Küçük boy telefon santrallarının bakım ve onarımı için bir uzman sistem tasarımı

Abstract

Bu tezin ilk bölümünde uzman sistemler konusuna giriş yapılmış ve uzman sistemlerin diğer sorun çözücü sistemlere olan üstünlüğü tartışılmıştır. İkinci bölüme uzman sistemlerin bir tanımıyla başlanmış ve uzman sistemlerin temel özellikleri anlatılmıştır. Uzman sistemlerde temel ilkeler buluşsallık, saydamlık ve esnekliktir. Temel düşünce, uzmanlık bilgisini olabildiğince bu bilgiyi isleyen denetim işlemlerinden bağımsız duruma getirmektir. Üçüncü bölümde uzman sistemlerde en çok kullanılan gösterim ve çıkarım yöntemleri kural tabanlı gösterime ağırlık verilerek anlatılmıştır. Kural tabanlı gösterimin dışında, çerçevelerle gösterim ve mantığa dayalı gösterim de tanıtılmıştır. Dördüncü bölüm uzman sistem geliştirme aşamalarına ayrılmıştır. Temel basamaklar sırasıyla: tanımlama, kavramsal tasarım, biçimlendirme, gerçekleştirme, sınama ve doğrulama, sürüm sonrası geliştirme ve bakımdır. Sistem geliştirilmesi sırasında karşılaşılan en önemli sorunlar bilginin kırılganlığı ve tutarsızlığıdır. Besinci bölümde, Bilgi Toplama sürecinin tanımı, teknikleri, kısıtları anlatılmıştır. Bilgi toplama yaklaşımları: karşılıklı görüşme, etkileşim yoluyla öğrenme, tümevarım yoluyla öğrenmedir. Altıncı bölüm uzman sistem geliştirme araçlarının özellikleri tanıtılmıştır. Uzman sistem geliştirme araçları dört sınıfa ayrılabilir: izlenceleme dilleri, bilgi mühendisliği dilleri, sistem geliştirme yardımcıları, destek araçları. Bu araçlardan birkaçını veya hepsini içeren kabuklar uzman sistem geliştiricilerine kolaylık sağlar. Yedinci bölümde, bakım, onarım ve hastalık teşhislerinde kullanılmak amacıyla tasarlanmış bazı uzman sistemlerin özellikleri sıralanmıştır. Gerçek zamanda çalışan sistemlerde karşılaşılan hata ve sorunlara çözümler üretebilecek uzman sistemlerin özellikleri ayrı bir başlık altında toplanmıştır. Sekizinci bölümde, bu tez kapsamında, NETAŞ firması tarafından üretilen DX1 telefon santrallinin bakımı ve onarımı için tasarlanan uzman sistemin geliştirme aşamalarını ve özelliklerini anlatmaktadır.

