Visual-assisted lightweight deflection monitoring method and system

Abstract

This invention relates to the field of structural monitoring technology, and more particularly to a visual-assisted lightweight deflection monitoring method and system. The method includes: collecting tilt angle data at mid-span, L / 4, and 3L / 4 sections; detecting vehicle passage events at the edge ends to obtain the absolute displacement of the supports; performing dynamic modal truncation based on acceleration data; constructing observation equations based on the tilt angles of the three measurement points, using visual displacement as boundary constraints, and reconstructing the full-bridge deflection curve using regularized least squares method. This invention solves the physical blind zone problem of not being able to observe the first-order mode at a single point tilt angle at mid-span of a simply supported beam by adding simple tilt angle probes at L / 4 and 3L / 4. It utilizes visual anchor points to provide an absolute reference, achieving low-power, high-precision full-bridge deflection inversion, and is particularly suitable for long-term health monitoring of distributed small-to-medium span bridges.

Description

Technical Field

[0001] This invention relates to the field of structural monitoring technology, and in particular to a visually assisted lightweight deflection monitoring method and system. Background Technology

[0002] Maintenance and monitoring of small- and medium-span simply supported beam bridges has long been lacking. Existing structural monitoring methods have the following shortcomings: GNSS / static leveling method: High accuracy but expensive, and requires full-bridge power supply wiring, making it unsuitable for scattered small and medium-sized bridges.

[0003] Pure tilt angle integration method: Deflection is calculated using mid-span tilt angle. Although it is low-cost, it suffers from serious integration drift problem and cannot distinguish between rigid body rotation and actual deformation.

[0004] Machine vision methods require the installation of a vision camera, are susceptible to weather conditions, and have high computational and power consumption, making them difficult to operate unattended for extended periods.

[0005] Multi-point fitting method: It relies on multiple tilt angle sensors to construct overdetermined equations, has high hardware redundancy, and cannot solve the engineering pain point of "single-point sensing".

[0006] Therefore, it is necessary to provide a new visual-assisted lightweight deflection monitoring method to solve the above-mentioned technical problems. Summary of the Invention

[0007] The main objective of this invention is to provide a visually assisted lightweight deflection monitoring method and system, which aims to solve the technical problems of integral drift and high computational cost and power consumption in existing structural monitoring methods.

[0008] To achieve the above objectives, the present invention proposes a visually assisted lightweight deflection monitoring method, comprising: S1. Collect the inclination angle data, mid-span acceleration data, and environmental parameters of the bridge at the mid-span, L / 4, and 3L / 4 sections; S2. Vehicle passing events are detected at the edge. Based on acceleration and environmental parameters, the vision module is triggered to capture the target at the beam end and obtain the absolute displacement at the left and right supports. S3. Based on the acceleration data, perform dynamic mode truncation according to the frequency response to obtain the mode order; S4. Construct observation equations based on tilt angle data, use the absolute displacements at the left and right supports as boundary constraints, and solve the modal coordinates using the regularized least squares method to reconstruct the full bridge deflection curve. S5. If the deflection value in the reconstructed full-bridge deflection curve exceeds the preset threshold, send an early warning message.

[0009] A further improvement of the visual-assisted lightweight deflection monitoring method of the present invention is that S3 specifically includes the following steps: S301. Excitation Frequency Identification: Perform a Fast Fourier Transform on the acceleration data to extract the dominant frequency under the current vehicle load. ; S302, Modal order adaptive mapping: based on the natural frequency characteristics of a simply supported beam. To determine the modal order, a mapping relationship between frequency and modal order is established. It is the natural frequency; S303, Matrix Scale Linkage: Dynamically adjust the dimension of the shape function derivative matrix according to the determined modal order.

[0010] A further improvement of the visual-assisted lightweight deflection monitoring method of the present invention is that S302 includes the following steps: like If the motion is determined to be rigid body motion or low-frequency gradual change, and only the N=1 order mode is activated, then only the calculation is performed. At this point, matrix A degenerates into a scalar, and the computational complexity is reduced to O(1), where Where N is the fundamental frequency and N is the truncation order of the mode; like This is determined to be vehicle main frequency excitation, activating the N=2nd order mode, where It is a second-order frequency; like If the load is determined to be an impact load, the N=3 or N=4 mode is activated.

