石化廠維修履歷到可靠度數據轉換研究

在高度電腦化作業的時代，幾乎所有事務的通訊、記載及查詢都有相應的軟體可供使用。設備維修記錄也不例外，但一方面由這個電腦軟體中又可回饋出怎麼的資訊。設備失效率數據便是一個例子，如果資料欄位的規劃得當的話。失效率數據來自維修紀錄的後處理。但當檢視業界實質維修記錄（工單）內容時，卻又發現多數工單可能因原設計的不同，以致資料記載不完全，最後導致多數資料還未進入後處理之失效率計算階段，即面臨必須被刪除的命運。這些被刪除的數據不僅可惜，更重要的是計算品質已因其受到汙染，影響失效率計算的準確性。有鑑於此，本研究中規劃了一個標準工單架構，其考量有三，首先應能符合絕大多數電腦軟體的運作需求；其次須能保留原始維修記載內容；再者則為大量採用編碼制度進行作業，避免人工手寫鍵入所造成的人因問題。由維修記錄中的故障起迄時間轉換成設備失效率的過程看似直接，實則繁複，其中有許多需要克服的問題，例如，同類設備失效模式之歸類及小樣本數的統計分佈計算程式等。本研究以石化設備的維修履歷為例，從數據過濾、格式轉換到失效率計算，透過網際網路平台進行處理、運算，並可從不同篩選條件查詢失效模式與失效率數據。

關鍵字

維修履歷；數據轉換；失效率； ISO 14224

並列摘要

In this highly computerized era, almost all tasks are done with a computer program from sending out requests, recording process events; to counting results queries are all have appropriate software to processing and recording information. Equipment maintenance history is of no exception. A lot of information has been stored in the computer database, yet very little is done to explore the proper use of those data. One of the very important data is the failure rate of equipment which requires huge amount of quality maintenance records and careful post-processing. However, most of plant repairs were poorly recorded let alone later on statistical processing, and therefore a big portion of the equipment maintenance records has but be discarded. Also, in order to form a failure data base in the shortest period as possible, original data should come from different plants in the same industry, among which a big issue would be the different naming schemes of equipment terms as well as the types and contents of repair work order is collected. ISO 14224 is thus adopted as the common naming platform of equipment hierarchy and terms. Another practical matter to be encountered is at the early stage where data is scarce and statistical accuracy is difficult to maintain. We have managed to counter this by including different methods to cope with the limitation in sample size. In this paper, we utilize Weibull's method to demonstrate how to handle small data sets. The process of obtaining the observed failure rate seems straightforward. In reality, we are forced to work with the data format problem and with insufficient knowledge of the equipment lifecycle, which makes it extremely harsh to come up with proper failure rate data. In this study, a web-based platform for exchanging the raw maintenance records and for users to look up equipment failure rates has been established for this purpose.