Multi-Target Particle Swarm Optimization with Machine Learning Surrogates for Efficient Concrete Mix Design
Malik Mushthofa (Universitas Diponegoro, Universitas Islam Indonesia)
John Thedy (National Taiwan University, Universitas Diponegoro)
Han Ay Lie (Universitas Diponegoro)
undefined Purwanto (Universitas Diponegoro)
Marc Ottelé (TU Delft - Materials and Environment)
Mochamad Teguh (Universitas Islam Indonesia)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
This study presents a multi-target particle swarm optimization (MT-PSO) approach for efficient concrete mix design. It simultaneously designs mixes with multiple predefined strengths under a constant water-cement ratio. A gradient boosting-based surrogate model, trained on experimental mix data, predicts compressive strength. The modified particle swarm optimization (PSO) algorithm accommodates multiple targets in parallel, allowing solution sharing across target groups. MT-PSO is compared with a repeated PSO (R-PSO) strategy that optimizes each target separately, both minimizing the absolute error between predicted and desired strengths. Across 30 independent trials, MT-PSO consistently achieves lower mean errors, smaller deviations, and faster convergence, often reaching R-PSO's final accuracy within only a few iterations. Moreover, MT-PSO requires over 85% fewer fitness evaluations. These results demonstrate the superior accuracy, robustness, and computational efficiency of MT-PSO for multi-target optimization problems.
help_outline