Search

This study seeks to present a sophisticated artificial intelligence (AI) framework to model the compressive strength (fc′) of concrete containing waste foundry sand (WFS), with the aim of minimizing the need for time‐consuming laboratory tests and skilled technicians. For this purpose, artificial neural network (ANN) is hybridized with two metaheuristic algorithms—particle swarm optimization (PSO) and ant colony optimization (ACO) to predict the fc′ of 340 samples containing WFS collected from the literature. Results indicated that the ACO + ANN model showed the best performance with the Pearson coefficient of 0.9971, mean absolute error of 0.0221 MPa, and root mean squared error of 0.7473 MPa. The values of prediction errors exhibited that more than 90% of them in the ACO + ANN model fall within the range of (−1.5 MPa, 1.5 MPa), while this range for the PSO + ANN and traditional ANN models was obtained as (−3 MPa, 3 MPa) and (−4 MPa, 4 MPa), respectively. Furthermore, the proposed ACO + ANN model predicted the fc′ in the range of 5.24–54.48 MPa. Besides, the results indicated that the water‐to‐cement ratio, cement strength class, and cement content had the most significant impact on the fc′ of WFS‐containing concrete. Finally, a comparison was made between the proposed ACO + ANN model and four other AI models recently proposed in the literature, in which the performance criteria demonstrated that the proposed ACO + ANN model outperformed the models in the literature. [ABSTRACT FROM AUTHOR]

Unsustainable development in the industry has increased the challenge of proper disposal of industrial by‐products and has caused environmental pollution on a large scale. Furthermore, minimizing the carbon dioxide emissions due to industrial and environmental activities is a crucial challenge to controlling the global warming crisis and climate change. The use of industrial by‐product‐based supplementary cementitious materials (SCMs) can reduce fossil fuel consumption through the reduction in cement production and contribute to suitable waste management. In this research, a meta‐model was developed using integration of the decision tree‐based method named reduced error pruning tree (REPtree) and ensemble approaches (boosting and bagging) to improve the pruning phase for strength simulation of concrete mixtures incorporating industrial by‐products. Then, ensemble multiple learning algorithms were applied to develop a meta‐model using the artificial neural network and Gaussian process regression. A comprehensive experimental‐based database incorporating 3836 records from short‐term to long‐term compressive strength (CS) was collected from the literature. Performance measures present that the proposed ensemble boosting‐REPtree model outperforms other benchmark and ensemble models with a correlation coefficient of 0.960 and root mean square error of 7.884 MPa. Also, uncertainty evaluation was done based on the Monte Carlo simulation algorithm in the modeling process. The FAST and SOBOL methods, as global sensitivity analysis, presented water‐to‐binder (W/B) ratio, curing time, and SCMs to binder ratio (SCM/B) as critical components to estimate the CS of concrete mixtures. The computational simulation of the influential parameters on concrete mixtures incorporating industrial by‐products revealed that the experimental factors yielding the highest CS of the mixtures were calculated as follows: cement content of 584 kg/m3, W/B ratio of 0.17, superplasticizer‐to‐cement ratio of 1.88%, and SCM/B ratio of 0.15. [ABSTRACT FROM AUTHOR]

Reinforced concrete structures can experience various harsh environments during their service life, among which chloride ion exposure, especially in marine environments, can cause the durability reduction and deterioration of concrete structures. Artificial intelligence (AI)‐based modeling of the non‐steady‐state apparent chloride diffusion coefficient (DC) of concrete for a long exposure time using the experimental field results can assist in identifying the influential factors and better estimating the service life of a concrete structure. In this study, two novel extensions of ensemble AI algorithms, including genetic programming forest (GPF) and linear genetic programming forest (LGPF) algorithms, were proposed to model the DC of concrete. The experimental field data were gathered from the literature. Different structures of the proposed ensemble methods were developed and examined, and the best‐developed model was selected for further analysis, including sensitivity analysis and parametric study. In addition, the random forest (RF) method was used as the control ensemble technique to have a comparison. The results show that the best LGPF model possesses superior performance than the best‐developed GPF and RF models. In addition, the results show that silica fume‐to‐binder ratio, exposure time, and exposure conditions have the most significant impacts on the DC of concrete. This study contributes to the civil engineering practice by developing a new tool to model the DC of concrete that facilitates the durability assessment of concrete structures. [ABSTRACT FROM AUTHOR]

The use of Class F fly ash (CFFA) as a partial replacement of cement in the concrete mixture can provide a wide variety benefits such as improving the mechanical properties, reducing the construction costs, and enhancing the environmental conditions. Compressive strength as one of the most critical mechanical parameters used in many design codes can be obtained either through costly and time‐consuming experiments or non‐destructive methods such as predictive models. In this study, a hybrid model of the radial basis function neural network (RBFNN) and firefly optimization algorithm (FOA) (RBFNN + FOA) was proposed for the prediction of the compressive strength of self‐compacting concrete (SCC) containing CFFA. For this purpose, a multi‐laboratory dataset containing 327 SCC samples made with CFFA was employed. The input parameters in the proposed models included the age of the specimen and the amounts of cement, water, CFFA, coarse aggregate, fine aggregate, and superplasticizer, and the only output parameter was the compressive strength of SCC. For comparison purpose, the artificial neural network (ANN) was also used to model the compressive strength of SCC containing CFFA. The results show that the compressive strength estimated by the proposed and ANN models have good accuracies compared to the experimental results. Moreover, several performance metrics showed that the accuracy of the proposed model is higher than that of the model developed based on ANN algorithm. [ABSTRACT FROM AUTHOR]