In recent years, there has been a surge in studies examining the feasibility of incorporating ceramic waste powder (CWP) as a partial substitute for cement in concrete production. Researchers worldwide have conducted these investigations to assess how CWP affects the compressive strength of concrete, a crucial factor in structural design and a widely used measure of concrete quality by engineers. Given that the conventional methods for determining compressive strength are resource-intensive and susceptible to human error, there is a growing need for more efficient approaches. This study presents a data-driven method for predicting concrete compressive strength that incorporates CWP as a partial replacement for Portland cement. Various ensemble boosting models (AdaBoost, Gradient Boost, XGBoost, CatBoost, and LightGBM) are deployed, and their performances are rigorously evaluated. Furthermore, this is the first study to employ an innovative deep neural network based on a tabular data learning architecture, called TabNet, for predicting compressive strength in ceramic waste powder-based concrete. This neural network, designed specifically for tabular data, is known for its interpretability and high performance. The effectiveness of all these models is assessed using metrics such as root mean square error (RMSE) and coefficient of determination (R2) values. Additionally, the study investigates the significance of different features, particularly CWP, in predicting the target variable. This is accomplished through SHAP (SHapley Additive exPlanations) analysis, which assigns importance values to each feature based on individual predictions made by the model. TabNet achieved the maximum accuracy, with R2 = 0.969 and RMSE = 4.126, outperforming methods such as XGBoost, which recorded R2 = 0.961 and RMSE = 4.674. CatBoost achieved the lowest R2 value of 0.92 with an RMSE score of 6.666. Feature importance analysis further indicated that ceramic waste powder, though less dominant than the water–cement ratio and cement, consistently ranked sixth, enhancing compressive strength at optimal replacement levels (10–20%). Hence, it is possible to adopt a data-driven approach to predict the compressive strength of ceramic waste-based concrete, promoting sustainable mix design over conventional techniques that require significant laboratory effort to predict compressive strength.

Automated compressive strength prediction for concrete composites with ceramic waste powder as partial cement replacement with data-driven techniques

期刊论文
发表日期：

2025年12月13日

查看全文

T. Jothi Saravanan,
K. I. Syed Ahmed Kabeer

摘要

In recent years, there has been a surge in studies examining the feasibility of incorporating ceramic waste powder (CWP) as a partial substitute for cement in concrete production. Researchers worldwide have conducted these investigations to assess how CWP affects the compressive strength of concrete, a crucial factor in structural design and a widely used measure of concrete quality by engineers. Given that the conventional methods for determining compressive strength are resource-intensive and susceptible to human error, there is a growing need for more efficient approaches. This study presents a data-driven method for predicting concrete compressive strength that incorporates CWP as a partial replacement for Portland cement. Various ensemble boosting models (AdaBoost, Gradient Boost, XGBoost, CatBoost, and LightGBM) are deployed, and their performances are rigorously evaluated. Furthermore, this is the first study to employ an innovative deep neural network based on a tabular data learning architecture, called TabNet, for predicting compressive strength in ceramic waste powder-based concrete. This neural network, designed specifically for tabular data, is known for its interpretability and high performance. The effectiveness of all these models is assessed using metrics such as root mean square error (RMSE) and coefficient of determination (R²) values. Additionally, the study investigates the significance of different features, particularly CWP, in predicting the target variable. This is accomplished through SHAP (SHapley Additive exPlanations) analysis, which assigns importance values to each feature based on individual predictions made by the model. TabNet achieved the maximum accuracy, with R² = 0.969 and RMSE = 4.126, outperforming methods such as XGBoost, which recorded R² = 0.961 and RMSE = 4.674. CatBoost achieved the lowest R² value of 0.92 with an RMSE score of 6.666. Feature importance analysis further indicated that ceramic waste powder, though less dominant than the water–cement ratio and cement, consistently ranked sixth, enhancing compressive strength at optimal replacement levels (10–20%). Hence, it is possible to adopt a data-driven approach to predict the compressive strength of ceramic waste-based concrete, promoting sustainable mix design over conventional techniques that require significant laboratory effort to predict compressive strength.