Abstract: In view of the high real-time requirements of turnout fault diagnosis system and the serious dependence of feature extraction on prior knowledge, a real-time fault diagnosis method for turnouts based on one-dimensional convolutional neural network（1D-CNN） is proposed. Taking the power curve of S700k switch machine as an example, a one-dimensional convolutional neural network structure model is established. The model integrates feature extraction and fault classification, optimizes network parameters, and improves the generalization ability of the model by using regularized dropout, uses t-SNE visualization method to reflect the effectiveness of model extraction features. Simulation results show that the adaptive feature extraction of the original time domain signals by the convolutional layer and the pooling layer can better capture the spatial dimension information of signal, reduce the calculation amount of the model, and improve the anti-noise performance of the model, achieving the end-to-end real-time fault diagnosis, and effectively improve the accuracy of the real-time fault diagnosis of the turnout.

Key words: one-dimensional convolution neural network, S700k switch machine, time series, fault diagnosis

摘要： 针对道岔故障诊断系统实时性要求高、特征提取严重依赖于先验知识的问题，提出了一种基于一维卷积神经网络（1D-CNN）的道岔实时故障诊断方法。以S700k转辙机的功率曲线为例，建立一维卷积神经网络的结构模型，该模型将特征提取与故障分类融合为一体，优化了网络参数，同时使用正则化Dropout提高模型的泛化能力，采用t-SNE可视化方法，来反映模型提取特征的有效性。仿真实验表明：卷积层和池化层对原始时域信号的自适应特征提取，能较好地捕捉信号空间维度信息，降低模型的计算量，提高模型的抗噪性能，实现了端到端的实时故障诊断，并有效地提高道岔故障实时诊断的准确率。

关键词: 一维卷积神经网络, S700k转辙机, 时间序列, 故障诊断