- Email: [email protected]
DEPUR: A knowledge-based tool for wastewater treatment plants
EngngApplic. Artif. Intell. Vol. 7, No. 1, pp. 23-30, 1994 Pergamon 0952-1976(93)E0004-R
Copyright© 1994ElsevierScienceLtd Printed in Great Britain. All rights reserved 0952-1976/94 $6.(X)+ 0.00
Contributed Paper DEPUR: a Knowledge-based Tool for Wastewater Treatment Plants PAU SERRA Universitat Aut6noma de Barcelona MIQUEL SANCHEZ Universitat Polit~cnica de Catalunya JAVIER LAFUENTE Universitat Aut6noma de Barcelona ULISES CORTI~S Universitat Polit6cnica de Catalunya MANEL POCH Universitat Aut6noma de Barcelona
(Received December 1992; in revised form May 1993) A malfunction of a wastewater treatment plant is a major social and biological problem. Poorly treated waste water outside the plant could provoke dangerous consequences for human beings as well as the environment itself. The conventional control systems, that are usually applied in this field, have to cope with some difficulties: complexity of the system, an ill-structured domain, qualitative information, uncertainty or approximate knowledge, real-time dynamic system . . . . This paper shows an application of artificial intelligence in order to help the operators of wastewater treatment plants in their task of process control. The main goal is to build a knowledge-based tool useful for the diagnosis and management of wastewater treatment plants. First, a survey of wastewater treatment plants describes the complexity of the system being modelled and outlines its difficulties. The development of the application, and the methodology employed in it, are discussed. A new methodology called L1NNEO + is introduced. It is used for the automatic knowledge acquisition process in order to build up a Knowledge Base. The prototype architecture constructed - D E P U R - is detailed, together with some obtained results. Keywords: Knowledge-based systems, process diagnosis and m a n a g e m e n t , wastewater treatment, knowledge acquisition tools, environmental engineering, biotechnology.
1. I N T R O D U C T I O N The aim of this work relies on the application of new techniques in the fields of automation and process control. The A I tools, as seen below, are very important aids in the automation of process management. A malfunction of a wastewater treatment plant is a big social and biological problem. Poorly treated waste water outside the plant could provoke dangerous
consequences for human beings, as well as for the natural environment itself. The setting of an automatic process control over the wastewater treatment plant system, has outlined some difficulties:
• The complexity of the problem. There are many facts of different natures present at the system: chemical facts, mechanical, biological, microbiological ones . . . . • A non-well structured domain. The relationships among the concepts or attributes of the domain
should be sent to: Dr P. Serra, Universitat Autbnoma de Barcelona, Unitat d'Enginyeria Bioqufmica (CSICUAB), Edifici C. E-08193 Bellaterra, Barcelona, Catalonia, Spain.
Correspondence
23
24
PAU SERRA et al.: DEPUR are not well-enough known, or the agreement among experts is not very high. • Most information is neither numeric nor quantified. It is therefore not suitable to be included in the context of a classical-numerical-control model (microbiological information . . . . ). • Uncertainty or approximate knowledge. For example, the subjective information supplied by the experience of the plants' managers, is often vague or uncertain. • A dynamic system. Continuous changes that directly modify the performance of the process cause a further complication.
Moreover, it can be said that the classical methods work well on the normal states of the plant but not in other abnormal states of working. How could they detect some mechanical faults? How could they use the available but incomplete information, to solve a specific problem? These problems with the conventional process-control systems such as feed-back control, feedforward control, optimal control, adaptive control, . . . have given rise during the last few years to considerable research effort in Artificial Intelligence.' 4 Bearing in mind the main advantages and characteristics of knowledge-based systems, specially those of expert systems, and the above difficulties, one could expect that they would match very well. So a prototype of a qualitative expert system with uncertainty reasoning has been developed, as well as a real time numerical control expert system, which is described in Ref. 5, for the diagnosis and management of wastewater treatment plants. 2. A W A S T E W A T E R T R E A T M E N T PLANT
The main goal of a wastewater treatment plant is to reduce the level of pollution of the wastewater at the lowest cost, that is, to r e m o v e - - a s far as possible-strange compounds (pollutants) of the inflow water to the plant, prior to discharge to the environment. So, the effluent water must have low levels of the pollutants (lower than maximum ones allowed by the law). The plants taken as models are based on the main biological technology usually applied: the activated sludge process. ~ The concrete wastewater plant studied is located in Manresa, near Barcelona (Catalonia). This plant receives about 35,000 m3/day inflow from 75,000 inhabitants. A wastewater treatment plant is usually composed of two stand-alone but interactive subsystems: sludge line and water line. This study has focused on the water line. It is formed by a sequence of unitary processes where the effluent of one process becomes the inflow to the next one. Usually there are 3 major processes: primary treatment, secondary treatment and tertiary treatment. Each one reduces or removes the concentration of several specific pollutants. The flow chart of a plant may be seen in Fig. 1.
Sand
INFLOW
PRE-TREATMENTJ
t Oil
BIOLOGICAL
~
PRIMARY
PROCESS
~,~
BiologicalSludge
~
PrimarySludge (sedimentable suspended solids )
TERTIARY TREATMENT EFFLUENT
Other pollutants
Fig. 1. Chart of a wastewater treatment plant.
3. DEVELOPMENT OF THE QUALITATIVE EXPERT SYSTEM An expert system, called D E P U R , 78 has been built in order to support the approximate or uncertain reasoning, using qualitative knowledge, in the managing of wastewater treatment plants. The main problem appearing on a wastewater treatment plant is the breakdown of the balance between biomass and substrate, as shown in Fig. 2. The biomass is formed by a set of microbiological beings that in presence of oxygen, degrade non-sedimentable organic suspended solids (i.e. the substrate) in the water. More information about the wastewater treatment processes can be found in the classical literature. 9~,0 Thus, the task of diagnosis of the system is to determine the causes of the problems detected. Often, the most serious problems are related to the biological process in the secondary treatment. The design process of a knowledge-based system can commonly be expressed as shown in Fig. 3. The work described here has been focused on the water line subsystem domain. The relevant attribute selection process has been developed in a twofold fashion: firstly, in a typical manner based on the attribute selection made by the experts in the domain, and, secondly, by means
BIOMASS SUBSTRATE + 02
=,,
BIOMASS + CO2 + H20
Fig. 2. Biological balance.