Expert systems are constructed when the real experts are rare at the problem domain or hard to reach. Acquisition of the knowledge that have been gained through the years of experience, and supplying this knowledge to nonexperts are the main motives to develope expert systems. In this thesis, the main components of the expert systems, expert system development tools, classical and recent expert system applications and expert systems developed for diagnostic purposes are studied in deep. And after this study a diagnosis and maintenence expert system for small size telephony switches is developed up to the prototype stage. This work is motivated by the need at the field support department of company NETAS. This company is producing and supporting telephony switches. Field support of telephony switches is a very complex job. In addition to that problems may happen again and again at many sites. And experienced real experts are always scarce. Classical methods in problem solution are ineffective because of the complexity of the problem domain. Expert system techniques are very convenient while dealing this kind of problem domains. The main aim to develop an expert system is to do the field diagnostic an field maintenance of the telephony switches without the help of real experts. The second aim is to train the new and unexperienced field support personel. Plans for the future development is to help the end-users to maintin and to diagnose their own switch without calling support. ?st First part of this paper makes a short introduction to the expert systems which is the most practical appliction of artifical intelligence. Here, a comparison between the problem solving methodology used in classical programs and expert systems is presented. Conceptual breakthrough which can be simply stated as "to make a program intelligent, provide it with lots of high quality, specific knowledge about some problem area" is also introduced at this first chapter. The most important gain from the expert system technology is that high quality knowledge can be provided easily for a wider area. The second chapter which begins with a complete definition of expert systems is devoted to the principles of the expert systems. Different definitions of the expert systems are given according to several sources. Main idea behind the knowledge based paradigma can be stated as 'seperation of control and the problem specific knowledge'. Basic principles are transparency, elasticity and heuristics. Main components of expert systems are the knowledge base and the inference engine which operates on the knowledge base. The knowledge base editor which is used to edit the knowledge base is one of the main requirements to the expert system. In the third chapter expert system inference and knowledge representation methods are explained with the emphasis on the rule based knowledge representation and inference on the rule bases. The three requirements of knowlegde representation are capability, expandibil ity, and simplicity. In addition to the rule based representation frame based knowledge representation and logical representation are explained too. Main inference control strategies are.forward reasoning, backward reasoning, and combination of forward and backward reasoning. At the last section of this chapter uncertanity in the reasoning is explained. The forth chapter is devoted to the expert system development stages. Although the stages are very application dependent common steps can be listed: identification, conceptual design, formalization, implementation, testing and var i f icat ion, field maintenance and support. Main difficulties during the expert system development process are inherent limitations of expert system technology and the lack of resources (both money and human resources). Limitations of the expert system technology are unconcistency of knowledge at w the knowledge base and brittelnesa of knowledge (narrowness of the problem domain that the expert systems can effectively works.) The fifth chapter is about the knowledge acquisition process. Although this is a subprocess of expert system development process, because of its importance, knowledge acquisition techniques are introduced in a seperate chapter. Approches to the knowledge acquisition are disccusitons, learning by induction, learning by interaction. The knowledge acquired may be grouped in two classes : compiled knowledge and deep knowledge. Knowledge aqisition methods differ according to the type of knowledge, developers intented to acquire. Compiled knowledge requires discussions with the experienced experts but deep knowledge is mostly based on theoretical background behind the system being supported. Expert systems can be classified according to the knowledge type in the knowledge base. Expert systems which have mostly deep knowledge in the knowledge base are causal systems, those have compiled knowledge in the knowledge base are called shallow expert systems. The sixth chapter gives a coverage of proporties of expert system tools. After discussing the roles of programming languages, knowledge engineering languages, and system support tools at ES development process; shells which includes most of these expert system development tools, are explained. Shells keep the developers from reimplement ing most of the code which are common to all expert system programs. Shells make the knowledge aquisition easier, provide an inference engine that makes inference from the aquired knowledge. Shells support a user interface that both helps the developers and end users of the developed expert system. ARITY/EXPERT devolepment shell is used at the application stage thus this tool is widely introduced at this chapter. The seventh chapter is about the usage of expert systems at diagnosis, fault finding and maintenance of electronic devices. With some examples, main principles and adventages of these kind of expert systems are presented. Principles and adventages of expert systems which are used for fault finding and problem solving purposes on realtime electronic systems are explained under a seperate title. V//A. At the eightieth chapter, development process and specifications of the DX1EX expert system for diagnosis and maintenance of DX1 telophony switch which is a real time electronic system, - are explained^ Properties of this expert system would be applicable for all small size telephony swithes. Knowledge required to construct the system is provided from the engineers working at the field support departments of the company NET AS. Some knowledge is also acquired from the maintenance manuals distributed by the same company. The expert system is developed on the ARITY expert system shell. The program is running on the IBM Personel Computer Compatible systems and needs little affort to be integrated with the DOS based small size telephony switches. Since the ARITY development shell is DOS dependent, the program is unportable outside the DOS enviroment. But the experience gained while developing this program is independent and can be reused in a similar application on different platforms. Bibliography section lists the books, articles, and other documents which were used during the thesis study. The book "A GUIDE TO EXPERT SYSTEMS" by D.A. Waterman is the first and classical book about knowledge representation, rule based systems, inference, expert system development methods. Old dated data and examples are disadvantages of this book. The book "EXPERT SYSTEMS FOR EXPERTS" by Kamran Par s aye presents comtemporary information about knowledge based systems, knowledge representation, inference, frame and rule based repsenentat ion methods, expert system development life cycle, and expert system development enviroments with a practitioners approach. The book "EXPERT SYSTEM TECHNOLOGY" by R. Keller, after introducing general concepts about expert systems, gives a structural analysis approach to application selection. Last chapter of this book "Knowledge Based Information Center" tries to predict the advances at near future in knowledge based paradigma. The article "Real-Time Fault Diagnostics" by S. Padalkar reviews the techniques and characteristics of expert systems which are designed for fault diagnostics on real time systems. This article was published by IEEE Expert Magazine in June 1991. People who like to get more Information about the knowledge based systems and ES for diagnosis, can find many usefull sources at bibliography pages. File list of programs and knowledge aquired to develop the DX1 expert system make the last chapter of this thesis. DX1.TAX file includes the taxonomy or hierarchical functional structure of the target system (DX1 telephony switch). DX1.RUL is the rule file (rule base) of the expert system. It includes the rules (knowledge) aquired during the aquisition process. It also includes the control and user interface parameters. The last file is DX1.ARI. It is the procedure file. The procedures are actually PROLOG predicates. Most useful adventages of the ARITY shell is its interface to the PROLOG language.