Bridge service state monitoring and early warning method based on machine vision

Abstract

The invention discloses a bridge service state monitoring and early warning method based on machine vision, and the method comprises the steps: firstly building a bridge numerical model through numerical analysis software, and then arranging a sensor to recognize the information of passing vehicles; before monitoring is started, a standard vehicle is selected and passes through a bridge along the center line of each lane, and the influence surface of each feature point is established by recording the vertical displacement of the feature point in the vehicle driving process; according to the vertical displacement influence surface of each feature point of the bridge and the identified vehicle information, calculating the vertical displacement of each feature point by using a superposition method; and finally, calculating the correlation between the actual measurement response and the theoretical value, and when the correlation of more than three feature points is lower than 95%, triggering an alarm to remind passing vehicles. According to the method, piezoelectric sensing and machine vision technologies are fused, and accurate inversion of vehicle loads and real-time evaluation of bridge structure states are realized through axle coupling dynamic response analysis and a data dual criterion mechanism.

Description

Technical Field

[0001] The present invention relates to a method for monitoring and warning the service state of a bridge based on machine vision, and belongs to the technical field of evaluating the service state of a bridge. Background Art

[0002] Since the reform and opening up, China has made remarkable achievements in bridge construction. According to statistics, as of 2021, the number of highway bridges built in China has exceeded 9.6 million, with a total length of 73,802 km, ranking first in the world. However, with the increase in service life, more and more early bridges in China have shown performance degradation and local diseases, and even serious collapse accidents have occurred from time to time, greatly endangering people's lives and property safety and causing adverse social impacts. Therefore, the monitoring of the working state and the safety performance evaluation of bridges during the service period have become the top priorities of bridge management and maintenance in China.

[0003] The existing bridge monitoring methods mainly rely on manual regular inspections or the installation of physical sensors for data collection and analysis. However, these traditional methods have many limitations and cannot meet the safety monitoring requirements of modern bridges:

[0004] 1. Insufficient monitoring real-time performance: Manual regular inspections usually have a long cycle and low efficiency, cannot achieve real-time monitoring of bridges, are easy to miss the subtle changes in the bridge structure, and lead to potential safety hazards not being discovered in time.

[0005] 2. High cost and complex maintenance: Although the installation of physical sensors such as strain gauges and accelerometers can achieve the monitoring of bridges, the sensor layout is complex and the maintenance cost is high. Especially in long-span bridges or harsh environments, the sensor system is easily affected by the external environment, resulting in data distortion.

[0006] 3. Limited coverage: Traditional monitoring means are difficult to cover multiple key parts of the bridge at the same time, especially the local structural deformation or fatigue damage is difficult to capture. Some monitoring points may not be able to represent the health state of the entire bridge, thus unable to accurately evaluate the overall structure of the bridge.

[0007] 4. Insufficient data processing ability: The data collection and processing ability of existing monitoring technologies is relatively limited, and it is difficult to cope with the dynamic response of complex bridge structures under various loads. Especially on bridges with a large traffic flow, it is difficult to accurately evaluate the real-time bearing capacity of the bridge through existing technologies.

[0008] The concept of bridge health monitoring was proposed in the 1980s, aiming to automate bridge management and maintenance using advanced computer and communication sensing technologies. Since the 1990s, my country has gradually begun theoretical research and system development for bridge health monitoring. As of 2019, over 400 bridges in my country had health monitoring systems installed, including large-scale bridges such as the Tsing Ma Bridge, Humen Bridge, Jiangyin Bridge, and Donghai Bridge. Bridge health monitoring has rapidly developed in recent decades, and governments around the world have introduced relevant policies to promote the development of intelligent and digital infrastructure monitoring. China's "14th Five-Year Plan for National Infrastructure Construction" explicitly calls for strengthening the health monitoring and management of key infrastructure and promoting the intelligent and information-based development of monitoring technologies. Simultaneously, several countries have emphasized the need to promote the in-depth application of key technologies such as machine vision in various industries in their "New Generation Artificial Intelligence Development Plan." With the continuous development of emerging technologies such as artificial intelligence, the Internet of Things, and big data, bridge monitoring systems are gradually transforming towards intelligence and automation. These new technologies not only improve the efficiency of data collection and processing but also enable real-time monitoring and early warning, better ensuring the safe operation of bridges.

