基于LSTM神经网络的桥梁损伤预警方法  

Bridge Damage Warning Method Based on LSTM Neural Network

作　　者：王超 王康迪 柳尚 宋杰 姚守峰 WANG Chao;WANG Kang-di;LIU Shang;SONG Jie;YAO Shou-feng(Shandong Provincial Communications Planning&Design Institute Group Co.,Ltd.,Jinan 250101,China)

机构地区：[1]山东省交通规划设计院集团有限公司,山东济南250101

出　　处：《世界桥梁》2023年第3期81-88,共8页World Bridges

基　　金：国家重点研发计划项目(2021YFE0112200);山东省交通运输厅科技计划项目(2021B66)。

摘　　要：为准确把握桥梁服役期内的健康状况,提高损伤评估效率,基于泗安塘大桥主桥某年多个测点的车致应变数据,建立了一种桥梁损伤预警方法。首先使用长短时记忆(LSTM)神经网络建立了车辆荷载作用下桥梁不同测点的多输入相关性模型,之后建立桥梁有限元模型模拟有损工况的桥梁状态,并提出损伤评估指标,最后根据损伤评估指标进行桥梁损伤预警。结果表明:采用LSTM神经网络建立的多输入相关性模型能够实现桥梁任意测点间车致响应的高精度预测;基于有限元模拟的车致应变提出的4个损伤评估指标能对假定的桥梁损伤进行明显识别;对泗安塘大桥主桥进行状态评估,4个损伤评估指标均稳定分布在一定的范围内,表明该桥近期没有增加新的损伤;建立了基于多测点的结构损伤评估指标体系,设定同时刻75%及以上的损伤评估指标超限时发出预警。In this paper,a bridge damage warning method that can improve damage assessing efficiency and accurately predict the health conditions of bridges in service stage is proposed.The vehicle-induced strain data at multiple measurement points in the Siantang Bridge in a year were consulted.First,a multi-input convolutional neural network model that contains data of various measurement points under vehicle loads is built up using the long-short time memory(LSTM)neural network,subsequently a finite element model of the bridge is developed to simulate the conditions of bridge with damages,based on which damage assessment indicators are given.At last,the bridge damage warning can be made in accordance with the damage assessment indicators.It is shown that the multi-input convolutional neural network model built up using LSTM neural network can predict the vehicle-induced responses of the bridge at any measurement point with high accuracy.The assumed bridge damages can be effectively identified according to the four damage assessment indicators proposed based on the numerically simulated vehicle-induced strains.The health conditions of the main bridge of Siantang Bridge were assessed,and all the four damage assessment indicators are stably distributed within a certain range,meaning that no new damage has occurred in the bridge recently.A structural damage assessment indicator system based on multiple measurement points is established,which can give an early warning of damages when the exceedance rates of damage assessment indicators simultaneously surpasses 75%or above.

关 键 词：混凝土箱梁桥 汽车荷载 内力重分布 响应相关性 损伤预警 LSTM神经网络 

分 类 号：U446[建筑科学—桥梁与隧道工程]