Uzman sistemler ve uygulamaları

Abstract

Uzman sistemlerin bir çok alanda uygulanmaya başlandığı günümüzde, usman sistemlerin ne olduğunun, nasıl kurulduğunun, nasıl kullanıldığının ve ne işe yaradığının anlaşılması gerekliliği bu çalışmanın temel amacını oluşturmaktadır. Bu çalışmada usman sistemlerin tanımı, amacı, kapsamı, yapısı ve kurulması için gerekli alt yapının nasıl oluşturulacağı gibi teorik bilgilerin yanında uygu lama alanları, yapılmış uygulamalardan örnekler ve açıklamalar da yer almaktadır. Herhangi bir alanda bir uzman gibi davranacak, problemlere usman gibi çösümler ve öneriler getirecek, yorumlar ve açıklamalar sunacak bilgisayar programı olarak tanımlanan usman sistemlerin özellikleri, yararları, problemleri ve kısıtları çalışmanın ikinci bölümünde açıklan maktadır. Usman sistemlerin temelini oluşturan bilgi temin etme, elde edilen bilgiyi uygun şekillerde ifade etme ve bilgiyi gerekli alanlara transfer etme gibi önemli olaylar da yine bu bölümde açıklanmıştır. üçüncü bölümde, usman sistemlerin kuruluş aşamaları; usman sisteme uygulanacak problemin alt problemlere ve gerçeklere ayrıştırılmasmdan, kullanılacak donanım ve yasılım seçimine, programlama tekniklerinin belirlenmesine ve kullanılacak programlama dil( ler) inin seçimine kadar olan başlıkları içerir. Dördüncü bölümde, usman sistemlerin hangi alanlara uygulandığı, hangi alanların bu uygulama için daha yararlı ve uygun olacağı anlatılmaya çalışılmıştır. Uygulama için seçilen alan uzman sistem kullanmaya ihtiyaç duyan bir alan olmalıdır. Yani usman sistem kullanmanın diğer yön temlere göre maliyet, saman ve emek açısından daha kasançlı; ulaşılan sonuçlar bakımından daha kaliteli ve uzman kişi kapasitesini aşan kompleksklikteki problemler için de uygun çösümler bulabilmenin mümkün olduğu alanlar uygulama için seçilmelidir. Çalışmanın bu bölümünde kalite kontrol, karar destek sistemleri, üretim programlama, ergonomide yük kaldırma, çalışma yeri dizaynı, fisiksel iş yükü analizi gibi alanlarda usman sistemlerin nasıl uygulana bileceği hakkında bilgi verilmekte ve şimdiye kadar yapılmış çalışır durumdaki usman sistemlerin yapıları ve amaçları anlatılmaktadır. Beşinci bölüm ösel olarak ergonomi alanındaki usman sistemlerle ilgilenmekte ve bunlar hakkında detaylı bilgiler sunmaktadır. Ayrıca bu sistemlere yapılabilecek ekle- vii malar, getirilebilecek yenilikler de önerilmektedir. Altıncı bölüm olayın biraz daha bilgisayar ağırlıklı yönünü incelemektedir. Uzman sistemleri kurmak için yararlanılan araçlar tanıtılmakta, gerekli yazılım ve donanım desteği hakkında bilgi verilekte ve programlamada kullanılan yapay zeka dilleri hakkında açıklamalar sunulmak tadır. Uzman sistemler ve uygulamaları hakkındaki bu çalışma, uzman sistemlerin gelecekte kaydedebileceği gelişmeleri ve ergonomi alanında kurulabilecek bir uzman sistemin içermesi gereken bazı bölüm ve değişkenleri sunan sonuç bölümüyle son bulmaktadır.

