An Information Management Method for High-Strength Bolts in Steel Bridges

Abstract

An information management method for high-strength bolts in steel bridges is proposed. This method involves collecting a dataset, building a bolt center point recognition model based on a fully convolutional neural network, and mounting the trained model onto a small computing unit. Before final tightening, the overall image of the bolt node panel is input into the model to identify the bolt distribution on the node plate. A QR code is scanned to confirm the correspondence between the QR code, the scan result, and the model's recognition result. For each bolt being tightened, a corresponding image is collected, and the row and column number of the bolt being tightened are identified. The bolt image is then input into a high-strength bolt construction management system for individual mapping, along with its corresponding number, final tightening angle, final tightening torque, and node plate number. This invention addresses the problems of lack of overall management, lack of correlation, and poor traceability in on-site construction quality management, enabling rapid querying of bolt-by-bolt construction information for high-strength bolts in steel bridges during construction and operation phases.

Description

Technical Field

[0001] This invention belongs to the field of steel bridge construction management and relates to a method for information management of steel bridge bolts; more specifically, it relates to an information management method for high-strength steel bridge bolts based on computer vision and an intelligent torque-controlled electric wrench. Background Technology

[0002] Steel structures have advantages such as light weight, strong load-bearing capacity, modular construction, and good recyclability, and their use is becoming increasingly widespread in the construction of various bridges. High-strength bolt connections are gradually replacing traditional riveting and welding methods due to their advantages of simple construction, replaceability, and high load-bearing capacity, becoming the main connection method for steel bridge nodes. Problems such as loosening and breakage of high-strength bolts are closely related to the quality of on-site installation. Traditional high-strength bolt construction information is mainly based on paper records, with low levels of informatization in process control, inability to obtain bolt tightening torque, and inability to achieve efficient management and bolt information traceability. It is also difficult to obtain tightening information during the operation and maintenance phases.

[0003] When using only intelligent torque-controlled electric wrenches for construction, the bolt tightening sequence may be inconsistent due to factors such as worker convenience during actual construction. This can lead to a problem where the final tightening torque and final tightening rotation angle recorded in the high-strength bolt construction management system cannot be matched with each bolt being tightened in the later management stage. Therefore, there is an urgent need for information-based management of high-strength bolts in steel structure bridges. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention provides an information management method for high-strength bolts in steel bridges. This method aims to address the problem that inconsistent bolt tightening sequences during actual construction, caused by factors such as worker convenience, prevent the recording of final tightening torque and rotation angle in the high-strength bolt construction management system from being matched with each bolt individually. It also aims to solve problems such as a lack of overall coordination, lack of correlation, and poor traceability in on-site construction quality management, enabling rapid querying of bolt-by-bolt construction information during the construction and operation phases of high-strength bolts in steel bridges.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] An information-based management method for high-strength bolts in steel bridges includes the following steps:

[0007] 1) Image acquisition and preprocessing: Acquire images of steel bridge bolt node plates, use Photoshop image processing software to crop each image to the set size, mark the center point of the original steel bridge bolt head, integrate the Mask, JSON file and original steel bridge bolt images into a steel bridge bolt image dataset, and then perform data augmentation operation on the steel bridge bolt image dataset to obtain the enhanced steel bridge bolt image dataset.

[0008] 2) Building a bolt center point recognition model: Build a target center point network on the Keras deep learning framework based on Tensorflow, and pre-train the model using the COCO dataset until the bolt center point recognition model converges;

[0009] 3) Training the bolt hole recognition model: The enhanced dataset is divided into a training set and a validation set according to a set ratio. The training set is input into the pre-trained model for training. The mean square error (MSE) function is used as the objective function of the model. A dynamic learning rate is used, and Adam is selected as the optimizer to train the model. The best model is selected using the validation set and then mounted into the small computing unit at the lower end of the intelligent constant torque electric wrench.