[0011] A further improvement of the visual-assisted lightweight deflection monitoring method of the present invention is that S4 specifically includes the following steps: S401. Calculate the rigid body rotation component caused by support settlement or rotation. S402. Separate the rigid body rotation component caused by the settlement or rotation of the support, construct the observation equation based on the tilt angle data, and obtain the observation vector by subtracting the measured tilt angle value from the rigid body rotation component. S403. Discretize the derivative of the deflection shape function into matrix form to obtain the shape function derivative matrix; S404. Based on the observation vector, the modal coordinate vector is obtained by introducing a regularization term to suppress noise amplification according to the derivative matrix of the shape function, and then solving it by regularized least squares method. S405. Reconstruct the full-bridge deflection curve based on the modal coordinate vector.

[0012] A further improvement of the visual-assisted lightweight deflection monitoring method of the present invention is that the calculation formula for S401 is as follows: ; in: Let the rigid body rotation component be... Absolute displacement at the left support Where L is the absolute displacement at the right support, and L is the calculated span of the bridge. For time.

[0013] A further improvement of the visual-assisted lightweight deflection monitoring method of the present invention is that the calculation formula for the observation vector in S402 is as follows: ; in: For the observation vector, For the actual measured inclination angle at mid-span, , These are measured tilt angle data.

[0014] A further improvement of the visual-assisted lightweight deflection monitoring method of the present invention is that the calculation formula for S403 is as follows: ; Where: A is the matrix of derivatives of the formal functions, and the rows correspond to respectively: , , The derivative value of the shape function at that point.

[0015] A further improvement of the visual-assisted lightweight deflection monitoring method of the present invention is that the calculation formula for S404 is as follows: ; in: For modal coordinate vectors, The regularization coefficient is . Let A be the transpose of the matrix of derivatives of the shape functions. It is an identity matrix.

[0016] A further improvement of the visual-assisted lightweight deflection monitoring method of the present invention is that the calculation formula for S405 is as follows: ; in: For the total bridge deflection, The longitudinal coordinates of the bridge are... For modal order hour Modal coordinate vector at time step, This is the index of the modal order.

[0017] The present invention also provides a system applying the vision-assisted lightweight deflection monitoring method described above, comprising: A cross-sectional integrated sensing node is configured to collect tilt and acceleration data in the cross-sectional area. The vision module includes passive reflective targets deployed at the supports at both ends of the bridge, used to construct an absolute displacement reference system; The edge computing module is used to decouple the full bridge deflection into a rigid body displacement term and an elastic deformation term. The rigid body displacement term is obtained by the observation of the vision module. Based on the tilt angle data, the observation equation is constructed, and the absolute displacement at the left and right supports is used as the boundary constraint. The modal coordinates are solved by regularized least squares method to reconstruct the full bridge deflection curve. If the deflection value in the reconstructed full bridge deflection curve exceeds a preset threshold, an early warning information is sent.

[0018] The technical solution of the present invention has the following beneficial effects: The visual-assisted lightweight deflection monitoring method provided by this invention no longer attempts to "eliminate errors," but rather avoids errors through mathematical reconstruction. The core innovation lies in the inversion paradigm of "multi-source data fusion + regularization constraints." This invention is based on the Euler-Bernoulli beam theory of simply supported beams, where the shape of the deflection curve of a simply supported beam under moving loads is known to be fixed. By adding auxiliary tilt angle probes at L / 4 and 3L / 4, the physical blind zone of the zero slope of the first-order mode at the mid-span of the simply supported beam is effectively avoided, solving the problem of unreliable single-point sensing. The absolute displacement of the target at the beam end is obtained through a low-power vision module as a boundary constraint, transforming the originally ill-conditioned underdetermined problem into a well-conditioned constraint optimization problem, achieving "hardware replacement with algorithm," and significantly reducing deployment costs and power consumption. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the layout of the visual-assisted lightweight deflection monitoring system of the present invention. Figure 1 ; Figure 2 This is a schematic diagram of the layout of the visual-assisted lightweight deflection monitoring system of the present invention. Figure 2 ; Figure 3 This is a flowchart of the visual-assisted lightweight deflection monitoring method of the present invention; Figure 4 This is a comparison chart of the deflection reconstruction accuracy of the visual-assisted lightweight deflection monitoring method of the present invention.

