基于概率神经网络的S700K转辙机故障诊断方法

Fault Diagnosis Method of S700K Switch Machine Based on PNN

转辙机故障对铁路运输安全影响重大。针对转辙机故障原因与现象之间的复杂不确定性关系,提出一种基于概率神经网络的S700K转辙机故障诊断方法。在对转辙机各个状态功率曲线动作原理进行分析的基础上,根据信号不同时域特征参数,对各工作区段的特征进行提取;依据故障现象与故障类型的关系建立概率神经网络模型,以F1-Score作为诊断准确性评价指标,通过测试不同平滑因子对F1-Score值的影响,确定0.019为概率神经网络进行故障诊断最优的平滑因子;最后选择来自某电务段的81组S700K转辙机故障数据作为测试数据,验证了该智能故障诊断方法的可靠性。

The switch fault has a great impact on the safety of railway transportation.In view of the complex and uncertain relationship between the causes and phenomena of the switch fault,a fault diagnosis method of S700K switch based on probabilistic neural networks(PNN)is presented.Based on the analysis of the power curve action principle for each state of the switch machine,the features of each operating zone are extracted according to the different time-domain feature parameters of the signal.The probabilistic neural network model is established according to the connection relationship between fault phenomena and fault types,and F1-Score is used as the diagnostic accuracy evaluation index.0.019 is determined as the optimal smoothing factor of the probabilistic neural network for fault diagnosis by testing the influence of different smoothing factors on F1-Score values.Finally,81 sets of S700K switch fault data from a certain communication and signal depot are selected as test data to verify the reliability of the intelligent fault diagnosis method.