Sparse norm matrix machine and its application in roller bearing fault diagnosis

Abstract

Roller bearings are an important part of rotating machinery, and bearing failure may lead to serious casualties and economic losses. Thus, the implementation of fault diagnosis to ensure the smooth operation of bearings is an essential step to maintain the safe and stable operation of modern machinery and equipment. Recent data-driven intelligent fault diagnosis methods have become widely used. However, traditional machine learning methods are limited when using matrix data, which are forcibly vectorized as the input, resulting in their structure information becoming lost. Moreover, these methods are often disturbed by outliers. To solve those issues, this paper proposes a robust classifier based on a supervised tensor learning framework, a named sparse norm matrix machine (SNMM). SNMM constructs a pair of nonparallel hyperplanes, whose optimization problems are established by using L 1-norm distance and hinge loss. L 1-norm distance can reduce the sensitivity of the model to outliers and improve the robustness of the model. Furthermore, the generated dual quadratic programming problems of SNMM avoid the need to invert the matrix in the calculation process, so as to reduces the amount of calculation and to make SNMM more suitable for large-scale data. The experimental results on roller bearing fault datasets show that SNMM has the highest diagnosis accuracy and superior diagnosis ability.