To ensure optimum system safety and availability at the minimized direct maintenance cost, fleet management decision making must be based on accurate forecasts of component reliability and proper maintenance policy. With the objective of minimizing maintenance cost while satisfying system safety and availability requirements, this paper presents an approach integrating Quantitative Risk Assessment (QRA) and non-linear programming concept to minimize the expected maintenance cost subject to various safety/reliability requirements. The approach begins with decomposition of a complex system into simpler sub-systems where hardware components are grouped together either based on logistics convenience or on common functions. Reliability Block Diagrams will be used to model each sub-system availability and reliability. Life data analysis is performed on each component and its statistical model is hypothesized, reviewed, and evaluated. Optimal maintenance policy using Non-Linear Programming concept is then derived for each component based on cost, reliability and component safety. Optimized results at the component level are then rolled up to the sub-system level. However, a local minimum might not lead to global minimum so search algorithms will be applied at a higher level in order to arrive at a global minimum for cost. Finally, there is a discussion of various simulation tools, the limitations and future activities that will account for uncertainties in the analysis and their impacts on optimal maintenance policy.