Machine Learning‐Assisted Optimization for Dynamic Supply Network Design in the Additive Manufacturing Platform

Distributed manufacturing enabled by additive manufacturing (AM) allows supply chains to rapidly adapt to changing market conditions through flexible facility relocation and capacity adjustment. However, existing studies largely overlook the logistical aspects of AM‐based supply chains and typically treat facility location and logistics decisions separately. This study addresses this gap by developing a novel AM supply network design model that simultaneously considers the dynamic relocation of AM facilities, the adjustment of production capacities, and delivery routing decisions over multiple periods. We formulate the problem as a mixed‐integer linear programming model (MILP). Given the computational challenges associated with the delivery routing component, we propose a machine learning genetic algorithm (MLGA) approach to solve the problem efficiently. MLGA incorporates a machine learning surrogate to estimate the cost of the capacitated vehicle routing subproblem (CVRP), thereby significantly accelerating the convergence of the genetic algorithm (GA). We train two different machine learning surrogates: One based on a neural network model and the other on an XGBoost model. The MLGA solution methodology, under both surrogate models, performs satisfactorily in terms of solution quality and computation time across a set of randomly generated instances. Extensive sensitivity analyses are also conducted on a realistic use case to evaluate the model's applicability under varying service area, demand, and cost scenarios, offering valuable managerial insights and practical implications.