Scholarlink

<p>The continued increased intensity and variability of seismic hazards warrant the development of design methods for resilient reinforced concrete (RC) frames to account for complicated ground-motion effects and improve quantifying risk compared with traditional time-history optimization. The approaches adopted by these traditional methods generally rely on arbitrary record selections, which are then modeled with expensive, nonlinear simulation, consequently sacrificing both reliability and computational tractability. To fill those gaps, paper introduce a fully integrated optimization framework, blending advanced ground motion encoding, differentiable structural-response modeling, risk-consistent multi-objective synthesis, and next-generation plasma-based optimizer-all validated under adversarial seismic scenarios. The pipeline begins with Ground Motion Latent Harmonic Encoder replacing subjective record selection with embedding wavelet-scattering features in a learned latent space and clustering them in a small set of hazard-weighted canonical motions, preserving over 97% of spectral similarity while compressed to six representative records. The Differentiable Nonlinear Response Operator-a physics-regularized Fourier neural operator-directly maps design variables and encoded motions to key would-be engineering demands with analytical gradients while maintaining median prediction errors below 6–9%. These outputs feed the Reliability-Consistent Multi-Objective Synthesizer, which formulates smooth, differentiable objectives and constraints through weighted conditional value at risk, enabling risk-aware cost and drift control. The search for Pareto-optimal designs is driven by the Plasma Temperature Trust Region Optimizer, balancing exploration and gradient-guided exploitation to achieve 8–18% cost reduction and 15–25% drift-risk mitigation at 3-6x faster convergence. Finally, Adversarial Hazard Closure and Validation tests design candidates with adversarial motions spectrum matched to ensure collapse margin ratios above 2.5 and annual loss predictions within 8% of full Monte Carlo estimates. This end-to-end framework is one of the major advances in seismic design of RC frames by integrating machine-learned ground-motion compression, differentiable physics, and robust optimization, giving rise to lighter, safer, and computationally more efficient structures under existing practices.</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

Optimization for seismic-resilient reinforced concrete frames under time history loading

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Sarala P. Adhau,
  • V. S. Rajguru,
  • Yashshree S. Dhale,
  • Yogesh Lanjewa,
  • Achal Rokade

摘要

The continued increased intensity and variability of seismic hazards warrant the development of design methods for resilient reinforced concrete (RC) frames to account for complicated ground-motion effects and improve quantifying risk compared with traditional time-history optimization. The approaches adopted by these traditional methods generally rely on arbitrary record selections, which are then modeled with expensive, nonlinear simulation, consequently sacrificing both reliability and computational tractability. To fill those gaps, paper introduce a fully integrated optimization framework, blending advanced ground motion encoding, differentiable structural-response modeling, risk-consistent multi-objective synthesis, and next-generation plasma-based optimizer-all validated under adversarial seismic scenarios. The pipeline begins with Ground Motion Latent Harmonic Encoder replacing subjective record selection with embedding wavelet-scattering features in a learned latent space and clustering them in a small set of hazard-weighted canonical motions, preserving over 97% of spectral similarity while compressed to six representative records. The Differentiable Nonlinear Response Operator-a physics-regularized Fourier neural operator-directly maps design variables and encoded motions to key would-be engineering demands with analytical gradients while maintaining median prediction errors below 6–9%. These outputs feed the Reliability-Consistent Multi-Objective Synthesizer, which formulates smooth, differentiable objectives and constraints through weighted conditional value at risk, enabling risk-aware cost and drift control. The search for Pareto-optimal designs is driven by the Plasma Temperature Trust Region Optimizer, balancing exploration and gradient-guided exploitation to achieve 8–18% cost reduction and 15–25% drift-risk mitigation at 3-6x faster convergence. Finally, Adversarial Hazard Closure and Validation tests design candidates with adversarial motions spectrum matched to ensure collapse margin ratios above 2.5 and annual loss predictions within 8% of full Monte Carlo estimates. This end-to-end framework is one of the major advances in seismic design of RC frames by integrating machine-learned ground-motion compression, differentiable physics, and robust optimization, giving rise to lighter, safer, and computationally more efficient structures under existing practices.