A method and system for laser welding

Abstract

This invention provides an LSW laser welding method and system, relating to the field of laser welding technology. The method includes acquiring three-dimensional information of a large structural component and information of multiple welding robots; using a stress analysis model based on the three-dimensional information of the large structural component to determine multiple high-stress regions and multiple low-stress regions; constructing a collaborative task map based on the multiple segmented sub-paths; processing the collaborative task map based on a graph neural network to determine preliminary welding information for one or more welding sub-paths corresponding to each welding robot; and controlling multiple welding robots to perform LSW laser welding based on the target welding information of one or more welding sub-paths corresponding to each welding robot. This method can efficiently and accurately determine a multi-robot LSW welding information adaptation scheme for large structural components.

Description

Technical Field

[0001] This invention relates to the field of laser welding technology, and specifically to an LSW laser welding method and system. Background Technology

[0002] Laser scanning welding (LSW), with its high precision and efficiency, has become a key technology in the manufacturing of large structural components. However, existing multi-robot LSW welding methods still face many prominent problems in practical applications. Large structural components have complex geometries and uneven inherent stress distributions. Traditional welding path planning does not fully consider stress distribution characteristics, easily leading to defects such as cracks and deformation in high-stress areas, seriously affecting the integrity and safety of the structural components. In multi-robot collaborative operations, the lack of a scientific overall welding path segmentation and task allocation mechanism makes it difficult to balance robot load balance and spatial coordination, often resulting in low work efficiency and motion interference. Furthermore, welding parameters are mostly preset fixed values, failing to consider the impact of dynamic factors such as thermal deformation and changes in working conditions during the actual welding process. Adjusting welding parameters relies on human experience, lacking adaptive optimization methods based on actual welding data, leading to poor weld quality consistency and difficulty in meeting the high-precision manufacturing requirements of large structural components. These problems collectively restrict the efficient application of multi-robot LSW welding technology in the manufacturing of large structural components.

[0003] Therefore, how to efficiently and accurately determine the adaptation scheme for multi-robot LSW welding information of large structural components is an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem addressed by this invention is how to efficiently and accurately determine the multi-robot LSW welding information adaptation scheme for large structural components.

[0005] According to a first aspect, the present invention provides an LSW laser welding method, comprising: acquiring three-dimensional information of a large structural component and information of multiple welding robots; determining multiple high-stress regions and multiple low-stress regions based on the three-dimensional information of the large structural component using a stress analysis model; determining multiple segmented sub-path information based on the three-dimensional information of the large structural component, the multiple high-stress regions, the multiple low-stress regions, and the information of the multiple welding robots; constructing a collaborative task graph based on the multiple segmented sub-paths, the collaborative task graph including multiple task nodes and edges between task nodes, each task node corresponding to a segmented sub-path, the node features of the task node including segmented sub-path information and multiple welding robot information, and the edges between nodes representing the spatial positional relationship between the segmented sub-paths; processing the collaborative task graph based on a graph neural network to determine preliminary welding information of one or more welding sub-paths corresponding to each welding robot; determining target welding information of one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot; and controlling multiple welding robots to perform LSW laser welding based on the target welding information of one or more welding sub-paths corresponding to each welding robot.

[0006] In one possible implementation, determining the target welding information for one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot includes: controlling multiple welding robots to perform LSW laser welding on a large structural component based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot, and acquiring multiple welding point information on each welding sub-path after welding is completed; clustering the multiple welding point information on each welding sub-path to obtain multiple welding information groups on each welding sub-path; determining the adjustment welding point information for each welding sub-path based on the multiple welding information groups on each welding sub-path; and determining the target welding information for one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information and the adjustment welding point information of each welding sub-path.

[0007] In one possible implementation, determining multiple segmented sub-path information based on the three-dimensional information of the large structural component, the multiple high-stress areas, the multiple low-stress areas, and the information of the multiple welding robots includes: determining multiple mandatory welding points and multiple welding avoidance areas based on the three-dimensional information of the large structural component, the multiple high-stress areas, and the multiple low-stress areas; determining overall welding planning path information based on the three-dimensional information of the large structural component, the multiple mandatory welding points, and the multiple welding avoidance areas; and determining multiple segmented sub-path information based on the overall welding planning path information and the information of the multiple welding robots.

[0008] In one possible implementation, the stress analysis model is a deep neural network model.

[0009] According to a second aspect, the present invention provides an LSW laser welding system, comprising: an acquisition module for acquiring three-dimensional information of a large structural component and information of multiple welding robots; a stress analysis module for determining multiple high-stress regions and multiple low-stress regions based on the three-dimensional information of the large structural component using a stress analysis model; a sub-path determination module for determining multiple segmented sub-paths based on the three-dimensional information of the large structural component, the multiple high-stress regions, the multiple low-stress regions, and the multiple welding robot information; and a graph construction module for constructing a collaborative task graph based on the multiple segmented sub-paths, wherein the collaborative task graph includes multiple task nodes and edges between task nodes, and each task node corresponds to a segmented sub-path. The task node features include segmented sub-path information and multiple welding robot information, with edges between nodes representing the spatial positional relationship between the segmented sub-paths; a preliminary welding information determination module is used to process the collaborative task graph based on a graph neural network to determine the preliminary welding information of one or more welding sub-paths corresponding to each welding robot; a target welding information determination module is used to determine the target welding information of one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot; and a welding control module is used to control multiple welding robots to perform LSW laser welding based on the target welding information of one or more welding sub-paths corresponding to each welding robot.

