Intelligent welding method and system for corrugated plate surface of container

Abstract

The present application relates to the technical field of intelligent welding, and discloses a kind of container corrugated plate curved surface intelligent welding method and system, the method comprises: using three-dimensional scanning equipment to carry out full coverage scanning to container corrugated plate, obtain the point cloud data of container corrugated plate and carry out pre-processing to point cloud data;Using the point cloud data after pre-processing carries out surface reconstruction, generates container corrugated plate curved surface three-dimensional model and carries out surface smoothing processing;Optimal welding path is generated using intelligent welding planning algorithm, and the container corrugated plate is welded using optimal welding path;Welding defect identification model is used to carry out welding defect identification to weld joint state data, and welding mode is adjusted based on welding defect identification result feedback.The present application generates optimal welding trajectory adapting to the geometric characteristics of corrugated plate in combination with intelligent path planning, and uses multi-modal sensor to perceive weld joint state in real time during welding process, and dynamically adjusts welding parameters and mode.

Description

Technical Field

[0001] This invention relates to the manufacture of welding equipment such as arc welding machines, and particularly to the field of intelligent welding, specifically an intelligent welding method and system for the curved surface of corrugated container plates. Background Technology

[0002] With the increasing automation and intelligence of the container manufacturing industry, the welding quality and production efficiency of corrugated container sheets have gradually become key factors restricting the overall production line's capacity. Corrugated container sheets, due to their unique structural design with periodic crests and troughs, can provide higher rigidity and strength within a limited thickness. However, this structural characteristic also makes their welding process more technically challenging. The surface of the corrugated sheet is a complex three-dimensional curved structure with frequent and significant local curvature changes. During welding, it is necessary not only to precisely control the three-dimensional movement trajectory of the welding torch but also to ensure the matching degree between the welding torch posture and the normal direction of the corrugated surface to avoid weld misalignment or uneven heat input. Furthermore, the control of the welding thermal cycle is particularly important; excessive heat input may lead to weld deformation and stress concentration, while insufficient heat input may cause quality defects such as incomplete penetration.

[0003] In actual production, even slight deviations can lead to various welding defects such as undercut, incomplete penetration, weld beads, porosity, and cracks. These defects not only reduce the mechanical properties of the weld but also affect the overall sealing and durability of the container, and may even cause safety hazards.

[0004] In existing research, patent CN105945399B proposes a visual recognition-based automatic weld seam tracking method and an intelligent welding robot. The method includes: a lifting device mounted on a mobile cart, a robotic arm mounted on the lifting device, and a welding torch mounted on the robotic arm. The welding torch is equipped with a visual recognition-based automatic weld seam tracker, which includes a vision sensor, a controller, and an actuator connected in sequence. The actuator is connected to the welding torch. A computer control system is used to control the mobile cart, lifting device, robotic arm, and welding torch. This solution enables mobile robot visual recognition and automatic weld seam tracking in arc welding in the heavy machinery industry, improving the level of automation.

[0005] However, this technology is mainly aimed at weld seams with flat surfaces or small curvatures. It relies on a single vision sensor for positioning and lacks the ability to comprehensively perceive the curvature, spatial posture, and welding thermal state of complex three-dimensional curved surfaces (such as corrugated container plates). It is difficult to cope with the real-time identification and dynamic correction of weld seam path changes, occlusions, and various defects, resulting in limited welding accuracy and consistency.

[0006] To address this problem, this invention proposes an intelligent welding method and system for the curved surface of container corrugated sheets, which reduces welding defects and ensures the stability of welding quality. Summary of the Invention

[0007] This invention provides an intelligent welding method and system for the curved surface of a container corrugated sheet. Since noise is easily introduced during the scanning process of the corrugated sheet surface, direct use for path planning will lead to unstable welding trajectories. This invention adopts a curvature-sensitive surface smoothing algorithm to suppress noise while preserving the corrugation characteristics. Furthermore, traditional welding path planning is difficult to balance efficiency and accuracy, especially when dealing with surface undulations and welding torch movement constraints, which can easily cause quality fluctuations. This invention introduces curvature, adaptive speed, and attitude deviation constraints to solve the technical problem of balancing welding efficiency and forming quality, as well as the problem of lagging manual quality inspection and inability to correct deviations in real time, thus realizing closed-loop quality control of the welding process.

[0008] To achieve the above objectives, the present invention provides an intelligent welding method for the curved surface of a container corrugated sheet, comprising the following steps:

[0009] S1: Use a 3D scanning device to perform a full-coverage scan of the container corrugated sheet to obtain point cloud data of the container corrugated sheet, and preprocess the point cloud data.

[0010] S2: Use the preprocessed point cloud data to reconstruct the surface, generate a 3D model of the container corrugated plate surface, and perform surface smoothing on the 3D model of the container corrugated plate surface to obtain a smoothed 3D model of the container corrugated plate surface.

[0011] S3: Based on the smoothed three-dimensional model of the container corrugated plate surface, an intelligent welding planning algorithm is used to generate the optimal welding path, and the container corrugated plate is welded using the optimal welding path.

[0012] S4: Use multimodal sensors to collect weld status data of container corrugated plates in real time during the welding process, use a welding defect identification model to identify welding defects in the weld status data, and adjust the welding method based on the welding defect identification results.

[0013] As a further improvement of the present invention:

[0014] Optionally, a 3D scanning device is used to perform a full-coverage scan of the container corrugated sheet to obtain point cloud data of the container corrugated sheet, including:

[0015] The three-dimensional scanning equipment includes a high-precision three-dimensional laser scanner and a laser rangefinder;

[0016] The point cloud data consists of three-dimensional position coordinates and the signal strength of the three-dimensional position coordinates.

[0017] Optionally, the point cloud data is preprocessed, including:

[0018] The preprocessing includes point cloud registration, noise reduction, and downsampling in sequence.

