This is the authors' abstract. We don't add key points for this paper.
Complex wired networks in automotive and aerospace systems require reliable load-bearing fault diagnosis, as critical issues like intermittent connections are only triggered and manifested by vibration and energization during equipment operation. However, load-bearing reflectometry diagnosis suffers from low signal-to-noise ratios (SNR), as excitation signals must remain low-energy to avoid disrupting connected operating equipment. We propose a robust 1D residual network (Robust 1D-ResNet) integrating bandpass filtering and data augmentation for load-bearing wired network fault diagnosis. A small-scale power network testbed and an LTSpice simulation model were developed to collect real-world and simulated fault datasets. We systematically evaluated the impacts of excitation signal characteristics and noise levels on diagnostic performance. Results show that the proposed method achieves exceptional noise robustness, attaining matching rates (MR) of 93.5% on simulated data and 98.6% on real-world data under severe noise conditions. Furthermore, practical deployment tests demonstrate an inference latency of only 7.08 ms on an embedded Rockchip RK3588 NPU (Batch Size = 1) with bandpass filtering (compared to 3.46 ms without filtering). This confirms the method’s high computational efficiency and its potential for real-time load-bearing fault diagnosis.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Huang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: