以預防保養為目的之失效分析與預測方法

Abstract

以預防保養為目的之失效分析與預測方法 研究生：楊善國 指導教授：呂宗熙 教授 國立交通大學機械工程研究所 摘要 如果裝置或系統的失效可以被預測得知，則經由預防保養可防止失效的發 生，從而提高裝置或系統的可靠度。本研究旨在分別以裴氏網路法及卡曼 濾波器法，進行失效分析及失效預測的探討。裴氏網路對於各樣系統的模 型建立均非常有用。本文以安全氣囊之充氣系統為例子，說明了如何以裴 氏網路法尋找最小割集以及描述系統失效的標識轉移和過程。為達到預防 保養的目的，本文提出了一個配合參數趨勢和故障樹的複式裴氏網路建模 法，用以執行早期失效偵測與隔離。根據此建模法，本文發展出一個可達 到警告、早期失效偵測、故障隔離、事件計數、系統狀態描述、以及自動 關機或調節的裴氏網路，並以空調設施的故障診斷系統為例子說明此方法 。接著本文以現場可程式邏輯閘陣列 (FPGA) 晶片執行了裴氏網路。除將 裴氏網路的基本符號和邏輯架構實現成邏輯電路外，早期失效偵測與隔離 的裴氏網路亦被裝設成一個特用積體電路。此外，本文藉卡曼濾波器於直 流馬達上進行狀態估測，提出了以預防保養為目的的失效預測方法。將一 指數衰減器置放於馬達模型的輸出端並監測狀態變數，以模擬老化失效。 失效時間係由蒙地卡羅模擬法產生，且由卡曼濾波器預測。文中所執行領 先一步及兩步等兩種預測之結果的誤差均可接受。

Failure Analysis and Prediction Methods for Preventive Maintenance Student: Shang-Kuo Yang Advisor: Prof. Tzong-Shi Liu Institute of Mechanical Engineering National Chiao Tung University ABSTRACT Failure can be prevented in time by preventive maintenance so as to promote reliability only if failures can be early predicted. This study aims at failure analysis and failure prediction by Petri nets and the Kalman filter methods, respectively. Petri nets are useful for modeling a variety of systems. An airbag inflator system is employed as an example to demonstrate a Petri net approach to finding minimum cut sets and describing marking transfer and scenario of system failure. To achieve preventive maintenance, a hybrid Petri net modeling method, endowed with parameter trend and fault trees, is presented to perform early failure detection and isolation. A Petri net is developed to facilitate alarm, early failure detection, fault isolation, event count, system state description, and automatic shutdown or regulation. A fault diagnosis system for district heating and cooling facilities is used to exemplify the proposed method. A Field Programmable Gate Array (FPGA) is implemented, so as to realize basic Petri net symbols and logic structures in Petri nets by logic circuits. An ASIC is implemented for the early failure detection and isolation Petri net. Besides, a failure prediction method for preventive maintenance by state estimation using the Kalman filter is presented dealing with a DC motor. An exponential attenuator is placed at the output end of the motor model to simulate aging failures by monitoring state variables. Failure times are generated by Monte Carlo simulation and predicted by the Kalman filter. One-step-ahead and two-step-ahead predictions are conducted. Resultant prediction errors are acceptable for both predictions.