Usually large complicated systems, such as weapon systems, airplanes, Mass Rapid Transit (MRT) system and high speed railway, have a product life cycle of several years or even several decades. Preventive maintenance plays a key role to keep system operate normally during the planned life cycle. Besides, the performance of maintenance operations is also heavily relied on the precision of spare part quantity determining. This research tries to find the optimal preventive maintenance policy for the mechanical system with series configuration between modules, determines the frequency of unexpected breakdowns and the related spare part quantity to maintain the system in a required level of normal operation, at the lowest cost. A dynamic mechnical reliability model is developed including preventive replacement and minimal repairs following a non-homogeneous Poisson process. Based on the trade-off analysis of average costs for performing the preventive replacement to keep system continue to operate or stoping replacement until system phase-out, a genetic algorithm is applied to find the optimal preventive replacement schedule and combination of modules under different system reliability requirements and preventive replacement cycle. A couple of numerical examples have been demonstrated. The results show that the lower system reliability level required, the longer preventive replacement cycle should be applied to find the lowest cost per unit cycle time. In addition to this, we also show that the average number of un-scheduled failures turns out to be very few while the proposed preventive replacement policy is taken.