Escolar Documentos
Profissional Documentos
Cultura Documentos
tion methodology and we have implemented it by ones. Process Unit has the choices of the enologist
building a knowledge-based simulator. This generator concerning the process control as structural variables;
has been structured around a hierarchical model as its state variables describe the evolution of phenomena
defined by Oussalah (1988) and Mozetic (1990). We such as thermal transfers or fermentations; its know-
have given a special attention to winemaking. We ledge functions evaluate the feasibility for accepting an
present here the validation of the simulator. The input flux or give an output flux according to the values
interest of this simulator under development is illus- of its state variables and they evaluate the opportunity
trated by some examples. of a change of activity of the Winemaking Subsystem
described in this first level of abstraction.
Second level of abstraction. A set of knowledge that
RESULTS we have called Winemaking System, has been defined
with two classes of objects: Operation, Material. We
Design of the expert simulator jump to this second level of abstraction by the
aggregation of several objects Process Unit which have
The model the same type of action on Material and can conse-
We started from a concept of the winery where wine- quently be considered as belonging to the same
making follows process lines, each one being a succes- Operation. Operation has global input and output
sion of operations, e.g. reception of grapes, pressing, fluxes of Material. Its knowledge functions evaluate the
thermal transfer, settling, centrifugation, maceration, feasibility of accepting an input or an output flux, and
fermentation, storage. Each operation gathers a set of they also evaluate the opportunity of a change of
units making the same kind of transformation on the activity of the Winemaking System described at this
material. We have imagined a hierarchy between units, second level of activity.
operations, process lines and the winery, with four Third level of abstraction. At this level we have
levels of abstraction, respectively (Figure 1). defined the Process Line linked to Material. We jump
First level of abstraction. A set of knowledge that we to this third level of abstraction by describing a
have called a Winemaking Subsystem has been defined succession of Process Operations to achieve. Process
by means of five objects, respectively Process Unit, Line has global input and output fluxes of Material.
Material (treated by the Process Unit), Phenomenon Fourth level of abstraction. This is the level of
(taking place in the Process Unit), Main Equipment Winery, linked to Vintage. We jump to this fourth level
(container of the Process Unit such as a tank, an of abstraction by aggregation of all Process Lines.
exchanger, . . .) and Accessory Equipment (such as a Vintage describes all grapes brought to the winery and
pump for refilling a tank). Each object is characterized the object Winery contains knowledge functions simu-
by its fixed data (structural variables), its parameters lating the decision of the winemaker to dispatch the
(state variables) and the functions manipulating the vintage among process lines.
information. In this way, Material has as structural
variables: inital sugar content, grape variety, ferment-
ability index; and as state variables sugar and alcohol Dynamic simulation
content, density, temperature, fermentation rate, The Winemaking Subsystem described at the first
phase (liquid, solid). We have not yet created calcula- level of the hierarchical model is a discrete events
tion functions in this class of objects, but it could system: its behaviour, imposing input and output
be a correlation between variety, initial sugar content fluxes, depends on the activity in which it is implied. A
and fermentability for some specific conditions. Winemaking Subsystem may have the activities ‘On’
Main Equipment and Accessory Equipment have and ‘Off’ if it is continuous or ‘Standby’, ‘Refilling’,
their physical characteristics as structural variables and Storage’, ‘Emptying’ if it is batch. The logics of
their available capacities as state variables. Pheno- evolution of the state variables of the objects of the
menon has only knowledge functions which read the Winemaking Subsystem rely on its activity.
state variable values of Process Unit and calculate new Three sets of functions regulate the activity of the
Winemaking Subsystems and Systems. We have called
them Determine TabRec, Determine Mode and Trans-
Main equipment form Action. Each one of these functions is defined
Phenomenon
twice, at the first and at the second level of the
Accessory equipment 2 hierarchical model, in a different way: this is called
/
.
