Bridge tower steel bar component forming precision control method

Abstract

The invention provides a bridge tower reinforcing steel bar component forming precision control method, which comprises the following steps of: arranging a support frame on a movable machine head for mounting a camera and a lens, performing photographing to calculate a forming angle after bending forming, and performing automatic length and angle detection by adopting an algorithm based on deep learning and image processing; a three-dimensional laser scanning technology is adopted, and the forming line shape and precision of the reinforcing mesh are measured through the steps of target ball installation, laser scanning, point cloud processing, denoising, circle center extraction, axis extraction and key size calculation. According to the rib block body automatic detection technology, the distance between main ribs is positioned through a main rib positioning hole plate, a servo motor is driven to incline to a design angle, and a full-temperature-compensation high-precision digital output type double-shaft inclination angle sensor is adopted to automatically measure the inclination angle and compare the inclination angle with a theoretical value; and with the assistance of a large complex component rapid detection method, the steel bar forming precision of each link is reasonably controlled, and finally the mounting precision of the steel bar component on the tower is guaranteed.

Description

[0001] Methodology Area The invention relates to the technical field of bridge concrete structure reinforcement construction, and in particular to a method for controlling the forming accuracy of a bridge tower reinforcement component.

[0002] Background Methods Bridge construction is a labor-intensive industry. The aging of construction workers is becoming more and more serious, and the labor force is gradually shifting to emerging service industries. How to ensure the high-quality development of the industry is an urgent problem that needs to be solved. In terms of steel bar construction, in recent years, bridge construction companies have gradually applied off-site binding of steel bar components, and installed them as a whole or in blocks, that is, manually loosely binding steel bar components on the steel bar binding cradle in the construction ground area, and hoisting the steel bar components as a whole or in blocks to the tower for installation through tower cranes and other lifting equipment. Furthermore, the industry has seen a transformation and upgrading of the modular steel bar industrial construction model. In the steel bar processing plant, steel mesh and blocks are mechanized through customized production lines, and then assembled as a whole to form steel bar components and hoisted to the tower.

[0003] With the iteration and upgrading of steel bar assembly construction technology, the control and detection technology of steel bar component forming accuracy has become increasingly important. If the forming accuracy in each link cannot be reasonably controlled, it will lead to large accuracy deviations in each link, making it difficult to adapt to modern industrial assembly requirements and affecting the final installation forming accuracy of steel bar components. Summary of the invention

[0004] The main purpose of the present invention is to provide a method for controlling the forming accuracy of bridge tower steel bar components, so as to solve the problem that large precision deviation occurs in each link of bridge tower steel bar components, making it difficult to adapt to modern industrial assembly requirements and affecting the final installation forming accuracy of steel bar components.

[0005] In order to solve the above method problems, the method scheme adopted by the present invention is: a method for controlling the forming accuracy of a bridge tower steel bar component, the method comprising: an automatic detection method for steel bar unit elements, an automatic detection method for steel bar meshes, an automatic detection method for steel bar blocks and an automatic detection method for steel bar components; The automatic detection method of steel bar unit element is as follows: by setting a support frame to install a camera and lens on the mobile head, taking pictures after bending and forming to calculate the forming angle, and using an algorithm based on deep learning and image processing to automatically detect the length and angle; The automatic detection method of steel mesh is: using 3D laser scanning technology, through the steps of target ball installation, laser scanning, point cloud processing, denoising, center extraction, axis extraction, and key dimension calculation, the forming line shape and accuracy of the steel mesh are measured; The automatic detection method of steel bar blocks is as follows: the automatic detection technology of steel bar blocks locates the main bar spacing through the main bar positioning hole plate, the servo motor drives the tilt to the designed angle, and the full temperature compensation high-precision digital output dual-axis inclination sensor is used to automatically measure the tilt angle and compare it with the theoretical value; The automatic detection method of steel bar components is as follows: the automatic detection technology of steel bar components uses 3D Disto equipment to select measurement points, and the drone system is equipped with a laser lens to collect data, and performs real-time measurement of steel bar components in the tire frame and on the top of the tower.

[0006] In the preferred solution, a camera and a lens are installed on the support frame. When the steel bar is bent and the bending head is returned to its position, the camera takes a picture of the steel bar, captures the image of the bending section, automatically calculates the forming angle, and informs the on-site workers through the on-site indicator light; If the angle is not enough, the worker enters the compensation value or modifies the bending angle on the equipment side, steps on the pedal, completes the second bending, and calculates the forming angle again. Continuous processing can be carried out after the requirements are met.

[0007] In the preferred solution, the algorithm steps for automatic detection of steel bar units are as follows: A1. Database establishment: Fix the camera at a certain place to collect images. This place must ensure that complete image information including the bending area and the steel bar target can be obtained. Then manually mark the collected images to mark the specific location and feature information of the bending area and the steel bar target. These marking information and image data together constitute the elements of the database, thus completing the database establishment. The collected image set is ,in is the number of images collected; for each image , whose tag information set is ,in is the number of labels for each image; database It can be expressed as ; This formula is used to clearly define the structure of the database, ensuring that image data and corresponding label information can be accurately stored and used, providing an accurate data basis for subsequent algorithm processing; A2. Image acquisition and processing: Since the object to be measured is large in size, multiple cameras are used to capture images. First, the images collected by multiple cameras are stitched together. The number of cameras is The collected images are ; During the stitching process, for each image captured by a camera, it is necessary to find the overlapping area with other camera images and stitch them together based on the feature information of the overlapping area; after the stitching is completed, the steel bar contour is detected on the stitched image; by analyzing the grayscale changes and edge feature information of the image, the approximate contour position of the steel bar in the image is determined; Next, based on the determined steel bar contour, the steel bar bending area is further identified; This step requires combining the existing bending area marking information in the database with the feature information in the image to find the part that meets the bending area characteristics; Finally, calculate the length and angle of the reinforcement; camera The collected images are ,in is the image coordinate; for two adjacent cameras and , and the overlapping area is , the matching function of the overlapping area is defined as ;when Less than a certain threshold When , the two images are considered to match well in the overlapping area and can be stitched; the stitched image The values ​​in the overlapping area can be calculated by weighted average, that is, ,in and is the weight, which can be determined according to factors such as image quality and resolution; This formula is used to accurately determine whether adjacent camera images match in the overlapping area, and to reasonably stitch them together if they match well, ensuring that the stitched image can accurately reflect the full picture of the object being measured; The grayscale matrix of the image is , define the edge detection operator as ,in for The neighborhood of the point; by setting a threshold ,when Greater than When , the point is considered to be an edge point, thus determining the outline of the steel bar; This formula determines the edge points by calculating the grayscale difference between each point in the image and its neighboring points, thereby accurately detecting the outline of the steel bars in the image; Rebar bending area identification: The point set on the rebar contour is , for each point , calculate its curvature ,in For point The change in the angle between the two adjacent points is: is the arc length between two points; when Greater than a threshold When , the area where the point is located is considered to be the bending area; this formula determines whether the point is located in the bending area by calculating the curvature of the point on the steel bar contour, thereby accurately identifying the steel bar bending area; Length and angle calculation formula: For steel bar length calculation, assume that the extracted skeleton line node is , the distance between nodes is ,in For Node Coordinates of the steel bar; length of the steel bar is the sum of the distances between all nodes, that is ; For the calculation of the bending angle, the slopes of the straight lines fitting the bending area are and , then the bending angle ; A3. Result visualization and alarm: The measured angle and length information will be visualized through the industrial computer deployed on site; A graphical interface can be used to display information such as the shape, length, and bending angle of the steel bar, so that operators can intuitively understand the status of the steel bar; If the measurement result deviates greatly from the theoretical value, an alarm will be triggered through the andon system. By setting a reasonable deviation threshold, when the difference between the measured value and the theoretical value exceeds the threshold, the andon system alarm will be triggered to remind the operator to check and handle the problem. The measured length is The theoretical length is , the deviation threshold is , then the length deviation judgment formula is Length deviation alarm is triggered; The measured angle is , the theoretical perspective is , the deviation threshold is , then the angle deviation judgment formula is Angle deviation alarm is triggered.