PAU SERRA et al.: DEPUR
25
4. THE LINNEO + M E T H O D O L O G Y FOR THE DOMAIN SELECTION
K N O W L E D G E ACQUISITION PROCESS
DOMAIN FOCUSING
v* RELEVANT A'i-I'RIBUTE SELECTION
DATA BASE
DOMAIN ~ MODELLING ATTRIBUTE STRUCTURING
KNOWLEDGE BASE
x,,. IMPLEMENTATION
NO-~OKI
VALIDATION
lOK
SYSTEM USE
Fig. 3. Design process of a Knowledge-Based System.
of a new automatic attribute selection approach based on the relevance and potential utility concepts." The attribute structuring process has been made using two approaches:
• Decision trees: the classical approach derived from the interactions between the expert and the engineer. This approach is governed by the causeeffect criterion. Multi-valued decision trees have been used, and inference rules are obtained from them. The rules point to the causes of a certain problem or situation previously detected. • Automatic classification techniques: used as an automatic knowledge acquisition tool, this is an approach recently derived from machine learning in Artificial Intelligence. The L I N N E O ÷ system, based on fuzzy non-well-structured domain classifications, has been used in this phase. From the characterization of the classes obtained, moregeneral inference rules are constructed. They lead to knowledge of the most likely working situation of the plant. This methodology will be explained more clearly in the next section. Decision trees have been used in the cause identification phase, and automatic classification techniques in the problem identification process, as shown in Fig. 4.
The L I N N E O + software 12' ,3 is a knowledge acquisition tool which is useful in ill-structured domains, and which uses both inductive learning methods and deductive ones. L I N N E O + uses (possibly noisy) observations to induce classes, as well as taking advantage of the expert's current state of knowledge (which may be incomplete), expressed as logical constraints on the possible output of the classification process. Figure 5 shows the general framework of the methodology. The human expert abstracts a collection of observations (sample) from the problem domain that he thinks of as relevant. He also selects a set of attributes (descriptors) that he assumes to be significant for the characterization of the domain objects. A sample is then modelled by a set of attribute/value vectors. At the same time, the expert can express what he already knows about the domain. This previous knowledge about the domain (that will be called Domain Constraints--DC) is represented as constraints and examples on classes already known to exist. Starting with this knowledge and information, L I N N E O + produces a classification. It uses induction over the observations, taking into account the DC which guide the inductive process. The results of this classification are an intensive and extensive description of the generated classes and a fuzzy membership matrix that relates observations to generated classes. The expert evaluates the results and possibly modifies some attributes or some DC. When a "correct" classification is obtained, rules are derived from the generated classes. Each rule includes in its antecedent the membership conditions and a fuzzy certainty factor for a particular class. Each antecedent is a conjunction of logical conditions on the
ATTRIBUTE III SYSTEM STATE CHARACTERISTICS
SIT1 SIT2 SIT3 ... SITN ~
O1.1 01.2 ... O2.1 ... CN.N
SITUATIONS
---- CAUSES
l lll SOLUTIONS ----
$1.1 $1.2
$2.1
SN,N
Fig. 4. Diagnostic process in an Expert System.
PAU SERRA et al.: DEPUR
26
Bulking-not-enough-oxygen Bulking-sulphurs Rising Foaming Overaeration - - Summer Winter - - Storm High plant Rotation-band-crash Primary-sludge-sedim.-problem Not-enough-primary-sludge-exit Overloading Turbine-crash By-pass-bad-closed - -
- -
STEPSOFTHE METHODOLOGY
- -
.l
bs/~action
- -
ATrRIBUTE SELECTION AND OBSERVATIONS Observations: Attdbute/Value matrix
- -
- -
Ngw attributes/ observations
- -
i n f l o w
- -
- -
FUZZY CLASSIFICATION
- -
A'II'R.IBUTE RELEVANCE MODULE
~e
- -
- -
/
Classes
att butcs/ ob ervations
New
- -
- -
- -
GER~FAsTION "'J
KNOWLEDGEI BASE
Fig. 5. The methodology of LINNEO ÷.
v a l u e s t h a t d e s c r i p t o r s m u s t show for m e m b e r s h i p of the class w h o s e identifier a p p e a r s on the r u l e ' s conseq u e n t . T h e s e rules are a n a l y s e d a n d , t h r o u g h s u b s u m p tion d e t e c t i o n a n d s y n o n y m y analysis, t h e y are organized in a h i e r a r c h i c a l way. T h e results are m a d e k n o w n to t h e user, w h o can a c c e p t a n d confirm t h e m . H e also has the c h a n c e to go b a c k to the classification m o d u l e , to t h e definition o f the D C o r to the s a m p l e selection s t e p , a n d o b t a i n o t h e r results. W h e n all these a c t i o n s a r e o v e r , rules can be v a l i d a t e d using new o b s e r v a t i o n s n o t p r e v i o u s l y in the s a m p l e . A s s o o n as the e x p e r t a c c e p t s the resulting rule set, it is transf o r m e d into I F - T H E N rules a n d m o d u l e s .
5.
ARCHITECTURE
OF
THE
APPLICATION
T h e m o d e l l i n g o f the d o m a i n has b e e n m a d e as follows: objects o r observations are p r o t o t y p i c a l w o r k ing s i t u a t i o n s o r p r o b l e m s in the w a s t e w a t e r p l a n t of Manresa, near Barcelona (Catalonia). The experts d e f i n e d a b o u t 19 d i f f e r e n t situations at the b e g i n n i n g of the w o r k . T h e attributes o r properties a r e the m o s t r e l e v a n t p a r a m e t e r s a n d f a c t s - - i n the v i e w p o i n t of the e x p e r t s - - t h a t c h a r a c t e r i z e the d o m a i n . T h e y have s e l e c t e d a b o u t 20 a t t r i b u t e s . So, the input m a t r i x to the classification p r o c e s s o f L I N N E O + is as follows: •
Objects-situations-defined: - -
- -
Normal Bulking-non-filamentous
T
o
x
i
c
-
s
u
b
s
t
a
n
c
e
s
-
l
o
a
d
i
n
g
Toxic-substances-increase
• A t t r i b u t e s defined: -- Rotate-band-clarifiers - - Bubbles-clarifiers -- Bubbles-primary-settlers -- Inflow-water -- Dissolved-Oxygen -- Inflow-DQO -- Outflow-DQO -- Sludge-age -- Foam-bassins -- Filamentous --SV1 -- Odor-bassins -- Sulphurs -- Sludge-exit-primary-settler - - S u s p e n d e d Solids-bassins - - S u s p e n d e d Solids-inflow - - S u s p e n d e d Solids-outflow - - V o l a t i l e S u s p e n d e d Solids-bassins - - V o l a t i l e S u s p e n d e d Solids-recirculat. -- Temperature -- Turbines -- Toxic-substances-bassins -- Toxic-substances-inflow A n e x a m p l e of a c o m p l e t e o b s e r v a t i o n with its values for the a t t r i b u t e s m a y be like this o n e : •
Bulking-non-filamentous: Outflow-DQO --~ H i g h Sludge age --->O l d Filamentous ~ Normal SVI ~ High Volatile SS-recirculation ~ Low All-other-attributes ~ Nought-value
T h e o t h e r a t t r i b u t e s have nought values '4 o r don't care values. This m e a n s that the values o f these attrib u t e s are not r e l e v a n t for the c h a r a c t e r i z a t i o n of the o b j e c t , a l t h o u g h t h e y c o u l d be r e l e v a n t to d e s c r i b e o t h e r w o r k i n g situations of the plant. F o r instance, in the a b o v e e x a m p l e , the value of the a t t r i b u t e tempera-
PAU SERRA et al.: DEPUR
27
ture could be any o n e I b u t don't care which it is--for the description of the bulking situation originated by non-filamentous causes.
tl
5.1. Results
Ten classes have been obtained (see Fig. 6) as a result of the classification process. Each one groups closely related prototypical situations. The classification shows the most common situations and problems in the working state of the wastewater treatment plant, and simplifies the arrangement of the underlying domain. These classes lead the inference process in the diagnosis.