[0021] Explanation of icon numbers: 1. Cross-sectional integrated sensing node; 2. Passive reflective target; 3. Simple tilt probe. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] Furthermore, in this invention, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0024] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0025] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0026] like Figures 1-4 As shown, this invention proposes a visually assisted lightweight deflection monitoring method, comprising: S1. Collects tilt angle data, bridge mid-span acceleration data, and environmental parameters at the mid-span, L / 4, and 3L / 4 sections of the bridge. The integrated sensing node at the mid-span incorporates a triaxial accelerometer, a triaxial inclinometer, and temperature / humidity / light sensors. Additionally, simplified tilt probes 3 are added at the L / 4 and 3L / 4 sections of the bridge. These simplified tilt probes 3 are connected to the integrated sensing node at the mid-span for unified processing, eliminating the need for independent power supply and communication modules, effectively reducing system complexity and power consumption. Environmental parameters include temperature, humidity, and light intensity (for adaptive visual exposure). Environmental parameters are collected by sensors built into the integrated sensing node 1. Temperature / humidity is collected using a digital temperature and humidity sensor (e.g., SHT35). Light intensity is collected using a photoresistor or ambient light sensor (e.g., BH1750) for the visual module to automatically adjust exposure parameters. The integrated sensing node 1 at the mid-span also includes a triaxial accelerometer and a triaxial inclinometer for collecting acceleration and tilt angle data. The integrated sensing node 1 at the mid-span also includes an AI processor for computational processing. The mid-span integrated sensing node 1 integrates the above components inside an integrated box installed on the mid-span web plate. It can operate with a built-in battery / solar energy, without the need for full-bridge wiring. It adopts an event-triggered mechanism and has an average power consumption of ≤12mW.

[0027] S2. Vehicle event detection is performed at the edge. Based on acceleration and environmental parameters, the vision module is triggered to capture images of the beam end target and obtain the absolute displacement at the left and right supports. The trigger threshold is an RMS (root mean square) value of acceleration > 0.05 g, where g is the acceleration due to gravity. The beam end target is a highly reflective sticker (passive target, reflectivity ≥ 80%) pasted on the side of the abutment or cap beam, requiring no power supply.

[0028] S3. Based on the acceleration data and frequency response, perform dynamic modal truncation to obtain the modal order; based on the spectral analysis results of the acceleration data, adaptively adjust the modal order N (N=1 for low vehicle speed, N=3 for impact load). Specifically, this includes the following steps: S301. Excitation Frequency Identification: Perform a Fast Fourier Transform on the acceleration data to extract the dominant frequency under the current vehicle load. ; S302, Modal order adaptive mapping: based on the natural frequency characteristics of a simply supported beam. To determine the modal order, a mapping relationship between frequency and modal order is established, where: The natural frequency is given by the formula for the natural frequency of a simply supported beam. Pre-calculated fundamental frequency and second-order frequency ,in: For bending stiffness, Linear density, For the first angular frequency, The ordinal number of the modal order ( =1, 2, 3…); natural frequency (Unit: Hz) and natural angular frequency (Unit: rad / s) Satisfies the relationship .

[0029] Specifically, the following steps are included: like If the motion is determined to be rigid body motion or low-frequency gradual change, and only the N=1 order mode is activated, then only the calculation is performed. At this point, matrix A degenerates into a scalar, and the computational complexity is reduced to O(1), where The fundamental frequency is N, and the cutoff order of the mode is N=1,2,3,4, which changes dynamically, representing the upper limit of the selected mode order; like This is determined to be vehicle main frequency excitation, activating the N=2nd order mode, where It is the second-order natural frequency; like If the load is determined to be an impact load (such as a heavy vehicle crossing a bridge), the N=3 or N=4 mode is activated.

