基于特征子空间的系统故障检测与诊断

Fault detection and diagnosis based on feature subspace

针对KPCA不适合于大样本的不足,提出一种基于特征子空间映射的主元分析(FSP-KPCA)及其故障检测方法。利用核函数技巧,从样本数据中选取基向量集,基向量集在特征空间F中扩展生成特征子空间,样本数据在特征子空间投影,然后,对映射数据进行主元分析,从而形成FSP_PCA及其故障检测方法。当故障产生时,计算该故障与已知故障的特征子空间的夹角(核主角),以此识别故障类型。将提出的方法应用于Tennessee Eastman(TE)过程故障监测,证明本算法的有效性。

For the problems that KPCA is not suitable for large sample,a new PCA method based on feature subspace projection(FSP_PCA) was proposed.The kernel trick is used to select from the sample data set to form a base vector set in a feature space F.Thus the selected vectors define a subspace in F.The data are projected onto this subspace where principal component analysis is applied to the projected data.Then,a new FSP_PCA and its application to fault detection are presented.When fault occurs,the angle(kernel principle angle) between the new fault subspaces and known fault subspace is calculated.Fault can be identified based on the kernel principle angle.The proposed method is used to monitor the Tennessee Eastman(TE) process to demonstrate the effectiveness and efficiency.