基于遗传规划和线性鉴别分析的故障特征提取模型及其应用

1引言航空发动机故障诊断需要从原始监测数据中获取合理的特征参数,如果依靠专家经验手工从众多的监测参数中选取特征参数,将是十分烦琐和低效的。因此,构造优质的特征参数是提高故障诊断效率及其准确性的关键。许多机器学习方法被应用到特征参数的自动选取中,例如神经网络1],

Feature extraction based on genetic programming and linear discriminant analysis for fault diagnosis and its application

A new feature extraction model based on genetic programming and linear discriminant analysis in fault diagnosis is proposed. In this model, genetic programming is first used to construct compound features from original feature set. Then linear discriminant analysis is employed to get rid of the correlation among features and reduce the dimension of features. Thus the more effective and smaller subset of features for classification can be gained. Fault recognition experiments in aeroengine lubricating oil system are carried out to test the performance of this model. Practical results show that the extracted features based on genetic programming and linear diseriminant analysis have better recognition ability and they are robust for various classifiers.