Multi-objective optimization of concrete mix design based on machine learning

This study proposes a multi-objective optimization (MOO) framework for optimizing concrete mixture proportions. Advanced methods such as K-fold cross-validation, Bayesian hyperparameter optimization, regression feature elimination and C-TAEA algorithm are used to develop suitable ML models. The results show that compared to the commonly used ML algorithms, the application of 5-fold cross-validation, hyperparameter and regression feature elimination significantly improves the prediction accuracy of compressive strength to 0.993 and binder intensity to 0.980. Moreover, quantitative correlations between different mix design parameters and concrete performances can be clearly observed. By using the proposed MOO model with C-TAEA algorithm, the errors between the predicted and tested values are only around 4%, exhibiting high prediction accuracy and the ability to solve multi-objective mixture optimization problems.