,180, 181, 182,
183,184,185, 186, 187, 188,194,199, 201,213, 214, 217, 218, 219, 220, 237, 254, 261
aeration......20, 25, 74, 77, 78, 79, 80, 117,130,154, 163, 184, 188,195,196, 197,198,
199, 200, 201, 202, 203, 204, 207, 211, 215, 219, 220, 221
ammonification.......................................................20,106,110,139,143,148,158,162
artificial neural network................................................................ 15,17, 87, 88, 93,198
ASM1 ... 11, 12, 13,14,17, 18,19, 20, 21, 23, 24, 25, 26, 36, 37, 58, 59, 60, 62, 81, 83,
101,102,103,104, 105, 106, 107, 108, 109,110, 111, 114,115, 118, 121, 122, 123,
124,128,129,131,137,138, 140, 143, 144, 145, 146, 147,150,152, 153, 159, 160,
162,163, 164,165, 166,168, 169,170, 171,173,174, 177,192, 209
ASM3 .. 102, 103,106,107,109,110, 111, 112,113,114,115, 147, 148, 162,163, 169,
170,180,192
attached growth systems.......................................................................................219,
................................................................................................................................220
ATU......7.......................................................................................................157, 159,160
autotrophic biomass.... 19, 20, 63,103,104, 107,108,121,139, 149,153, 172, 184, 214
batch experiment..... ...................................132,138,151, 156,157,162, 180,199, 261
batch test......................................................110,130, 136,137,138,139,152,180,183
benchmark..................................................73, 74, 75, 77, 78, 79, 80, 81,82, 83, 84, 85
biodegradation 17, 102,103, 111, 140,144, 145, 146,182,185, 231, 232, 233, 235, 236,
237,243, 245, 246, 247, 248, 250, 254, 259, 260, 261, 262, 263, 264,265, 268,270,
272, 273
biofilttation.. 229, 233, 239,240, 242, 244,245, 247, 248, 249, 252, 253, 254, 255, 256,
259,260
biological characterisation............................................................ 119, 129,145, 161, 180
biological nitrogen removal. 101,187, 188, 189,191, 193,194, 195,202, 204, 205, 206,
207, 208, 210, 211, 213,214,215, 216, 221, 222
bioprocess.................. ...... . .................... . ....1, 11,13, 14, 15, 26,28, 31,37, 57, 60, 61
bioremediation..................................................................................................................1
bioscrubbers..........................................................................................................239, 249
calibration......65, 67, 69, 72, 91,101, 115,116,117, 118, 119,120,121,122, 123,124,
146, 163, 166, 168, 169,170,171,172, 180, 181,182, 183,184, 186, 197
conttol.... 11, 12,13, 16, 17,18, 19, 20, 21, 23, 24, 25, 26, 28, 36, 37,58, 59, 60, 61, 65,
73, 74, 75, 77, 78, 79, 80, 81, 82, 83, 84, 85, 87, 88, 90, 91,92, 94, 95, 96, 98, 99, 100,
143,161, 182,185, 187,188,189,193,194, 195, 196,197, 198, 199, 200,201, 202, 203,
204, 205, 207, 208, 209, 210, 211, 212, 213, 217,219, 221, 222, 224, 237, 240, 241,
242, 244, 245, 246, 251, 252, 253, 254, 255, 256, 257
conttol strategies.. 17, 60, 61,73, 74, 75, 78, 79, 81, 82, 84, 85,143, 195,196, 198,199,
202,207, 221
correction factors..................................................................................................109, 170
decay..22, 23, 25, 58, 62, 63, 67, 102, 103, 106, 108,109, 110, 111, 114, 115, 117,125,
275
127,132,134, 146,149, 152,153, 154,156, 163,166,167,169,173, 174,183, 184,
185,186,191, 192, 214, 215 .................................
decay rate.................. 62, 67,108, 110,117, 149, 153, 156,166, 169, 184, 191,192, 215
denitrification .. 13,18,19, 20, 21, 22, 24, 25,60, 75,107,109,110,112, 114, 118,121, 129,
134, 135,140,143, 161,162, 170, 181, 182,183,185, 188,190,191, 192,193, 194,195,
196,197, 198, 199, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210,
211,217,218,219,220,221,223
endogenous respkation.. 83, 108, 110,112, 114,115, 137,152, 153, 157,163,169,173,
174,175, 176,181,185, 186, 192, 211, 219
enzymology.....................................................................................................................1
evaluation of conttol strategies.......................................................................................73
experimental design............................................................. 101,137,158,159, 179,184
external carbon addition...............................................................................17, 18, 60,
193
fault detection........................................................................................88, 89,93, 99, 100
fuzzy logic....................................................................................................87, 88, 93,99
growth.... 17,19, 22, 23, 25, 29,43,58, 62, 67, 68, 74, 79,102, 103,104, 106, 107, 108,
109,110, 111, 112, 113,114,117, 120, 121, 132,134,137, 139,147,151, 152, 154,
156,157, 158,161,162, 163,166, 167, 168, 169,170, 172, 173, 174, 175, 176,178, 179,
