Research Projects
Regional Fulfillment Network Design
Amazon recently reevaluated how the US fulfillment network was organized and made significant structural changes to deliver lower costs and faster speed for many years to come. Until recently, the company operated one national US fulfillment network that distributed inventory from fulfillment centers spread across the entire country. If a local fulfillment center didn’t have the product a customer ordered, it ended up shipping from other parts of the country, increasing the cost to serve and delivery times. Amazon recently started rearchitecting the inventory placement strategy and leveraging the larger fulfillment center footprint to move from a national fulfillment network to a regionalized network model. It made significant changes to create eight interconnected regions in smaller geographic areas. Each of these regions has a broad, relevant selection to operate in a largely self-sufficient way while still being able to ship nationally when necessary. Shorter travel distances mean lower costs to serve, less environmental impact, and customers getting their orders faster. This project focuses on building a rigorous optimization framework for designing the regions.
Hyperconnected Network Design
The Physical Internet (PI) presents a transformative vision for logistics systems, where assets are shared openly, and flow consolidation is achieved through standardization, modularization, interfaces, and protocols. Hyperconnected logistics networks have emerged as a promising implementation of the PI, leveraging multi-tier meshed hubs and interconnectivity to achieve greater efficiency, resilience, and sustainability in the transportation of physical goods. However, a lack of clarity in the literature regarding the definition and design of hyperconnected logistics networks presents a significant obstacle to realizing their full potential. To address this gap, we propose a comprehensive definitional framework integrating key concepts such as tiered network topology, hub interconnectivity, consolidation, and containerization. Moreover, we present a practical design approach for a hyperconnected logistics network in the United States, utilizing a representative demand scenario and accompanying network visualizations to enhance comprehension. Our research aims to unlock the potential of hyperconnected logistics networks as a crucial component of the PI, offering significant benefits to the global logistics industry and society as a whole.
Conference paper: IPIC 2023 paper
Middle-mile Network Design with Routing Constraints
The tactical decision of deciding the connectivity of middle-mile logistics hubs with last-mile hubs (nodes) depends on the operational decision of how the vehicles are routed through the nodes. When the demand of the nodes is less than a truckload, making such connectivity decisions without modeling vehicle routing constraints such as time windows and vehicle capacities may result in sub-optimal or infeasible solutions. We model the problem as a multi-depot vehicle routing problem with time windows, as the nodes have time windows in which volume can be delivered.
We approached the problem using column generation, where the pricing problem is an elementary shortest-path problem with resource constraints. We solve the pricing problem using an exact approach with the Pulse algorithm. This problem is particularly challenging due to the size of the real-life instances we used, even for one demand scenario. We need to decide the assignments of nodes to depots, which optimize for total routing cost over different demand scenarios. We learned that assigning nodes to depots such that the assignments are contiguous and feasible for vehicle routing is a good heuristic for the practical problem with multiple demand scenarios.
Parcel Consolidation in Modular Handling Containers
In Physical Internet-based hyperconnected logistics, parcels are dynamically consolidated in modular containers and routed through a multi-tier meshed network of logistics hubs. The dynamic optimization of parcel loading in modular handling containers at a logistics hub encapsulated in transport containers is particularly important. We use mathematical programming alongside analytical solutions and heuristics developed for several problem variants. Our models minimize induced costs and environmental impact, accounting for operational constraints to ensure timely departure and improve packing and handling efficiency. We rigorously account for each parcel's dimensions, expected arrival time, target departure time, and sequence of hubs to the final destination. We also account for the sizes and availability of handling containers and carriers. Leveraging our collaborative research with a major logistics service provider, we provide an empirical assessment of our approach.
Conference paper: IISE paper
Conference poster: IPIC poster
Please reach out to me if you are interested in learning more about any of the above projects.