Scholarlink

<p>The optimization of reinforced concrete (RC) structural elements remains a fundamental aspect of sustainable and resilient building design. However, traditional methodologies are often constrained by manual heuristics, deterministic load assumptions, and a lack of transparency in computational modeling. This study introduces an integrated framework that combines nonlinear multi-objective evolutionary optimization with interpretable machine learning (ML) to derive and predict the optimal beam and column configurations for mid-rise residential buildings. The structural design space was derived from full-scale STAAD.Pro simulations of two reinforced concrete buildings, incorporating real-world boundary conditions, code-based loading (IS 875 and IS 1893), and ductility requirements as per IS 13920. Pareto-optimal designs were initially identified by minimizing the total cost and material usage while maximizing a safety index that reflects flexural, shear, and axial performance. The resulting configurations were subsequently used to train two predictive models: extreme learning machines (ELM) and elastic net regression (ENR) to estimate the optimal cross-sections and reinforcement requirements under varying design scenarios. ELM consistently outperformed ENR, achieving a higher predictive accuracy with lower mean errors across all target variables. The SHAP analysis further elucidated the structural influence of input variables such as span, end condition, and floor level, ensuring model transparency and physical interpretability. Interaction surface plots derived from the trained models revealed highly nonlinear and position-sensitive relationships between geometric attributes and reinforcement demands, aligned with structural mechanics principles such as moment redistribution, load path variation, and biaxial force interaction. This framework advances the paradigm of data-informed, human-centric design in accordance with Industry 5.0 objectives by integrating AI-driven optimization with rigorous compliance to structural codes. It offers a scalable and deployable path toward real-time, sustainable, and explainable structural engineering solutions.</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

Design of structural elements for residential buildings utilizing non-linear multi-objective optimization and interpretable data-driven learning

  • 期刊论文

  • 发表日期：

    2026年1月2日

查看全文
  • Suraj Kumar Parhi,
  • Ashutosh Mishra,
  • Sambhunath Biswal,
  • Bharadwaj Nanda,
  • Sanjaya Kumar Patro

摘要

The optimization of reinforced concrete (RC) structural elements remains a fundamental aspect of sustainable and resilient building design. However, traditional methodologies are often constrained by manual heuristics, deterministic load assumptions, and a lack of transparency in computational modeling. This study introduces an integrated framework that combines nonlinear multi-objective evolutionary optimization with interpretable machine learning (ML) to derive and predict the optimal beam and column configurations for mid-rise residential buildings. The structural design space was derived from full-scale STAAD.Pro simulations of two reinforced concrete buildings, incorporating real-world boundary conditions, code-based loading (IS 875 and IS 1893), and ductility requirements as per IS 13920. Pareto-optimal designs were initially identified by minimizing the total cost and material usage while maximizing a safety index that reflects flexural, shear, and axial performance. The resulting configurations were subsequently used to train two predictive models: extreme learning machines (ELM) and elastic net regression (ENR) to estimate the optimal cross-sections and reinforcement requirements under varying design scenarios. ELM consistently outperformed ENR, achieving a higher predictive accuracy with lower mean errors across all target variables. The SHAP analysis further elucidated the structural influence of input variables such as span, end condition, and floor level, ensuring model transparency and physical interpretability. Interaction surface plots derived from the trained models revealed highly nonlinear and position-sensitive relationships between geometric attributes and reinforcement demands, aligned with structural mechanics principles such as moment redistribution, load path variation, and biaxial force interaction. This framework advances the paradigm of data-informed, human-centric design in accordance with Industry 5.0 objectives by integrating AI-driven optimization with rigorous compliance to structural codes. It offers a scalable and deployable path toward real-time, sustainable, and explainable structural engineering solutions.