Scholarlink

<p>Self-compacting concrete requires a careful balance between binder content, water content, aggregate skeleton, and chemical admixtures. This balance leads to a complex and nonlinear relationship between mixture composition and compressive strength. In this study, a physics-guided and uncertainty-aware framework is developed to predict and optimize the 28-day compressive strength of self-compacting concrete. A curated dataset consisting of 555 experimental mixtures is transformed into physically meaningful descriptors based on binder content, water-to-binder ratio, paste-related measures, aggregate-related measures, and interaction terms. Unlike conventional data-driven models, the proposed approach explicitly enforces physically meaningful monotonic trends. Compressive strength is constrained to increase with binder content and to decrease with increasing water-to-binder and aggregate-to-paste ratios, thereby improving physical consistency and interpretability. The proposed model demonstrates strong predictive performance. The mean absolute error is approximately 3.9 megapascals. The root mean square error is approximately 6.3 megapascals. The coefficient of determination is approximately 0.87. These results are superior to those obtained using linear regression and random forest models. A key innovation of this work is the integration of distribution-free conformal prediction with physics-guided machine learning to generate calibrated uncertainty intervals for compressive strength. Accumulated local effects and permutation importance analyses further confirm that the learned relationships are consistent with established material behavior. In addition, an optimization framework is formulated to support uncertainty-aware modification of mixture proportions while satisfying feasibility constraints and limiting deviations from a reference mixture. The proposed methodology provides a transparent and reliable tool for predicting compressive strength and guiding practical mix design decisions for self-compacting concrete.</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

Reliable strength prediction and mix optimization of self-compacting concrete using physics-constrained boosting and conformal intervals

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Rupesh Kumar Tipu,
  • Sagar Paruthi

摘要

Self-compacting concrete requires a careful balance between binder content, water content, aggregate skeleton, and chemical admixtures. This balance leads to a complex and nonlinear relationship between mixture composition and compressive strength. In this study, a physics-guided and uncertainty-aware framework is developed to predict and optimize the 28-day compressive strength of self-compacting concrete. A curated dataset consisting of 555 experimental mixtures is transformed into physically meaningful descriptors based on binder content, water-to-binder ratio, paste-related measures, aggregate-related measures, and interaction terms. Unlike conventional data-driven models, the proposed approach explicitly enforces physically meaningful monotonic trends. Compressive strength is constrained to increase with binder content and to decrease with increasing water-to-binder and aggregate-to-paste ratios, thereby improving physical consistency and interpretability. The proposed model demonstrates strong predictive performance. The mean absolute error is approximately 3.9 megapascals. The root mean square error is approximately 6.3 megapascals. The coefficient of determination is approximately 0.87. These results are superior to those obtained using linear regression and random forest models. A key innovation of this work is the integration of distribution-free conformal prediction with physics-guided machine learning to generate calibrated uncertainty intervals for compressive strength. Accumulated local effects and permutation importance analyses further confirm that the learned relationships are consistent with established material behavior. In addition, an optimization framework is formulated to support uncertainty-aware modification of mixture proportions while satisfying feasibility constraints and limiting deviations from a reference mixture. The proposed methodology provides a transparent and reliable tool for predicting compressive strength and guiding practical mix design decisions for self-compacting concrete.