# Fault Diagnosis of Inter-turn Short Circuit in Permanent Magnet Synchronous Motors with Current Signal Imaging and Unsupervised Learning

**Authors:** W. Jung, S. H. Yun, Y. S. Lim, S. Cheong, J. Bae, Y. H. Park  
**arXiv:** [2206.07651](https://arxiv.org/abs/2206.07651)  
**Date:** 2022-06-09  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2206.07651v1/

## Key points

- The paper presents a novel approach for diagnosing inter-turn short circuit faults in PMSMs using electrical current signals.
- The proposed method involves machine-independent feature engineering that can be applied to PMSMs with different capacities and under varying environmental and operational conditions.
- The overall procedure consists of three key steps: transforming the time-series current signal into a two-dimensional image using recurrence plots (RP), extracting features from the images using convolutional neural networks, and calculating a health indicator based on Mahalanobis distance.
- The RP method is designed to maximize the features of inter-turn short faults while minimizing noise from systems with various capacities.
- Two case studies were conducted using an artificial fault-seeded testbed with two different motor capacities to demonstrate the validity of their proposed method.
- By calculating the feature using only the electrical current signal of the motor without considering parameters related to its capacity, they showed that their proposed feature can be applied to motors with different capacities while maintaining high performance.
- The proposed method is not only machine-independent but also robust against various environmental and operational conditions, making it suitable for real-world applications.

## Summary

This paper presents a novel approach for diagnosing inter-turn short circuit faults in permanent magnet synchronous motors (PMSMs) using electrical current signals. The proposed method involves machine-independent feature engineering that can be applied to PMSMs with different capacities and under varying environmental and operational conditions. To address these challenges, the authors developed a robust current signal imaging method and deep learning-based feature extraction technique. The overall procedure consists of three key steps: (1) transforming the time-series current signal into a two-dimensional image using recurrence plots (RP), (2) extracting features from the images using convolutional neural networks, and (3) calculating a health indicator based on Mahalanobis distance. The RP method is designed to maximize the features of inter-turn short faults while minimizing noise from systems with various capacities. To demonstrate the validity of their proposed method, the authors conducted two case studies using an artificial fault-seeded testbed with two different motor capacities. By calculating the feature using only the electrical current signal of the motor without considering parameters related to its capacity, they showed that their proposed feature can be applied to motors with different capacities while maintaining high performance. Overall, this study provides a promising approach for detecting inter-turn short circuit faults in PMSMs using electrical current signals. The proposed method is not only machine-independent but also robust against various environmental and operational conditions, making it suitable for real-world applications.