EUVIMEDEUVIMED entdecken
Crossref

Deep Learning-Based Detection of Electrical Faults and Power Quality Disturbances in Aerospace Power Systems

Ian C. Guzmán, Radu Babiceanu, Berker Peköz

IEEE Transactions on Aerospace and Electronic Systems · 2026 · 62 · 19422-19442

DOI: 10.1109/taes.2026.3733949 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Aerospace and Electronic Systems, 2026. Verfasst von Ian C. Guzmán und 2 weiteren Autor:innen. Die Arbeit nennt 82 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The transition toward more electric aircraft has introduced complex electrical architectures with strict reliability, safety, and real-time requirements. However, most research on power quality disturbances (PQDs) and electrical fault diagnosis focuses on conventional power grids and low-frequency analysis, limiting applicability to aircraft electrical systems operating at higher frequencies. This article presents a deep learning framework for automated multiclass classification of electrical faults and PQDs in aircraft electrical systems, emphasizing accuracy, robustness, and aerospace constraints. A high-fidelity aircraft power system model inspired by the Boeing 787 architecture was developed to simulate operation at a 400 Hz fundamental frequency. Two datasets were generated: one containing augmented 1-D time-domain signals, publicly available on IEEE DataPort, for 1-D convolutional neural networks (1D-CNNs), and another containing 2-D time–frequency representations derived from the short-time Fourier transform for 2-D CNNs. Architectures including CNNs, LSTMs, hybrid CNNLSTM models, ResNet, MobileNet, and VGG were evaluated. A Compact ResNet achieved the best balance between classification performance and model complexity, attaining 96.94% software test accuracy with 175,685 parameters. Following 8-bit quantization and deployment on a Xilinx Zynq UltraScale+ Multiprocessor System-on-Chip (MPSoC) ZCU102, the FPGA accelerator achieved 95.87% accuracy, 6.90 ms mean inference latency, and 14.40 ms worst-case latency per input record. Resource utilization was 15.52% of digital signal processing (DSP) slices, 63.71% of Block RAM, and 93.77% of CLB limited look-up tables (LUTs). These results demonstrate the potential of lightweight deep learning models for accurate multiclass classification under aerospace-relevant timing and hardware constraints, while highlighting limited look-up table LUT capacity for further FPGA integration.

Zitieren

APA

Guzmán, I. C., Babiceanu, R., & Peköz, B. (2026). Deep Learning-Based Detection of Electrical Faults and Power Quality Disturbances in Aerospace Power Systems. IEEE Transactions on Aerospace and Electronic Systems, 62, 19422-19442. https://doi.org/10.1109/taes.2026.3733949

Vancouver

Guzmán IC, Babiceanu R, Peköz B. Deep Learning-Based Detection of Electrical Faults and Power Quality Disturbances in Aerospace Power Systems. IEEE Transactions on Aerospace and Electronic Systems. 2026;62:19422-19442. doi:10.1109/taes.2026.3733949

BibTeX
@article{Guzmn2026,
  author = {Guzmán, Ian C. and Babiceanu, Radu and Peköz, Berker},
  title = {Deep Learning-Based Detection of Electrical Faults and Power Quality Disturbances in Aerospace Power Systems},
  journal = {IEEE Transactions on Aerospace and Electronic Systems},
  year = {2026},
  volume = {62},
  pages = {19422--19442},
  doi = {10.1109/taes.2026.3733949}
}
Kostenlos starten →