基于数据扩充的深度置信网络地铁轨排故障识别方法研究

Research on the Method of Subway Track Fault Identification of Deep Belief Network Based on Data Enrichment

针对轨排铺设过程中钢轨与轨枕之间扣配件丢失产生的故障,振动加速度数据特征表征不明显的问题,为了提高铺轨过程中轨排故障识别精度,提出了一种基于数据扩充的深度置信网络的地铁轨排故障识别方法。首先,将采集到的轨排振动加速度数据利用改进的相似度准则进行样本扩充,再通过小波降噪处理得到新的样本作为数据集,并按比例划分为训练集和测试集;然后,将训练样本集输入到DBN模型中进行训练,利用训练好的DBN模型对测试样本集进行故障识别,从而确定故障位置和故障类型;最后,将本方法应用于轨排振动加速度数据进行故障识别,同时与基于原始数据的故障识别方法相比较,结果表明:所提出的方法对轨排故障类型、故障位置的识别准确率有了明显提升。

Aiming at the problem that the fault caused by the loss of fastener fittings between rail and sleeper during track panel laying and the characterization of vibration acceleration data is not obvious,in order to improve the accuracy of track panel fault identification in the process of rail laying,a subway track panel fault identification method of deep belief network based on data augmentation is proposed.Firstly,the collected track panel vibration acceleration data is expanded by the improved similarity criterion,and then the new sample is obtained as a dataset by wavelet noise reduction processing,and it is proportionally divided into training set and test set.Then,the training sample set is input to the DBN model for training,and the trained DBN model is used to identify the fault of the test sample set to determine the fault location and fault type.Finally,the proposed method is applied to the vibration acceleration data of track panel for fault identification,and compared with the fault identification method based on the original data,the results show that the proposed method has significantly improved the identification accuracy of fault type and fault location of track panel.