Scholarlink

<p>Accurately and efficiently evaluating the stochastic dynamic responses of cable-stayed bridges equipped with nonlinear tuned inerter dampers (NTIDs) under stochastic near-fault seismic excitations remains a significant challenge. Although deep learning-based surrogate models offer a promising alternative for rapid seismic response prediction, their integration with random vibration analysis for response’s probability density function and reliability evaluation is rarely explored. More critically, when the structures incorporate friction-type nonlinear control devices such as NTIDs, the discontinuity of Coulomb friction induces gradient vanishing and error accumulation in conventional physics-informed neural network (PINN), making this class of nonlinearity difficult to handle. To address these challenges, this paper proposes an accelerated stochastic dynamics analysis framework with two primary techniques. First, a mechanism-enhanced explicit time-domain physics-informed neural network (ME-PINN) is developed, featuring a tri-head prediction module and a tri-state gating mechanism that imposes velocity-regime-based physical constraints to accurately capture the non-smooth nature of Coulomb friction. Second, the ME-PINN is integrated with the direct probability integral method to enable rapid and accurate computation of stochastic seismic responses and dynamic reliabilities. Results indicate that the proposed deep learning empowered framework attains the accuracy comparable to Quasi-Monte Carlo simulation, while reducing the online computing time from 10350&#xa0;s to 83.5&#xa0;s. Moreover, quantitative assessments reveal distinct pulse effects, namely, forward-directivity pulses reduce the dynamic reliability with respect to the tower-top drift by 28.0% compared to fling-step pulses, whereas fling-step pulses decrease the reliability to main girder acceleration by 38.6%. NTIDs exhibit remarkable vibration mitigation, enhancing the dynamic reliability of tower-top drift by up to 51.6% and substantially improving the reliability of main girder acceleration under near-fault ground motions.</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 empowered stochastic dynamics analysis of single-tower cable-stayed bridge with nonlinear tuned inerter dampers

  • 期刊论文

  • 发表日期：

    2026年8月3日

查看全文
  • Zheng Zhou,
  • Ningchang Xiao,
  • Hui Li,
  • Dixiong Yang

摘要

Accurately and efficiently evaluating the stochastic dynamic responses of cable-stayed bridges equipped with nonlinear tuned inerter dampers (NTIDs) under stochastic near-fault seismic excitations remains a significant challenge. Although deep learning-based surrogate models offer a promising alternative for rapid seismic response prediction, their integration with random vibration analysis for response’s probability density function and reliability evaluation is rarely explored. More critically, when the structures incorporate friction-type nonlinear control devices such as NTIDs, the discontinuity of Coulomb friction induces gradient vanishing and error accumulation in conventional physics-informed neural network (PINN), making this class of nonlinearity difficult to handle. To address these challenges, this paper proposes an accelerated stochastic dynamics analysis framework with two primary techniques. First, a mechanism-enhanced explicit time-domain physics-informed neural network (ME-PINN) is developed, featuring a tri-head prediction module and a tri-state gating mechanism that imposes velocity-regime-based physical constraints to accurately capture the non-smooth nature of Coulomb friction. Second, the ME-PINN is integrated with the direct probability integral method to enable rapid and accurate computation of stochastic seismic responses and dynamic reliabilities. Results indicate that the proposed deep learning empowered framework attains the accuracy comparable to Quasi-Monte Carlo simulation, while reducing the online computing time from 10350 s to 83.5 s. Moreover, quantitative assessments reveal distinct pulse effects, namely, forward-directivity pulses reduce the dynamic reliability with respect to the tower-top drift by 28.0% compared to fling-step pulses, whereas fling-step pulses decrease the reliability to main girder acceleration by 38.6%. NTIDs exhibit remarkable vibration mitigation, enhancing the dynamic reliability of tower-top drift by up to 51.6% and substantially improving the reliability of main girder acceleration under near-fault ground motions.