Simply supported girder bridge deflection identification method based on main girder reconstruction strain
By deploying sensors on the upper edge of the outer beam of a simply supported beam bridge and reconstructing the deflection of the middle beam using distributed optical fiber strain and finite element models, the problems of difficult sensor deployment and poor data interpretability are solved, achieving high-precision and low-cost bridge deflection monitoring, which is suitable for practical engineering applications.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-07
- Publication Date
- 2026-03-10
AI Technical Summary
Existing bridge deflection monitoring methods suffer from stringent deployment conditions and poor data interpretability, especially in the case of simply supported beam bridges where sensors are difficult to deploy in the concealed areas of the beam. Furthermore, purely data-driven models lack generalization ability.
A method based on reconstructing strain from the main beam is adopted. By deploying sensors on the upper edge of the outer beam, distributed optical fiber strain data is used, combined with a finite element model and a one-dimensional convolutional neural network, to reconstruct the strain and deflection data of the upper edge of the middle beam, thereby simplifying the sensor deployment and enhancing the physical interpretability.
It achieves high-precision deflection recognition in areas where sensors are difficult to deploy, reducing the difficulty and cost of sensor deployment, while improving the interpretability of the method and its generalization ability under unknown working conditions, thus meeting the needs of engineering applications.
Smart Images
Figure CN121632503A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent monitoring of the operational safety of civil engineering structures, and relates to a method for identifying bridge deflection, specifically a method for identifying the deflection of a simply supported beam bridge based on the reconstructed strain of the main beam. Background Technology
[0002] Structural health monitoring is crucial for ensuring the safe operation of bridges, and deflection is the most direct and effective indicator reflecting structural condition. Traditional contact measurement methods (such as dial gauges and LVDTs) rely on reference points, are difficult to implement in real time, and cannot obtain continuous deflection distributions. Non-contact technologies (such as GPS and lasers) are limited by environmental interference and high costs, making them unsuitable for modern monitoring needs. Against this backdrop, strain-based indirect measurement methods have shown great potential. Among them, distributed fiber optic sensing technology, especially Brillouin optical time-domain analysis (BOTDA), has become an ideal sensing method for bridge structural deflection monitoring due to its outstanding advantages, including ultra-high spatial resolution, ultra-long sensing distance, automation, and the ability to obtain continuous strain fields.
[0003] In strain-based bridge deflection identification methods, two main categories exist: explicit function analysis and implicit function mapping. Explicit methods, represented by the conjugate beam method (CBM) and its improved models, offer clear physical meaning and high identification accuracy. However, their theoretical premise requires the simultaneous deployment of strain sensors on the upper and lower edges of the main beam. This presents stringent deployment conditions in many practical bridges (especially in concealed areas like the middle beam), significantly limiting their engineering applications. Implicit mapping methods, represented by neural networks, can address uncertain conditions such as structural cracking by learning the complex nonlinear relationship between strain and deflection. However, their "black box" nature leads to poor physical interpretability, and their generalization ability is often constrained by the completeness of the training data. Currently, a solution that can simultaneously overcome both the limitations of deployment conditions and the insufficient generalization ability of the model remains elusive. Summary of the Invention
[0004] To address the challenges of deploying distributed fiber optic deflection monitoring sensors and the poor interpretability of data in simply supported beam bridges, this invention integrates the advantages of both data-driven and physical model-driven approaches, providing a deflection identification method for simply supported beam bridges based on reconstructed strain of the main beam. This method achieves high-precision deflection identification in concealed areas of the bridge without requiring the deployment of strain sensors on the upper edge of the beam, thus resolving a core bottleneck in engineering implementation. Simultaneously, by introducing physical mechanisms, it ensures that the reconstruction process and results conform to the laws of structural mechanics, enhancing the method's interpretability and generalization ability under unknown conditions, and overcoming the inherent limitations of purely data-driven models.
[0005] The objective of this invention is achieved through the following technical solution:
[0006] A method for deflection identification of simply supported beam bridges based on reconstructed strain of the main girder includes the following steps:
[0007] Step 1: For the outer beams of simply supported beam bridges where upper edge strain sensors are easily installed, collect distributed fiber optic strain monitoring data at the upper edge of the outer beams.
[0008] Step 2: For the middle beam of a simply supported beam bridge where it is difficult to install upper edge strain sensors, collect distributed fiber optic strain monitoring data at the lower edge of the middle beam.