[0010] In one possible implementation, the target welding information determination module is further configured to: control multiple welding robots to perform LSW laser welding on a large structural component based on preliminary welding information of one or more welding sub-paths corresponding to each welding robot, and acquire multiple welding point information on each welding sub-path after welding is completed; cluster the multiple welding point information on each welding sub-path to obtain multiple welding information groups on each welding sub-path; determine the adjustment welding point information of each welding sub-path based on the multiple welding information groups on each welding sub-path; and determine the target welding information of one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information and the adjustment welding point information of each welding sub-path.

[0011] In one possible implementation, the sub-path determination module is further configured to: determine multiple welding necessary points and multiple welding avoidance areas based on the three-dimensional information of the large structural component, the multiple high-stress areas, and the multiple low-stress areas; determine overall welding planning path information based on the three-dimensional information of the large structural component, the multiple welding necessary points, and the multiple welding avoidance areas; and determine multiple segmented sub-path information based on the overall welding planning path information and the information of the multiple welding robots.

[0012] In one possible implementation, the stress analysis model is a deep neural network model.

[0013] According to a third aspect, embodiments of the present invention provide an electronic device, including: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method described above, the method including: acquiring three-dimensional information of a large structural component and information of multiple welding robots; determining multiple high-stress regions and multiple low-stress regions based on the three-dimensional information of the large structural component using a stress analysis model; determining multiple segmented sub-path information based on the three-dimensional information of the large structural component, the multiple high-stress regions, the multiple low-stress regions, and the multiple welding robot information; and constructing a collaborative task graph based on the multiple segmented sub-paths, the collaborative task graph... The system includes multiple task nodes and edges between them. Each task node corresponds to a segmented sub-path. The node features of each task node include segmented sub-path information and information about multiple welding robots. The edges between nodes represent the spatial positional relationship between the segmented sub-paths. The collaborative task graph is processed based on a graph neural network to determine the preliminary welding information of one or more welding sub-paths corresponding to each welding robot. Based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot, the target welding information of one or more welding sub-paths corresponding to each welding robot is determined. Based on the target welding information of one or more welding sub-paths corresponding to each welding robot, multiple welding robots are controlled to perform LSW laser welding.

[0014] According to a fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the aforementioned LSW laser welding method. The method includes: acquiring three-dimensional information of a large structural component and information of multiple welding robots; determining multiple high-stress regions and multiple low-stress regions based on the three-dimensional information of the large structural component using a stress analysis model; determining multiple segmented sub-path information based on the three-dimensional information of the large structural component, the multiple high-stress regions, the multiple low-stress regions, and the multiple welding robot information; and constructing a collaborative task graph based on the multiple segmented sub-paths, the collaborative task graph including multiple task nodes and tasks. The edges between task nodes represent the spatial relationships between them. Each task node corresponds to a segmented sub-path. The node features of a task node include segmented sub-path information and information about multiple welding robots. The edges between nodes represent the spatial relationships between the segmented sub-paths. The collaborative task graph is processed based on a graph neural network to determine the preliminary welding information of one or more welding sub-paths corresponding to each welding robot. Based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot, the target welding information of one or more welding sub-paths corresponding to each welding robot is determined. Based on the target welding information of one or more welding sub-paths corresponding to each welding robot, multiple welding robots are controlled to perform LSW laser welding.

[0015] This invention provides an LSW laser welding method and system. The method includes acquiring three-dimensional information of a large structural component and information of multiple welding robots; determining multiple high-stress regions and multiple low-stress regions based on the three-dimensional information of the large structural component using a stress analysis model; determining multiple segmented sub-path information based on the three-dimensional information of the large structural component, the multiple high-stress regions, the multiple low-stress regions, and the multiple welding robot information; constructing a collaborative task graph based on the multiple segmented sub-paths, the collaborative task graph including multiple task nodes and edges between task nodes, each task node corresponding to a segmented sub-path, the node features of the task node including segmented sub-path information and multiple welding robot information, and the edges between nodes representing the spatial positional relationship between segmented sub-paths; processing the collaborative task graph based on a graph neural network to determine preliminary welding information for one or more welding sub-paths corresponding to each welding robot; determining target welding information for one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information for one or more welding sub-paths corresponding to each welding robot; and controlling multiple welding robots to perform LSW laser welding based on the target welding information for one or more welding sub-paths corresponding to each welding robot. This method can efficiently and accurately determine the multi-robot LSW welding information adaptation scheme for large structural components. Attached Figure Description

[0016] Figure 1 A schematic flowchart of an LSW laser welding method provided in an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of a large structural component provided in an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of a process for determining multiple segmented sub-path information provided in an embodiment of the present invention;

[0019] Figure 4 This invention provides a schematic flowchart for determining target welding information for one or more welding sub-paths corresponding to each welding robot.

[0020] Figure 5 This is a schematic diagram of an LSW laser welding system provided in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0022] In this embodiment of the invention, the following are provided: Figure 1 The LSW laser welding method shown includes steps S1 to S7:

[0023] Step S1: Obtain the three-dimensional information of the large structural component and information of multiple welding robots.

[0024] Large structural components refer to core load-bearing components in the aerospace field that need to be formed by welding multiple units, such as aircraft fuselage reinforcement frames and wing main load-bearing beams. Figure 2 This is a schematic diagram of a large structural component provided in an embodiment of the present invention.

[0025] The three-dimensional information of large structural components is obtained by comprehensively scanning the large structural components with a laser 3D scanner, providing precise information describing their three-dimensional structure. The three-dimensional information of large structural components includes global reference coordinate point cloud information and key dimension information.

[0026] The global reference coordinate point cloud information includes complete spatial coordinate data of the outer contour of the structure, internal cavities, surface protrusions, surface depressions, and holes.

[0027] Key dimensional information includes precise values ​​for the length, width, height, thickness, diameter, hole diameter, and depth of each component, as well as related parameters such as the spacing, angle, and mating clearance between components.

