Efficient Sensor-Based Condition Monitoring Methodology for the Detection and Localization of Faults on the Railway Track (ConMoRAIL)

该项目的目的是开发一种有效的轨道故障检测方法，以支持智能的、基于条件的维护规划，以防止基础设施损坏，同时提高安全性和降低维护成本。该监测系统应具有成本效益、车板自主和免许可(不需要德国铁路监管机构的特别授权)，并可安装在正常服务期间使用的车辆上，以便能够连续记录轨道状况。连续的测量将允许开发科学的方法，这些方法将用于定义轨道质量，通过使用机器学习(ML)算法检测和分类铁路缺陷。此外，相同的多传感器系统将协同使用，以有效地定位并在空间和时间上分离已识别的缺陷。本研究项目的主要目标是：目标1：分析车辆在正常服务过程中获得的惯性数据，以检测和识别轨道几何不平顺。从实验室规模缩小的模型中转移知识，以检测和识别跟踪结构缺陷。考虑车辆的动态行为及其对采集信号的影响。这种方法将包括ML的算法以及动力系统和控制理论，以开发一种混合处理方法。目标2：考虑时间相关性的运动学传感器融合，使用一种类似形状过滤器的Unscented卡尔曼滤波扩展。以数字航迹图为基础，用非线性随机等式和不等条件约束传感器融合。此外，数字轨道地图将根据时刻表以及最高速度和平均速度进行完善和更新。这不仅对于可靠、准确地定位某一断层是必要的，而且对于重叠断层的空间分离也是必要的。目标3：集成、测试和评估定位系统和故障诊断算法，以进行铁轨故障检测和有效定位。根据定位的准确性和可靠性以及故障密度以及其他统计和动态参数制定总体质量指标，以便快速、有效地了解轨道质量和故障定位性能。

The aim of this project is to develop a methodology for efficient track fault detection to support intelligent, condition-based maintenance planning that will prevent infrastructure damage while increasing safety and reducing maintenance costs. The monitoring system should be cost-effective, board-autonomous and permit-free (it should not require special authorization from the German railway regulatory authorities) and can be installed on vehicles for use during regular service, so that continuous recording of the track condition is possible. The continuous measurements will allow for the development of scientific methods that will be used to define the quality of the track, detect and classify railway defects by using Machine Learning (ML) algorithms. Additionally, the same multi-sensor system will synergistically be used to efficiently localize and spatially and temporally separate the identified defects. The main goals of this research project are: Goal 1: Analysis of inertial data coming from vehicles during regular service for detection and identification of track geometrical irregularities. Knowledge transfer from laboratory down-scaled models to detect and identify track structural defects. Consideration of the vehicle’s dynamic behavior and its impact on the collected signals. This approach will include algorithms from ML as well as dynamical systems and control theory to develop a hybrid processing methodology. Goal 2: Kinematic sensor fusion considering the temporal correlations using a shape-filter-like Unscented Kalman Filter extension. Constraining the sensor fusion by non-linear stochastic equality and inequality conditions based on a digital track map. Furthermore, the digital track map will be refined and updated with schedules as well as maximum and average speeds. This is not only necessary for a reliable and accurate localization of a certain fault, but also for the spatial separation of overlapping faults. Goal 3: Integration, testing and evaluation of the positioning system and fault diagnosis algorithms for railway track fault detection and efficient localization. Development of an overall quality indicator dependent on the accuracy and reliability of the positioning, as well as fault density and further statistical and dynamic parameters, which shall give fast and efficient insight into the track quality and fault location performance.