[0009] Step 3: Establish a finite element model of the simply supported beam bridge, obtain strain measurement data of the upper edge of the outer beam and the middle beam, and construct a distributed strain-position data relationship model;
[0010] Step 4: Based on the distributed fiber optic strain monitoring data of the outer beam obtained in Step 1, use the distributed strain and position data relationship model obtained in Step 3 to obtain the distributed fiber optic strain reconstruction data of the middle beam.
[0011] Step 5: Based on the distributed fiber optic strain monitoring data of the lower edge of the middle beam obtained in Step 2 and the distributed fiber optic strain reconstruction data of the upper edge of the middle beam obtained in Step 4, the deflection reconstruction data of the middle beam is obtained.
[0012] Compared with the prior art, the present invention has the following advantages:
[0013] This invention, through a data-physics hybrid driving strategy, achieves accurate reconstruction of continuously distributed deflection data in concealed areas of bridges, requiring only the deployment of sensors at the lower edge of the beam. This significantly reduces the difficulty and cost of sensor deployment. Simultaneously, the core components of this invention are constrained by physical mechanisms, ensuring that the reconstruction process and results conform to the laws of structural mechanics, enhancing the interpretability and generalization ability of the method under unknown conditions. This invention is applicable to solving the implementation bottlenecks and reliability problems of deflection monitoring in simply supported beam bridges in practical engineering. Attached Figure Description
[0014] Figure 1 Photograph of a simply supported beam test bridge;
[0015] Figure 2 This is a finite element model diagram of a simply supported beam test bridge.
[0016] Figure 3 To simulate the load frequency histogram;
[0017] Figure 4 This is a convergence graph of the 1D-CNN network during iteration.
[0018] Figure 5 The result is the distributed fiber strain reconstruction of the upper edge of beam B.
[0019] Figure 6 The result is the deflection reconstruction of beam B;
[0020] Figure 7 This is a flowchart of a method for deflection identification of simply supported beam bridges based on the reconstructed strain of the main girder. Detailed Implementation
[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.
[0022] This invention provides a method for deflection identification of simply supported beam bridges based on reconstructed strain of the main beam. First, using strain monitoring data from the upper edge of the outer beam (a beam where sensors are easily deployed), a strain mapping relationship is constructed from the outer beam to the middle beam (a beam where sensors are difficult to deploy) through a distributed strain-position data relationship model, reconstructing the strain data of the upper edge of the middle beam. Then, the measured strain at the lower edge of the middle beam is fused with the reconstructed strain at the upper edge, and the high-precision deflection reconstruction of the middle beam is achieved using the method for deflection identification of simply supported beam bridges based on reconstructed strain of the main beam. Figure 7 As shown, the specific steps are as follows:
[0023] Step 1: For the outer beams of simply supported beam bridges where upper edge strain sensors are easily installed, collect distributed fiber optic strain monitoring data at the upper edge of the outer beams.
[0024] Step 2: For the middle beam of a simply supported beam bridge where it is difficult to install upper edge strain sensors, collect distributed fiber optic strain monitoring data at the lower edge of the middle beam.
[0025] Step 3: Establish a finite element model of the simply supported beam bridge, obtain strain measurement data at the upper edges of the outer and middle beams, and construct a distributed strain-location data relationship model. The specific steps are as follows:
[0026] Step 31: Assume that beam A of the simply supported beam bridge is an easily arranged beam (outer beam), and beam B is a difficult-to-arrange beam (middle beam). The span of the simply supported beam is... Evenly distributed on the upper and lower edges of beam A. The beam is divided along its length by several measuring points. Units, each unit length .