[0030] S303, Matrix Scale Linkage: Dynamically adjust the dimension of the shape function derivative matrix according to the determined modal order.

[0031] Perform a Fourier transform on the acceleration data to extract the current dominant frequency of the bridge. If the dominant frequency falls within the fundamental frequency range, only the modal coordinate vector with modal order N=1 is activated; if it falls within a higher order range, the modal order is automatically increased to N=2 or N=3. The vibration energy of small-to-medium span bridges is mainly concentrated in low-order modes (usually 1-3). Fixing N=4 for calculation would result in unnecessary waste of computational resources.

[0032] S4. Construct observation equations based on tilt angle data, use the absolute displacements at the left and right supports as boundary constraints, and solve the modal coordinates using the regularized least squares method to reconstruct the full bridge deflection curve. S4 specifically includes the following steps: S401. Calculate the rigid body rotation component caused by support settlement or rotation; the calculation formula is as follows: ; in: Let the rigid body rotation component be... The absolute displacement at the left support (provided by the visual anchor point). The absolute displacement at the right support (provided by the visual anchor point), and L is the calculated span of the bridge. For time.

[0033] S402. Rigid body rotation components caused by settlement or rotation of the separate bearing, respectively, in the longitudinal coordinate system of the bridge. and An auxiliary simple tilt probe 3 is installed at the mid-span. An inclinometer was used to acquire measured inclination data at three points (mid-span, L / 4, and 3L / 4 sections). An observation vector was then constructed. The calculation formula is as follows: ; in: For the observation vector, For the actual measured inclination angle at mid-span, , These are measured tilt angle data.

[0034] This invention effectively avoids the physical blind spot where the slope of the first-order mode at the mid-span of a simply supported beam is zero by fusing multiple measurement points. Although the number of observation equations (3) may be less than or equal to the number of unknowns (N-order modes), the innovation of "replacing hardware with algorithms" is still achieved by introducing regularization constraints to replace traditional data redundancy.

[0035] S403. Discretize the derivative of the deflection shape function into matrix form to obtain the shape function derivative matrix; the calculation formula is as follows: ; Where: A is the matrix of derivatives of the shape functions, and each row corresponds to the following: , and The derivative value of the shape function at that point.

[0036] S404. Based on the observation vector, the modal coordinate vector is obtained by introducing a regularization term to suppress noise amplification according to the derivative matrix of the shape function, and then solving it by regularized least squares method. Constructing a regularized least squares objective function : ; in, For modal coordinate vectors, The regularization coefficient is 0.1 to 1.0. The first part ensures that the calculation results are close to the measured data, and the second part forces the magnitude of the solution vector to not be too large, so as to prevent unreasonable bridge deformation shapes from being calculated due to sensor noise.

[0037] Taking the derivative of the objective function with respect to modal coordinates and setting it to zero, we derive the analytical solution (normal equation) as follows: ; in: For modal coordinate vectors, The regularization coefficient is . Let A be the transpose of the matrix of derivatives of the shape functions. It is an identity matrix.

[0038] In particular, when N=1 or N=2, the above matrix operations are extremely small and can be completed in milliseconds, which greatly reduces the peak power consumption at the edge.

[0039] S405. Reconstruct the full-bridge deflection curve based on the modal coordinate vectors, and express the full-bridge deflection as a linear combination of rigid body displacement and elastic deformation terms, as shown in the following formula: ; in: Let x represent the total bridge deflection, and x be the longitudinal coordinate of the bridge. For modal order The modal coordinate vector at time t. This is the index of the modal order. For rigid body displacement terms, The elastic deformation term. Rigid body displacement term: describes the rigid body displacement caused by support settlement / overall rotation, measured by the vision module as the absolute displacement of the left and right supports. , Directly determined, with clear physical meaning. The rigid body rotation component of the beam caused by differential settlement at the supports varies linearly with the longitudinal coordinate x.