[0019] The point cloud registration includes coarse registration and fine registration. Coarse registration is to perform preliminary alignment on point cloud data obtained from multiple and multi-view scans so that the point cloud data from each view roughly overlap in the same coordinate system. Fine registration is to further optimize the point cloud after coarse registration, remove redundant point cloud data, and construct the point cloud data set after point cloud registration using the retained point cloud data.

[0020] The denoising process is as follows: after point cloud registration, calculate the number of neighboring points of the point cloud data and the average distance between the point cloud data and the neighboring points. Mark the point cloud data with the number of neighboring points below a preset threshold as noise points, and mark the point cloud data with the average distance above the preset threshold as noise points. Remove the noise points and construct the denoised point cloud data set from the remaining point cloud data.

[0021] The downsampling processing method is the voxel grid method, which is used to downsample the denoised point cloud data set, and the point cloud data obtained by downsampling is used as the preprocessed point cloud data.

[0022] Optionally, the preprocessed point cloud data is used to reconstruct the surface, generating a 3D model of the corrugated container surface, including:

[0023] Calculate the normal vector of the preprocessed point cloud data, and construct a gradient field from the normal vectors of all the preprocessed point cloud data. The normal vector of the preprocessed point cloud data is the sampling vector in the gradient field.

[0024] Spatial interpolation is performed on the gradient field so that the preprocessed point cloud data forms a continuous gradient field on the discrete voxel grid.

[0025] The space where the container corrugated sheet is located is divided into a multi-scale octree grid, where different scale levels correspond to different resolutions, and a discrete Poisson equation is constructed on the multi-scale octree network.

[0026] The Poisson equation is solved iteratively using the finite element method to generate global continuous scalar fields at different scale levels. The global continuous scalar field at the highest resolution is selected as the implicit scalar field obtained by the solution.

[0027] The Marching Cubes algorithm is used to extract isosurfaces from the implicit scalar field, resulting in a closed, continuous, and smooth three-dimensional surface mesh. Mesh vertices and triangular patches are then generated to serve as a three-dimensional model of the corrugated container surface.

[0028] Optionally, the 3D model of the corrugated container surface is smoothed to obtain a smoothed 3D model of the corrugated container surface, including:

[0029] Calculate the local curvature of the mesh vertices in the 3D model of the corrugated surface of a shipping container;

[0030] Based on the local curvature of the mesh vertices, the position coordinates of the mesh vertices in the 3D model of the corrugated container surface are smoothed to obtain a smoothed 3D model of the corrugated container surface.

[0031] Optionally, based on the smoothed 3D model of the container corrugated plate surface, an intelligent welding planning algorithm is used to generate the optimal welding path, including:

[0032] The process of the intelligent welding planning algorithm includes:

[0033] Based on the local curvature of the mesh vertices in the smoothed 3D model of the corrugated container surface, mesh vertices are selected and included in the candidate trajectory point set.

[0034] Candidate trajectory points are selected from the set of candidate trajectory points as welding start points to generate multiple sets of local welding paths;

[0035] A path optimization function for the local welding path is constructed, where the welding control parameters of the local welding path are used as optimization variables, and the local curvature of the coordinate points of the welding path is used as the control parameter. Adaptive dynamic welding control parameters for the local welding path are generated. During the welding process of the local welding path, the welding speed, welding acceleration, and welding direction are adaptively controlled. The functional form of the constructed path optimization function is as follows:

[0036] ;

[0037] ;

[0038] ;

[0039] ;

[0040] ;

[0041] ;

[0042] ;

[0043] in, This represents the path optimization function. This represents the path optimization function value for the local welding path Load.

[0044] Indicates the local welding path The Middle Velocity smoothing constraint term for each welding path coordinate point. Indicates the local welding path The welding speed at the coordinate point of the nth welding path. Indicates the local welding path The welding speed at the (n-1)th welding path coordinate point. Indicates the local welding path The local curvature at the coordinate point of the nth welding path. This indicates the preset maximum speed value. Represents the smoothing coefficient of the velocity term. The modulus of the computation vector is represented by N, where N represents the local welding path. The number of coordinate points along the welding path;

[0045] Set as weighting coefficient The values ​​are 0.2, 0.2, 0.3, and 0.3, respectively.

[0046] Indicates the local welding path The acceleration smoothing constraint term for the nth welding path coordinate point. Indicates the local welding path The welding acceleration at the nth welding path coordinate point. Indicates the local welding path The welding acceleration at the (n-1)th welding path coordinate point. This indicates the preset maximum acceleration value. Indicates the smoothing coefficient of the acceleration term;

[0047] Indicates the local welding path The welding direction constraint term for the nth welding path coordinate point. This represents the direction vector of the welding torch when welding the nth welding path coordinate point. With mesh normal The angular deviation between them, where For local welding path The normal vector of the grid where the nth welding path coordinate point is located. Indicates the deviation smoothing coefficient. Represents the inverse cosine function;

[0048] Indicates the local welding path The welding efficiency loss term at the nth welding path coordinate point. Indicates the local welding path The expected speed of the nth welding path coordinate point. Indicates selection The maximum value in the value, wherein the welding efficiency loss term is used to penalize unnecessary speed limits;

[0049] The path optimization function of the local welding path is optimized and solved to form the optimal welding parameter sequence of the local welding path. The optimal welding parameter sequence of all local welding paths is taken as the optimal welding path. The optimal welding parameter sequence consists of the position coordinates of the welding path coordinate points in the local welding path, the optimal welding speed, the optimal welding acceleration, and the optimal welding torch orientation vector. The optimal welding speed, the optimal welding acceleration, and the optimal welding torch orientation vector are the solution results of the path optimization function.

[0050] Optionally, candidate trajectory points are selected from the candidate trajectory point set as welding start points to generate a local welding path, including:

[0051] From the candidate trajectory point set, select the candidate trajectory point closest to the edge of the container corrugated plate as the welding start point, where the candidate trajectory point is the grid vertex in the candidate trajectory point set;

[0052] Starting from the welding start point, an iterative method is used to generate a sequence of welding path coordinate points with continuous positions, forming a set of local welding paths.

