基于一维卷积神经网络的地铁钢轨波磨识别方法

来源期刊:中南大学学报(自然科学版)2021年第4期

论文作者:温泽峰 谢清林 陶功权

文章页码:1371 - 1380

关键词:钢轨波磨;轴箱振动;深度学习;一维卷积;“空间域”切割

Key words:rail corrugation; axle box vibration; deep learning; 1-DCNN; "spatial domain" cutting

摘    要:利用轴箱振动加速度构建一种基于一维卷积神经网络(1-dimensional convolutional neural network, 1-DCNN)的地铁钢轨波磨智能识别方法。提出“空间域”切割的方法制作样本集,构建恰当的1-DCNN结构与配置参数可对输入样本集数据进行自动特征提取并学习分类。通过设置“空间窗”长度任意调节钢轨波磨智能分类时的定位分辨率。研究结果表明:提出的1-DCNN方法能有效、快速且稳定地对钢轨波磨进行智能识别与定位,在车辆复杂的运营条件及速度时变工况下仍然能保持较高的识别精度,稳定在99.20%(标准差为0.1);与此同时,对每一条样本的识别时间均少于0.2 ms,满足钢轨波磨在线监测的时效性。

Abstract: A method based on 1-dimensional convolutional neural network(1-DCNN) was developed to identify the rail corrugation intelligently using the axle box acceleration. A mean of "spatial domain" cutting was proposed to make the sample set, and an appropriate 1-DCNN structure and configuration parameters were constructed, so that the data of the input sample set could be automatically extracted and classified. In addition, by setting the length of "spatial window", the resolution of rail corrugation positioning detection could be adjusted arbitrarily. The results show that the proposed 1-DCNN method can effectively, rapidly and stably carry out intelligent identification and positioning of the rail corrugation, and maintain a high identification accuracy of 99.20% (standard deviation of 0.1) even under the complicated operating conditions and at time-varying speed of the vehicle. Meanwhile, the recognition time of each sample data is less than 0.2 ms, which can meet the requirement of online monitoring of rail corrugation.

有色金属在线官网  |   会议  |   在线投稿  |   购买纸书  |   科技图书馆

中南大学出版社 技术支持 版权声明   电话:0731-88830515 88830516   传真:0731-88710482   Email:administrator@cnnmol.com

互联网出版许可证:(署)网出证(京)字第342号   京ICP备17050991号-6      京公网安备11010802042557号