High-Precision Fault Identification in Motor Bearings via Enhanced Multi-Kernel ELM Framework

Abstract

Motor bearings are among the most critical components in rotating machinery, and their failure often leads to costly downtime and maintenance. Accurate and early detection of bearing faults can significantly enhance the reliability and longevity of industrial systems. This study introduces a novel Enhanced Multi-Kernel Extreme Learning Machine (EMK-ELM) framework for high-precision fault identification in motor bearings. The model leverages a dynamic integration of Gaussian and polynomial kernels to improve classification performance over traditional single-kernel methods. By capturing both linear and nonlinear relationships in vibration signal data, the EMK-ELM framework demonstrates superior generalization and faster learning capabilities. The experimental evaluation is carried out using benchmark datasets such as the Case Western Reserve University (CWRU) bearing dataset, highlighting the framework’s robustness across different fault types and severities. Performance is compared with standard machine learning classifiers, including Support Vector Machines and traditional ELM, showing that EMK-ELM offers improved diagnostic accuracy, training efficiency, and fault sensitivity. This paper provides a detailed discussion of model architecture, feature engineering, experimental methodology, and statistical evaluation, supporting its claim as a high-precision, real-time-capable solution for motor bearing fault detection.