This work is concerned with the problem of optimizing maintenance policies in terms of economical rewards and availability. We consider a system with multiple states in terms of healthy mode (good state, degraded state and failure state) and maintenance action (running state, stopped for maintenance). The level of maintenance (perfect or not) is also taken into account. We propose semi-Markovian model highlighting the effects of dwell times and transitions on economical rewards. We determine an optimal policy conditionally upon the current state according to eight decision parameters related to time intervals between two preventive maintenances and the level of maintenance. We show through a sensitive analysis that decision parameters have nonlocal effects that imply a multiple objective function. Hence, we propose a compromise by optimizing the asymptotic average reward.