KNOWLEDGE BASE
INFERENCEENGINES
6. I M P L E M E N T A T I O N OF DEPUR IN MILORD
MILORD ~5'~ is a classical second-generation expertsystem shell, but consideration has been given to some important features related to this particular domain, such as the microbiological and temporal ones. The application developed is called DEPUR, and reflects the experts' knowledge (see Fig. 7). The Data Base is a representation of the current state of the system, that could include some past or future facts of interest. In this representation the attributes most relevant to the domain characterization are selected. The dictionary of the Data Base is composed of 167 attributes, with a decrease by means of relevant and potential utility concepts. An example of dictionary entries may be:
CLASS- 1 CLASS-3 TANC
N-FI
[ BULKING-NOT-ENOUGHOXYGEN~ BULKING-SULFURES ] \ FOAMING I ~-BAD-Cy
CLASS-4
CLASS-I 0 CLASS-6
CLASS-9
T~URBINE-CR~H~
CLASS-@ CLASS-5
@ CLASS-8 ~E-SEDIME~
Fig. 6. Resultsof the classiticationprocess.
CONTROLMODULE
Fig. 7. General architectureof DEPUR. (INFLOW-WATER-NORMAL "Has the inflow water normal (35000-45000) m3/day?'' FUZZY)
values
(LOW-TEMPERATURE "Is the temperature low ( < 5)°C? '' FUZZY) The Knowledge Base is the arrangement of the knowledge available over the domain, made explicit by the expert and by information coming from the analysis of the system being modelled. Usually, this knowledge is expressed by means of inference rules. Examples of rules may be: (RMENU003B (IF (DQO-EFLUENT-WATER-HIGH SLUDGE-AGE-YOUNG)) SURE (THEN (INFER (CLASS-3B)))) (RC3004 (IF (CLASS-3A DISSOLVED-OXYGEN-LOW ODOR-BASSINS)) SURE (THEN (INFER (BULKING-O2))))
PAU SERRA et al.: DEPUR
28
The system is formed of 79 rules, grouped in 12 modules. A module is a structural element of the Knowledge Base, composed of a set of related rules. The Control Module has the goal to control how and when the knowledge supported by the system may be used. Often it is expressed by means of metarules. The prototype has 15 metarules. An example of a metarule may be:
KNOWLEDGE-BASEDSUPERVISORSYSTEM
(M003B (IF (CLASS-3B)) SURE (THEN (TREAT-MODULES C3)))
l
The linguistic terms used to manage the approximate knowledge are, in order of certainty: Sure Almost-sure Very-possible Possible Moderately-possible Slightly-possible Almost-impossible Impossible
7. VALIDATION The prototype diagnostics, in contrast to the answers of human experts, have yielded very good results. They have been tested with several prototypical situations in the working of the plant (bulking, foaming . . . . ). Currently, this classification is validated against a data stream obtained daily from direct observation in the wastewater treatment plant over the last two years, with very high degree of performance. A short dialogue with the application could be like the following one: > Has the effluent DBO high values (25-40) mg/l? -Sure > Has the water inflow high values (45,000-55,000) m~/day ? -Almost impossible > Is the temperature low (<5)°C? -Impossible > Has the inflow water high concentration of Zn (4-20) mg/l? -Impossible > Is the Volumetric index of sludge (SV1) high (150-300) rag~l? -Very possible > Are the Suspended Solids (SS) in the recirculation low (1300-4600) mg/l -Sure
EVOLUTIVE DATA BASE
NUMERICAL CONTROL ALGORITHMS
\
/
~IAL
The modularity of the application, the selfexplanation and the user-friendly interface afforded by the shell are other outstanding properties.
II
TI
KNOWLEDGE BASE
PR~E~
Fig. 8. A SupervisoryKnowledge-BasedSystem. > Are the Volatile Suspended Solids (SSV) in the recirculation low (100-3500) mg/l? Sure > Has the dissolved oxygen (DO) high values (2.5-10) rag~l? Impossible > Is the age of sludge old (8-20) days? Impossible > Conclusion Bulking-not-enough-oxygen--Very possible 8.
C O N C L U S I O N S
A N D
F U T U R E
W O R K
The building of the DEPUR system as a prototype fuzzy expert system applied to managing a wastewater treatment plant, has been the consequence of the research effort during recent years (study and modelling of wastewater treatment plants, development of new techniques of selection and arrangement of attributes, building of computer systems . . . . ) by both scientific groups involved. A diagnostic and management tool based on knowledge-based systems for a wastewater treatment plant may be a useful approach in the field of intelligent systems applied to managing industrial processes. The prototype may be helpful to the manager of the plant. The treatment of uncertainty and the new fuzzy techniques for automatic classification are some of its characteristics. This methodology
PAU SERRA et al.: DEPUR has b e e n successfully a p p l i e d to o t h e r d o m a i n s (see R e f s 11, 12). This p r o j e c t p r o v i d e s o n e s t e p t o w a r d s the b u i l d i n g of i n t e l l i g e n t s y s t e m s o r i e n t e d to c o n s e r v e w a t e r , as a p a r t of n a t u r e (intelligent e c o l o g i c a l systems). A n o t h e r i n t e r e s t i n g p o i n t is the possibility to a d a p t the system to the specific c h a r a c t e r i s t i c s of the actual w a s t e w a t e r treatment plant under control. In t h e f u t u r e , it is p r o p o s e d to r e a c h a s e c o n d step: the d e v e l o p m e n t o f a S u p e r v i s o r K n o w l e d g e - B a s e d S y s t e m - - m o r e p o w e r f u l t h a n an e x p e r t s y s t e m - - t h a t c o m b i n e d i a g n o s i s a n d c o n t r o l , including the n u m e r i c a l c o n t r o l a l g o r i t h m s ~7with q u a l i t a t i v e a n d heuristic inform a t i o n a n d r e a s o n i n g (see Fig. 8). A l s o , t h e r e is m u c h w o r k to do: to e x t e n d t h e d o m a i n to the sludge line, to r e s e a r c h the m i c r o b i o l o g i c a l c o m p o u n d o f the biological p r o c e s s in a m o r e - a c c u r a t e w a y , a n d to a c h i e v e b e t t e r t e m p o r a l i n f o r m a t i o n t r e a t ment.
