Method for calculating and flattening non-straight position of medium-thickness plate based on three-dimensional visual reconstruction

Abstract

The invention discloses a medium-thickness plate non-straight position calculation and flattening method based on three-dimensional visual reconstruction. Point cloud data of the surface of a medium-thickness steel plate is obtained through three-dimensional visual scanning and preprocessed, an improved fitting algorithm is used for three-dimensional reconstruction of the steel plate, and an optimal three-dimensional curved surface equation is obtained through optimization by combining a genetic algorithm with a nonlinear least square method. And calculating non-straight position parameters according to the three-dimensional curved surface equation, wherein the non-straight position parameters comprise a bending extreme point, a cushion block position point, the bending curvature of the extreme point and the like. Then, according to the position parameters, designing an intelligent flattening amount calculation model in combination with the bending curvature at the extreme point to calculate flattening parameters. And finally, parameters are optimized through a deep learning model, and a flattening machine is guided to perform flattening operation on the non-straight position in combination with closed-loop control. According to the method, accurate three-dimensional reconstruction of the medium-thickness steel plate is achieved through a high-precision three-dimensional visual scanning technology in combination with an improved fitting algorithm, an intelligent flattening amount calculation model and a deep learning model are introduced to accurately calculate the non-straight position and flattening parameters, and the flattening precision and efficiency are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of steel plate flatness detection, and specifically relates to a method for calculating and flattening the uneven position of a medium-thick plate based on three-dimensional visual reconstruction. Background Art

[0002] Medium and thick steel plates are one of the main products of the steel industry and are widely used in the fields of construction engineering, shipbuilding and bridges, machinery manufacturing, nuclear power equipment, etc. With the innovative development of rolling technology and the continuous increase in the thickness and width of steel plates, the problems of uneven deformation and severe internal stress in the rolling process of medium and thick plates have become prominent. Partial bending, warping and other deformations are prone to occur during the heat treatment process, resulting in a decrease in the flatness index. The steel plate application industry has strict requirements on the flatness of steel plates, and usually requires flattening of uneven positions.

[0003] For the flattening and straightening of medium and thick steel plates, the traditional flattening position judgment is mainly based on manual experience. The flattening process is generally divided into six stages: loading, bending degree inspection, feeding, placing pads, flattening and discharging. During the whole process, the flattening position control mainly relies on manual control or semi-automatic control through displacement sensors, resulting in low quality and efficiency of flattening. Therefore, the present invention proposes a method for calculating and flattening the uneven position of medium and thick plates based on three-dimensional visual reconstruction. The point cloud data of the surface of the medium and thick steel plate is obtained by three-dimensional visual scanning and preprocessed, and the steel plate is reconstructed in three dimensions using an improved fitting algorithm, and the optimal three-dimensional surface equation is obtained by using a genetic algorithm combined with nonlinear least squares optimization. The uneven position parameters are calculated according to the three-dimensional surface equation, including the bending extreme point, the pad position point, and the bending curvature of the extreme point. Then, according to the position parameters, the intelligent flattening amount calculation model is designed in combination with the bending curvature at the extreme point to calculate the flattening parameters. Finally, the parameters are optimized by a deep learning model, and the flattening machine is guided to flatten the uneven position in combination with closed-loop control. The present invention aims to realize accurate three-dimensional reconstruction of medium and thick steel plates through high-precision three-dimensional visual scanning technology combined with an improved fitting algorithm, introduce an intelligent flattening calculation model and a deep learning model to accurately calculate the uneven position and flattening parameters, and improve the flattening accuracy and efficiency. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides a method for calculating and flattening the uneven position of a medium-thick plate based on three-dimensional visual reconstruction, the method comprising the following steps: Step 1.1: Use a 3D camera to collect 3D point cloud data of the steel plate surface. To ensure that the point cloud data is in a suitable coordinate system, perform rotation correction on the point cloud. Step 1.2: Perform point cloud segmentation, calculate consistency by using a smooth estimation loss function, set a distance threshold, and segment the surface of the steel plate; Step 1.3: Use adaptive filtering technology to process the segmented point cloud data. First, calculate the distance between each point and the adjacent points, determine the distribution density of the points, and adjust the filtering intensity according to the distribution density to obtain the final steel plate surface point cloud data. .

