Real-Time Induction Motor Condition Monitoring Using Machine Learning Approaches

Abstract

Induction motors are widely used in the industrial field thanks to their reliability and robustness, so understanding their health state is essential to prevent costly failures. Condition monitoring techniques measure parameter deviations from the normal performance. When these deviations reach dangerous levels, a warning can be issued and a maintenance task scheduled. Although several types of sensors can be used, a combination of vibration, thermal, and electrical signals is preferred. Condition monitoring has also been approached using machine learning techniques. Nevertheless, the deployment of effective techniques in a real-time setup is still a challenge. A real-time end-to-end system is proposed, integrating the acquisition of the sensor signals, their processing, and the application of different machine learning techniques.

Condition monitoring aims to collect sensor signals continuously, detect deviations from the normal behavior of the system, and issue warnings. Considering that accidents, for instance, due to bearing failure, are rare, the monitoring techniques also need to be reliable; thus, performance in terms of sensitivity and specific are key. Commonly, information from different types of sensors is combined so that a drastic event in one sensor can be compensated by the others. The three most common sensors used for induction motor condition monitoring are vibration, thermal, and electrical. Vibration information has been widely used due to its high sensitivity to faults, such as bearing wear, misalignment, and imbalance. Electrical monitoring is also a promising approach since a fault in the motor will generate changes in the electrical parameters because the electrical signals are the input and output of the system. The thermal behavior of the induction motor is also an important aspect to monitor.