基于多通道输入和1DCNN-LSTM的道岔转辙机故障诊断

Fault Diagnosis of Switch Machine Based on Multi-channel Input and 1DCNN-LSTM

针对三相交流道岔转辙机故障诊断问题,提出一种基于多通道输入和一维卷积神经网络(1DCNN)-长短期记忆神经网络(LSTM)的故障诊断方法。首先使用经验模态分解算法对动作功率信号进行分解,获得若干个尺度特征不同的固有模态函数信号;其次建立基于1DCNN和LSTM的组合故障诊断模型,使用1DCNN提取功率信号中的局部特征,使用LSTM选择性提取局部特征中的长距离特征;然后通过所建模型诊断出道岔转辙机的故障类型,并结合t-分布随机近邻嵌入展示诊断效果;最后与经典的诊断方法进行对比分析。对比实验结果表明:本方法在道岔转辙机故障诊断中具有较高的准确性和稳定性,且具有较好的泛化性。

Aiming at the problem of fault diagnosis of three-phase AC switch machines,a fault diagnosis method based on multi-channel input and one-dimensional convolutional neural network(1DCNN)-long-short term memory(LSTM)network was proposed.Firstly,the action power signal was decomposed by empirical mode decomposition algorithm to obtain several intrinsic mode function signals with different scale characteristics.Secondly,a combined fault diagnosis model was built based on the 1DCNN and the LSTM,which used 1DCNN to extract local features in the power signal,and LSTM to selectively extract long-distance features in local features.Finally,the fault type of the switch machine was diagnosed by the model,and the diagnosis effect was demonstrated by t-distributed stochastic neighbor embedding.Simultaneously,the model was compared with the classical diagnosis method.The results of the experiment show that this method has relatively high accuracy,stability as well as good generalization in fault diagnosis of switch machines.