高速铁路不仅适合我国人口众多、幅员辽阔的基本国情，而且符合我国可持续发展的战略要求，在近二十年的快速发展下，我国高速铁路总体技术水平已进入世界领先行列，高速铁路已经成为国民经济发展的大动脉和对外交流合作的新名片。然而随着高速铁路的飞速发展，不断提升的车速、运输量及载重使钢轨的滚动接触疲劳裂纹成为威胁高速铁路安全运营的重要问题。现有检测技术及方式存在检测速度低、无法避免检测盲区、对微小伤损不敏感的固有缺陷，不满足高速铁路钢轨探伤的需求。声发射(acoustic emission，AE)技术作为被动式动态无损检测技术，具有敏感性强、实时性好、对被测物几何要求低等优势，可用于钢轨健康监测和伤损检测。然而高速情况下较强的轮轨滚动噪声是阻碍AE技术应用于钢轨探伤的主要问题，由于缺少对轮轨滚动噪声的了解，常用的AE信号分析方法及特征不再适用，现有去噪方法的效果并不理想。因此，亟需研究针对高速铁路轮轨滚动噪声的分析及特征提取方法、理论模型和去除方法。 研究在高速轮轨滚动接触模拟实验设备和拉伸试验机上分别采集了不同车速下的轮轨滚动噪声和裂纹扩展信号，基于轮轨粗糙接触面的分形特性对两种信号进行分形分析，证明轮轨滚动噪声具有分形特性，可用分形布朗运动描述。进一步，利用小波系数方差与Hurst指数的标度关系估计两种信号的分形维，得到轮轨滚动噪声的分形维在小于2的小区域内随机分布，与车速无关，是噪声的固有特征，可用于不同车速下噪声的统一描述；裂纹扩展信号的分形维均大于2，分形维是区别二者的本质特征，可用于高速铁路钢轨伤损检测。在此基础上，从轮轨滚动噪声的产生机理出发，针对粗糙接触面统计理论和分形理论的分歧提出变形率指标，确定了统一的粗糙点临界状态表达式，并基于粗糙接触面分形理论和Hertz接触理论，建立了轮轨滚动噪声的法向能量模型。在分形理论的基础上，提出同时考虑粗糙点法向及切向弹-塑性变形和温度影响的粗糙点切向应力-应变模型，并结合Kalker轮轨蠕滑理论以及FASTSIM对轮轨接触面的网格划分，建立了粗糙点弹-塑性变形轮轨蠕滑理论以及轮轨滚动噪声的切向能量模型。能量模型确定了轮轨接触面粗糙点弹-塑性变形产生的轮轨滚动噪声功率与载重、车速、蠕滑率及粗糙接触面分形参数的关系，通过与机械密封试验和干态稳定轮轨滚动试验以及经典轮轨蠕滑理论结果的对比，验证了所建模型的正确性和有效性。最后基于轮轨滚动噪声和裂纹扩展信号的分形维区别，提出了基于分形维和自适应谱线增强的轮轨滚动噪声消除方法，将分形维作为新的指标引入自适应滤波器的代价函数，并通过对含噪信号去噪证明所提方法既能抑制高速、强轮轨滚动噪声又能增强钢轨裂纹伤损信号。 分形分析提取了轮轨滚动噪声的固有特征，分形布朗运动模型确定了不同车速下轮轨滚动噪声的分形特性、统计特性、时域特性和频谱特性，对产生轮轨滚动噪声的随机过程给出了有效的数学描述。轮轨滚动噪声能量模型明确了轮轨滚动噪声功率与各影响因素的关系，解释了轮轨滚动噪声幅值变化的原理，为高速情况下轮轨滚动噪声研究奠定了理论基础。基于分形维的自适应谱线增强去噪为高速铁路钢轨健康监测和伤损检测提供了有效方法，保证了高速强噪情况下钢轨伤损信号的获取。本研究为实现真正意义上的AE技术高速铁路钢轨探伤提供了理论基础和方法支持，使进一步的伤损信号分析和分类诊断成为可能。 关键词：高速铁路；轮轨滚动噪声；声发射；分形理论；机理建模；自适应去噪

 High-speed railway not only fits the basic domestic conditions of the large population and the vast territory, but also meet the requirement of the sustainable development strategy of our country. Under the rapid development in the recent two decades, the overall technical level of the national high-speed railway has reached an internationally leading position. In the meantime, high-speed railway has become the main artery of national economic development and a fresh-new business card in foreign exchanges and cooperation. However, along with the fast development of the high-speed railway, and due to the increasing vehicle speed, transportation, and load, the rolling contact fatigue crack of rail has become a fatal threat to the safe operation of the high-speed railway. Existing detection technologies and approaches have inherent deficiencies of low-detection speed, unavoidable dead zone, and insensitivity to subliminal defects. As a result, the rail defect detection demand for the high-speed railway is not meet. Acoustic emission (AE) technology, as a passive nondestructive detection technology, can be applied in rail health monitoring and defect detection, owing to its merits of sensitivity, instantaneity and low-geometry request of the objective. Whereas, the strong wheel-rail rolling noise (WRRN) under high speed is the primary issue that impedes the application of AE technology on rail defect detection. Because of the lack of knowledge regarding the WRRN, usual signal analysis methods and features no longer apply, and the performance of existing denoising methods are unsatisfactory as well. Therefore, there is an urgent need to investigate analysis and feature extraction methods, theoretical model and denoising approach respect to the WRRN in high-speed railway. In our research, the WRRNs at different vehicle speeds and the crack propagation signals are acquired from an experimental rig for high-speed wheel-rail rolling contact simulation and a tensile test machine, respectively. Based on the fractality of wheel-rail rough surfaces, fractal analysis is performed on the two kinds of signals. It is demonstrated that WRRN has fractality as well and it can be described by the fractional Brownian motion (FBM). Furthermore, the power-law relation between the variance of wavelet