[0008] In the preferred solution, the algorithm steps for extracting the steel bar bending area and steel bar contour based on deep learning are as follows: B1. Target detection: Use the YOLO v5 series model for target detection. Input the collected image into the YOLOv5 model, and the model will automatically identify the approximate location of the steel bar target and the bending area in the image. The formula can be expressed as: ,in is the input image, are the parameters of the model, is the output prediction result; the formula is used to describe the basic operation process of the YOLO v5 model. By adjusting the model parameters , which enables the model to accurately detect objects for different input images; B2. Segmentation of bending area sub-images: For the segmented bending area sub-images, the semantic segmentation method is used to segment the steel bar targets at the pixel level. The bending area sub-images are input into the UNet network, and the network classifies each pixel as a steel bar target pixel or a background pixel according to the feature information of the image, thereby achieving pixel-level segmentation of the steel bar targets. The formula can be expressed as: ,in is the input bending region sub-image, are the parameters of the network, The output segmentation result; the formula is used to describe the basic operation process of the UNet network, by adjusting the network parameters , which enables the network to perform accurate pixel-level segmentation of sub-images in different bending regions; B3. Calculation of steel bar length and bending angle based on image processing: For the steel bar segmentation map of the entire image obtained in the previous processing steps, the image processing method is directly used to analyze the characteristics of the connected domain; by analyzing the characteristic information such as the area, perimeter, shape, etc. of the connected domain, the structure and state of the steel bar are further determined; The connected domain is , its area formula is The perimeter formula is ,in Connected domain By calculating the area and perimeter of the connected domain, the thickness and shape of the steel bar can be preliminarily determined; The shape factor of the connected domain is ,The shape factor can be used to more accurately describe the shape characteristics of the connected domain, and plays an important role in judging whether the shape of the steel bar meets the requirements; Length and angle calculation: For the angle of the bending position, fit the straight line of the bending area; for the calculation of the length of the steel bar, directly extract the skeleton line of the connected domain, extract the nodes for each bending position, calculate the length of the skeleton line between each node, that is, the length of each section of the steel bar; perform skeleton extraction based on the previously obtained contour, observe the pixels through the extracted skeleton line, count the pixel length, and convert the pixel length to the real-world coordinate length, convert the number of pixels to the real length, and measure the length of each section of the steel bar; The specific formula is as follows: For the calculation of steel bar length, assume that the extracted skeleton line node is , the distance between nodes is ,in For Node Coordinates of the steel bar; length of the steel bar is the sum of the distances between all nodes, that is ; For the calculation of the bending angle, the slopes of the straight lines fitting the bending area are and , then the bending angle .

[0009] In the preferred solution, the automatic detection method of the steel mesh is: C1. Target ball installation: Place at least three target balls on the steel mesh to be tested. The target balls should be placed in a position that ensures that three target balls in the same position can be scanned at least twice when scanning the target in different directions. The centers of any three target balls cannot be on the same straight line. C2. Laser scanning: Use a ground-based 3D laser scanner to perform a panoramic scan of the area to be tested to obtain the original 3D point cloud data including the steel mesh. The distance between the 3D laser scanner and the mesh should be greater than 3m. The point cloud accuracy obtained is between 1.3-1.8mm, and the point cloud density is 90-120 points / square centimeter; C3. Point cloud processing: Process the original 3D point cloud data obtained by laser scanning, retain the target steel mesh, target ball and the point cloud around the target ball, delete other irrelevant point cloud data, use point cloud processing software to splice the point cloud data obtained by multiple scans, and finally obtain the complete 3D point cloud data of the steel mesh; C4, Denoising: Define the coordinate system, where the horizontal plane direction is the X and Y axes, the height direction is the Z axis, the base surface direction is the X axis, and the coordinate origin is selected at the center of the base surface; in order to reduce the interference with the later surface generation, after the point cloud is spliced, first manually remove the obvious redundant noise points outside the target steel bar, and then use the algorithm to remove the discrete isolated noise points; the specific algorithm is to determine whether the point is a discrete point by specifying the number of point clouds within the range of the spatial radius n, that is, draw a sphere with a radius of 0.01m with any point as the center of the sphere, and calculate the number of point clouds contained in the sphere. If the number of point clouds is less than 30-40, the point is identified as an isolated discrete noise point, and the point is deleted to obtain the denoised steel mesh point cloud data; C5. Circle center extraction: intercept any steel bar point cloud data, and use the RANSAC algorithm to extract the circle center coordinates (Xv, Yv) from the point cloud; C6. Axis extraction: After extracting the center of each section, the steel bar axis can be directly fitted based on the least squares method; C7. Calculation of key dimensions: Based on the previously obtained coordinates of the circle center and the steel bar axis, key dimensions such as main reinforcement stirrup spacing, end length, flatness, plane angle, etc. can be automatically calculated; Calculation of main reinforcement and stirrup spacing: select any two main reinforcements, cut the point cloud sections at equal intervals perpendicular to the Z direction, obtain the center coordinates of the two sections under the Zh plane (Xh1, Yh2, Zt) and (Xh2, Yh2, Zt), and calculate the main reinforcement spacing Lh at this height; Calculation of end flushness: The coordinates of the center of the main reinforcement end face are (Xti, Yti, Zti), where i = 1, 2, …, m, m is the number of main reinforcements, and the end flushness can be obtained after the average height of the main reinforcement is calculated; Mesh surface flatness calculation: manually select the mesh area to be measured, find the least square fitting plane for the measured point cloud data (xi, yi, zi), where i = 0, 1, ..., m, obtain the fitting plane equation, calculate the distance from each point to the fitting plane, and the statistical value of the distance can be regarded as the flatness of the measured mesh; Calculation of joint length: The joint length includes the top length and the bottom length. Plane angle calculation: After the above steps, we can get the equations of two adjacent planes. The angle between the two planes is the angle between the normal vectors of the two planes. C8. Docking judgment: By calculating the relevant parameters of transverse docking, longitudinal docking and radial docking, the quality of the steel mesh is judged; when any one of the parameters ΔXmax, ΔYmax, ΔLmax exceeds the normal range, it indicates that the mesh quality is unqualified; the calculated data is exported and analyzed to determine whether the accuracy of the steel mesh meets the requirements.

[0010] In the preferred embodiment, the specific method of step C7 is: C71. Calculation of main reinforcement spacing: Select any two main reinforcements and cut the point cloud sections at equal intervals in the vertical Z direction to obtain Z h Coordinates of the centers of the two cut planes under the plane （X h1 ,Y h2 ,Z t ） and （X h2 ,Y h2 ,Z t ） , calculate the main reinforcement spacing at this height L h ; C72. Calculation of end flushness: The better the end flushness, the more convenient it is for the steel mesh to be quickly connected to the tower; the coordinates of the center of the main reinforcement end face are （X ti ,Yti ,Z ti ） , average height of main reinforcement: ; The flatness of the end is: ; C73, mesh surface flatness calculation: Manually select the mesh surface area to be measured, and calculate the measured point cloud data. (xi,yi,zi)(i= 0,1,…,m) Find the least squares fitting plane and get the fitting plane equation as follows:

[0011] Calculate the distance from each point to the fitting plane: ; The flatness of the mesh is: ; in, d is the distance from the measuring point to the fitting plane, m is the mean error of the point from the measurement point to the fitting plane, which can be regarded as the flatness of the measured mesh. m The smaller the value, the flatter the mesh surface; the smaller the value, the more uneven the mesh surface. C74, joint length: The joint length includes the top length and the bottom length. Taking the top length as an example, the joint length L AB for A and B The distance between the ends of the main reinforcement is OXY Projection on a plane, B The axis of the main reinforcement and the axis of the end stirrups are OXY Intersection points of plane projections; C75. Calculation of mesh angle: After the above steps, the equations of two adjacent planes can be obtained as follows: and , then the normal vectors of the two planes are {A1, B1, C1} and {A2, B2, C2} , let the angle between these two normal vectors be φ , then the angle between these two planes is φ Calculated by the following formula: .

[0012] In the preferred solution, the automatic detection method of steel bar blocks is specifically as follows: D1. The main controlled data in the steel bar block forming stage include: inclination ≤1 / 1000 and ≤8mm, spacing between main bars ≤±5mm, spacing between main bars in the same row ≤±10mm, spacing between stirrups ≤±10mm; D2. The spacing between main reinforcement rows and the spacing between main reinforcement rows are limited by the main reinforcement positioning hole plate on the equipment to meet the accuracy requirements. The spacing between stirrups is controlled by the comb rod structure on the equipment to also meet the accuracy requirements. D3. The inclination angle of the steel bar block is controlled by the angle adjustment device on the equipment, which is driven by a servo motor. Theoretically, the accuracy can reach 0.05°, which meets the use requirements; In order to verify that the angle accuracy of the formed steel bar block meets the requirements, a full temperature compensation high-precision digital output dual-axis inclination sensor is used to automatically measure the inclination angle and compare it with the theoretical value, so as to achieve the purpose of detecting the production angle of the steel bar block; Specifically, four full-temperature-compensated high-precision digital output dual-axis inclination sensors are installed on the four angle adjustment poles of the equipment. By comparing and analyzing the measured data after the block angle adjustment with the equipment input data, the purpose of automatically detecting the block forming accuracy is achieved.

