Intelligent Fault Diagnosis of Roller Bearings using Bond Graph -Transformer-based Model

Abstract

This study introduces an advanced hybrid framework for intelligent fault diagnosis of bearings by merging Bond Graph-based physical modelling with a Transformer-based deep learning technique. Traditional data-driven methods rely heavily on large, labeled datasets. This limitation is addressed in this study by employing a synthetic vibration signal generated from bond graph-derived differential equations representing various fault conditions, such as healthy, inner race fault, outer race fault, and ball fault. To extract discriminative features from these synthetic signals, a modified Transformer model is created that has been tailored for temporal signal processing. The model was trained and assessed through numerous trials, demonstrating approximately 92.2 % accuracy in both the training and validation phases.