The effectiveness of a maintenance policy for a production system is generally measured by its availability. Nevertheless, for some production systems, the failure is a dangerous event for people or in any case too much expensive; therefore the failure probability must be maintained less than an accepted low value by a proper preventive maintenance policy. The system considered hereafter is schematized by a series of components some of them constituted by parallel-series sub-systems. The problem is to individuate the set of elements on which to operate the maintenance actions, providing the system functioning with the required reliability until the next planned maintenance instant by the minimum maintenance cost. Moreover, the optimal size of the maintenance crews is individuated. The problem is formulated as a mathematical programming model that becomes very hard to solve when the number of parallel-series components increases. In order to solve large practical problem, a genetic approach is proposed and tested on a case study comparing the results with those obtained solving the mathematical model by the commercial software.