Corrosion Damage Detection in Induction Motor Bearings Using Ensemble Learning Algorithms and Data Fusion

Early fault detection in induction motors is essential to ensure operational continuity, improve energy efficiency, and reduce costs associated with unexpected repairs or downtime. These motors are critical in modern industry due to their reliability and versatility, but their natural wear, especially in components like bearings, can account for over 40% of recorded failures. This work proposes a comprehensive approach to analyzing magnetic flux, current, and vibration signals in motors subj ect to different levels of corrosion. Both statistical and non-statistical parameters are used to characterize the signals, combining advanced techniques such as Linear Discriminant Analysis for dimensionality reduction and the classification algorithm based on Gradient Boosting with the Bagging ensemble learning method to enhance model accuracy. Additionally, a genetic algorithm is employed to optimize parameter selection and achieve the best damage-level classification model, achieving an accuracy of 99%.