Cyclic Spectral Coherence Map-Based Filter Design With Geometric Mean Optimization for Industrial Fault Diagnosis

This study introduces a denoising filter designed to enhance non-stationarities within signals for industrial fault diagnosis. The filter calculates kurtosis from each frequency bin of the spectral coherence map, optimized using the geometric mean optimizer (GMO) with Rényi entropy as a fitness function. The optimized kurtosis vectors are multiplied with the system impulse to extract impulsive components from the signal. Validation on simulated, vibration, and acoustic signals from a defective bearing in the mining industry demonstrates the filter's ability to extract impulsiveness, improve kurtosis, signal-to-noise ratio (SNR) and envelope spectrum-based indicator (ENVSI), and reduce interferences from other machinery and the environment. Compared to Kurtogram analysis, spectral kurtosis-based filter, wavelet filter and variational mode decomposition (VMD), the proposed method achieves superior noise reduction and defect signal enhancement. The designed filter is a valuable tool for condition monitoring and predictive maintenance, contributing to operational efficiency, preventing costly breakdowns, and maintaining a safe working environment.