可修復系統現場失效數據之分析與預測

Abstract

在分析可修復系統之平均累積失效數時,一般文獻所提的方法均欠缺對季 節效果的完善考量,但由於自然或人為因素導致的季節效果,對系統在使用 時的平均累積失效數會造成顯著的影響 ,所以如果在分析與預測時能考慮 到季節效果,將可使預測結果準確,同時也可提供系統作為改善設計與後勤 支援的參考.因此,本論文提出新的分析與預測可修復系統現場失效數據之 方法.新的方法採用疊加更新過程來分析可修復系統.首先利用NELSON之圖 形法來估計系統的平均累積失效數,其次再利用季節性ARIMA模式來分析失 效數據的季節效果,並預測系統未來的平均累積失效數與失效率變動的情 形.最後,本論文利用汽車失效數據的實際資料來比較本論文所提之預測方 法與其他預測方法所得到的結果,驗證了本論文所提方法較接近實際資料. When analyzing the mean cumulative number of failures for repairable systems, the usual prediction methods are : MIL- HDBK-217, Lars Rimested's method and regression analysis. All of these methods ingore the seasonal effect of the failure data However, due to some natural and manual factors, the seasonal effect sometimes do affect the failure data significantly for repairable systems. Failure to consider the seasonal effect in failure data would cause the inaccurate prediction of the mean cumulative number of failures. The main objective of this study is to propose a procedure which considers the seasonal effect to analyze and predict the failure data for repairable systems. The procedure applies Surperimposed Renewal Process and nonparametric graphical method, developed by Nelson in 1988, to analyze the failure data . Seasonal Arima models is also applied to determine the seasonal effect and is used to predict the mean cumulative number of failures and failure rate. Finally, several prediction methods are compared through a real- world failure data from automobiles industry. It was shown that Seaonal Arima model can effectively depict the seasonal effects and therefore is a better method for analyzing and predicting the mean cumulative number of failures for repairable systems.