Unsupervised multi-target domain transfer learning for inverter open-circuit fault diagnosis in PMSM drives

View at publisher Summarise with VIRAT
Authors: Kailong Fang, Deqiang He, Baofu Qin, Song Zhang, Zhenzhen Jin
Cited by: 0
References: 32

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract Data-driven approaches are extensively utilized for open-circuit fault diagnosis in permanent magnet synchronous motor (PMSM) drive inverters. However, current methodologies require substantial labeled data and exhibit vulnerability to discrepancies caused by operating condition variations. To address these constrictions, this paper introduces an unsupervised multi-target domain transfer learning approach with dynamic signal standardization. Firstly, a dynamic window sampling mechanism is proposed to ensure the integrity of fault information within the window during speed variations. Secondly, the variable-modulus-length normalization is applied to standardize the signal energy scale, which effectively removes current amplitude variations caused by load variations. On this basis, a multi-target domain adaptive network is developed. It maps the original signal into a high-dimensional space through intra-channel multi-scale feature adaptive weighted aggregation and inter-channel interaction mechanisms. Furthermore, a joint optimization framework based on conditional domain adversarial learning is constructed, which combines fault classification loss and domain discrimination loss. The framework effectively optimizes the model’s domain-invariant fault feature extraction capability. Finally, verification of the method’s effectiveness is carried out using simulation and physical experiments. The results show that the proposed method achieves a diagnostic accuracy exceeding 97.45% in the simulation experiments at an SNR of −10 dB, significantly outperforming existing methods.

Fields of Study & Subjects

Machine Fault Diagnosis Techniques Electrical Fault Detection and Protection Sensorless Control of Electric Motors

Publication details

Journal / source
Engineering Research Express
Volume
8
Issue
8
Pages
085324–085324
Publication date
27 March 2026
Language
EN