An expert system for bottling plant design M. Novak & A. Jezernik Faculty of Technical Sciences, Mechanical Engineering Department, Maribor, Slovenia

Size: px

Start display at page:

Download "An expert system for bottling plant design M. Novak & A. Jezernik Faculty of Technical Sciences, Mechanical Engineering Department, Maribor, Slovenia"

Phyllis Logan
5 years ago
Views:

1 An expert system for bottling plant design M. Novak & A. Jezernik Faculty of Technical Sciences, Mechanical Engineering Department, Maribor, Slovenia Abstract A prototype of an expert system (ES) for designing various types of bottling plants is presented in this paper. ES is implemented as a production system. A bottling plant design knowledge has been represented by production rules using programming language Prolog. Interviewing has been applied as a knowledge acquisition method to extract a design knowledge from a human expert. Program, which controls the bottling plant design process and the communication between the user and the ES, was developed using Prolog. Prolog's built-in goal driven reasoning was used as an inference engine. The system was tested on several examples for bottling e.g.: mineral water, oil, juice. A comparison with the conventional design procedures was made. An ES evaluation confirms that the artificial intelligence techniques application is an effective approach to solve a bottling plant design problem. 1 Introduction Design is a basic activity in product development process that requests special domain knowledge, experiences, capabilities like a creativity and decision making ability. Designer has to make a decision about description of the structure considering the design requirements. Nowadays many computer aided design systems provide the assistance to designer. Yet the applications, which use the methods of artificial intelligence, like an expert systems (ES), enable user to use computer as a direct aid to design decision making.

198 Artificial Intelligence in Engineering The factory of bottling plant and equipment, Radenska & Steinle (R&S), manufactures several types of machines, which are components of the bottling system.

2 198 Artificial Intelligence in Engineering The factory of bottling plant and equipment, Radenska & Steinle (R&S), manufactures several types of machines, which are components of the bottling system. Each product is made for a known customer. Therefore the manufacture requires a very careful planning, preparation and a complete and thorough knowledge of the bottling technology. Engineers, specialized in bottling plant technology, examine and diagnose customer problems of bottling, washing, packaging and transportation of liquids and than recommend a project design and an (nearly) optimal technological solution of the bottling and packaging line. Development of an offer of a bottling plant requests the skill of an expert, who is familiar with the manufacturing process in the factory, with the machines on the market and with the characteristics of the machines. The designer has to select the appropriate type of machines and the layout of the production line, considering the requirements and the relations between the machines. He can help himself with some catalogues. His goal is to make an (nearly) optimal bottling plant. The bottling plant includes the following machines: depalletizing machine, bottle removal machine, plastic crate washing machine, bottle washing machine, bottling and capping machine with crown caps and screwed caps, labelling machine, bottle crating machine and palletizing machine. The machines are linked with the bottle conveyors and the plastic crates conveyors. The basic requirements for bottling line design are the capacity, the type of the bottling substance and the size of the bottles. The capacity determines the size of the machines. The types of the machines are fixed with the bottling liquid and the performances of the machines are determined with the bottle size. The bottling machine is the first selected unit of the bottling line. It is very important that the capacities of the machines on the left and right side of the bottling machine increase in steps by 10 per cent. 2 Bottling Plant Design Problem The bottling plant design requires a considerable amount of information (for example: basic specification of machines, data about the bottling substance, etc.), which can be find in many catalogues, tables or diagrams. Very important role in design process presents the experience, (for example: the capacity of the bottling machine reduces for 30 per cent, if the bottling liquid is beer, because of frothing) and the knowledge of the specialists, who know the bottling technology very well. The design process is time consuming process and it can be repeated until all design requirements are not satisfied. All this reasons lead to the conclusion that it makes sense to build an ES which will help the user to design an appropriate bottling plant in time. The bottling plant design is a nonnumerical problem with structured object and with the relationship between the objects.

Artificial Intelligence in Engineering 199 Most of the knowledge for solving the bottling plant design problem can be represented in the form of the rules that express relationships among objects.