[0009] Building on this foundation, machine vision-based monitoring methods are gradually emerging. Leveraging high-definition cameras and image processing technology, these methods enable efficient monitoring of bridge conditions and are becoming an emerging monitoring tool. Using high-definition cameras, machine vision systems can monitor bridge displacement changes in real time, automatically capturing displacement data and component deformation information at key locations. This not only improves monitoring accuracy and timeliness, but also significantly reduces the workload of manual inspections, providing a scientific basis for bridge condition assessment. Furthermore, the intelligent development of bridge monitoring systems has facilitated the establishment of early warning mechanisms. Once monitoring data exceeds a set safety threshold, the system promptly issues an alert, prompting relevant management personnel to take action to prevent accidents. While effectively reducing safety risks, these intelligent early warning mechanisms also provide a scientific basis for bridge operations and maintenance.

[0010] With the rapid development of intelligent technology, how to design a low-cost and accurate bridge health monitoring and early warning method based on real-time monitoring data, which can evaluate and predict the bearing capacity of the bridge in real time and issue early warning signals in time, has become a technical problem that needs to be solved urgently. Summary of the Invention

[0011] To address the issues of insufficient vehicle load identification accuracy and poor real-time structural response prediction in traditional bridge monitoring, this paper proposes a bridge service condition monitoring and early warning method based on machine vision. This method builds an intelligent monitoring system through machine vision-driven dynamic inversion of vehicle parameters, prediction of impact surface superposition response, and a data-based dual-criteria early warning mechanism. The method includes the following steps:

[0012] S1. Use numerical analysis software to establish a numerical model of the bridge. Take one side of one end of the bridge deck as the origin with coordinates (0, 0), and the coordinates of the other side of this end are (0, B), where B is the width of the bridge; the coordinates of the two sides of the other end of the bridge are (L, 0) and (L, B) respectively, where L is the length of the bridge; divide the bridge length into 10 equal parts, combine the center lines of each lane, determine each load point, apply unit loads to each load point, select 10 positions with relatively large vertical displacements under the action of each point load as characteristic points, and arrange QR code targets at the characteristic points.

[0013] S2. Arrange a right-angled trapezoidal step embedded with piezoelectric materials at the bridge entrance, and arrange a camera at a height of 35 cm from the ground (the height of the vehicle axle from the ground). Based on the piezoelectric signals excited when the vehicle passes and the vertical vibration displacement of the axle, identify the information of the passing vehicle, including vehicle type, vehicle speed, and axle weight.

[0014] S3. Before the monitoring starts, select standard vehicles and drive them across the bridge along the center lines of each lane. By recording the vertical displacements at the characteristic points during the vehicle driving process, establish the influence surface of each characteristic point.

[0015] S4. According to the influence surface of the vertical displacements of each characteristic point of the bridge and the identified vehicle information, including vehicle speed, axle weight, and axle distance, use the superposition method to calculate the vertical displacements at each characteristic point. If the vertical displacement of a certain characteristic point exceeds 5% of the design value, use tools such as alarm bells and warning lights to warn that the vehicle is not suitable to pass the bridge.

[0016] S5. Use the high-definition cameras fixed on the bridge piers and the QR code targets at each characteristic point to identify the changes in the vertical displacements at each characteristic point during the passing of the vehicle flow, obtain the time history of the changes in the vertical displacements, take 5 minutes as the time interval, calculate the correlation between the measured response and the theoretical value. When the correlation of more than three characteristic points is lower than 95%, trigger an alarm to remind the passing vehicles. In addition, when the correlation of a certain characteristic point is lower than 95% for three consecutive time periods, also trigger an alarm.

[0017] Furthermore, the specific operations of step S2 include:

[0018] S2.1. Arrange a right-angled trapezoidal step with a lower base of 2 cm, a height of 1 cm, and an upper base of 1 cm at the bridge entrance. The inclined surface faces the direction of vehicle entry. The length of the right-angled trapezoidal step is horizontally covered along the lane width, and one right-angled trapezoidal step is arranged every 20 cm, with a total of three right-angled trapezoidal steps arranged; when the vehicle passes, each axle rises along the inclined surface of the right-angled trapezoidal step and then undergoes free fall after rising to the upper base.

[0019] S2.2. The hypotenuse and the upper base of the right trapezoidal step arranged in step S2.1 are both embedded with high-strength piezoelectric materials. For example: quartz crystal. The planes of the piezoelectric materials are parallel to the corresponding surfaces, that is, the piezoelectric material embedded in the hypotenuse is placed obliquely, and the piezoelectric material embedded in the upper base is placed vertically. The piezoelectric material on the hypotenuse of the right trapezoidal step is used to measure the impact force of the vehicle tire and output piezoelectric signal 1, and piezoelectric signal 1 is affected by the axle load and vehicle speed; the piezoelectric material at the upper base is used to measure the vertical pressure of the vehicle tire and output piezoelectric signal 2, and piezoelectric signal 2 is affected by the axle load;

