基于声发射信号轮轨耦合传导特征的车载移动式高铁钢轨伤损检测方法研究

The rapid development of high-speed railway promotes the economic and social development, and how to ensure a safe, stable and reliable operation will be an important problem in railway transportation. In order to ensure the safe operation of train and obtain rail safety status in time, the existing detection model must be improved, and a new, fast and convenient high-speed rail defect detection method is very necessary. Based on propagation feature research of rail defect signals in wheel-rail coupling, this project proposes the vehicle-mounted detection method of high-speed rail defect. By finite volume method and three dimensional non-Hertzian rolling contact model, the solid elastic wave model of wheel-rail contact coupling can be established. Acoustic emission features in different rail defect stages will be analyzed and multi-dimension features will be extracted. Combining with the wheel-rail coupling model and analyzing the effect from the load and speed of train, the propagation feature rules of signals from rail to wheel will be researched, and the signal features of rail defects on wheel can be obtained. Finally, multi-sensor adaptive noise cancelling method and strategy will be investigated by multi-sensor signal acquisition on multi-wheel, the classification rules of rail defect stages can be obtained by deep learning algorithm, and the rail defects in the whole railway can be detected by the vehicle-mounted sensors. This project presents a basic research for a new acoustic emission detection method in high-speed rail defects.

高速铁路的快速发展，给经济和社会发展带来了巨大的推动作用，如何保障高铁安全、可靠的运行成为铁路运输面临的重大问题。为保障列车的安全运行，及时掌握钢轨的安全状态，必须改变现有的检测模式，研究一种新型、快速和便捷的高铁钢轨伤损检测方法迫在眉睫。本项目基于钢轨伤损声发射信号在轮轨耦合中的传导特征研究，提出车载移动式高铁钢轨伤损检测的新方法。通过有限体积法和三维非赫兹滚动接触模型，建立轮轨接触耦合的固体弹性波模型；分析不同钢轨伤损阶段的声发射信号特征，进而构建信号的多维特征；结合轮轨耦合模型，并分析列车载荷和速度的影响，研究信号从钢轨到车轮传导过程中的特征规律，获得车轮端的钢轨伤损信号特征；最后，利用多轮上多传感器进行信号采集，研究多传感器自适应消噪方法及策略，通过深度学习算法建立钢轨伤损阶段的识别分类规则，随列车对整个路段钢轨伤损进行检测，为高铁钢轨的新型声发射探伤技术应用做基础性研究。

为保障高速铁路安全、稳定、可靠的运行，针对利用声发射传感器随车对整个路段钢轨伤损进行检测的实际应用问题，本项目基于钢轨裂纹伤损声发射信号在轮轨耦合中的传导特征，研究了车载移动式高铁钢轨伤损检测的方法。基于弹性体赫兹接触理论对轮轨接触中的接触范围及接触面上的应力场进行了研究，并利用子模型技术建立了轮轨接触耦合的有限元模型，研究了车轮端接收信号的特征变化规律；基于混沌原理，采循环神经网络中的NARX结构，建立了轮轨接触运动中的混沌系统模型，并结合裂纹特征对噪声信号进行了有效的去除；提出了基于变步长最小均方的多级自适应消噪方法，达到了同时去除白噪声和有色噪声的目的，能够有效地对高速情况下被噪声淹没的伤损信号进行检测；同时，进一步研究了采用自适应数据融合与卡尔曼滤波算法相结合的多传感器裂纹声发射信号检测方法，该方法能够根据不同传感器的测量结果快速在线估计系统状态，提高了检测结果的准确度；再者，研究了基于双层LSTM深度学习网络的裂纹信号检测方法，该方法能够自主学习噪声模型，有效检测出强噪背景下的裂纹声发射弱信号；利用多声发射事件概率改进了基于卷积神经网络的钢轨伤损分类识别方法，该方法充分利用时间序列信息与频谱信息，达到了提高伤损阶段识别准确度的目标。此外，构建了用于车载移动式钢轨探伤模拟试验系统一套，并研发出钢轨裂纹声发射检测工程样机一套，为后续的应用推广奠定了基础。