A novel symplectic relevance matrix machine method for intelligent fault diagnosis of roller bearing

Abstract

In the fault classification of roller bearing, it is often encountered that the input sample is composed of feature matrix with rich structural information. As a new classifier, support matrix machine (SMM) makes full use of the structural information of input samples to establish prediction model. However, SMM lacks necessary probability information, and its sparsity and robustness are not clear. To resolve the above issues, a new matrix classifier - symplectic relevance matrix machine (SRMM) is proposed based on the probability framework and symplectic geometry theory. In SRMM, the model takes the original signal matrix as the input, and constructs the model elements with rich structural information. Then, symplectic geometry matrix is obtained by symplectic geometry similarity transformation, which makes SRMM robust. Meanwhile, the model elements are constructed by symplectic geometry, which can solve the difficulty of constructing recursive kernel function, reduce the complexity of the model, avoid the time consumption in the process of model parameter optimization. The experimental results of three roller bearing datasets show that SRMM has good classification performance in roller bearing fault diagnosis by comparing recognition rate, time, kappa, accuracy, recall rate and F1 and statistical test.