CN-122021159-A - Bridge structure health real-time monitoring method and system

CN122021159ACN 122021159 ACN122021159 ACN 122021159ACN-122021159-A

Abstract

The invention relates to the field of intersection of fault prediction and health management technologies, and discloses a bridge structure health real-time monitoring method and system. The method comprises the steps of constructing a high-fidelity parameterized finite element model, deploying a multi-source heterogeneous sensing network, constructing a neural network fused with physical constraints, correcting model parameters on line based on a joint loss function, and outputting a health assessment result. The system comprises a sensing network, an edge-cloud cooperative computing architecture, a historical data backtracking calibration module and a multi-scale verification mechanism. By adopting the technical scheme, the self-evolution and high-fidelity maintenance of the digital twin model can be realized, and the real-time performance, reliability and engineering practicability of bridge health monitoring are improved.

Inventors

YU SHUIJING
NIU YI
Dan Yapeng
ZHANG JIAHAO
ZHU GUANGPENG
ZHANG HAN
JIAO YANGYANG
LI FEI
Shao Ruheng
MEN HAO
LI QIJIE
YANG CHENGLONG
ZHANG ZHILEI
Hao Yana
LIANG SHIQI

Assignees

郑州市公路工程公司

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (10)

1. The real-time monitoring method for bridge structure health is characterized by comprising the following steps: Establishing a high-fidelity parameterized finite element model, establishing a refined finite element model comprising a beam body, a bridge pier, a support and a connecting node based on a bridge design drawing, material properties and construction records, and parameterizing key physical parameters to form a dynamically adjustable digital twin initial model; Disposing a multi-source heterogeneous sensing network, disposing a strain sensor, an accelerometer, a displacement meter, a temperature sensor and an inclinometer at key parts of a bridge, collecting structural response data and environmental load information in real time, wherein the sampling frequency is not less than a preset working frequency, and ensuring the time consistency of the multi-source data through a time synchronization protocol; Constructing a physical information neural network, embedding the high-fidelity parameterized finite element model into a deep neural network architecture as physical constraint, wherein the deep neural network comprises an input layer, a plurality of hidden layers and an output layer, a loss function of the deep neural network is formed by forward propagation prediction errors and physical control equation residuals, and a physical control equation comprises a structural dynamics balance equation, a material constitutive relation and boundary condition constraint; On-line joint optimization and parameter correction, driving the physical information neural network to forward propagate by utilizing real-time monitoring data, calculating a joint loss function value, and reversely updating network weight and key parameters in a finite element model by a gradient descent algorithm to realize on-line, automatic and continuous correction of a digital twin model; Outputting a structural health state evaluation result, calculating stress, strain, displacement and modal parameters of a bridge key section in real time based on the corrected high-fidelity digital twin model, identifying an abnormal response mode, and generating a structural damage index and safety margin evaluation report.
2. The method according to claim 1, wherein the key physical parameters of the high-fidelity parameterized finite element model include elastic modulus, poisson ratio, density, damping ratio, support rigidity and boundary constraint coefficient, all of which are embedded in the model in the form of a learnable variable and give an initial physical reasonable range.
3. The bridge structure health real-time monitoring method according to claim 2, wherein the multi-source heterogeneous sensing network adopts a wired and wireless mixed networking mode, a key stress area realizes micro-strain level precision monitoring by using a fiber bragg grating sensor, drift caused by environmental temperature and humidity changes is corrected by a built-in temperature compensation algorithm, and data transmission delay is smaller than a preset time threshold.
4. The method for monitoring the health of a bridge structure in real time according to claim 3, wherein the number of hidden layers of the physical information neural network is a preset layer number range, the number of neurons of each layer is a preset neuron number range, the activation function adopts a modified linear unit, and the residual term of the physical control equation is embedded into the loss function by applying a Lagrange multiplier method to discrete nodes of the finite element model.
5. The method for real-time monitoring of bridge construction health according to claim 4, wherein the joint loss function , To monitor the mean square error between the data and the model predictions, The sum of squares of residuals of the physical equation at discrete space-time points is a weight coefficient And (3) with Dynamically adjusting according to data quality and automatically lifting in a data sparse period To enhance physical constraint dominance.
6. The method for real-time monitoring of bridge structure health according to claim 5, wherein the parameter correction process adopts a sliding time window mechanism, the window length is a predetermined time period, on-line optimization iteration is executed at intervals of a predetermined time, each iteration is not greater than a preset maximum iteration step number, and parameter change rate constraint is introduced to prevent overfitting.
7. The method for monitoring the health of a bridge structure in real time according to claim 6, wherein the structural damage index is calculated based on a rigidity matrix singular value decomposition result of the modified model, a first-level early warning is triggered when a minimum singular value is smaller than a preset singular value threshold, a safety margin evaluation comprehensively considers the ratio of a current load effect to a limit bearing capacity, and a high risk state is determined when the ratio is larger than the preset safety margin threshold.
8. The method for real-time monitoring of bridge construction health according to claim 7, further comprising a step of retrospectively calibrating historical data, wherein all corrected model parameters in a predetermined period of time are stored in a time sequence database, and when a sensor fault or data abnormality is detected, a latest valid parameter set is automatically invoked for model state recovery.
9. The method for monitoring bridge structure health in real time according to claim 8, further comprising the steps of verifying in a multi-scale mode that a strain field output by a neural network is compared with an actual measurement value of an optical fiber sensor point by point on a local component level, and a model recalibration process is started when frequency deviation is larger than a preset frequency tolerance by comparing the first several orders of self-vibration frequencies predicted by a model with environmental excitation recognition results on an overall structure level.
10. The utility model provides a bridge construction health real-time monitoring system which characterized in that includes: The high-fidelity parameterized finite element modeling module is used for establishing a refined finite element model comprising a beam body, a bridge pier, a support and a connecting node based on bridge design drawings, material properties and construction records, and parameterizing key physical parameters to form a dynamically adjustable digital twin initial model; The multi-source heterogeneous sensing network module is used for distributing a strain sensor, an accelerometer, a displacement meter, a temperature sensor and an inclinometer at key parts of the bridge, collecting structural response data and environmental load information in real time, wherein the sampling frequency is not less than a preset working frequency, and ensuring the time consistency of the multi-source data through a time synchronization protocol; the physical information neural network construction module is used for embedding the high-fidelity parameterized finite element model into a deep neural network architecture as physical constraint, wherein the deep neural network comprises an input layer, a plurality of hidden layers and an output layer, and a loss function of the deep neural network is formed by forward propagation prediction errors and physical control equation residuals; The on-line joint optimization and parameter correction module is used for driving the physical information neural network to forward propagate by utilizing real-time monitoring data, calculating a joint loss function value, and reversely updating the network weight and key parameters in the finite element model by a gradient descent algorithm; The structural health state evaluation module is used for calculating stress, strain, displacement and modal parameters of a bridge key section in real time based on the corrected high-fidelity digital twin model, identifying an abnormal response mode and generating a structural damage index and safety margin evaluation report.