[0013] In the preferred solution, the automatic detection method of steel bar parts is specifically to measure the steel bar parts in three states during the production process: assembly in the tire frame, storage outside the tire frame, and hoisting on the tower; E1. Measurement inside the tire frame: Measurement before welding after assembly is completed: 8 corner points are measured at the top, middle and bottom, and the measurement points are converted to the outer measurement of the tire frame through the conversion measurement method for easy observation and measurement. Specifically, 2 marker poles are set on each side of the top and bottom of the component, and the Leica 3D Disto equipment is used to measure the target points on the marker poles. The spatial position of the component connection point is calculated based on the target point position and the pole length. Three measuring poles are arranged at the top, middle and bottom of each control side; Measurement after welding: Also measure 8 corner points at the top, middle and bottom, because there may be welding deformation after welding, which needs to be measured and compared with the theory; After undoing the adjustment screw: still measure the 8 corner points of the top, middle and bottom. Since stress rebound deformation may occur after the external force of the top rod is removed, it is necessary to measure and compare with the theory; E2. Measurement outside the tire frame: On flat ground, measure 8 corner points of the top, middle and bottom in a free state, as well as the main reinforcement spacing, including the outer circle, middle circle and inner circle of the top; the outer circle and inner circle of the bottom, measure the dimensions, coordinates, inclination and other data of the components in a completely free state, and compare them with the theory; E3. Tower hoisting connection measurement: After inserting the bell mouth guide device and before connecting the cone sleeve, measure: measure the spatial position of the 8 corner points of the top of the component to guide the docking of the components and visualize the docking points of the previous component and the current component; After the cone sleeve is connected: measure the spatial position of the 8 corner points of the top or bottom of the component to check the installation accuracy of the component; After concrete pouring: measure the spatial positions of the 8 corner points of the top opening, analyze the impact of pouring on the linear shape of the tower limbs, give pre-deflection suggestions for the next segment, measure the position of the main reinforcement of the top opening, establish an actual tower column model, and compare and analyze it with the theory.

[0014] In the preferred solution, the algorithm steps of the drone system equipped with a laser lens for data collection in the automatic detection method of steel bar parts are as follows: F1. Data collection: When performing close-range high-resolution image acquisition at the construction site, considering that mobile equipment and personnel seriously affect the safety of the UAV's automatic route planning flight, and the relatively small workload of steel bar component data collection, manual flight is used to obtain UAV images to ensure the safety and image resolution of the UAV; images are acquired according to the "Z"-shaped flight trajectory to obtain sufficient and appropriate steel bar component image data; F2. Image 3D reconstruction: Based on the comprehensive advantages of popular software on the market in terms of computing speed and accuracy, DJI Intelligent Map was selected to perform 3D reconstruction on the collected image data. DJI Intelligent Map can process image data and reconstruct point cloud data. The obtained point cloud data can better reflect the shape and structural information of steel bar parts. F3, spatial positioning algorithm: Slice separation: Slice the point cloud data of the steel bar parts along the Z axis based on straight-through filtering to separate the point cloud data at different elevation positions; count the number of point cloud slice data from large to small along the elevation direction; Feature positioning: When longitudinal reinforcement appears, the amount of point cloud data will jump in magnitude, and this jump will only occur once, that is, the first section that appears is the longitudinal reinforcement section; in this way, the top surface positioning of the longitudinal reinforcement of the point cloud data is achieved, thereby establishing the association and mapping relationship between the image and the point cloud data in spatial position; F4, Single steel bar fitting algorithm: Fitting model selection, based on sampling consistency algorithm, set the circle model and cylinder model to fit the point cloud data; Fitting process and optimization: According to the design requirements, the threshold is set to 3-4cm, the number of iterations is set to 10,000-20,000 times, and the steel bar data is fitted; through the fitting process, the longitudinal steel bar data of the steel bar component point cloud is fitted, and the relevant data information of the steel bar is further extracted; F5. Point cloud data filtering: In the above steps, the original point cloud data is filtered to remove noise points and abnormal points to improve data quality.

[0015] In the preferred solution, the steps of the single steel bar fitting algorithm are: Based on the sampling consistency algorithm, the circle model and the cylinder model are set to fit the point cloud data. The circle and cylinder models are shown in the following formula: ; In the formula, is a three-dimensional circle in space, is a point on a spatial sphere or a spatial plane, is the center of the sphere, is the radius of the sphere, is the coefficient on the spatial plane; .

[0016] The present invention provides a method for controlling the forming accuracy of bridge tower steel bar components. Aiming at two industrialized construction modes of bridge tower steel bar components, a new steel bar component precision control system is proposed. By controlling the steel bar forming progress of each production link, the "tolerance" control requirements in the industrial processing field are met. Assisted by a new large-scale complex component rapid detection method, reasonable, efficient and rapid control of the steel bar component forming accuracy is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a process flow chart of data processing of the present invention; Figure 2 It is a flow chart of position segmentation of the detection of the present invention; Figure 3 It is a diagram of steel bar length and bending angle based on image processing of the present invention; Figure 4 It is the technical roadmap of the three-dimensional laser scanning application of the present invention; Figure 5 It is a schematic diagram of the placement of the target ball of the present invention; Figure 6 It is the laser scanning diagram of the present invention; Figure 7 It is a point cloud processing diagram of the present invention; Figure 8 is a point cloud data graph after denoising of the present invention; Fig. 9 It is the circle center extraction diagram of the present invention; Fig.10 It is the axis extraction and center point diagram of the present invention; Fig.11 It is a docking schematic diagram of the present invention; Fig.12 It is the application effect diagram of the present invention; Fig.13 It is a schematic diagram of the installation of the dual-axis tilt sensor of the present invention; Fig.14This is a flow chart of measuring steel bar parts in a tire frame of the present invention; Fig.15 It is a structural diagram of the cleaning head of the protective housing of the present invention for cleaning the measuring head; Fig.16 It is a key control point diagram of the components of the present invention; Fig.17 It is the technical roadmap of the steel bar parts of the present invention; Fig.18 This invention is the DJI Figure 3 dimensional point cloud reconstruction; Fig.19 It is the image reconstruction point cloud data map of the present invention; Fig. 20 It is a schematic diagram of point cloud data slicing of the present invention; Fig.21 It is the fitting effect diagram of the present invention; Fig. 22 It is the cylindrical fitting diagram of the present invention. DETAILED DESCRIPTION

[0018] Example 1 like Figures 1 to 22 As shown, a method for controlling the forming accuracy of a bridge tower steel bar component, the method comprising: an automatic detection method for steel bar unit elements, an automatic detection method for steel bar meshes, an automatic detection method for steel bar blocks, and an automatic detection method for steel bar components; The automatic detection method of steel bar unit element is as follows: by setting a support frame to install a camera and lens on the mobile head, taking pictures after bending and forming to calculate the forming angle, and using an algorithm based on deep learning and image processing to automatically detect the length and angle; The automatic detection method of steel mesh is: using 3D laser scanning technology, through the steps of target ball installation, laser scanning, point cloud processing, denoising, center extraction, axis extraction, and key dimension calculation, the forming line shape and accuracy of the steel mesh are measured; The automatic detection method of steel bar blocks is as follows: the automatic detection technology of steel bar blocks locates the main bar spacing through the main bar positioning hole plate, the servo motor drives the tilt to the designed angle, and the full temperature compensation high-precision digital output dual-axis inclination sensor is used to automatically measure the tilt angle and compare it with the theoretical value; The automatic detection method of steel bar components is as follows: the automatic detection technology of steel bar components uses 3D Disto equipment to select measurement points, and the drone system is equipped with a laser lens to collect data, and performs real-time measurement of steel bar components in the tire frame and on the top of the tower.

[0019] In the preferred solution, a camera and a lens are installed on the support frame. When the steel bar is bent and the bending head is returned to its position, the camera takes a picture of the steel bar, captures the image of the bending section, automatically calculates the forming angle, and informs the on-site workers through the on-site indicator light; If the angle is not enough, the worker enters the compensation value or modifies the bending angle on the equipment side, steps on the pedal, completes the second bending, and calculates the forming angle again. Continuous processing can be carried out after the requirements are met.

[0020] Example 2 Further illustrate with reference to Example 1, Figure 1-13 For the structure shown, the algorithm steps for automatic detection of steel bar units are as follows: A1. Database establishment: Fix the camera at a certain place to collect images. This place must ensure that complete image information including the bending area and the steel bar target can be obtained. Then manually mark the collected images to mark the specific location and feature information of the bending area and the steel bar target. These marking information and image data together constitute the elements of the database, thus completing the database establishment. The collected image set is ,in is the number of images collected; for each image , whose tag information set is ,in is the number of labels for each image; database It can be expressed as ; This formula is used to clearly define the structure of the database, ensuring that image data and corresponding label information can be accurately stored and used, providing an accurate data basis for subsequent algorithm processing; A2. Image acquisition and processing: Since the object to be measured is large in size, multiple cameras are used to capture images. First, the images collected by multiple cameras are stitched together. The number of cameras is The collected images are ; During the stitching process, for each image captured by a camera, it is necessary to find the overlapping area with other camera images and stitch them together based on the feature information of the overlapping area; after the stitching is completed, the steel bar contour is detected on the stitched image; by analyzing the grayscale changes and edge feature information of the image, the approximate contour position of the steel bar in the image is determined; Next, based on the determined steel bar contour, the steel bar bending area is further identified; This step requires combining the existing bending area marking information in the database with the feature information in the image to find the part that meets the bending area characteristics; Finally, calculate the length and angle of the reinforcement; camera The collected images are ,in is the image coordinate; for two adjacent cameras and , and the overlapping area is , the matching function of the overlapping area is defined as ;when Less than a certain threshold When , the two images are considered to match well in the overlapping area and can be stitched; the stitched image The values ​​in the overlapping area can be calculated by weighted average, that is, ,in and is the weight, which can be determined according to factors such as image quality and resolution; This formula is used to accurately determine whether adjacent camera images match in the overlapping area, and to reasonably stitch them together if they match well, ensuring that the stitched image can accurately reflect the full picture of the object being measured; The grayscale matrix of the image is , define the edge detection operator as ,in for The neighborhood of the point; by setting a threshold ,when Greater than When , the point is considered to be an edge point, thus determining the outline of the steel bar; This formula determines the edge points by calculating the grayscale difference between each point in the image and its neighboring points, thereby accurately detecting the outline of the steel bars in the image; Rebar bending area identification: The point set on the rebar contour is , for each point , calculate its curvature ,in For point The change in the angle between the two adjacent points is is the arc length between two points; when Greater than a threshold When , the area where the point is located is considered to be the bending area; this formula determines whether the point is located in the bending area by calculating the curvature of the point on the steel bar contour, thereby accurately identifying the steel bar bending area; Length and angle calculation formula: For steel bar length calculation, assume that the extracted skeleton line node is , the distance between nodes is ,in For Node Coordinates of; Steel bar length is the sum of the distances between all nodes, that is ; For the calculation of the bending angle, the slopes of the straight lines fitting the bending area are and , then the bending angle ; A3. Result visualization and alarm: The measured angle and length information will be visualized through the industrial computer deployed on site; A graphical interface can be used to display information such as the shape, length, and bending angle of the steel bar, so that operators can intuitively understand the status of the steel bar; If the measurement result deviates greatly from the theoretical value, an alarm will be triggered through the andon system. By setting a reasonable deviation threshold, when the difference between the measured value and the theoretical value exceeds the threshold, the andon system alarm will be triggered to remind the operator to check and handle the problem. The measured length is The theoretical length is , the deviation threshold is , then the length deviation judgment formula is Length deviation alarm is triggered; The measured angle is , the theoretical perspective is , the deviation threshold is , then the angle deviation judgment formula is Angle deviation alarm is triggered.

