Integrated analytical performance evaluation models of warehouse

Warehouse design process is a complex process with numerous alternatives at all design stages, with focus on throughput capacity, inventory size and material handling equipment requirements. Enumerating all feasible solutions that satisfy the throughput and storage capacity requirements is not practical. Analytical models play a key role in the preliminary design stages in identifying several good initial warehouse configurations. This research effort pertains to the development of integrated analytical models that address capacity/congestion and inventory issues simultaneously in warehouse systems.;The first part of the dissertation focuses on the development of a queueing network model of the "shared-server system," which is an inventory store with a server performing both storage and retrieval operations. First, we modeled the shared-server system using Continuous Time Markov Chains (CTMC) under exponential assumptions. We then developed an approximate queueing network model for general arrivals and general service time distribution, and designed a solution procedure based on the parametric-decomposition method. Later, we extended these models to include multi-server cases.;The second part of the dissertation focuses on the development of a queueing-inventory (QI) model of an order-picking system. The configuration of the unit-load that is stored (pallets) is different from that which is retrieved (cases). We developed a single stage QI model with batch processing to represent the material movement in and out of the forward inventory store. We then extended these models to include multi-server cases.;The last part of the dissertation focuses on the development of an integrated model that demonstrates the applicability of these key building blocks (the shared-server system and the order-picking system) in developing an end-to-end model of the warehouse system. Extensive numerical experiments indicate that the proposed analytical models can be solved in a computationally efficient manner and are accurate for a wide range of parameter values when compared with simulation estimates.