[0010] 4) Use the mobile app connected to the wrench to scan the QR code and identify the entire node plate: After the initial tightening is completed, scan the personnel QR code and the construction work order. Then, use the camera on the upper part of the smart torque electric wrench to take an overall image of the bolt node plate. Input the image into the bolt center point recognition model for recognition, output the number of bolts in the current node plate, compare and confirm with the QR code scanning result, and number each bolt.

[0011] 5) Final tightening and photo taking: Use the intelligent torque-controlled electric wrench to final tighten the bolt. While pressing the tightening button, the camera on the top of the wrench takes a photo.

[0012] 6) Current bolt number identification: Input the photo taken by the camera into the small calculation unit at the bottom of the wrench to identify the position number of the bolt being tightened;

[0013] 7) Recording current bolt tightening information: After tightening each bolt, use a green frame to outline the bolt position in the previously captured overall image until all bolts on the node plate are outlined; transmit the current tightening image, the overall image after the green frame selection, and information such as the bolt number, tightening rotation angle, current node plate number, and operator to the high-strength bolt construction management system; when the overall image of a node plate shows that all bolts have been selected with green frames, the indicator light on the upper part of the intelligent torque-controlled electric wrench will turn green;

[0014] Furthermore, in step 2), the bolt center point recognition model uses a fully convolutional network with a deepened ResNet50 as the feature extraction network. The network model includes:

[0015] (1) Input a 704*512 preprocessed three-channel image; where Conv Block and IdentityBlock are consistent with ResNet, and Denconv Block is a transposed convolutional kernel with size (4,4) and stride 2 with BatchNormalization and ReLU activation function;

[0016] (2) CONV1 layer: a convolutional layer with BatchNormalization and ReLU activation function, with a kernel size of 7*7, a stride of 2, and an output of 64 feature maps of size 352*256; a max pooling layer with a filter size of (3,3) and a stride of 2, and an output of 64 feature maps of size 176*128.

[0017] (3) CONV2 layer: contains one Conv Block and two Identity Blocks. The Conv Block outputs 256 feature maps of size 256*128; the outputs of the two Identity Blocks are both 256 feature maps of size 176*128.

[0018] (4) CONV3 layer: contains one Conv Block and three Identity Blocks. The Conv Block outputs 512 feature maps of size 88*64; the outputs of the three Identity Blocks are all 512 feature maps of size 88*64.

[0019] (5) CONV4 layer: contains one Conv Block and five Identity Blocks. The Conv Block outputs 1024 feature maps of size 44*32; the outputs of the five Identity Blocks are all 1024 feature maps of size 44*32.

[0020] (6) CONV5 layer: contains one Conv Block and two Identity Blocks. The Conv Block outputs 2048 feature maps of size 22*16; the two Identity Blocks output 2048 feature maps of size 66*16.

[0021] (7) DECONV6 layer: contains a Deconv Block, outputting 256 feature maps of size 44*32;

[0022] (8) DECONV7 layer: contains a Deconv Block, outputting 256 feature maps of size 88*64;

[0023] (9) DECONV8 layer: contains a Deconv Block, outputting 256 feature maps of size 176*128;

[0024] (10) DECONV9 layer: contains a Deconv Block, outputting 128 feature maps of size 352*256;

[0025] (11) CONV10 layer: contains a convolution kernel of size (1,1) with a stride of 1, and outputs a heat map of the bolt center point of size 352*256.

[0026] Furthermore, in step 4), the construction work order QR code is generated by the project technicians through the construction plan management interface of the high-strength bolt construction management system by associating the bolt area, the laboratory-calibrated construction torque, and the tightening time. It includes the specific location of the tightening node and the number of bolts on the tightening node plate. The intelligent torque-controlled electric wrench integrates a torque sensor and a CNC system, which can set, display, and record the final output torque of the electric wrench during the tightening process.

