EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Condition‐Constrained Imbalance Learning for Incipient PMSM Winding Fault Diagnosis Under Multiple Operating Conditions

Ming Jing, Guanglong Jia, Fengge Zhang

IET Electric Power Applications · 2026-01 · 20 (1)

DOI: 10.1049/elp2.70249 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IET Electric Power Applications, online erschienen am 19.11.2026. Verfasst von Ming Jing und 2 weiteren Autor:innen. Die Arbeit nennt 23 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: National Science and Technology Major Project, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Reliable diagnosis of incipient interturn short‐circuit (ITSC) faults in PMSMs remains challenging because experimentally validated fault data are scarce, fault severities are highly imbalanced and fault signatures vary across operating conditions. To address these challenges, a PMSM prototype with controllable ITSC faults is developed, and three‐phase stator currents are acquired under PWM inverter supply at multiple steady‐state speed–load operating points. An experimental‐data‐driven, condition‐constrained imbalance learning framework is proposed. Feature selection first constructs a compact diagnostic representation to reduce the adverse effect of redundant dimensions on sparse minority neighbourhoods. Condition‐constrained Borderline‐SMOTE (CC‐BSMOTE) then incorporates operating‐condition consistency into borderline‐sample identification, neighbourhood construction and synthetic‐sample generation, thereby avoiding cross‐condition interpolation while preserving informative local variations in minority‐fault distributions. The framework is evaluated using record‐independent testing and repeated class‐ and condition‐stratified downsampling over imbalance ratios from r = 2 to r = 20. At r = 20, the proposed method achieves a Macro‐F1 of 0.8925 ± 0.0061, an A1 (the fault with the fewest samples) recall of 0.7007 ± 0.0182 and a fault false‐negative rate of 0.0805 ± 0.0057. Compared with global Borderline‐SMOTE, CC‐BSMOTE improves Macro‐F1 and A1 recall by 3.28 and 5.39 percentage points, respectively, while reducing the fault false‐negative rate by 16.3% in relative terms. Moreover, its performance degrades more slowly as the imbalance increases. These results demonstrate that severity‐dependent fault imbalance is conditionally structured rather than merely numerical and that preserving operating‐condition consistency during minority augmentation improves incipient ITSC recognition and robustness across multiple operating points.

Förderung

  • National Science and Technology Major Project
  • National Natural Science Foundation of China

Zitieren

APA

Jing, M., Jia, G., & Zhang, F. (2026). Condition‐Constrained Imbalance Learning for Incipient PMSM Winding Fault Diagnosis Under Multiple Operating Conditions. IET Electric Power Applications, 20(1). https://doi.org/10.1049/elp2.70249

Vancouver

Jing M, Jia G, Zhang F. Condition‐Constrained Imbalance Learning for Incipient PMSM Winding Fault Diagnosis Under Multiple Operating Conditions. IET Electric Power Applications. 2026;20(1). doi:10.1049/elp2.70249

BibTeX
@article{Jing2026,
  author = {Jing, Ming and Jia, Guanglong and Zhang, Fengge},
  title = {Condition‐Constrained Imbalance Learning for Incipient PMSM Winding Fault Diagnosis Under Multiple Operating Conditions},
  journal = {IET Electric Power Applications},
  year = {2026},
  volume = {20},
  number = {1},
  doi = {10.1049/elp2.70249}
}
Kostenlos starten →