[0053] Optionally, multimodal sensors are used to collect weld condition data of the container corrugated sheet in real time during the welding process, and a welding defect identification model is used to identify welding defects from the weld condition data, including:

[0054] The multimodal sensor includes a laser vision sensor, an infrared thermal imaging sensor, and an acoustic sensor. The multimodal sensor is deployed in front of the welding torch and collects weld status data of the container corrugated plate in real time during the welding process.

[0055] Feature extraction is performed on weld condition data to obtain multimodal weld features;

[0056] A welding defect recognition model based on a deep neural network is used to receive multimodal weld features and output welding defect categories as welding defect recognition results. The welding defect categories include normal weld, undercut, weld bead, incomplete penetration, porosity, and crack.

[0057] Optionally, the welding method can be adjusted based on the feedback from the welding defect identification results, including:

[0058] The adjustment strategy for the welding method is as follows:

[0059] When undercut is detected, reduce the welding speed and welding current;

[0060] When weld beads are detected, reduce the wire feed speed or current and appropriately increase the welding torch movement speed.

[0061] When porosity is identified, maintain a stable welding torch position during welding, use dry welding wire, or clean the base material surface.

[0062] When cracks are identified, reduce the welding speed, weld in sections, or preheat the base material to reduce thermal stress.

[0063] When incomplete penetration is detected, increase the welding current or increase the heat input, and slow down the welding speed.

[0064] This invention also proposes an intelligent welding system for the curved surface of a container corrugated sheet, the intelligent welding system for the curved surface of a container corrugated sheet includes a three-dimensional reconstruction device, a path planning module, and a feedback adjustment device:

[0065] The three-dimensional reconstruction device is used to perform a full-coverage scan of the container corrugated plate using a three-dimensional scanning device to obtain point cloud data of the container corrugated plate. The point cloud data is preprocessed, and the preprocessed point cloud data is used to perform surface reconstruction to generate a three-dimensional model of the container corrugated plate surface. The three-dimensional model of the container corrugated plate surface is then smoothed to obtain a smoothed three-dimensional model of the container corrugated plate surface.

[0066] The path planning module is used to generate the optimal welding path based on the smoothed three-dimensional model of the container corrugated plate surface using an intelligent welding planning algorithm, and then use the optimal welding path to weld the container corrugated plate.

[0067] The feedback adjustment device is used to collect weld status data of container corrugated plates in real time during the welding process using multimodal sensors, identify welding defects in the weld status data using a welding defect identification model, and adjust the welding method based on the welding defect identification results.

[0068] This enables an intelligent welding method for the curved surface of a container corrugated sheet, as described above.

[0069] Compared with existing technologies, this invention proposes an intelligent welding method and system for the curved surface of container corrugated sheets, which has the following beneficial effects:

[0070] First, the path optimization function extracted in this application comprehensively considers the smoothness of the welding trajectory, attitude deviation control, and welding efficiency. It achieves curvature-adaptive global speed optimization by introducing a welding efficiency loss term. Traditional cost functions only constrain the speed, acceleration, and welding torch normal deviation of adjacent points. While this ensures trajectory smoothness and local quality, it cannot actively improve welding efficiency in low-curvature regions, leading to lower welding speeds in flat or low-curvature sections and thus prolonging the overall welding time. The path optimization function applies a secondary penalty to welding speeds below the desired speed, maximizing the welding speed within the allowable quality range and significantly shortening the welding cycle. The desired speed automatically adjusts with local curvature; high-curvature or complex corrugated regions reduce the desired speed to ensure welding quality remains unaffected. Simultaneously, smoothness, acceleration, and attitude constraints ensure trajectory executability and molten pool stability. The welding efficiency loss term is continuously differentiable, facilitating stable convergence of the optimization algorithm and avoiding speed fluctuations caused by boundary switching. Furthermore, this path optimization function allows adaptive adjustment of the welding speed on different corrugated plate surfaces, eliminating the need for manual setting of a global speed benchmark and improving the versatility and automation level of the welding system.

[0071] Meanwhile, this application integrates multimodal sensors, including laser vision, infrared thermal imaging, and acoustic sensing, in front of the welding torch to achieve simultaneous acquisition of weld seam images, heat distribution, and high-frequency acoustic signals, comprehensively reflecting the geometric, thermal, and acoustic characteristics of the welding process. Combined with feature extraction and a deep neural network defect recognition model, it can accurately and in real-time determine weld seam defect types and automatically adjust parameters such as welding speed, current, wire feed speed, and attitude according to different defect types, achieving closed-loop adaptive control of the welding process. Attached Figure Description

[0072] Figure 1 This is a flowchart illustrating an intelligent welding method for the curved surface of a container corrugated sheet, provided in an embodiment of the present invention.

[0073] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0074] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0075] This application provides an intelligent welding method for the curved surface of a container corrugated sheet. The executing entity of this intelligent welding method for the curved surface of a container corrugated sheet includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the intelligent welding method for the curved surface of a container corrugated sheet can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0076] Reference Figure 1 Embodiment 1 of the present invention is as follows:

[0077] A smart welding method for the curved surface of a container corrugated sheet includes the following steps:

[0078] S1: Use a 3D scanning device to perform a full-coverage scan of the container corrugated sheet to obtain point cloud data of the container corrugated sheet, and preprocess the point cloud data.

[0079] A 3D scanning device was used to perform a full-coverage scan of the container corrugated sheet, obtaining point cloud data of the container corrugated sheet, including:

[0080] The three-dimensional scanning equipment includes a high-precision three-dimensional laser scanner and a laser rangefinder;

[0081] The scanning process for the full-coverage scan is as follows:

[0082] The length, width, and corrugation direction of the container corrugated plate are measured using a laser rangefinder. The location distribution of the crest and trough positions of the corrugations is identified. Based on the crest positions, dividing lines are generated along the direction perpendicular to the corrugation direction to divide the container corrugated plate into multiple strip-shaped scanning areas.

