Early Rolling Bearing Fault Detection Using A Gini Index Guided Adaptive Morlet Wavelet Filter

Rolling element bearings play a fundamental role in critical applications such as aerospace, automotive, and energy systems. Reducing operation and maintenance costs have increased the urge for more efficient detection and diagnosis of bearing faults during service time. However, early fault signals are difficult to detect as they are obscured by noise and vibration signals from other sources. Envelope analysis is typically used for bearing fault detection although the success of this technique is conditioned by proper bandpass filtration. An adaptive optimum Morlet wavelet filter is used in this work to automatically prepare the signal for envelope extraction. Searching for a proper criterion for optimum Morlet wavelet filter selection is a recurrent issue in the literature. In this context, this manuscript uniquely introduces Gini index for this purpose. The proposed work applies particle swarm optimization (PSO) algorithm to adaptively optimize the parameters of the Morlet wavelet filter by maximizing the Gini index of the squared envelope of the filtered signal. Validation of the proposed method is accomplished through both simulated and real rolling bearing fault signals. A comparative study is made based on real rolling bearing fault signals. The results verified the supremacy of the proposed method in incipient fault detection. It is also proved that Gini index is advantageous over the improved correlated kurtosis, kurtosis, and spectral kurtosis when used as a criterion for an accurate Morlet wavelet filter selection.