3 Artificial Intelligence in Engineering 199 Most of the knowledge for solving the bottling plant design problem can be represented in the form of the rules that express relationships among objects. Prolog is a language of rules and it is an excellent tool for prototyping of ES [1], so we used it as a programming language and developed a prototype of ES. The problem solving process is very complex, thus we divided it into three tasks: the machine selection and the bottling plant offer elaboration, the bottling plant layout elaboration without conveyors and the connection of the machines with conveyors and the elaboration of the complete layout. 3 An ES for Bottling Plant Design The power of the ES is in its knowledge, therefore the most important part of every ES is the knowledge base. Knowledge acquisition - the process of extracting knowledge from the source, structuring and organizing it in a way that can be stored in the knowledge base and explored by a problem solving mechanism - can be performed in different ways: by interviewing experts, by using the problem-dependent systems for knowledge formalization or by using the machine learning. Since we did not have data about many representative examples, so the most correspondent way to create the knowledge base has been to extract the design knowledge from the expert through interviewing. We have visited the experts in R&S several times. We have studied the concepts and basic function of machine and bottling plants. Afterwards the relations between the machines, selection processes, design process and various criterions have been extracted. The bottling plant design knowledge has been represented by production rules using programming language Prolog. After the knowledge is represented, we need a reasoning procedure to draw conclusions from the knowledge base. For production rules, there are two basic ways of reasoning: backward chaining and forward chaining. As we want to select the appropriate type of machines, we can start from the hypothesis and continue to the pieces of evidence, like an expert in his classical work and we can use the first way - the goal driven reasoning. Since we used Prolog, which has its own built-in backward chaining reasoning procedure, built-in unification mechanism as well as backtracking that is done automatically and is hidden from the user, we have not had difficulties by implementing the inference engine. In the first step our work was concentrated on the correct writing down the information and the knowledge into the knowledge base. After the knowledge base definition we have developed the program, which performs the bottling plant design process and the communication between the user and the ES. The program is written in Prolog and can be considered as a meta-interpreter.

4 200 Artificial Intelligence in Engineering The weakness of that access is the syntax of the rules. Namely, we hardly distinguish the knowledge base rules from the program rules. On the other hand a development of the ES, which is implemented as a production system, has also advantages, for example: easy tracing of rules performance, easy modifying and extending the knowledge base, etc. [4]. Our system is composed out of three parts: the data and knowledge base, the user interface and the inference engine. The configuration of the system can be seen on Figure 1. It is an example of the synthetic type of the ES [2]. The data and knowledge base contains information and knowledge of machines, bottling substances, bottling process, etc. The user interface caters for smooth communication between the user and the system. It depends on the user interface how the ES will be accepted by the users. DATA AND KNOWLEDGE BASE FACT: washmach(^rcx',1,4,10,9000,30000). RULE: corfact(carbonat,_,t,c,0.83) :- T>4, T=<20, C=<8,! corfact(beer,k,t,_,0.7) :- T=<16, KX20000,!. corfact(beer,_,t,c,0.8) :- T=<2, C<9,!. corfact(champagne,_,t,c,0.4357) : - T=<1, C=<12,!7 PROCEDURE: lis_ sub( [], [],0). lisjsub([x Rx],[X Rxl],Ist):- writelnl [nl,x,' : ']), answer(a), (A=yes, lis_sub(rx,rxl,1st) ;A=no, fail ;A=end_of_file, end(1st),! ). lis_sub([_ Rx],Sub,Ist) :- lis sub(rx,sub,1st). BOTTLING PLANT DESIGN EXPERT SYSTEM USER INTERFACE tipcont(1,[-1],end_of_file) :-!. tipcont(0,[-1],-999) :-!. tipcont(ind,asclist,answer) :- contasc(ind,asclist), contnum(ind,asclist,list), transf(ind,list,answer), tipcont(ind,_,_) :- tip(ind,text), nl,write(^fail! Data isn't ^) write(text), nl,write(^ New data: ^), fail. tip(0,number). tip(1,text). INFERENCE ENGINE: BUILT-IN PROLOG BACKWARD CHAINING REASONING CONTROL RULES: pro]ect :- heading, datareading, def_substance. def_substance :- defsubs(1st), ( Ist=l, interruption, end ; def charact subst ). PROGRAM Figure 1: The bottling plant design expert system.

Artificial Intelligence in Engineering 201 In our case the user interface gives to the user the following possibilities: data input type control (number or text), limited data control, the use of

