强噪声背景下高速铁路钢轨伤损的声发射检测与诊断方法研究

In order to keep pace with the fast development and high running speed of high speed railway, as well as changing the inefficiency situation in field of rail defect detection, this project proposes a real time monitoring method of rail state under high speed and strong background noise by acoustic emission technology. Based on the fracture mechanics theory and seismic moment tensor, the multi-point excitation crack defect model is established with the speed as the dominant factor. From the mechanism of wheel-rail moving noise, the contact surface excitation noise model is established based on the principle of tribology and abrasion stimuli, and the noise and defect signals in high running speed condition can be obtained. For the strong noise with complicated components in high speed, the de-noising method based on colored noise Kalman filter and multi-layer auto-correlation is proposed, and modal acoustic emission theory is used to extract valid defect waveform. Then the comprehensive defect features from time domain, frequency domain and energy field can be obtained and used to establish a feature library of rail defects. Subsequently, the structural features of the signal sequences are learned by the recurrent neural network method, and the real-time defect classifier is established. Finally, activity intensity analysis is applied to acoustic emission signals so as to diagnose the health stages of the rail and realize rail health monitoring under high running speed and strong background noise conditions.

为与当今高速铁路快速发展及运行速度的不断提高相匹配，改变目前钢轨探伤耗人耗时、效率低下的局面，本项目针对高速、强噪声背景条件下钢轨伤损的实时监测难题，提出一种高速强噪条件下的钢轨声发射技术检测与诊断方法。以速度为主要影响因素，利用断裂力学理论和矩张量方法，建立速度相关多点激励裂纹伤损模型；从轮轨运动噪声产生机理出发，运用摩擦学和磨损激励声发射原理，建立速度相关激励噪声模型，从而获得高速轮轨运动下的噪声与伤损声发射信号。针对成分复杂的高速强噪声，提出基于多重自相关的有色卡尔曼滤波去噪方法，根据模态声发射理论提取钢轨有效伤损波形；继而获取伤损信号的时域、频域及能量综合特征，建立钢轨伤损信号特征库，采用循环神经网络方法对信号序列的结构特征进行学习，建立实时伤损分类器，再对声发射的活动强度进行分析，诊断钢轨寿命阶段，最终实现高速轮轨运动强噪声情况下的钢轨健康监测。

为与当今运行速度不断提高的高速铁路系统相匹配，改变目前钢轨探伤耗人耗时、效率低下的局面，本项目对高速铁路强噪影响下的钢轨伤损声发射检测与识别问题进行了探究。首先研究了钢轨裂纹的声发射特性提取，基于自适应小波基构建出了裂纹声发射特征高精度提取方法，同时基于最小均方反卷积方法给出钢轨裂纹模拟的方法；而后研究了轮轨滚动噪声的建模问题，基于统计学与分形理论分别建立了轮轨滚动接触的噪声能量模型，拟合出轮轨噪声能量受运动速度、轮轨接触等条件的影响，并通过分形维描述出滚动噪声在时域及频域的本质属性，为钢轨伤损、轮轨噪声的判别提供理论基础；继而，为实现对轮轨滚动高速强噪声的抑制，研究了时变自回归与卡尔曼滤波方法；同时为了增强去噪结果中的裂纹信号信噪比，研究了香农熵改进的小波自适应谱线增强器，在去噪同时增强裂纹的固有特征；再通过参数优化下的变分模态分解挖掘提取噪声淹没下的潜在裂纹信号时频特征；最后，为识别真实铁路环境下的裂纹伤损，基于对抗式循环神经网络实现了噪声干扰下声发射时频特征的提取与钢轨伤损的识别，并通过自编码器与无监督算法的结合，进一步降低循环神经网络识别系统的误检风险，构建出可面向实际铁路应用的伤损检测与识别系统。本项目共发表高质量科技论文12篇，其中SCI收录9篇，EI收录3篇，并培养博士后1名，博士研究生3名，硕士研究生7名，参加出国及国内会议共5人次。