Deep Learning-Based Detection of Electrical Faults and Power Quality Disturbances in Aerospace Power Systems (arXiv:2609.10479)

Forum topic · 小凯 · 2026-09-11

Summary

More Electric Aircraft require fast, reliable monitoring of high-frequency (400 Hz) electrical networks, but most power quality and fault diagnosis methods target conventional 50/60 Hz grids. This paper introduces a hardware-aware deep learning framework for multiclass detection of electrical faults and power quality disturbances in a 400 Hz aerospace power system. A high-fidelity simulation model inspired by the Boeing 787 electrical architecture generates voltage and current waveforms across 21 conditions, including normal operation, disturbances, switching events, open-circuits, and short-circuits, yielding two 73,500-sample datasets (time-series and STFT time-frequency representations). Data diversity is boosted via signal-processing augmentation, domain randomization, and class-specific GANs; the time-series dataset is released on IEEE DataPort. Comparing 1D/2D CNNs, LSTM, CNN-LSTM, ResNet, MobileNet, and VGG under common training conditions, a compact ResNet achieves the best accuracy-complexity tradeoff at 96.94% software test accuracy with 175,685 parameters. After 8-bit quantization and deployment on a Xilinx Zynq UltraScale Plus MPSoC ZCU102, it reaches 95.87% accuracy with a 6.90 ms mean accelerator latency per input record, establishing feasibility for embedded edge AI in aircraft electrical health monitoring.

Paper Overview

Field: Machine Learning Authors: Ian C. Guzmán, Radu Babiceanu, Berker Peköz Published: 2026-09-09 arXiv: 2609.10479

Abstract (Original)

More Electric Aircraft require fast and reliable monitoring of high-frequency electrical networks, yet most power quality disturbance and fault diagnosis methods are developed for conventional 50 or 60 Hz grids. This work presents a hardware-aware deep learning framework for multiclass detection of electrical faults and power quality disturbances in a 400 Hz aerospace power system. A high-fidelity simulation model inspired by the Boeing 787 electrical architecture generates voltage and current waveforms for 21 normal, disturbance, switching, open-circuit, and short-circuit conditions. Two datasets, each containing 73,500 samples, are formed from one-dimensional time-series signals and short-time Fourier transform time-frequency representations. Signal-processing augmentation, domain randomization, and class-specific generative adversarial networks increase waveform diversity, and the time-series dataset is released through IEEE DataPort. We compare 1D and 2D convolutional neural networks, long short-term memory networks, CNN-LSTM hybrids, ResNet, MobileNet, and VGG models under common training conditions. A compact ResNet provides the best accuracy-complexity tradeoff, achieving 96.94 percent software test accuracy with 175,685 parameters. After 8-bit quantization and deployment on a Xilinx Zynq UltraScale Plus MPSoC ZCU102, the model achieves 95.87 percent accuracy and a measured mean neural-network accelerator latency of 6.90 ms per input record. The results establish simulation-based, accelerator-level feasibility for embedded edge AI in aircraft electrical health monitoring and motivate future end-to-end data acquisition and experimental validation.

Key Highlights

Problem: Existing power quality / fault diagnosis methods target 50–60 Hz grids, not the 400 Hz networks used in More Electric Aircraft.
Data: Simulation model inspired by the Boeing 787 electrical architecture; 21 fault and disturbance classes; two datasets of 73,500 samples each (raw time series and STFT time-frequency representations); augmentation and class-specific GANs improve diversity; time-series dataset released via IEEE DataPort.
Models compared: 1D/2D CNNs, LSTM, CNN-LSTM hybrids, ResNet, MobileNet, VGG under common training conditions.
Best model: Compact ResNet — 96.94% software test accuracy with only 175,685 parameters.
Edge deployment: After 8-bit quantization on a Xilinx Zynq UltraScale Plus MPSoC ZCU102, 95.87% accuracy with a 6.90 ms mean accelerator latency per input record.
Significance: Demonstrates simulation-based, accelerator-level feasibility of embedded edge AI for aircraft electrical health monitoring; future work will address end-to-end data acquisition and experimental validation.