Solver-based heuristics for scheduling railway rolling stock corrective and predictive maintenance

Rolling stock units are critical resources because their schedule is constrained, and they are not available on demand. Given that all rolling stock units in the network share the same resources, such as tracks and infrastructure, the time and place to repair malfunctions while keeping the network in its optimal state are not easy to find. The goal of this research is to find an efficient solution to this scheduling problem. We define the problem as it occurs at the French national railway company and propose modeling it using a mixed-integer programming (MIP) model and a constraint programming (CP) model. We then solve the models using heuristic algorithms: a local branching heuristic for the MIP model and a variable partitioning local search heuristic for the CP model. We then evaluate these solutions on both generated and real data.