[0028] Multiple welding robot information refers to the specific information of each robot participating in the welding operation. Welding robot information includes parameters such as robot model, welding range, motion accuracy, welding speed, load capacity parameters, laser emission power, and welding head angle adjustment range.

[0029] Step S2: Based on the three-dimensional information of the large structural component, a stress analysis model is used to determine multiple high-stress areas and multiple low-stress areas.

[0030] The stress analysis model is a deep neural network model. The input to the stress analysis model is the three-dimensional information of the large structural component, and the output of the stress analysis model is multiple high-stress regions and multiple low-stress regions.

[0031] Deep neural network models include deep neural networks (DNNs). A deep neural network is an artificial neural network containing multiple hidden layers. Deep neural network models simulate the connection patterns of neurons in the human brain and utilize multi-layer nonlinear transformations to extract features and abstractly represent data. Deep neural network models compute output values ​​through forward propagation and adjust network weights using the backpropagation algorithm, thus enabling them to handle complex nonlinear relationships and high-dimensional data patterns.

[0032] High-stress areas are regions of high stress concentration and large values ​​in large structural components under stress conditions, as determined by stress analysis models.

[0033] High-stress areas can be located at corners, joints, or points of geometric abrupt change in structural components. These areas require special attention during the welding process to prevent cracking or deformation.

[0034] Low-stress regions are areas in large structural components where the stress values ​​are relatively small and the distribution is relatively uniform under stress, as determined by stress analysis models.

[0035] Low-stress areas can be located on flat surfaces of structural components or in non-critical load-bearing areas. In welding planning, low-stress areas can be considered as secondary areas of concern or transitional zones along the welding path.

[0036] Three-dimensional information of large structural components can fully reflect the geometric structure and spatial form of the structure. Since the stress distribution of the structure is closely related to its geometric characteristics, three-dimensional information can provide comprehensive structural data for stress analysis models, enabling the models to accurately determine the areas of stress concentration and dispersion.

[0037] Deep neural networks, through their multi-layered network structure, can learn the nonlinear mapping relationship between geometric features and stress distribution in three-dimensional information. Deep neural networks can automatically identify the mechanical properties hidden within complex geometric shapes, thereby accurately dividing the surface of structural components into high-stress and low-stress regions.

[0038] Step S3: Determine multiple segmentation sub-path information based on the three-dimensional information of the large structural component, the multiple high-stress areas, the multiple low-stress areas, and the information of the multiple welding robots.

[0039] In some embodiments, Figure 3This is a flowchart illustrating a process for determining multiple segmented sub-path information according to an embodiment of the present invention. The determination of multiple segmented sub-path information includes steps S31 to S33:

[0040] Step S31: Based on the three-dimensional information of the large structural component, the multiple high-stress areas, and the multiple low-stress areas, determine multiple necessary welding points and multiple welding avoidance areas.

[0041] In some embodiments, a welding target analysis model can be used to determine multiple mandatory welding points and multiple welding avoidance areas. The welding target analysis model is a convolutional neural network. The inputs to the welding target analysis model are the three-dimensional information of the large structural component, the multiple high-stress areas, and the multiple low-stress areas; the outputs of the welding target analysis model are the multiple mandatory welding points and the multiple welding avoidance areas.

[0042] A Convolutional Neural Network (CNN) is a deep feedforward neural network that incorporates convolutional computations. CNNs can extract local features through convolutional layers, reduce feature dimensionality using pooling layers, and perform classification or regression through fully connected layers. CNNs possess translation invariance and local awareness capabilities, enabling them to effectively extract key features from spatial data and identify complex patterns.

[0043] Multiple essential welding points are specific locations on large structural components where welding operations must be performed, determined by the welding target analysis model.

[0044] Multiple welding points are key connection locations determined based on structural strength requirements and connection needs, and cover key stress points in high-stress areas as well as necessary joints in structural assembly.

[0045] Multiple welding avoidance zones are areas on large structural components that need to be protected from direct laser beam irradiation and heat effects during the welding process, as determined by the welding target analysis model.

[0046] Multiple welding avoidance areas include material-sensitive areas, reserved holes, non-welded functional surfaces, and low-stiffness areas that are severely deformed due to heat input.

[0047] The three-dimensional information of large structural components provides the model with a global spatial environment, and the high-stress and low-stress areas further clarify the mechanically critical and non-critical objects. This information together constitutes the constraints and objectives for welding path planning, enabling the model to distinguish where welding is necessary and where it is not.

[0048] Convolutional neural networks (CNNs) use convolutional kernels to scan three-dimensional information and stress region distribution maps, extracting spatial geometric and mechanical features related to welding process requirements. Through powerful image and spatial pattern recognition capabilities, CNNs can classify locations that meet welding connection characteristics as essential welding points and locations with risky or non-connection characteristics as welding avoidance zones.

[0049] In some embodiments, determining multiple welding necessity points and multiple welding avoidance areas based on the three-dimensional information of the large structural component, the multiple high-stress areas, and the multiple low-stress areas includes steps S311 to S313:

[0050] Step S311: Based on the three-dimensional information of the large structural component, the multiple high-stress regions, and the multiple low-stress regions, determine the set of connection points of the high-stress region structure, the set of support points of the high-stress region structure, the set of auxiliary points of the low-stress region structure, the set of edge points of the low-stress region structure, and the set of overlapping points of the stress region and the structure connection edge.

[0051] In some embodiments, a convolutional neural network can be used to determine the set of connection points of the high-stress zone structure, the set of support points of the high-stress zone structure, the set of auxiliary points of the low-stress zone structure, the set of edge points of the low-stress zone structure, and the set of overlapping points of the stress zone and the structure connection edge.

