We study an inventory system with two potential supply modes: regular order placement opportunities happen at a fixed frequency, which defines regular review cycles, while emergency orders can be placed on every period within a regular review cycle. Regular and emergency orders incur a per-unit cost while the latest incurs an additional setup cost For a regular order lead-time equal to two periods, we show that the optimal policy with respect to the expected total discounted cost, is of (s; S) type for emergency orders while the size of a regular order depends on the inventory position following a potential emergency order. Although we could not establish the optimality for regular order lead-times exceeding two periods, substantial experimental evidence supports the conjecture that the optimal policy retains the same structure. Since the optimal policy algorithm requires significant computational effort, we develop and evaluate two heuristic policies whose operational parameters can be computed with relatively small computational effort and compare them against the optimal policy in terms of implementation difficulty, speed and accuracy for an experimental design of 3888 cases. The evaluation of the proposed optimal policy and the two heuristics requires a simulation suite flexible enough to capture the specific problem dynamics. This motivates the development of the Inventory Simulator Workbench (ISW). This inventory system simulator, written in Java, provides the user with a graphical interface and the ability to model a large range of supply chain structures and inventory ordering and distribution policies.

The thesis contributes to the body of knowledge in analyzing and optimizing inventories of multiple stocking locations in a supply chain system. Optimization model is developed for planning inventories with respect to the proposed inventory-pooling strategy. The model is solved under stochastic environment using a Hybrid Genetic Algorithm technique.