[0020] S2.3. High-definition cameras are arranged at both ends of the right trapezoidal step arranged in step S2.1. The cameras are located at the position of the middle right trapezoidal step, 35 cm above the ground, and are used to collect the deformation of the tire when passing through the step and the bounce of the axle after landing;

[0021] S2.4. A high-definition camera is arranged above the bridge to capture the outer contour of the moving vehicle and the vehicle speed v. Combining the number of signals generated by the piezoelectric material in step S2.2, the number of axles of the vehicle is determined; based on the time difference Δt of the piezoelectric signal and the vehicle speed v, the vehicle wheelbase is determined. Finally, combining the number of axles, wheelbase and vehicle outer contour of the vehicle, using the trained classification network, the type of the moving vehicle is determined;

[0022] S2.5. The input parameters of the classification network described in step s2.4 are the number of axles, wheelbase, and vehicle outer contour dimensions, including vehicle length, width, and height. The output parameter of the classification network is the vehicle type. A mapping relationship between the input parameters and output parameters is established using the trained recurrent neural network RNN;

[0023] S2.6. Based on the bounce situation after free fall after rising along the inclined plane of the right trapezoidal step and reaching the upper base, a vertical displacement curve of the axle is drawn. This vertical displacement curve of the axle is a free vibration decay curve. By reading two peaks, with a difference of m periods between the front and the back, dividing the two read peaks and calculating their natural logarithm, it is used as the vehicle weight coefficient. The heavier the vehicle weight, the greater the vibration decay, and the greater the vehicle weight coefficient.

[0024]

[0025] Where: ζ: represents the vehicle weight coefficient here and is related to the vehicle weight. The heavier the vehicle weight, the faster the vibration decay, and the greater the vehicle weight coefficient ζ; m: the number of vibration periods; T d : the vibration period, with the time in seconds (s); x i : the peak vertical displacement at the i-th vibration period; The peak displacement after m vibration periods.

[0026] S2.7. Select multiple vehicles, measure the axle weights of the axles of each vehicle, control the vehicle speed to pass the bridge, collect multiple groups of signals, use the vehicle speed, piezoelectric signal 1, piezoelectric signal 2, and vehicle weight coefficient as inputs, and the vehicle axle weight as the output to establish the mapping relationship between each parameter and the vehicle axle weight, so as to be used for the rapid estimation of the moving vehicle axle weight of the bridge.

[0027] Further, the value condition of m in the step S2.6 is: such that the two peaks are approximately half different.

[0028] Further, the specific method of the step S3 is as follows:

[0029] S3.1. According to the number of lanes of the bridge, select the same number of identical vehicles, weigh the vehicles to ensure that the vehicle weights of each vehicle are equal;

[0030] S3.2. Take one side of one end of the bridge as the origin with coordinates (0,0), the coordinates of the other side of this end are (0,B), B is the width of the bridge, and the coordinates of both sides of the other end of the bridge are (L,0) and (L,B) respectively, where L is the length of the bridge;

[0031] S3.3. Select one of the vehicles and drive it from one end to the other end along one of the lanes at vehicle speeds of 10 km / h, 20 km / h, 50 km / h, and 100 km / h respectively, and then return from the other end; use a camera to record the position of the vehicle on the bridge. The distance of the vehicle from the origin is x1, the distance of the center line of the lane from the origin is y1, and select the vertical displacement response at the characteristic point as zgo1, so as to establish the mapping model zgo1~(x1,y1) between the response zgo1 and the vehicle coordinates (x1,y1);

[0032] S3.4. To eliminate the influence of the longitudinal force along the bridge caused by the vehicle, the vehicle returns in the lane in step S3.3. There is a difference between the mapping model zback1~(x1,y1) of the vertical displacement response zback1 at the characteristic point and the vehicle abscissa (x1,y1) due to the influence of the longitudinal force along the bridge caused by the vehicle. Average zgo1~(x1,y1) and zback1~(x1,y1) to obtain the mapping model z1~(x1,y1) between the vertical displacement z1 at the characteristic point and the vehicle coordinates (x1,y1);

[0033] S3.5. Select the remaining lanes and repeat steps S3.3~S3.4 to obtain the influence lines of the vertical displacement at the selected characteristic point in the remaining lanes;

[0034] S3.6. Interpolate in the transverse direction of the bridge to obtain the mapping relationship between the vertical displacement response z at the selected characteristic points and any point (x, y) on the bridge deck. Then, considering the vehicle weight w, obtain the influence surface z / w ~ (x, y) of the vertical displacement response at the characteristic points;

