Passenger trains have a very precise schedule due to the transportation demand andrailway systems aim to exploit rolling stocks to their maximum capacity. Maintaining ahealthy network of rolling stocks can be really difficult because it must rely on an effec-tive maintenance schedule that does not impact the transportation plan. But while somemaintenance operations are known beforehand, some repairing that could not have beenpredicted still needs to be done. These jobs are brought to our knowledge through the trainitself. The time allowed to fix these malfunctions is relatively short (from a few hours to afew days). It is allowed to schedule a complete repair, or a partial repair named diagnosisthat ensures that the train can be used in normal condition even if the operation is notcompletely done. The aim of our study is to find an efficient way to schedule the startingtimes of the maintenance jobs, completely or not, so that their due dates are met.Section 2 defines the considered problem, next presents a Mixed Integer Linear Pro-gramming (MILP) model and a Constraint Programming (CP) model. Section 3 introducestwo local search heuristics based on these models and Section 4 provides an overview ofthe computational results.