【摘要】 针对高速铁路工况复杂、不同工况下牵引电机定子电流差异较大、传统逆变器开路故障诊断算法泛化能力较差等问题,提出新的牵引电机定子电流数据预处理方法和基于卷积神经网络的逆变器开路故障诊断算法。该数据预处理算法先对三相定子电流进行角度域同步重采样,再采用基于积分的波形修复算法对电流波形进行修复,并对数据进行规范。提出用于单相定子电流故障诊断的卷积神经网络(T-CNN),以及用于解决两个绝缘栅双极型晶体管同时故障可能导致T-CNN误诊断问题的故障冲突解决规则。实验证明,相比于传统的智能诊断算法,当仅采用单个工况的故障数据训练故障诊断算法时,所提算法在其他工况下具有更好的诊断精度和泛化能力。更多还原

【Abstract】 The working conditions of the high-speed railway are complex. Under different working conditions, the stator current of traction motor is quite different, and the domain adaptability of traditional open-circuit fault diagnosis algorithms is poor. To solve these problems, a new data preprocessing method of stator current and an open-circuit fault diagnosis algorithm based on convolution neural network were proposed. The data preprocessing algorithm firstly synchronously resampled the three-phase stator current in angle domain, then the integral based waveform repair algorithm was used to repair the current waveform, and finally the current value was normalizes. A Convolutional Neural Network(CNN) named T-CNN was proposed to diagnose single-phase stator current open-circuit faults. The solution rules of fault conflict were proposed to solve the problem that two IGBTs fail at the same time, which might lead to T-CNN misdiagnosis. Experimental results showed that the proposed diagnosis algorithm had better diagnosis accuracy and generalization ability in other working conditions than traditional intelligent diagnosis algorithms when only using fault data of one working condition to train these algorithms.更多还原