現今各型戰鬥機在作戰任務的需求上，加強了飛機雷達等多種功能的航空電子 組件，相對造成任務失效或儀器降落系統故障等飛安事件機率大幅提昇，故環境控 制系統的致冷功能在提供飛機電子裝備艙降溫所需的冷空氣，在戰機航電系統運作 維持上可謂至要關鍵。 本研究是以本國某戰鬥機環境控制系統致冷效能之組件為例，首先採用德菲法 (Delphi Method)製作第一次專家調查問卷，蒐集影響效能之重要組件，再製作發出 李克特五點量表專家問卷進行評分，再根據專家一致性指標評選出四項影響重要組 件壽命之關鍵檢測數據，分別為渦輪已裝機飛時(操作時間)、發動機高階引流空氣 壓力、渦輪空氣壓縮比及渦輪出口溫度，再載入2009年至2013年相關的維修資料數 據為樣本，載入倒傳遞類神經網路軟體Alyuda NeuroIntelligence來測試輸入與輸出 之間的關係以建置預測模式，測試所得到的設定參數為隱藏層神經元數目:6、學習 速:0.1、學習循環次數:10,000，做為BPN預測的條件，藉以預測重要組件剩餘的使 用壽命。 研究結果顯示，經由倒傳遞類神經網路軟體學習訓練後，關聯性(Correlation)與 模式配適度(R-squared)分別達到0.984129及0.963150，預測準確度高達93%，可見倒 傳遞類神經網路確實能有效作為飛機組件壽命預測的方法。 本研究利用倒傳遞類神經網路之預測能力，針對飛機環境控制系統關鍵組件之 故障時距建構最適預測模型，並擬訂一套戰機環境控制系統之維修策略，期能以最 少資源成本之投入，獲取最大效益，以維持飛機妥善，達成各項戰、演訓任務，發 揮最大戰力。

Today the demand for various types of fighter combat missions in strengthening aviation aircraft radar and other electronic components functions, and the relative failure of the task or instrument landing system failure probability significantly enhance flight safety incident, the cooling function of the environmental control system in providing aircraft cabin cooling electronic equipment required for cold air in the aircraft avionics systems can be described to be critical to maintain operations. This study use the cooling component of the environmental control system for one of the R.O.C air force fighter as an example. Firstly, this study selects the Delphi Method to collect experts’ key factors to collect critical component. Secondly, this study uses Likert 5-Piont Scale to request experts to evaluate importance and give a score on the key factors. According to experts’ agreed indicators, four key factors have chosen including TSC, Engine High Stage Bleed Air, Turbine compressed discharge air ratio & Turbine Discharge Temperature. According to the historical maintenance reports from 2009 to 2013, put the data of the four key factors into the BPN (Back-Propagation Network) to train the relationship between input and output to build the prediction mode via Alyuda NeuroIntelligence software. For best prediction mode used to predict the remaining life of critical components, it resulted to 1 hidden layer with 6 neurons, learning rate at 0.1 and iteration at 10,000. The research result indicates that correlation and R-squared model could reach 0.969 and 0.932 respectively after BPN training. Certainly, BPN serves as an effective method of predicting the aircraft component life span. This study uses the BPN prediction capability, the time to construct a critical component for the failure of the aircraft environmental control system for optimum predictive model, thereby enabling the aircraft component to develop a failure reaction strategy. Hopefully, able to ensure maximum output and minimum input, in order to maintain proper aircraft to perfect the war exercises tasks to maximize combat power.