Scholarlink

<p>The application of adaptable cementitious composites can promote sustainable development in the building sector. Integrating nanoparticles into cement composites can produce materials with superior performance and a broad spectrum of applications. Carbon nanotubes (CNTs) hold significant potential for creating effective solutions that contribute to a sustainable ecosystem with diverse properties. However, due to their complex structure and nonlinear behaviour, it is difficult to predict the properties of these composites. It is challenging but also costly, and time-consuming to plan and carry out laboratory studies on various samples at different ages. Furthermore, an accurate predictive model for estimating concrete reinforced with nanoparticles’ compressive strength (CS) has yet to be developed. This study aims to develop several machine learning techniques, including Light Gradient Boosting Machines (LightGBM), Learning Vector Quantization (LVQ), AdaBoost Regressor, Random Forest (RF), Extreme Gradient Boosting (XGB), and Categorical Boosting (CatBoost) models, to predict the CS of concrete containing CNTs. A total of 282 samples came from several reputable and trustworthy experiments. Out of them, 57 samples (20 per cent) were set aside for testing, while 225 data (eighty per cent) were used for training. The statistical indicators demonstrated the CatBoost model’s superiority over the other models, achieving an R<sup>2</sup> value of 0.95, MAE of 1.98, RMSE of 2.51, MSE of 6.30, WI of 0.99, and NSE of 0.95 during the testing phase. The XGBoost and RF models also showed strong predictive capabilities compared to the LightGBM, LVQ, and AdaBoost Regressor models. We employed Partial Dependency Plots, Local Interpretable Model-agnostic Explanations, and Shapley Additive Explanations to analyse the models and their predictions. The post-hoc explanations included information about the following: (a) the importance of each feature; (b) how they interacted with one another; and (c) the fundamental reasoning behind the predictions. According to explainable artificial intelligence, the curing period had the most influence on CS prediction for concrete with CNTs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable AI approach exploring the impact of carbon nanotubes on the compressive strength of concrete through various machine learning techniques

  • 期刊论文

  • 发表日期：

    2025年11月17日

查看全文
  • Suhaib Rasool Wani,
  • Manju Suthar,
  • Rayeh Nasr Al-Dala’ien

摘要

The application of adaptable cementitious composites can promote sustainable development in the building sector. Integrating nanoparticles into cement composites can produce materials with superior performance and a broad spectrum of applications. Carbon nanotubes (CNTs) hold significant potential for creating effective solutions that contribute to a sustainable ecosystem with diverse properties. However, due to their complex structure and nonlinear behaviour, it is difficult to predict the properties of these composites. It is challenging but also costly, and time-consuming to plan and carry out laboratory studies on various samples at different ages. Furthermore, an accurate predictive model for estimating concrete reinforced with nanoparticles’ compressive strength (CS) has yet to be developed. This study aims to develop several machine learning techniques, including Light Gradient Boosting Machines (LightGBM), Learning Vector Quantization (LVQ), AdaBoost Regressor, Random Forest (RF), Extreme Gradient Boosting (XGB), and Categorical Boosting (CatBoost) models, to predict the CS of concrete containing CNTs. A total of 282 samples came from several reputable and trustworthy experiments. Out of them, 57 samples (20 per cent) were set aside for testing, while 225 data (eighty per cent) were used for training. The statistical indicators demonstrated the CatBoost model’s superiority over the other models, achieving an R2 value of 0.95, MAE of 1.98, RMSE of 2.51, MSE of 6.30, WI of 0.99, and NSE of 0.95 during the testing phase. The XGBoost and RF models also showed strong predictive capabilities compared to the LightGBM, LVQ, and AdaBoost Regressor models. We employed Partial Dependency Plots, Local Interpretable Model-agnostic Explanations, and Shapley Additive Explanations to analyse the models and their predictions. The post-hoc explanations included information about the following: (a) the importance of each feature; (b) how they interacted with one another; and (c) the fundamental reasoning behind the predictions. According to explainable artificial intelligence, the curing period had the most influence on CS prediction for concrete with CNTs.