Acknowledgements--This work has been supported by the "Junta de Sanejament de la Generalitat de Catalunya" and the Spanish CICyT projects ROB89-0479-C03-02, ROB91-1139 and TIC91-1087-C03-01. Also, the authors wish to acknowledge the cooperation of Ricard Tom~s, manager of the Manresa's wastewater treatment plant.
REFERENCES 1. Coleman J. J., Krovvidy S., Scott Summers R. and Wee W. G. An AI approach for wastewater treatment systems. J. Appl. lntell. I, 247-261 (1991). 2. Floris V. Artificial intelligence in the operation and management of water resources in South Florida. Personal Communication (1989). 3. Maeda K. A knowledge-based system for the wastewater treatment plant. Future Generation Comput. Syst. 5, 29-32 (1989). 4. Sanz R., Aguilar J., Sierra C., God6 LI. and Ollero A. Adaptive control with a supervisor level using a rule-based inference system with approximate reasoning. Artif. lntell. Sci. Comput. towards 2nd generation systems. IMACS (1989).
29
5. Serra P., Sfinchez M., Lafuente J., Cort6s U. and Poch M. ISCWAP: a knowledge based system for supervising activated sludge processes. Submitted for publication. 6. Robust6 J. Modelitzaci6 i identificaci6 del proc6s de fangs activats. Ph.D. Thesis, Departament de Quimica, Universitat AutSnoma de Barcelona (1990). 7. S~mchezM., Serra P., Belanche LI. and Cort6s U. A knowledgebased system for the diagnosis of waste-water treatment plants. Proc. of 5th International Conference on Industrial & Engineering Applications of Artificial Intelligence and Expert Systems (lEA~ ALE-92). Lecture Notes in Artificial Intelligence 604, pp. 324-336. Springer Paderborn (1992). 8. Sfinchez M. DEPUR: aplicaci6 dels sistemes basats en el coneixement al diagnSstic en plantes de tractament d'aigiies residuals. Master Thesis, Department de Llenguatges i Sistemes Informhtics, Universitat Polit~cnica de Catalunya (1991). 9. Benefield L. D. and Randall C. W. Biological Process Design for Wastewater Treatment. Prentice-Hall, Englewood Cliffs, NJ (1980). 10. Metcalf and Eddy Inc. Wastewater Engineering: Treatment~Disposal~Reuse, 3rd edn. McGraw-Hill, New York (1991). 11. Belanche LI. and Cort6s U. The nought attributes in knowledgebased systems. Proc. of the European workshop on Verification & Validation, EUROVAV-91, pp. 77-103. Jesus College, Cambridge (1991). 12. B6jar J. and Cort6s U. LINNEO÷: Herramienta para la adquisici6n de conocimiento y generaci6n de reglas de clasificaci6n en dominios poco estructurados. Actas 3er Congreso Iberoamericano de lnteligencia Artificial, IBERAMIA-92, pp. 471-481. La Habana, Cuba (1992). 13. Martin M, Sangiiesa R. and Cort6s U. Biasing induction with previous knowledge for knowledge acquisition in imprecise domains. Proc. Avignon '91, pp. 359-370 (1991). 14. S~inchezE. Importance in knowledge systems. Information Syst. 14, 455-464 (1989). 15. God6 L., L6pez dc Mtintaras R., Sierra C. and Verdaguer A. MILORD: the architecture and management of linguistically expressed uncertainty. Int. J. Intell. Syst. 4,471-501 (1989). 16. Sierra C. MILORD: arquitectura multi-nivell per a sistemes experts en classificaci6. Ph.D. Thesis. Department de Llenguatges i Sistemes lnformhtics, Universitat Polit~cnica de Catalunya (1989). 17. Moreno R., de Prada C., Lafuente J., Poch M. and Montague G. Non-linear predictive control of dissolved oxygen in the activated sludge process. In Modelling and Control of Biotechnological Processes, pp. 289-293 (Karim M. N. and Stephanopoulos G., Eds). Pergamon Press, Oxford (1992).
AUTHOR'S BIOGRAPHIES Pau Serra i Prat graduated in Chemical Engineering from the Universtitat AutSnoma de Barcelona in 1988. Since 1988 he has researched in the area of control of environmental processes. In 1993 he presented his Ph.D. Thesis on the development and application of a knowledge-based system for the management of a wastewater treatment plant. Dr Serra is currently working in an enterprise related to the exploitation of potable and waste water treament plants in Spain. Miquel Sknehez i Marr/~ (Barcelona, Catalonia, 1964) graduated in Computer Science at the Technical University of Catalonia (UPC) in 1989. In 1991 he presented his Master's thesis, and is currently working towards his Ph.D. He has been on the staff of the Department of Computer Systems and Languages since 1988. Since 1989 he has participated in the Interest Group on EuLISP, funded by the ECC and has also been working on projects supported by the Spanish Council and the ECC. His main research topics are distributed and concurrent AI, knowledge engineering, automatic knowledge acquisition, real-time systems and LISP-like languages. Javier Lafuente Sancho, Associate Professor in Chemical Engineering at the University AutSnoma de Barcelona, received his Ph.D. in the area of water quality modelling. This work is now being used by the Catalan Government. He undertook NATO-funded postdoctoral research work at MIT, and subsequently has been involved in the EC BAP project on the monitoring and control of bioreactors. He is a member of the "European Laboratory Without Walls". His work in Artificial Intelligence has recently been enhanced by a British Council scheme in collaboration with the University of Newcastle-upon-Tyne, U.K. He is currently involved in a Spanish project in the area of "Control of Municipal Waste Water Treatment Plants". Ulises C o r t ~ (Mexico D.F., M~xico 1960) received a B.Sc. in Industrial and Systems Engineering at the Instituto Tecnol6gico y de Estudios Superiores de Monterrey in 1982, and a Ph.D. from the Technical University of Catalonia (UPC) in 1984. He is an Associate Professor in the Software Department of the UPC, Head of its Artificial Intelligence Section, and an Adjunct
3(I
PAU SERRA
et al.:
DEPUR
Researcher at the CSIs Center for Advanced Studies of Blanes (Spain). He participates in the ECC-funded Interest Group on EuLISP, and is the Spanish representative on the JTC1/SC22/WG16 LISP ISO Group. His main research topics are machine learning, knowledge acquisition and LISP-like languages. He has an extensive list of publications, conference papers, and tutorial and supervisory contributions. Manel Pooh Espallargas, Associate Professor in Chemical Engineering at the Universitat Autbnoma de Barcelona, is the author of more than 50 papers, 30 congress communications and advisor to eight Ph.D. Theses. He has been involved in monitoring and control of biotechnological and environmental processes during the last 10 years, being responsible for several Spanish research projects in this topic, and a researcher in an European project about advanced monitoring and computer control of biotechnological processes. He has been responsible for several contracts with companies in monitoring and control projects involved in biotechnological and environmental topics. His main research areas are related to the optimal control and management of biotechnological processes including monitoring, control and applications of artificial intelligence.