[0052] The structural connection points in high-stress areas are determined by a convolutional neural network, consisting of all points within the high-stress region of a large structural component that serve as connections between parts.

[0053] The structural support points in the high-stress zone are determined by a convolutional neural network, consisting of all points within the high-stress area of ​​a large structural component that provide structural support.

[0054] The auxiliary points of the low-stress zone structure are determined by a convolutional neural network and consist of all points in the low-stress zone of a large structural component that play an auxiliary connection or support role.

[0055] The structural edge points in the low-stress zone are determined by a convolutional neural network, consisting of all points located at the edge of the structure within the low-stress region of a large structural component.

[0056] The overlap points between the stress zone and the structural connection edge are determined by a convolutional neural network, which are the points where the high-stress area, low-stress area and the structural connection edge of the large structural component overlap.

[0057] Convolutional neural networks (CNNs) can perform multi-layer convolution operations on the three-dimensional information of large structural components to accurately extract spatial information such as component connection contours, node distribution features, and edge morphology details from the structural surface. CNNs can also use the range data of multiple high-stress and low-stress regions as constraints, and through feature matching and spatial localization algorithms, can filter out connection points located on component connection contours within high-stress regions and support points at key nodes of the supporting structure. Simultaneously, CNNs can identify auxiliary points at auxiliary connection locations within low-stress regions and edge points on the structural edge contours. Furthermore, CNNs can locate overlapping points where high-stress or low-stress regions coincide with structural connection edges in terms of coordinates through spatial coordinate comparison.

[0058] Step S312: Based on the set of connection points of the high stress zone structure, the set of support points of the high stress zone structure, the set of auxiliary points of the low stress zone structure, the set of edge points of the low stress zone structure, and the set of overlapping points of the stress zone and the structure connection edge, determine multiple necessary welding points for ensuring the structure, multiple optional welding points that do not affect safety, a process feasibility score for each welding point, areas with welding risks and hidden dangers, and areas where welding is not necessary.

[0059] In some embodiments, a convolutional neural network can be used to determine multiple necessary welding points for the protective structure, multiple optional welding points that do not affect safety, a process feasibility score for each welding point, areas with potential welding risks, and areas where welding is unnecessary.

[0060] Multiple essential welding points for structural protection are identified through convolutional neural networks, which are indispensable for ensuring the overall strength and stability of large structural components.

[0061] Failure to weld essential welding points in the structure will directly affect structural safety.

[0062] Multiple optional welding points that do not affect safety are determined by a convolutional neural network. These welding points will not affect the core safety performance of large structural components when not welded, but can further optimize the structural connection effect after welding.

[0063] The feasibility score for each welding point is determined by a convolutional neural network after evaluating each welding point. It is a score that reflects the difficulty and feasibility of implementing the welding process at that point.

[0064] The welding risk hazard area is the area that is prone to welding defects, safety problems, or structural damage during the welding operation, as determined by the convolutional neural network.

[0065] The areas where welding is not necessary are those identified by convolutional neural networks. These areas are where welding will not be performed and will not affect the strength, stability, or performance of large structural components.

[0066] Convolutional neural networks (CNNs) can perform in-depth analysis of point information in various input point sets, and combine structural mechanics principles and welding process requirements to determine the degree of influence of each point on the structural strength. This allows them to distinguish between points requiring welding and those that can be welded. Simultaneously, CNNs can assess the weldability and operational difficulty of each point, thereby calculating a process feasibility score and identifying areas prone to welding risks and areas where welding is unnecessary.

[0067] Step S313: Based on the multiple necessary welding points of the protection structure, the multiple optional welding points that do not affect safety, the process feasibility score of each welding point, the welding risk hazard area, and the area where welding is not necessary, determine multiple necessary welding points and multiple welding avoidance areas.

[0068] In some embodiments, deep neural networks can be used to determine multiple welding must-pass points and multiple welding avoidance areas.

[0069] Deep neural networks can perform multi-dimensional feature extraction and nonlinear fusion analysis on the spatial distribution characteristics, stress bearing weights, and process feasibility scores of multiple essential welding points for structural protection and multiple optional welding points that do not affect safety. Then, through gradient iterative learning of multi-layer neurons, the optional welding points can be prioritized to select those with high process feasibility scores and that can significantly improve structural stability. Essential welding points are then integrated and identified as multiple mandatory welding points. Simultaneously, deep neural networks can quantitatively assess the risk level of areas with welding risks and the structural redundancy of areas without necessary welding. By combining the spatial adjacency relationships of different areas and the chain reaction effects of welding operations, and through feature mapping and classification output, the model can dynamically delineate multiple welding avoidance areas with clear boundaries, no overlap, and adapted to the overall welding plan. This ensures that the selection of mandatory welding points meets structural strength requirements while reducing welding risks through the precise delineation of avoidance areas.

[0070] Step S32: Determine the overall welding planning path information based on the three-dimensional information of the large structural component, the multiple welding necessary points, and the multiple welding avoidance areas.

[0071] In some embodiments, a welding planning model can be used to determine the overall welding planning path information. The welding planning model is a Transformer model. The inputs to the welding planning model are the three-dimensional information of the large structural component, the multiple mandatory welding points, and the multiple welding avoidance areas; the output of the welding planning model is the overall welding planning path information.

[0072] The core architecture of the Transformer model consists of an encoder and a decoder. The encoder is responsible for deep feature mining and representation learning from the input data. Internally, the encoder integrates a self-attention mechanism and a feedforward neural network. The self-attention mechanism captures the relationships between different positions in the input data, while the feedforward neural network performs non-linear transformations and enhancements on the extracted features. The decoder, in addition to receiving the encoder's output features, adds a multi-head attention mechanism module. This module can simultaneously focus on multi-dimensional key information in the encoder's output, thereby achieving accurate decoding of the encoded features and generating a target sequence that meets the task requirements.