[0021] The hardware part mainly includes industrial cameras, lenses, industrial computers (including communication modules), display screens, and photo switches. The software part includes algorithm programs, visualization interfaces, and data interfaces. The algorithm part mainly detects automatic length and angle based on deep learning and image processing. The algorithm processing includes recognition and marking of bending areas, segmentation of bending areas, segmentation of steel bar targets within the area, straight line detection based on connected domains, and bending angle calculation.

[0022] Set up a multi-camera system, stitch images together through image synthesis and proceed to the next step of processing. Considering that the camera is set up far away from the bending point of the steel bar, two cameras can meet the requirements. The required hardware is two high-definition cameras and a processor with built-in algorithms. The high-definition camera is mainly responsible for data collection of steel bar units and steel bar sheets, and the processor is mainly used for real-time data processing and output results.

[0023] Taking G1-1 steel bar as an example, the workpiece size is 6250 mm×1382 mm. Taking into account the error caused by the edge transition of the image, the field of view is appropriately enlarged compared to the object. The field of view size can be set to 6300 mm×1432 mm. Assuming the measurement accuracy is 1mm / pixel, the required camera resolution is 6300 / 1=6300pixel, and the other direction is 1432 / 1=1432pixel, that is, the camera resolution needs to be at least 6300×1432=9021600pixel.

[0024] The Basler a2A2448-75ucBAS USB3.0 camera was selected. The pixel of the camera is 2448×2048. It meets the measurement requirements in width but not in length. Consider taking photos at three fixed points and fusing the three photos to meet the measurement requirements in the length direction.

[0025] Rebar bending area and rebar contour extraction algorithm based on deep learning, such as Figure 1 shown.

[0026] Since industrial applications require online detection as fast as possible and the computing platform on which the model relies is as cheap as possible, the development direction of deep learning in image detection in recent years has focused on high frame rate and low computing performance requirements. Research on lightweight models has emerged in recent years, among which the YOLO series model is undoubtedly one of the models with the best performance and the most extensive application. Considering that the latest YOLO v5 series model already has a very high recognition rate and detection accuracy, it has been shown in a large number of database verifications that it meets engineering requirements and the YOLOv5 series model has strong multi-platform portability, so the latest YOLOv5 series model is directly applied for target detection.

[0027] The above method is used to automatically identify the bending position of the steel bar from the real-time video and segment the detected position in real time for the next step of processing. Figure 2 shown.

[0028] In the preferred solution, the algorithm steps for extracting the steel bar bending area and steel bar contour based on deep learning are as follows: B1. Target detection: Use the YOLO v5 series model for target detection. Input the collected image into the YOLOv5 model, and the model will automatically identify the approximate location of the steel bar target and the bending area in the image. The formula can be expressed as: ,in is the input image, are the parameters of the model, is the output prediction result; the formula is used to describe the basic operation process of the YOLO v5 model. By adjusting the model parameters , which enables the model to accurately detect objects for different input images; B2. Segmentation of bending area sub-images: For the segmented bending area sub-images, the semantic segmentation method is used to segment the steel bar targets at the pixel level. The bending area sub-images are input into the UNet network, and the network classifies each pixel as a steel bar target pixel or a background pixel according to the feature information of the image, thereby achieving pixel-level segmentation of the steel bar targets. The formula can be expressed as: ,in is the input bending region sub-image, are the parameters of the network, The output segmentation result; the formula is used to describe the basic operation process of the UNet network, by adjusting the network parameters , which enables the network to perform accurate pixel-level segmentation of sub-images in different bending regions; B3. Calculation of steel bar length and bending angle based on image processing, such as Figure 3 shown.

[0029] For the steel bar segmentation map of the entire image obtained in the previous processing steps, the image processing method is directly used to analyze the characteristics of the connected domain; by analyzing the characteristic information such as the area, perimeter, shape, etc. of the connected domain, the structure and state of the steel bar are further determined; The connected domain is , its area formula is The perimeter formula is ,in Connected domain By calculating the area and perimeter of the connected domain, the thickness and shape of the steel bar can be preliminarily determined; The shape factor of the connected domain is ,The shape factor can be used to more accurately describe the shape characteristics of the connected domain, and plays an important role in judging whether the shape of the steel bar meets the requirements; Length and angle calculation: For the angle of the bending position, fit the straight line of the bending area; for the calculation of the length of the steel bar, directly extract the skeleton line of the connected domain, extract the nodes for each bending position, calculate the length of the skeleton line between each node, that is, the length of each section of the steel bar; perform skeleton extraction based on the previously obtained contour, observe the pixels through the extracted skeleton line, count the pixel length, and convert the pixel length to the real-world coordinate length, convert the number of pixels to the real length, and measure the length of each section of the steel bar; The specific formula is as follows: For the calculation of steel bar length, assume that the extracted skeleton line node is , the distance between nodes is ,in For Node Coordinates of; Steel bar length is the sum of the distances between all nodes, that is ; For the calculation of the bending angle, the slopes of the straight lines fitting the bending area are and , then the bending angle .

[0030] Example 3 During the bending process, the steel mesh will rebound due to the influence of residual stress, which will cause a deviation between the steel mesh forming line shape and the designed line shape, which will directly affect the hoisting and mold insertion of the steel mesh, the connection between the steel meshes, the forming accuracy of the steel parts, and other issues, thus affecting the construction quality. By using 3D laser scanning technology to measure the forming line shape and accuracy of the steel mesh, the forming line shape and accuracy of the steel mesh can be ensured, and the construction quality of the steel parts can be guaranteed. Figure 4 shown.

[0031] Select a 3D laser scanner to perform a panoramic scan of the area to be tested to obtain 3D point cloud data of the steel mesh, which provides a basis for subsequent steel point cloud data processing.

[0032] In the preferred solution, the automatic detection method of the steel mesh is: C1. Target ball installation: Place at least three target balls on the steel mesh to be tested. The target balls should be placed in a position that ensures that three target balls in the same position can be scanned at least twice when scanning the target in different directions. The centers of any three target balls cannot be on the same straight line. Figure 5 shown.

[0033] C2. Laser scanning: Use a ground-based 3D laser scanner to perform a panoramic scan of the area to be tested to obtain the original 3D point cloud data including the steel mesh. The distance between the 3D laser scanner and the mesh should be greater than 3m. The point cloud accuracy obtained is between 1.3-1.8mm, and the point cloud density is 90-120 points / square centimeter; Figure 6 shown.

[0034] C3. Point cloud processing: Process the original 3D point cloud data obtained by laser scanning, retain the target steel mesh, target ball and the point cloud around the target ball, delete other irrelevant point cloud data, use point cloud processing software to splice the point cloud data obtained by multiple scans, and finally obtain the complete 3D point cloud data of the steel mesh; Figure 7 shown.

[0035] C4. Denoising: define a coordinate system, where the horizontal plane direction is the X and Y axes, the height direction is the Z axis, the base surface direction is the X axis, and the coordinate origin is selected at the center of the base surface; to reduce interference with the later surface generation, after the point cloud is spliced, first manually remove the obvious redundant noise points outside the target steel bar, and then use the algorithm to remove the discrete isolated noise points; the specific algorithm is to determine whether the point is a discrete point by specifying the number of point clouds within the range of the spatial radius n, that is, draw a sphere with a radius of 0.01m with any point as the center of the sphere, and calculate the number of point clouds contained in the sphere. If the number of point clouds is less than 30-40, the point is identified as an isolated discrete noise point, and the point is deleted to obtain the denoised steel mesh point cloud data; such as Figure 8 shown.

