A Novel Multi-Kernel Extreme Learning Machine Model for Accurate Diagnosis of Motor Bearing Faults

Abstract

Accurate fault diagnosis of motor bearings is essential for ensuring the reliability and safety of rotating machinery in various industrial applications. Traditional machine learning models often struggle to balance generalization and precision, particularly when processing complex, non-linear features present in motor vibration signals. To address this challenge, this research introduces a novel Multi-Kernel Extreme Learning Machine (MKELM) model tailored for precise diagnosis of motor bearing faults. The proposed model integrates multiple kernel functions—specifically Gaussian, polynomial, and wavelet kernels—to exploit diverse data characteristics and enhance classification performance. Experimental evaluations using the Case Western Reserve University (CWRU) bearing dataset demonstrate the effectiveness of MKELM in achieving superior accuracy, robustness, and generalization compared to conventional single-kernel ELM and other baseline classifiers such as Support Vector Machines and Random Forests. Through rigorous analysis, the study validates the hypothesis that multi-kernel learning, when fused with ELM's fast learning capabilities, provides a powerful framework for bearing fault detection in real-world scenarios.