基于CNN-LSTM的车辆悬挂系统故障识别方法研究  

Fault identification method for vehicle suspension system based on CNN-LSTM

作　　者：梁君海 李智强 葛天 郑志伟 LIANG Junhai;LI Zhiqiang;GE Tian;ZHENG Zhiwei(R&D Center,CRRC Chengdu Locomotive and Rolling Stock Co.Ltd.,Chengdu 610511,China;State Key Laboratory of Rail Transit Vehicle System,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]中车成都机车车辆有限公司技术中心,成都610511 [2]西南交通大学轨道交通运载系统全国重点实验室,成都610031

出　　处：《铁路计算机应用》2024年第2期1-6,共6页Railway Computer Application

摘　　要：高速铁路车辆(简称:车辆)运行条件恶劣多变,车辆悬挂系统的可靠性关系到行车安全和乘坐舒适性。当车辆的悬挂系统发生故障时,振动信号呈现非线性、非平稳的特征。为此,提出了一种基于卷积神经网络(CNN,Convolutional Neural Network)-长短时记忆(LSTM,Long Short-Term Memory)模型的车辆悬挂系统故障识别方法。通过SIMPACK平台建立了包含悬挂系统的车辆-轨道耦合动力学模型,获得了车辆系统各部件在健康状态及各类故障状态下的振动信号;以与故障元件关联部件的振动加速度信号作为模型输入,通过构建的CNN-LSTM模型对时序信号进行特征提取和分类预测,进而实现对车辆悬挂系统的故障识别;通过构建不同工况的故障数据集对该方法进行评估。试验结果表明,该方法在速度等级相同的情况下,故障识别准确率可达98%;在速度等级不同的情况下,故障识别准确率可达99%,验证了该方法的有效性。The operating conditions of high-speed vehicles were harsh and changeable,and the reliability of suspension system was related to the safety and riding comfort of trains.When the suspension system of the high-speed vehicles fails,the vibration signal was non-linear and non-stationary.Therefore,this paper proposeded a fault identification method for vehicle suspension system based on Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)model.The paper established a vehicle track coupling dynamic model with suspension system on the SIMPACK platform,and obtained vibration signals of various components of the vehicle system under healthy and various fault states,and took the vibration acceleration signal of the components associated with the faulty component as the input of the model,through the construction of a CNN-LSTM model,extracted features and predicted classification of time-series signals,thereby achieved fault recognition of the vehicle suspension system,evaluated the method by constructing fault datasets for different operating conditions.The experimental results showed that the fault recognition accuracy of the method can reach up to 98%under the same speed level,and the fault recognition accuracy can reach up to 99%under different speed levels,which verifies the effectiveness of the method.

关 键 词：高速铁路车辆 悬挂系统 卷积神经网络 长短时记忆模型 故障识别 

分 类 号：U270.331[机械工程—车辆工程] TP39[交通运输工程—载运工具运用工程]