Support Matrix Machine (SMM), as a novel classification method using the matrix as the input, has been widely used in the fault diagnosis by fully utilizing the structured information between rows and columns of the input matrix. However, For the diagnosis problem of multiple tasks, SMM does not consider the direct correlation of multiple tasks, thus unable to achieve accurate diagnosis. Therefore, a Multi-task Support Matrix Machine (MTSMM) is proposed in this paper, in which ascending terms are defined to make full use of the correlation between the rows and columns of matrices while dealing with complex matrix samples brought by multi-task learning. Meanwhile, and the multi-task learning theory is introduced to construct multitask classification hyperplanes, so as to achieve the sharing of matrix feature data for multiple tasks and realize the multitask classification performance. Finally, the proposed method is applied to mechanical fault diagnosis, and the results show that the proposed MTSMM has good multi-task classification performance.

Multi-task learning; fault diagnosis; support matrix machine; multi-task support matrix machine