Deep Learning-Based Detection of Electrical Faults in Aerospace Power Systems
Original title:Deep Learning-Based Detection of Electrical Faults and Power Quality Disturbances in Aerospace Power Systems
Unlike standard 50 or 60 Hz utility grids, 400 Hz electrical networks in modern aircraft require specialized disturbance and fault monitoring. Modeling after the Boeing 787 electrical architecture, this paper evaluates multiple deep learning architectures across 21 operating and fault conditions. A compact ResNet with 175,685 parameters was quantized to INT8 and implemented on a Xilinx Zynq UltraScale+ MPSoC board, achieving 95.87% accuracy with an accelerator latency of 6.90 ms per record, demonstrating the feasibility of embedded edge intelligence in flight systems.
Why it's worth reading
It offers an end-to-end implementation and hardware benchmark for deploying compact neural networks on FPGA hardware, addressing the unique constraints of 400 Hz aerospace grids.