Bearing Fault Diagnosis Using Multichannel Broad Learning System Based on Positive–Negative Weighted Voting Mechanism

Abstract

Bearing fault diagnosis is essential for improving the efficiency of industrial operations. The inherently multi-modal nature of bearing vibration signals presents significant challenges in bearing fault diagnosis. To address this issue, we propose a novel Multi-Channel Broad Learning System (MCBLS). MCBLS comprises multiple Broad Learning Systems (BLSs) and incorporates aPositive-Negative Weighted Voting mechanism (PNWV). MCBLS first employs empirical mode decomposition to break down the vibration signal into multiple intrinsic mode functions (IMFs) and subsequently transforms these IMFs into concise spectrogram distributions (CSDs). Following this, MCBLS employs multiple BLSs, with each one being trained based on a CSD associated with an IMF. The final diagnosis result is obtained by aggregating all BLSs’ outcomes using the PNWV. To validate MCBLS, we employ three bearing fault datasets collected from Guangdong University of Petrochemical Technology, China (GDUPT), Case Western Reserve University (CWRU) and Southeast University, China (SEU). The accuracies of MCBLS in three datasets can reach up to 98.11%, 100%, and 94.33% respectively, demonstrating its superior classification performance and applicable generalization for bearing fault diagnosis.