Expert systems have recently become popular and are attracting more and more attention. The high quality performance achieved by some systems in areas previously not considered practical for computational solutions has lead to great interest from many different disciplines. Most expert systems use a subset of techniques from the general area of computer science research known as artificial intelligence (AI). However, some expert systems have been developed that incorporate more traditional mathematical modelling techniques. The combination of artificial intelligence techniques has been shown to be quite effective in developing several high quality performance computer software systems. AI which is a topic of basic computer science research, is how being applied to problems of scientific, technical and commercial interest. Some consultation programs, although limited in versatility, have achieved levels of performance rivaling those of human experts. Research in AI has several goals. One is the development of computational models of intelligent behavior, including both its cognitive and perceptual aspects. A more engineering-oriented goal is the development of computer programs that can solve problems normally thought to require human intelligence. Expert systems as a branch of AI has already achieved significant success in helping to solve problems in a number of scientific fields. The expert system term generally implies a computer software system that has achieved a level of performance in a limited domain which approaches that of an acknowledged human expert in the same domain. On the other hand, an expert system is a computer program that solves problems that heretofore required significant human expertice by using expilicitly represented domain knowledge and computational decision procedures. Expert systems can be grouped into three major categories depending on the kind of problems they were designed to address. By far the most common are classification systems dealing with problems of diagnosis. Medical diagnosis and interpretation of geological data have been two most widely explored application areas in this category. The second category include systems dealing with problems of design. The third category ix comprises expert systems designed for decision support, It is not enough to develop an expert system only with a computer programmer. It is needed the interaction between some different disciplines. These can be explained as follows: Expert system (the computer program that can behaive like a domain expert), domain expert (the human expert who has detailed knowledge and a lot of experiences in that domain), knowledge engineer (who designs the expert's knowledge and data bases), expert system building tools (these are used to simplify to build an expert system such as shells, tools and specific expert systems) and user interface (who use the program). An expert system includes three different modules: 1) A knowledge base, 2) An inference mechanism, 3) A user interface. There have been two distinct approaches taken to expert system construction. One relies on using empricial associations to construct realitionships between concepts, without considering the cause and effect factors. An alternative approach to empricial associations is to use qualitative, model-based reasoning, derived from knowledge of structure and behavior. This is sometimes described as reasining from first principles. This approach has been especially used in problems of design and fault diagnosis in computers, but it has also been explored in developing causal reasoning for medical diagnosis. The choice of either type of approach to reasoning an expert systems largely depends on the characteristics of the domain knowledge. A rule-based, surface reasoning may be appropx'iate in situations where knowledge about application is essentially dependent on experience gained through exposure to a large collection of bases and where a suitably large assembly of typical examples can be obtained. Many expert systems researchers have recently focused on determining the appropriate knowledge representations to use in order to achieve high quality performance from knowledge-based systems. After the knowledge is acquisited, the appropriate knowledge representation technique must be chosen. There are three x types o£ knowledge representation techniques: 1) Rules 2) Networks 3) Frames Rules generally are in the form of "IF conditions THEN conclusion". If the conditions are satisfied then the conclusion is obtained. Networks are formed by dividing the problem into subproblams to understand the problem more easily. Finally frames are the more detailed structures according to the networks. Frames can present more detailed knowledge about the problem and subproblems to the programmer or user. And by using frames the errors and gaps between the facts and rules can be recognised easily. After the knowledge representation is completed the inference mechanism can work. The goals can be satisfied by using the inference mechanism. Inference prosses if completed in two different ways: Forward chaining and backward chaining. In a forward chaining system, rules for which the conditional part is satisfied are evaluated. The evaluation of therir cosequent parts will update working memory. This will cause the conditional part of other rules to be satisfied and they will become candidate for evaluation. The process cotiniues until the conditional part of no rule is satisfied. This strategy is also called data-driven because the choise of which rule to evaluate next depends on updates in working memory. In a backward chaining system, a special rule is chosen for evaluation. In order to evaluate that rule, the goal of satisfying the conditional part of that rule is established. Therefore rules whose consequent parts will update working memory such that the conditional part of the chosen rule will be satisfied are selected for evaluation. This, then, establishes