[0036] C5. Circle center extraction: intercept any steel bar point cloud data, and use the RANSAC algorithm to extract the circle center coordinates (Xv, Yv) from the point cloud; Fig. 9 shown.

[0037] C6. Axis extraction: After extracting the center of each section, the steel bar axis can be directly fitted based on the least squares method; Fig.10 shown.

[0038] C7. Calculation of key dimensions: Based on the previously obtained coordinates of the circle center and the steel bar axis, key dimensions such as main reinforcement stirrup spacing, end length, flatness, plane angle, etc. can be automatically calculated; Calculation of main reinforcement and stirrup spacing: select any two main reinforcements, cut the point cloud sections at equal intervals perpendicular to the Z direction, obtain the center coordinates of the two sections under the Zh plane (Xh1, Yh2, Zt) and (Xh2, Yh2, Zt), and calculate the main reinforcement spacing Lh at this height; Calculation of end flushness: The coordinates of the center of the main reinforcement end face are (Xti, Yti, Zti), where i = 1, 2, …, m, m is the number of main reinforcements, and the end flushness can be obtained after the average height of the main reinforcement is calculated; Mesh surface flatness calculation: manually select the mesh area to be measured, find the least square fitting plane for the measured point cloud data (xi, yi, zi), where i = 0, 1, ..., m, obtain the fitting plane equation, calculate the distance from each point to the fitting plane, and the statistical value of the distance can be regarded as the flatness of the measured mesh; Calculation of joint length: The joint length includes the top length and the bottom length. Plane angle calculation: After the above steps, we can get the equations of two adjacent planes. The angle between the two planes is the angle between the normal vectors of the two planes. C8. Docking judgment: By calculating the relevant parameters of transverse docking, longitudinal docking and radial docking, the quality of the steel mesh is judged; when any one of the parameters ΔXmax, ΔYmax, ΔLmax exceeds the normal range, it indicates that the mesh quality is unqualified; the calculated data is exported and analyzed to determine whether the accuracy of the steel mesh meets the requirements.

[0039] In the preferred embodiment, the specific method of step C7 is: C71. Calculation of main reinforcement spacing: Select any two main reinforcements and cut the point cloud sections at equal intervals in the vertical Z direction to obtain Z h Coordinates of the centers of the two cut planes under the plane （X h1 ,Y h2 ,Zt ） and （X h2 ,Y h2 ,Z t ） , calculate the main reinforcement spacing at this height L h ; C72. Calculation of end flushness: The better the end flushness, the more convenient it is for the steel mesh to be quickly connected to the tower; the coordinates of the center of the main reinforcement end face are （X ti ,Y ti ,Z ti ） , average height of main reinforcement: ; The flatness of the end is: ; C73, mesh surface flatness calculation: Manually select the mesh surface area to be measured, and calculate the measured point cloud data. (xi,yi,zi)(i= 0,1,…,m) Find the least squares fitting plane and get the fitting plane equation as follows:

[0040] Calculate the distance from each point to the fitting plane: ; The flatness of the mesh is: ; in, d is the distance from the measuring point to the fitting plane, m is the mean error of the point from the measurement point to the fitting plane, which can be regarded as the flatness of the measured mesh. m The smaller the value, the flatter the mesh surface; the smaller the value, the more uneven the mesh surface. C74, joint length: The joint length includes the top length and the bottom length. Taking the top length as an example, the joint length L AB for A and B The distance between the ends of the main reinforcement is OXY Projection on a plane, B The axis of the main reinforcement and the axis of the end stirrups are OXY Intersection points of plane projections; C75. Calculation of mesh angle: After the above steps, the equations of two adjacent planes can be obtained as follows: and , then the normal vectors of the two planes are {A1, B1, C1} and {A2, B2, C2} , let the angle between these two normal vectors be φ , then the angle between these two planes is φ Calculated by the following formula: .

[0041] Mesh quality evaluation Transverse butt joint: If the spacing of the main bars is not well controlled, a cumulative effect will be formed at the end of the mesh surface. In severe cases, it will cause the steel bars of two adjacent segments to be misaligned when they are butt jointed up and down, and they cannot be butt jointed normally. Taking the first main bar of the mesh as the reference, calculate the deviation ΔX in the X direction between the two main bars to be butt jointed in the adjacent segments. If ΔXmax>ΔXmax0, it indicates that the steel mesh is unqualified and needs to be rectified.

[0042] Longitudinal butt joint: The butt joint deviation reflects the level of the main reinforcement misalignment of the adjacent segment steel mesh. If it is not well controlled, when the main reinforcement of the adjacent segments is butt jointed up and down, it may be impossible to use the steel connector for normal connection because the distance difference is too far. The butt joint deviation of the adjacent segment top is ΔLi=LAB_i+LAB_i+1-2L0, where L0 is the theoretical distance and LAB_i is the actual length of the i-th joint. When ΔLi is greater than ΔL0, it indicates that the steel mesh is unqualified and needs to be rectified.

[0043] Radial docking: If the flatness of the mesh surface is not well controlled, the radial deviation of the two adjacent mesh surfaces may be too large, and normal docking cannot be completed. The distance from the center of the end steel bar to the theoretical installation line is defined as ΔY. When ΔYmax≥ΔY0, it indicates that the mesh quality is unqualified and the docking cannot be completed, and rectification is required. Fig.11 shown.

[0044] Export data and analysis: Analyze the calculated data to determine whether the steel mesh meets the requirements. Fig.12 shown.

[0045] Example 4 At present, the construction of steel bars in reinforced concrete structures has begun to change from traditional manual in-situ binding to a modular construction method based on block construction. The use of modular construction can greatly reduce labor intensity, improve construction efficiency, and improve the accuracy of steel bar construction. However, the modular construction method of steel bars based on block assembly requires strict control of the construction accuracy of steel bar blocks to avoid the phenomenon that steel bar blocks cannot enter the steel bar assembly frame, and ensure the quality of steel bar construction.

[0046] For the detection of steel bar blocks in the forming stage, a set of automatic steel bar block detection technology is designed. The main bar positioning hole plate on the equipment is used to locate the main bar spacing of the steel bar block, and the servo motor is used to drive the steel bar block to tilt it to the designed angle; and the ACA626T-full temperature compensation high-precision digital output dual-axis inclination sensor is used to automatically measure the inclination angle of the steel bar block, and compare it with the theoretical value at all times to achieve real-time monitoring of the steel bar block.

[0047] In the preferred embodiment, the automatic detection method of steel bar blocks is specifically as follows: Fig.13 shown.

[0048] D1. The main controlled data in the steel bar block forming stage include: inclination ≤1 / 1000 and ≤8mm, spacing between main bars ≤±5mm, spacing between main bars in the same row ≤±10mm, spacing between stirrups ≤±10mm; D2. The spacing between main reinforcement rows and the spacing between main reinforcement rows are limited by the main reinforcement positioning hole plate on the equipment to meet the accuracy requirements. The spacing between stirrups is controlled by the comb rod structure on the equipment to also meet the accuracy requirements. D3. The inclination angle of the steel bar block is controlled by the angle adjustment device on the equipment, which is driven by a servo motor. Theoretically, the accuracy can reach 0.05°, which meets the use requirements; In order to verify that the angle accuracy of the formed steel bar block meets the requirements, a full temperature compensation high-precision digital output dual-axis inclination sensor is used to automatically measure the inclination angle and compare it with the theoretical value, so as to achieve the purpose of detecting the production angle of the steel bar block; Specifically, four full-temperature-compensated high-precision digital output dual-axis inclination sensors are installed on the four angle adjustment poles of the equipment. By comparing and analyzing the measured data after the block angle adjustment with the equipment input data, the purpose of automatically detecting the block forming accuracy is achieved.

[0049] Example 5 As bridge construction gradually develops towards industrialization and intelligence, the construction of reinforced concrete bridge towers is also developing towards industrial construction. For the construction of bridge tower reinforcement, the traditional bridge tower construction method has begun to change in recent years, and most of the reinforcement construction has been transferred to the ground factory for production, and then assembled and formed in the reinforcement assembly frame before being hoisted to the tower for installation, which greatly improves the efficiency of reinforcement construction. However, the construction accuracy of the reinforcement components must be strictly controlled during construction under the tower, otherwise the reinforcement components will be misaligned and unable to be connected after being hoisted onto the tower, thus affecting the efficiency of reinforcement construction. Therefore, the construction accuracy of bridge tower reinforcement must be strictly controlled when using the component construction method.

[0050] For the inspection of steel bar components, Leica 3D Disto equipment is used to accurately select measurement points through wide-angle images, and data is collected through the DJI M300 RTK drone system equipped with the Sail PSDK102S V3 lens to achieve real-time measurement of steel bar components in the tire frame and on the top of the tower, ensuring the accuracy and safety of steel bar component construction.

[0051] The Leica 3D Disto device can accurately select measurement points through wide-angle images. The laser will automatically align with the selected points on the image. One station can measure 3 to 4 target points on the measuring rod on the control edge. It is easy to operate and has high measurement efficiency, meeting the time requirements of requirement 1. Due to the high cost-effectiveness of the device, two devices can be used for simultaneous measurement if necessary to further improve efficiency.

