Predicting crack width after self-healing for experimental concrete mixture samples containing environmentally friendly supplementary cementitious materials
摘要
Self-healing concrete, which repairs cracks through mechanisms like cement hydration and calcium carbonate precipitation, is enhanced with materials such as fibers and microorganisms. Recent advancements use eco-friendly SCMs like Fly Ash and Silica Fume. Due to variability in concrete composition, machine learning is increasingly used to predict concrete properties. In this study, Random Forest Regression (RFR), Extreme Gradient Boosting Regression (XGBR), Stochastic Forest (SFO), and Stochastic Gradient Boosting (SGB). Optimization techniques like the Dung beetle optimization (DBO), Greylag Goose Optimization (GGO), and sled dog optimizer (SDO) fine-tune models’ hyperparameters for improved performance, Dempster Shafer Theory (DST) assist in ensemble predictions and the Fourier Amplitude Sensitivity Test (FAST) and Analysis of Variance (ANOVA) methods are used for sensitivity analysis. The results of the study revealed that among the models evaluated, XGDB (XGBR optimized by DBO) delivered the highest performance, achieving an RMSE of