[0035] S3.7. Select half of the vehicles and arrange them on the upstream side or the downstream side. The vehicles drive side by side and travel from one end to the other end; then switch to the other side and again drive side by side from one end to the other end. Use a camera to record the positions of the vehicles on the bridge and record the vertical displacement responses at the characteristic points to verify the accuracy of the influence surface in step S3.6;

[0036] S3.8. Select all the vehicles and arrange them on the upstream side and the downstream side respectively. The vehicles drive side by side and the vehicles on both sides drive towards each other and travel from one end to the other end. Use a camera to record the positions of the vehicles on the bridge and record the vertical displacement responses at the characteristic points to verify the accuracy of the influence surface in step S3.6;

[0037] When selecting vehicles, mainly use two-axle vehicles. The wheelbase of the vehicles should be the same, and the vehicle load distribution should be balanced to avoid too large a difference in the axle weights of the two axles. In addition, when the vehicle is driving in the lane, the vehicle should drive along the center line of the lane.

[0038] The beneficial effects of the present invention compared with the prior art are as follows:

[0039] 1. The method for monitoring and warning the service status of bridges based on machine vision of the present invention establishes a bridge numerical model through numerical analysis software and combines machine vision technology to monitor the bridge status in real time, realizing the organic combination of theoretical analysis and actual monitoring. This method can accurately identify the vertical displacement changes of bridge characteristic points and give early warnings in time when the vertical displacement exceeds 5% of the design value, significantly improving the accuracy and efficiency of bridge safety management.

[0040] 2. The present invention adopts a right-angled trapezoidal step embedded with piezoelectric materials and a high-definition camera. Through the collaborative analysis of piezoelectric signals and axle vibration displacements, it realizes the rapid and accurate identification of passing vehicle types, vehicle speeds, and axle weights. This innovative method provides accurate data support for bridge load analysis and greatly improves the intelligent level and response speed of the monitoring system.

[0041] 3. By establishing the influence surface of the vertical displacement at the characteristic points through standard vehicle driving tests and using the superposition method to calculate the stress state of the bridge in real time when the vehicle passes, the present invention provides a reliable benchmark for the dynamic monitoring of bridges. This method can quickly evaluate the service status of the bridge when the vehicle passes and give early warnings in case of abnormalities, effectively ensuring the safety of the bridge structure.

[0042] 4. The present invention utilizes a high-definition camera and a QR code target to monitor the vertical displacement changes at characteristic points, and realizes multi-dimensional monitoring through correlation analysis. When the correlation of more than three characteristic points is lower than 95%, or the correlation of a certain characteristic point is lower than 95% for three consecutive time periods, the system triggers an alarm. This multi-dimensional monitoring and intelligent alarm mechanism ensure the comprehensive control of the bridge state and reduce the risks caused by unnoticed local anomalies.

[0043] 5. Based on the model update mechanism of measured data, the present invention can optimize the monitoring model in real time according to the actual usage conditions and environmental changes of the bridge, ensuring the accuracy and adaptability of the system. This method is applicable to bridges of different types and service states, has strong generality and practical value, and provides technical support for the long-term safety management of bridges. Description of the Drawings

[0044] Figure 1 Overall framework diagram of the bridge service status monitoring and early warning method based on machine vision;

[0045] Figure 2 Schematic deployment diagram of the bridge service status monitoring and early warning system based on machine vision;

[0046] Figure 3 Side view of the right-angled step arrangement;

[0047] Figure 4 Top view of the right-angled step arrangement;

[0048] Figure 5 Flow chart of the vertical displacement early warning of characteristic points;

[0049] Figure 6 Flow chart of the correlation early warning between measured response and theoretical value;

[0050] The figure includes 1. High-definition camera A; 2. Right-angled trapezoidal step; 3. Standard vehicle; 4. QR code target; 5. High-definition camera B. Detailed Implementation Modes

[0051] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the following embodiments will specifically describe the bridge service status monitoring and early warning method based on machine vision of the present invention in conjunction with the drawings. The specific steps of the bridge service status monitoring and early warning method based on machine vision are as Figure 1 shown.

[0052] The first step: Establishment of the bridge numerical model

[0053] Use numerical analysis software (such as ANSYS or ABAQUS) to construct a three-dimensional finite element model of the bridge. Take one side of the bridge deck at one end of the bridge as the coordinate origin (0, 0), and the other side of this end as (0, B), where B is the width of the bridge; the two sides at the other end of the bridge are (L, 0) and (L, B) respectively, and L is the length of the bridge. The X-axis is along the length direction of the bridge, the Y-axis is along the width direction of the bridge, and the Z-axis is the vertical direction.