Copyright© 1994ElsevierScienceLtd Printed in Great Britain. All rights reserved 0952-1976/94 $6.(X)+ 0.00
Contributed Paper DEPUR: a Knowledge-based Tool for Wastewater Treatment Plants PAU SERRA Universitat Aut6noma de Barcelona MIQUEL SANCHEZ Universitat Polit~cnica de Catalunya JAVIER LAFUENTE Universitat Aut6noma de Barcelona ULISES CORTI~S Universitat Polit6cnica de Catalunya MANEL POCH Universitat Aut6noma de Barcelona
(Received December 1992; in revised form May 1993) A malfunction of a wastewater treatment plant is a major social and biological problem. Poorly treated waste water outside the plant could provoke dangerous consequences for human beings as well as the environment itself. The conventional control systems, that are usually applied in this field, have to cope with some difficulties: complexity of the system, an ill-structured domain, qualitative information, uncertainty or approximate knowledge, real-time dynamic system . . . . This paper shows an application of artificial intelligence in order to help the operators of wastewater treatment plants in their task of process control. The main goal is to build a knowledge-based tool useful for the diagnosis and management of wastewater treatment plants. First, a survey of wastewater treatment plants describes the complexity of the system being modelled and outlines its difficulties. The development of the application, and the methodology employed in it, are discussed. A new methodology called L1NNEO + is introduced. It is used for the automatic knowledge acquisition process in order to build up a Knowledge Base. The prototype architecture constructed - D E P U R - is detailed, together with some obtained results. Keywords: Knowledge-based systems, process diagnosis and m a n a g e m e n t , wastewater treatment, knowledge acquisition tools, environmental engineering, biotechnology.
1. I N T R O D U C T I O N The aim of this work relies on the application of new techniques in the fields of automation and process control. The A I tools, as seen below, are very important aids in the automation of process management. A malfunction of a wastewater treatment plant is a big social and biological problem. Poorly treated waste water outside the plant could provoke dangerous
consequences for human beings, as well as for the natural environment itself. The setting of an automatic process control over the wastewater treatment plant system, has outlined some difficulties:
• The complexity of the problem. There are many facts of different natures present at the system: chemical facts, mechanical, biological, microbiological ones . . . . • A non-well structured domain. The relationships among the concepts or attributes of the domain
should be sent to: Dr P. Serra, Universitat Autbnoma de Barcelona, Unitat d'Enginyeria Bioqufmica (CSICUAB), Edifici C. E-08193 Bellaterra, Barcelona, Catalonia, Spain.
Correspondence
23
24
PAU SERRA et al.: DEPUR are not well-enough known, or the agreement among experts is not very high. • Most information is neither numeric nor quantified. It is therefore not suitable to be included in the context of a classical-numerical-control model (microbiological information . . . . ). • Uncertainty or approximate knowledge. For example, the subjective information supplied by the experience of the plants' managers, is often vague or uncertain. • A dynamic system. Continuous changes that directly modify the performance of the process cause a further complication.
Moreover, it can be said that the classical methods work well on the normal states of the plant but not in other abnormal states of working. How could they detect some mechanical faults? How could they use the available but incomplete information, to solve a specific problem? These problems with the conventional process-control systems such as feed-back control, feedforward control, optimal control, adaptive control, . . . have given rise during the last few years to considerable research effort in Artificial Intelligence.' 4 Bearing in mind the main advantages and characteristics of knowledge-based systems, specially those of expert systems, and the above difficulties, one could expect that they would match very well. So a prototype of a qualitative expert system with uncertainty reasoning has been developed, as well as a real time numerical control expert system, which is described in Ref. 5, for the diagnosis and management of wastewater treatment plants. 2. A W A S T E W A T E R T R E A T M E N T PLANT
The main goal of a wastewater treatment plant is to reduce the level of pollution of the wastewater at the lowest cost, that is, to r e m o v e - - a s far as possible-strange compounds (pollutants) of the inflow water to the plant, prior to discharge to the environment. So, the effluent water must have low levels of the pollutants (lower than maximum ones allowed by the law). The plants taken as models are based on the main biological technology usually applied: the activated sludge process. ~ The concrete wastewater plant studied is located in Manresa, near Barcelona (Catalonia). This plant receives about 35,000 m3/day inflow from 75,000 inhabitants. A wastewater treatment plant is usually composed of two stand-alone but interactive subsystems: sludge line and water line. This study has focused on the water line. It is formed by a sequence of unitary processes where the effluent of one process becomes the inflow to the next one. Usually there are 3 major processes: primary treatment, secondary treatment and tertiary treatment. Each one reduces or removes the concentration of several specific pollutants. The flow chart of a plant may be seen in Fig. 1.
Sand
INFLOW
PRE-TREATMENTJ
t Oil
BIOLOGICAL
~
PRIMARY
PROCESS
~,~
BiologicalSludge
~
PrimarySludge (sedimentable suspended solids )
TERTIARY TREATMENT EFFLUENT
Other pollutants
Fig. 1. Chart of a wastewater treatment plant.
3. DEVELOPMENT OF THE QUALITATIVE EXPERT SYSTEM An expert system, called D E P U R , 78 has been built in order to support the approximate or uncertain reasoning, using qualitative knowledge, in the managing of wastewater treatment plants. The main problem appearing on a wastewater treatment plant is the breakdown of the balance between biomass and substrate, as shown in Fig. 2. The biomass is formed by a set of microbiological beings that in presence of oxygen, degrade non-sedimentable organic suspended solids (i.e. the substrate) in the water. More information about the wastewater treatment processes can be found in the classical literature. 9~,0 Thus, the task of diagnosis of the system is to determine the causes of the problems detected. Often, the most serious problems are related to the biological process in the secondary treatment. The design process of a knowledge-based system can commonly be expressed as shown in Fig. 3. The work described here has been focused on the water line subsystem domain. The relevant attribute selection process has been developed in a twofold fashion: firstly, in a typical manner based on the attribute selection made by the experts in the domain, and, secondly, by means
BIOMASS SUBSTRATE + 02
=,,
BIOMASS + CO2 + H20
Fig. 2. Biological balance.
