Fault Diagnosis Based On Fusion Of Multi-Channel Observations By Sensors With Independent Component Analysis And Mutual Information
Sx Qian,Sx Yang,Xj Gu
2005-01-01
Abstract:Independent Component Analysis (ICA), as expanding of Principal Component Analysis (PCA) to higher statistic than second order, has such features as signal processing blindly and redundancy reduction, is fit for de-noising and feature extraction in fault diagnosis. In this paper, a new method for fault diagnosis based on fusion of multi-channel observations by sensors with independent component analysis and mutual information (MI) was proposed, by which redundancy among observations by sensors (vibrations or acoustical signals) can be reduced, high dimensional fault features can be compressed, and, feature extraction can be made simpler and faster. By means of such characteristics as redundancy reduction of ICA and higher statistic than second order of MI, two-step fusion and compression of multi-channel observations by sensors were implemented. Firstly, multi-channel observations were inputted into an ICA processing unit, by which additive noise and interference were removed, and information embedded into observations was re-adjusted. Then, a procedure for feature extraction named MI based Wavelet was used for capturing higher statistical features than second order. Finally, by means of clustering analysis of the fused feature vectors, capacity of the new method in classification for different fault patterns was tested. Results of experiment showed that multi-channel observations by sensors were fused sufficiently, dimension of observations was reduced remarkably, and, at the same time, good performance in classifying different fault patterns was obtained. Thus, it is possible to construct an on-line and real-time fault classifier by the application of this new method to fault diagnosis of a complex machine.