基于改进ConvNeXt的接触网吊弦故障诊断算法的研究

Research on High-Speed Railway Catenary Dropper Fault Diagnosis Based on Improved ConvNeXt

摘要: 随着高铁运行速度与频次的增加，接触网吊弦故障的发生频率也随之增加，这不仅给高铁运行安全带来巨大隐患，同时也对铁路维修工作提出了严峻挑战。现有的检测方法难以满足对吊弦故障高效准确诊断的需求，鉴于此现状，本文提出了一种创新的接触网吊弦故障诊断方法——InCANeXt网络模型。该模型由ConvNeXt模型基础上改进而来，主要改进措施包括：通过减少模型基础块的堆叠数以降低模型复杂度，引入Inception结构以优化卷积计算过程，并在模型中集成了CA与SimAM注意力机制以增强特征提取能力。上述优化措施使检测模型能更准确、快速地检测吊弦的运行状态。将所提出方法与多种图像分类算法在真实接触网吊弦图像数据集上进行对比实验，InCANeXt模型在吊弦故障诊断任务上达到了88.71%的准确率。实验结果证明，所提出的方法在吊弦故障检测任务上展现了卓越的性能。

Abstract: As the operating speed and frequency of high-speed trains increase, the incidence of catenary dropper failures also rises, posing significant safety hazards to high-speed rail operation and presenting severe challenges to railway maintenance work. Existing detection methods struggle to meet the demand for efficient and accurate diagnosis of dropper faults. In light of this situation, this paper proposes an innovative method for diagnosing catenary dropper faults—the InCANeXt network model. This model is an improvement on the ConvNeXt model, with major enhancements including reducing the number of basic blocks stacked to decrease model complexity, incorporating the inception structure to optimize the convolution process, and integrating CA and SimAM attention mechanisms to enhance feature extraction capability. These optimizations enable the detection model to diagnose the operational status of droppers more accurately and swiftly. By comparing the proposed method with various image classification algorithms on a real catenary dropper image dataset, the InCANeXt model achieved an accuracy of 88.71% in the task of dropper fault diagnosis. The experimental results demonstrate that the proposed method exhibits outstanding performance in the task of dropper fault detection.