A Generalized Convolutional Neural Network Model Trained on Simulated Data for Fault Diagnosis in a Wide Range of Bearing Designs

Rolling element bearings (REBs) are critical components in rotating machinery and a leading cause of machine failures. Traditional fault detection methods rely on signal processing, but advances in machine learning (ML) and deep learning (DL) have dramatically improved diagnostic accuracy. However, existing DL models struggle with data availability, generalization, and domain adaptation, making industrial applications challenging. This study proposes a convolutional neural network (CNN) model trained on numerically simulated vibration data generated for a wide range of bearing designs. A novel hybrid signal processing method is employed to enhance feature extraction and reduce domain shifts between simulated and real-world data. The optimized CNN model, trained on simulated data, is tested using experimental and real-world vibration signals from laboratory bearings and jet engine components. The results show high classification accuracy using data from the Case Western Reserve University experimental dataset and successful fault detection in real-world Safran jet engine ground tests. The findings demonstrate the effectiveness of the developed CNN-based model for bearing fault classification, tackling training data scarcity and generalizability challenges while contributing to the development of intelligent fault diagnosis models for several industrial applications.