動態案例式推論診斷維修系統

Dynamic Case-Based Reasoning for Remote Diagnosis and Maintenance System

指導教授 : 黃榮堂、陳正光  研究生 : 李汪龍  製造科技研究所 91年


摘要

  本研究係開發一動態監測系統,透過網際網路動態地擷取現場設備之相關參數資料,並結合統計及案例式推論的觀念成為動態式CBR監控系統,本系統利用經由統計後的資料作為品質管制的依據,來分析設備是否有發生故障的趨勢,藉以事先預防設備故障,進而在故障發生前先行通知客戶進行零件更換或保養,降低停工期的發生次數。另一方面,當設備發生突發性故障或動態監控系統未偵測到故障趨勢而發生故障時,維修工程師可連上遠端伺服器使用靜態式CBR診斷維修系統進行診斷維修,經由網頁瀏覽器輸入問題,透過CBR推論流程,協助解決問題。此系統結合Nearest Neighbor Algorithm、Fuzzy Set Theory及SQL等演算法,藉以提高新案例與案例庫中已存案例之案例相似度(Similarity),並輔以權重值進行加權計算,增加案例推論之正確性。
在案例學習的步驟中,當案例新增時,案例搜尋的權重值需重新調配,以維持案例搜尋的正確性。在本篇論文中,我們使用類神經網路Levenberg- Marquardt演算法來調整權重值,以增加下次搜尋之正確性。

ABSTRACT

  This paper presents a Dynamic Case-Based Reasoning (DCBR) system which consists a dynamic monitor system to obtain equipment parameters through Internet. The DCBR monitors the parameters and applies the technique of statistics to determine whether the status of any one of the parameter is toward abnormal range so as to prevent faults occur. Therefore, it allows the manufacturer to notify their customers to maintain equipment before any fault occurs. On the other hand, when any fault suddenly occurs or DCBR doesn’t discover faults while faults occur, the system can use Static CBR (SCBR) system for remote diagnosis. It uses web browser as a single interface to input the inquiry. Then, through the CBR, it can find the cause and solution of the fault. The SCBR system combines nearest neighbor algorithm, fuzzy set theory and SQL as well as weighting values to increase the accuracy of similarity during comparing the faulted case with the previous cases stored in the case-base.
When a new case is added, the weighting values will be needed to adjust in the process of Case Retain and Learning to keep the correctness of Case Retrieve. In this paper, the system applies Levenberg-Marquardt Algorithm of Backpropagation in Matlab Neural Network Toolbox to adjust the weighting values and raise the correctness of Case Retrieve at next time.

Key Words: Case-Based Reasoning、Maintenance、Diagnosis