Scholarlink

<p>Most design methods use simplified earth-pressure theories and decoupled analyses that struggle to represent soil–structure interaction, uncertainty, and code logic in a united, optimizable framework. The growing demand for retaining-wall systems that balance safety, material efficiency, and lifecycle performance highlights a gap in current practice When walls must achieve stringent dependability goals, seismic loading distribution modifications, and cost- and carbon-efficiency, these limits are very constraining. To bridge this gap, this work integrates physics, uncertainty, code compliance, robustness, and validation into a five-stage computational architecture differentiable design pipeline. To predict active/passive pressures and soil–structure stiffness quickly and differently, physics informed topology-optimization surrogate PITO-SSI enforces equilibrium, Mohr–Coulomb yield, compatibility, and wall flexure. The outputs feed BMRDO, a Bayesian multi-fidelity reliability-guided optimizer that directly maximizes sliding, overturning, bearing, and flexural limit state reliability indices utilizing low/medium-fidelity surrogates and sparse high-fidelity simulations GCO-DCC graphs and validates wall components using differentiable ACI/Eurocode clauses for gradient-driven geometry and reinforcement optimization. Designs that survive DR-SAPE construct seismic earth-pressure envelopes with Wasserstein distributional resilience to withstand ground-motion statistics and dynamic soil characteristics. Finally, L3P V&amp;V integrates lifespan cost, carbon, and conformal-prediction-based methods to produce defensible Pareto-optimal solutions. These models enable replicable high-fidelity retaining-wall engineering for optimization by reducing material use, improving dependability, and quantifying robustness.</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

Reliability-based design optimization of retaining walls using machine learning techniques

  • 期刊论文

  • 发表日期：

    2025年12月15日

查看全文
  • Pallavi S. Chakole,
  • Snehal K. Kamble,
  • Sangita Meshram,
  • Shradhesh Marve,
  • Abhinay Gudadhe,
  • Latika Pinjarkar,
  • Vikash R. Agrawal

摘要

Most design methods use simplified earth-pressure theories and decoupled analyses that struggle to represent soil–structure interaction, uncertainty, and code logic in a united, optimizable framework. The growing demand for retaining-wall systems that balance safety, material efficiency, and lifecycle performance highlights a gap in current practice When walls must achieve stringent dependability goals, seismic loading distribution modifications, and cost- and carbon-efficiency, these limits are very constraining. To bridge this gap, this work integrates physics, uncertainty, code compliance, robustness, and validation into a five-stage computational architecture differentiable design pipeline. To predict active/passive pressures and soil–structure stiffness quickly and differently, physics informed topology-optimization surrogate PITO-SSI enforces equilibrium, Mohr–Coulomb yield, compatibility, and wall flexure. The outputs feed BMRDO, a Bayesian multi-fidelity reliability-guided optimizer that directly maximizes sliding, overturning, bearing, and flexural limit state reliability indices utilizing low/medium-fidelity surrogates and sparse high-fidelity simulations GCO-DCC graphs and validates wall components using differentiable ACI/Eurocode clauses for gradient-driven geometry and reinforcement optimization. Designs that survive DR-SAPE construct seismic earth-pressure envelopes with Wasserstein distributional resilience to withstand ground-motion statistics and dynamic soil characteristics. Finally, L3P V&V integrates lifespan cost, carbon, and conformal-prediction-based methods to produce defensible Pareto-optimal solutions. These models enable replicable high-fidelity retaining-wall engineering for optimization by reducing material use, improving dependability, and quantifying robustness.