轨道平顺性检测方法现状及发展综述  

Overview on Existing and Developing Methods for Track Irregularity Detection

作　　者：李奇[1,2] 戴宝锐 杨飞 石龙 吴阅 LI Qi;DAI Baorui;YANG Fei;SHI Long;WU Yue(School of Civil Engineering,Tongji University,Shanghai 200092,China;Xizang Agricultural and Animal Husbandry University,Nyingchi 860000,China;State Key Laboratory for Track System of High-speed Railway,Beijing 100081,China;China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;Postgraduate Department,China Academy of Railway Sciences,Beijing 100081,China)

机构地区：[1]同济大学土木工程学院,上海200092 [2]西藏农牧学院,西藏林芝860000 [3]高速铁路轨道系统全国重点实验室,北京100081 [4]中国铁道科学研究院集团有限公司,北京100081 [5]中国铁道科学研究院研究生部,北京100081

出　　处：《铁道学报》2024年第7期101-116,共16页Journal of the China Railway Society

基　　金：西藏自治区自然科学基金(XZ202301ZR0040G);国家自然科学基金(52178432);中国铁道科学研究院集团有限公司科研项目(2021YJ054)。

摘　　要：概述3种常见的轨道平顺性检测方法,从数据驱动与模型驱动2个方面对轨道平顺性车载检测方法的研究和发展进行详细评述,并分析其面临的主要挑战。数据驱动方法多应用于铁路系统健康状态检测,在轨道不平顺动态检测及轨下结构变形提取方面的研究正逐步展开。在模型驱动方法中,逆模型法目前主要用于获取轨道不平顺的频域特征;卡尔曼滤波类方法相比于惯性基准法具有融合多个传感器数据来提升轨道不平顺检测精度的优势。未来的研究应将物理模型和机制引入数据驱动的机器学习和深度学习模型中,在减少训练样本的情况下保证轨道平顺性的预测精度。多种传感器数据融合的动态检测,以及从动态不平顺中分离出不同成分,也是未来亟需发展的技术。This study firstly introduced three general methods for track irregularity detection.Then,a detailed review was conducted on the research and development of vehicle-mounted detection methods in detail from the aspects of both data-driven and model-driven methods.The main challenge of the vehicle-mounted detection method was also analyzed.Data-driven methods are widely used in the healthy monitoring of railway system,and research on dynamic detection of track irregularities and extraction of structural deformation under track is underway.As one of the model-driven methods,the inverse model method is mainly used to obtain the frequency domain characteristics of track irregularity at present.Compared with the inertial reference method,the Kalman filtering method has the advantage of integrating multiple sensor data to improve the detection accuracy of track irregularity.Future research should combine physical models and mechanisms with data-driven machine learning models and deep learning models to reduce the number of training samples while ensuring the estimation accuracy of track irregularity.In order to ensure the comprehensive,timely and accurate identification of the state of the track and the structure below the track under the condition of frequent train operation,it is urgent to develop dynamic detection technologies that integrate vehicle-mounted vibration,optical and acoustic sensors.Dynamic detection by fusion of data from multiple sensors and separation of different components from dynamic track irregularity are also technologies that need to be developed urgently in the future.

关 键 词：轨道平顺性 车载检测 数据驱动 模型驱动 卡尔曼滤波 

分 类 号：U216.3[交通运输工程—道路与铁道工程] U213.2[建筑科学—桥梁与隧道工程] U448.1