Deep neural network-enhanced prediction and carbon footprint analysis of early-age high-performance manufactured sand concrete's stress-strain behavior

Abstract This study aims to investigate the stress–strain curve relationship and construction efficiency of manufactured sand concrete (MSC) at early ages thr ough uniaxial compression tests conducted on 216 specimens at ages of 2 d, 3 d, 4 d, 5 d, 6 d, 7 d, 14 d, and 28 d. The research examines the influence of age, water-cement ratio, and fly ash content on the peak stress and peak str ain of MSC. The results suggest that the variation pattern of characteristic poin ts on the stress-strain curve of MSC is influenced by the age. Within the 2-d to 4-d age range, the characteristic points on the stress-strain curve of MSC de monstrate the most prominent alterations. In the age range of 5 d to 7 d, the hydration reactions have already been completed, in contrast to the 2-d to 4-d age range. Consequently, the rate of change in characteristic points on the curv e becomes relatively gradual. During the extended period from 14 d to 28 d, t he rate of change in characteristic points on the curve exhibits no significant d ifference compared to other periods. This study evaluates the carbon emissions of MSC as a sustainable building material over its entire lifecycle and conclud es that the replacement of supplementary cementitious material (SCM) for ceme nt is essential for emission reduction. Furthermore, a deep neural network (DN N) model with four hidden layers and 100 neurons in each layer was develope d based on experimental results. The model was trained to predict the stress–st rain curves of MSC under varying water cement ratios, ages, and fly ash conte nt. DNN model was trained and validated through pre-processing and segmenti ng the original dataset using Pytorch deep learning (DL) libraries. DNN model accurately predicts stress–strain curves that closely align with MSC test curves.