PAU SERRA et al.: DEPUR
25
4. THE LINNEO + M E T H O D O L O G Y FOR THE DOMAIN SELECTION
K N O W L E D G E ACQUISITION PROCESS
DOMAIN FOCUSING
v* RELEVANT A'i-I'RIBUTE SELECTION
DATA BASE
DOMAIN ~ MODELLING ATTRIBUTE STRUCTURING
KNOWLEDGE BASE
x,,. IMPLEMENTATION
NO-~OKI
VALIDATION
lOK
SYSTEM USE
Fig. 3. Design process of a Knowledge-Based System.
of a new automatic attribute selection approach based on the relevance and potential utility concepts." The attribute structuring process has been made using two approaches:
• Decision trees: the classical approach derived from the interactions between the expert and the engineer. This approach is governed by the causeeffect criterion. Multi-valued decision trees have been used, and inference rules are obtained from them. The rules point to the causes of a certain problem or situation previously detected. • Automatic classification techniques: used as an automatic knowledge acquisition tool, this is an approach recently derived from machine learning in Artificial Intelligence. The L I N N E O ÷ system, based on fuzzy non-well-structured domain classifications, has been used in this phase. From the characterization of the classes obtained, moregeneral inference rules are constructed. They lead to knowledge of the most likely working situation of the plant. This methodology will be explained more clearly in the next section. Decision trees have been used in the cause identification phase, and automatic classification techniques in the problem identification process, as shown in Fig. 4.
The L I N N E O + software 12' ,3 is a knowledge acquisition tool which is useful in ill-structured domains, and which uses both inductive learning methods and deductive ones. L I N N E O + uses (possibly noisy) observations to induce classes, as well as taking advantage of the expert's current state of knowledge (which may be incomplete), expressed as logical constraints on the possible output of the classification process. Figure 5 shows the general framework of the methodology. The human expert abstracts a collection of observations (sample) from the problem domain that he thinks of as relevant. He also selects a set of attributes (descriptors) that he assumes to be significant for the characterization of the domain objects. A sample is then modelled by a set of attribute/value vectors. At the same time, the expert can express what he already knows about the domain. This previous knowledge about the domain (that will be called Domain Constraints--DC) is represented as constraints and examples on classes already known to exist. Starting with this knowledge and information, L I N N E O + produces a classification. It uses induction over the observations, taking into account the DC which guide the inductive process. The results of this classification are an intensive and extensive description of the generated classes and a fuzzy membership matrix that relates observations to generated classes. The expert evaluates the results and possibly modifies some attributes or some DC. When a "correct" classification is obtained, rules are derived from the generated classes. Each rule includes in its antecedent the membership conditions and a fuzzy certainty factor for a particular class. Each antecedent is a conjunction of logical conditions on the
ATTRIBUTE III SYSTEM STATE CHARACTERISTICS
SIT1 SIT2 SIT3 ... SITN ~
O1.1 01.2 ... O2.1 ... CN.N
SITUATIONS
---- CAUSES
l lll SOLUTIONS ----
$1.1 $1.2
$2.1
SN,N
Fig. 4. Diagnostic process in an Expert System.
PAU SERRA et al.: DEPUR
26
Bulking-not-enough-oxygen Bulking-sulphurs Rising Foaming Overaeration - - Summer Winter - - Storm High plant Rotation-band-crash Primary-sludge-sedim.-problem Not-enough-primary-sludge-exit Overloading Turbine-crash By-pass-bad-closed - -
- -
STEPSOFTHE METHODOLOGY
- -
.l
bs/~action
- -
ATrRIBUTE SELECTION AND OBSERVATIONS Observations: Attdbute/Value matrix
- -
- -
Ngw attributes/ observations
- -
i n f l o w
- -
- -
FUZZY CLASSIFICATION
- -
A'II'R.IBUTE RELEVANCE MODULE
~e
- -
- -
/
Classes
att butcs/ ob ervations
New
- -
- -
- -
GER~FAsTION "'J
KNOWLEDGEI BASE
Fig. 5. The methodology of LINNEO ÷.
v a l u e s t h a t d e s c r i p t o r s m u s t show for m e m b e r s h i p of the class w h o s e identifier a p p e a r s on the r u l e ' s conseq u e n t . T h e s e rules are a n a l y s e d a n d , t h r o u g h s u b s u m p tion d e t e c t i o n a n d s y n o n y m y analysis, t h e y are organized in a h i e r a r c h i c a l way. T h e results are m a d e k n o w n to t h e user, w h o can a c c e p t a n d confirm t h e m . H e also has the c h a n c e to go b a c k to the classification m o d u l e , to t h e definition o f the D C o r to the s a m p l e selection s t e p , a n d o b t a i n o t h e r results. W h e n all these a c t i o n s a r e o v e r , rules can be v a l i d a t e d using new o b s e r v a t i o n s n o t p r e v i o u s l y in the s a m p l e . A s s o o n as the e x p e r t a c c e p t s the resulting rule set, it is transf o r m e d into I F - T H E N rules a n d m o d u l e s .
5.
ARCHITECTURE
OF
THE
APPLICATION
T h e m o d e l l i n g o f the d o m a i n has b e e n m a d e as follows: objects o r observations are p r o t o t y p i c a l w o r k ing s i t u a t i o n s o r p r o b l e m s in the w a s t e w a t e r p l a n t of Manresa, near Barcelona (Catalonia). The experts d e f i n e d a b o u t 19 d i f f e r e n t situations at the b e g i n n i n g of the w o r k . T h e attributes o r properties a r e the m o s t r e l e v a n t p a r a m e t e r s a n d f a c t s - - i n the v i e w p o i n t of the e x p e r t s - - t h a t c h a r a c t e r i z e the d o m a i n . T h e y have s e l e c t e d a b o u t 20 a t t r i b u t e s . So, the input m a t r i x to the classification p r o c e s s o f L I N N E O + is as follows: •
Objects-situations-defined: - -
- -
Normal Bulking-non-filamentous
T
o
x
i
c
-
s
u
b
s
t
a
n
c
e
s
-
l
o
a
d
i
n
g
Toxic-substances-increase
• A t t r i b u t e s defined: -- Rotate-band-clarifiers - - Bubbles-clarifiers -- Bubbles-primary-settlers -- Inflow-water -- Dissolved-Oxygen -- Inflow-DQO -- Outflow-DQO -- Sludge-age -- Foam-bassins -- Filamentous --SV1 -- Odor-bassins -- Sulphurs -- Sludge-exit-primary-settler - - S u s p e n d e d Solids-bassins - - S u s p e n d e d Solids-inflow - - S u s p e n d e d Solids-outflow - - V o l a t i l e S u s p e n d e d Solids-bassins - - V o l a t i l e S u s p e n d e d Solids-recirculat. -- Temperature -- Turbines -- Toxic-substances-bassins -- Toxic-substances-inflow A n e x a m p l e of a c o m p l e t e o b s e r v a t i o n with its values for the a t t r i b u t e s m a y be like this o n e : •
Bulking-non-filamentous: Outflow-DQO --~ H i g h Sludge age --->O l d Filamentous ~ Normal SVI ~ High Volatile SS-recirculation ~ Low All-other-attributes ~ Nought-value
T h e o t h e r a t t r i b u t e s have nought values '4 o r don't care values. This m e a n s that the values o f these attrib u t e s are not r e l e v a n t for the c h a r a c t e r i z a t i o n of the o b j e c t , a l t h o u g h t h e y c o u l d be r e l e v a n t to d e s c r i b e o t h e r w o r k i n g situations of the plant. F o r instance, in the a b o v e e x a m p l e , the value of the a t t r i b u t e tempera-
PAU SERRA et al.: DEPUR
27
ture could be any o n e I b u t don't care which it is--for the description of the bulking situation originated by non-filamentous causes.
tl
5.1. Results
Ten classes have been obtained (see Fig. 6) as a result of the classification process. Each one groups closely related prototypical situations. The classification shows the most common situations and problems in the working state of the wastewater treatment plant, and simplifies the arrangement of the underlying domain. These classes lead the inference process in the diagnosis.