*-__ r polymorphism in AI. A first Determine TabRec func-
-\ I
- - - Press Unit : tion sets up a table describing the state in which Process
‘\ Unit would like to be, and a higher level Determine
\
\ TabRec function synthesizes at the level of Operation
ll ‘4
Process Operation - - - - :n Material the changes required by the Process Units composing
/ this Operation. The Determine Mode function decides
,
,
I at the level of Operation whether it is opportune or
l-i I
Process Line ’ not to modify its mode according to the calculations
of Determine TabRec and to flux constraints with the
previous Operation of the Process Line. The lower
3 Vintage level Determine Mode function accordingly affects a
mode to each Process Unit. The Transform Action
Figure 1 Presentation of the hierarchical model of the winery. Each
name represents a class of objects; =+, link is composed of; +. function modifies the state variables of Process Units
link is associated to according to the new modes of activity. Finally a higher
Validation
tanks and refrigeration power. The maximum number refrigeration power makes a good tool for optimization
of used tanks simulated passed from 26 to 17 and the of these resources. If we want the simulator to do more
maximum refrigeration needs passed from 70000 to than simulating the expert, flexible constraint satisfac-
45 000 kcal/h, which is correlated with the ratio of input tion problem is a promising way. This new computer
fluxes. For more subtle variations in grape inputs, it is tool mixes recent concepts of computer science with
clear that only a simulation can predict the dynamics valorization of expertise and this will be important in
of the needs. the future of wine engineering.
Besides the use of the dynamic simulation of
Influence of the process technology on the availability of resources in the winery, AI will have industrial
resources applications at the level of vintage reception and
We have assumed a maceration duration of 48 h dispatching for optimizing the links between selection
instead of 72 h and we noticed that the number of tanks in the vineyard and processing, and also for supervision
required did not change significantly whereas the of pressing or for piloting alcoholic fermentations
refrigeration needs drastically increased (Figure 4). In (Grenier et al., 1988).
this case we could predict this result because the
operation for maceration was without temperature
control whereas the operation of fermentation in liquid
ACKNOWLEDGEMENTS
phase was controlled.
The authors are grateful to Mr Orange, Director of
DISCUSSION the Winery of Saint-Genies-les-Mourgues (France,
Henault) for his welcome during the 1991 grape picking
and Dr Lopez, Professor at the University of Lerida
We plan to extend the simulation to several parallel
(Spain), for his help in the model validation.
process lines. We have taken another example where
the Champagne method is applied to winemaking with
two process lines: maceration prior to pressing, and
no maceration. The knowledge source of the dynamic REFERENCES
model has had new Operations added such as settling
clarification, centrifugation and various continuous Aarts, R.J., Suviranta, A., Rauman-Aalto, P. and Linko, P. (1990)
thermal processes. The knowledge functions describing An expert system in enzyme production control. Food Biotechnol
the related phenomena taking place in the correspond- 4 (1). pp. 301-315
ing process units are being implemented. Daubas B., Pingaud, H. and Koehret, B. (1991). Developpement
This will require improvements in the daily task d’un simulateur de procedes discontinus. semi-continus et
continus: ProSim. Recenfs frog. Genie Proc. 5, 307-311
planning. At present it works on a ‘first come first
Dubois, D. and Prade. H. (1988) La Theorie des Possibilites. Edition
served’ basis. To fit reality better, the planning set Masson
up every morning by the winemaker has to be taken
Fargier, H. (1992) Probltmes de satisfaction de contraintes floues.
into account. We have started to develop a computer Rapport IRIT/92-29R. UPS, Toulouse, France
solution that makes 24-h cycles of calculations. During
Grenier, P. and Niviere, V. (1990) Systeme expert de gestion
the first simulation of a given day, we set up the list of d’informations techniques pour I’utilisation optimale des
tasks to achieve during the day. We then classify these ressources en froid d’une cave de vinification au tours de
tasks as the winemaker would. Then we make a second I’elaboration des vins blancs. Dixiemes Journees Internationales
sur les Sysdmes Experts et leurs Applications, Avignon 28 May-l
definitive simulation and we jump to the next day. As
June Proceedings EC2. EC2, Paris, France, pp. 181-189
the knowledge of the winemaker is imprecise, a fuzzy
Grenier, P., Feuilloley, P. and Sablayrolles, J.-M. (1988) Develop-
representation results. We may want to provide the ment of an expert system for the optimization of the wine
winemaker’s simulator with a module of constraints alcoholic fermentation. International Conference on Agricultural
analysis for the software to more than just simulate Engineering, 2-5 March, Paris, paper 88.399. Eur. Sot. Ag. Eng.
