MODELING AN EXPERT SYSTEM IN A TYPICAL DIAGNOSTIC ENVIRONMENT

Abstract

Expert systems are computer programs designed to emulate the logic and reasoning processes, human experts would use to solve a problem in their field of expertise. Interest in the area has grown rapidly with the emerging availability of Artificial Intelligence (AI) based develop ment techniques and tools. By emulating human reasoning in its ability to comine objective and subjective knowledge, expert systems can expand both our capabilities and availa bility of specialized expertise. Much of the work in modeling through expert systems involved abstracting the essence of a systen and representing it in suitable formalism. The chief advantage in modeling through expert systems is that they contain the knowledge necessary to generate computable structures from abstract symbolic models and find the natural place for incorporating heuristic methods to perform iterative changes leading to meaningful solutions. The technologies and techniques that have been developed in this area have wide applications covering many aspects including creativity and imagination. An important application area of this technology has been found via diagnosis of systems in operational/industrial settings. The work undertaken in this dissertation is motivated with a view to maintain safe, orderly and efficient 11 flow of air traffic in the country by automating manual diagnostics of VOR (very High Frequency Omnidirectional Range), a typical navigational aid, by expert system modeling techniques. There are several reasons for choosing this particular system for diagnosis. The major problem is due to increasing installations of the navigational aid throughout the country resulting in a scarcity of field maintenance experts. The problem is further aggravated due to the fact that existing troubleshooting techniques make complete and comprehensive maintenance of the system difficult, laborious and time consuming. Another problem arises due to turbulent emotions and anxieties which often seem to be interfering when breakdown occurs. The preliminary investigations during the conceptual stages have revealed that the diagnosis of the system is radically different from those developed earlier in the domain of diagnosis. This realization came while studies were made on radiated signals which indicated their deterioration due to several factors. This revelation put additional demands on software development-. The disserta tion, in general, exploits the typicality of diagnostic environment for modeling an expert system named VORFAD (VOR Fault Diagnostician) by abstracting the essen.ce of the system from various sources and replicating its

performance in software. In an attempt to meet the diagnostic requirements of the system two expert modules, CEAAM (Course Error Analysis and Alignment Module) and FDM (Fault Diagnosis Module) are developed. The primary purpose of CEAAM is to recognize deterio ration of VOR performance prior to complete out-of-tolerance conditions and to suggest alignment(s)/adjustment(s) and also location of malfunctioning components. The development of CEAAM is based on mathematical modeling of the error samples obtained from VOR equipment and analyzing them using heuristical techniques and bit strings. The concept of bit strings in data interpretation task has been effectively inducted. The FDM on the other hand, captures the abnormalities of the system for fixing the faults. The diagnostic capabi lities of the system are developed through extensive collaboration with field experts, guidance from manufacturers and maintenance manuals of the system. The result of this interaction has been consolidated in the form of general signal-tracing skills and construction of Knowledge Base (KB). The signal-tracing philosophy elucidated in our work is based on pragmatism of the field techniques applicable to most troubleshooting situations;in general. Various knowledge representation schemes are examined while seeking the most appropriate methodology for the

development of FDM. The experiential knowledge of the experts utilizes rule-based approach of the problem solving. The signal-tracing skills, structural and functional interdependence are incorporated in a data structure called here as pseudoframes. The pseudoframes clubbed with bit representation techniques, an entirely new methodo logy, are developed for building a deep knowledge base of the system. The Inference Mechanism (IM) is designed to interpret the information available and taking decisions by integrating various sources of knowledge. One of the major responsibilities of IM in VORFAD is to determine when and how to use which source of knowledge. To achieve the objectives IM is developed in three modules working together (i.e., Data Interpreter (DI), Rule Interpreter (Ri) and Pseudoframe Interpreter (PSFI)). The DI uses bit strings and heuristic rules to analyze course error data, while RI takes advantage of the techniques for interpreting domain specific knowledge in rule-based formalism. The development of PSFI is based around PSF structure conceptually denoting a computer word(s). The Explanation Module (EM) is designed to enhance the system's transparency by examining its reasoning strategy to expert and/or user. In doing so, EM makes use of system's static KB and dynamic consultation records. As the incremental development of the system continues throughout the life cycle of an expert system, knowledge acquisition module is incorporated. This module is invoked when an expert user discovers an error or lacuna in the system's performance and/or new knowledge is required to be added. Currently, the knowledge acquisition process is dependent upon handcrafting and various updaters. In an attempt to demonstrate the generality of VORDAD's approach, a sample example is taken. The accuracy of the system's decisions has been checked by actual field data obtained from the office of NAA (National Airports Authority), New Delhi. The VORFAD system will serve the mission of the Civil Aviation for all those who are engaged in the manual diagnostics of VOR systems all over the world. The developed system would also help reducing manpower requirements for ground servicing, increased availability and sustainability rates and expose fewer ground personnel to drudgery of the maintenance work. The system has been implemented on DEC-2050 mainframe computer in LISP programming environment.