[0054] Divide the bridge evenly into 10 equal parts along the X-axis (the length direction of the bridge) to form 10 cross-sections with equal spacing. Each cross-section is further subdivided into grids in combination with the position of the lane center line to determine a series of load points. These load points are usually located on the lane center line and are used to simulate the stress condition of the bridge when the vehicle passes.

[0055] Apply a unit vertical load (such as 1 kN) at each load point, and then calculate the vertical displacement distribution of the bridge under these loads through finite element analysis. The displacement distribution reflects the stiffness and sensitivity of the bridge at different positions. By comparing the displacement responses caused by each load point, find the 10 positions with the largest vertical displacement as characteristic points. These points are usually located in the key areas of the bridge, such as the mid-span (the middle part of the bridge, where the stress is the greatest), near the supports (the supporting points, where the displacement is more sensitive), or other structurally weak areas. These characteristic points are the focus of subsequent monitoring.

[0056] Step 2: System Deployment

[0057] The system deployment of the bridge service status monitoring and early warning method based on machine vision is as Figure 2 shown.

[0058] Deploy a high-definition camera A1 above the bridge (such as fixed on the pier or crossbeam through a bracket) to capture the outer contour (such as length, width, height) of the driving vehicle 3 and the real-time vehicle speed v. The vehicle speed is calculated through the displacement between consecutive frames of the vehicle and the time difference. The camera A1 needs to have a high frame rate (such as 60 frames per second) and high resolution (such as 1080p or higher) to ensure accurate identification of its characteristics even when the vehicle passes at high speed.

[0059] Arrange QR code targets 4 at the positions of the 10 characteristic points selected in the first step. These targets 4 are the key to machine vision recognition. Each target has a unique code (such as a QR code) to facilitate accurate positioning and tracking by the camera. The targets 4 are usually fixed on the bridge surface to ensure that their position changes can be clearly captured when the vehicle passes. The size and material of the QR code targets 4 need to be selected according to the bridge environment. For example, they are made of corrosion-resistant and wear-resistant metal or plastic, with a size of about 10 cm × 10 cm, ensuring that they can be recognized by the long-distance camera.

[0060] At the bridge entrance, three right trapezoidal steps 2 are arranged horizontally along the lane width direction. The geometric parameters of each step 2 are as follows: the lower base is 2 cm, the height is 1 cm, the upper base is 1 cm, and the inclined plane faces the vehicle driving-in direction. The step spacing is 20 cm to ensure that the vehicle tires pass through one by one. The hypotenuse and the upper base of the right trapezoidal step 2 are both embedded with high-strength piezoelectric materials to analyze the dynamic characteristics of the vehicle through piezoelectric signals, such as axle weight and vehicle speed. The placement methods of the piezoelectric materials on the two sides are as follows:

[0061] Hypotenuse piezoelectric material: It is placed obliquely to measure the impact force of the vehicle tire and output piezoelectric signal 1. The signal size is affected by the axle weight and vehicle speed. For example, a high-speed and heavy-load vehicle will generate a greater impact force.

[0062] Upper base piezoelectric material: It is placed vertically to measure the vertical pressure of the tire and output piezoelectric signal 2, which mainly reflects the axle weight.

[0063] Camera B5 is installed at both ends of the right trapezoidal step, 35 cm above the ground, and is used to capture the tire deformation when the vehicle passes the step and the bounce after the axle lands. Camera B5 needs to have a high-speed shooting function (such as 120 frames per second) to record the dynamic response of the axle (such as amplitude and frequency), and these data are used for subsequent axle weight estimation.

[0064] Step 3: Establish the influence surface of each characteristic point

[0065] Before the monitoring starts, select a standard vehicle 3 (such as a truck with a load of 5 tons), and drive from one end of the bridge to the other end along the center line of each lane at different vehicle speeds (10 km / h, 20 km / h, 50 km / h, 100 km / h), and then return.

[0066] Use camera A1 to record the real-time position of the vehicle (x1 is the distance of the vehicle from the origin, and y1 is the distance of the lane center line from the origin), and at the same time record the vertical displacement responses of 10 characteristic points: zgo1 for the forward journey and zback1 for the return journey. Based on the data of the forward and return journeys, establish the relationship models of zgo1(x1,y1) and zback1(x1,y1) respectively. To eliminate the influence of the vehicle's longitudinal force (such as acceleration and deceleration), take the average of the two to obtain the mapping model z1~(x1,y1) of the vertical displacement z1 of the characteristic point and the vehicle position (x1,y1). Interpolate in the transverse direction of the bridge to obtain the mapping relationship between the vertical displacement response z at the selected characteristic point and any point (x,y) on the bridge deck, and then consider the vehicle weight w to obtain the influence surface z / w~(x,y) of the vertical displacement response at the characteristic point.

