Detection of vibration faults in induction motors using automatic features selection

Abstract

The use of machine learning techniques in the diagnosis of induction motors (IM) is becoming increasingly common in modern industry. Employing the right indicators that reflect the behavior of IMs directly impacts the accuracy and effectiveness of diagnostic systems, enabling not only a reduction in maintenance costs but also an improvement in operational efficiency and safety in industrial operations. However, identifying these indicators is complex and often leads to the choice of more robust algorithms, which in turn complicates the implementation of models in real-world environments. Therefore, this work focuses on developing a methodology for fault detection through vibration in IMs using random forest and logistic regression for automatic feature selection, and support vector machines, K-nearest neighbors, and logistic regression as classification models. The results demonstrate the importance of identifying these features and how their synergy improves accuracy and effectiveness in fault classification.