机器学习驱动的砂土场地冲刷桥梁群桩基础地震耗能机制高效识别与分析方法

机器学习驱动的砂土场地冲刷桥梁群桩基础地震耗能机制高效识别与分析方法

王靖程 ^{1,2} , 叶爱君 ^{1,2} , 王晓伟 ^{1,2} , 周连绪 ^{3}

(1. 同济大学土木工程防灾减灾国家重点实验室,上海 200092;2. 同济大学桥梁工程系,上海 200092;
3. 英属哥伦比亚大学工程学院,基洛纳,加拿大,BC V1V 1V7)

摘要:河床冲刷会导致桥梁的地震易损部位从桥墩向群桩基础转移,增加了传统能力保护设计策略的实施成本与难度,因而利用群桩基础耗能的抗震设计成为一种潜在的替代策略。冲刷桥梁群桩基础的地震耗能机制主要分为三类,即摇摆耗能、塑性耗能、以及摇摆 - 塑性共同耗能。对于不同土性、冲刷深度、结构参数的群桩基础,准确、高效地识别它们的地震耗能机制是开展冲刷桥梁抗震设计的前提,然而,传统的群桩基础非线性地震行为分析依赖复杂的有限元模拟,且常伴有计算收敛性问题。为此,本文提出了一种机器学习驱动的砂土场地冲刷桥梁群桩基础地震耗能机制高效识别与分析方法。利用经试验验证的有限元分析方法和随机抽样技术建立了冲刷桥梁群桩基础地震耗能机制数据集,采用经优化的支持向量机、神经网络、以及集成树算法,建立了机器学习驱动的地震耗能机制高效识别和分析模型。结果表明,神经网络能更准确地识别冲刷桥梁群桩基础的地震耗能机制,查准率和查全率基本均超过 90\% ;桩排数、桩长、桩配筋率、墩高以及桩轴压比是冲刷群桩基础地震耗能机制识别的重要变量;桩排数、桩长、桩轴压比越小,桩配筋率、墩高越大,冲刷群桩基础越倾向于摇摆耗能,反之,越倾向于塑性耗能。为了方便应用,本研究建立的地震耗能机制数据集和神经网络模型已公开于 https://bit.ly/JW912。

关键词:桥梁工程;冲刷群桩基础;地震耗能机制;机器学习;变量重要性分析

中图分类号:TU473.1 文献标志码:A doi: 10.6052/j.issn.1000-4750.2024.03.0185

MACHINE LEARNING-DRIVEN EFFICIENT IDENTIFICATION AND ANALYSIS OF SEISMIC ENERGY DISSIPATION MECHANISMS FOR SCOURED BRIDGE PILE-GROUP FOUNDATIONS IN COHESIONLESS SOILS

WANG Jing-cheng ^{1,2} , YE Ai-jun ^{1,2} , WANG Xiao-wei ^{1,2} , ZHOU Lian-xu ^{3}

(1. State Key Laboratory of Disaster Reduction in Civil Engineering, Tongji University, Shanghai 200092, China;
2. Department of Bridge Engineering, Tongji University, Shanghai 200092, China;
3. School of Engineering, University of British Columbia, Kelowna, BC V1V 1V7, Canada)

Abstract: Scour can transfer the seismic vulnerable portions of a bridge from piers to pile-group foundations, which increases the cost and difficulty of implementing the conventional capacity-protection design strategy for pile-group foundations. Therefore, leveraging the energy dissipation of pile-group foundations becomes a potential alternative strategy. The seismic energy dissipation mechanisms of scoured bridge pile-group foundations are mainly categorized into three types, i.e., rocking-induced energy dissipation, plasticity-induced energy dissipation, and the energy dissipation induced by both the rocking and plastic behaviors. For pile-group
foundations with different soil properties, scour depths, and structural parameters, accurate and efficient identification of their seismic energy dissipation mechanisms is a premise for the seismic design of scoured bridges. However, the conventional nonlinear seismic behavior analysis of pile-group foundations relies on complex finite element simulations frequently with computational convergence problems. In this study, a machine learning-driven method is proposed for efficient identification and analysis of the seismic energy dissipation mechanism of scoured bridge pile-group foundations in cohesionless soils. A dataset for seismic energy dissipation mechanism of scoured bridge pile-group foundations is established by using the experimentally validated finite element analysis method and by random sampling technique. The optimized support vector machine, neural network, and ensemble tree algorithms are used to establish a machine learning-driven model for efficient identification and analysis of seismic energy dissipation mechanisms. The study results show that the neural network can more accurately identify the seismic energy dissipation mechanism of scoured bridge pile-group foundations with both the precision and recall basically larger than 90\% . The number of pile rows, of pile length, of pile longitudinal reinforcement ratio, of pier height and, of pile axial load ratio are important variables for the identification of the seismic energy dissipation mechanism of scoured pile-group foundations. The smaller the number of pile rows, of pile length and, of pile axial load ratio, and the larger the pile longitudinal reinforcement ratio and pier height, the more the scoured pile-group foundation tends to consume energy by the rocking behavior, and conversely, the more it tends to dissipate energy by the structural plasticity. For easy implementation, the dataset and neural network established in this study are available at https://bit.ly/JW912.

Key words: bridge engineering; scoured pile-group foundation; seismic energy dissipation mechanism; machine learning; variable importance analysis

评论
Copyright © 2025 桥梁抗震小站 All Rights Reserved. Powered by Halo.