摘要
提出基于核方法(KM)和尾气信息(ED)的内燃机故障诊断法。该法先通过非线性映射将尾气样本集从输入空间映射到高维特征空间,而后在高维特征空间中进行Fisher判别分析(FDA),使样本集在投影空间中按各自所属类得到较好分离;计算正常工况下的样本在特征空间中与其均值间的距离,利用核密度估计拟合其分布,并计算出99%置信限用于监控新采集样本是否为异常样本。若异常则将其转换到投影空间并判断其具体故障类型。诊断试验证明了该法的有效性。
ICE fault diagnosis method based on kernel method and emission data was proposed.The emission sample was first mapped on high-dimension feature space from input space with non-linear map and then the Fisher discriminant analysis(FDA) was carried out to separate the samples according to their respective class in the projection space.The distances between normal condition samples and their mean value in feature space were calculated,their distribution was fitted with the kernel density estimation(KDE),and the 99% confidence limit was calculated to monitor whether the new samples were abnormal or not.If the samples were abnormal,they would be mapped on the projection space and the specific fault type was judged.The diagnosis experiment shows that the method is effective.
出处
《车用发动机》
北大核心
2012年第6期80-84,89,共6页
Vehicle Engine
关键词
内燃机
核方法Fisher判别分析
尾气信息
核密度估计
故障诊断
internal combustion engine
kernel Fisher discriminant analysis
emission data
kernel density estimation
fault diagnosis
