Multiresolution Bidirectional Long Short Term Memory Network-Based Motor Bearing Fault Diagnosis Method

In the realm of long-term production and operation processes, the motor system is often susceptible to performance degradation, consequently leading to frequent operational failures that substantially threaten overall system safety. The conventional approach to motor bearing fault diagnosis, predicated on deep learning techniques, grapples with limitations arising from noise interference during the acquisition of bearing vibration signals and incomplete analysis of said signals, thus leading to restricted diagnostic precision. In response to these limitations, this paper introduces a novel motor-bearing fault diagnosis methodology grounded in Multiresolution Bi-Directional Long and Short-Term Memory Networks (MR-BILSTM). Initially, we construct a multiple optimization variational modal decomposition model tailored to enhance the filtering of motor-bearing vibration signals. Subsequently, a weighted kurtosis factor is introduced to meticulously identify the decomposition-filtered vibration signals characterized by diverse resolutions that exhibit the highest sensitivity to fault signals. Following this selection process, a bi-directional long and short-term memory network model is enlisted to comprehensively scrutinize the critical vibrational signal features extracted from various resolutions, ultimately culminating in diagnosing motor bearing faults. Ultimately, the proposed method's efficacy is rigorously validated using the Case Western Reserve bearing fault diagnosis dataset.