基于特征增维和近邻成分分析的民航发动机故障分类方法

Aeroengine Fault Classification Method Based on Feature Expand and Neighbourhood Components Analysis

为提高航空发动机故障诊断准确度,提出了一种从快速存取记录器(QAR)数据中提取最合适故障特征的方法。对原始QAR数据进行缺失值填补和巡航点提取操作,选择部分发动机性能参数差值作为初始特征值;再采用特征增维方法挖掘隐藏特征信息,进而采用近邻成分分析算法进行特征筛选优化,将所提方法与朴素贝叶斯等4种分类算法相结合,对某航空公司CFM56-7B发动机的QAR数据进行试验验证。结果表明:从QAR数据中提取最合适故障特征的方法能有效地提高发动机故障分类算法的准确率,且适用于不同的诊断算法,准确率优于80%。

In order to improve the accuracy of aeroengine fault diagnosis,a method for extracting the most suitable fault features from QAR data was proposed.Firstly,the missing value filling and cruising point extraction were performed on the original QAR data,and some engine performance parameter differences were selected as the initial characteristic values.Then the method of feature expand was used to excavate the hidden feature information and furthermore,feature selection and optimization were carried out by using the nearest neighbor component analysis algorithm(NCA).Finally,the method proposed in this paper was combined with four classification algorithms such as Naive Bayes to carry out experimental verification on the CFM56-7B engines’QAR data of an airline.The results show that the optimal features can effectively improve the accuracy of engine fault diagnosis algorithm and are applicable to different diagnosis algorithms,accuracy of diagnosis exceeds 80%.