Drying shrinkage and crack width prediction using machine learning in mortars containing different types of industrial by-product fine aggregates

Concrete is a material that loses water and changes shape while hardening due to its structure. Over time, this water loss results in some shrinkage of the hardened concrete, referred to as drying shrinkage. In addition, water loss of concrete also causes the formation of various cracks. The aggregate used in concrete plays an important role in the shrinkage and cracking of concrete. The focus of this study is to accurately estimate the amount of crack width and drying shrinkage over time after the substitution of fine aggregates with other types of aggregates (consisting of various industrial by-products or wastes at different percentages) in the concrete mortar. For this purpose, various experimental results of the ‘substituted fine aggregate concrete mortars’ were converted into a data set. Following this a model was developed to predict the drying shrinkage and crack width of concrete mortars. The machine learning model was trained with the measurement results of 60-day drying shrinkage and crack widths of concrete mortars with different proportions of bottom ash (BA), granulated blast furnace slag (GBFS), fly ash (FA), and crushed tiles (CT). To enhance the detection/prediction capability of the model, the model hyperparameters were optimized. It is observed that the developed model was able to detect the drying shrinkage and crack width with an accuracy exceeding 99.6 %. In addition, the physical properties such as grain shape (angular or round) of components like fine aggregates may be effective for improved performance of the machine learning models in predictions of the drying shrinkage values or drying shrinkage cracking widths.