[0040] Elastic deformation term: describes the elastic deflection caused by the bending of the beam itself, expanded using orthogonal modes (sine basis functions) of the free vibration of a simply supported beam, and consists only of modal coordinates. Control and satisfy the boundary conditions of the simply supported beam. .

[0041] Minimize objective function The first layer is the "problem to be solved (optimization model)," and the second layer is the "analytical solution to this problem (solution tool)." These are the description and computation layers of the same mathematical process. This invention first constructs a regularized least squares objective function. A mathematical model was established to address underdeterminacy and noise issues, and then the analytical solution formula (S404) corresponding to the model was derived. In practical applications, this analytical formula can be directly used to calculate modal coordinates without iteration, thereby achieving low-power edge computing.

[0042] S5. If the deflection value in the reconstructed full-bridge deflection curve exceeds a preset threshold, a warning message is sent. A yellow warning message is issued when |w|>L / 600, and a red warning message is issued when |w|>L / 400. |w| is the absolute value of the full-bridge deflection.

[0043] The present invention also includes a system applying the vision-assisted lightweight deflection monitoring method described above, comprising: The mid-span integrated sensing node 1 is configured to collect tilt and acceleration data in the mid-span region; it also includes two sets of simple tilt probes 3, which are respectively deployed at the longitudinal coordinates of the bridge. and At this location, a wire is connected to the integrated sensing node across the mid-span to supplement the collection of tilt data at non-mid-span locations.

[0044] The vision module includes passive reflective targets 2 deployed at the supports at both ends of the bridge, which are used to construct an absolute displacement reference system. The edge computing module is used to decouple the full bridge deflection into a rigid body displacement term and an elastic deformation term. The rigid body displacement term is obtained by the observation of the vision module. Based on the tilt angle data, the observation equation is constructed, and the absolute displacement at the left and right supports is used as the boundary constraint. The modal coordinates are solved by regularized least squares method to reconstruct the full bridge deflection curve. If the deflection value in the reconstructed full bridge deflection curve exceeds a preset threshold, an early warning information is sent.

[0045] The overall average power consumption of the visual-assisted lightweight deflection monitoring system of this invention is ≤12 mW (including sleep and event triggering); normal sleep is ≤1mW; visual wake-up capture peak is ≤300mW (single time <200ms); edge computing peak is ≤150mW.

[0046] The vision module is directly mounted on the camera. It includes a perspective transformation anti-shake unit, which uses at least four auxiliary feature points around a passive reflective target 2 to calculate camera pose changes via a homography matrix, outputting a pure support displacement after deducting camera shake. and It utilizes a visual displacement measurement algorithm to convert changes in pixel points into actual structural displacement, which is an existing technology. The specific steps are as follows: 1) Feature point extraction: Select no less than four passive auxiliary feature points (such as corner points or circular markers) around the passive reflective target 2, and locate the pixel coordinates of the feature points using a sub-pixel level corner detection algorithm (such as Harris or FAST).

[0047] 2) Homography matrix calculation: Based on feature point matching between the current frame and the reference frame, the homography matrix is ​​solved by direct linear transformation (DLT) to describe the projection relationship between the camera plane and the bridge end face.

[0048] 3) Attitude Decoupling: Rotation and translation components are obtained using homography matrix decomposition. Camera attitude changes caused by wind vibration or support sway are eliminated, and the pure support displacement after deducting camera jitter is output. ,and .

[0049] Additional explanation: The calculation logic is as follows: 1) Feature point extraction and coordinate normalization: Let the pixel coordinates of the k-th feature point in the current frame be... To improve numerical stability, normalized coordinates are used: ; in: This is a similarity transformation matrix used to translate the coordinate center to the centroid and scale it to a unit scale. The pixel coordinates (in pixels) of the k-th feature point after normalization by similarity transformation are improved by shifting the centroid and scaling to a unit scale.

