Noise-Robust Bearing Fault Diagnosis via Enhanced EFD and QPSOL-Optimized SVM
This paper presents an optimization‑driven framework for noise‑robust bearing fault diagnostics aimed at enhancing the reliability and design performance of rotating machinery systems. The proposed approach integrates an enhanced empirical ensemble Fourier decomposition (EEFD) with a support vector machine (SVM) whose hyperparameters are optimized using a quadratic interpolation particle swarm optimization with local search (QPSOL) algorithm. To address signal degradation under harsh industrial environments, EEFD is employed to decompose vibration signals and extract high‑quality intrinsic components. A compact yet discriminative multi‑domain feature set, including Root Mean Square (RMS), Kurtosis, and Hjorth Mobility (HM), is constructed to characterize the dynamic behaviour of the system. The QPSOL algorithm is then utilized to optimize the SVM parameters, improving convergence accuracy and classification robustness. Experimental validation under a 10 dB signal‑to‑noise ratio demonstrates that the proposed method achieves superior diagnostic accuracy and convergence performance compared with conventional optimization techniques. Beyond fault classification, the developed framework provides a basis for simulation‑driven design optimization and reliability‑oriented decision‑making in mechanical systems, enabling more effective predictive maintenance strategies and lifecycle performance improvement.