A hybrid model based on convolution neural network and long short-term memory for qualitative assessment of permeable and porous concrete

Abstract

Estimating design factors like concrete strength and durability is complicated by the cement industry's practice of producing multiple grades of cement for different uses, necessitating substantial labor hours and monetary investment. The experimental findings of accelerated carbonation-induced corrosion and associated durability characteristics of concrete built with high-volume Class F Fly Ash (FA), including AC impendence, half-cell potential, water permeability, and volume of permeable voids. FA was added to ordinary portland cement at varied replacement amounts (0–70%) to create concrete specimens. The concrete specimen has been prepared by varying different proportions of water cement ratio (0.45, 0.40, and 0.35). To predict the compressive strength and carbonation level of concrete, this study presents a simulation environment based on Artificial Intelligence (AI) that makes use of input parameters such as water/cement ratio, fly-ash percentage, and time duration. Here, One-Dimensional Convolution Neural Network based Long Short-Term Memory (1D-CNN-LSTM) has been proposed for estimating the carbonation depth and compressive strength of concrete. The developed model will be compared with other state-of-the-art techniques, including DL and ML-based techniques. The obtained R2 values from the proposed 1D-CNN-LSTM regression network deliver accuracy of 80% for estimating carbonation depth and 96% for predicting compressive strength. The proposed methodology demonstrates the use of modern AI-based techniques in the actual design model and illustrates the development of DL methods such as LSTM and CNN.

Keywords

• Carbonation depth
• Concrete strength
• Convolutional neural network
• Deep learning
• Fly ash
• Long short-term memory