数据驱动的高速铁路轨道几何状态劣化机理与预测方法

The research objective of this project is to diagnose and predict the high-speed rail geometry degradation through two parallel technique trajectories: “Big Data Computing” and “Physical-based Modeling”. In the trajectory of “Big Data Computing”, this project will carry out field studies, summarize and analyze the characteristics and causalities, and propose the methods and standards to classify geometry irregularities of high-speed rail，supported by railway agencies. This research also adopts crowdsourced data from onboard smartphones to identify the event and the type of rail geometry irregularities. In addition, this project will further enhance the missing data filling theory through tensor decomposition. Specifically, block-wise missing rail data will be interpolated with the consideration of spatial-temporal relationship and constraints. Finally, a variety of measurement data will be fused via coupled tensor decomposition. The coupled factor will be found and provided as critical influential information for forecasting rail geometry irregularities. In the trajectory of “Physical-based Modeling”, this project will select several classical types of high-speed rail geometry irregularities, build state transition equations via the plastic theory for track settlement. A particle-based filter will be established within a Bayesian framework. Such filter allows to combine observations from big data and state transition via physical-based models within the discrete-time state-space equation, and build a hybrid prediction model for the remaining useful life of rail geometry irregularities. This project will eventually provide solutions for high-speed rail safe service performance monitoring.

本项目采用“大数据计算”与“物理建模”双线并行的技术路线，研究高速铁路几何状态的劣化与预测。在“大数据”路线上，首先利用路局支持开展调研，统计分析各类不平顺的特性与成因，提出高速铁路不平顺的分类方法与标准。基于车载手机数据，提出众筹数据的轨道不平顺事件与类别的确认方法。针对轨道检测数据，进一步通过时空关联张量分解完善数据的补齐理论，对缺失的大块轨道测量数据进行补齐。最后采用耦合张量分解，对各种测量数据进行融合，找到各数据的耦合因子，对不平顺预测提供准确的影响因子。在“物理建模”路线上，考虑几种典型的高速铁路不平顺类别，利用塑性理论推导轨道下沉的物理状态转移公式，通过数据驱动参数采样与仿真进行模型选择，构建一个粒子过滤器状态空间模型。此过滤器结合观测的数据与物理模型，在贝叶斯框架中构建轨道裂化的混合预测模型并估计剩余服役寿命。该项目最终为高速轨道不平顺安全服役性能监控提供解决方案。

高速铁路轨道几何的不平顺性会降低列车运行平稳性，严重时危害行车安全，同时存在性能劣化速率较快、检测耗时费力、评价标准不完善、维修效率低下等一系列问题。因此，课题科研团队采用了“大数据计算“和“物理建模”双线并行的技术路线，应用理论研究、仿真分析、现场试验相结合的方法，开展了轨道不平顺车载检测研究，建立了轨道不平顺动、静态运维管理指标评价方法，提出了无砟轨道精调方法优化模型，取得的成果包括：（1）基于自主研发的车载多源数据采集装置，建立了车体振噪数据和轨道不平顺之间的正逆向映射关系；（2）提出轨道不平顺动、静态管理指标关联分析模型和评价方法，实现动、静态数据里程对齐与轨道不平顺管理值优化；（3）建立用于轨道不平顺机理分析和维修控制的数据物理融合模型，实现车辆-轨道动力学高效评估和轨道不平顺精调方案优化。以上的研究成果为我国高速铁路智能运维领域，尤其是车载式便携式检测、轨道服役状态评价、养护维修方面提供了指导方法。