Deep learning and supervised learning techniques for modeling and prediction of strength of ground granulated blast furnace slag based sustainable mortar

Abstract

This paper presents an analysis and synthesis of deep learning and supervised learning techniques for predicting the flexural tensile strength and compressive strength of mortars containing ground granulated blast furnace slag (GGBFS). The goal is to save time and reduce the costs associated with experimental testing. Initially, mechanical tests were conducted experimentally, using GGBFS as a partial replacement for cement in the mortar, with replacement levels ranging from 0 wt% to 80 wt% in 1% intervals. The results indicated a nearly linear decrease in flexural tensile strength and a non-linear change in compressive strength with increasing GGBFS content. To address the limitations of traditional experimental methods and improve processes, deep learning and supervised learning approaches were explored and compared. Evaluation metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), and R-squared demonstrated that the deep learning model achieved higher sensitivity and efficiency for both 7-day and 28-day flexural tensile strength, as well as 7-day and 28-day compressive strength. Quantitatively, the deep learning model outperformed traditional models for 28-day flexural tensile strength, with an RMSE of 0.6779, an R-squared value of 0.90, an MSE of 0.4596, and an MAE of 0.6045, and for 28-day compressive strength, with an RMSE of 0.6808, an R-squared value of 0.99, an MSE of 0.4636, and an MAE of 0.4891. These findings suggest that deep learning is a promising method for accurately modeling and predicting the mechanical properties of GGBFS-based mortars, particularly for the 28-day strengths.