Description

Bridge structure health real-time monitoring method and system Technical Field The invention belongs to the technical intersection field of fault prediction and health management, and particularly relates to a bridge structure health real-time monitoring method and system. Background Along with the continuous improvement of the intelligent level of the infrastructure, the health monitoring of the bridge structure has become a core link for guaranteeing the safe operation of important traffic engineering, As a typical life line engineering, the service state of the bridge is influenced by multiple factors such as material degradation, environmental erosion, load change and the like, and a real-time evaluation means with high precision and high reliability is needed. In recent years, a bridge health monitoring method based on digital twinning is widely focused, and the method aims at realizing dynamic mapping and prediction of structural states by constructing a virtual model which synchronously evolves with physical entities. However, the prior art still has the following problems in the aspect of realizing high-fidelity digital twin, that the traditional finite element model has good physical interpretability, but the model drift is generated between the model and a real structure due to the fact that the traditional finite element model cannot adapt to actual working conditions such as material aging, support settlement, boundary condition time variation and the like, and the monitoring result is gradually distorted, while the deep learning model driven by pure data can excavate complex nonlinear relation from massive monitoring data, but lacks physical rule constraint, has weak extrapolation capability and poor generalization performance under the condition of sparse data or abrupt change of working conditions, and is difficult to provide interpretable mechanical basis. The prior art is in dilemma that the physical model is stiff and cannot be updated automatically, the data model is flexible and lacks credibility, the two cracks to cause the digital twin body to be difficult to maintain high fidelity in the whole life cycle of the bridge, the realization of key functions such as accurate perception of the health state of the structure, early warning of damage, reliable evaluation of the residual life and the like is severely restricted, and a novel self-evolution modeling paradigm integrating a physical mechanism and data intelligence is needed. Disclosure of Invention The invention aims to provide a real-time monitoring method and system for bridge structure health, which can effectively solve the problems in the background technology. In order to achieve the above purpose, the invention provides a real-time monitoring method for bridge structure health, comprising the following steps: Establishing a high-fidelity parameterized finite element model, establishing a refined finite element model comprising a beam body, a bridge pier, a support and a connecting node based on a bridge design drawing, material properties and construction records, and parameterizing key physical parameters to form a dynamically adjustable digital twin initial model; Disposing a multi-source heterogeneous sensing network, disposing a strain sensor, an accelerometer, a displacement meter, a temperature sensor and an inclinometer at key parts of a bridge, collecting structural response data and environmental load information in real time, wherein the sampling frequency is not less than a preset working frequency, and ensuring the time consistency of the multi-source data through a time synchronization protocol; Constructing a physical information neural network, embedding the high-fidelity parameterized finite element model into a deep neural network architecture as physical constraint, wherein the deep neural network comprises an input layer, a plurality of hidden layers and an output layer, a loss function of the deep neural network is formed by forward propagation prediction errors and physical control equation residuals, and a physical control equation comprises a structural dynamics balance equation, a material constitutive relation and boundary condition constraint; On-line joint optimization and parameter correction, driving the physical information neural network to forward propagate by utilizing real-time monitoring data, calculating a joint loss function value, and reversely updating network weight and key parameters in a finite element model by a gradient descent algorithm to realize on-line, automatic and continuous correction of a digital twin model; Outputting a structural health state evaluation result, calculating stress, strain, displacement and modal parameters of a bridge key section in real time based on the corrected high-fidelity digital twin model, identifying an abnormal response mode, and generating a structural damage index and safety margin evaluation report. Preferably, the key physical parameters of t