EUVIMEDEUVIMED entdecken
Crossref

A Remaining Useful Life Prediction Framework for IGBT Based on Dynamic Precursor Weighting

Jianhua Zhang, Maoliang Jian, Haorui Zhu, Lianqiao Yang

IEEE Transactions on Power Electronics · 2026-09 · 41 (9) · 16129-16143

DOI: 10.1109/tpel.2026.3680653 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Power Electronics, 2026-09. Verfasst von Jianhua Zhang und 3 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Gefördert von: Natural Science Foundation, China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In power electronic systems, accurately predicting the remaining useful life (RUL) of insulated gate bipolar transistors is essential for ensuring system reliability. This article presents a robust RUL prediction framework that integrates dynamic precursor weighting, health indicator (HI) construction, and deep learning. First, power cycling tests are conducted to obtain aging-sensitive precursors. A sliding-window-based degradation rate analysis is employed to assign time-varying weights for adaptive precursor selection and fusion, resulting in an HI that captures the degradation trend. Wavelet threshold denoising is then applied to the constructed HI to reduce noise and enhance smoothness. Next, a comprehensive set of time-domain features is extracted from the denoised HI, and relevant features are selected using correlation, monotonicity, similarity, and redundancy criteria. To further capture frequency-domain information, variational mode decomposition is applied, producing intrinsic mode functions that are also used as predictive inputs. The selected features are fed into a one-dimensional convolutional neural network (1-D CNN)-Informer model for RUL prediction. Two evaluation strategies are employed to assess the generalization performance. The proposed HI consistently outperforms classical aging precursors across all tasks, demonstrating its superior capability in capturing degradation patterns. In addition, comparative experiments with 1-D CNN-Transformer and 1-D CNN-long short-term memory (LSTM) models show that the CNN-Informer achieves higher prediction accuracy and better training efficiency. These results confirm the effectiveness and robustness of the proposed framework.

Förderung

  • Natural Science Foundation, China

Zitieren

APA

Zhang, J., Jian, M., Zhu, H., & Yang, L. (2026). A Remaining Useful Life Prediction Framework for IGBT Based on Dynamic Precursor Weighting. IEEE Transactions on Power Electronics, 41(9), 16129-16143. https://doi.org/10.1109/tpel.2026.3680653

Vancouver

Zhang J, Jian M, Zhu H, Yang L. A Remaining Useful Life Prediction Framework for IGBT Based on Dynamic Precursor Weighting. IEEE Transactions on Power Electronics. 2026;41(9):16129-16143. doi:10.1109/tpel.2026.3680653

BibTeX
@article{Zhang2026,
  author = {Zhang, Jianhua and Jian, Maoliang and Zhu, Haorui and Yang, Lianqiao},
  title = {A Remaining Useful Life Prediction Framework for IGBT Based on Dynamic Precursor Weighting},
  journal = {IEEE Transactions on Power Electronics},
  year = {2026},
  volume = {41},
  number = {9},
  pages = {16129--16143},
  doi = {10.1109/tpel.2026.3680653}
}
Kostenlos starten →