SNR-adaptive bearing fault diagnosis using NRBO-optimized SVMD and kernel extreme learning machine

Abstract

To overcome the problems such as the empirical parameters tuning, suboptimal feature selection, and poor classifier generalization in the existing bearing fault diagnosis methods, an NRBO - SVMD - KELM method is proposed in this paper. Newton-Raphson-based Optimizer (NRBO) adaptively optimizes the penalty factor of successive variational mode decomposition (SVMD). Multi-domain features from reconstructed signals are selected via permutation importance, and NRBO further optimizes the kernel parameters and regularization coefficient of kernel extreme learning machine (KELM) for classification. On Case Western Reserve University (CWRU) and Jiangnan University (JNU) datasets, the method achieves 100% and 99.68% accuracy, demonstrating high accuracy and strong generalization.