[0027] Preferably, in step 4), the method for numbering each bolt is as follows: first, based on the two-dimensional coordinates of the bolt center point output by the bolt positioning model (the origin of which is the top left corner of the image), the two-dimensional coordinates are converted to one-dimensional coordinates before the bolts are numbered. The coordinate transformation method is as follows: Figure 4 As shown:

[0028] The coordinate conversion formula is as follows: ;

[0029] In the formula, C is the converted center point coordinate, y is the y-axis coordinate value in the output coordinates of the bolt center point model, and x is the x-axis coordinate value in the output coordinates of the bolt center point model; a and b are the threshold parameters for the y-axis coordinates of the bolt center points in the same row. The range of a and b is 1~9, and their size is determined by the size of the image area occupied by the bolt image. The initial values ​​are set to a=9 and b=6.

[0030] After converting the two-dimensional coordinates of all bolt center points to a one-dimensional coordinate system, the K-means clustering algorithm is used to cluster the center points. The number of center points in each cluster is set according to the number of bolts in each row displayed by the QR code scanning results. After clustering the center points, the center points of each cluster are sorted from left to right on the one-dimensional coordinate axis. The center point in the first cluster is the bolt in the first row. The coordinates of the center points in each cluster are sorted according to the size of the coordinate axis, and then the bolt number of each center point is marked. For example, the first bolt in the first row should be marked as 0101 to facilitate the identification of the currently tightened bolt.

[0031] In step 5), the bolt node panel image includes the boundary of the panel.

[0032] In step 6), after converting the two-dimensional coordinates of the bolt center point output from the bolt center point model to a one-dimensional coordinate system, the distance between the transformed one-dimensional coordinates of the bolt center point in step 4) and the one-dimensional coordinates of the bolt center point in the current image is calculated one by one. The distance calculation formula is as follows:

[0033] ;

[0034] In the formula, d represents the distance between each center point in the overall image and the tightening image after coordinate transformation. The coordinates of the center point in the overall image taken before final tightening, transformed into one-dimensional coordinates. The coordinates of the center point in the current tightening image captured during tightening, transformed into one-dimensional coordinates;

[0035] The distance between the corresponding center points in two images should be equal. When the distance between the center points of the same bolt with the same number in two images is significantly greater than the distance between the center points of the corresponding bolt with the same number in the previous image, the bolt number corresponding to that number in the overall image is the bolt number to be tightened.

[0036] Preferably, in step 6), the high-strength bolt construction management system can control the bolt construction process from multiple aspects such as tightening tools, operators, tightening torque, bolt images, bolt hole numbers, and node plate numbers.

[0037] In step 7), the high-strength bolt construction management system uses QR codes to coordinate and associate four pieces of information: operator, tightening tool, tightening task, and tightened bolt; the high-strength bolt construction management system uses GPRS or Bluetooth wireless transmission to achieve real-time uploading of construction data.

[0038] In step 1), each image is cropped to a 4:3 ratio using Photoshop image processing software, and the center point of the original steel bridge bolt head is marked using Labelme annotation software; the data augmentation operations include flipping, rotating, cropping, deforming, and scaling.

[0039] In step 3), the enhanced dataset is divided into a training set and a validation set in an 8:2 ratio.

[0040] The beneficial effects of this invention are mainly reflected in the following aspects: This invention realizes the information management of high-strength bolts in steel structure bridges, which facilitates the quick query of various construction information of bolts during operation and maintenance, such as tightening torque, bolt number, node plate number, tightening tool, operator and other information. Combining a constant torque electric wrench, a high-strength bolt construction management system and deep learning, an information management method for high-strength bolts in steel bridges is proposed. The electric wrench is a constant torque electric wrench that integrates a torque sensor and a CNC system, which can display and record the final output torque of the electric wrench. Attached Figure Description

[0041] Figure 1 This is a flowchart of the information management method for high-strength bolts in steel bridges proposed in this invention.

[0042] Figure 2 Network structure diagram for identifying bolt center points.

[0043] Figure 3 The intelligent constant torque electric wrench used in this invention comprises: 1, a small computing unit; 2, a motor; 3, an information transmission module; 4, a data storage module; 5, a digital display control unit; 6, a fisheye lens; 7, an indicator light; and 8, a socket.

[0044] Figure 4 This is a schematic diagram showing the coordinate transformation of the bolt center point.