[0050] Similarity transformation matrix, used to transform the original pixel coordinates Convert to normalized coordinates The format is: ; in: Scaling factor , The coordinates of the centroid of the feature point set are given.

[0051] 2) Homography Matrix Calculation (DLT): Based on the matching point pairs between the current frame and the reference frame Construct a system of linear equations: ; Stack all matching points and solve for the homography matrix H in the least squares sense using singular value decomposition (SVD). (The homography matrix H is a column vector expansion with a length of 9, since H is a 3×3 matrix.) ; Where: h is the column vector expansion of H, The optimal homography matrix obtained by Singular Value Decomposition (SVD) (satisfying...) Minimum and ).

[0052] 3) Attitude decoupling and support displacement output: Decomposing the homography matrix H yields the rotation matrix R and the translation vector t: ; Where: K is the camera intrinsic parameter matrix (pre-calibrated). , is the column vector of the rotation matrix.

[0053] After removing spurious displacements caused by rotational components, the pure rigid body translation components are output. Let ΔX be the physical displacement corresponding to the homography matrix in the reference coordinate system; then the absolute displacements of the left and right supports are: ; ;in The longitudinal (along the beam length) unit vector in the bridge coordinate system is obtained by transposing the unit vector. Extract the longitudinal component of the displacement increment. However, since the support mainly undergoes vertical displacement, the vertical component can also be directly taken in practical applications (it must be consistent with the coordinate system definition); T is the matrix / vector transpose operator, the first element corresponds to the longitudinal displacement component, and the second element corresponds to the vertical displacement component. , These represent the displacement increments of the left and right supports in the bridge coordinate system. Since the rotation matrix R describes the camera attitude change, after separating it from H, the remaining translation vector t only reflects the true rigid body displacement of the bridge end face.

[0054] The vision module normally remains in sleep mode, only waking up to take pictures when the accelerometer detects a vehicle crossing the bridge (impact vibration), ensuring extremely low average power consumption. When the vision module fails to capture images multiple times consecutively, the system switches to a pure inertial prediction mode, using the previous cycle's modal coordinates and structural damping ratio to perform short-term deflection prediction and issue a maintenance alarm. The structural damping ratio ξ describes the attenuation characteristics of the bridge's free vibration. In the event of vision failure, the system uses the previous cycle's modal coordinates... And the damping ratio ξ, through the modal decay model Predict short-term deflection and maintain monitoring continuity until visual recovery. Specifically: The nth natural angular frequency, For the damping frequency, .

[0055] The invention will be illustrated below using a prestressed concrete simply supported beam bridge with a span of L=30m as an example: S1. Data Acquisition and Triggering: The AI ​​processor monitors the effective value of acceleration in real time. When RMS{a(t)} ≥ 0.05g and lasts for more than 200ms, it is determined that the vehicle is on the bridge, and the vision module is activated.

[0056] S2, Visual Anchoring: The vision module captures a passive reflective target 2 at the beam end 50ms before the vehicle reaches the mid-span. Sub-pixel-level corner detection is used to obtain the displacement of the left and right supports. , .

[0057] S3. Inversion calculation based on dynamic mode truncation: Frequency identification: Perform FFT on acceleration data to extract dominant frequencies. .like ( (where N is the fundamental frequency), only the N=1 mode is activated; otherwise, the N=3 mode is activated.

[0058] S4. Edge-end solution process and numerical stability handling: To ensure computational efficiency and numerical stability on low-power AI processors (such as the ARM Cortex-M4 series), this invention adopts the following specific implementation steps: 1) Data preprocessing and filtering: Raw tilt data collected by the integrated node This includes installation errors and temperature drift. During system power-on initialization, the average value during the first 10 minutes of static operation is recorded as the zero-point drift. For real-time monitoring, a sliding window mean filter is used: ; Meanwhile, the image acquired from the visual anchor point undergoes perspective transformation anti-shake processing: the homography matrix is ​​calculated using four auxiliary feature points around the target, eliminating camera micro-motion caused by wind vibration, and outputting the pure support displacement. and .

