Scholarlink

<p>This study presents a physically informed machine learning framework for predicting the coupled hygrothermal performance of sustainable lime mortars reinforced with bamboo fibers and natural jaggery. Despite growing interest in data-driven approaches for construction materials, predictive modeling of coupled moisture and thermal transport behavior in bio-modified lime mortars remains limited. The proposed framework simultaneously predicts six interrelated hygrothermal properties, namely diffusion resistance factor, vapor permeability, moisture capacity, heat of sorption, capillary suction coefficient, and thermal expansion. An experimentally anchored dataset comprising 3,600 records was developed from controlled laboratory measurements and physically derived transport descriptors for bamboo fiber contents ranging from 0.5 to 2.0% and jaggery contents ranging from 1.0 to 5.0%. Random Forest, Gradient Boosting, and K-Nearest Neighbors regression models were developed and optimized using grid-search hyperparameter tuning and 20-fold cross-validation. Additional robustness assessments, including bootstrap uncertainty analysis, compositional hold-out validation, and data dependency evaluation, were conducted to examine model reliability and potential information leakage. Among the investigated algorithms, Gradient Boosting achieved the highest overall predictive performance with an average coefficient of determination (R²) of 0.9955, root mean square error (RMSE) of 9.38, and mean absolute error (MAE) of 2.87. Random Forest also demonstrated strong predictive capability with an average R² of 0.9939, RMSE of 7.62, and MAE of 2.06, whereas K-Nearest Neighbors exhibited comparatively lower performance. Feature contribution analysis identified equilibrium moisture content, capillary uptake mass, heat exchange, vapor transport descriptors, and moisture-induced mass variation as influential predictors governing coupled hygrothermal behavior. The results demonstrate the potential of physically informed machine learning as a decision-support framework for material screening and formulation optimization within the investigated compositional and environmental domains, while maintaining the necessity of experimental validation for practical implementation.</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

Prediction of coupled hygrothermal performance in bamboo fiber–jaggery reinforced lime mortars

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月17日

查看全文
  • P Maheswar Reddy,
  • Srinivasarao Naik B,
  • Ankit Srivastava,
  • Neeraj Kumar Vidhyarthi,
  • M Amareswara Reddy,
  • Jangiti Ravi Yadav,
  • Neha Jhakra,
  • Amrita Ratnani,
  • Amit Kumar Singh

摘要

This study presents a physically informed machine learning framework for predicting the coupled hygrothermal performance of sustainable lime mortars reinforced with bamboo fibers and natural jaggery. Despite growing interest in data-driven approaches for construction materials, predictive modeling of coupled moisture and thermal transport behavior in bio-modified lime mortars remains limited. The proposed framework simultaneously predicts six interrelated hygrothermal properties, namely diffusion resistance factor, vapor permeability, moisture capacity, heat of sorption, capillary suction coefficient, and thermal expansion. An experimentally anchored dataset comprising 3,600 records was developed from controlled laboratory measurements and physically derived transport descriptors for bamboo fiber contents ranging from 0.5 to 2.0% and jaggery contents ranging from 1.0 to 5.0%. Random Forest, Gradient Boosting, and K-Nearest Neighbors regression models were developed and optimized using grid-search hyperparameter tuning and 20-fold cross-validation. Additional robustness assessments, including bootstrap uncertainty analysis, compositional hold-out validation, and data dependency evaluation, were conducted to examine model reliability and potential information leakage. Among the investigated algorithms, Gradient Boosting achieved the highest overall predictive performance with an average coefficient of determination (R²) of 0.9955, root mean square error (RMSE) of 9.38, and mean absolute error (MAE) of 2.87. Random Forest also demonstrated strong predictive capability with an average R² of 0.9939, RMSE of 7.62, and MAE of 2.06, whereas K-Nearest Neighbors exhibited comparatively lower performance. Feature contribution analysis identified equilibrium moisture content, capillary uptake mass, heat exchange, vapor transport descriptors, and moisture-induced mass variation as influential predictors governing coupled hygrothermal behavior. The results demonstrate the potential of physically informed machine learning as a decision-support framework for material screening and formulation optimization within the investigated compositional and environmental domains, while maintaining the necessity of experimental validation for practical implementation.