SURROGATE-BASED VS SIMULATION-BASED OPTIMIZATION FOR BUFFER ALLOCATION: A QUANTITATIVE ASSESSMENT

This paper presents a comparative assessment of surrogate-based and simulation-based optimization for the Buffer Allocation Problem (BAP) within a digital shadow architecture, aiming to determine the optimal configuration of a production line that minimizes total buffer capacity under a minimum requirement on system throughput. A prototype of a five-station production line serves as the physical system. An artificial neural network (ANN) is adopted as the surrogate model and trained on simulation samples to approximate system throughput as a function of buffer capacities. Results show that the surrogate-based approach achieved a 17.6% reduction in total buffer capacity (14 vs 17) compared to the simulation-based one, while cutting optimization time from hours to seconds. Validation on the physical system revealed prediction errors of 8.06% and 5.79% for the surrogate- and simulation-based approaches, respectively. The findings demonstrate that surrogate models offer a favorable trade-off between solution quality and computational efficiency for digital shadow applications in manufacturing optimization.