Supervised Learning Approaches for Vibration Analysis and Anomaly Detection in Rotating Machines

Safety and reliability are crucial for modern systems and technologies. This paper presents an approach for real-time anomaly detection in rotating machines using vibration analysis and machine learning algorithms for predictive maintenance. Rotating machines such as motors, engines, gearboxes and turbines are widely used systems and faults in these systems can result in complete shutdowns. Preventing unexpected downtime while ensuring equipment reliability and safety is essential. In this context, anomaly detection plays a vital role in monitoring the healthy operation of rotating machines. A test-bed was designed to generate datasets for both balanced (healthy) and unbalanced (faulty) motor conditions. The experimental setup consists of a three-phase induction motor, a signal acquisition system (Raspberry Pi), a vibration sensor (MPU6050) and a coupling system capable of adjusting motor balance. Supervised machine learning algorithms, Support Vector Machine (SVM) and Random Forest (RF) were applied to classify and detect anomalies in the collected datasets. A grid search algorithm was employed for hyperparameter tuning of the SVM and RF models. The proposed technique achieved a maximum accuracy of approximately 98% using SVM and 100% using the RF algorithm leveraging nine optimally selected temporal features.