[0045] Figure 5 This is a diagram illustrating the identification of the bolt number currently being tightened. Detailed Implementation

[0046] The present invention will now be further described with reference to the accompanying drawings.

[0047] Reference Figures 1-5 An information management method for high-strength bolts in steel bridges includes the following steps:

[0048] 1) Image acquisition and preprocessing: Images of steel bridge bolt node plates were acquired. The size of each image was cropped to 4:3 using Photoshop image processing software. The center point of the original steel bridge bolt head was marked using Labelme annotation software. The Mask, JSON file and original steel bridge bolt images were integrated into a steel bridge bolt image dataset. Then, data augmentation operations such as flipping, rotating, cropping, deforming and scaling were performed on the steel bridge bolt image dataset to obtain the enhanced steel bridge bolt image dataset.

[0049] 2) Building a bolt center point recognition model: Build a target center point network on the Keras deep learning framework based on Tensorflow, and pre-train the model using the COCO dataset until the bolt center point recognition model converges;

[0050] In step 2), the bolt center point identification model structure is as follows: Figure 2 As shown, the network model is a fully convolutional network with a deepened ResNet50 as the feature extraction network. The network model includes:

[0051] (1) Input a 704*512 preprocessed three-channel image; where Conv Block and IdentityBlock are consistent with ResNet, and Denconv Block is a transposed convolutional kernel with size (4,4) and stride 2 with BatchNormalization and ReLU activation function;

[0052] (2) CONV1 layer: a convolutional layer with BatchNormalization and ReLU activation function, with a kernel size of 7*7, a stride of 2, and an output of 64 feature maps of size 352*256; a max pooling layer with a filter size of (3,3) and a stride of 2, and an output of 64 feature maps of size 176*128.

[0053] (3) CONV2 layer: contains one Conv Block and two Identity Blocks. The Conv Block outputs 256 feature maps of size 256*128; the outputs of the two Identity Blocks are both 256 feature maps of size 176*128.

[0054] (4) CONV3 layer: contains one Conv Block and three Identity Blocks. The Conv Block outputs 512 feature maps of size 88*64; the outputs of the three Identity Blocks are all 512 feature maps of size 88*64.

[0055] (5) CONV4 layer: contains one Conv Block and five Identity Blocks. The Conv Block outputs 1024 feature maps of size 44*32; the outputs of the five Identity Blocks are all 1024 feature maps of size 44*32.

[0056] (6) CONV5 layer: contains one Conv Block and two Identity Blocks. The Conv Block outputs 2048 feature maps of size 22*16; the two Identity Blocks output 2048 feature maps of size 66*16.

[0057] (7) DECONV6 layer: contains a Deconv Block, outputting 256 feature maps of size 44*32;

[0058] (8) DECONV7 layer: contains a Deconv Block, outputting 256 feature maps of size 88*64;

[0059] (9) DECONV8 layer: contains a Deconv Block, outputting 256 feature maps of size 176*128;

[0060] (10) DECONV9 layer: contains a Deconv Block, outputting 128 feature maps of size 352*256;

[0061] (11) CONV10 layer: contains a convolution kernel of size (1,1) with a stride of 1, and outputs a heat map of the bolt center point of size 352*256.

[0062] 3) Training the bolt hole recognition model: The enhanced dataset is divided into training set and validation set in an 8:2 ratio. The training set is input into the pre-trained model for training. The mean square error (MSE) function is used as the objective function of the model. A dynamic learning rate is used, and Adam is selected as the optimizer to train the model. The best model is selected using the validation set and then mounted into the small computing unit at the lower end of the intelligent constant torque electric wrench.

[0063] 4) Use the mobile app connected to the wrench to scan the QR code and identify the entire node plate: After the initial tightening is completed, scan the personnel QR code and the construction work order. Then, use the camera on the upper part of the smart torque electric wrench to take an overall image of the bolt node plate. Input the image into the bolt center point recognition model for recognition, output the number of bolts in the current node plate, compare and confirm with the QR code scanning result, and number each bolt.

