列车空气制动机测试类型自动辨识方法研究  

Research on Automatic Identification Method of Train Air Braking Machine Test Type

作　　者：王杰 刘泽[1] 郭成[2] 曹景铭 李俊杰 马丁一 WANG Jie;LIU Ze;GUO Cheng;CAO Jing-ming;LI Jun-jie;MA Ding-yi(School of Electronic Information Engineering,Beijing Jiaotong University,Beijing 100044,China;Scientific Research Institution,Kunming Railway Administration,Kunming 650011,China)

机构地区：[1]北京交通大学电子信息工程学院,北京100044 [2]昆明铁路局集团科研所,云南昆明650011

出　　处：《测控技术》2023年第3期19-23,43,共6页Measurement & Control Technology

基　　金：国家自然科学基金(61771041)。

摘　　要：列车空气制动机测试系统工作时记录大量的测试数据,测试数据的自动分析可提高试验系统测试效率,拓展测试模式。针对列车制动机测试数据的特点,提出了基于格拉姆角场(GAF)图像编码与卷积神经网络(CNN)相结合的列车制动测试类型辨识方法。可对制动测试采集的时序气压数据编码重构,扩展为二维网格图像数据,由CNN主干网络实现图像特征信息的自动提取,进而设计分类器进行列车制动测试类型辨识分类。基于现场数据的分析结果表明,所提出的GAF-CNN模型对制动机测试类型识别的辨识率可达到98%以上,可应用于现场测试系统。另外,测试表明使用求和(Summation)计算的GASF比差分(Difference)计算的格拉姆角差域(GADF)辨识率更高。The train air brake test system records a large amount of test data during operation,and the automatic analysis of the test data can improve the test system test efficiency and expand the test mode.Aiming at the characteristics of train brake test data,a train brake test type identification method based on the combination of Gramian angular field(GAF) image coding and convolutional neural network(CNN) is proposed.The data is encoded and reconstructed and expanded into two-dimensional grid image data.The CNN backbone network realizes the automatic extraction of image feature information,and then a classifier is designed to identify and classify the type of train braking test.The analysis results based on field data show that the recognition rate of the GAF-CNN model proposed can reach more than 98% in the recognition of brake test types,which can be applied to field test systems.In addition,the test shows that Gramian angular summation field(GASF) calculated using the summation has a higher recognition rate than Gramian angular difference field(GADF) calculated by the difference.

关 键 词：列车制动机测试 格拉姆角场 卷积神经网络 模式辨识 时间序列分类 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] O235[自动化与计算机技术—控制科学与工程]