[0052] The DJI M300 RTK UAV system is equipped with the Xaar PSDK102S V3 lens for data collection. The UAV system includes the aircraft, remote controller, visual system, infrared sensing system, etc. The key parameters of the UAV system and lens are shown in Table 5.2.4-1 and Table 5.2.4-2.

[0053] During the manufacturing process, the steel bar parts are mainly measured in three states: assembly inside the tire frame, storage outside the tire frame, and hoisting on the tower. The assembly inside the tire frame is divided into three working conditions: before assembly and welding, after welding, and after loosening the adjustment screw. The working condition outside the tire frame is the deadweight state. The hoisting connection on the tower is divided into three working conditions: before the cone sleeve connection is completed after the posture adjustment, after the cone sleeve connection is completed, and after the concrete pouring is completed. There are a total of 7 working conditions. In the preferred solution, the automatic detection method of steel bar parts is specifically to measure the steel bar parts in three states during the production process: assembly in the tire frame, storage outside the tire frame, and hoisting on the tower; E1. Measurement inside the tire frame: Figure 14-16 shown.

[0054] Measurement before welding after assembly is completed: 8 corner points are measured at the top, middle and bottom, and the measurement points are converted to the outer measurement of the tire frame through the conversion measurement method for easy observation and measurement. Specifically, 2 marker poles are set on each side of the top and bottom of the component, and the Leica 3D Disto equipment is used to measure the target points on the marker poles. The spatial position of the component connection point is calculated based on the target point position and the pole length. Three measuring poles are arranged at the top, middle and bottom of each control side; Measurement after welding: Also measure 8 corner points at the top, middle and bottom, because there may be welding deformation after welding, which needs to be measured and compared with the theory; After undoing the adjustment screw: still measure the 8 corner points of the top, middle and bottom. Since stress rebound deformation may occur after the external force of the top rod is removed, it is necessary to measure and compare with the theory; E2. Measurement outside the tire frame: On flat ground, measure 8 corner points of the top, middle and bottom in a free state, as well as the main reinforcement spacing, including the outer circle, middle circle and inner circle of the top; the outer circle and inner circle of the bottom, measure the dimensions, coordinates, inclination and other data of the components in a completely free state, and compare them with the theory; E3. Tower hoisting connection measurement: After inserting the bell mouth guide device and before connecting the cone sleeve, measure: measure the spatial position of the 8 corner points of the top of the component to guide the docking of the components and visualize the docking points of the previous component and the current component; After the cone sleeve is connected: measure the spatial position of the 8 corner points of the top or bottom of the component to check the installation accuracy of the component; After concrete pouring: measure the spatial positions of the 8 corner points of the top opening, analyze the impact of pouring on the linear shape of the tower limbs, give pre-deflection suggestions for the next segment, measure the position of the main reinforcement of the top opening, establish an actual tower column model, and compare and analyze it with the theory.

[0055] Example 6 The above reasons cause the 3D laser scanner to be in a shaking state, so it is difficult to obtain high-precision point cloud data. In order to obtain high-precision point cloud data under the above-mentioned influencing conditions, this paper selects the drone image reconstruction 3D point cloud method to obtain point cloud data. The drone obtains image data from high altitudes to solve the station setting, shaking, and occlusion problems caused by the mobility of equipment and personnel. The feature point extraction of the image reconstruction point cloud process and the joint settlement of spatial points can effectively overcome the platform shaking problem. On the basis of determining the data collection method, such as Fig.17 shown.

[0056] In the preferred solution, the algorithm steps of the drone system equipped with a laser lens for data collection in the automatic detection method of steel bar parts are as follows: F1. Data collection: When performing close-range high-resolution image acquisition at the construction site, considering that mobile equipment and personnel seriously affect the safety of the UAV's automatic route planning flight, and the relatively small workload of steel bar component data collection, manual flight is used to obtain UAV images to ensure the safety and image resolution of the UAV; images are acquired according to the "Z"-shaped flight trajectory to obtain sufficient and appropriate steel bar component image data; F2. Image 3D reconstruction: Based on the comprehensive advantages of popular software in the market in terms of computing speed and accuracy, DJI Intelligent Map was selected to perform 3D reconstruction on the collected image data; DJI Intelligent Map can process image data and reconstruct point cloud data. The obtained point cloud data can better reflect the shape and structural information of steel bar parts; Figure 18-19 shown.

[0057] F3. Spatial positioning algorithm: Slice separation: Slice the point cloud data of the steel bar parts along the Z axis based on the straight-through filter to separate the point cloud data at different elevation positions; count the number of point cloud slice data from large to small along the elevation direction; Fig. 20 shown.

[0058] Feature positioning: When longitudinal reinforcement appears, the amount of point cloud data will jump in magnitude, and this jump will only occur once, that is, the first section that appears is the longitudinal reinforcement section; in this way, the top surface positioning of the longitudinal reinforcement of the point cloud data is achieved, thereby establishing the association and mapping relationship between the image and the point cloud data in spatial position; F4, Single steel bar fitting algorithm: Fitting model selection, based on sampling consistency algorithm, set the circle model and cylinder model to fit the point cloud data; Fitting process and optimization: According to the design requirements, the threshold is set to 3-4cm, the number of iterations is set to 10,000-20,000 times, and the steel bar data is fitted; through the fitting process, the longitudinal steel bar data of the steel bar component point cloud is fitted, and the relevant data information of the steel bar is further extracted; F5. Point cloud data filtering: In the above steps, the original point cloud data is filtered to remove noise points and abnormal points to improve data quality.

[0059] In the preferred solution, the steps of the single steel bar fitting algorithm are: Based on the sampling consistency algorithm, the circle model and the cylinder model are set to fit the point cloud data. The circle and cylinder models are shown in the following formula: ; In the formula, is a three-dimensional circle in space, is a point on a spatial sphere or a spatial plane, is the center of the sphere, is the radius of the sphere, is the coefficient on the spatial plane; .

[0060] Taking circle fitting as an example, the steps for point cloud fitting are as follows: 1. Project the point cloud data to the XOY plane; 2. Initialize the number of cycles and the point set within the bureau; 3. Extract 3 points from the boundary point set and solve the parameters of the circle, i.e. the center data and radius r; 4. Calculate the distance d from each boundary point to the center of the circle obtained in step 2). If dr , then the point is counted into the internal point set; 5. Calculate the number of points on the circle, recorded as M .like M Greater than threshold M min , then the estimation is considered successful and go to step 5, otherwise go to step 6; 6. Recalculate the parametric model of the circle using the least squares method for all points in the point set; 7. Set the number of iterations. K = K +1 if , then end, otherwise go to 2.

[0061] From the above steps, we can see that the key parameters of point cloud data fitting include threshold, number of iterations and many parameters of random sampling consistency algorithm. This paper sets is 3 cm; set the number of iterations to 10,000, fit the steel bar data, and visualize the fitting results. The results are shown in 21. The center of the steel bar fitting circle is , the fitting diameter is 4.2 cm. Compared with the designed diameter of the steel bar of 3.6 cm, this result is too large. This is because the steel bar has a certain inclination. Therefore, when projected to the XOY plane, the distribution range of the point cloud becomes larger, resulting in a larger fitting of the steel bar diameter. Therefore, this method needs to be further optimized.

[0062] The model obtained based on the cylindrical model fitting is as follows Fig. 22 The cylinder fitting diameter is 3.4 cm, and the coordinates on the axis are , the axis direction vector is Compared with circle fitting, the cylindrical model directly fits the real point cloud data, so its result is relatively accurate. However, the cylindrical model requires too many parameters to be set and relies on manual experience. The above embodiments are only preferred method solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the method solutions recorded in the claims, including equivalent replacement solutions of the method features in the method solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for controlling the forming accuracy of a bridge tower steel bar component, characterized by: The method includes: an automatic detection method for steel bar unit elements, an automatic detection method for steel bar meshes, an automatic detection method for steel bar blocks and an automatic detection method for steel bar parts; The automatic detection method of steel bar unit element is as follows: by setting a support frame to install a camera and lens on the mobile head, taking pictures after bending and forming to calculate the forming angle, and using an algorithm based on deep learning and image processing to automatically detect the length and angle; The automatic detection method of steel mesh is: using 3D laser scanning technology, through the steps of target ball installation, laser scanning, point cloud processing, denoising, center extraction, axis extraction, and key dimension calculation, the forming line shape and accuracy of the steel mesh are measured; The automatic detection method of steel bar blocks is as follows: the automatic detection technology of steel bar blocks locates the main bar spacing through the main bar positioning hole plate, the servo motor drives the tilt to the designed angle, and the full temperature compensation high-precision digital output dual-axis inclination sensor is used to automatically measure the tilt angle and compare it with the theoretical value; The automatic detection method of steel bar components is as follows: the automatic detection technology of steel bar components uses 3D Disto equipment to select measurement points, and the drone system is equipped with a laser lens to collect data, and performs real-time measurement of steel bar components in the tire frame and on the top of the tower.