[0005] Step 2.1: Using the improved fitting algorithm, in order to avoid falling into the local optimal solution, it is necessary to first estimate the initial parameters of the fitting function and define the function form as follows: , In the formula, Represents the probability density function value of the point cloud at (x, y), where μ x and μ y are the means of x and y respectively, where ,in are the variances of x and y, σ xy is the covariance of x and y. The genetic algorithm is used to determine the initial parameters of the three-dimensional surface equation. The genetic algorithm selects parameter combinations with higher fitness by simulating the evolutionary process, so that the algorithm can jump out of the local optimum and quickly obtain suitable initial parameters. Step 2.2: Based on the initial parameters, use the nonlinear least squares method to minimize the residual between the actual data and the three-dimensional surface equation, and introduce a dynamic adjustment strategy to dynamically adjust the step size according to the change of the error to determine the optimal characteristic parameters , The steel plate point cloud data obtained after preprocessing in step 1 , using the nonlinear least squares principle, solve the least squares solution of equation (2) above, dynamically adjust the step size and convergence threshold according to the fitting error to ensure that it converges as quickly as possible while meeting the accuracy requirements, and obtain the three-dimensional surface equation .

[0006] Step 3.1: Find the maximum point of the three-dimensional surface equation , as the downward pressure point, calculate the bending curvature K of this point, and the three-dimensional surface equation obtained by fitting , calculate its second-order partial derivative and For the calculation of curvature, the bending curvature K of the lower pressure point is obtained as: , K can directly reflect the flatness information of the steel plate. The size of K reflects the curvature of the steel plate surface. If K is small, it means that the surface is relatively flat; if K is large, it means that the curvature of the surface is large. Step 3.2: Solve the pad position point, pad 1 position point Set as the local lowest point of the three-dimensional surface equation, that is, the minimum point. The two pad position points are roughly symmetrically distributed relative to the downward pressure position point. During the downward pressure process, the two pad position points and the downward pressure position point are required to be in the same plane, and the plane is a vertical plane. According to the position point of pad 1 and the downward pressure position point, the plane equation of the vertical plane can be determined: in , C As a constant term, let the position point of pad 2 be on the three-dimensional surface equation, and the point is also located in the vertical plane mentioned above, then the constraint condition can be established to determine the position point of pad 2 : .

[0007] Step 4.1: According to the bending curvature of the extreme point and the specific location of the extreme point, design an intelligent flattening calculation model. The downward pressing position point is the position point of the two pads. and The fulcrum is the point. Based on the material mechanics and the parameters of the flattening equipment, the downward pressure is calculated by calculating the fulcrum spacing, bending moment and deformation. Step 4.2: Apply the calculated position parameters and pressure amount to the flattening equipment and guide the flattening machine to operate accurately at the pressure point. At the same time, combined with the fulcrum position and force analysis, ensure that the steel plate remains stable during the flattening process, realize the intelligent flattening operation of the steel plate, reduce manual intervention, and improve the flattening efficiency.

[0008] Step 5.1: Introduce the convolutional neural network deep learning model, train the historical data, and optimize the position parameters and flattening parameters. Utilize the powerful learning ability of the neural network to automatically identify and extract the uneven position features based on the point cloud data, and then obtain the optimal parameters; Step 5.2: During the flattening process, the 3D visual scanner is used to monitor the state changes of the steel plate surface in real time, and the data is fed back to the deep learning model to automatically adjust the parameters. This closed-loop control system can make dynamic adjustments during the steel plate flattening process to ensure that the flatness after each flattening meets the requirements and improve the overall operation efficiency and effect.

[0009] The flattening objects described in the present invention include computational flattening of medium-thick steel plates or other curved plate materials; a three-dimensional visual scanner obtains point cloud data of the surface of the curved steel plate, pre-processes the initial point cloud data, obtains a three-dimensional surface equation through an improved fitting algorithm, and uses a genetic algorithm combined with a nonlinear least squares method to adjust the characteristic parameters of the three-dimensional surface equation. The uneven position parameters are calculated according to the three-dimensional surface equation, including the bending extreme point, the pad position point, and the bending curvature of the extreme point. Then, based on the position parameters, an intelligent flattening calculation model is designed in combination with the bending curvature at the extreme point. Finally, the prediction of the position parameters is optimized through a deep learning model, and closed-loop control guides the flattening machine to flatten the uneven position.

