This paper concerns the maintenance optimization of a multistate system subject to random failure of its components occurring with different distributions with or without aging. A dynamic maintenance policy is a sequence of condition-based intervention dates and actions. A policy generates a cost that is composed of an unavailability cost when the equipment is down or in the workshop as well as repair or replacement costs for the components. The aim of this work is to compute an approximation of the minimal mean cost over all admissible policies, and propose a policy that performs as close as possible to this minimum.