[0027] Step 32: Establish a refined finite element model of the simply supported beam bridge and simulate it using the Monte Carlo method. A static loading simulation of the bridge was performed using random vehicle loads, and a strain simulation dataset of the upper edge of beam A was collected. And the strain simulation dataset of the upper edge of beam B The two datasets are defined as follows, serving as training samples for the network:
[0028]
[0029]
[0030] In the formula, For the first Simulation data of distributed fiber strain at the upper edge of beam A, obtained from the second sampling. , This represents the number of simulation samples; For the first Simulation data of distributed fiber strain at the upper edge of beam B from the second sampling;
[0031] Step 33: A one-dimensional convolutional neural network is used to extract the spatial dimension features of the strain simulation data at the upper edge of beam A. The strain simulation data at the upper edge of beam B is used as the training target for the entire model output. The model is trained using the backpropagation algorithm, thus constructing a data relationship model of the distributed strain and position at the upper edges of beams A and B in a simply supported beam bridge. The training objective of the one-dimensional convolutional neural network is to obtain the true value of the strain simulation data at the upper edge of beam B. With reconstructed values Minimize the mean squared error, as shown in the following equation:
[0032]
[0033]
[0034] In the formula, This is a parameter vector for a one-dimensional convolutional neural network. This represents the optimal parameter vector for a one-dimensional convolutional neural network. The loss function; It is a one-dimensional convolutional neural network model.
[0035] Step 4: Based on the distributed fiber optic strain monitoring data of the outer beam obtained in Step 1, and using the distributed strain-position data relationship model obtained in Step 3, the reconstructed distributed fiber optic strain data of the middle beam's upper edge is obtained. The specific steps are as follows:
[0036] Step 41: Based on the distributed fiber optic strain monitoring data of the upper edge of beam A obtained in Step 1, assume that at a certain moment, the set of distributed fiber optic strain monitoring data of the upper edge of beam A is a vector. The definition is as follows:
[0037]
[0038] In the formula, For the first Distributed fiber optic strain monitoring data at the upper edge of beam A at each measuring point. .
[0039] Step 42: Collect the distributed fiber optic strain monitoring data of the upper edge of beam A at this moment. Inputting the data relationship model between distributed strain and position, the distributed fiber optic strain reconstruction data of the upper edge of beam B at that moment is obtained. The process is shown in the following formula:
[0040]
[0041] In the formula, Let be the strain reconstruction vector at the upper edge of beam B. .
[0042] Step 43: For The set of distributed fiber optic strain monitoring data obtained from the second sampling of beam A. The definition is as follows:
[0043]
[0044] In the formula, For the first The second sample of distributed fiber optic strain monitoring data at the upper edge of beam A. .
[0045] The distributed fiber optic strain monitoring data obtained from each sampling of the upper edge of beam A are input into the distributed strain-position data relationship model to obtain the distributed fiber optic strain reconstruction data set of the upper edge of beam B. :
[0046]
[0047] In the formula, For the first Strain reconstruction data of distributed optical fiber at the upper edge of beam B from the second sampling. .
[0048] Step 5: Based on the distributed fiber optic strain monitoring data of the lower edge of the middle beam obtained in Step 2 and the distributed fiber optic strain reconstruction data of the upper edge of the middle beam obtained in Step 4, the deflection reconstruction data of the middle beam is obtained. The specific steps are as follows:
[0049] Step 51: Based on the distributed fiber optic strain monitoring data of the lower edge of beam B obtained in Step 2, assume that at a certain moment, the set of distributed fiber optic strain monitoring data of the lower edge of beam B is a vector. :
[0050]
[0051] In the formula, For the first Distributed fiber optic strain monitoring data at the lower edge of beam B at each measuring point. .
[0052] Step 52: Based on the distributed fiber strain reconstruction data of the upper edge of beam B obtained in Step 4, the average curvature reconstruction vector of beam B at a certain moment can be obtained. :
[0053]
[0054] In the formula, Let be the measured average lower edge strain vector of beam B. ; This is the reconstructed vector of the average upper edge strain of beam B. ; The distance between the upper and lower edge sensors.
[0055] Step 53: Obtain the deflection reconstruction data of beam B at this moment according to the following formula. :
[0056]
[0057] In the formula, For beam B Deflection at each measuring point ; For beam B The average curvature reconstruction value of each unit, ; For bending stiffness; Shear modulus; This represents the effective shear area.
[0058] Step 54: For The set of distributed fiber strain reconstruction data obtained from the second sampling of beam B upper edge Substitute the samples into steps 51-53 in the sampling order to obtain the deflection reconstruction dataset of beam B. :
[0059]
[0060] In the formula, For the first The deflection reconstruction data of beam B from the second sampling, , .
