透過您的圖書館登入
IP:3.145.184.113
  • 學位論文

應用長短期記憶法建立品質績效指標預測模式-以飛機維修護廠為例

Using the LSTM Method to Establish Quality Performance Prediction Model - Case Study of An Aircraft Maintenance Organization

指導教授 : 蕭育霖

摘要


航空運輸業的核心就是飛航安全,根據統計飛航事故主因除了人員因素之外,其次就是與飛機本體相關其佔比接近五成,因此飛行器的維修保養在飛航安全中是很重要的一環,故飛行器的維修組織就變得格外重要。近幾年,我國民用航空局除主動執行安全查核外,亦要求維修組織建立自我查核,並視需求接受外部主管機關或客戶執行查核,因此維修組織可收集到不同來源的安全查核資料。如何充分分析這些安全資訊,用以提升組織風險管理,是值得深入探討的研究課題。本研究目的即在於運用上述安全查核資料建立品質績效指標,並深入探討績效指標的預測能力。研究分析數據來自國內某業者所提供之查核紀錄資料,再經由專家分析查核紀錄之主要發生原因,最後將其分類為人員類、環境類及管理三類績效指標。後續則使用長短期記憶(Long Short-Term Memory, LSTM)建立品質績效指標模型,用以提醒維修組織事先提出防範措施。研究結果顯示,使用LSTM方法在人員類正確率為88%,環境類正確率為90%,管理類正確率為80%,雖然仍有過度描述現象,但研究結果仍可驗證使用類神經網路法建立趨勢分析機制在實務上的可能性。此模型預測準確率還有再進步的空間,未來可繼續深入探討,期望所預測的品質指標可應用於航空業界,並提高飛航安全。

並列摘要


According to statistics, the causes of flight accidents were related to the aircraft itself, accounting for nearly 50%. Therefore, aircraft maintenance and repair organization (MRO) is an important part of flight safety management. This study uses the inspection report of a maintenance organization to establish its Quality Performance Indicators (QPI). The qualitative inspection data was classified into Personnel, Environment and Management, and then transformed into quantitative rates. Long Short-Term Memory (LSTM) was used to establish a prediction model for QPI. The research results showed the accuracy rate for Personnel, Environment and Management were 88%, 90%, and 80% respectively. The results could prove that it is feasible in practice to use the neural network-like method to establish safety trend analysis for MRO. The prediction of QPI can be considered as a possible application to the aviation industry to improve flight safety management.

參考文獻


外文部分:
1. S. Hochreiter.; J. Schmidhuber. (1997). Long short-term memory. Neural Computation, 9(8):1735–1780.
2. Successful SMS Implementation Leads to a Successful Business. (2014). Flight Safety Quarterly, Autumn
3. ICAO (3rd ,2013) Doc 9859 Safety Management Manual
4. Boritz, J., Kennedy, D. (1995). Effectiveness of Neural Network Types for Prediction of Business Failure. Expert Systems with Applications, 9(4), 503-512.

延伸閱讀