[0067] To verify the accuracy of the influence surface, two working conditions are adopted for verification. The first working condition: half of the standard vehicles drive side by side, pass through the bridge from the upstream side or the downstream side, and then switch to the other side to repeat; The second working condition: all standard vehicles drive relatively on the upstream side and the downstream side respectively, and record the vehicle positions and the vertical displacements of the characteristic points. Compare the measured vertical displacements with the vertical displacements predicted based on the influence surface. If the error is less than a certain threshold (such as 5%), the influence surface is considered accurate.

[0068] Step 4: Predict the vertical displacements of each characteristic point

[0069] The early warning flowchart of the vertical displacement of the characteristic point is as Figure 5 shown.

[0070] According to the outer contour size and vehicle speed v of vehicle 3 captured by the high-definition camera A1, combined with the number of signals generated by the piezoelectric material, determine the number of vehicle axles; Based on the time difference Δt of the piezoelectric signals and the vehicle speed v, determine the vehicle wheelbase. Finally, input the number of axles, wheelbase and outer contour into the recurrent neural network (RNN) to output the vehicle type (such as car, truck, bus).

[0071] Based on the vertical displacement curve of the axle (free vibration decay curve) recorded by the high-definition camera B5, select two peaks with a difference of m cycles (m makes the peak value halved), and calculate the natural logarithm as the vehicle weight coefficient.

[0072] Through the measured data of multiple vehicles (vehicle speed, piezoelectric signal 1, piezoelectric signal 2, vehicle weight coefficient and axle weight), establish an axle weight estimation model, so as to be used for the rapid estimation of the axle weight of moving vehicles on the bridge.

[0073] According to the influence surface model and vehicle information (vehicle speed, axle weight, wheelbase), use the superposition method to calculate the vertical displacements of each characteristic point. If the vertical displacement of a certain characteristic point exceeds 5% of the design value, use tools such as alarm bells and warning lights to warn that the vehicle is not suitable to pass through the bridge.

[0074] Step 5: Calculate the correlation between the measured response and the theoretical value

[0075] The early warning flowchart of the correlation between the measured response and the theoretical value is as Figure 6 shown.

[0076] Use the high-definition cameras fixed on the bridge piers and the QR code targets 4 at each characteristic point to identify the changes in the vertical displacements at each characteristic point during the passing of the vehicle flow, and obtain the time history of the vertical displacement changes. Taking 5 minutes as the time interval, calculate the correlation between the measured response and the theoretical value. When the correlation of more than three characteristic points is lower than 95%, trigger an alarm to remind passing vehicles. In addition, when the correlation of a certain characteristic point is lower than 95% for three consecutive time periods, an alarm is also triggered.

[0077] The following aspects need to be noted in this implementation plan:

[0078] a) Equipment protection and durability

[0079] In the bridge environment, high-definition cameras and piezoelectric materials on right-angled trapezoidal steps are exposed to harsh conditions such as wind, rain, dust, high temperature or low temperature for a long time and are easily damaged. To ensure the stable operation of the equipment, it is recommended to install waterproof and dustproof protective covers for high-definition cameras and regularly check the fixing status and protection of piezoelectric materials to avoid performance degradation caused by environmental erosion, thereby extending the service life of the equipment.

[0080] b) Camera resolution and equipment selection

[0081] When selecting a high-definition camera, its resolution needs to be given priority to ensure that details such as the outer contour of the vehicle, the QR code target and the bounce of the axle can be clearly captured. If the resolution is insufficient, it may lead to identification errors in the number of vehicle axles, wheelbase or vertical displacement of feature points, affecting the accuracy of bridge condition monitoring. In addition, high-strength and wear-resistant models of piezoelectric materials need to be selected for the right-angled trapezoidal steps to withstand the repeated impacts and pressures of vehicle tires.

[0082] c) Regular calibration and maintenance

[0083] After long-term operation, high-definition cameras and piezoelectric sensors may experience parameter drift due to factors such as vibration and temperature changes, affecting data accuracy. It is recommended to calibrate the camera once every quarter, adjust the focal length and exposure parameters, and at the same time check the sensitivity of the piezoelectric material and the stability of signal output to ensure that the monitoring system can continuously provide reliable data support.

[0084] d) Parameter adjustment to adapt to environmental changes

[0085] The bridge monitoring system needs to operate efficiently under different seasons and weather conditions (such as rainy days, foggy days, strong light or at night). Changes in light and visibility may affect the recognition effect of the camera on the QR code target and vehicle information. Therefore, it is necessary to dynamically adjust parameters such as the exposure and focal length of the camera according to the actual situation to ensure that the image quality and monitoring accuracy are not interfered by environmental factors.

