Bearing Fault Diagnosis with a Hybrid CWT-ResNet-LSTM Model

Bearings’ reliability is very important for the seamless operation of rotating machinery, as their failures can cause much downtime and expenses. In applications of bearing fault diagnosis (BFD), where deep learning (DL) has performed well, problems remain in making sure that the model remains robust under different operating conditions. This study introduces a novel hybrid model which integrates both Continuous Wavelet Transform (CWT) and Residual Networks (ResNet) together with Long Short-Term Memory (LSTM) networks to improve fault detection. The process begins with CWT that transforms time signals into 2-D time-frequency (TF) images, which are used by ResNets to extract spatial features. Temporal dynamics of these features are analyzed using an LSTM network in order to enhance predictive accuracy and adaptability to different operational scenarios. Tested against the famous CWRU dataset, our hybrid model demonstrates superior diagnostic performance, outperforming existing methods in accuracy.