[0073] The overall welding planning path information is the information of the overall welding trajectory output by the welding planning model, which covers all necessary welding points and avoids welding avoidance areas.

[0074] The overall welding planning path information includes the welding start point, end point, path direction, suggested movement speed, and connection sequence between each welding point.

[0075] The three-dimensional information of large structural components defines the feasible space of the path, the necessary welding points determine the must-reach nodes of the path, and the welding avoidance areas define the prohibited areas of the path. These input data provide a complete map environment and constraints for the welding planning model, ensuring that the path planned by the model meets process requirements and conforms to geometric logic.

[0076] The Transformer model, through its self-attention mechanism, can simultaneously focus on the complex spatial relationships between the three-dimensional information of large structural components, multiple necessary welding points, and multiple welding avoidance zones, capturing the impact of each element on path planning while also incorporating the constraints of the avoidance zones. The Transformer model uses an encoder to understand the complex spatial constraint context and encode features into the input data. The decoder then generates an overall welding path that meets the constraints based on the encoding results, thus constructing an efficient and safe overall welding planning path.

[0077] Step S33: Determine multiple segmented sub-path information based on the overall welding planning path information and the multiple welding robot information.

[0078] In some embodiments, a path segmentation model can be used to determine multiple segmented sub-path information. The path segmentation model is a deep neural network model. The input to the path segmentation model is the overall welding planning path information and the information of the multiple welding robots, and the output of the path segmentation model is the multiple segmented sub-path information.

[0079] Multiple sub-path information is a collection of information about several path segments formed by splitting the overall welding planning path through a path segmentation model. Each path segment is assigned a welding robot to be adapted to it.

[0080] The sub-path segmentation information includes the geometric features of the sub-path, welding difficulty coefficient, weld quality requirement parameters, robot adaptation parameters, and estimated welding time.

[0081] Geometric feature information includes three-dimensional spatial coordinate sequence, path shape, curvature variation parameters, total path length and segment length, normal angle between the path and the surface of large structural components, and connection angle between adjacent segments of the path.

[0082] The welding difficulty coefficient is a numerical indicator that describes the complexity of the process for that section of the path.

[0083] The weld quality requirements parameters include weld penetration standards, weld width tolerances, and non-destructive testing qualification levels.

[0084] Robot adaptation parameters include the motion axis travel adaptation range of the welding robot and the welding head posture adjustment threshold.

[0085] The overall welding planning path information provides a complete framework for the welding task path, clarifying the overall scope and sequence of welding. Information on multiple welding robots provides available execution resources and their capability limitations. This input data enables the model to find a balance between the total number of tasks and execution capabilities, providing a data foundation for reasonable task allocation.

[0086] Deep neural networks can fuse and analyze features from overall welding planning path information and information from multiple welding robots to learn optimization rules for path segmentation. The deep neural network discretizes the overall welding planning path information into a series of path units and matches the coordinates and curvature of each unit with the base position and arm span range from multiple welding robot information. In the hidden layers, the model can construct a matching matrix between task load and robot capabilities, thereby evaluating the reachability and execution efficiency of each robot for different segments of the path. The deep neural network uses optimization algorithms to find the optimal cutting point, ensuring that each segmented path falls completely within the corresponding robot's workspace. The deep neural network can calculate the welding difficulty coefficient based on the curvature change rate of the path segment and the required welding torch posture adjustment range, encapsulate these attributes, and output them to ultimately determine multiple segmented sub-path information suitable for each robot.

[0087] Step S4: Construct a collaborative task graph based on the multiple segmented sub-paths. The collaborative task graph includes multiple task nodes and edges between task nodes. Each task node corresponds to a segmented sub-path. The node features of the task node include segmented sub-path information and multiple welding robot information. The edges between nodes represent the spatial positional relationship between the segmented sub-paths.

[0088] A collaborative task graph is a graph-structured data representation of the relationships and attributes among multiple welding tasks. It structurally expresses the characteristics of segmented sub-paths and their spatial relationships through nodes and edges.

[0089] Multiple task nodes are the basic units that carry welding task information in the collaborative task graph. Each task node corresponds to a segmented sub-path, and the node characteristics of the task node include segmented sub-path information and information of multiple welding robots.

[0090] The edges between task nodes are the associations connecting different task nodes in the collaborative task graph. The edges between task nodes represent the spatial positional relationship between the corresponding sub-paths.

[0091] The spatial relationship between sub-paths describes the relative positions, distances, and overlaps of different sub-paths in three-dimensional space.

[0092] For example, if two sub-paths are too close together, it may cause a collision between robots. This proximity relationship is reflected through spatial position.

[0093] Step S5: Process the collaborative task map based on the graph neural network to determine the preliminary welding information of one or more welding sub-paths corresponding to each welding robot.

[0094] Graph Neural Networks (GNNs) are neural network models that can directly operate on graph-structured data. GNNs update node feature representations by passing messages between nodes and aggregating information from neighboring nodes. GNNs can capture topological dependencies and complex interactions between nodes in a graph structure and are suitable for processing data in non-Euclidean spaces. The input to the GNN is the cooperative task graph, and the output is preliminary welding information for one or more welding sub-paths corresponding to each welding robot.

[0095] The initial welding information for one or more welding sub-paths corresponding to each welding robot is determined by a graph neural network, which identifies the specific welding task assigned to each robot and its initial process parameters.

[0096] Each welding robot can perform welding operations in one or more welding subpaths.

