機械設備故障診斷及維修支援之多代理人系統研發

Abstract

本文發展一套用於機械設備故障診斷及維修支援之多代理人專家系統。首先根據美軍之MIL-STD-1388-2B規範建立一個知識庫，此知識庫包含後勤資訊、錯誤症狀、症狀之失效原因、維修程序、維修敘述等資訊。接著從知識庫擷取設備故障症狀與其可能原因間的關係，並以知識精鍊策略粹取精鍊後之知識。另外提出一種混合式之故障診斷推論方法，此推論方法整合知識為基與案例為基推論方法，並根據零件可靠度及症狀發生機率以找出故障原因。最後以Prolog、Java及C++語言發展一多代理人專家系統程式，此程式透過網際網路進行故障診斷、知識精鍊、以線上手冊及多媒體方式呈現維修工作及敘述給使用者、藉分析維修紀錄自我學習等；此程式也提供服務給網際網路上經授權使用之電腦。最後此多代理人系統以一複雜機械設備之知識庫進行測試，測試結果顯示本文提出之知識精鍊策略及混合式推論方法能有效率地辨識設備的各種故障原因。 關鍵字：故障診斷、維修支援、混合式推論、多代理人系統

In this thesis, a multi-agent system for fault diagnosis and maintenance support of mechanical equipments is developed. First, a knowledge base is created based on the MIL-STD-1388-2B format. It contains logistic information, fault symptoms, causes of symptoms, maintenance procedures and statements, and repair records. Then, relationships between the symptoms and causes of faults are extracted from the knowledge base and refined by proposed knowledge refinement strategy. A hybrid reasoning method is proposed for fault diagnosis. It is formed by combining the knowledge-based reasoning and the case-based reasoning, and based on reliabilities of parts and probabilities of symptoms for identifying fault causes. Finally, a program of multi-agent system is developed using Prolog, Java, and C++ for knowledge refinement, fault diagnosis, presenting maintenance procedure and statements with on-line manual, and self-learning from analyzing maintenance records. The program also provides Internet services such that it can be run through any authorized client computer. The developed program is tested with the database of complex mechanical equipments. The diagnosis process shows that the proposed knowledge refinement and new inference strategy can identify the causes of fault symptoms efficiently. Keywords：Fault diagnosis, Maintenance support, Hybrid reasoning, Multi-agent system