[0010] The present invention has the following beneficial effects: 1. The present invention overcomes the problem that the traditional uneven position detection method only considers the deformation of the steel plate in one direction and cannot fully reflect the overall deformation of the steel plate. For steel plates with complex shapes, the overall surface of the steel plate is fitted, which is low in cost, high in efficiency, and improves the calculation accuracy; Second, the present invention solves the three-dimensional surface equation through an improved fitting algorithm, thereby achieving high-precision fitting of the steel plate surface, and can effectively overcome the problems of low detection efficiency and poor flattening quality; Third, the present invention avoids the accuracy problems caused by steel plate placement errors, teaching errors, etc., and effectively reduces the residual between actual data and characteristic parameters; 4. The present invention performs three-dimensional reconstruction of the entire steel plate, which is suitable for computational detection of steel plates in complex environments. Compared with manual intervention methods, the present invention has higher accuracy and is suitable for computational flattening of steel plates of various specifications. It can be widely used in steel plate quality control, intelligent flattening equipment, steel plate rolling and other fields.

[0011] It should be understood that the foregoing general description and the following detailed description are exemplary illustrations and explanations and should not be used as limitations on the content of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. The drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 It is the overall flow chart of the present invention; Figure 2 It is a schematic diagram of the three-dimensional visual acquisition structure of the present invention; Figure 3 It is a schematic diagram of the downward pressing structure of the present invention. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] The calculation objects described in the present invention include the calculation of medium-thick steel plates or other curved plate materials.

[0015] In this embodiment, if Figure 1 As shown, the present invention includes four major steps: visual data processing, three-dimensional reconstruction, uneven position calculation and intelligent flattening.

[0016] In this embodiment, the visual system is arranged as follows Figure 2 As shown, the three-dimensional camera in the device does not have a very strict position requirement, as long as the steel plate to be tested is in the optimal imaging range of the system. The three-dimensional camera is installed on the detection frame and is located directly above the steel plate to be tested.

[0017] When scanning the surface, the 3D vision system starts to collect data, and the image data enters the industrial computer through the acquisition card to obtain the surface point cloud, obtain the initial data coordinates of the point cloud, and then perform point cloud rotation correction to put the point cloud data in a suitable coordinate system.

[0018] The RANSAC algorithm is used to segment the point cloud. The consistency is calculated by adopting a smooth estimation loss function, and the distance threshold is set to segment the surface of the steel plate. This can reduce data redundancy, focus on retaining data in areas with obvious features, and improve processing efficiency.

[0019] In the preprocessing of the sampled point cloud data, adaptive filtering technology is used. First, the distance between each point and the adjacent points is calculated to determine the density of the point distribution. For densely distributed areas, the filtering intensity is reduced to retain more details; for points with more noise and farther distance, the filtering intensity is increased to remove outliers, and the final steel plate surface point cloud data is obtained. This adaptive filtering process can effectively reduce noise while preserving surface details and improve the accuracy of subsequent fitting.

[0020] In this embodiment, the function form is defined as The initial parameters of the Gaussian distribution are determined using a genetic algorithm. The genetic algorithm selects a parameter combination with a higher fitness by simulating the evolutionary process, so that the algorithm can jump out of the local optimum and quickly obtain the appropriate initial parameters.

[0021] Based on the initial parameters, the nonlinear least squares method is used to minimize the residual between the actual data and the three-dimensional surface equation to determine the optimal characteristic parameters.

[0022] According to the steel plate point cloud data obtained after preprocessing , using the nonlinear least squares principle, solve The least squares solution is obtained, and a dynamic adjustment strategy is introduced to dynamically adjust the step size according to the change of the error. If the error decreases significantly during the iteration process, the step size is increased to accelerate convergence; if the error decreases slowly or oscillates, the step size is reduced to improve accuracy. Set the convergence threshold. When the error change is less than the threshold, stop the iteration, determine the optimal feature parameters, and obtain the three-dimensional surface equation. .

[0023] In this embodiment, the non-straight position parameters are calculated according to the three-dimensional surface equation, including the bending extreme point, the pad position point, and the bending curvature of the extreme point.

[0024] Finding the maximum point of a three-dimensional surface equation , as the downward pressure point. Calculate the curvature K of this point, and the three-dimensional surface equation obtained by fitting , calculate its second-order partial derivative and For the calculation of curvature, the bending curvature K of the lower pressure point is obtained as: .