182,183,184, 186, 187,188, 212, 217, 218, 219, 220, 221, 222, 232, 233,234,
241,252,262,264 ....
half-saturation concentration................................................................ 154,156, 157,161
heterottophic biomass....20, 63, 66, 67, 68,103, 104,107,108,109,114, 132, 138,139, 140,
152, 153,154,157,172, 182,183,192
hydrolysis .... 18, 20, 21,63, 106,108, 109,110, 111, 115, 124,125,131,132, 137,138, 139,
147, 157,159,161,163, 164,169, 170,171, 172
hydrolysis rate................... ......................................108,125,137, 138,139,157,164
identifiability..................................12, 17,21,60, 115,116,117, 173, 181,183, 184,186
kinetic parameters..60,109,110,121,150,152, 155, 157, 158, 159,160, 161, 162, 163, 165,
170, 171, 172,173,180, 182,183,189,199 . . ..
mathematical model.....................................................................61,72, 88, 99, 182,184
maximum specific autotrophic growth rate........................................................157,167
maximum specific heterottophic growth rate................................................................167
membrane.....................................................................125, 138, 147,231, 233, 236, 257
model reduction . 11, 12,13, 14, 15, 16, 17, 20, 25, 28, 30, 32, 37,51, 57, 58,59, 61, 62
modellingll, 12, 15, 16,17, 25, 31, 61, 62, 74, 75, 83, 99, 108, 117, 118, 120, 172,173,
182, 183, 184,185,186
Monod . . ......... ................17,19, 21, 22, 25,43,44,45, 50, 52, 57, 59,60,107, 154, 252
nitrate utilisation rate............................129, 130, 140,145, 150, 160,161, 162, 170,185
nitrification....15, 17, 18, 19, 21, 22, 60, 82,92, 99,107,109,110,114,118, 121,129,
134,135,138, 139,140,141, 142,143,144,146,151,152,157, 158, 159,160,161,
162,167,180, 181, 182,187,188,189, 190,191, 193, 194,195, 196, 197,198, 199, 200,
201, 203, 204, 211, 212, 213, 214, 215, 216, 217, 218, 219,220, 221, 273
nitrogen removal.. 13,17, 20, 61,102,121,183, 185, 186,187,188, 189, 193,194, 195, 196,
197, 198,199, 204, 206,211, 213, 216, 219,221, 222
optimisation.... 12, 13, 20, 25, 59, 116,117, 121,133, 157,186, 187, 195, 204, 211, 213
order reduction.........................................................11,14,16, 26, 28, 31, 37, 38,58,59
276
parameter estimation.............. 115, 116,173,209
performance assessment........ ................. 82
process design........................ 118, 187,188, 189, 194, 195, 213, 221, 222
quasi steady state assumption ............ . .
.................... 16, 58
respừation rate.....73, 74, 78, 79, 80, 82, 83, 85, 114,129, 131,133, 136, 137,138, 139,
140,150,152, 153, 154,155,157,158,159,160, 173,183, 184
respừometric 74, 78, 81, 83, 127,133,136,138, 139,144, 146,148, 149, 150, 151,152,
153,155,158,159, 161, 162, 163,170, 173, 180, 181,184, 185
respirometry... 73, 74, 75, 78, 81, 82, 83, 84, 85,126,129, 130,131, 139,140, 143, 145,
146,149,150, 152, 161, 162, 164,182,184, 185, 199, 222
simulation 11, 12, 53, 54, 55, 69, 70, 71, 73, 74, 75, 76, 77, 78, 83, 84, 85, 95, 109, 148,
181, 203, 204, 208, 209, 210, 213, 215, 237
singular perturbation..... 11,14, 16, 24, 26, 27, 28, 30, 31, 32, 36, 37, 38, 51, 59, 61,62
sludge composition...............................................................................................148, 180
So/Xo....................................................................................................................181,183
SRT... 79, 80, 81, 109,110, 188,189, 190,191, 192, 193, 194,195, 211, 212, 213, 214,
215, 216, 217, 218, 221, 223, 224
storage ...90,110, 111, 112, 113, 114,115, 147,151, 162, 163, 169, 170, 176, 178,179,
182, 183, 192, 214, 215, 220, 262
switching functions.......................................................................................................158
tittimetric.......................................................129, 141,143, 161,162,180,182,183,186
tittimetry........................................................................................130, 146,150, 161,
162
transferability....................................................................... 101, 166,167,168,173,181
trickling filters..............................................................................................217, 241, 242
wastewatercharacterisation.nl, 118, 121, 128,129, 130,133, 140, 141, 143,144, 146, 147,
150,163, 171, 180,185,186
wastewater tteatmentll, 12,19, 25, 38, 61, 65, 66, 67, 69, 70, 72, 73, 74, 84, 85, 87, 99,
101, 125,131, 181, 182,183,184, 197, 201, 214, 246, 247, 251, 252, 259, 262, 270
yield coefficient.... .67,109, 136,139, 146,148, 149, 150,151, 152, 154,157,171, 172
• the daily oxidized nitrogen in the mainstream system is RN (mass N/day). Note that
RN is smaller than the influent nitrogen loading rate because part of the loaded
nitrogen is assimilated into biomass cells, and that,
• the daily oxidized nitrogen in the side stream reactor is RN,rej = S*RN (mass
N/day)
277