[0061] The aforementioned deflection identification method for simply supported beam bridges based on reconstructed strain of the main beam utilizes strain data from the upper edge of the outer beam to reconstruct strain data from the upper edge of the middle beam. Combined with strain monitoring data from the lower edge of the middle beam, this method achieves accurate reconstruction of the middle beam deflection through a data-physical hybrid approach. This overcomes the stringent dependence of existing bridge deflection identification methods on sensor deployment conditions, improves the method's engineering applicability and ease of implementation, compensates for the shortcomings of pure data-driven models in terms of physical interpretability and extrapolation generalization ability, and ensures the reliability and robustness of the identification results.
[0062] The following experiments were conducted to verify the effectiveness of the present invention:
[0063] This experiment is based on Figure 1 Taking a 2.5-meter span aluminum alloy simply supported beam test bridge as an example, distributed fiber optic strain sensors are deployed along the length of the beams on the webs of beams A and B, forming two upper and lower edge strain sensing fiber optic measurement loops. Distributed strain measurement points are set at 0.2 meters each, dividing the beam body into 12 evenly spaced units. The upper edge strain sensing fiber optic cable is located 0.025 meters below the bottom surface of the top plate, and the lower edge strain sensing fiber optic cable is located 0.025 meters above the bottom surface of the web, with a spacing of 0.1 meters between the upper and lower edge cables. Displacement gauges are placed at 1 / 4, 1 / 2, and 3 / 4 spans of beam B to measure the deflection of the simply supported beam test bridge under the test loading conditions.
[0064] The specific details of this experiment are as follows:
[0065] Assuming beam A is easy to install strain sensors on its upper edge, and beam B is difficult to install strain sensors on its upper edge, in order to verify the effectiveness of the proposed deflection identification method for simply supported beam bridges based on reconstructed strain of the main beam, with strain measurement points only installed on the lower edge of beam B, the deflection of beam B is obtained by using only the strain data of the upper edge of beam A and the strain data of the lower edge of beam B. A static loading condition is set at the mid-span of the simply supported beam test bridge as shown in Table 1, using a concentrated force loading mode.
[0066]
[0067] Using the finite element model of a simply supported beam test bridge, such as Figure 2 As shown, in Figure 3 Data from strain sensors at the upper edges of beam A and beam B were collected under random loading conditions to form the network training sample dataset.
[0068] Using the generated training sample data, the first 80% of the data sets were designated as the training set, and the last 20% as the validation set. The 1D-CNN network was trained using strain sensor data from the upper edge of beam A as input and strain sensor data from the upper edge of beam B as output. The model iteratively converged as follows: Figure 4 As shown.
[0069] The distributed fiber optic strain monitoring data of the upper edge of beam A, collected under working condition 1, is input into the trained 1D-CNN network to obtain the output distributed fiber optic strain reconstruction data of the upper edge of beam B, such as... Figure 5 As shown. By Figure 5 It can be seen that the distributed strain-location data relationship model can effectively reconstruct the strain sensor data at the upper edge of beam B (a beam where sensors are difficult to deploy), and its strain distribution pattern has a high degree of fit with the true value. The reconstructed deflection of beam B is obtained using a simply supported beam bridge deflection identification method based on the reconstructed strain of the main beam, as shown below. Figure 6 As shown. By Figure 6 It is evident that the deflection identification method for simply supported beam bridges based on the reconstructed strain of the main girder can effectively identify the deflection of beam B (a beam where sensors are difficult to deploy). The relative error between the identification result and the true value (displacement gauge) is less than 4%, and its accuracy meets the requirements of bridge online monitoring systems. Furthermore, compared to point-type sensing devices (displacement gauges), this method can provide a continuous distribution of deflection along the longitudinal direction of the bridge. In summary, this invention not only accurately quantifies the deflection value but also precisely reconstructs the distribution pattern of the deflection, demonstrating excellent overall performance and possessing significant engineering application and promotion value.
Claims
1. A simply supported beam bridge deflection identification method based on reconstruction of strain of a main beam, characterized in that The method comprises the following steps: Step one: collecting the upper edge distributed fiber strain monitoring data of the outer edge beam of the simply supported beam bridge; Step two: collecting the lower edge distributed fiber strain monitoring data of the middle beam of the simply supported beam bridge; Step three: establishing a finite element model of the simply supported beam bridge, obtaining the upper edge strain measurement point data of the outer edge beam and the middle beam, and constructing a data relationship model of the distributed strain and the position; Step four: obtaining the middle beam upper edge distributed fiber strain reconstruction data by using the data relationship model of the distributed strain and the position obtained in step three, according to the outer edge beam upper edge distributed fiber strain monitoring data obtained in step one; Step five: obtaining the deflection reconstruction data of the middle beam according to the middle beam lower edge distributed fiber strain monitoring data obtained in step two and the middle beam upper edge distributed fiber strain reconstruction data obtained in step four.