[0083] Specifically, a laser line emitted by a laser rangefinder is used to scan along a certain direction of the corrugated plate of the container. If the scanning curve shows obvious periodic fluctuations, then the direction is perpendicular to the corrugation direction, and thus the corrugation direction can be determined.

[0084] A high-precision 3D laser scanner was used to scan the surface height of the container corrugated sheet in the corrugated direction, and the sequence of the surface height of the container corrugated sheet in the corrugated direction was obtained.

[0085] The first derivative is used to process the plate height sequence to identify the peak and trough positions in the plate height sequence. The peak position is when the first derivative of the position changes from positive to negative and the position is a local maximum. The trough position is when the first derivative of the position changes from negative to positive and the position is a local minimum.

[0086] A high-precision 3D laser scanner is used to perform serpentine scanning at different angles (e.g., 0 degrees, -30 degrees, 30 degrees, -35 degrees, 45 degrees, etc.) on a strip-shaped scanning area to acquire point cloud data of the container corrugated sheet. The point cloud data consists of three-dimensional position coordinates and the signal intensity of the three-dimensional position coordinates. The serpentine scanning method involves scanning back and forth along the corrugation direction, covering the entire strip-shaped scanning area line by line. Each time a line is changed, the high-precision 3D laser scanner moves along a path perpendicular to the scanning direction to ensure that the inter-line overlap rate reaches 20% to 30%. As an embodiment of this application, the posture (position, rotation angle) and timestamp of the high-precision 3D laser scanner are recorded in real time during the scanning process to ensure that the point clouds from multiple perspectives can be registered subsequently.

[0087] As an embodiment of this application, the crest region can be scanned quickly (small curvature, simple geometric information), while the trough region is scanned at a slower speed (large curvature, rapid geometric changes, requiring higher sampling density), thereby improving the overall scanning efficiency. The crest region is a 10 cm wide container corrugated plate area generated along the direction perpendicular to the corrugation direction at the crest position, and the trough region is a 10 cm wide container corrugated plate area generated along the direction perpendicular to the corrugation direction at the trough position. It should be noted that in the container corrugated plate structure, in order to ensure strength and ease of forming, the crest region has a certain flattened transition section rather than a sharp vertex, so the local radius of curvature is relatively large (the curvature value is relatively small), while the trough region is designed as a deeper depression to enhance rigidity, which makes the bottom radius of the trough smaller (the curvature value is relatively large).

[0088] Specifically, this application first uses a laser rangefinder to accurately measure the length, width, and corrugation direction of the container corrugated plate. Combined with the spatial distribution of the peaks and troughs, a strip-shaped area is divided, effectively avoiding repeated acquisition and data redundancy caused by blind scanning of a large area. The scanning task is limited to the strip-shaped scanning area by dividing lines, and a multi-view serpentine scanning strategy is adopted, which not only ensures full coverage and high overlap of the scanning data, but also improves the ability to capture complex surface geometric information. The peak areas with small curvature and simple geometric information are scanned quickly to reduce invalid sampling time. The trough areas with large curvature and rapid geometric changes are scanned at a slower speed to improve sampling density and data accuracy. Thus, while ensuring the quality of the point cloud, the total scanning time and data processing burden are significantly reduced.

[0089] Preprocessing the point cloud data includes:

[0090] The preprocessing includes point cloud registration, noise reduction, and downsampling in sequence.

[0091] The point cloud registration includes coarse registration and fine registration. Coarse registration involves initially aligning point cloud data obtained from multiple scans and multiple viewpoints, so that the point cloud data from each viewpoint roughly overlaps in the same coordinate system. Fine registration further optimizes the point cloud after coarse registration, minimizing the overall point cloud fusion error. It also involves fusing the same point cloud data from each viewpoint, removing redundant point cloud data, and constructing the registered point cloud data set from the retained point cloud data. As an embodiment of this application, the SAC-IA algorithm is used for coarse registration, and the ICP algorithm is used for fine registration.

[0092] The denoising process is as follows: After point cloud registration, calculate the number of neighboring points of the point cloud data and the average distance between the point cloud data and its neighboring points. Mark the point cloud data with the number of neighboring points below a preset threshold as noise points, and mark the point cloud data with the average distance above the preset threshold as noise points. Remove the noise points and construct the denoised point cloud data set from the remaining point cloud data. For any point cloud data c, the neighboring points of point cloud data c are the point cloud data in the point cloud data set after point cloud registration that are within a preset radius (e.g., 10 cm) from point cloud data c. The distance between point cloud data is the Euclidean distance between the three-dimensional position coordinates in the point cloud data.

[0093] The downsampling processing method is the voxel grid method, which is used to downsample the denoised point cloud data set, and the point cloud data obtained by downsampling is used as the preprocessed point cloud data.

[0094] Specifically, the voxelization mesh method involves dividing the spatial location of the container corrugated plate into a fixed-size cubic grid, mapping the point cloud data in the denoised point cloud data set to the corresponding cubic grid according to the three-dimensional position coordinates, calculating the centroid of the three-dimensional position coordinates and the mean signal intensity of all point cloud data in the cubic grid, and using the centroid of the three-dimensional position coordinates and the mean signal intensity as the point cloud data obtained by downsampling the point cloud data in the cubic grid.

[0095] S2: Use the preprocessed point cloud data to reconstruct the surface, generate a 3D model of the container corrugated plate surface, and perform surface smoothing on the 3D model of the container corrugated plate surface to obtain a smoothed 3D model of the container corrugated plate surface.

[0096] Using preprocessed point cloud data, surface reconstruction is performed to generate a 3D model of the corrugated container surface, including:

[0097] Calculate the normal vector of the preprocessed point cloud data, and construct a gradient field from the normal vectors of all the preprocessed point cloud data. The normal vector of the preprocessed point cloud data is the sampling vector in the gradient field.