[0097] Preliminary welding information includes the ID of one or more welding sub-paths assigned to the corresponding welding robot, multiple welding points to be verified on each welding sub-path, initial value of welding laser power, initial value of welding robot operating speed, and initial value of welding defocusing amount.

[0098] Preliminary welding information for one or more welding sub-paths corresponding to each welding robot can provide a basic basis for welding operations. Through targeted welding operations in the preliminary welding information, the actual welding effect can be verified, providing a real reference for optimizing and improving the welding plan, and subsequently determining the target welding information that meets the actual welding needs and ensures welding quality.

[0099] In the collaborative task graph, each node represents a segmented sub-path, and its features include sub-path information and information about multiple welding robots. These features directly determine the compatibility between the sub-path and the welding robot, as well as the feasibility of the welding operation. Edges between nodes represent the spatial relationships between sub-paths, reflecting the collaborative constraints between welding operations on different sub-paths, such as job sequence connection and spatial interference avoidance. By processing the graph data using a graph neural network, preliminary welding information for one or more welding sub-paths corresponding to each welding robot can be determined. This helps to achieve reasonable allocation of welding tasks and collaborative operation of robots, thereby avoiding job conflicts or resource waste. Through graph convolution operations, the graph neural network can capture comprehensive information about node features and inter-node relationships, thus generating effective feature representations of the nodes. These representations integrate the inherent attributes of the sub-paths and the collaborative requirements between paths, facilitating accurate matching of robots and welding sub-paths and the formulation of preliminary welding parameters.

[0100] Graph neural networks (GNNs) enable each task node in a collaborative task graph to perceive the state of its neighbors through message passing mechanisms. GNNs can aggregate the spatial relationships between adjacent sub-paths and robot information, and learn high-order feature representations of nodes through graph convolution operations. This allows them to deduce preliminary welding information for one or more welding sub-paths corresponding to each welding robot, while considering global collaboration and obstacle avoidance.

[0101] Step S6: Determine the target welding information for one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot.

[0102] In some embodiments, Figure 4This invention provides a flowchart illustrating the process of determining target welding information for one or more welding sub-paths corresponding to each welding robot, wherein determining the target welding information for one or more welding sub-paths corresponding to each welding robot includes steps S61 to S64:

[0103] Step S61: Based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot, control multiple welding robots to perform LSW laser welding on the large structural component being tested, and obtain information on multiple welding points on each welding sub-path after welding is completed.

[0104] Information on multiple welding points along each welding sub-path is obtained by real-time acquisition and recording of data at multiple key points during actual welding testing using laser displacement sensors, infrared thermometers, and weld appearance inspection instruments. This data includes the actual welding position coordinates, weld penetration depth, weld width, surface finish quality data, and real-time feedback signals during the welding process.

[0105] Surface forming quality data are quantitative indicators characterizing the appearance and physical state of welds. Surface forming quality data include weld reinforcement height and uniformity, weld undercut depth and length, number and diameter of weld pores, slag inclusion area ratio, weld surface roughness, and weld-base transition angle.

[0106] Real-time feedback signals during the welding process include real-time monitoring values ​​of molten pool temperature, arc voltage fluctuation values, welding arc stability level, weld metal deposition rate, shielding gas back pressure value, and real-time distance deviation between the welding torch and the base material.

[0107] The information on multiple welding points on each welding sub-path after welding is completed can objectively present the actual welding status of the preliminary welding information of one or more welding sub-paths corresponding to each welding robot, and accurately reflect the welding quality details and parameter adaptation results of each point to be verified, providing direct data basis for subsequent identification of welding quality differences and positioning parameter optimization direction.

[0108] Step S62: Cluster the multiple welding point information on each welding sub-path to obtain multiple welding information groups on each welding sub-path.

[0109] The clustering method used is K-means clustering. K-means clustering is an iterative clustering analysis algorithm. It pre-determines the number of clusters K, randomly selects K objects as initial cluster centers, and then calculates the distance between each object and each cluster center to assign each object to the nearest cluster center. K-means clustering continuously updates the cluster centers and reallocates objects until a convergence condition is met, thus dividing the data into K clusters.

[0110] Each welding sub-path corresponds to multiple welding information groups, and the number of welding information groups is consistent with the preset number of clusters in the K-means clustering algorithm. In some embodiments, the K value in the K-means clustering algorithm can be preset manually.

[0111] Each welding information group on each welding sub-path is a set of welding point information with highly consistent data characteristics, obtained by classifying multiple welding point information contained in that welding sub-path according to feature similarity using the K-means clustering algorithm. There are significant differences in the collected data characteristics between different welding information groups on each welding sub-path.

[0112] Within each welding information group, the welding point information exhibits high consistency across dimensions such as actual welding position coordinates, weld penetration, weld width, surface finish quality data, and real-time feedback signals from the welding process. However, significant differences exist in the welding point information across different welding information groups within each welding sub-path across the dimensions of the collected data.

[0113] As an example, specifically, K welding point information on the welding sub-path are randomly selected as initial cluster centers. For each welding point information on the welding sub-path, its feature similarity with all initial cluster centers is calculated in terms of the collected data dimensions such as actual welding position coordinates, weld penetration, weld width, surface forming quality data, and real-time feedback signals of the welding process. Then, each welding point information is assigned to the cluster center with the highest feature similarity, thus forming K welding information groups. For each welding information group, the feature average value of the collected data corresponding to all welding point information within the group is calculated, and this average value is used as the new cluster center. The above steps are repeated until the feature change of the cluster center is less than a preset threshold or a predetermined number of iterations is reached, at which point the clustering ends.