2. The method of claim 1, wherein The specific steps of step three are as follows: Step three one: assume that the simply supported beam bridge A beam is easy to set up beam, namely the outer beam, B beam is not easy to set up beam, namely the middle beam, simply supported beam span is , on the A beam, the upper and lower edges are evenly arranged measuring points, the beam is divided into units along the length direction, and the length of each unit is ; Step three two: Establish the refined finite element model of simply supported beam bridge, simulate the random vehicle load of the group through the Monte Carlo method, and simulate the static loading of the bridge to collect the simulation data set of the upper edge strain of A beam and the simulation data set of the upper edge strain of B beam As network training samples, the two data sets are defined as follows: In the formula, is the first sampled A-beam upper edge distributed optical fiber strain simulation data, , is the simulation sample number; is the first sampled B-beam upper edge distributed optical fiber strain simulation data; Step three: using a one-dimensional convolutional neural network to extract the spatial dimension features of the A-beam upper edge strain simulation data, taking the B-beam upper edge strain simulation data as the training target of the whole model output, and realizing the training of the model through the error back propagation algorithm, so as to construct the data relationship model of the upper edge distributed strain and the position of the A-beam and the B-beam of the simply supported beam bridge.
3. The method of claim 2, wherein The training target of the one-dimensional convolutional neural network is to make the true value of the B beam upper edge strain simulation data with the reconstructed values mean square error minimization, as shown in the following formula: In the formula, is a one-dimensional convolutional neural network parameter vector; is a one-dimensional convolutional neural network optimal parameter vector; is a loss function; is a one-dimensional convolutional neural network model.
4. The method of claim 2, wherein The specific steps of step four are as follows: Step four one: according to the A beam upper edge distributed optical fiber strain monitoring data obtained in step one, assuming that at a certain moment, the A beam upper edge distributed optical fiber strain monitoring data set is a vector , defined as follows: In the formula, A beam top edge distributed optical fiber strain monitoring data of the 1st ; Step four two: collect the distributed optical fiber strain monitoring data of the upper edge of A beam at this moment Input the data relationship model of distributed strain and position to obtain the distributed optical fiber strain reconstruction data of the upper edge of B beam at this moment The process is as follows: wherein is the upper edge strain reconfiguration vector for the B-beam. Step four three: for The A-beam upper edge distributed optical fiber strain monitoring data set obtained by subsampling is defined as follows: In the formula, is the first A-beam upper edge distributed optical fiber strain monitoring data of the n-th sampling, ; The distributed optical fiber strain monitoring data of the upper edge of the A beam obtained by each sampling is respectively input into a data relationship model of distributed strain and position to obtain a distributed optical fiber strain reconstruction data set of the upper edge of the B beam : In the formula, is the first B-beam upper edge distributed optical fiber strain reconstruction data of the n-th sampling, .
5. The method of claim 4, wherein The specific steps of step five are as follows: Step five: According to the B-beam lower edge distributed optical fiber strain monitoring data obtained in step two, assuming that at a certain moment, the B-beam lower edge distributed optical fiber strain monitoring data set is vector : In the formula, B-beam bottom edge distributed optical fiber strain monitoring data of the 1st ; Step five two: according to the B beam upper edge distributed optical fiber strain reconstruction data obtained in step four, the average curvature reconstruction vector of the B beam at a certain time is obtained : wherein is the average lower edge strain measured vector for the B-beam, is the average upper edge strain reconstructed vector for the B-beam, is the upper and lower edge sensor spacing; Step five three: get the deflection reconstruction data of the B beam at this moment according to the following formula : wherein is the deflection of the B-beam at the ; is the average curvature reconstruction value of the B-beam at the ; is the bending stiffness; is the shear modulus; is the effective shear area; Step five four: for The B-beam upper edge distributed fiber strain reconstruction data set obtained by sub-sampling , substitute into step five one~step five three in turn according to the sampling order, obtain the B-beam deflection reconstruction data set : In the formula, is the first sampling of B-beam deflection reconstruction data, , .