A small and medium span bridge damage identification method based on computer vision scanning

Abstract

This invention discloses a method for damage identification of small-to-medium span bridges based on computer vision scanning. The method includes: first, determining the size of the scanning frame; second, performing rapid vibration testing on the bridge using computer vision scanning, completing the vibration test of the entire bridge or a key area of ​​interest while the scanning frames overlap and advance; third, identifying the structural displacement response from real-time video images of each scanning frame under high-density measurement points; and fourth, performing damage identification individually for each scanning frame by identifying modal parameters from vibration data at multiple measurement points within the scanning frame, and using the identified multi-order modal curvature fragments to identify the location of structural damage. This method combines the concept of scanning examination from medical examinations with computer vision, enabling damage identification to be completed within a single field of view. It overcomes the dual problems of limited field of view and insufficient measurement points in computer vision, and can quickly identify minute structural damage under high-density measurement points.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring, and is a method for damage identification based on computer vision to collect data from bridges by scanning. Background Technology

[0002] According to relevant data, by the end of 2022, my country had 1,033,200 highway bridges, totaling 85,764,900 meters in length. Among them, there were 8,816 extra-large bridges (16,214,400 meters) and 159,600 large bridges (44,319,300 meters), representing an increase of 276,100 bridges in the past five years. It is evident that small and medium-span bridges constitute the vast majority of the total number of bridges. Bridge construction requires substantial investment of manpower and resources, and the costs of bridge maintenance and upkeep are even greater. Therefore, how to conduct economical bridge structural health monitoring has become a research hotspot.

[0003] With the rapid development of computer vision technology, camera-based non-contact vision sensors are gradually becoming an important means of replacing traditional contact sensors for structural dynamic response measurement and structural monitoring. Traditional contact sensors are cumbersome and expensive to deploy and maintain, and are easily limited in practical engineering applications. Computer vision-based measurements, on the other hand, can achieve long-distance, non-contact operation, minimizing the impact on the structure. Furthermore, vision sensors are low-cost, easy to set up and operate, and offer the flexibility to extract the displacement of any point on the structure from a single video feed. However, because computer vision-based structural measurements are easily limited by the field of view, most current research focuses on using computer vision technology to obtain the displacement time history of key parts of the structure, while research on using computer vision technology to identify structural damage is scarce. Currently, structural damage identification is mainly accomplished by comparing changes in modal parameters before and after structural damage. Modal parameters that can be used for damage identification include natural frequencies, damping, and mode shapes. Mode shape-based damage identification is the most common method, but it requires a high density of measurement points, making vibration-based structural damage identification difficult to apply to actual bridges.

[0004] In view of the above problems, this invention integrates the concept of medical scanning with computer vision to present a novel method for damage identification of small-to-medium span bridges based on computer vision scanning. This method uses a single field of view of computer vision to achieve damage identification based on modal vibration, eliminating the limitations of field of view range and high measurement point density requirements. Compared with other computer vision-based methods for structural damage identification, it is more convenient and efficient. Summary of the Invention

[0005] This invention provides a high-density vibration testing and damage identification method for small and medium-span bridges based on computer vision scanning. It utilizes a similar scanning processing technique to perform high-density sensing measurements on small and medium-span bridges, develops new measurement techniques, and establishes a process for structural vibration response identification and damage identification, enabling non-contact health monitoring and damage identification of bridge structures.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] 1) Determine the scan frame size. The size of the computer vision scan frame should be determined based on the computer vision hardware used, the measurement environment, and the size of the damage to be identified.

[0008] 2) Conduct rapid vibration testing of the bridge using computer vision scanning. Sampling points are evenly spaced on the lateral surface of the bridge. Based on the scanning frame size, the system moves sequentially from one side of the bridge to the other, identifying damage to beam segments within the scanning frame as it scans. Finally, vibration testing of the entire bridge or key bridge areas is completed while the scanning frames overlap and advance.

[0009] 3) For the video images measured in each scanning frame, digital image correlation and other technologies are used to identify the structural displacement response of each equally spaced point in the scanning frame with high-density measurement points;

[0010] 4) Damage identification is performed individually for each scan frame. Modal parameters are identified from the vibration data of high-density measurement points within the scan frame, and the location of structural damage is identified using the obtained multi-order mode shape curvature fragments.

