Application of physics-informed machine learning models for diagnosis of inter-turn short circuits in an induction motor

Abstract

Aim. To develop a physics-informed machine learning model capable of diagnosing inter-turn short circuits in the windings of squirrel-cage induction motors with high accuracy. Materials and Methods. This study utilized a neural network training method using stator current data and FFT analysis, along with programming in Python and C++. A laboratory test bench was created to simulate the signs of inter-turn short circuits in the motor stator winding. Results. A machine learning model capable of detecting inter-turn short circuits in the stator windings of induction motors with high accuracy was developed. The volume of labeled data required for model training was determined. A laboratory test bench was created to simulate inter-turn short circuits in the motor winding. Conclusion. The obtained results allow us to address a number of issues associated with the effective use of neural networks for diagnosing induction motors, including the need for large volumes of data, low interpretability, poor generalization ability, and high sensitivity to the quality of the recorded data.