A Novel Fault Diagnosis Method for Rolling Bearings Based on ELTSA and CGWO-WGKELM

Abstract

Fault diagnosis method for rolling bearings by using entropy-weighted distance-based local tangent space alignment and Weighted Gaussian Kernel extreme learning machine with Circle chaotic mapping-based grey wolf optimization (ELTSA-CGWGKELM) is presented in this paper. This study introduces two innovative methodologies for rolling bearing fault diagnosis. First, an entropy-weighted distance-based local tangent space alignment (ELTSA) technique is developed to address feature dimensionality reduction in rolling bearing data. This approach effectively resolves limitations associated with conventional Euclidean distance measurement while significantly enhancing critical data information preservation capabilities. Secondly, a weighted Gaussian kernel extreme learning machine optimized through circle chaotic mapping-enhanced grey wolf optimization (CGWGKELM) is proposed for fault classification. The Gaussian kernel implementation substantially improves nonlinear processing performance and robustness compared to traditional weighted ELM architectures. The circle chaotic mapping strategy integrated into the grey wolf optimization algorithm (CGWO) enables superior optimization of the weighted Gaussian kernel ELM training parameters, ensuring enhanced global search capability and convergence efficiency. The experimental results indicate the following fault diagnosis accuracy rates for rolling bearings: ELTSA-CGWGKELM achieves 99.5%, LTSA-WGKELM attains 95.5%, LTSA-ELM reaches 94%, and PCA-ELM attains 92%.It can be seen that ELTSA-CGWGKELM is the better fault diagnosis ability for rolling bearings than LTSA-WGKELM, LTSA-WGKELM, and PCA-ELM.