By Nidhi DhullReviewed by Susha Cheriyedath, M.Sc.Aug 13 2024A recent article published in Scientific Reports utilized multi-expression programming (MEP), gene expression programming (GEP), and the boosting-based algorithm AdaBoost to predict the 28-day compressive strength of bentonite plastic concrete (BPC). The study focused on determining how the mixture composition of BPC affects its compressive strength after 28 days.
Methodology followed by the genetic programming techniques. Image Credit: https://www.nature.com/articles/s41598-024-69271-0
Background
Bentonite, a naturally occurring clay mineral, is widely used in the concrete industry for its ability to enhance key properties of concrete and remove toxic metals. When mixed with conventional concrete, it forms BPC, which is particularly effective for constructing cut-off walls to prevent seepage under dams.
Accurately predicting the 28-day compressive strength of BPC is crucial for large dam construction projects, as it impacts critical decisions and ensures the concrete meets the required standards. Rapid and precise predictions of this strength can significantly reduce the time and cost associated with concrete testing procedures.
To address this need, the study employed evolutionary programming techniques such as GEP and MEP, along with the AdaBoost boosting algorithm. GEP and MEP, both considered grey-box models, offer empirical expressions that provide transparency and practical applicability for widespread use. AdaBoost, with its straightforward algorithm and efficient implementation, enhances prediction accuracy by combining multiple weak models to form a robust predictive framework.
Computational Methods
In this study, three soft computing models were employed to predict the 28-day compressive strength of BPC: MEP, GEP, and AdaBoost. MEP and GEP are both based on genetic programming (GP) and evolutionary algorithms, whereas AdaBoost improves a simple weak regression algorithm through iterative training.
To develop these models, a comprehensive literature review was conducted, resulting in a dataset with 246 BPC strength instances sourced from international publications. This dataset was split into a training set with 172 data points (70 %) and a testing set with 74 points (30 %).
Six key input parameters were identified as critical for predicting BPC compressive strength. These parameters include cement and water content, quantities of coarse and fine aggregates, bentonite content, and clay content.
The model development process focused on optimizing hyperparameters to achieve the highest accuracy. Model performance was evaluated using several statistical metrics, including mean absolute error (MAE), root mean square error (RMSE), performance index (PI), and objective function (OF).
The most accurate model, after comparison, was selected for Shapley Additive Explanations (SHAP) analysis to provide insights into the prediction process. Finally, a graphical user interface (GUI) was developed using Python to automate the calculation of BPC strength based on the input variables.
Results and Discussion
The output of the MEP algorithm was generated as C++ code, which was then decoded to derive an equation for predicting compressive strength. Through multiple trials with the MEP software, a straightforward and accurate equation using arithmetic functions was then developed. This equation effectively predicted the compressive strength of BPC with good accuracy for both the training and testing sets, though some discrepancies were observed.
In the GEP model, addition was used as a linking function. The results were expressed as four subexpressions, which were decoded and combined to yield the final result. The GEP model performed well, showing minor differences between actual and predicted values at certain points.
The AdaBoost algorithm achieved the highest accuracy in predicting strength values. It demonstrated smaller differences between actual and predicted values compared to the GEP and MEP models. AdaBoost also exhibited the strongest correlation with actual values during both the training and testing phases, resulting in the lowest error metrics, such as RMSE and average error. However, it did not provide an empirical relation for its output like the MEP and GEP algorithms.
The performance of the developed algorithms in predicting BPC compressive strength was compared using radar plots. AdaBoost consistently showed the lowest values for nearly all error metrics. In contrast, GEP and MEP had similar error metrics except for RMSE, where GEP had a lower RMSE in the training phase, while MEP performed better in the testing phase.
Shapley Additive Explanations (SHAP) analysis on the most accurate AdaBoost model highlighted that cement, coarse aggregate, and fine aggregate were the most influential factors in predicting BPC strength.
Conclusion and Future Prospects
Overall, this study successfully applied different machine-learning techniques to establish empirical models for predicting the 28-day compressive strength of BPC. The AdaBoost algorithm with a correlation of 0.962 proved more accurate than GEP and MEP with correlations of 0.936 and 0.926, respectively. However, AdaBoost exhibited lower PI values and did not yield empirical equations like MEP and GEP.
The proposed soft models are limited due to the small dataset of 246 variables. The researchers suggest considering larger datasets from variable sources to develop more generalized and robust models. In addition, other material properties like the diameter of aggregates, cement type, different properties of bentonite and clay, etc., should be considered as factors influencing the compressive strength of BPC.
The GUI developed in this study can be an invaluable tool for professionals in the construction sector and promote using BPC to improve concrete properties.
Journal Reference
Inqiad, W. B. (2024). Soft computing models for prediction of bentonite plastic concrete strength. Scientific Reports, 14(1). DOI: 10.1038/s41598-024-69271-0, https://www.nature.com/articles/s41598-024-69271-0
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.