[0086] e) Data security and storage management

[0087] This implementation plan involves the collection, transmission and storage of a large amount of real-time data (such as piezoelectric signals, vehicle information, vertical displacement data). To prevent data loss or tampering, encryption measures (such as SSL / TLS protocol) need to be taken to protect data transmission security. At the same time, establish a multi-point backup mechanism (such as combining cloud storage and local backup) to minimize the data risk caused by hardware failures or cyber attacks, thereby ensuring the security and reliability of the system.

[0088] The above are only the preferred embodiments of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can still be made without departing from the principle of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A method for monitoring and warning the service status of a bridge based on machine vision, characterized in that, The method includes the following steps: S1. Establish a numerical model of the bridge using numerical analysis software. Take one side of one end of the bridge deck as the origin with coordinates (0,0), and the coordinates of the other side of this end are (0,B), where B is the width of the bridge; the coordinates of the two sides of the other end of the bridge are (L,0) and (L,B) respectively, where L is the length of the bridge; divide the bridge length into 10 equal parts, combine the center lines of each lane, determine each load point, apply a unit load to each load point, select 10 positions with relatively large vertical displacements under the action of each point load as characteristic points, and arrange QR code targets at the characteristic points. S2. Arrange a right-angled trapezoidal step embedded with piezoelectric materials at the bridge entrance, and arrange a camera 35 cm above the ground. Based on the piezoelectric signals excited when the vehicle passes and the vertical vibration displacement of the axle, identify the information of the passing vehicle, including vehicle type, vehicle speed, and axle weight. S3. Before the monitoring starts, select a standard vehicle and drive it across the bridge along the center line of each lane. By recording the vertical displacements at the characteristic points during the vehicle's driving, establish the influence surface of each characteristic point. S4. According to the influence surface of the vertical displacements of each characteristic point of the bridge and the identified vehicle information, including vehicle speed, axle weight, and wheelbase, use the superposition method to calculate the vertical displacements at each characteristic point. If the vertical displacement of a certain characteristic point exceeds 5% of the design value, use tools such as alarm bells and warning lights to warn that the vehicle is not suitable to pass the bridge. S5. Use the high-definition cameras fixed on the bridge piers and the QR code targets at each characteristic point to identify the changes in the vertical displacements at each characteristic point during the passing of the vehicle flow, obtain the time history of the vertical displacement changes, take 5 minutes as the time interval, calculate the correlation between the measured response and the theoretical value. When the correlation of more than three characteristic points is lower than 95%, trigger an alarm to remind the passing vehicles. In addition, when the correlation of a certain characteristic point is lower than 95% for three consecutive time periods, an alarm is also triggered.

2. The method for monitoring and warning the service status of a bridge based on machine vision according to claim 1, wherein, The specific operations of step S2 include: S2.

1. Arrange a right-angled trapezoidal step with a lower base of 2 cm, a height of 1 cm, and an upper base of 1 cm at the bridge entrance. The inclined surface faces the direction of vehicle entry. The length of the right-angled trapezoidal step is horizontally covered along the lane width, and one right-angled trapezoidal step is arranged every 20 cm, with a total of three right-angled trapezoidal steps arranged; when the vehicle passes, each axle rises along the inclined surface of the right-angled trapezoidal step and then undergoes free fall after rising to the upper base. S2.

2. Both the inclined side and the upper base of the right-angled trapezoidal step arranged in step S2.1 are embedded with high-strength piezoelectric materials. The piezoelectric material on the inclined side of the right-angled trapezoidal step is used to measure the impact force of the vehicle tire and output piezoelectric signal 1, and piezoelectric signal 1 is affected by the axle weight and vehicle speed; the piezoelectric material at the upper base is used to measure the vertical pressure of the vehicle tire and output piezoelectric signal 2, and piezoelectric signal 2 is affected by the axle weight. S2.

3. Arrange high-definition cameras at both ends of the right-angled trapezoidal step arranged in step S2.

1. The cameras are located at the position of the middle right-angled trapezoidal step, 35 cm above the ground, and are used to collect the deformation of the tire when passing the step and the bounce of the axle after landing. S2.

4. Arrange high-definition cameras above the bridge to capture the outer contour and vehicle speed v of the moving vehicle. Combine the number of signals generated by the piezoelectric material in step S2.2 to determine the number of vehicle axles. Based on the time difference Δt of the piezoelectric signals and the vehicle speed v, determine the vehicle wheelbase. Finally, combine the number of vehicle axles, wheelbase, and vehicle outer contour, and use the trained classification network to determine the type of the moving vehicle; S2.

