Multi-objective Optimization of the Concrete Mixture Blended with Mineral Admixture Using Machine Learning and NSGA-II Algorithms

Abstract

The trade-off between all the competing objectives is necessary for the multi-objective concrete mix optimization. For this purpose, a multi-objective optimization (MOO) method is needed which can generate Pareto optimal solutions. In this study, a MOO method based on non-dominated sorting genetic algorithm II (NSGA-II) and a hyper parameter tuned machine learning is proposed. Firstly, all the objectives are formulated using a machine learning algorithm called random forest regressor. To optimize the concrete mixture, ingredients are considered as variables of the objective functions, then NSGA-II algorithm is used to optimize concrete mixture proportion to get optimal combination of ingredients. The results of the proposed algorithm show that model can be successfully employed before the construction phase as guide to get optimal concrete mix proportion blended with mineral admixtures. Required trade-off between objectives have been successfully obtained using proposed MOO model known as Pareto front solutions. It has been also found that the hyper parameters tuning even boosts the performance of the machine learning model while defining objective functions. Proposed model can be used at the construction area before preparation of concrete where it can propose optimal concrete mixture proportion for the selected objectives.