KNOWLEDGE BASE
INFERENCEENGINES
6. I M P L E M E N T A T I O N OF DEPUR IN MILORD
MILORD ~5'~ is a classical second-generation expertsystem shell, but consideration has been given to some important features related to this particular domain, such as the microbiological and temporal ones. The application developed is called DEPUR, and reflects the experts' knowledge (see Fig. 7). The Data Base is a representation of the current state of the system, that could include some past or future facts of interest. In this representation the attributes most relevant to the domain characterization are selected. The dictionary of the Data Base is composed of 167 attributes, with a decrease by means of relevant and potential utility concepts. An example of dictionary entries may be:
CLASS- 1 CLASS-3 TANC
N-FI
[ BULKING-NOT-ENOUGHOXYGEN~ BULKING-SULFURES ] \ FOAMING I ~-BAD-Cy
CLASS-4
CLASS-I 0 CLASS-6
CLASS-9
T~URBINE-CR~H~
CLASS-@ CLASS-5
@ CLASS-8 ~E-SEDIME~
Fig. 6. Resultsof the classiticationprocess.
CONTROLMODULE
Fig. 7. General architectureof DEPUR. (INFLOW-WATER-NORMAL "Has the inflow water normal (35000-45000) m3/day?'' FUZZY)
values
(LOW-TEMPERATURE "Is the temperature low ( < 5)°C? '' FUZZY) The Knowledge Base is the arrangement of the knowledge available over the domain, made explicit by the expert and by information coming from the analysis of the system being modelled. Usually, this knowledge is expressed by means of inference rules. Examples of rules may be: (RMENU003B (IF (DQO-EFLUENT-WATER-HIGH SLUDGE-AGE-YOUNG)) SURE (THEN (INFER (CLASS-3B)))) (RC3004 (IF (CLASS-3A DISSOLVED-OXYGEN-LOW ODOR-BASSINS)) SURE (THEN (INFER (BULKING-O2))))
PAU SERRA et al.: DEPUR
28
The system is formed of 79 rules, grouped in 12 modules. A module is a structural element of the Knowledge Base, composed of a set of related rules. The Control Module has the goal to control how and when the knowledge supported by the system may be used. Often it is expressed by means of metarules. The prototype has 15 metarules. An example of a metarule may be:
KNOWLEDGE-BASEDSUPERVISORSYSTEM
(M003B (IF (CLASS-3B)) SURE (THEN (TREAT-MODULES C3)))
l
The linguistic terms used to manage the approximate knowledge are, in order of certainty: Sure Almost-sure Very-possible Possible Moderately-possible Slightly-possible Almost-impossible Impossible
7. VALIDATION The prototype diagnostics, in contrast to the answers of human experts, have yielded very good results. They have been tested with several prototypical situations in the working of the plant (bulking, foaming . . . . ). Currently, this classification is validated against a data stream obtained daily from direct observation in the wastewater treatment plant over the last two years, with very high degree of performance. A short dialogue with the application could be like the following one: > Has the effluent DBO high values (25-40) mg/l? -Sure > Has the water inflow high values (45,000-55,000) m~/day ? -Almost impossible > Is the temperature low (<5)°C? -Impossible > Has the inflow water high concentration of Zn (4-20) mg/l? -Impossible > Is the Volumetric index of sludge (SV1) high (150-300) rag~l? -Very possible > Are the Suspended Solids (SS) in the recirculation low (1300-4600) mg/l -Sure
EVOLUTIVE DATA BASE
NUMERICAL CONTROL ALGORITHMS
\
/
~IAL
The modularity of the application, the selfexplanation and the user-friendly interface afforded by the shell are other outstanding properties.
II
TI
KNOWLEDGE BASE
PR~E~
Fig. 8. A SupervisoryKnowledge-BasedSystem. > Are the Volatile Suspended Solids (SSV) in the recirculation low (100-3500) mg/l? Sure > Has the dissolved oxygen (DO) high values (2.5-10) rag~l? Impossible > Is the age of sludge old (8-20) days? Impossible > Conclusion Bulking-not-enough-oxygen--Very possible 8.
C O N C L U S I O N S
A N D
F U T U R E
W O R K
The building of the DEPUR system as a prototype fuzzy expert system applied to managing a wastewater treatment plant, has been the consequence of the research effort during recent years (study and modelling of wastewater treatment plants, development of new techniques of selection and arrangement of attributes, building of computer systems . . . . ) by both scientific groups involved. A diagnostic and management tool based on knowledge-based systems for a wastewater treatment plant may be a useful approach in the field of intelligent systems applied to managing industrial processes. The prototype may be helpful to the manager of the plant. The treatment of uncertainty and the new fuzzy techniques for automatic classification are some of its characteristics. This methodology
PAU SERRA et al.: DEPUR has b e e n successfully a p p l i e d to o t h e r d o m a i n s (see R e f s 11, 12). This p r o j e c t p r o v i d e s o n e s t e p t o w a r d s the b u i l d i n g of i n t e l l i g e n t s y s t e m s o r i e n t e d to c o n s e r v e w a t e r , as a p a r t of n a t u r e (intelligent e c o l o g i c a l systems). A n o t h e r i n t e r e s t i n g p o i n t is the possibility to a d a p t the system to the specific c h a r a c t e r i s t i c s of the actual w a s t e w a t e r treatment plant under control. In t h e f u t u r e , it is p r o p o s e d to r e a c h a s e c o n d step: the d e v e l o p m e n t o f a S u p e r v i s o r K n o w l e d g e - B a s e d S y s t e m - - m o r e p o w e r f u l t h a n an e x p e r t s y s t e m - - t h a t c o m b i n e d i a g n o s i s a n d c o n t r o l , including the n u m e r i c a l c o n t r o l a l g o r i t h m s ~7with q u a l i t a t i v e a n d heuristic inform a t i o n a n d r e a s o n i n g (see Fig. 8). A l s o , t h e r e is m u c h w o r k to do: to e x t e n d t h e d o m a i n to the sludge line, to r e s e a r c h the m i c r o b i o l o g i c a l c o m p o u n d o f the biological p r o c e s s in a m o r e - a c c u r a t e w a y , a n d to a c h i e v e b e t t e r t e m p o r a l i n f o r m a t i o n t r e a t ment.