2. According to claim 1, a method for controlling the forming accuracy of a bridge tower steel bar component is characterized by: A camera and lens are installed on the support frame. When the steel bar is bent and the bending head is returned to its position, the camera takes a picture of the steel bar, captures the image of the bending section, automatically calculates the forming angle, and informs the on-site workers through the on-site indicator light; If the angle is not enough, the worker enters the compensation value or modifies the bending angle on the equipment side, steps on the pedal, completes the second bending, and calculates the forming angle again. Continuous processing can be carried out after the requirements are met.

3. According to the method for controlling the forming accuracy of a bridge tower steel bar component as described in claim 2, it is characterized by: The steps of the automatic detection algorithm for steel bar units are as follows: A1. Database establishment: Fix the camera at a certain place to collect images. This place must ensure that complete image information including the bending area and the steel bar target can be obtained; then manually mark the collected images to mark the specific location and feature information of the bending area and the steel bar target; These marking information together with the image data constitute the elements of the database, thus completing the establishment of the database; The collected image set is ,in is the number of images collected; for each image , whose tag information set is ,in is the number of labels for each image; database It can be expressed as ; This formula is used to clearly define the structure of the database, ensuring that image data and corresponding label information can be accurately stored and used, providing an accurate data basis for subsequent algorithm processing; A2. Image acquisition and processing: Since the object to be measured is large in size, multiple cameras are used to capture images. First, the images collected by multiple cameras are stitched together. The number of cameras is The collected images are ; During the stitching process, for each camera image, it is necessary to find the overlapping area with other camera images and stitch them together based on the feature information of the overlapping area. After the stitching is completed, the steel bar contour is detected on the stitched image. The approximate contour position of the steel bar in the image is determined by analyzing the grayscale changes and edge feature information of the image. Next, based on the determined steel bar contour, the steel bar bending area is further identified; This step requires combining the existing bending area marking information in the database with the feature information in the image to find the part that meets the bending area characteristics; Finally, calculate the length and angle of the reinforcement; camera The collected images are ,in is the image coordinate; for two adjacent cameras and , and the overlapping area is , the matching function of the overlapping area is defined as ;when Less than a certain threshold When , the two images are considered to match well in the overlapping area and can be stitched; the stitched image The values ​​in the overlapping area can be calculated by weighted average, that is, ,in and is the weight, which can be determined according to factors such as image quality and resolution; This formula is used to accurately determine whether adjacent camera images match in the overlapping area, and to reasonably stitch them together if they match well, ensuring that the stitched image can accurately reflect the full picture of the object being measured; The grayscale matrix of the image is , define the edge detection operator as ,in for The neighborhood of a point; by setting a threshold ,when Greater than When , the point is considered to be an edge point, thus determining the outline of the steel bar; This formula determines the edge points by calculating the grayscale difference between each point in the image and its neighboring points, thereby accurately detecting the outline of the steel bars in the image; Rebar bending area identification: The point set on the rebar contour is , for each point , calculate its curvature ,in For point The change in the angle between the two adjacent points is is the arc length between two points; when Greater than a threshold When , the area where the point is located is considered to be a bending area; This formula calculates the curvature of a point on the steel bar contour to determine whether the point is located in the bending area, thereby accurately identifying the bending area of ​​the steel bar; Length and angle calculation formula: For steel bar length calculation, assume that the extracted skeleton line node is , the distance between nodes is ,in For Node Coordinates of the steel bar; length of the steel bar is the sum of the distances between all nodes, that is ; For the calculation of the bending angle, the slopes of the straight lines fitting the bending area are and , then the bending angle ; A3. Result visualization and alarm: The measured angle and length information will be visualized through the industrial computer deployed on site; A graphical interface can be used to display information such as the shape, length, and bending angle of the steel bar, so that operators can intuitively understand the status of the steel bar; If the measurement result deviates greatly from the theoretical value, an alarm will be triggered through the andon system. By setting a reasonable deviation threshold, when the difference between the measured value and the theoretical value exceeds the threshold, the andon system alarm will be triggered to remind the operator to check and handle the problem. The measured length is , the theoretical length is , the deviation threshold is , then the length deviation judgment formula is Length deviation alarm is triggered; The measured angle is , the theoretical perspective is , the deviation threshold is , then the angle deviation judgment formula is Angle deviation alarm is triggered.

4. According to claim 2, a method for controlling the forming accuracy of a bridge tower steel bar component is characterized by: The algorithm steps for extracting the bending area and contour of steel bars based on deep learning are as follows: B1. Target detection: Use the YOLO v5 series model for target detection. Input the collected image into the YOLO v5 model, and the model will automatically identify the approximate location of the steel bar target and the bending area in the image. The formula can be expressed as: ,in is the input image, are the parameters of the model, is the output prediction result; the formula is used to describe the basic operation process of the YOLO v5 model. By adjusting the model parameters , which enables the model to accurately detect objects on different input images; B2. Segmentation of bending area sub-images: For the segmented bending area sub-images, the semantic segmentation method is used to segment the steel bar targets at the pixel level. The bending area sub-images are input into the UNet network, and the network classifies each pixel as a steel bar target pixel or a background pixel according to the feature information of the image, thereby achieving pixel-level segmentation of the steel bar targets. The formula can be expressed as: ,in is the input bending region sub-image, are the parameters of the network, The output segmentation result; the formula is used to describe the basic operation process of the UNet network, by adjusting the network parameters , which enables the network to perform accurate pixel-level segmentation of sub-images in different bending regions; B3. Calculation of steel bar length and bending angle based on image processing: For the steel bar segmentation map of the entire image obtained in the previous processing steps, the image processing method is directly used to analyze the characteristics of the connected domain; by analyzing the characteristic information such as the area, perimeter, shape, etc. of the connected domain, the structure and state of the steel bar are further determined; The connected domain is , its area formula is The perimeter formula is ,in Connected domain By calculating the area and perimeter of the connected domain, the thickness and shape of the steel bar can be preliminarily determined; The shape factor of the connected domain is ,The shape factor can be used to more accurately describe the shape characteristics of the connected domain, and plays an important role in judging whether the shape of the steel bar meets the requirements; Length and angle calculation: For the angle of the bending position, fit the straight line of the bending area; for the calculation of the length of the steel bar, directly extract the skeleton line of the connected domain, extract the nodes for each bending position, calculate the length of the skeleton line between each node, that is, the length of each section of the steel bar; perform skeleton extraction based on the previously obtained contour, observe the pixels through the extracted skeleton line, count the pixel length, and convert the pixel length to the real-world coordinate length, convert the number of pixels to the real length, and measure the length of each section of the steel bar; The specific formula is as follows: For the calculation of steel bar length, assume that the extracted skeleton line node is , the distance between nodes is ,in For Node Coordinates of the steel bar; length of the steel bar is the sum of the distances between all nodes, that is ; For the calculation of the bending angle, the slopes of the straight lines fitting the bending area are and , then the bending angle .

5. According to claim 1, a method for controlling the forming accuracy of a bridge tower steel bar component is characterized by: The automatic detection method of steel mesh is: C1. Target ball installation: Place at least three target balls on the steel mesh to be tested. The target balls should be placed in a position that ensures that three target balls in the same position can be scanned at least twice when scanning the target in different directions. The centers of any three target balls cannot be on the same straight line. C2. Laser scanning: Use a ground-based 3D laser scanner to perform a panoramic scan of the area to be tested to obtain the original 3D point cloud data including the steel mesh. The distance between the 3D laser scanner and the mesh should be greater than 3m. The point cloud accuracy obtained is between 1.3-1.8mm, and the point cloud density is 90-120 points / square centimeter; C3. Point cloud processing: Process the original 3D point cloud data obtained by laser scanning, retain the target steel mesh, target ball and the point cloud around the target ball, delete other irrelevant point cloud data, use point cloud processing software to splice the point cloud data obtained by multiple scans, and finally obtain the complete 3D point cloud data of the steel mesh; C4, Denoising: Define the coordinate system, where the horizontal plane direction is the X and Y axes, the height direction is the Z axis, the base surface direction is the X axis, and the coordinate origin is selected at the center of the base surface; in order to reduce the interference with the later surface generation, after the point cloud is spliced, first manually remove the obvious redundant noise points outside the target steel bar, and then use the algorithm to remove the discrete isolated noise points; the specific algorithm is to determine whether the point is a discrete point by the number of point clouds within the specified space radius n, that is, draw a sphere with a radius of 0.01m with any point as the center of the sphere, and calculate the number of point clouds contained in the sphere. If the number of point clouds is less than 30-40, the point is identified as an isolated discrete noise point, and the point is deleted to obtain the denoised steel mesh point cloud data; C5. Circle center extraction: intercept any steel bar point cloud data, and use the RANSAC algorithm to extract the circle center coordinates (Xv, Yv) from the point cloud; C6. Axis extraction: After extracting the center of each section, the steel bar axis can be directly fitted based on the least squares method; C7. Calculation of key dimensions: Based on the previously obtained coordinates of the circle center and the steel bar axis, key dimensions such as main reinforcement stirrup spacing, end length, flatness, plane angle, etc. can be automatically calculated; Calculation of main reinforcement and stirrup spacing: select any two main reinforcements, cut the point cloud sections at equal intervals perpendicular to the Z direction, obtain the center coordinates of the two sections under the Zh plane (Xh1, Yh2, Zt) and (Xh2, Yh2, Zt), and calculate the main reinforcement spacing Lh at this height; Calculation of end flushness: The coordinates of the center of the main reinforcement end face are (Xti, Yti, Zti), where i = 1, 2, …, m, m is the number of main reinforcements, and the end flushness can be obtained after the average height of the main reinforcement is calculated; Mesh surface flatness calculation: manually select the mesh area to be measured, find the least square fitting plane for the measured point cloud data (xi, yi, zi), where i = 0, 1, ..., m, obtain the fitting plane equation, calculate the distance from each point to the fitting plane, and the statistical value of the distance can be regarded as the flatness of the measured mesh; Calculation of joint length: The joint length includes the top length and the bottom length. Plane angle calculation: After the above steps, we can get the equations of two adjacent planes. The angle between the two planes is the angle between the normal vectors of the two planes. C8. Docking judgment: By calculating the relevant parameters of transverse docking, longitudinal docking and radial docking, the quality of the steel mesh is judged; when any one of the parameters ΔXmax, ΔYmax, ΔLmax exceeds the normal range, it indicates that the mesh quality is unqualified; the calculated data is exported and analyzed to determine whether the accuracy of the steel mesh meets the requirements.

