Making “Dependable Engines”: From Policy Search to Stochastic Lookaheads in Dynamic Supply Chain Planning

Abstract

Pratt & Whitney is an aerospace company based in East Hartford, Connecticut, that manufactures jet engines and turbines. To produce these highly complex and sophisticated goods, the company outsources an estimated 80% of production, which creates a flow of parts across a large network of independently operated suppliers. By their nature, these parts are highly expensive, and require complex machining that results in long production times and small yields. These factors restrict the storage of large inventories, forcing suppliers to plan weeks to months in advance against uncertainties in supply and demand.

As part of a DARPA-funded research project conducted jointly by Princeton and United Technologies researchers, this thesis seeks to address the problem described above, and model Pratt & Whitney’s supply chain operations as a sequential decision problem using a five-part modelling framework developed by Professor Warren Powell. Upon applying this framework, we will suggest several decision-making policies for solving the problem, including a simple heuristic policy frequently used in industry, as well as a more complicated lookahead policy that uses a model of the system to project inventories into the future.