Original scientific paper
https://doi.org/10.15255/KUI.2023.063
AI-supported Causal Analysis and Optimisation of Sustainable Concrete Mixture Compositions
Želimir Kurtanjek
orcid.org/0000-0001-5453-6255
; University of Zagreb Faculty of Food Technology and Biotechnology, Pierrotijeva 6, 10 000 Zagreb, Croatia
*
* Corresponding author.
Abstract
A causal analysis of the effects of environmentally sustainable cement mixtures on the compressive strength of concrete and the reduction of CO2 emissions was conducted in this study. A Bayesian causal model, decision tree sets, and deep neural networks were applied. The model is based on a large dataset with a sample size of n = 1030, and p = 9 composition variables: Portland cement, slag, fly ash, water, plasticiser, coarse gravel, fine gravel, time, and compressive strength of concrete. The model is a directed acyclic graph (DAG) determined by a heuristic procedure for optimising the Bayesian Information Criterion (BIC). The resulting AI model, using machine learning, enables the prediction of the compressive strength of concrete with an average absolute error of 3 MPa (4.3 %) compared to an error of 10 MPa of the multiple linear model. To eliminate interfering effects among variables, a criterion of directed separation (d-separation) was applied to determine the causal effects of individual variables on the compressive strength of concrete. These effects are expressed as average treatment effects (ATE). The analysis of the causal effect of time reveals a two-phase, zero-order kinetics dynamics. The highest ATE values (MPa/kg m–3) during the first phase of the process were: coarse gravel 0.53, plasticiser 0.35, and fine gravel 0.19, while the largest negative effect was water −0.3. In the second phase of the process, the highest positive ATE of 0.5 was shown by the plasticiser, and the largest negative was for coarse gravel −0.23. Given the complex interactions between variables and the dynamic nature of the process, a genetic algorithm is proposed for optimising the mixture composition. The AI model predicts potential CO2 emission reductions of up to 30 % when using fly ash, and up to 50 % when using slag.
Keywords
artificial intelligence; causality; optimisation; concrete
Hrčak ID:
321436
URI
Publication date:
17.11.2024.
Visits: 0 *