[0098] Spatial interpolation is performed on the gradient field so that the preprocessed point cloud data forms a continuous gradient field on the discrete voxel grid.

[0099] The space where the container corrugated sheet is located is divided into a multi-scale octree grid, where different scale levels correspond to different resolutions, and a discrete Poisson equation is constructed on the multi-scale octree network.

[0100] The Poisson equation is solved iteratively using the finite element method to generate global continuous scalar fields at different scale levels. The global continuous scalar field at the highest resolution is selected as the implicit scalar field obtained by the solution.

[0101] Specifically, a global continuous scalar field is generated first at a low resolution, and then the resolution is gradually increased to optimize local details, thereby improving computational efficiency and preserving smoothness.

[0102] The Marching Cubes algorithm is used to extract isosurfaces from the implicit scalar field, resulting in a closed, continuous, and smooth three-dimensional surface mesh. Mesh vertices and triangular patches are then generated to serve as a three-dimensional model of the corrugated container surface.

[0103] As an embodiment of this application, the three-dimensional surface mesh is composed of scalar values ​​of the surface position in the implicit scalar field. The three-dimensional surface mesh is composed of multiple voxel meshes. Each voxel mesh has 8 vertices, and each vertex has its scalar value in the implicit scalar field. For each voxel mesh, if the scalar values ​​of the 8 vertices have a sign change (crossing 0), then the 8 vertices of the voxel mesh are taken as mesh vertices.

[0104] The 3D model of the corrugated container surface is smoothed to obtain the smoothed 3D model of the corrugated container surface, including:

[0105] Calculate the local curvature of the mesh vertices in the 3D model of the corrugated container surface; the calculation process for the local curvature is as follows:

[0106] Obtain the position coordinates of the neighboring vertices of the mesh vertex node whose local curvature is to be calculated, where the neighboring vertices of the mesh vertex node are the mesh vertices directly connected to the mesh vertex node in the 3D model of the corrugated container surface.

[0107] Calculate the diagonal angle between the two triangular faces containing the mesh vertex node and its neighboring vertex, wherein the mesh vertex node is directly connected to the neighboring vertex and both are located on the two triangular faces.

[0108] The local curvature of the mesh vertex node is estimated using the following formula:

[0109] ;

[0110] in, This represents 1 / 3 of the area of ​​the triangular facet connected to the mesh vertex node. This represents the set of neighborhood vertices of a grid vertex node. Represents the set of vertices in the domain Any vertex in the domain, Let be the diagonal angle between the two triangular faces containing the grid vertex node and its neighboring vertex e. Represents the cotangent function. This represents the position coordinates of vertex e in the domain. This represents the position coordinates of a grid vertex node. This indicates the magnitude of the calculated vector;

[0111] It should be noted that this curvature estimation method is based on the discrete geometry principle of neighboring vertices. By constructing local patches and calculating the geometric differences between vertices and their neighbors, it can perform high-precision approximate curvature estimation of mesh surfaces with arbitrary topological structures without relying on the equations of continuous surfaces. Furthermore, the introduction of vertex-related local area weights during the calculation process effectively suppresses the interference of irregular mesh distribution on curvature estimation, improving the stability and accuracy of the calculation results. Compared with traditional curvature fitting methods, this method has lower computational cost and is suitable for the rapid processing of large-scale point clouds or meshes. Simultaneously, because it directly utilizes local geometric information, it can balance low-noise response in smooth regions and sensitive capture of high-curvature regions.

[0112] Based on the local curvature of the mesh vertices, the position coordinates of the mesh vertices in the 3D model of the corrugated container surface are smoothed to obtain a smoothed 3D model of the corrugated container surface.

[0113] Specifically, the smoothness correction formula for the position coordinates of the mesh vertex node is:

[0114] ;

[0115] in, This represents the smoothing correction result for the position coordinates of the mesh vertex nodes. Indicates the smoothing step size, set It is 0.1. Indicates sensitive control parameters, settings It is 0.3. This represents the local curvature of vertex e in the neighborhood of a mesh vertex node. This represents the local curvature difference of a vertex *e* in the neighborhood of a grid vertex *node*. The smoothing weight of vertex e in the neighborhood is represented;

[0116] When the local curvature difference is large, the smoothing weight is reduced to decrease the pulling across features, thereby effectively protecting key geometric features such as peaks and troughs; when the local curvature difference is small, the weight is close to 1, making the smoothing effect more uniform and helping to eliminate noise.

[0117] S3: Based on the smoothed three-dimensional model of the container corrugated plate surface, an intelligent welding planning algorithm is used to generate the optimal welding path, and the container corrugated plate is welded using the optimal welding path.

[0118] Based on the smoothed 3D model of the corrugated container surface, an intelligent welding planning algorithm is used to generate the optimal welding path, including:

[0119] The process of the intelligent welding planning algorithm includes:

[0120] Based on the local curvature of the mesh vertices in the smoothed 3D model of the corrugated container surface, mesh vertices are selected and included in the candidate trajectory point set.

[0121] Specifically, grid vertices with local curvature exceeding a preset maximum curvature threshold, as well as those located in peak or trough regions, are included in the candidate trajectory point set.

[0122] Candidate trajectory points are selected from the set of candidate trajectory points as welding start points to generate multiple sets of local welding paths;

[0123] A path optimization function for the local welding path is constructed, where the welding control parameters of the local welding path are used as optimization variables, and the local curvature of the coordinate points of the welding path is used as the control parameter. Adaptive dynamic welding control parameters for the local welding path are generated. During the welding process of the local welding path, the welding speed, welding acceleration, and welding direction are adaptively controlled. The functional form of the constructed path optimization function is as follows:

[0124] ;

[0125] ;

[0126] ;

[0127] ;

[0128] ;

[0129] ;

[0130] ;

[0131] in, This represents the path optimization function. This represents the path optimization function value for the local welding path Load.

