HPASC – OPCC bi-surface shear strength prediction model using deep learning

Abstract

ABSTRACT This research focused on high-performance alkali-activated slag concrete (HPASC) as an overlay material on ordinary Portland cement concrete (OPCC). HPASC specimens were designed with different NaOH molarity, silica fume (SF) content, steel fiber content, age of repair material, and proportion of grooved surface area to the gross area. The results indicated that the steel fibers in HPASC improved the compressive strength and increased the BSS strength. The convolutional neural network (CNN) and multilayered perceptron (MLP) were used to determine a comprehensive model for simulating the effect of each parameter on the BSS shear strength. The simulation results suggest that the CNN, as compared to MLP, can more accurately predict experimental data due to the less MSE and higher R-factor. According to the conducted simulations, HPASC mixed with 2% fibers, 12 molars NaOH solution, 20% SF content, and 30% grooved surface was the most suitable mixture for repair and rehabilitation.