[0059] 2) Rigid-Elastic Decoupling: Due to the existence of support settlement or rigid body rotation in actual bridges, simple sinusoidal basis functions cannot satisfy non-zero boundary conditions. This invention decomposes the deflection into rigid body displacement (directly measured by the visual module) and elastic deformation (to be determined), with the following expression: ; Crossing China At this location, the measured inclination angle It also includes rigid body rotation components. .

[0060] 3) Construction and regularized solution of underdetermined equations: Construct a corrected observation vector b to eliminate interference from known quantities: ; Construct the derivative matrix A of the shape functions (1×N dimensions): ; Since N≥1, the equation Ac=b is an underdetermined equation. We introduce Tikhonov regularization to solve it: ; in, This is the regularization coefficient (default value is 0.5). This is an identity matrix. The calculation involves only small-scale matrix operations and can be completed in milliseconds.

[0061] S5. Result Verification and Early Warning: Suppose that at a certain moment the vehicle is located in the middle of the span: Measured tilt angle: θ = 0.85°.

[0062] Visual anchor point indication: , (Minor settlement).

[0063] The modal coordinates obtained by solving are: .

[0064] The reconstructed mid-span deflection is -14.3 mm, which, compared to the measured value of -14.5 mm by the high-precision laser displacement gauge, has a relative error of less than 1.5%. Figure 4 As shown.

[0065] Set warning thresholds: Yellow warning |w|>L / 600 (50mm), Red warning |w|>L / 400 (75mm).

[0066] S6, Visual Failure Tolerance: If visual recognition fails due to heavy fog or rain, the system switches to pure inertial prediction mode, uses the structural damping ratio (ξ) to predict modal attenuation, maintains short-term monitoring capabilities, and issues maintenance alarms.

[0067] The forecast is based on a structural dynamics attenuation model: Based on the modal coordinates of the previous period And the structural damping ratio ξ, predicting the modal coordinates within the future Δt according to the modal attenuation law. Combined with the last displacement constraint before the visual anchor point fails. , The short-term deflection field is reconstructed. If three consecutive visual captures fail, a maintenance alarm is triggered and the system switches to pure inertial mode.

[0068] This invention reduces the dimensionality of the originally infinite-dimensional continuous deflection field to ≤4 unknown coefficients, thus reducing the computational complexity from... Down to This meets the millisecond-level edge computing requirements. The passive reflective target 2 provides the absolute displacement observation value at the support. Since the tilt angle integral can only determine the "shape" of the deflection curve and cannot fix the "zero displacement reference," the vision module measures the actual displacement of the support and compares it with the laser displacement meter. The relative error is less than 1.5%, eliminating the uncertainty of the integration constant and thus achieving absolute deflection measurement. The sinusoidal function basis sin(iπx / L) in existing technologies naturally satisfies w(0)=w(L)=0, which cannot handle the actual support settlement of bridges. This invention decomposes the deflection into a rigid body displacement term (directly measured by the camera in conjunction with the passive reflective target) and an elastic deformation term (inverted from the tilt angle), achieving a perfect unity of physical meaning and mathematical model.

[0069] The above description is only a preferred embodiment of the present invention and does not limit the scope of the present invention. All equivalent structural transformations made under the concept of the present invention using the description and drawings of the present invention, or direct / indirect applications in other related technical fields, are included within the protection scope of the present invention.

Claims

1. A visually assisted lightweight deflection monitoring method, characterized in that, include: S1. Collect the inclination angle data, mid-span acceleration data, and environmental parameters of the bridge at the mid-span, L / 4, and 3L / 4 sections; S2. Vehicle passing events are detected at the edge. Based on acceleration and environmental parameters, the vision module is triggered to capture the target at the beam end and obtain the absolute displacement at the left and right supports. S3. Based on the acceleration data, perform dynamic mode truncation according to the frequency response to obtain the mode order; S4. Construct observation equations based on tilt angle data, use the absolute displacements at the left and right supports as boundary constraints, and solve the modal coordinates using the regularized least squares method to reconstruct the full bridge deflection curve. S5. If the deflection value in the reconstructed full-bridge deflection curve exceeds the preset threshold, send an early warning message.

