A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Strength Prediction of Geopolymer Concrete With Wide-Ranged Binders and Properties Using Artificial Neural Network
Geopolymer concrete (GPC) is an efficient alternative to traditional construction materials, utilizing amorphous or semicrystalline waste–based binders. It enhances durability than Portland cement, remaining environment friendly and cost-effective, while promoting waste reuse. Despite extensive use and research, predicting concrete strength based on the mixed chemical composition of constituents remains challenging due to consideration of limited variables, small dataset sizes, and traditional models struggling with nonlinearity. The primary objective is to develop an artificial neural network (ANN) model to enhance the accuracy of compressive strength (C-S) predictions of GPC, which is crucial for the sustainable use of waste-based binders in construction. For this, the study incorporates a comprehensive dataset of 1018 experimental data points of binary and ternary GPC from 43 research sources addressing these issues. Thirteen input parameters, including three major geopolymer binder’s oxide compositions (viz, SiO2, CaO, and Al2O3), are considered to generalize the model across various binder types. With all these, an ANN model was developed to handle the inherent nonlinearity of input parameters, validated through fivefold cross-validation. The relationship between individual parameters and C-S is analyzed. Additionally, regression models (viz., linear regression (LR), Lasso regression, Ridge regression, and XGBoost regressor) were developed to thoroughly assess the performance of the ANN model in comparison to the conventional regression model. The ANN model, superior to other regression techniques, excels in recognizing nonlinear correlations between features and an objective variable, a challenge for traditional regression models. Various metrics assess model performance (e.g., mean absolute error (MAE), root mean square error (RMSE), mean square error (MSE), R-squared (R2)), revealing the ANN model’s superiority over regression models. The ANN model’s proficiency in nonlinear correlations contributes to its effectiveness, yielding low MAE ( = 2.58), RMSE ( = 4.13), and MSE ( = 17.86) values, and high (R2 = 0.93) value, signifying minimal deviation between predicted and actual C-S values. By focusing on the oxide composition of binders, the model is more generalized, making it applicable across different binder types and not limited to specific materials. Also, the parametric analysis confirmed the ANN’s ability to capture the effects of input parameters, offering a comprehensive prediction tool. This study highlights the potential of employing the ANN model for predicting material behavior, offering a resource-efficient approach, and showcasing the viability of mixed GPC for sustainable industrial waste utilization. Future research directions include exploring the model’s application under varying environmental conditions and expanding the dataset to enhance its diversity and representativeness.
Strength Prediction of Geopolymer Concrete With Wide-Ranged Binders and Properties Using Artificial Neural Network
Geopolymer concrete (GPC) is an efficient alternative to traditional construction materials, utilizing amorphous or semicrystalline waste–based binders. It enhances durability than Portland cement, remaining environment friendly and cost-effective, while promoting waste reuse. Despite extensive use and research, predicting concrete strength based on the mixed chemical composition of constituents remains challenging due to consideration of limited variables, small dataset sizes, and traditional models struggling with nonlinearity. The primary objective is to develop an artificial neural network (ANN) model to enhance the accuracy of compressive strength (C-S) predictions of GPC, which is crucial for the sustainable use of waste-based binders in construction. For this, the study incorporates a comprehensive dataset of 1018 experimental data points of binary and ternary GPC from 43 research sources addressing these issues. Thirteen input parameters, including three major geopolymer binder’s oxide compositions (viz, SiO2, CaO, and Al2O3), are considered to generalize the model across various binder types. With all these, an ANN model was developed to handle the inherent nonlinearity of input parameters, validated through fivefold cross-validation. The relationship between individual parameters and C-S is analyzed. Additionally, regression models (viz., linear regression (LR), Lasso regression, Ridge regression, and XGBoost regressor) were developed to thoroughly assess the performance of the ANN model in comparison to the conventional regression model. The ANN model, superior to other regression techniques, excels in recognizing nonlinear correlations between features and an objective variable, a challenge for traditional regression models. Various metrics assess model performance (e.g., mean absolute error (MAE), root mean square error (RMSE), mean square error (MSE), R-squared (R2)), revealing the ANN model’s superiority over regression models. The ANN model’s proficiency in nonlinear correlations contributes to its effectiveness, yielding low MAE ( = 2.58), RMSE ( = 4.13), and MSE ( = 17.86) values, and high (R2 = 0.93) value, signifying minimal deviation between predicted and actual C-S values. By focusing on the oxide composition of binders, the model is more generalized, making it applicable across different binder types and not limited to specific materials. Also, the parametric analysis confirmed the ANN’s ability to capture the effects of input parameters, offering a comprehensive prediction tool. This study highlights the potential of employing the ANN model for predicting material behavior, offering a resource-efficient approach, and showcasing the viability of mixed GPC for sustainable industrial waste utilization. Future research directions include exploring the model’s application under varying environmental conditions and expanding the dataset to enhance its diversity and representativeness.
Strength Prediction of Geopolymer Concrete With Wide-Ranged Binders and Properties Using Artificial Neural Network
Md Merajul Islam (author) / Md Al-Mamun Provath (author) / G. M. Sadiqul Islam (author) / Md Tariqul Islam (author)
2024
Article (Journal)
Electronic Resource
Unknown
Metadata by DOAJ is licensed under CC BY-SA 1.0
Prediction equations of creep and drying shrinkage for wide-ranged strength concrete
| British Library Conference Proceedings | 2001
Viscosity prediction of CRM binders using artificial neural network approach
| Online Contents | 2011