Acknowledgements--This work has been supported by the "Junta de Sanejament de la Generalitat de Catalunya" and the Spanish CICyT projects ROB89-0479-C03-02, ROB91-1139 and TIC91-1087-C03-01. Also, the authors wish to acknowledge the cooperation of Ricard Tom~s, manager of the Manresa's wastewater treatment plant.
REFERENCES 1. Coleman J. J., Krovvidy S., Scott Summers R. and Wee W. G. An AI approach for wastewater treatment systems. J. Appl. lntell. I, 247-261 (1991). 2. Floris V. Artificial intelligence in the operation and management of water resources in South Florida. Personal Communication (1989). 3. Maeda K. A knowledge-based system for the wastewater treatment plant. Future Generation Comput. Syst. 5, 29-32 (1989). 4. Sanz R., Aguilar J., Sierra C., God6 LI. and Ollero A. Adaptive control with a supervisor level using a rule-based inference system with approximate reasoning. Artif. lntell. Sci. Comput. towards 2nd generation systems. IMACS (1989).
29
5. Serra P., Sfinchez M., Lafuente J., Cort6s U. and Poch M. ISCWAP: a knowledge based system for supervising activated sludge processes. Submitted for publication. 6. Robust6 J. Modelitzaci6 i identificaci6 del proc6s de fangs activats. Ph.D. Thesis, Departament de Quimica, Universitat AutSnoma de Barcelona (1990). 7. S~mchezM., Serra P., Belanche LI. and Cort6s U. A knowledgebased system for the diagnosis of waste-water treatment plants. Proc. of 5th International Conference on Industrial & Engineering Applications of Artificial Intelligence and Expert Systems (lEA~ ALE-92). Lecture Notes in Artificial Intelligence 604, pp. 324-336. Springer Paderborn (1992). 8. Sfinchez M. DEPUR: aplicaci6 dels sistemes basats en el coneixement al diagnSstic en plantes de tractament d'aigiies residuals. Master Thesis, Department de Llenguatges i Sistemes Informhtics, Universitat Polit~cnica de Catalunya (1991). 9. Benefield L. D. and Randall C. W. Biological Process Design for Wastewater Treatment. Prentice-Hall, Englewood Cliffs, NJ (1980). 10. Metcalf and Eddy Inc. Wastewater Engineering: Treatment~Disposal~Reuse, 3rd edn. McGraw-Hill, New York (1991). 11. Belanche LI. and Cort6s U. The nought attributes in knowledgebased systems. Proc. of the European workshop on Verification & Validation, EUROVAV-91, pp. 77-103. Jesus College, Cambridge (1991). 12. B6jar J. and Cort6s U. LINNEO÷: Herramienta para la adquisici6n de conocimiento y generaci6n de reglas de clasificaci6n en dominios poco estructurados. Actas 3er Congreso Iberoamericano de lnteligencia Artificial, IBERAMIA-92, pp. 471-481. La Habana, Cuba (1992). 13. Martin M, Sangiiesa R. and Cort6s U. Biasing induction with previous knowledge for knowledge acquisition in imprecise domains. Proc. Avignon '91, pp. 359-370 (1991). 14. S~inchezE. Importance in knowledge systems. Information Syst. 14, 455-464 (1989). 15. God6 L., L6pez dc Mtintaras R., Sierra C. and Verdaguer A. MILORD: the architecture and management of linguistically expressed uncertainty. Int. J. Intell. Syst. 4,471-501 (1989). 16. Sierra C. MILORD: arquitectura multi-nivell per a sistemes experts en classificaci6. Ph.D. Thesis. Department de Llenguatges i Sistemes lnformhtics, Universitat Polit~cnica de Catalunya (1989). 17. Moreno R., de Prada C., Lafuente J., Poch M. and Montague G. Non-linear predictive control of dissolved oxygen in the activated sludge process. In Modelling and Control of Biotechnological Processes, pp. 289-293 (Karim M. N. and Stephanopoulos G., Eds). Pergamon Press, Oxford (1992).
AUTHOR'S BIOGRAPHIES Pau Serra i Prat graduated in Chemical Engineering from the Universtitat AutSnoma de Barcelona in 1988. Since 1988 he has researched in the area of control of environmental processes. In 1993 he presented his Ph.D. Thesis on the development and application of a knowledge-based system for the management of a wastewater treatment plant. Dr Serra is currently working in an enterprise related to the exploitation of potable and waste water treament plants in Spain. Miquel Sknehez i Marr/~ (Barcelona, Catalonia, 1964) graduated in Computer Science at the Technical University of Catalonia (UPC) in 1989. In 1991 he presented his Master's thesis, and is currently working towards his Ph.D. He has been on the staff of the Department of Computer Systems and Languages since 1988. Since 1989 he has participated in the Interest Group on EuLISP, funded by the ECC and has also been working on projects supported by the Spanish Council and the ECC. His main research topics are distributed and concurrent AI, knowledge engineering, automatic knowledge acquisition, real-time systems and LISP-like languages. Javier Lafuente Sancho, Associate Professor in Chemical Engineering at the University AutSnoma de Barcelona, received his Ph.D. in the area of water quality modelling. This work is now being used by the Catalan Government. He undertook NATO-funded postdoctoral research work at MIT, and subsequently has been involved in the EC BAP project on the monitoring and control of bioreactors. He is a member of the "European Laboratory Without Walls". His work in Artificial Intelligence has recently been enhanced by a British Council scheme in collaboration with the University of Newcastle-upon-Tyne, U.K. He is currently involved in a Spanish project in the area of "Control of Municipal Waste Water Treatment Plants". Ulises C o r t ~ (Mexico D.F., M~xico 1960) received a B.Sc. in Industrial and Systems Engineering at the Instituto Tecnol6gico y de Estudios Superiores de Monterrey in 1982, and a Ph.D. from the Technical University of Catalonia (UPC) in 1984. He is an Associate Professor in the Software Department of the UPC, Head of its Artificial Intelligence Section, and an Adjunct
3(I
PAU SERRA
et al.:
DEPUR
Researcher at the CSIs Center for Advanced Studies of Blanes (Spain). He participates in the ECC-funded Interest Group on EuLISP, and is the Spanish representative on the JTC1/SC22/WG16 LISP ISO Group. His main research topics are machine learning, knowledge acquisition and LISP-like languages. He has an extensive list of publications, conference papers, and tutorial and supervisory contributions. Manel Pooh Espallargas, Associate Professor in Chemical Engineering at the Universitat Autbnoma de Barcelona, is the author of more than 50 papers, 30 congress communications and advisor to eight Ph.D. Theses. He has been involved in monitoring and control of biotechnological and environmental processes during the last 10 years, being responsible for several Spanish research projects in this topic, and a researcher in an European project about advanced monitoring and computer control of biotechnological processes. He has been responsible for several contracts with companies in monitoring and control projects involved in biotechnological and environmental topics. His main research areas are related to the optimal control and management of biotechnological processes including monitoring, control and applications of artificial intelligence.