[0011] In step one, the size of the scanning frame is determined by the selected computer vision hardware, the measurement environment, and the degree of damage to the structure of interest. Better computer vision hardware parameters result in a larger field of view and higher measurement accuracy, leading to a larger scanning frame size. Less interference in the measurement environment ensures high measurement accuracy even with a large number of measurement points, allowing for a larger scanning frame size. Greater structural damage leads to greater changes in modal curvature at the damage location, making damage easier to identify, thus allowing for a larger scanning frame.

[0012] In step two, markers are placed at equal intervals on the bridge surface, with denser spacing in areas of suspected damage. Smaller marker intervals result in more precise data measurements within the same scan frame length, leading to higher damage identification accuracy. The camera's field of view, i.e., the size of the scan frame, is used to set the movement step size; smaller step sizes result in larger overlapping areas. The camera moves sequentially from one side of the bridge to the other, completing the test of the entire bridge while maintaining overlapping scan frame steps.

[0013] The scanning frame is divided into a core area and an edge area. The measurement points in the core area are actually used for structural damage identification, while the edge area is used to calculate the mode shape curvature of the core area to assist in damage identification. The core area must be continuous or overlapping. A continuous core area means that the core area of ​​the scanning frame is continuous during the scanning frame's movement. A repetitive core area means that the core area of ​​the scanning frame overlaps during the scanning frame's movement, implying repeated scanning of this segment. This can be performed in areas suspected of being damaged to improve the accuracy of damage identification.

[0014] In step three, the key is to identify the displacement vibration response of each equally spaced point in the scanning area under high-density measurement points. High-contrast natural markers or artificially arranged equally spaced high-contrast markers can be used to improve the recognition accuracy of the vibration displacement response algorithm.

[0015] In step three, following the method described in step two, an industrial camera is used to scan the area. The results of each scan frame are stored in the computer as files. Each video segment is processed in real time using corresponding program algorithms. The structural displacement within each scan frame is obtained using technologies such as digital image correlation. Then, by locating the marker points, the displacement time history at each sampling point is extracted.

[0016] The digital image correlation technique in step three uses the first frame of the processed video as a reference image. The i-th frame is cross-correlated with the first frame to obtain the displacement of the structure within the scanning frame. Specifically, the cross-correlation operation involves first obtaining the grayscale values ​​I(x,y) and Ii of the initial and deformed structures. d (x, y); Perform cross-correlation on the two images and calculate the correlation C between them:

[0017]

[0018] In the formula, B is the area of ​​the reference image sub-region, x and y are the pixel coordinates of the image; Δx and Δy are the positional differences between the reference sub-region and the deformed sub-region, and I and I d Let be the grayscale values ​​of the pixels in the image before and after deformation, respectively. Δx and Δy, which maximize C(Δx,Δy), represent the pixel displacements of the structure. Performing this cross-correlation operation between the deformed image and the reference image yields the displacement response of each sampling point within the scan frame.

[0019] Step four: Based on the displacement time histories of each sampling point in each scan frame obtained in step three, the multi-mode vibration patterns of the scan frame are obtained using the random subspace algorithm. Theoretically, modal curvature is the second derivative of the displacement mode, and the modal curvature of the structural vibration pattern can be approximately calculated using the central difference method. Therefore, this method obtains multi-mode vibration pattern curvature segments to identify the location of structural damage.

[0020] Taking a scanning frame n as an example, the number of measurement points obtained by the scanning frame is N. i The number of core area points is N. c The number of measurement points in the edge region is (N) i -N c Assume the identified r-th mode shape is... After normalizing the maximum value of the mode shape, calculate its mode shape curvature. Then the mode shape curvature of the r-th order mode at node i is:

[0021]

[0022] In the formula, Let be the curvature of the bridge structure at point i in the r-th mode within the nth scanning frame; Let i be the r-th mode shape of the bridge structure within the nth scanning frame; n Let be the i-th measurement point within the n-th scan frame; h is the distance between adjacent measurement points. The distance h between adjacent measurement points is related to the accuracy of the identified damage; the smaller h is, the higher the identification accuracy.

[0023] When multiple mode shapes are identified, the curvatures of the multiple mode shapes are fused using the following formula:

[0024]

[0025] Where, N m This represents the number of identified vibration modes.