5 Artificial Intelligence in Engineering 201 In our case the user interface gives to the user the following possibilities: data input type control (number or text), limited data control, the use of default data, data transformation, user's influence on decisions, help for the user, messages, warnings, information and results printing, etc. We already mentioned that we have used Prolog's own built-in inference engine. Besides that we have needed the rules, to control the bottling plant design process. The design process was divided into sub tasks. The data definition is followed with the selection of the machines starting with the bottling machine, which is a bottling plant base. At the end the results are listed on the screen and the bottling plant offer is written on the file. The sequence of sub tasks is determined, so the design task is always performed in the same order. The procedure 'project', starts the execution and activates the first sub task - the data definition by calling the procedure 'def_substance'. After the task is successfully completed, the next sub task is activated. A bottling plant design problem is concluded after all the sub tasks are completed. As the result the bottling plant configuration offer is made. In the case of sub task failure, the message is written on the screen and the first sub task is activated again. The procedure which writes the part of results on the file is activated by the interruption of the first sub task and the design process is broken off. Regarding to the fact that our system is a prototype, some restrictions were implemented as follows. The system can select only the R&S machines, which is an advantage for the company and its interest. All bottling substance are divided into groups: oil, syrup, brandy, wine, juice, beer, mineral water, carbonated drink and champagne, but more specific definition of the substance, for example "Coca-Cola", can determine also the bottle, the labels and the crate. The bottle selection is limited with volume from 0.2 litre to 2 litre and with standard, for example: the bottle 0.5 litre determines the beer bottle, which is 228 mm in height and its diameter is 70.5 mm. However all restrictions are going to be abolished with the knowledge base extension in the future development. 4 Results The input data are entered interactively, some of them during the execution of the machines selection process. After the selection of the machines is successfully completed, the system extracts the list of the selected machines and basic information about them on the screen. The user has two possibilities. The first one enables him to return to the start and to modify the input data in order to get the alternative solution. The ES result - the bottling plant offer is printed on the file as a second possibility. On the heading page of the offer the addresses of the producer and the customer are given as well as the basic input data.

202 Artificial Intelligence in Engineering After that the selected machines with the capacities and the prices are listed and the total price of the bottling plant is calculated.

6 202 Artificial Intelligence in Engineering After that the selected machines with the capacities and the prices are listed and the total price of the bottling plant is calculated. The machine descriptions with the technical data are included at the end of the bottling plant offer if the user asks for it. The system was tested on several examples for bottling e.g.: mineral water, oil, juice, beer, champagne. A comparison with the conventional design procedures was made. The testing process has been performed in two steps. First, we have collected the old bottling plant offers and tested the system using the correspondent input data. After we have been satisfied with the results of the primary test, the system has been installed in R&S and tested by the users. The users have no problems with the system and they are satisfied with the results. Previously they needed about one day for the elaboration of the bottling plant offer. Today the usage of the system enables them to prepare the data and to design the bottling plant in about half an hour at most. 5 Conclusions An ES evaluation confirms that the artificial intelligence techniques application is an effective approach to solve a bottling plant design problem. Until now, a prototype of the ES, that solves only the first task of the problem - the selection of the machines and the bottling plant offer elaboration - has been developed. A number of suggestions for the future work have been proposed during the testing process, therefore we are planning to continue with the further development of the bottling plant design ES. First of all we would like to solve the problem how to arrange the selected machines into the available room. After that the machines should be linked with the appropriate conveyors. The graphical library of all bottling plant machines and equipment exists already in the computer aided design system for two dimensional drafting. The connection between this system and the ES for bottling plant design is planed for the future. Our vision of how this integration should look like can be seen in Figure 2. Acknowledgements Our gratitude should go to the experts at Radenska & Steinle and to prof. Ivan Bratko from AI laboratory of Jozef Stefan Institute, Ljubljana. Research was financially supported by the Ministry of Science and Technology, Republic of Slovenia.

7 Artificial Intelligence in Engineering 203 USER INTERFACE INFERENCE ENGINE BOTTLING PLANT OFFER THE BOTTLING PLANT LAYOUT THE MACHINES CONECTION WITH CONVEYORS Figure 2: The ES integration with 2D-CAD system. References 1. Bratko I. Fast Prototyping of Expert Systems Using Prolog, Part 3: Design techniques, Topics in Expert System Design, G. Guida and C. Tasso (Editors), pp , Elsevier Science Publishers B.V. (North- Holland), Gero J.S. Expert Systems for Design: A Framework, in Proceedings of The World Congress on Expert Systems, Orlando, Florida, USA, Novak M. Expert System for Bottling Plant Design Using Artificial Intelligence Techniques (in Slovene with English abstract), M.Sc. thesis, University of Maribor, Faculty of Technical Sciences, Maribor, Slovenia, Schalkoff R. J. Artificial Intelligence: An Engineering Approach, McGraw- Hill Book Company, Singapore, 1990.

Capturing and Adapting Traces for Character Control in Computer Role Playing Games

Capturing and Adapting Traces for Character Control in Computer Role Playing Games Jonathan Rubin and Ashwin Ram Palo Alto Research Center 3333 Coyote Hill Road, Palo Alto, CA 94304 USA Jonathan.Rubin@parc.com,