[0064] In step 4), the construction work order QR code is generated by the project technicians through the construction plan management interface of the high-strength bolt construction management system by associating the bolt area, the laboratory-calibrated construction torque, and the tightening time. It contains the specific location of the tightening node and the number of bolts on the tightening node plate. The intelligent torque-controlled electric wrench integrates a torque sensor and a CNC system, and can set, display, and record the final output torque of the electric wrench during the tightening process.

[0065] In step 4), the numbering method for each bolt is as follows: First, based on the two-dimensional coordinates of the bolt center point output by the bolt positioning model (the origin of which is the top left corner of the image), the two-dimensional coordinates are converted to one-dimensional coordinates before the bolts are numbered. The coordinate transformation method is as follows: Figure 4 As shown:

[0066] The coordinate conversion formula is as follows: ;

[0067] In the formula, C represents the converted center point coordinates, y represents the y-axis coordinate value in the output coordinates of the bolt center point model, and x represents the x-axis coordinate value in the output coordinates of the bolt center point model; a and b are the threshold parameters for the y-axis coordinates of the bolt center points in the same row, both ranging from 1 to 9, and their values ​​are determined by the area of ​​the image occupied by the bolt image. The initial values ​​are set to a=9 and b=6.

[0068] After converting the two-dimensional coordinates of all bolt center points to a one-dimensional coordinate system, the K-means clustering algorithm is used to cluster the center points. The number of center points in each cluster is set according to the number of bolts in each row displayed by the QR code scanning results. After clustering the center points, the center points of each cluster are sorted from left to right on the one-dimensional coordinate axis. The center point in the first cluster is the bolt in the first row. The coordinates of the center points in each cluster are sorted according to the size of the coordinate axis. Then, the bolt number of each center point is marked. For example, the first bolt in the first row should be marked as 0101 to facilitate the identification of the bolts being tightened in the future.

[0069] 5) Final tightening and photo taking: Use the intelligent torque-controlled electric wrench to final tighten the bolt. While pressing the tightening button, the camera on the top of the wrench takes a photo.

[0070] In step 5), the bolt node panel image includes the boundary of the panel;

[0071] 6) Current bolt number identification: Input the photo taken by the camera into the small calculation unit at the bottom of the wrench to identify the position number of the bolt being tightened;

[0072] In step 6), the method for identifying the position number of the currently tightened bolt is as follows: Figure 5As shown, after converting the two-dimensional coordinates of the bolt center point output by the bolt center point model to a one-dimensional coordinate system, the distance between the transformed one-dimensional coordinates of the bolt center point in step 4) and the one-dimensional coordinates of the bolt center point in the current image is calculated one by one. The distance calculation formula is as follows:

[0073] ;

[0074] In the formula, d represents the distance between each center point in the overall image and the tightening image after coordinate transformation. The coordinates of the center point in the overall image taken before final tightening, transformed into one-dimensional coordinates. The coordinates of the center point in the current tightening image captured during tightening, transformed into one-dimensional coordinates;

[0075] The distance between the corresponding center points in two images should be equal. When the distance between the center points of the same bolt number in two images is significantly greater than the distance between the center points of the corresponding bolt number in the previous image, the bolt number corresponding to that bolt in the overall image is the bolt number to be tightened.

[0076] In step 6), the high-strength bolt construction management system can control the bolt construction process from multiple aspects, such as tightening tools, operators, tightening torque, bolt images, bolt hole numbers, and node plate numbers.

[0077] 7) Recording current bolt tightening information: After tightening each bolt, use a green frame to outline the bolt position in the previously captured overall image until all bolts on the node plate are outlined; transmit the current tightening image, the overall image after the green frame selection, and information such as the bolt number, tightening rotation angle, current node plate number, and operator to the high-strength bolt construction management system; when the overall image of a node plate shows that all bolts have been selected with green frames, the indicator light on the upper part of the intelligent torque-controlled electric wrench will turn green;

[0078] In step 7), the high-strength bolt construction management system uses QR codes to coordinate and associate four pieces of information: operator, tightening tool, tightening task, and tightened bolt; the high-strength bolt construction management system uses GPRS or Bluetooth wired transmission to achieve real-time uploading of construction data.

