Low-carbon and low-cost optimization framework of concrete under chloride environments with text-enhanced deep learning

Existing low-carbon concrete optimization approaches typically treat performance indicators such as compressive strength and durability as optimization objectives. However, in engineering practice, these indicators are required only to meet design code specifications and service-life requirements rather than to be continuously improved. This study proposes a multi-objective optimization framework for concrete under chloride environments, in which compressive strength and chloride diffusivity are treated as constraints rather than optimization objectives, while carbon emissions and cost are minimized as the primary objectives. In this framework, deep neural network (DNN) was employed to predict the compressive strength of concrete. Given the complexity of factors influencing chloride diffusion in concrete, including material properties and environmental conditions that are difficult to quantify and often described in textual form, this study developed a text-enhanced deep learning model to accurately predict the chloride diffusivity. The results demonstrate that both models can effectively capture the nonlinear relationships between the input variables and compressive strength/chloride diffusivity. Subsequently, the constraint-handled non-dominated sorting genetic algorithm II (NSGA-II) and technique for order preference by similarity to ideal solution (TOPSIS) were adopted to obtain the optimal solution. The optimization results of the concrete with low-carbon raw materials indicate that reductions of 20.35% in carbon emissions and 12.18% in cost can be achieved. In the benchmark case based on pier concrete from the Hong Kong-Zhuhai-Macao Bridge, the optimized concrete mix can reduce carbon emissions and cost by 17.34% and 3.00%, respectively.