Study on Fault Diagnosis for Bearing Based on Hierarchical Extreme Learning Machine

Abstract

Rolling bearings are widely used in mechanical systems but have a high damage rate. Its running state is related to the production safety and stable operation of various industries. Nowadays, scholars have applied so many signal processing methods such as differential entropy, energy entropy, and empirical mode decomposition methods in conjunction with various algorithms which likes particle swarms and neural networks to implement pattern classification in the process of the vibration signals of rolling bearings (Qin et al. in Mech Des Manuf 08:11–14, 2018 [1]). On this basis of it, this paper presents the variational mode decomposition–singular value decomposition (VMD-SVD) method based on the previous studies by other scholars with good verification effect that is developed and used to extract the characteristics of different IMF components under different operating conditions in order to establish the characteristic matrix. The latest and better effect of hierarchical extreme learning machine (H-ELM) is applied for training and verification. Besides, by comparing with the traditional ELM method, it verifies its superiority in rolling bearing fault diagnosis.