Intelligent fault detection in rotor-bearing system using optimization-based learning algorithms

• Published in: Journal of the Brazilian Society of Mechanical Sciences and Engineering Vol. 47; no. 6; p. 305
• Main Authors: Suryawanshi, Ganesh L., Desavale, Ramchandra G.
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.06.2025; Springer Nature B.V
• Subjects: Algorithms; Artificial neural networks; Axial loads; Ball bearings; Defects; Dimensional analysis; Dynamic characteristics; Engineering; Fast Fourier transformations; Fault detection; Fault diagnosis; Grooves; Journal bearings; Machine learning; Machining; Mechanical Engineering; Misalignment; Neural networks; Optimization; Parameter identification; Prediction models; Predictive maintenance; Regularization; Rotor speed; Rotor-bearing systems; Signal processing; Sugar industry; Technical Paper; Vibration analysis; Vibration response; Dynamic characteristics; ANN; Bearing element defects; Bearing system; Misalignment; Predictive maintenance
• ISSN: 1678-5878; 1806-3691
• DOI: 10.1007/s40430-025-05617-7

Deep groove ball bearings can handle radial as well as axial loads effectively, making them suitable for heavy duty applications like rotating systems in sugar factories. These systems often succumbed primarily to bearing element defects and misalignment. Hence, proper maintenance strategy must be adopted to avoid sudden failure in the system. This study presents investigation on dynamic characteristics of rotor-bearing system under rotor misalignment and outer race bearing defect at variable rotor speed and load using experimental and artificial neural network (ANN) approach. Vibration responses of rotor-bearing system comprising artificially damaged bearings, created using electrode discharge machining and rotor misalignment are collected using fast Fourier transform technique on developed test rig. This experimental data are then utilized to train and test the ANN algorithm. Three different approaches of weight optimisation, namely Levenberg–Marquardt (LM), Bayesian regularization (BR), and scaled conjugate gradient (SCG), have been compared along with optimizing hidden layer nodes (4, 5, and 6) to identify the best predictive model. In the presence of combined bearing and misalignment defects, LM algorithm provides the most accurate prediction up to 96% of accuracy for vibration responses. Moreover, paired t-test proved reliability and robustness of LM algorithm with p-value greater than 0.05. The findings support the use of artificial neural networks to improve diagnostic and predictive maintenance strategies in high-speed rotor-bearing systems.