Optimization-based multitarget stacked machine-learning model for estimating mechanical properties of conventional and fiber-reinforced preplaced aggregate concrete
Michael Saleh, Farzin Kazemi, Hakim S. Abdelgader, Haytham F. Isleem
Abstract
Abstract Nowadays, using advanced structural materials such as preplaced aggregate concrete (PAC) and fiber-reinforced preplaced aggregate concrete (FR-PAC) are widely investigated due to their benefits in designing infrastructures. Therefore, finding the mechanical characteristics of PAC and FR-PAC can be help structural engineers. This study explores the material characteristics, performance, and potential challenges associated with using PAC and FR-PAC, aiming to provide insights into their practical implementation and long-term benefits in construction. In addition, a superior estimation tool based on multi-target stacked machine-learning (ML) model was introduced to reduce the cost of experimental tests and increase the accuracy and speed of finding the best mixture for PAC and FR-PAC. Experimental tests were conducted to prepare unseen dataset to validate the general ability of the ML models. Results show that the proposed multi-target stacked ML models can estimate the compressive and tensile strengths of PAC specimens with an accuracy of 97.4% and 94.7%, respectively; however, for compressive, flexural, and tensile strengths FR-PAC specimens, the accuracy of 97.7%, 98.0% and 98.3%, were determined, respectively. The proposed predictive model was turned into a graphical user interface (GUI) with ability on predicting the mechanical properties of PAC and FR-PAC in different curing day, and updating the database in future.