Scholarlink

<p>The modern construction industry relies heavily on high-performance concrete (HPC) because of its superior mechanical properties as well as long-lasting nature and excellent workability characteristics. Traditional procedures used for strengthening HPC face limitations because they depend on empirical restrictions and require extensive experimental periods. The proposed model is a hybrid of deep learning with SHapley Additive exPlanations and multi-objective Bayesian optimization for data-driven interpretable prediction and improvement of HPC compressive strength. A feedforward neural network with full connection trained concrete data from an 8-parameter material constituent and curing age dataset. The predicted strength outcomes from this model reached an exceptionally high standard (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^2 = 0.926\)</EquationSource> </InlineEquation>; RMSE = 3.49 MPa) beyond the performance level of baseline methods Random Forest, XGBoost, and Support Vector Regression. SHAP analysis showed that cement content together with curing age and superplasticizer dosage made the most significant contributions to the predictions. The optimized mixture design represented a target compressive strength of 160 MPa with sustainable material usages. Furthermore, we parameterize embodied carbon and cost objectives and detail how to integrate them into the MOBO stage when project-specific unit factors are available, thereby operationalizing the sustainability framing. The proposed work helps civil engineering adopt explainable sustainable AI tools through which it aligns with digital transformation and green infrastructure initiatives.</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

Deep learning-based optimization for predicting and enhancing compressive strength of high-performance concrete

  • 期刊论文

  • 发表日期：

    2026年2月27日

查看全文
  • Rupesh Kumar Tipu,
  • Gaurang A. Patel

摘要

The modern construction industry relies heavily on high-performance concrete (HPC) because of its superior mechanical properties as well as long-lasting nature and excellent workability characteristics. Traditional procedures used for strengthening HPC face limitations because they depend on empirical restrictions and require extensive experimental periods. The proposed model is a hybrid of deep learning with SHapley Additive exPlanations and multi-objective Bayesian optimization for data-driven interpretable prediction and improvement of HPC compressive strength. A feedforward neural network with full connection trained concrete data from an 8-parameter material constituent and curing age dataset. The predicted strength outcomes from this model reached an exceptionally high standard ( \(R^2 = 0.926\) ; RMSE = 3.49 MPa) beyond the performance level of baseline methods Random Forest, XGBoost, and Support Vector Regression. SHAP analysis showed that cement content together with curing age and superplasticizer dosage made the most significant contributions to the predictions. The optimized mixture design represented a target compressive strength of 160 MPa with sustainable material usages. Furthermore, we parameterize embodied carbon and cost objectives and detail how to integrate them into the MOBO stage when project-specific unit factors are available, thereby operationalizing the sustainability framing. The proposed work helps civil engineering adopt explainable sustainable AI tools through which it aligns with digital transformation and green infrastructure initiatives.