coefficients and the Hurst exponent is utilized to estimate the fractal dimensions of the two kinds of signals. The fractal dimensions of WRRNs distribute in a small region below 2 and are irrelevant to the vehicle speed, which indicates that the fractal dimension is an intrinsic feature and can be utilized in the uniform description of the WRRNs at different speeds. Fractal dimensions of crack propagation signals are all above 2,therefore the fractal dimension is a distinct feature to differ the two kinds of signals and can be taken advantage of in rail defect detection for high-speed railway. On this basis and departing from the generation mechanism of WRRN, a deformation ratio is proposed to settle the contradiction between the statistical theory and the fractal theory, then the critical status of each asperity can be described by a uniform expression. Afterward, on the basis of the fractal theory of rough surfaces in contact and the Hertzian contact theory, the normal energy model of the WRRN is established. Based on the fractal theory, a tangential stress-strain model of the asperity is proposed, which considers the elastic and plastic deformations both in the normal and the tangential directions as well as the influence of the temperature. Then the model is combined with the Kalker's wheel-rail creep theory and the mesh partition of wheel-rail contact patch by FASTSIM, and a wheel-rail creep theory based on elastic-plastic deformations of asperities and a tangential energy model of the WRRN are built. Thus, the relations between the power generated by the deformed asperities on the wheel-rail contact patch and the load, the vehicle speed, the creepage as well as the fractal characteristics of the rough surfaces are determined by the energy models. Comparing the results of a mechanical-seal test, dry and steady wheel-rail rolling test, and the classical wheel-rail creep theory, accuracy and effectiveness of the proposed models are validated. At last, on the basis of the difference between the fractal dimensions of the WRRN and the crack propagation signal, a denoising method of WRRN is proposed based on the fractal dimension and the adaptive line enhancer (ALE), where the fractal dimension is introduced as a new criterion into the cost function of the adaptive filter. Through processing the noisy signals acquired from the experimental rig, it is demonstrated that the improved adaptive denoising method not merely has the ability to suppress the strong WRRN under high speed, but also be able to enhance the rail crack defect signal. Fractal analysis extracts the intrinsic feature of WRRN. The FBM model determines the fractal, stochastic, time-domain, and frequency-domain properties of WRRNs at different vehicle speeds, offering an effective mathematical description for the stochastic process of the generation of WRRN. The energy model of WRRN specifies the relations between the power and the influence factors, and the principle of the varying amplitude of WRRN is explained. Thus, the theoretical basis of investigations on WRRN under the high-speed condition is established. The ALE denoising method improved by the fractal dimension provides an effective way for rail health monitoring and defect detection in high-speed railway, which guarantees the acquisition of rail defect signals under the condition with high speed and strong WRRN. The study offers theoretical basis and method support for a real sense of applying AE technology on high-speed rail defect detection, and enables further signal analysis and classification diagnosis of the defects. Keywords: high-speed railway, wheel-rail rolling noise, acoustic emission, fractal theory, mechanical modelling, adaptive denoising