面向高速轮轨滚动系统安全预警的正则化辨识研究

The high-speed wheel/rail rolling system is an important part of train transport system. The operation quality and safety of the high-speed train are directly affected by the wheel/rail rolling system. This project intends to carry out regularization identification for security warning of high-speed wheel/rail rolling system. .Firstly, sum-of-norms SVM sparse identification method is proposed for modeling the wheel/rail fatigue crack evolution behavior. For the formulated fatigue crack evolution model, the scientific problems, such as the solution of high-dimensional nonlinear programming problem, sparse feature extraction of crack evolution, computation of regularization factor, the crack evolution trend forecasting, will be studied in this project. And the initiation and propagation mechanism of wheel/ rail fatigue crack will also be expounded..Secondly, manifold weighted smoothing identification method is proposed for wheel/rail wear regression model. For the formulated wheel/rail wear model, the scientific problems, such as estimation of semi-supervised non-parametric regression function , trade-off between model fitting accuracy and generalization performance, the calculation of the weighted coefficient, will be considered in this project. And the obtained identification model is used to reveal the wheel/ rail wear mechanism..Thirdly, maximum likelihood identification method is proposed for wheel/rail contact generalized state space model. For the formulated wheel/rail contact model, the scientific problems, such as the realization of expectation maximization algorithm for maximum likelihood identification problem, estimation of the state conditional probability density, design of parameter search algorithm, will be studied in our work. .Fourthly, model validation methods, which are suitable for wheel/rail identification model, will be studied in this project. .Finally, service simulation of wheel/rail rolling system will be implemented in our work..In a word, the research of this project will provide theoretical support for security warning of high-speed wheel/rail rolling system.

高速轮轨滚动系统是列车运输系统的重要组成部分，轮轨滚动系统直接影响高速列车运行品质和安全性。本项目开展面向高速轮轨滚动系统安全预警的正则化辨识研究。提出轮轨疲劳裂纹演化模型的范和SVM稀疏辨识方法。针对构建的裂纹演化模型研究高维非线性规划问题的求解、裂纹演化稀疏特征的提取、正则化因子的计算、裂纹演化趋势的预报等问题，阐明轮轨疲劳裂纹萌生和扩展的机理。提出轮轨磨损回归模型的流形加权平滑辨识方法。围绕构建的磨损模型研究半监督非参数回归函数的估计、模型拟合精度与泛化性能的折中、加权系数的计算等问题，揭示轮轨磨损机理。提出轮轨接触广义状态空间模型的极大似然辨识方法。针对构建的轮轨接触模型研究极大似然辨识问题的期望最大算法实现、状态条件概率密度的估计、参数搜索算法的设计等问题。研究适合于轮轨辨识模型的有效性检验方法。开展轮轨滚动系统的服役模拟仿真研究。课题研究将为高速轮轨系统的安全预警提供理论支撑。

高速轮轨滚动系统是列车运输系统的重要组成部分，轮轨滚动系统直接影响高速列车运行品质和安全性。本项目开展面向高速轮轨滚动系统安全预警的正则化辨识研究。提出轮轨疲劳裂纹演化模型的范和SVM稀疏辨识方法。针对构建的裂纹演化模型研究高维非线性规划问题的求解、裂纹演化稀疏特征的提取、正则化因子的计算、裂纹演化趋势的预报等问题，阐明轮轨疲劳裂纹萌生和扩展的机理。提出轮轨磨损回归模型的流形加权平滑辨识方法。围绕构建的磨损模型研究半监督非参数回归函数的估计、模型拟合精度与泛化性能的折中、加权系数的计算等问题，揭示轮轨磨损机理。提出轮轨接触广义状态空间模型的极大似然辨识方法。针对构建的轮轨接触模型研究极大似然辨识问题的期望最大算法实现、状态条件概率密度的估计、参数搜索算法的设计等问题。研究适合于轮轨辨识模型的有效性检验方法。开展轮轨滚动系统的服役模拟仿真研究。.提出基于分布式ICA-PCA( independent component analysis-principal component analysis)模型的故障监测方法，适合于轮轨滚动系统难以自动划分子块及过程数据存在非高斯信息的情况。首先，对过程数据进行PCA分解，并在PCA主成分不同的方向上构建不同的子块，把原始特征空间自动划分为不同子空间。然后，对各个子块采用ICA-PCA两步信息提取的策略，提取出高斯信息和非高斯信息，并构建新的统计量和统计限。.课题研究将为高速轮轨系统的安全预警提供理论支撑。