基于深度门控循环单元网络的转辙机健康状态评估

Health Condition Assessment of Point Machine Based on a Deep GRU Model

转辙机的健康状态评估是实现状态修的重要技术之一。提出基于时域、时频域特征与循环神经网络的状态评估方法。首先利用集合经验模态分解将原始信号分解,得到不同时间尺度的固有模态分量,并提取每个分量的模糊熵与6个时域特征组合作为原始特征集;然后利用局部加权回归对特征曲线进行平滑以捕获退化趋势,并利用特征的固有属性进行特征选择;最后将最优特征子集作为门控循环单元神经网络的输入,建立时间序列特征与转辙机健康指数的非线性关系。实验结果表明,该方法可以利用转辙机原始功率数据准确提取并选择混合特征,有效识别早期故障,评估转辙机的健康状态。

The assessment of health condition of point machine is one of the important technologies to implement Condition-based Maintenance.A condition assessment method based on time-domain features,time-frequency features and Recurrent Neural Network was proposed.Firstly,the Ensemble Empirical Mode Decomposition was used to decompose the original signal to obtain the Intrinsic Mode Function(IMF)with different time scales,and extract the fuzzy entropy of each IMF component.The fuzzy entropy of each component and six time-domain features were regarded as the original feature set.Then,locally weighted regression was carried out in the features to smooth the curve and capture the degradation trend.Moreover,the better hybrid features were extracted by the inherent attributes.Finally,the optimal feature subset was used as the input of the Gated Recurrent Unit(GRU)network,and the nonlinear relationship between the time series features and point machine health indicator was established.The experimental results show that the proposed method can accurately extract and select the hybrid features by using the original power data of the point machine,effectively identifying the early fault and evaluating the health status of the point machine.