[0025] Among them, K can intuitively reflect the flatness information of the steel plate, and the size of K reflects the curvature of the steel plate surface. If K is small, it means that the surface is relatively flat; if K is large, it means that the surface curvature is large.

[0026] Solve the pad position point, pad 1 position point Set as the local lowest point of the three-dimensional surface equation, that is, the minimum point. The two pad position points are roughly symmetrically distributed relative to the downward pressure position point. During the downward pressure process, the two pad position points and the downward pressure position point are required to be in the same plane, and the plane is a vertical plane. According to the position point of pad 1 and the downward pressure position point, the plane equation of the vertical plane can be determined: .

[0027] Let the position point of pad 2 be on the three-dimensional surface equation, and the point is also located in the above vertical plane, then the constraint condition can be established , to determine the position point of pad 2 .

[0028] In this embodiment, if Figure 3As shown, pressure leveling is based on the three-point bending model. Different amounts of pressure are applied according to the original bending amounts to ensure that the leveling purpose is achieved. When the steel plate bends upward, pads are placed on both sides of the lower surface of the steel plate, and then the pressure head of the press presses down the steel plate to make the steel plate bend in the opposite direction. The pressure head is lifted and the steel plate rebounds to a plane.

[0029] The flattening machine is based on the obtained position parameters to bend the steel plate to the maximum point The downward pressure position point is the position point of the two pads. and The fulcrum is used as the support. Based on the material mechanics and the parameters of the flattening equipment, the downward pressure and force are calculated by calculating the support distance, bending moment and deformation.

[0030] According to the calculated position parameters and pressure amount, they are applied to the flattening equipment, and the flattening machine is guided to operate accurately at the pressure point. At the same time, combined with the fulcrum position and force analysis, the steel plate is guaranteed to remain stable during the flattening process, realizing the intelligent flattening operation of the steel plate, reducing manual intervention and improving the flattening efficiency.

[0031] The convolutional neural network deep learning model is used to train historical data and optimize the position parameters and flattening parameters. The powerful learning ability of the neural network is used to automatically identify and extract the uneven position features based on the point cloud data to obtain the optimal parameters.

[0032] During the flattening process, the 3D visual scanner monitors the state changes of the steel plate surface in real time, feeds the data back to the deep learning model, and automatically adjusts the parameters. This closed-loop control system can make dynamic adjustments during the steel plate flattening process to ensure that the flatness after each flattening meets the requirements, improving the overall operation efficiency and effect.

[0033] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions only describe the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for calculating and flattening the uneven position of a medium-thick plate based on three-dimensional visual reconstruction, characterized in that: The following steps are involved: Step 1: Obtain point cloud data of the steel plate surface through a 3D visual scanner, and pre-process the initial point cloud data to optimize noise while retaining important surface feature information; Step 2: Use the improved fitting algorithm to perform 3D reconstruction on the processed 3D point cloud data, and use the genetic algorithm to dynamically adjust the characteristic parameters of the 3D surface equation to improve the fitting accuracy and convergence speed; Step 3: Calculate the uneven position parameters according to the three-dimensional surface equation, including the bending extreme point, the pad position point, and the bending curvature of the extreme point; Step 4: Based on the position parameters, combined with the bending extreme point and the fulcrum position and their corresponding bending curvature, an intelligent flattening amount calculation model is proposed to automatically calculate the downward pressure; Step 5: Use the deep learning model to train historical data, optimize the position parameters and flattening parameters, form a closed-loop control system, and perform real-time monitoring and feedback.

2. The method of obtaining point cloud data of the steel plate surface using a three-dimensional visual scanner and preprocessing the initial point cloud data according to claim 1 is characterized in that: The steps include: Step 1.1: Use a 3D camera to collect 3D point cloud data of the steel plate surface. To ensure that the point cloud data is in a suitable coordinate system, perform rotation correction on the point cloud. Step 1.2: Perform point cloud segmentation, calculate consistency by using a smooth estimation loss function, set a distance threshold, and segment the surface of the steel plate; Step 1.3: Use adaptive filtering technology to process the segmented point cloud data. First, calculate the distance between each point and the adjacent points, determine the distribution density of the points, and adjust the filtering intensity according to the distribution density to obtain the final steel plate surface point cloud data. .