[0079] During the tightening process, on-site management personnel can monitor the current construction status in real time through cloud connection, check for bolts missing from node plate positions that are not displayed in green, and send prompt messages through the platform. Construction personnel can receive information about the missing bolts, including the node plate number and the missing bolt number, on their mobile app, and then go to the corresponding node plate position to check the bolt. In this way, even if the management personnel are not on-site, they can keep track of the current construction progress and site conditions in real time.

[0080] When the difference between the final tightening torque of the bolt and the preset torque in the laboratory exceeds the threshold, the bolt location is selected in red in the overall image transmitted from the high-strength bolt construction management system.

[0081] When the final tightening torque of a bolt is not up to standard, even after all bolts on the node plate have undergone a final tightening, the indicator light on the top of the intelligent electric torque wrench will not turn green. At this time, the system will automatically send the number of the bolt that failed to meet the final tightening standard. On-site construction personnel will then re-tighten the corresponding bolts according to the information prompts. Once the final tightening torque of all bolts on the node plate meets the standard, the indicator light on the top of the wrench will turn green.

[0082] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. An information-based management method for high-strength bolts in steel bridges, characterized in that, The method includes the following steps: 1) Image acquisition and preprocessing: Acquire images of steel bridge bolt node plates, use Photoshop image processing software to crop each image to the set size, mark the center point of the original steel bridge bolt head, integrate the Mask, JSON file and original steel bridge bolt images into a steel bridge bolt image dataset, and then perform data augmentation operation on the steel bridge bolt image dataset to obtain the enhanced steel bridge bolt image dataset. 2) Building a bolt center point recognition model: Build a target center point network on the Keras deep learning framework based on Tensorflow, and pre-train the model using the COCO dataset until the bolt center point recognition model converges; 3) Training the bolt hole recognition model: The enhanced dataset is divided into a training set and a validation set according to a set ratio. The training set is input into the pre-trained model for training. The mean square error (MSE) function is used as the objective function of the model. A dynamic learning rate is used, and Adam is selected as the optimizer to train the model. The best model is selected using the validation set and then mounted into the small computing unit at the lower end of the intelligent constant torque electric wrench. 4) Use the mobile app connected to the wrench to scan the QR code and identify the entire node plate: After the initial tightening is completed, scan the personnel QR code and the construction work order. Then, use the camera on the upper part of the smart torque electric wrench to take an overall image of the bolt node plate. Input the image into the bolt center point recognition model for recognition, output the number of bolts in the current node plate, compare and confirm with the QR code scanning result, and number each bolt. The method for numbering each bolt is as follows: First, based on the two-dimensional coordinates of the bolt center point output by the bolt positioning model, which takes the top left corner of the image as the origin, the two-dimensional coordinates are converted to one-dimensional coordinates and then the bolts are numbered. After converting the two-dimensional coordinates of all bolt center points to a one-dimensional coordinate system, the K-means clustering algorithm is used to cluster the center points. The number of center points in each cluster is set according to the number of bolts in each row displayed by the QR code scanning results. After clustering the center points, the center points of each cluster are sorted from left to right on the one-dimensional coordinate axis. The center point in the first cluster is the bolt in the first row. The coordinates of the center points of each cluster are sorted according to the size of the coordinate axis, and then the bolt number of each center point is marked. 5) Final tightening and photo taking: Use the intelligent torque-controlled electric wrench to final tighten the bolt. While pressing the tightening button, the camera on the top of the wrench takes a photo. 6) Current bolt number identification: Input the photo taken by the camera into the small calculation unit at the bottom of the wrench to identify the position number of the bolt being tightened; 7) Recording current bolt tightening information: After tightening each bolt, use a green frame to outline the bolt position in the previously captured overall image until all bolts on the node plate are outlined; transmit the current tightening image, the overall image after the green frame selection, and the bolt number, tightening rotation angle, current node plate number, and operator information to the high-strength bolt construction management system; when the overall image transmitted for a node plate shows that all bolts have been selected with green frames, the indicator light on the upper part of the intelligent torque-controlled electric wrench will turn green.