what the enologist does. Taking imprecise or uncertain Grenier, P., Sablayrolles, J.M., Chabas, J. and Barre, P. (1992)
information into account by the fuzzy formalism of the Refrigeration des mouts en fermentation: progres r¢s. Rev.
possibilities theory, the optimization of daily task Gen. Froid 5, 31-35
classification would have to be treated within the Hitzmann, B., Liibbert, A. and Shtigerl, K. (1991) An expert system
approach for the control of a bioprocess. I: Knowledge represen-
domain of problems of flexible constraint satisfaction
tation and processing. Biotechnol. Bioeng. 39, 33-43
(Dubois and Prade, 1988; Grenier et al. 1988; Aarts
Konstantinov, K.B. and Yoshida, T. (1992) Mini review. Knowledge-
et al., 1990; Oishi et al., 1991; Fargier, 1992; Niviere
based control of fermentation processes. Eiorechnol. Bioeng. 39,
et al., 1993). 479-486
Finally, further improvements would mean integrat- Le Roux, J.M.W., Purchas, K. and Nell, B. (1986) Refrigeration
ing other aspects of the plant such as the pollution requirements for precooling and fermentation control in wine-
charge. For that, it is enough to include new state making. S. Afr. J. Enol. Viticulture 7 (l), 6-13
variables in the Process Unit and in the Operation, Lopez, A. and Secanell, P.A. (1992) A simple mathematical empirical
describing the consumed quantity of water as well as model for estimating the rate of heat generation during fermenta-
tion in white-wine making. Inr. J. Refrig. 15 (4). l-5
calculations and expert estimations of the organic
matter content of the rejected waters. Lucas, P., Isambert, A., Depeyre, D., Cuille, P. and Rossiny, P.
(1988) Dynamic simulation and control strategy of a continuous
separation system with unsteady inputs. Chemdata Congress, 13-
15 June, Goteborg, Sweden. Eur. Fed. Chem. Eng.
CONCLUSIONS
Maron, J-M. and Blouin, J. (1989) Logiciel BTC de Bilun Thermique
de Chai. Federation Departementale des Centres d’Etude et de
The dynamic simulation of needs in equipment and Formation Oenologique, Bordeaux. France
Mozetic, I. (1YYO)Hierarchical model-based diagnosis. Inf. J. Man- Pokkinen, M., Flares Bustamante, Z.R., Endo, I., Aarts, R. and
Machine Studies 35, 329-362 Linko, P. (1992) A knowledge based system for diagnosing
Niviere, V., Grenier, P. and SCvila, F. (1993) Simulation d’usine et microbial activities during a fermentation process. Bioprocess
ensembles flous dans I’industrie vinicole. Troisitimes Journkes Eng. 7, 33 l-334
Nationales sur les Applications des Ensembles Flous. 26-27 Sablayrolles, J-M. and Barre, P. (1989) Pilotage automatique de la
October, Nimes. EC2, Paris, France temperature de fermentation en conditions oenologiques. Sci.
Oishi, K., Tominaga, M., Kawato, A., Abe, Y., Imayasu, S. and Aliments 9, 239-25 1
Nanba, A. (1991) Application of fuzzy control theory to the sake
brewing process. J. Ferment. Bioeng. 72 (2), 115-121
Oussalah, M. (1988) Modbles hierarchises multi-vues pour le support Received 3 November lYY2
de raisonnement dans les domaines techniques. Thesis, Univer- Revised 30 June 1993
sity of Aix. Marseille Accepted S October 1993