This paper deals with a class of large-scale stochastic control problems typically associated with the maintenance of complex systems. For these systems the maintenance and replacement actions are numerous and the dynamics is complex (involving jumps and diffusions).

We propose to approach the control problem by using a multilayer control scheme. At a lower layer a regulator implements a control law picked in a class of admissible control laws by a supervisor whose interventions are triggered at controlled stopping times. The supervisor constitutes the upper-layer controller. It is shown in the paper how the control problem can be lifted to the supervisor level in the form of a semi-Markov decision process. A numerical illustration shows how the approach can be implemented.