[0132] Indicates the local welding path The velocity smoothing constraint term for the nth welding path coordinate point. Indicates the local welding path The welding speed at the coordinate point of the nth welding path. Indicates the local welding path The welding speed at the (n-1)th welding path coordinate point. Indicates the local welding path The local curvature at the coordinate point of the nth welding path. This indicates the preset maximum speed value (e.g., 1 meter per second). This represents the smoothing coefficient for the velocity term (e.g., 0.3). The modulus of the computation vector is represented by N, where N represents the local welding path. The number of coordinate points along the welding path;

[0133] Set as weighting coefficient The values ​​are 0.2, 0.2, 0.3, and 0.3, respectively.

[0134] Indicates the local welding path The acceleration smoothing constraint term for the nth welding path coordinate point. Indicates the local welding path The welding acceleration at the nth welding path coordinate point. Indicates the local welding path The welding acceleration at the (n-1)th welding path coordinate point. This indicates the preset maximum acceleration value (e.g., 0.5 meters per second squared). This represents the smoothing coefficient for the acceleration term (e.g., 0.2).

[0135] Indicates the local welding path The welding direction constraint term for the nth welding path coordinate point. This represents the direction vector of the welding torch when welding the nth welding path coordinate point. With mesh normal The angular deviation between them, where For local welding path The normal vector of the grid where the nth welding path coordinate point is located. This represents the deviation smoothing coefficient (e.g., 0.2). The inverse cosine function is represented; the normal vector of the mesh is the mean of the normal vectors of each face of the mesh.

[0136] Indicates the local welding path The welding efficiency loss term at the nth welding path coordinate point. Indicates the local welding path The expected speed of the nth welding path coordinate point. Indicates selection The maximum value in the value, wherein the welding efficiency loss term is used to penalize unnecessary speed limits;

[0137] The path optimization function of the local welding path is optimized and solved to form the optimal welding parameter sequence of the local welding path. The optimal welding parameter sequence of all local welding paths is taken as the optimal welding path. The optimal welding parameter sequence consists of the position coordinates of the welding path coordinate points in the local welding path, the optimal welding speed, the optimal welding acceleration, and the optimal welding torch orientation vector. The optimal welding speed, the optimal welding acceleration, and the optimal welding torch orientation vector are the solution results of the path optimization function.

[0138] As an embodiment of this application, the optimization solution of the path optimization function is an improved genetic algorithm. The improvement strategy of the genetic algorithm is as follows: increase the mutation probability of the low fitness part of the local welding path to explore more welding parameters; increase the crossover probability of the high fitness part of the path to enhance the local optimization effect. The low fitness segment may have speed, acceleration or direction deviation from the optimal, and increasing mutation can help escape the local bad solution. The high fitness segment maintains stable crossover to speed up the convergence speed and avoid disturbing the welding quality. The welding parameters of any welding path coordinate point in the current optimal welding parameter sequence are perturbed, the path optimization function value is re-evaluated, and the improved sequence is retained.

[0139] Candidate trajectory points are selected from the candidate trajectory point set as welding start points to generate local welding paths, including:

[0140] From the candidate trajectory point set, select the candidate trajectory point closest to the edge of the container corrugated plate as the welding start point, where the candidate trajectory point is the grid vertex in the candidate trajectory point set;

[0141] Starting from the welding start point, an iterative method is used to generate a sequence of continuously positioned welding path coordinate points, forming a set of local welding paths. The iterative formula is as follows:

[0142] ;

[0143] ;

[0144] in, This represents the coordinates of the m-th welding path obtained through iteration. This represents the coordinates of the (m+1)th welding path point obtained through iteration; specifically, when m=0, This is the selected welding start point;

[0145] This represents the iteration step size for iterating over the coordinates of the m-th welding path point. The direction vector representing the direction of the ripples. Indicates the basic step size. This represents the minimum threshold for local curvature (e.g., 0.05). ), Represents the coordinates of the welding path. Local curvature;

[0146] If the welding path coordinates Local curvature change rate Higher than the expected minimum rate of change threshold (e.g., 0.1) If the current local welding path is not found, the iteration will terminate, the current local welding path will be output, and a new welding starting point will be selected.

[0147] S4: Use multimodal sensors to collect weld status data of container corrugated plates in real time during the welding process, use a welding defect identification model to identify welding defects in the weld status data, and adjust the welding method based on the welding defect identification results.

[0148] Multimodal sensors are used to collect real-time weld condition data of corrugated container plates during the welding process. A welding defect identification model is then used to identify welding defects from this data, including:

[0149] The multimodal sensor includes a laser vision sensor, an infrared thermal imaging sensor, and an acoustic sensor. The multimodal sensor is deployed in front of the welding torch and collects the weld status data of the container corrugated plate in real time during the welding process. The laser vision sensor is used to acquire weld images, the infrared thermal imaging sensor is used to acquire weld heat distribution, and the acoustic sensor is used to acquire high-frequency acoustic signals during the welding process. The weld images, weld heat distribution, and high-frequency acoustic signals are used as weld status data.

[0150] Feature extraction is performed on weld condition data to obtain multimodal weld features;

[0151] As an embodiment of this application, the multimodal weld features include weld contour features, heat distribution uniformity index and frequency spectrum features of high-frequency acoustic signals. The weld contour features are extracted using convolutional filtering for noise reduction and the Canny edge detection algorithm. The absolute value of the heat distribution difference on both sides of the welding torch is calculated as the heat distribution uniformity index.

[0152] A welding defect recognition model based on a deep neural network is used to receive multimodal weld features and output welding defect categories as welding defect recognition results. The welding defect categories include normal weld, undercut, weld bead, incomplete penetration, porosity, and crack.