3. The method of claim 1, wherein the three-dimensional point cloud data is processed using an improved fitting algorithm to perform three-dimensional reconstruction, and the characteristic parameters of the three-dimensional surface equation are dynamically adjusted using a genetic algorithm to improve the fitting accuracy and convergence speed, wherein: The steps include: Step 2.1: Using the improved fitting algorithm, in order to avoid falling into the local optimal solution, it is necessary to first estimate the initial parameters of the fitting function and define the function form as follows: , In the formula, Represents the probability density function value of the point cloud at (x, y), where μ x and μ y are the means of x and y respectively, where ,in are the variances of x and y, σ xy is the covariance of x and y. The genetic algorithm is used to determine the initial parameters of the three-dimensional surface equation. The genetic algorithm selects parameter combinations with higher fitness by simulating the evolutionary process, so that the algorithm can jump out of the local optimum and quickly obtain suitable initial parameters. Step 2.2: Based on the initial parameters, the nonlinear least squares method is used to minimize the residual between the actual data and the three-dimensional surface equation, and a dynamic adjustment strategy is introduced to dynamically adjust the step size according to the change of the error to determine the optimal characteristic parameters: , The steel plate point cloud data obtained after preprocessing in step 1 , using the nonlinear least squares principle, solve the least squares solution of equation (2) above, dynamically adjust the step size and convergence threshold according to the fitting error to ensure that it converges as quickly as possible while meeting the accuracy requirements, and obtain the three-dimensional surface equation .

4. The method of calculating the uneven position parameters according to the three-dimensional surface equation according to claim 1, including the bending extreme point, the pad position point and the bending curvature of the extreme point, is characterized in that: The steps include: Step 3.1: Find the maximum point of the three-dimensional surface equation , as the downward pressure point, calculate the bending curvature K of this point, and the three-dimensional surface equation obtained by fitting , calculate its second-order partial derivative and For the calculation of curvature, the bending curvature K of the lower pressure point is obtained as: , K can directly reflect the flatness information of the steel plate. The size of K reflects the curvature of the steel plate surface. If K is small, it means that the surface is relatively flat; if K is large, it means that the curvature of the surface is large. Step 3.2: Solve the pad position point, pad 1 position point Set as the local lowest point of the three-dimensional surface equation, that is, the minimum point. The two pad position points are roughly symmetrically distributed relative to the downward pressure position point. During the downward pressure process, the two pad position points and the downward pressure position point are required to be in the same plane, and the plane is a vertical plane. According to the position point of pad 1 and the downward pressure position point, the plane equation of the vertical plane can be determined: in , C As a constant term, let the position point of pad 2 be on the three-dimensional surface equation, and the point is also located in the vertical plane mentioned above, then the constraint condition can be established to determine the position point of pad 2 : .

5. According to claim 1, based on the position parameter, an intelligent flattening amount calculation model is proposed to automatically calculate the downward pressure, characterized in that: The steps include: Step 4.1: According to the bending curvature of the extreme point and the specific location of the extreme point, design an intelligent flattening calculation model. The downward pressing position point is the position point of the two pads. and The fulcrum is the point. Based on the material mechanics and the parameters of the flattening equipment, the downward pressure is calculated by calculating the fulcrum spacing, bending moment and deformation. Step 4.2: Apply the calculated position parameters and pressure amount to the flattening equipment and guide the flattening machine to operate accurately at the pressure point. At the same time, combined with the fulcrum position and force analysis, ensure that the steel plate remains stable during the flattening process, realize the intelligent flattening operation of the steel plate, reduce manual intervention, and improve the flattening efficiency.

6. The method of claim 1, wherein the historical data is trained using a deep learning model to form a closed-loop control system for real-time monitoring and feedback, is characterized in that: The steps include: Step 5.1: Introduce the convolutional neural network deep learning model, train the historical data, and optimize the position parameters and flattening parameters. Utilize the powerful learning ability of the neural network to automatically identify and extract the uneven position features based on the point cloud data, and then obtain the optimal parameters; Step 5.2: During the flattening process, the 3D visual scanner is used to monitor the state changes of the steel plate surface in real time, and the data is fed back to the deep learning model to automatically adjust the parameters. This closed-loop control system can make dynamic adjustments during the steel plate flattening process to ensure that the flatness after each flattening meets the requirements and improve the overall operation efficiency and effect.

Images