Deep Learning-Based Identification of Shaft Imbalance Faults in Rotating Machinery Using the NARX Model

Abstract A robust, reliable, and online fault diagnosis is crucial for the efficient operation of transmission systems. Rotating shafts are critical components of these systems and faults like imbalance and misalignment can compromise their structural integrity, ultimately affecting the system’s overall performance. In the context of fault diagnosis, vibration-based techniques combined with machine learning methods typically rely on feature extraction from acquired vibration data. However, challenges arising from the low impact of such faults on the extracted features, along with a gap in the literature gap regarding the application of recurrent neural networks for fault identification on rotating shafts, have motivated this study. This study proposes the Nonlinear Auto-Regressive with exogenous Inputs model as a time-series estimation tool for diagnosing imbalance faults in a transmission shaft. The model leverages the estimation error, resulting from significant variations in signals acquired before and after the occurrence of imbalance, to evaluate the structural integrity of the operating structure. The efficiency of the model is validated using experimental data obtained under various imbalance scenarios. The results demonstrate that the proposed method effectively detects and quantifies imbalance, presenting a promising tool for improving existing condition monitoring techniques for transmission shafts.