A novel fault diagnosis model for rolling bearings in electric drive systems using multi-scale improved weighted permutation entropy and radial basis function neural network

Abstract

Listen

Translate article

The operational environment of rolling bearings in electric drive systems (EDSs) is highly complex, necessitating effective fault monitoring. However, due to the limited computing power in vehicles, conventional high-quality vibration sensors are rarely employed for monitoring. The weighting method of permutation entropy (PE) has been enhanced in this study to improve the stability of fault diagnosis. A novel method for extracting fault features named composite multi-scale improved weighted permutation entropy (CMIWPE) has been proposed. The study presents a novel fault diagnosis model that includes a radial basis function neural network (RBFNN), demonstrating its validated capability and stability in fault diagnosis using public datasets available from Case Western Reserve University (CWRU). A subsequent bearing fault experiment for EDSs is conducted using a laboratory test-bench to validate the identification features of three typical bearing faults. The utilisation of machine learning for fault prediction demonstrates remarkable accuracy and stability. The results unequivocally indicate that the enhanced weighting method exhibits superior stability. The composite multi-scale improved weighted permutation entropy-radial basis function neural network (CMIWPE-RBFNN) consistently maintains a stable diagnostic capability for subtle faults, exhibiting an approximate accuracy rate of 95%.