@@@@@ Biotechnology For The Environment

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

@@@@@ Biotechnology For The Environment

Uploaded by

Copyright:

Available Formats

FOCUS ON BIOTECHNOLOGY

Biotechnology for the

Springer-Science+Business Media, B.v.

SPRINGER-SCIENCE+BUSINESS MEDIA, B.v.

Printed on acid-free paper

All Rights Reserved

Spiros N. Agathos Walter Reineke

1.1. NEED FOR REDUCTION OF RIGOROUS, MECHANISTIC MODELS

Dynamic simulation based on rigorous, physics-based mechanistic modelling has become

analysis in electrical engineering and dynamic flowsheeting in chemical engineering. In

1.2. PROBLEM STATEMENT AND METHODOLOGY

provides the starting point for this method.

1.3. REDUCTION APPROACHES FOR PROCESS ENGINEERING SYSTEMS

1.3.1. New model building from "scratch"

1.3.2. Simplifying assumptions

13.4. Order reduction methods

7.3.5. Black-box identification

2. Reduction approaches for ASM1

2.1. NEW MODEL BUILDING FROM ‘SCRATCH’

Marsili-Libelli (1989) developed a low-order model for conventional activated sludge

This relational model is estimated from measured DO and nitrification rates in

2.2. SIMPLIFYING ASSUMPTIONS

2.2.1. Simplification with respect to components

• No suspended organic bound entrapped N (No XND)

• Do not consider alkalinity

2.2.2. Simplification due to separation of aerobic and anoxic conditions

• No denitrification under aerobic conditions

2.2.3. Simplifying assumptions with respect to kinetics

2.2.4. Simplification with respect to dynamics

2.2.5 Reduced order models

• One overall maximal denitrification rate rđenit,max is used.

d$NO _ --r. . . SNO_________________rload V /Ọ

• For nonlinear (open-loop) optimisation (optimisation horizon 1 day, Lukasse et al.,

• timescale of interest of hours. Assumptions

d I SNH j _ I -rNH , I DinSNH,in

• Do not consider alkalinity

• Approximate Monod expressions by Blackman kinetics (it is assumed that substrate

SNH =-iXB(rH * XCODXBH -bHXBH —^AXBA)

XC0D = -•77“rH XCODXBH +bHXBH +bAXBA

= D(So,in - So) + kLa(So - So) - rs0 dt u

2.3. DYNAMICS OF VARIABLES

2.5. BLACK-BOX IDENTIFICATION

Lindberg (1997) applied subspace state space identification to identify a fourth-order

2.6. DISCUSSION AND CONCLUSIONS

3. Singular perturbation of bioprocess systems: theory and review

Singular perturbation application to bioprocess systems is reviewed in this section. This

3.1. SINGULAR PERTURBATION THEORY

Let the system be described by n+m equations in state-space notation

X = f(x,z,u,t,e), x(t0) = x0,xe Rn , ueRp (3.1)

ez = g(x,z,u,t,e), z(t0) = z0,ze Rm. (3.2)

Condition 3.3: The eigenvalues of 3g / 3z evaluated, for 8 = 0, along x(t),z(t), have

3.2. REVIEW OF MODEL REDUCTION OF (BIO)PROCESS SYSTEMS USING

3.2.1. General rule for order reduction

Ặ = z (±)kij<pj - D^i - Qi + F1 ’ (3.5)

22(±)kijcPj =Qi -Fi . (3.6)

-7- = (p-DS + DSin dt (3.7)

= k'Q-DS + DSin- dt (3.9)

In this case, the rule holds (for psat sufficiently small).

the component balance equations are written as:

= - ki |1X - D s + D Si, (3.11)

However, it can be argued the volume is not the adequate perturbation

3.2.2. Other reduction approaches

3.3. SCALING IN MODEL REDUCTION