[0153] Welding methods are adjusted based on feedback from welding defect identification results, including:

[0154] The adjustment strategy for the welding method is as follows:

[0155] When undercut is detected, reduce the welding speed and welding current;

[0156] When weld beads are detected, reduce the wire feed speed or current and appropriately increase the welding torch movement speed.

[0157] When porosity is identified, maintain a stable welding torch position during welding, use dry welding wire, or clean the base material surface.

[0158] When cracks are identified, reduce the welding speed, weld in sections, or preheat the base material to reduce thermal stress.

[0159] When incomplete penetration is detected, increase the welding current or increase the heat input, and slow down the welding speed.

[0160] Example 2:

[0161] A smart welding system for the curved surface of a container corrugated sheet includes a three-dimensional reconstruction device, a path planning module, and a feedback adjustment device.

[0162] The three-dimensional reconstruction device is used to perform a full-coverage scan of the container corrugated plate using a three-dimensional scanning device to obtain point cloud data of the container corrugated plate. The point cloud data is preprocessed, and the preprocessed point cloud data is used to perform surface reconstruction to generate a three-dimensional model of the container corrugated plate surface. The three-dimensional model of the container corrugated plate surface is then smoothed to obtain a smoothed three-dimensional model of the container corrugated plate surface.

[0163] The path planning module is used to generate the optimal welding path based on the smoothed three-dimensional model of the container corrugated plate surface using an intelligent welding planning algorithm, and then use the optimal welding path to weld the container corrugated plate.

[0164] The feedback adjustment device is used to collect weld status data of container corrugated plates in real time during the welding process using multimodal sensors, identify welding defects in the weld status data using a welding defect identification model, and adjust the welding method based on the welding defect identification results.

[0165] This enables an intelligent welding method for the curved surface of a container corrugated sheet, as described above.

[0166] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0167] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0168] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0169] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A smart welding method for the curved surface of a container corrugated sheet, characterized in that, The method includes: S1: Use a 3D scanning device to perform a full-coverage scan of the container corrugated sheet to obtain point cloud data of the container corrugated sheet, and preprocess the point cloud data. S2: Use the preprocessed point cloud data to reconstruct the surface, generate a 3D model of the container corrugated plate surface, and perform surface smoothing on the 3D model of the container corrugated plate surface to obtain a smoothed 3D model of the container corrugated plate surface. S3: Based on the smoothed three-dimensional model of the container corrugated plate surface, an intelligent welding planning algorithm is used to generate the optimal welding path, and the container corrugated plate is welded using the optimal welding path. S4: Use multimodal sensors to collect weld status data of container corrugated plates in real time during the welding process, use a welding defect identification model to identify welding defects in the weld status data, and adjust the welding method based on the welding defect identification results. Using preprocessed point cloud data, surface reconstruction is performed to generate a 3D model of the corrugated container surface, including: Calculate the normal vector of the preprocessed point cloud data, and construct a gradient field from the normal vectors of all the preprocessed point cloud data. The normal vector of the preprocessed point cloud data is the sampling vector in the gradient field. Spatial interpolation is performed on the gradient field so that the preprocessed point cloud data forms a continuous gradient field on the discrete voxel grid. The space where the container corrugated sheet is located is divided into a multi-scale octree grid, where different scale levels correspond to different resolutions, and a discrete Poisson equation is constructed on the multi-scale octree network. The Poisson equation is solved iteratively using the finite element method to generate global continuous scalar fields at different scale levels. The global continuous scalar field at the highest resolution is selected as the implicit scalar field obtained by the solution. The Marching Cubes algorithm is used to extract isosurfaces from the implicit scalar field, resulting in a closed, continuous, and smooth three-dimensional surface mesh. Mesh vertices and triangular patches are then generated to serve as a three-dimensional model of the corrugated container surface.

2. The intelligent welding method for the curved surface of a container corrugated sheet as described in claim 1, characterized in that, A 3D scanning device was used to perform a full-coverage scan of the container corrugated sheet, obtaining point cloud data of the container corrugated sheet, including: The three-dimensional scanning equipment includes a high-precision three-dimensional laser scanner and a laser rangefinder; The point cloud data consists of three-dimensional position coordinates and the signal strength of the three-dimensional position coordinates.

3. The intelligent welding method for the curved surface of a container corrugated sheet as described in claim 2, characterized in that, Preprocessing the point cloud data includes: The preprocessing includes point cloud registration, noise reduction, and downsampling in sequence. The point cloud registration includes coarse registration and fine registration. Coarse registration is to perform preliminary alignment on point cloud data obtained from multiple and multi-view scans so that the point cloud data from each view roughly overlap in the same coordinate system. Fine registration is to further optimize the point cloud after coarse registration, remove redundant point cloud data, and construct the point cloud data set after point cloud registration using the retained point cloud data. The denoising process is as follows: after point cloud registration, calculate the number of neighboring points of the point cloud data and the average distance between the point cloud data and the neighboring points. Mark the point cloud data with the number of neighboring points below a preset threshold as noise points, and mark the point cloud data with the average distance above the preset threshold as noise points. Remove the noise points and construct the denoised point cloud data set from the remaining point cloud data. The downsampling processing method is the voxel grid method, which is used to downsample the denoised point cloud data set, and the point cloud data obtained by downsampling is used as the preprocessed point cloud data.

4. The intelligent welding method for the curved surface of a container corrugated sheet as described in claim 1, characterized in that, The 3D model of the corrugated container surface is smoothed to obtain the smoothed 3D model of the corrugated container surface, including: Calculate the local curvature of the mesh vertices in the 3D model of the corrugated surface of a shipping container; Based on the local curvature of the mesh vertices, the position coordinates of the mesh vertices in the 3D model of the corrugated container surface are smoothed to obtain a smoothed 3D model of the corrugated container surface.

