A New Approach for Large Non-Linear Integer Optimization Suitable for Implementation on a Distributed Collection of Computers
Abstract We present a new methodology called Multi-Indenture, Multi-Echelon Readiness-Based Sparing (MIMERBS) for solving large, non-linear integer optimization problems that arise in determining the retail and wholesale sparing policies that support the aircraft operating from a deployed aircraft carrier. MIMERBS determines the minimum cost mix of spare parts that meets required levels of expected aircraft availability. The size (thousands of variables), the nonlinear relationship between spare parts and aircraft availability, and the requirement that the variables be integers make this problem hard. We provide a concise description of the MIMERBS model and present data to show how it improves on earlier sparing models. This improvement comes at the price of significant computationally complexity, which in turn makes the optimization problem hard to solve. We describe how we integrated an interior point method with a direct search algorithm to solve this optimization problem. This hybrid algorithm is well suited for implementation on a home-made virtual super-computer made up of several dozen Windows NT computers connected by an office LAN. A description of the virtual super-computer is given in a separate paper. We report on three specific cases we solved using the MIMERBS model, having from 1,000 to 8,000 optimization variables.