[0026] The location and severity of structural damage are identified by anomalies in the mode shape curvature within each scan frame. When localized damage occurs, the mode shape curvature at that location will be significantly greater than that at other locations, thus allowing for damage identification using mode shape curvature. Since each scan continues or overlaps with a portion of the core area of ​​the previous scan frame, damage identification at segmentation points is not required. This means damage identification can be performed only on the specific mode shapes identified within each scan frame, without needing to assemble the entire scan frame for overall mode shape analysis.

[0027] Beneficial Effects: This invention utilizes computer vision technology for health monitoring of small-to-medium span bridge structures. Damage identification is achieved through scanning, with the scanning frame completing the scan in an overlapping, step-by-step manner. For each scanning frame, modal curvature obtained by using edge region parameters to assist core region parameters is used for segmented damage identification. The flexibility in scanning frame size and the method of damage identification within the scanning frame enable high-density sensing measurements under industrial camera conditions with varying parameter levels, providing an effective new method for damage identification. Attached Figure Description

[0028] Figure 1 This is a flowchart of the present invention.

[0029] Figure 2 This is a scanned schematic diagram of a simply supported beam bridge.

[0030] Figure 3 (a) is a schematic diagram of the scanning frame division, and (b) is a schematic diagram of the scanning frame of the damage unit.

[0031] Figure 4 For the identified modal shapes, (a) is the first modal shape within the first scan frame identified after structural damage, and (b) is the first modal shape diagram of the scan frame where the damage location is identified after structural damage.

[0032] Figure 5 For the identified mode curvature, (a) is the first mode curvature within the first scan frame identified after structural damage, and (b) is the first mode curvature map of the scan frame where the damage location is identified after structural damage. Detailed Implementation

[0033] This invention combines the concept of medical scanning with computer vision and applies it to bridge structure monitoring. It develops new structural recognition models and establishes a structural recognition technology process to achieve health monitoring and damage identification of bridge structures. The following detailed explanation uses a simply supported beam as an example to illustrate the technical solution of this invention. The embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0034] Step 1: Determine the scan frame size. The size of the computer vision scan frame is determined based on the computer vision hardware used, the measurement environment, and the size of the damage to be identified.

[0035] Step Two: Conduct rapid vibration testing of the bridge using computer vision scanning. Sampling points are evenly spaced on the lateral surface of the bridge. Based on the scanning frame size, the system moves sequentially from one side of the bridge to the other, identifying damage to beam segments within the scanning frame as it scans. Finally, vibration testing of the entire bridge or key bridge areas is completed while the scanning frames overlap and advance.

[0036] Step 3: For the video images measured in each scanning frame, digital image correlation and other techniques are used to identify the structural displacement response of each equally spaced point within the scanning frame with high-density measurement points;

[0037] Step 4: Perform damage identification for each scan frame individually. Modal parameter identification is performed on the vibration data from high-density measurement points within the scan frame. Using the obtained multi-order mode shape curvature fragments, the location of structural damage is identified.

[0038] Specifically, in step one, the computer vision hardware used is an MV-SUA134GC-T industrial camera with a resolution of 1280*1024, a frame rate of 211FPS, and a lens focal length of 8mm. The measurement environment is a simulated white noise environment. The dimensions of the simply supported beam are 1.68m in length and 0.06*0.003m in cross-section. The simply supported beam is divided into 42 units, each with a length of 0.04m, resulting in a total of 41 sampling points. The sampling frequency is 200Hz, and the sampling time is 5000s. The damage is set to 50% stiffness damage in the 16th unit. Therefore, the scanning frame size can be set to 0.45m, meaning that each scanning frame will contain 13 sampling points.

[0039] Step two involves attaching markers to the sampling points on the simply supported beam. These markers are manually created discrete points with high contrast. Based on the scanning frame determined in step one, a movement step size of 0.16m is set, meaning nine sampling points are overlapped each time. To ensure the segmented data more clearly expresses the damage identification effect, the two sampling points at each end of each segment are defined as the edge zone, and the nine sampling points in the middle are defined as the core zone. The sampling points at both ends are used to assist in calculating the mode shape curvature in the core zone and improve the expression of the segmented mode shape curvature. Figure 3 (a) is a schematic diagram of the scanning frame division. The core area must be continuous or overlapping. A continuous core area means that the core area of ​​the scanning frame is continuous during the scanning frame movement. A repeated core area means that the core area of ​​the scanning frame is repeated during the scanning frame movement, which means that this segment can be scanned repeatedly. This can be done in areas suspected of being damaged to improve the accuracy of damage identification.

