Diagnosis Expert System Design And Implementation Of The Bayesian Network-based Semiconductor Equipment Failure

Since the equipments in semi-conductor factory are much too expensive, and the equipment performance will significantly affect total capacity of the factory, enhancing the maintenance efficiency and thus reducing the equipment down time is an effective way to increase factory capacity and reduce cost. Therefore, providing convenient and reliable decision-making for equipment fault diagnosis has become essential,while the structures of semi-conductor equipments become even complicated, which brings great challenge to their fault diagnosis.Intellectualization has become the tendency of fault diagnostic field, while expert system technology is the most widely used branch of artificial intelligence. But most of the researches tend to rely on sensor data and equipment structure, thus the developed expert system can only be used to the specific equipments or the equipments with the similar structure. According to this, making full use of the expert knowledge and developing a universal expert system is the focus of this paper.In this paper, the technology of fault diagnosis and expert system were firstly introduced, and the necessity and feasibility of developing fault diagnostic expert system in the semi-conductor factory are discussed. Based on this, a fault diagnostic expert system was designed. In the system, for the causal relationship between failure mode and corresponding reasons in the expert knowledge, production rule was adopted to represent the expert knowledge; also because the evidence, the rule and the conclusion are all uncertain with the uncertainty represented by possibilities, Bayesian Network is used as the inference machine. While applying Bayesian Network to the fault diagnosis, the improved causal relationship questionnaire was used to construct the structure of Bayesian Network, and the probability scale method was employed to define the probabilities more conveniently and quickly by connecting the verbal description of probability people frequently used and the actual numerical probability. Finally, the user demands were defined and the realization of the system was stated. The system consists of three modules: knowledge module, inference module and auxiliary module, the former two of which are the core modules. The system has already been developed using Microsoft Visual Studio 2000 and Microsoft SQL Server 2000 on Windows platform, and an application instance was provided to prove its rapidity and accuracy. The system, which can provide the possible fault causes and corresponding probability and help the maintenance engineer with fault diagnosis decision-making, has been successfully applied in a chipset assembly and test factory.