subgoals that must be satiafied to satisfy the first goal. This strategy is also called goal-driven because the choice of which rule to evaluate next depends on established goals. One of the most widely explored aspects of human expert behavior in expert systems has been the ability to explain and justify conclusions. In most cases explanation mechanism consists of a direct record of rules and facts used by the program to reach the current stage of the problem-solving task. Most current expert systems have used intuitive methods for dealing with inexact knowledge and incomplete evidence. This is because statistical samples are often xi impossible to obtain. Another method commanly used in expert systems for drawing conclusions under conditions of uncertainty is based on fuzzy sets and fuzzy logic. It is led to use verbal veriables instead of numerical veriables in fuzzy sets. Each verbal veriable describes a numerical value and when we use a verbal veriable it is understood numerical value by computer. Fuzzy sets theory let us to use the principles of naturel language. Therefor it is possible to present real-world problems in a better way. The selected method and used programming language should let the programmer to use structured knowledge unstructured knowledge and incomplete or ill-structured knowledge. Among the special languages are PROLOG (Programming With Logic), LISP (List Processing), SAIL, RLL, KRL, PLANNER etc. Many of these have been used in expert systems research. In recent years, there has been a growing interest in development of tools designed specif icially to assist the expert system's builder in the execution of the task. One Important class of expert systems tools is that of "shells". Among the most widely used expert systems shells are EMYCIN, EXPERT, KAS, HEARSAY-III and SAGE. One of the main problems associated with using shells is that the characteristics and requirements of the problem-domain are very rarely adequately matched by the design facilities offered in the shell. The reason is that it is not enough just to put the knowledge in a knowledge-base. The expert system development process also includes following steps: Identify the problem, select expert (s), select the hardware and software tools' knowledge acquisition, construct a prototype expert system which includes a justifier, a knowledge-base, an inference mechanism as a sample of the complete expert system, evaluate the system performance, evaluate the acceptance of the system, parallel usage with current systems, dokumentation and develop maintenance plans, test the validation and relaibility of the system and finally use the system. Expert systems can be applied various areas. Before the application the area must be evaluated. because to develop an expert system is very expensive and toconstruct an expert system should support a lot of advantages. By far many expert systems have been developed in various areas such as decision support systems, diagnosis in medical, medical consultation, quality control, geologic explorations, chemical compounds, production schedualing, layout planning, ergonomics, debugging in computers etc. In this thesis, the expert systems, developed in this areas, are tried to introduce. Their tasks, problem-solving methods and design tolls are presented. xii types o£ knowledge representation techniques: 1) Rules 2) Networks 3) Frames Rules generally are in the form of "IF conditions THEN conclusion". If the conditions are satisfied then the conclusion is obtained. Networks are formed by dividing the problem into subproblams to understand the problem more easily. Finally frames are the more detailed structures according to the networks. Frames can present more detailed knowledge about the problem and subproblems to the programmer or user. And by using frames the errors and gaps between the facts and rules can be recognised easily. After the knowledge representation is completed the inference mechanism can work. The goals can be satisfied by using the inference mechanism. Inference prosses if completed in two different ways: Forward chaining and backward chaining. In a forward chaining system, rules for which the conditional part is satisfied are evaluated. The evaluation of therir cosequent parts will update working memory. This will cause the conditional part of other rules to be satisfied and they will become candidate for evaluation. The process cotiniues until the conditional part of no rule is satisfied. This strategy is also called data-driven because the choise of which rule to evaluate next depends on updates in working memory. In a backward chaining system, a special rule is chosen for evaluation. In order to evaluate that rule, the goal of satisfying the conditional part of that rule is established. Therefore rules whose consequent parts will update working memory such that the conditional part of the chosen rule will be satisfied are selected for evaluation. This, then, establishes subgoals that must be satiafied to satisfy the first goal. This strategy is also called goal-driven because the choice of which rule to evaluate next depends on established goals. One of the most widely explored aspects of human expert behavior in expert systems has been the ability to explain and justify conclusions. In most cases explanation mechanism consists of a direct record of rules and facts used by the program to reach the current stage of the problem-solving task. Most current expert systems have used intuitive methods for dealing with inexact knowledge and incomplete evidence. This is because statistical samples are often xi There are still largely unexplored applications for expert systems in management and operational research. The combination of new AI techniques with more traditional mathematics will probably produce a new generation of high quality performance expert system. Consequently expert systems are seen as they will have a hopefull future as a branch of computer science.