[0040] Based on the scanning frame size, the beam is moved sequentially from one side to the other, with damage identification of the beam segment within the scanning frame performed simultaneously. Finally, the vibration test of the entire beam is completed while the scanning frames are overlapping and progressing. A schematic diagram of the experiment is shown below. Figure 2 As shown. Figure 3 (b) is a schematic diagram of the single scan frame division of the cell where the damage is located.

[0041] Step 3: Following the method described in Step 2, area scanning is performed using an industrial camera. The results of each scan frame are stored in the computer as files. Each video segment is processed in real time using corresponding program algorithms. The structural displacement within each scan frame is obtained using digital image correlation technology. Then, by locating the marker points, the displacement time history at each sampling point is extracted.

[0042] Digital image correlation (DIC) technology uses the first frame of the processed video as a reference image. The i-th frame is cross-correlated with the first frame to obtain the displacement of the structure within the scanning frame. Specifically, the cross-correlation operation involves first obtaining the grayscale values ​​I(x,y) and Ii of the initial and deformed structures. d(x, y); Perform cross-correlation on the two images and calculate the correlation C between them:

[0043]

[0044] In the formula, B is the area of ​​the reference image sub-region, x and y are the pixel coordinates of the image; Δx and Δy are the positional differences between the reference sub-region and the deformed sub-region, and I and I d Let be the grayscale values ​​of the pixels in the image before and after deformation, respectively. Δx and Δy, which maximize C(Δx,Δy), represent the pixel displacements of the structure. Performing this cross-correlation operation between the deformed image and the reference image yields the displacement response of each sampling point within the scan frame.

[0045] Step four: Based on the displacement time histories of each sampling point in each scan frame obtained in step three, the multi-mode shapes of the scan frame are obtained using the random subspace method. Theoretically, modal curvature is the second derivative of the displacement mode, and the modal curvature of the structural mode shape can be approximately calculated using the central difference method. Therefore, this method obtains multi-mode shape curvature segments to identify the location of structural damage. Taking the third scan frame as an example, the number of measurement points obtained from the scan frame is N. i =13, of which N are core area points. c =9, the number of measurement points in the edge region is (N) i -N c =4. The identified r-th mode is The mode curvature of the first-order mode at node i is:

[0046]

[0047] In the formula, Let be the curvature of the beam structure at point i in the first mode of vibration within the third scanning frame; The first mode shape of the beam structure within the third scanning frame; i 3 The i-th measurement point is located within the third scanning frame; h is the distance between adjacent measurement points, which is 0.04m in this experiment. The distance h between adjacent measurement points is related to the accuracy of the identified damage; the smaller h is, the higher the identification accuracy.

[0048] Figure 4 These are the first-order mode shape diagrams of the first scan frame and the third scan frame where the damage occurred after structural damage. Figure 5 The first scan frame after structural damage and the first mode shape curvature diagram of the third scan frame where the damage is located are shown.

[0049] As can be seen from the identified modal curvature diagrams, when local damage occurs in the structure, the curvature of the mode shapes at that location will be significantly greater than that at other locations. Therefore, mode curvature can be used to identify damage. Since each scan continues or overlaps with a portion of the core area of ​​the previous scan frame, there is no need to consider damage identification at segmentation points. That is, damage identification can be performed only on the partial mode shapes identified in each scan frame, without needing to assemble them into a complete modal shape for damage identification. Figure 5 The location of the damage can be seen.

