Research on Mechanical Fault Diagnosis Method Based on Matrix Product State

In mechanical fault diagnosis, it is difficult for traditional neural networks to process high-level data, and many network parameters consume a lot of computing resources. Therefore, this paper proposes a tensor network fault diagnosis method based on matrix product state. By inputting high-order tensor fault data into the matrix product state fault diagnosis model, the high-order tensor is represented as multiple low-order tensors, thus simplifying the data structure and reducing the parameter number. In order to verify the effectiveness of the method, it is applied to the fault diagnosis of gears and compared with the traditional convolutional neural network fault diagnosis model. Moreover, the effect of bond dimension on the accuracy of the model was assessed. The experimental results show that the bond dimension of the proposed model affects the model accuracy, demonstrating a higher accuracy when the bond dimension is 16 in comparison to that of the model with a bond dimension of 8. While reducing the data complexity, the model can also identify different fault types with an accuracy of about 90%, which outperforms the traditional convolutional neural network fault diagnosis model.