5. The input parameters of the classification network described in step S2.4 are the number of axles, wheelbase, and vehicle outer contour dimensions, including vehicle length, width, and height. The output parameter of the classification network is the vehicle type. Use the trained recurrent neural network (RNN) to establish the mapping relationship between the input parameters and output parameters; S2.

6. Based on the rise along the inclined plane of the right trapezoidal step, after rising to the upper base, conduct the bouncing situation after free fall and landing, and draw the vertical displacement curve of the axle. This vertical displacement curve of the axle is a free vibration decay curve. By reading two peaks, with a difference of m cycles between the front and the back, divide the two read peaks and calculate the natural logarithm of the result, which is used as the vehicle weight coefficient. The heavier the vehicle weight, the greater the vibration decay, and the larger the vehicle weight coefficient; Where: ζ: Here it represents the vehicle weight coefficient, which is related to the vehicle weight. The greater the vehicle weight, the faster the vibration attenuation, and the greater the vehicle weight coefficient ζ; m: The number of vibration cycles; T d : The vibration period, with the time in seconds (s); x i : The peak vertical displacement at the i-th vibration cycle; The peak displacement after m vibration cycles; S2.

7. Select multiple vehicles, measure the axle weight of each vehicle axle, control the vehicle speed to pass the bridge, collect multiple groups of signals, and use the vehicle speed, piezoelectric signal 1, piezoelectric signal 2, and vehicle weight coefficient as inputs and the vehicle axle weight as the output to establish the mapping relationship between each parameter and the vehicle axle weight, so as to be used for the rapid estimation of the axle weight of moving vehicles on the bridge.

3. The method for monitoring and warning the service status of a bridge based on machine vision according to claim 2, wherein, The value condition of m is: making the difference between the two peaks be half.

4. The method for monitoring and warning the service status of a bridge based on machine vision according to claim 1 or 2 or 3, characterized in that, The specific method of step S3 is as follows: S3.

1. According to the number of bridge lanes, select the same number of identical vehicles and weigh the vehicles to ensure that the vehicle weights of each vehicle are equal; S3.

2. Take one side of one end of the bridge as the origin with coordinates (0,0), and the coordinates of the other side of this end are (0,B), where B is the width of the bridge. The coordinates of the two sides of the other end of the bridge are (L,0) and (L,B) respectively, where L is the length of the bridge; S3.

3. Select one vehicle and drive it from one end to the other end along one lane at vehicle speeds of 10 km / h, 20 km / h, 50 km / h, and 100 km / h respectively, and then return from the other end. Use a camera to record the position of the vehicle on the bridge. The distance of the vehicle from the origin is x1, and the distance of the center line of the lane from the origin is y1. Select the vertical displacement response at the characteristic point as zgo1, so as to establish the mapping model zgo1~(x1,y1) between the response zgo1 and the vehicle coordinates (x1,y1); S3.

4. To eliminate the influence of the longitudinal force along the bridge caused by the vehicle, the vehicle returns to the lane in step S3.

3. For the mapping model zback1~(x1, y1) of the vertical displacement response zback1 at the characteristic point and the vehicle abscissa (x1, y1), due to the influence of the longitudinal force along the bridge caused by the vehicle, there are differences between zgo1~(x1, y1) and zback1~(x1, y1). Average zgo1~(x1, y1) and zback1~(x1, y1) to obtain the mapping model z1~(x1, y1) of the vertical displacement z1 at the characteristic point and the vehicle coordinates (x1, y1); S3.

5. Select the remaining lanes and repeat steps S3.3 - S3.4 to obtain the influence line of the vertical displacement at the selected characteristic point in the remaining lanes; S3.

6. Interpolate in the transverse direction of the bridge to obtain the mapping relationship between the vertical displacement response z at the selected characteristic point and any point (x, y) on the bridge deck. Then, considering the vehicle weight w, obtain the influence surface z / w~(x, y) of the vertical displacement response at the characteristic point; S3.

7. Select half of the vehicles and arrange them on the upstream side or the downstream side. The vehicles drive side by side and travel from one end to the other end; then switch to the other side and drive side by side from one end to the other end again. Use a camera to record the position of the vehicle on the bridge and record the vertical displacement response at the characteristic point to verify the accuracy of the influence surface in step S3.6; S3.

8. Select all the vehicles and arrange them on the upstream side and the downstream side respectively. The vehicles drive side by side and the vehicles on both sides drive towards each other and travel from one end to the other end. Use a camera to record the position of the vehicle on the bridge and record the vertical displacement response at the characteristic point to verify the accuracy of the influence surface in step S3.6.

Images