Empirical-based DA and ANN to diagnose misalignment in rotor-bearing system

Timely and accurate misalignment fault detection in rotor bearing systems is crucial for reliable and efficient operation, especially in Industry 4.0 based condition monitoring. The method described in this paper uses artificial neural networks (ANN) and empirical modelling to identify misalignment defects in rotor bearing systems. To capture the dynamic behaviour of the rotor bearing system under different operating conditions and fault scenarios, an empirical model based on Dimensional Analysis (DA) is developed. This model is trained using vibration data obtained from a well-aligned system. By comparing the predicted response with the actual vibration response, an accuracy of 8% to 10% is achieved. However, accurately quantifying the severity of misalignment poses challenges due to system nonlinearities. To address this issue, an ANN is employed to learn the mapping between vibration features and misalignment levels. To train the ANN, a comprehensive dataset is created through experiments conducted on a test rig. Artificial misalignment is introduced to the rotor, and vibration signals are analysed under various conditions of misalignment, rotor speed, and static load. The trained ANN is validated using experimental data, demonstrating an impressive accuracy of 99.46%. This integrated approach holds significant potential for predictive maintenance, leading to improved operational efficiency.