2. The visual-assisted lightweight deflection monitoring method according to claim 1, characterized in that, S3 specifically includes the following steps: S301. Excitation Frequency Identification: Perform a Fast Fourier Transform on the acceleration data to extract the dominant frequency under the current vehicle load. ; S302, Modal order adaptive mapping: based on the natural frequency characteristics of a simply supported beam. To determine the modal order, a mapping relationship between frequency and modal order is established. It is the natural frequency; S303, Matrix Scale Linkage: Dynamically adjust the dimension of the shape function derivative matrix according to the determined modal order.

3. The visual-assisted lightweight deflection monitoring method according to claim 2, characterized in that, S302 includes the following steps: like If the motion is determined to be rigid body motion or low-frequency slow change, only the N=1th order mode is activated, where N is the cutoff order of the mode; like This is determined to be vehicle main frequency excitation, activating the N=2nd order mode, where It is a second-order frequency; like If the load is determined to be an impact load, the N=3 or N=4 mode is activated.

4. The visual-assisted lightweight deflection monitoring method according to claim 1, characterized in that, S4 specifically includes the following steps: S401. Calculate the rigid body rotation component caused by support settlement or rotation. S402. Separate the rigid body rotation component caused by the settlement or rotation of the support, construct the observation equation based on the tilt angle data, and obtain the observation vector by subtracting the measured tilt angle value from the rigid body rotation component. S403. Discretize the derivative of the deflection shape function into matrix form to obtain the shape function derivative matrix; S404. Based on the observation vector, the modal coordinate vector is obtained by introducing a regularization term to suppress noise amplification according to the derivative matrix of the shape function, and then solving it by regularized least squares method. S405. Reconstruct the full-bridge deflection curve based on the modal coordinate vector.

5. The visual-assisted lightweight deflection monitoring method according to claim 4, characterized in that, The formula for calculating S401 is as follows: ； in: Let the rigid body rotation component be... Absolute displacement at the left support Where L is the absolute displacement at the right support, and L is the calculated span of the bridge. For time.

6. The visual-assisted lightweight deflection monitoring method according to claim 5, characterized in that, The formula for calculating the observation vector in S402 is as follows: ； in: For the observation vector, For the actual measured inclination angle at mid-span, , These are measured tilt angle data.

7. The visual-assisted lightweight deflection monitoring method according to claim 6, characterized in that, The formula for calculating S403 is as follows: ； Where: A is the matrix of derivatives of the formal functions, and the rows correspond to respectively: , , The derivative value of the shape function at that point.

8. The visual-assisted lightweight deflection monitoring method according to claim 7, characterized in that, The formula for calculating S404 is as follows: ； in: For modal coordinate vectors, The regularization coefficient is . Let A be the transpose of the matrix of derivatives of the shape functions. It is an identity matrix.

9. The visual-assisted lightweight deflection monitoring method according to claim 8, characterized in that, The formula for calculating S405 is as follows: ； in: For the total bridge deflection, The longitudinal coordinates of the bridge are... For modal order hour Modal coordinate vector at time step, This is the index of the modal order.

10. A system applying the vision-assisted lightweight deflection monitoring method as described in claim 1, characterized in that, include: A cross-span integrated sensing node (1) is configured to collect tilt angle data and acceleration data of the cross-span region; The vision module includes passive reflective targets (2) deployed at the supports at both ends of the bridge, used to construct an absolute displacement reference system; An edge computing module is used to decouple the full-bridge deflection into a rigid body displacement term and an elastic deformation term, wherein the rigid body displacement term is observed by the vision module. The observation equation is constructed based on the tilt angle data, and the absolute displacement at the left and right supports is used as the boundary constraint. The modal coordinates are solved by regularized least squares method to reconstruct the full bridge deflection curve. If the deflection value in the reconstructed full bridge deflection curve exceeds the preset threshold, an early warning information is sent.

Images