建立直升機主旋翼葉片維修策略

Developing A Maintenance Strategy For The Main Rotor Blade Of Helicopters

因應不同任務需求，直升機維修需求量不斷增加，但零附件故障影響比例甚高，若能對直升機的關鍵零附件建置故障預測系統，並於故障前即執行檢修，將有助於降低因非預期性故障所導致的航班延誤，作為直升機零附件維修管理上的一項參考。本研究以A型直升機的主旋翼葉片為例，首先採用德菲法(Delphi Method)製作第一次專家調查問卷，蒐集關鍵因素，再運用李克特五點量表（Likert）評選出重要關鍵因素，再選定2012年至2014年主旋翼葉片的維修資料數據為樣本，載入倒傳遞類神經網路軟體（Neuro Intelligence）建置預測模式，以預測主旋翼葉片下一次故障。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0.999及0.998，預測準確度高達97%，可有效作為直升機零附件壽命預測的方法。

In response to operation readiness, disaster prevention, disaster relief, transportation and a variety of different mission requirements, the demand for helicopter maintenance is keeping increased, but the part failures highly affect flights and missions. If we can create prediction systems for the helicopters’ critical parts failures, so as to perform maintenance before the failure occurs, and that system will help to reduce the rate of flight delays caused by unpredictable failures. The creation may serve as a reference for our Armed Forces to manage its maintenance and spare parts. This study takes the main rotor blade used on the helicopters as an example. At the beginning, it uses Delphi Method to prepare the first expert questionnaire to collect the key factors affecting the life of the main rotor blades. After that, the study uses Likert five-point scale to score the expert questionnaires, and, in accordance with experts’ consistency index, Further, the data of main rotor blade maintenance performing during year 2012 to 2014 is used as samples to be loaded into back-propagation neural network software (Neuro Intelligence) to test the relationship between input and output to build predictive models. The number set is used as BPN prediction criteria in order to predict the usage life of a main rotor blade after installation. After learning and training by the inverted neural network software, the relevance (Correlation) and mode of fit (R-squared) reaches 0.999386 and 0.998655, respectively, and the accuracy of prediction is as high as 97%, and that proves back-propagation neural network is indeed an effective method to predict the life of helicopter parts.