Title: Data-Driven Reconfigurable Supply Chain Design and Inventory Control

Advisors: Dr. Alan Erera, Dr. Chelsea C. White III

Committee Members:

Dr. Alejandro Toriello

Dr. Enlu Zhou

Dr. Soumen Ghosh (Scheller College of Business)

Dr. David Goldberg (Cornell University)

Date and time: Thursday, May 3rd, 2018, 12:30 PM

Location: Groseclose 402 (Advisory Boardroom)

Abstract:

In this dissertation, we examine resource mobility in a supply chain that attempts to satisfy geographically distributed demand through resource sharing, where the resources can be inventory and manufacturing capacity.  Our objective is to examine how resource mobility, coupled with data-driven analytics, can result in supply chains that without customer service level reduction blend the advantages of distributed production-inventory systems (e.g., fast fulfillment) and centralized systems (e.g., economies of scale, less total buffer inventory, and reduced capital expenditures).

Firstly, we introduce the problem of planning the logistics of a multi-location production-inventory system equipped with transportable production capacity and propose near-optimal heuristic methodologies to effectively manage its response to uncertainty. For instances with 20 locations, the best heuristic solution cost provides 13% savings over a system with an optimal fixed capacity allocation. Greater savings result when the number of locations increases. 

In the second part, we present an analysis of a single location inventory control problem with a partially observed, demand-influencing Markov-modulation process. We present an easy-to-compute newsvendor-styled criterion that linearly partitions the modulation belief space into regions with unique optimal myopic base stock levels. We prove that the optimal myopic base stock policy is optimal for finite and infinite horizon problems when an inventory position attainability assumption holds. 

Finally, we consider the problem of planning the logistics of a multi-location production-inventory system with the options of transportability of production capacity and inventory transshipment, while facing demands influenced by a partially observed, Markov-modulation process. We propose efficient heuristic methods that result in cost savings as high as 26% on some instances with 10 locations over a system with no flexibility. Our results reinforce the value addition due to the production capacity portability, independent of the option of transshipment flexibility.