Experimental Investigation Using Robust Deep VMD-ICA and 1D-CNN for Condition Monitoring of Roller Element Bearing

Abstract

A rotor-bearing system experiences numerous vibrations during the operation that frequently degrade performance and endanger operational safety. Roller-bearing failure has significant consequences, leading to downtime or a complete outage of rotating machinery. It is crucial to detect and diagnose incipient bearing defects promptly to ensure optimal operation of the machinery and minimize potential disruptions to the process. Deep independent component analysis is a necessity used in modern condition monitoring to detect bearing failures prior to their occurrence. To address this issue, the feasibility of utilizing the deep independent component analysis (ICA) method based on the variational modal decomposition (VMD) with a one-dimensional convolutional neural network (1D-CNN) to diagnose the incipient bearing defect. Fast Fourier techniques are utilized to extract the vibration signatures of artificially damaged bearings on a newly built test bed. VMD addresses to minimize data noise by allowing data to decompose into various sub-datasets for the extraction of incipient defect features. With weak defect characteristic signal and noise interference, the deep VMD-ICA model and 1D-CNN simplicity improved the accuracy of diagnosis corresponding to the experimental results. Moreover, deep VMD-ICA with 1D-CNN has demonstrated strong performance compared to experimental results and is helpful in monitoring the condition of industrial machinery. The results reveal that this fault diagnosis approach is reliable, with a diagnostic accuracy of 98.93% for bearing faults.