物理信息双驱动的长距离盾构隧道结构纵向力学性态智能诊断方法  

A physics and information dual-driven intelligent diagnosis method for longitudinal mechanical behavior of long-distance shield tunnels

作　　者：张东明[1,2] 周烨璐 黄宏伟[1,2] 张晋彰 ZHANG Dong-ming;ZHOU Ye-lu;HUANG Hong-wei;ZHANG Jin-zhang(Department of Geotechnical Engineering,Tongji University,Shanghai 200092,China;Key Laboratory of Geotechnical and Underground Engineering of the Ministry of Education,Tongji University,Shanghai 200092,China)

机构地区：[1]同济大学地下建筑与工程系,上海200092 [2]同济大学岩土及地下工程教育部重点实验室,上海200092

出　　处：《岩土力学》2023年第10期2997-3010,共14页Rock and Soil Mechanics

基　　金：国家重点研发计划项目(No.2021YFF0502200,No.2021YFB2600804)。

摘　　要：针对目前盾构隧道纵向结构安全诊断面临的瓶颈,提出了一种物理信息双驱动的隧道纵向结构力学性态智能诊断方法。通过将表征隧道纵向结构力学性态的物理方程嵌入物理神经元中,利用实测数据作为信息神经元综合构建了物理信息双驱动的神经网络(physics-informed neural networks,PINNs)模型,可实时更新反演盾构隧道结构参数、周围地层参数以及荷载分布规律,继而正演求解隧道的纵向结构力学性态。将反演得到的参数进一步用于其他隧道段的分析,以实现长距离盾构隧道结构纵向智能诊断。算例与工程实例应用表明,提出的PINNs模型能有效求解隧道结构纵向问题,且相较传统的纯数据驱动的深度神经网络(deep neural network,DNN)模型,PINNs模型表现出了显著的泛化能力与鲁棒性,具有十分可观的工程应用前景。This paper proposes a physics and information dual-driven intelligent diagnostic method for the longitudinal mechanical behavior of shield tunnels,aiming to address the current bottlenecks in its safety diagnosis.By embedding the physical equations that characterize the longitudinal mechanical behavior of the tunnel into physical neurons,and integrating the measured data as information neurons,a physics and information dual-driven neural network model called physics-informed neural networks(PINNs)is constructed.This model enables real-time updating and inversion of the structural parameters of shield tunnels,surrounding geological parameters,and load distribution patterns,thereby forward solving the longitudinal structural mechanics state of the tunnel.The inverted parameters obtained are further utilized to analyze the longitudinal mechanical behavior of other tunnel sections,so as to realize the diagnosis of the longitudinal long-distance shield tunnels.Case studies and engineering applications show that the proposed PINNs model can effectively solve the longitudinal structural problems in tunnels.Furthermore,compared to the traditional purely data-driven deep neural network(DNN)models,the PINNs model exhibits significant generalization capability and robustness,presenting promising prospects for engineering applications.

关 键 词：物理信息双驱动 盾构隧道 纵向力学性态 反演 

分 类 号：U456[建筑科学—桥梁与隧道工程]