Optimization of VMD using kernel-based mutual information for the extraction of weak features to detect bearing defects

In this work, genetic algorithm (GA), kernel based mutual information (KEMI) fitness function and variational mode decomposition (VMD) based strategy is proposed for the purpose of easy identification of single and multiple defects of bearing, both at fixed and varying speed. To make the multiple defects identification possible at varying speed, Fourier synchro squeezed transform (FSST) based processing is proposed to extract instantaneous frequency (IF) from the vibration signal itself. Extracted IF is used for converting time domain signal into angular domain. For finding optimizing parameters of VMD, KEMI based fitness function is developed. Thereafter, optimum parameters of VMD are found by GA using proposed fitness function. Then, optimized VMD is carried out. After, applying optimized VMD, KEMI of modes is calculated. Finally, envelope of mode having minimal KEMI is computed to find out the defect by comparing with defect order. It has been proved that selection of VMD parameters using kurtosis-based criteria can cause loss of defect features while decomposition, as a result defect order could not be identified in the envelope spectrum. The proposed method founds to outperform existing methods while extracting weak defect features.

Automated summary

This work proposes a method using genetic algorithm, kernel based mutual information fitness function, and variational mode decomposition strategy for easy identification of single and multiple bearing defects. By optimizing VMD and calculating KEMI, defects can be located efficiently.