The heavy maintenance centers of SNCF carry out heavy maintenance of rolling stock (levels 4 and 5), including component repairs. The operating procedures vary from

one rolling stock unit to another, and most operations are carried out by maintenance agents. The uncertainties encountered in this process are numerous

(additional work, uncertain estimation of operating times, etc.).

This thesis focuses on modeling and solving the scheduling problem of maintenance operations, considering various uncertainties. The problem is modeled as a

Multi-Skill Resource-Constrained Multi-Project Scheduling Problem. A mathematical formulation is proposed, and a new memetic algorithm is implemented to solve large-scale

industrial instances. For the deterministic problem, two objective functions are independently considered: (i) Minimization of the sum of the weighted tardiness of the projects

and (ii) Minimization of the sum of the weighted duration of the projects. The memetic algorithm is compared to two heuristics and three metaheuristics. The effectiveness of the

proposed approach is validated on both industrial instances and instances from the literature.

The memetic algorithm is extended to integrate uncertainties and maximize the sum of the weighted tardiness service level of the projects. This criterion corresponds to maximizing

the probabilities of meeting customer deadlines. A new scenario generation approach, limited by different uncertainty budgets, is proposed. To generate these scenarios,

real historical data are used. The relevance of our robust scheduling approach is validated on ten industrial instances.

The research conducted in this thesis is industrialized, and a decision support tool is implemented for the heavy maintenance centers of SNCF.

Jury: