Fault diagnosis of multi-channel data by the CNN with the multilinear principal component analysis

Abstract

The multi-channel sensor data are widely collected during a manufacturing process to detect the variation of product quality. Multi-channel data can provide comprehensive information for the fault diagnosis, while the cross-correlation and redundant information in the data make it difficult to analyze using common methods. In this paper, the tensor structure and characteristics of a multi-channel dataset are investigated. After that, a novel fault diagnosis method is proposed by introducing the multilinear subspace learning algorithm into deep learning technologies. The dimension of the multi-channel data is reduced using the Multilinear Principal Component Analysis that does not destroy the tensor structure. The CNN is then used to extract features and build a classification model for fault diagnosis. The proposed method is compared with existing methods in the case study about a practical multi-operation forging process. Results show that the proposed fault diagnosis method for multi-channel data has superior performance and lower computational cost than existing approaches.