[0114] Clustering into multiple welding information groups effectively integrates the complex welding point information on the same welding sub-path. Since the welding point information on each welding sub-path encompasses multiple dimensions of the collected data, and the characteristics of each welding point differ, directly analyzing all welding point information is quite difficult. However, by grouping welding point information with similar collected data characteristics into multiple welding information groups, the data structure is greatly simplified, facilitating the rapid extraction of valuable core information. By dividing the welding point information on each welding sub-path into multiple welding information groups, the distribution and characteristic differences of different types of welding points on that sub-path can be visually displayed. Analysis of these welding information groups allows for the rapid identification of the main data characteristics and quality performance of welding points on that sub-path. Further comparison of the characteristic differences between different welding information groups within the same welding sub-path clearly reveals the impact of welding condition changes on welding quality in different areas of that sub-path, providing precise data support for subsequent adjustments to welding parameters.

[0115] Step S63: Determine the adjustment welding point information for each welding sub-path based on the multiple welding information groups on each welding sub-path.

[0116] In some embodiments, a welding information analysis model can be used to determine the adjusted welding point information for each welding sub-path. The welding information analysis model is a Transformer model. The input to the welding information analysis model is multiple welding information groups on each welding sub-path, and the output of the welding information analysis model is the adjusted welding point information for each welding sub-path.

[0117] The adjusted welding point information for each welding sub-path is obtained by modifying the preliminary welding information corresponding to each welding sub-path using a welding information analysis model to obtain optimized welding parameter information.

[0118] The adjusted welding point information for each welding sub-path includes the welding laser power correction value, welding robot operation speed correction value, and welding defocusing amount correction value for each welding point used to verify the welding effect in each welding sub-path.

[0119] Multiple welding information groups on each welding sub-path represent different types of welding result patterns and clearly present the quality status of different types of weld points on that sub-path. These grouped data reveal which areas have good welding results and which areas have deviations, providing a basis for the model to infer parameter optimization from actual results.

[0120] The Transformer model, through its self-attention mechanism, can capture the correlation features between multiple welding information groups on each welding sub-path, and analyze the welding quality issues reflected by different groups of welding point information. The Transformer model can focus on information groups that are clustered with large deviations and calculate the amount of parameter compensation required to achieve the ideal welding effect, thereby accurately determining the adjustment of welding point information for each welding sub-path.

[0121] Step S64: Based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot and the adjusted welding point information of each welding sub-path, determine the target welding information of one or more welding sub-paths corresponding to each welding robot.

[0122] In some embodiments, a target information determination model can be used to determine the target welding information for one or more welding sub-paths corresponding to each welding robot. The target information determination model is a deep neural network model. The inputs to the target information determination model are the preliminary welding information for one or more welding sub-paths corresponding to each welding robot and the adjusted welding point information for each welding sub-path. The output of the target information determination model is the target welding information for one or more welding sub-paths corresponding to each welding robot.

[0123] The target welding information for one or more welding sub-paths corresponding to each welding robot is the final welding scheme information generated after the target information determination model is optimized. The target welding information for one or more welding sub-paths corresponding to each welding robot includes the path welding sequence of each welding robot, the sequence of welding laser power variation with position in each welding sub-path, the sequence of welding robot operating speed variation with position, and the sequence of welding defocusing amount variation with position.

[0124] The preliminary welding information for one or more welding sub-paths corresponding to each welding robot defines the core process parameter types and basic value ranges for the target welding information. The adjusted welding point information for each welding sub-path can quantitatively output the parameter correction magnitude and adaptation rules at each verification point, providing a specific correction basis for the fitting calculation of the full path parameters in the target welding information.

[0125] Deep neural networks can comprehensively analyze preliminary welding information and adjusted welding point information, integrate the optimization content in the adjusted welding point information into the preliminary welding information, and make precise adjustments to welding parameters, welding sequence, etc., to ensure that the adjusted welding scheme can avoid problems that occur in the test welding, and finally form the target welding information.

[0126] Deep neural networks can analyze the basic value boundaries of the initial values ​​of welding laser power, welding robot speed, and welding defocusing amount in the preliminary welding information. Combined with multiple welding point information on each welding sub-path acquired during trial welding, the model extracts and adjusts the parameter correction amplitude and adaptation rules for each welding point used to verify the welding effect. Through feature fusion algorithms, the model can map the optimization logic of the welding points used to verify the welding effect to the complete position range of the welding sub-path. Furthermore, by calibrating the values ​​of welding laser power, welding robot speed, and welding defocusing amount segment by segment, it can generate sequences of welding laser power variation with position, welding robot speed variation with position, and welding defocusing amount variation with position. Simultaneously, the deep neural network can combine the path connection defects reflected in the welding point information acquired after laser welding controlled by the welding robot based on the preliminary welding information to quantify and sort the execution priorities of multiple sub-paths, determining the path welding sequence.

[0127] Step S7: Based on the target welding information of one or more welding sub-paths corresponding to each welding robot, control multiple welding robots to perform LSW laser welding.

[0128] Once the target welding information for one or more welding sub-paths corresponding to each welding robot is determined, multiple welding robots are controlled based on the target welding information for one or more welding sub-paths corresponding to each welding robot to perform LSW laser welding operations on large structural components.

[0129] Based on the same inventive concept Figure 5 This is a schematic diagram of an LSW laser welding system provided in an embodiment of the present invention. The LSW laser welding system includes:

[0130] The acquisition module 81 is used to acquire three-dimensional information of large structural components and information of multiple welding robots;

[0131] The stress analysis module 82 is used to determine multiple high-stress areas and multiple low-stress areas based on the three-dimensional information of the large structural component using a stress analysis model.

