Scholarlink

<p>This study proposes a digital twin (DT) and machine learning (ML) framework to predict nonlinear pushover responses of reinforced concrete (RC) shear walls using analytically derived data. Two hundred SAP2000 layered shell models were analyzed, and monotonic lateral capacity curves were processed via SPO2FRAG for bilinear parameter extraction. Key response features—initial stiffness (<i>K</i><sub>0</sub>), yield displacement (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\delta ^{y}\)</EquationSource> </InlineEquation>), and post-yield stiffness ratio (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\alpha\)</EquationSource> </InlineEquation>)were identified. Ten input variables including wall geometry, material properties, reinforcement ratios, axial load, and opening ratio were used to train Random Forest regressors for predicting the pushover curve descriptors. Model accuracy was validated using nested cross-validation, yielding mean test <i>R</i><sup>2</sup> values of 0.996 for <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\({K}_{0}\)</EquationSource> </InlineEquation>, 0.995 for <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(\delta ^{y}\)</EquationSource> </InlineEquation>, and 0.925 for <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\alpha\)</EquationSource> </InlineEquation>, while uncertainty measures (Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), 95% confidence intervals) supported robustness. The DT surrogate reconstructs pushover curves in under 2 seconds per specimen, supporting rapid parametric analysis and seismic scenario assessments without the need for repetitive finite element simulations. The study also documents model limitations and outlines guidance for extending the approach to shear-dominated walls and experimental validation.</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

Digital twin-based machine learning framework for predicting nonlinear seismic response of reinforced concrete shear walls using analytical data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月26日

查看全文
  • Kevin Karanja Kuria,
  • Orsolya Kegyes-Brassai

摘要

This study proposes a digital twin (DT) and machine learning (ML) framework to predict nonlinear pushover responses of reinforced concrete (RC) shear walls using analytically derived data. Two hundred SAP2000 layered shell models were analyzed, and monotonic lateral capacity curves were processed via SPO2FRAG for bilinear parameter extraction. Key response features—initial stiffness (K0), yield displacement ( \(\delta ^{y}\) ), and post-yield stiffness ratio ( \(\alpha\) )were identified. Ten input variables including wall geometry, material properties, reinforcement ratios, axial load, and opening ratio were used to train Random Forest regressors for predicting the pushover curve descriptors. Model accuracy was validated using nested cross-validation, yielding mean test R2 values of 0.996 for \({K}_{0}\) , 0.995 for \(\delta ^{y}\) , and 0.925 for \(\alpha\) , while uncertainty measures (Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), 95% confidence intervals) supported robustness. The DT surrogate reconstructs pushover curves in under 2 seconds per specimen, supporting rapid parametric analysis and seismic scenario assessments without the need for repetitive finite element simulations. The study also documents model limitations and outlines guidance for extending the approach to shear-dominated walls and experimental validation.