【摘要】 针对铁路道岔转辙机设备状态监测数据不完备、非平衡和局部缺失等问题，提出基于改进生成式对抗网络（GAN）的转辙机柱塞泵故障诊断方法。首先，构建条件深度卷积生成式对抗网络（C-DCGAN）模型，利用一维卷积层处理柱塞泵振动信号的时域和频域特征，通过博弈对抗机制优化生成器和判别器，提高模型的泛化能力和故障特征提取能力；其次，引入双时间尺度更新规则（TTUR）来解决判别器正则化过程中的缓慢学习问题，提升模型训练的稳定性；最后，采用实测数据进行案例分析，验证所提方法的有效性。结果表明：在4种不同工况下，生成样本的JSD值分别为0.190,0.235,0.240和0.185；在正常样本与故障样本比例分别为100∶1,50∶1,20∶1和10∶1时，故障分类精度依次达到91.24%,94.13%,94.96%和96.08%。该方法在样本生成方面具有更好的性能，尤其在处理数据不平衡问题时，可达到较高的故障分类精度，为铁路安全运行提供有力保障。更多还原

【Abstract】 To address the issues of incomplete, imbalanced, and partially missing condition monitoring data from railway switch machine, a fault diagnosis method of the plunger pump in switch machines with an improved Generative Adversarial Network(GAN) is proposed in this paper. First, a Conditional Deep Convolutional Generative Adversarial Network(C-DCGAN) model is established, with one-dimensional convolutional layers designed to capture the timing and frequency-domain features of the plunger pump vibration signal. The generator and discriminator are optimized through game confrontation mechanism to enhance the model’s generalization and fault feature extraction capability. Then, Two Time-scale Update Rule(TTUR) is added to solve the slow learning issue in discriminator regularization, improving the stability of model training.Finally, a case study using measured data provided by the Electrical Equipment Materials Company of a railway bureau is conducted to validate the effectiveness of the method. The results show that the JSD obtained from the generated samples are 0. 190, 0. 235, 0. 240 and 0. 185 under the four working conditions. When the normal-to-fault sample ratios are 100∶1, 50∶1, 20∶1 and 10∶1, respectively, the fault classification accuracies reach 91. 24%, 94. 13%, 94. 96% and 96. 08%, consecutively. The proposed method achieves better performance in handling imbalanced data, achieving high fault classification accuracy, thereby providing robust support for ensuring the safe operation of railways.更多还原