[0050] The above describes the application of scanning technology to computer vision monitoring of bridge damage and the characteristics of the scanning frame, but the present invention is not limited to the above embodiments. All equivalent changes or modifications made according to the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A computer vision scanning-based small and medium span bridge damage identification method, characterized in that, Includes the following steps: 1) Step 1: Determine the scanning frame size. The size of the computer vision scanning frame is determined based on the computer vision hardware used, the measurement environment, and the size of the damage to be identified. 2) Step 2: Conduct rapid vibration testing of the bridge based on computer vision scanning. Sampling points are arranged at equal intervals on the lateral surface of the bridge. According to the size of the scanning frame, the sampling points are moved from one side of the bridge to the other side one by one. Damage identification of beam segments within the scanning frame is performed while scanning. Finally, the vibration test of the entire bridge or key bridge areas is completed in the state of overlapping and stepping of the scanning frames. 3) Step 3: For the video images measured in each scanning frame, digital image correlation technology is used to identify the structural displacement response of each equally spaced point within the scanning frame with high-density measurement points; 4) Step 4: Perform damage identification for each scanning frame individually, identify modal parameters of the structural displacement response of high-density measurement points within the scanning frame, and identify the location of structural damage using the obtained multi-order mode curvature fragments. In step two, the scanning frame is divided into a core area and an edge area. The measurement points in the core area are actually used for structural damage identification, while the edge area is used to calculate the mode curvature of the core area to assist in damage identification. The core area must be continuous or overlapping. A continuous core area means that the core area of ​​the scanning frame is continuous during the scanning frame movement. A repetitive core area means that the core area of ​​the scanning frame is repeated during the scanning frame movement, which means that this section can be scanned repeatedly in the suspected damage area to improve the accuracy of damage identification. The digital image correlation technique in step three uses the first frame of the processed video as a reference image. Each frame image is cross-correlated with the first frame image to obtain the displacement of the structure within the scanning frame; the cross-correlation operation specifically involves: first obtaining the grayscale values ​​of the initial structure and the deformed structure images; , Perform cross-correlation processing on the two images to calculate their correlation. : In the formula, The area of ​​the reference image sub-region, These are the pixel coordinates of the image; To reference the positional difference between the sub-region and the deformed sub-region, and Let be the grayscale values ​​of the pixels in the image before and after deformation, respectively. To obtain the maximum value It refers to the pixel displacement of the structure; The cross-correlation operation is performed between the deformed image and the reference image to obtain the displacement response of the sampling points within each scan frame. In step four, based on the structural displacement response of the sampling points within each scan frame identified in step three, the modal parameters within each scan frame are obtained through a random subspace algorithm. The location of structural damage is identified using the identified multi-order modal curvature segments. In step four, the location and severity of structural damage are identified based on the abnormal curvature of the modal shapes within each scan frame. Since each scan will continue or overlap with part of the core area of ​​the previous scan frame, there is no need to consider the problem of damage identification at the segmentation points. That is, damage identification can be performed only for the partial modal shapes identified in each scan frame without assembling them into a whole modal shape for damage identification.

2. The method for damage identification of small and medium span bridge based on computer vision scanning according to claim 1, characterized in that: In step one, the determination of the scanning frame size is related to the selected computer vision hardware conditions, the measurement environment, and the degree of damage to the structure of interest. The better the selected computer vision hardware parameters, the larger the field of view and the higher the measurement accuracy, and the larger the scanning frame size. The less interference in the measurement environment, the more measurement points can be measured while maintaining high measurement accuracy, and therefore the scanning frame size can be larger. The greater the degree of structural damage, the greater the change in modal curvature at the damage location, and the easier it is to identify the damage, so the scanning frame can be set to be larger.

3. The computer vision scanning based damage identification method for small and medium span bridges according to claim 1, characterized in that: In step two, marker points are arranged at equal intervals on the bridge surface, and denser arrangement is made in areas suspected of damage. The smaller the interval between marker points, the more precise the measured data within the same scanning frame length, and the higher the accuracy of damage identification. The field of view of the camera is the size of the scanning frame. Based on the size of the scanning frame, the movement step size is set. The smaller the movement step size, the larger the overlapping area. The camera moves from one side of the bridge to the other side one by one, and the test of the entire bridge is completed in the state of overlapping stepping of the scanning frames.

4. The computer vision scanning based damage identification method for small and medium span bridges according to claim 1, characterized in that: In step three, the key is to identify the displacement vibration response of each equally spaced point in the scanning area under high-density measurement points. High-contrast natural markers or artificially arranged equally spaced high-contrast markers can be used to improve the recognition accuracy of the vibration displacement response algorithm.

5. The computer vision scanning based method for damage identification of small and medium span bridges according to claim 1, characterized in that: The identified first order mode is The mode curvature of the first order mode at the node is calculated after normalizing the mode to its maximum value. In the formula, For the bridge structure in the first Within the first scan frame The first mode of vibration The curvature of a point; For the bridge structure in the first Within the first scan frame Mode shape; For the first The first scan frame within the first scan frame One measurement point; The distance between adjacent measuring points; It is related to the accuracy of the identified damage. The smaller the value, the higher the recognition accuracy. When multiple mode shapes are identified, the curvatures of the multiple mode shapes are fused using the following formula: wherein, represents the number of identified modes.

Images