5. The intelligent welding method for the curved surface of a container corrugated sheet as described in claim 4, characterized in that, Based on the smoothed 3D model of the corrugated container surface, an intelligent welding planning algorithm is used to generate the optimal welding path, including: The process of the intelligent welding planning algorithm includes: Based on the local curvature of the mesh vertices in the smoothed 3D model of the corrugated container surface, mesh vertices are selected and included in the candidate trajectory point set. Candidate trajectory points are selected from the set of candidate trajectory points as welding start points to generate multiple sets of local welding paths; A path optimization function for the local welding path is constructed, where the welding control parameters of the local welding path are used as optimization variables, and the local curvature of the coordinate points of the welding path is used as the control parameter. Adaptive dynamic welding control parameters for the local welding path are generated. During the welding process of the local welding path, the welding speed, welding acceleration, and welding direction are adaptively controlled. The functional form of the constructed path optimization function is as follows: ； ； ； ； ； ； ； in, This represents the path optimization function. This represents the path optimization function value for the local welding path Load. Indicates the local welding path The Middle Velocity smoothing constraint term for each welding path coordinate point. Indicates the local welding path The welding speed at the coordinate point of the nth welding path. Indicates the local welding path The welding speed at the (n-1)th welding path coordinate point. Indicates the local welding path The local curvature at the coordinate point of the nth welding path. This indicates the preset maximum speed value. Represents the smoothing coefficient of the velocity term. The modulus of the computation vector is represented by N, where N represents the local welding path. The number of coordinate points along the welding path; Set as weighting coefficient The values ​​are 0.2, 0.2, 0.3, and 0.3, respectively. Indicates the local welding path The acceleration smoothing constraint term for the nth welding path coordinate point. Indicates the local welding path The welding acceleration at the nth welding path coordinate point. Indicates the local welding path The welding acceleration at the (n-1)th welding path coordinate point. This indicates the preset maximum acceleration value. Indicates the smoothing coefficient of the acceleration term; Indicates the local welding path The welding direction constraint term for the nth welding path coordinate point. This represents the direction vector of the welding torch when welding the nth welding path coordinate point. With mesh normal The angular deviation between them, where For local welding paths The normal vector of the grid where the nth welding path coordinate point is located. Indicates the deviation smoothing coefficient. Represents the inverse cosine function; Indicates the local welding path The welding efficiency loss term at the nth welding path coordinate point. Indicates the local welding path The expected speed of the nth welding path coordinate point. Indicates selection The maximum value in the value, wherein the welding efficiency loss term is used to penalize unnecessary speed limits; The path optimization function of the local welding path is optimized and solved to form the optimal welding parameter sequence of the local welding path. The optimal welding parameter sequence of all local welding paths is taken as the optimal welding path. The optimal welding parameter sequence consists of the position coordinates of the welding path coordinate points in the local welding path, the optimal welding speed, the optimal welding acceleration, and the optimal welding torch orientation vector. The optimal welding speed, the optimal welding acceleration, and the optimal welding torch orientation vector are the solution results of the path optimization function.

6. The intelligent welding method for the curved surface of a container corrugated sheet as described in claim 5, characterized in that, Candidate trajectory points are selected from the candidate trajectory point set as welding start points to generate local welding paths, including: From the candidate trajectory point set, select the candidate trajectory point closest to the edge of the container corrugated plate as the welding start point, where the candidate trajectory point is the grid vertex in the candidate trajectory point set; Starting from the welding start point, an iterative method is used to generate a sequence of welding path coordinate points with continuous positions, forming a set of local welding paths.

7. The intelligent welding method for the curved surface of a container corrugated sheet as described in claim 1, characterized in that, Multimodal sensors are used to collect real-time weld condition data of corrugated container plates during the welding process. A welding defect identification model is then used to identify welding defects from this data, including: The multimodal sensor includes a laser vision sensor, an infrared thermal imaging sensor, and an acoustic sensor. The multimodal sensor is deployed in front of the welding torch and collects weld status data of the container corrugated plate in real time during the welding process. Feature extraction is performed on weld condition data to obtain multimodal weld features; A welding defect recognition model based on a deep neural network is used to receive multimodal weld features and output welding defect categories as welding defect recognition results. The welding defect categories include normal weld, undercut, weld bead, incomplete penetration, porosity, and crack.

8. The intelligent welding method for the curved surface of a container corrugated sheet as described in claim 7, characterized in that, Welding methods are adjusted based on feedback from welding defect identification results, including: The adjustment strategy for the welding method is as follows: When undercut is detected, reduce the welding speed and welding current; When weld beads are detected, reduce the wire feed speed or current and appropriately increase the welding torch movement speed. When porosity is identified, maintain a stable welding torch position during welding, use dry welding wire, or clean the base material surface. When cracks are identified, reduce the welding speed, weld in sections, or preheat the base material to reduce thermal stress. When incomplete penetration is detected, increase the welding current or increase the heat input, and slow down the welding speed.

9. An intelligent welding system for the curved surface of a container corrugated sheet, characterized in that, The intelligent welding system for the curved surface of the container corrugated sheet includes a three-dimensional reconstruction device, a path planning module, and a feedback adjustment device. The three-dimensional reconstruction device is used to perform a full-coverage scan of the container corrugated plate using a three-dimensional scanning device to obtain point cloud data of the container corrugated plate. The point cloud data is preprocessed, and the preprocessed point cloud data is used to perform surface reconstruction to generate a three-dimensional model of the container corrugated plate surface. The three-dimensional model of the container corrugated plate surface is then smoothed to obtain a smoothed three-dimensional model of the container corrugated plate surface. The path planning module is used to generate the optimal welding path based on the smoothed three-dimensional model of the container corrugated plate surface using an intelligent welding planning algorithm, and then use the optimal welding path to weld the container corrugated plate. The feedback adjustment device is used to collect weld status data of container corrugated plates in real time during the welding process using multimodal sensors, identify welding defects in the weld status data using a welding defect identification model, and adjust the welding method based on the welding defect identification results. To achieve the intelligent welding method for the curved surface of a container corrugated sheet as described in any one of claims 1-8.

Images