Optimization of supplementary cementitious concrete mixes for marine structures using a one-dimensional convolutional neural network under strength and cost constraints

Abstract

Concrete used in marine infrastructure must satisfy not only compressive strength and durability requirements but also practical cost constraints. Conventional mix design methods usually rely on empirical formulas and repeated laboratory trials, which are time-consuming and often unable to fully capture the nonlinear relationships between material constituents and concrete strength. This study proposes a framework for optimizing supplementary cementitious concrete mixtures for marine applications using a machine learning, one-dimensional Convolutional Neural Network (CNN-1D), under strength and cost constraints. The results showed that the best CNN-1D configurations achieved a coefficient of determination (R²) of up to 0.8999 on the testing set and 0.8273 on the validation set. From the generated search space of 1,312,200 candidate mixtures, the proposed procedure identified mix designs with the lowest material cost while still meeting the required compressive strength. These findings indicate that a one-dimensional convolutional neural network can serve as an effective surrogate model for compressive strength prediction and cost-oriented mix design optimization of supplementary cementitious concrete for marine applications.