FAULTS IDENTIFICATION WITH DEEP CNN FINE-TUNED RESNET50 MODEL FOR ROLLING BEARINGS

There is broad use of rotating machinery in various industrial practices. The diagnosis of rolling bearing defects in rotating machinery is a broad area in all engineering disciplines since these errors might result in production losses and downtime. Machine learning approaches have been successfully used in recent years to diagnose these faults, with transfer learning being a promising approach that can be used with limited data. Transfer learning refers to the technique of leveraging knowledge gained from one domain to improve prediction and learning in another. The proposed approach in this paper uses a pre-trained ResNet50-TL deep neural network to extract features from 2D RGB scalogram images, which were created from 1D vibration signals by using the continuous wavelet transform technique. These 2D scalogram images were then used to train a ResNet50-TL classifier model for fault classification and identification. The proposed method was tested on a real-world public dataset of rolling bearings created by Case Western Reserve University (CWRU), outperforming other cutting-edge methods in terms of accuracy. An accurate fault identification method using transfer learning with the proposed method has been shown to be highly effective, demonstrating its applications in rotating machinery and achieving 99.6% test accuracy.