2. The information management method for high-strength bolts in steel bridges as described in claim 1, characterized in that, In step 2), the bolt center point recognition model uses a fully convolutional network with a deepened ResNet50 as the feature extraction network. The network model includes: (1) Input a 704*512 preprocessed three-channel image; where Conv Block and Identity Block are consistent with ResNet, and Denconv Block is a transposed convolutional kernel with size (4,4) and stride 2 with BatchNormalization and ReLU activation function; (2) CONV1 layer: a convolutional layer with BatchNormalization and ReLU activation function, with a kernel size of 7*7, a stride of 2, and an output of 64 feature maps of size 352*256; a max pooling layer with a filter size of (3,3) and a stride of 2, and an output of 64 feature maps of size 176*128. (3) CONV2 layer: contains one Conv Block and two Identity Blocks. The Conv Block outputs 256 feature maps of size 256*128; the outputs of the two Identity Blocks are both 256 feature maps of size 176*128. (4) CONV3 layer: contains one Conv Block and three Identity Blocks. The Conv Block outputs 512 feature maps of size 88*64; the outputs of the three Identity Blocks are all 512 feature maps of size 88*64. (5) CONV4 layer: contains one Conv Block and five Identity Blocks. The Conv Block outputs 1024 feature maps of size 44*32; the outputs of the five Identity Blocks are all 1024 feature maps of size 44*32. (6) CONV5 layer: contains one Conv Block and two Identity Blocks. The Conv Block outputs 2048 feature maps of size 22*16; the two Identity Blocks output 2048 feature maps of size 66*16. (7) DECONV6 layer: contains a Deconv Block, outputting 256 feature maps of size 44*32; (8) DECONV7 layer: contains a Deconv Block, outputting 256 feature maps of size 88*64; (9) DECONV8 layer: contains a Deconv Block, outputting 256 feature maps of size 176*128; (10) DECONV9 layer: contains a Deconv Block, outputting 128 feature maps of size 352*256; (11) CONV10 layer: contains a convolution kernel of size (1,1) with a stride of 1, and outputs a heat map of the bolt center point of size 352*256.

3. The information management method for high-strength bolts in steel bridges as described in claim 1 or 2, characterized in that, In step 4), the construction work order QR code is generated by the project technicians through the construction plan management interface of the high-strength bolt construction management system by associating the bolt area, the laboratory-calibrated construction torque, and the tightening time. It includes the specific location of the tightening node and the number of bolts on the tightening node plate. The intelligent torque-controlled electric wrench integrates a torque sensor and a CNC system, and can set, display, and record the final output torque of the electric wrench during the tightening process.

4. The information management method for high-strength bolts in steel bridges as described in claim 1 or 2, characterized in that, In step 5), the bolt node plate image includes the boundary of the panel.

5. The information management method for high-strength bolts in steel bridges as described in claim 1 or 2, characterized in that, In step 6), the high-strength bolt construction management system can control the bolt construction process from multiple aspects, including tightening tools, operators, tightening torque, bolt images, bolt hole numbers, and node plate numbers.

6. The information management method for high-strength bolts in steel bridges as described in claim 1 or 2, characterized in that, In step 7), the high-strength bolt construction management system uses QR codes to coordinate and associate four pieces of information: operator, tightening tool, tightening task, and tightened bolt; the high-strength bolt construction management system uses GPRS or Bluetooth wireless transmission to achieve real-time uploading of construction data.

7. The information management method for high-strength bolts in steel bridges as described in claim 1 or 2, characterized in that, In step 1), each image is cropped to a 4:3 ratio using Photoshop image processing software, and the center point of the original steel bridge bolt head is marked using Labelme annotation software; the data augmentation operations include flipping, rotating, cropping, deforming, and scaling.

8. The information management method for high-strength bolts of steel bridges as described in claim 1 or 2, characterized in that, In step 3), the enhanced dataset is divided into a training set and a validation set in an 8:2 ratio.

Images