[0132] The sub-path determination module 83 is used to determine multiple segmented sub-path information based on the three-dimensional information of the large structural component, the multiple high-stress areas, the multiple low-stress areas, and the information of the multiple welding robots;

[0133] The graph construction module 84 is used to construct a collaborative task graph based on the multiple segmented sub-paths. The collaborative task graph includes multiple task nodes and edges between task nodes. Each task node corresponds to a segmented sub-path. The node features of the task node include segmented sub-path information and multiple welding robot information. The edges between nodes represent the spatial positional relationship between the segmented sub-paths.

[0134] The preliminary welding information determination module 85 is used to process the collaborative task map based on a graph neural network to determine the preliminary welding information of one or more welding sub-paths corresponding to each welding robot.

[0135] The target welding information determination module 86 is used to determine the target welding information of one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot.

[0136] The welding control module 87 is used to control multiple welding robots to perform LSW laser welding based on the target welding information of one or more welding sub-paths corresponding to each welding robot.

[0137] It should be noted that, in order to simplify the descriptions disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments of this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0138] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A method for LSW laser welding, characterized in that, include: Acquire three-dimensional information of large structural components and information of multiple welding robots; Based on the three-dimensional information of the large structural component, a stress analysis model was used to determine multiple high-stress areas and multiple low-stress areas; Based on the three-dimensional information of the large structural component, the multiple high-stress regions, the multiple low-stress regions, and the information of the multiple welding robots, multiple segmented sub-path information is determined. This determination of multiple segmented sub-path information based on the three-dimensional information of the large structural component, the multiple high-stress regions, the multiple low-stress regions, and the information of the multiple welding robots includes: Based on the three-dimensional information of the large structural component, the multiple high-stress areas, and the multiple low-stress areas, multiple necessary welding points and multiple welding avoidance areas are determined. The overall welding planning path information is determined based on the three-dimensional information of the large structural component, the multiple welding necessary points, and the multiple welding avoidance areas; Based on the overall welding planning path information and the information of the multiple welding robots, multiple segmented sub-path information is determined; A collaborative task graph is constructed based on the multiple segmented sub-paths. The collaborative task graph includes multiple task nodes and edges between task nodes. Each task node corresponds to a segmented sub-path. The node features of the task node include segmented sub-path information and multiple welding robot information. The edges between nodes represent the spatial positional relationship between the segmented sub-paths. The collaborative task map is processed based on a graph neural network to determine the preliminary welding information of one or more welding sub-paths corresponding to each welding robot. Based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot, the target welding information of one or more welding sub-paths corresponding to each welding robot is determined, wherein determining the target welding information of one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot includes: Based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot, multiple welding robots are controlled to perform LSW laser welding on the test large structural component, and information on multiple welding points on each welding sub-path after welding is completed is obtained. Clustering is performed on the multiple welding point information on each welding sub-path to obtain multiple welding information groups for each welding sub-path; The adjustment welding point information for each welding sub-path is determined based on multiple welding information groups on each welding sub-path; Based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot and the adjusted welding point information of each welding sub-path, the target welding information of one or more welding sub-paths corresponding to each welding robot is determined. Based on the target welding information of one or more welding sub-paths corresponding to each welding robot, multiple welding robots are controlled to perform LSW laser welding.

2. The LSW laser welding method as described in claim 1, characterized in that, The stress analysis model is a deep neural network model.

3. An LSW laser welding system, characterized in that, include: The acquisition module is used to acquire three-dimensional information of large structural components and information of multiple welding robots; The stress analysis module is used to determine multiple high-stress areas and multiple low-stress areas based on the three-dimensional information of the large structural component using a stress analysis model. The sub-path determination module is used to determine multiple segmented sub-path information based on the three-dimensional information of the large structural component, the multiple high-stress regions, the multiple low-stress regions, and the information of the multiple welding robots. The sub-path determination module is also used for: Based on the three-dimensional information of the large structural component, the multiple high-stress areas, and the multiple low-stress areas, multiple necessary welding points and multiple welding avoidance areas are determined. The overall welding planning path information is determined based on the three-dimensional information of the large structural component, the multiple welding necessary points, and the multiple welding avoidance areas; Based on the overall welding planning path information and the information of the multiple welding robots, multiple segmented sub-path information is determined; The graph construction module is used to construct a collaborative task graph based on the multiple segmented sub-paths. The collaborative task graph includes multiple task nodes and edges between task nodes. Each task node corresponds to a segmented sub-path. The node features of the task node include segmented sub-path information and multiple welding robot information. The edges between nodes represent the spatial positional relationship between the segmented sub-paths. The preliminary welding information determination module is used to process the collaborative task map based on a graph neural network to determine the preliminary welding information of one or more welding sub-paths corresponding to each welding robot. The target welding information determination module is used to determine the target welding information for one or more welding sub-paths corresponding to each welding robot based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot. The target welding information determination module is further used to: Based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot, multiple welding robots are controlled to perform LSW laser welding on the test large structural component, and information on multiple welding points on each welding sub-path after welding is completed is obtained. Clustering is performed on the multiple welding point information on each welding sub-path to obtain multiple welding information groups for each welding sub-path; The adjustment welding point information for each welding sub-path is determined based on multiple welding information groups on each welding sub-path; Based on the preliminary welding information of one or more welding sub-paths corresponding to each welding robot and the adjusted welding point information of each welding sub-path, the target welding information of one or more welding sub-paths corresponding to each welding robot is determined. The welding control module is used to control multiple welding robots to perform LSW laser welding based on the target welding information of one or more welding sub-paths corresponding to each welding robot.

4. The LSW laser welding system as described in claim 3, characterized in that, The stress analysis model is a deep neural network model.

5. An electronic device, characterized in that, include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the LSW laser welding method as claimed in any one of claims 1 to 2.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the LSW laser welding method as described in any one of claims 1 to 2.

Images