6. The method for controlling the forming accuracy of a bridge tower steel bar component according to claim 5 is characterized by: The specific method of step C7 is: C71. Calculation of main reinforcement spacing: Select any two main reinforcements and cut the point cloud sections at equal intervals in the vertical Z direction to obtain Z h Coordinates of the centers of the two cut planes under the plane （X h1 ,Y h2 ,Z t ） and （X h2 ,Y h2 ,Z t ） , calculate the main reinforcement spacing at this height L h ； C72. Calculation of end flushness: The better the end flushness, the more convenient it is for the steel mesh to be quickly connected to the tower; the coordinates of the center of the main reinforcement end face are （X ti ,Y ti ,Z ti ） , average height of main reinforcement: ； The flatness of the end is: ； C73, mesh surface flatness calculation: Manually select the mesh surface area to be measured, and calculate the measured point cloud data. (xi,yi,zi)(i=0, 1,…,m) Find the least squares fitting plane and get the fitting plane equation as follows: Calculate the distance from each point to the fitting plane: ； The flatness of the mesh is: ； in, d is the distance from the measuring point to the fitting plane, m is the mean error of the point from the measurement point to the fitting plane, which can be regarded as the flatness of the measured mesh. m The smaller the value, the flatter the mesh surface; the smaller the value, the more uneven the mesh surface. C74, joint length: The joint length includes the top length and the bottom length. Taking the top length as an example, the joint length L AB for A and B The distance between the ends of the main reinforcement is OXY Projection on a plane, B The axis of the main reinforcement and the axis of the end stirrups are OXY Intersection points of plane projections; C75. Calculation of mesh angle: After the above steps, the equations of two adjacent planes can be obtained as follows: and , then the normal vectors of the two planes are {A1, B1, C1} and {A2, B2, C2} , let the angle between these two normal vectors be φ , then the angle between these two planes is φ Calculated by the following formula: 。 7. The method for controlling the forming accuracy of a bridge tower steel bar component according to claim 1 is characterized by: The specific method for automatic detection of steel bar blocks is as follows: D1. The main controlled data in the steel bar block forming stage include: inclination ≤1 / 1000 and ≤8mm, spacing between main bars ≤±5mm, spacing between main bars in the same row ≤±10mm, spacing between stirrups ≤±10mm; D2. The spacing between main reinforcement rows and the spacing between main reinforcement rows are limited by the main reinforcement positioning hole plate on the equipment to meet the accuracy requirements. The spacing between stirrups is controlled by the comb rod structure on the equipment to also meet the accuracy requirements. D3. The inclination angle of the steel bar block is controlled by the angle adjustment device on the equipment, which is driven by a servo motor. Theoretically, the accuracy can reach 0.05°, which meets the use requirements; In order to verify that the angle accuracy of the formed steel bar block meets the requirements, a full temperature compensation high-precision digital output dual-axis inclination sensor is used to automatically measure the inclination angle and compare it with the theoretical value, so as to achieve the purpose of detecting the production angle of the steel bar block; Specifically, four full-temperature-compensated high-precision digital output dual-axis inclination sensors are installed on the four angle adjustment poles of the equipment. By comparing and analyzing the measured data after the block angle adjustment with the equipment input data, the purpose of automatically detecting the block forming accuracy is achieved.

8. The method for controlling the forming accuracy of a bridge tower steel bar component according to claim 1 is characterized by: The specific method of automatic inspection of steel bar parts is to measure the steel bar parts in three states during the production process: assembly in the tire frame, storage outside the tire frame, and hoisting on the tower; E1. Measurement inside the tire frame: Measurement before welding after assembly is completed: 8 corner points are measured at the top, middle and bottom, and the measurement points are converted to the outer measurement of the tire frame through the conversion measurement method for easy observation and measurement. Specifically, 2 marker poles are set on each side of the top and bottom of the component, and the Leica 3D Disto equipment is used to measure the target points on the marker poles. The spatial position of the component connection point is calculated based on the target point position and the pole length. Three measuring poles are arranged at the top, middle and bottom of each control side; Measurement after welding: Also measure 8 corner points at the top, middle and bottom, because there may be welding deformation after welding, which needs to be measured and compared with the theory; After undoing the adjustment screw: measure the 8 corner points at the top, middle and bottom. Since stress rebound deformation may occur after the external force of the top rod is removed, it is necessary to measure and compare with the theory. E2. Measurement outside the tire frame: On flat ground, measure 8 corner points of the top, middle and bottom in a free state, as well as the main reinforcement spacing, including the outer circle, middle circle and inner circle of the top; the outer circle and inner circle of the bottom, measure the dimensions, coordinates, inclination and other data of the components in a completely free state, and compare them with the theory; E3. Tower hoisting connection measurement: After inserting the bell mouth guide device and before connecting the cone sleeve, measure: measure the spatial position of the 8 corner points of the top of the component to guide the docking of the components and visualize the docking points of the previous component and the current component; After the cone sleeve is connected: measure the spatial position of the 8 corner points of the top or bottom of the component to check the installation accuracy of the component; After concrete pouring: measure the spatial positions of the 8 corner points of the top opening, analyze the impact of pouring on the linear shape of the tower limbs, give pre-deflection suggestions for the next segment, measure the position of the main reinforcement of the top opening, establish an actual tower column model, and compare and analyze it with the theory.

9. A method for controlling the forming accuracy of a bridge tower steel bar component according to claim 8, characterized in that: In the automatic detection method of steel bar parts, the algorithm steps of the drone system equipped with a laser lens for data collection are as follows: F1. Data collection: When performing close-range high-resolution image acquisition at the construction site, considering that mobile equipment and personnel seriously affect the safety of the UAV's automatic route planning flight, and the relatively small workload of steel bar component data collection, manual flight is used to obtain UAV images to ensure the safety and image resolution of the UAV; images are acquired according to the "Z"-shaped flight trajectory to obtain sufficient and appropriate steel bar component image data; F2. Image 3D reconstruction: Based on the comprehensive advantages of popular software on the market in terms of computing speed and accuracy, DJI Intelligent Map was selected to perform 3D reconstruction on the collected image data. DJI Intelligent Map can process image data and reconstruct point cloud data. The obtained point cloud data can better reflect the shape and structural information of steel bar parts. F3, spatial positioning algorithm: Slice separation: Slice the point cloud data of the steel bar parts along the Z axis based on straight-through filtering to separate the point cloud data at different elevations; count the number of point cloud slice data from large to small along the elevation direction; Feature positioning: When longitudinal reinforcement appears, the amount of point cloud data will jump in magnitude, and this jump will only occur once, that is, the first section that appears is the longitudinal reinforcement section; in this way, the top surface positioning of the longitudinal reinforcement of the point cloud data is achieved, thereby establishing the association and mapping relationship between the image and the point cloud data in spatial position; F4, Single steel bar fitting algorithm: Fitting model selection, based on sampling consistency algorithm, set the circle model and cylinder model to fit the point cloud data; Fitting process and optimization: According to the design requirements, the threshold is set to 3-4cm, the number of iterations is set to 10,000-20,000 times, and the steel bar data is fitted; through the fitting process, the longitudinal steel bar data of the steel bar component point cloud is fitted, and the relevant data information of the steel bar is further extracted; F5. Point cloud data filtering: In the above steps, the original point cloud data is filtered to remove noise points and abnormal points to improve data quality.

10. A method for controlling the forming accuracy of a bridge tower steel bar component according to claim 9, characterized in that: The steps of single steel bar fitting algorithm are as follows: Based on the sampling consistency algorithm, the circle model and the cylinder model are set to fit the point cloud data. The circle and cylinder models are shown in the following formula: ； In the formula, is a three-dimensional circle in space, is a point on a sphere or plane, is the center of the sphere, is the radius of the sphere, is the coefficient on the spatial plane; 。

Images