Simulation of Big Data from Railway Braking

Condition Monitoring (CM) offers opportunities for improvement of safety, availability while typically also reducing maintenance cost. For some asset classes however, large up-front investments are required. Most notably, for the wagon subsystem of the freight rail system, a close monitoring of brake performance using on-board sensors for monitoring of the brake hardware using e.g. force sensor appears expensive and economically not viable. The approach outlined in this paper picks up on the approach of the author's contribution to Eurobrake 2019, where a big data approach merely based on accelerometer measurements was proposed. This yields the advantage of reducing hardware cost, both initial investment as well as maintenance cost, since accelerometers are everday electronic components which typically show a large lifespan with little degradation. This reduction in hardware cost needs to be compensated by more elaborate computational approaches, since it is based on the contribution of each individual wagon's brake to the braking performance of the entire train consist. By help of randomised mixture of the wagons during train formation, such an approach may be used to detect degradations of individual wagon brakes. Since such an approach relies on multiple train operations, it is obviously not suited for the replacement of the classical operator based pre-departure check. It may however increase safety and availability on an individual wagon basis due to the ability to observe degradation prior to becoming a functional limitation to be handled during a pre-departure check. Furthermore, it is well suited to supplement novel forms of brake assessments, such as automated brake tests or automated visual inspections and thus provide a higher level of safety and availability at reduced overall cost. The paper features a brief recap of the proposed method, integrated into the Wagon4.0 concept, which is one potential hardware and software basis as well as a review of big data approaches. The development of a simulation environment suitable for the generation of big data sets based on a headless MATLAB/Simulink setup on a cluster architecture is described. The model features an approximation of the pneumatic behaviour of the brake pipe, a simplified wheel-rail contact model, varying brake block friction as well as a parameter based model of a distributor valve for each simulated wagon. Thanks to the hardware and software setup, the simulation environment is able to generate braking data in the range of multiple Terabytes, which can then be analysed using big data approaches. Wagon parameters are sourced from a partly randomised digital wagon pool, which can be sent on virtual mission while observing their braking behaviour. The paper finishes with a preview of the intended analysis, which is currently being implemented on the FH Aachen cluster.