Physics-regularized neural network and multi-objective optimization for sustainable multi-binder concrete mix design
发布时间:2026-11-12 | 来源:Nature Scientific Reports | 作者:Ahmed Almutairi
This study examined multi-age compressive-strength prediction and multi-objective mixture selection for concrete containing fly ash, ground-granulated blast-fur
This study examined multi-age compressive-strength prediction and multi-objective mixture selection for concrete containing fly ash, ground-granulated blast-furnace slag, and silica fume. The dataset comprised 112 mixture-age mean strengths from 28 formulations tested at 3, 7, 28, and 90 days (336 cylinders). Complete mixture groups were assigned to training (20), validation (4), and testing (4) partitions. The physics-regularized neural network (PRNN) yielded an R² of 0.962 and an RMSE of 4.84 MPa on the 16 held-out mixture-age observations; the data-only multilayer perceptron yielded 0.790 and 11.42 MPa, respectively. Additional ridge, SVR-RBF, random forest, and gradient boosting models were evaluated using mixture-grouped validation and nine composition-and-age inputs. Random forest achieved a held-out R² of 0.743 and RMSE of 12.62 MPa; Gradient Boosting achieved 0.740 and 12.69 MPa. NSGA-III was used to examine candidate proportions under strength, cost, and global-warming-potential objectives. Three candidate laboratory mixtures yielded 90-day compressive strengths of 31.8, 41.5, and 51.2 MPa. The findings demonstrate the value of grouping formulations during evaluation and provide a comparative benchmark for predictive modeling of multi-binder concrete. Model evaluation was limited to four held-out formulations, and economic and environmental comparisons depend on their constituent inventories.
返回资讯频道