Approximate Dynamic Programming for a Remanufacture-to-Order System

TL;DR

The paper addresses remanufacture-to-order (RtO) systems under uncertain core quality and stochastic demand. It extends prior work by developing an approximate dynamic programming (ADP) framework with a linear value function approximation and an approximate policy iteration scheme to solve large RtO instances, building on a Markov decision process formulation where core quality is revealed upon inspection. The key contributions are demonstrating that a policy which fulfills demand whenever cores are available and prioritizes higher-quality cores remains optimal-like in larger settings, and showing that an order-up-to-like threshold persists under the ADP scheme; the approach scales to more core types than previous studies. The results highlight the potential of ADP to address remanufacturing planning problems, providing actionable insights for core acquisition and demand fulfillment in a sustainable manufacturing context.

Abstract

Remanufacturing is pivotal in transitioning to more sustainable economies. While industry evidence highlights its vast market potential and economic and environmental benefits, remanufacturing remains underexplored in theoretical research. This study revisits a state-of-the-art remanufacture-to-order (RtO) system and develops an alternative approach by developing an approximate dynamic programming (ADP) algorithm to solve larger RtO instances. The proposed methodology yields results consistent with existing state-dependent, non-congestive policies. Key findings include the optimality of fulfilling demand whenever remanufacturable cores are available and prioritizing cores of the highest quality level. This work highlights the potential of ADP in addressing problems in remanufacturing domains.

Figures (2)

• Figure 1: Average values of $\theta_i$ coefficients for core types $i \in \mathcal{K}$
• Figure 2: Expected $T_1$ operator costs associated with different total core inventory levels