Abstract: In order to improve the performance of gas path fault diagnosis for turbo shaft engine， a ReliefF-LMBP based method was proposed to fault feature extraction and pattern recognition. Firstly， the available engine sensor measurements were analyzed and assigned feature weights， and the fault feature subsets were ordered and determined after iterative selection by ReliefF algorithm. The effective feature measured parameters were gathered by similar samples， and the rest parameters fell in the discrete heterogeneous sample subsets. Afterwards， the LMBP Neural Network algorithm was employed to build up the relationship between the fault modes and features of reduced measurements. The tests of gas path fault diagnosis are carried out on a turbo shaft engine， and results show the capability of feature extraction and superiority of fault pattern recognition.

Key words: Turbo shaft engine;Gas path fault diagnosis;Feature extraction;Neural network;ReliefF analysis

摘要： 针对涡轴发动机气路故障模式识别精度不高的问题，提出了一种基于ReliefF-LMBP故障特征提取的发动机故障模式识别方法。应用ReliefF算法对发动机传感器参数赋予权值，对传感器参数特征权重值进行迭代更新和排序，聚集好的特征样本，离散异类样本。根据筛选出的特征子集，利用LMBP神经网络算法进行发动机故障模式识别。以涡轴发动机为对象进行气路故障诊断验证，结果表明所提方法能提取特征传感器参数并实现有效的故障模式识别。

关键词: 涡轴发动机;气路故障诊断;特征提取;神经网络;ReliefF分析