Industrial robot automatic welding method and control device

Abstract

The application provides an industrial robot automatic welding method and a control device, and relates to the technical field of machine control. The method comprises the following steps: obtaining three-dimensional point cloud data of a workpiece, adjusting a BIM model of the workpiece which is constructed in advance according to the three-dimensional point cloud data, aligning the adjusted BIM model with the actual position of the workpiece, obtaining a theoretical welding seam path from the adjusted BIM model, obtaining an actual welding seam image of the workpiece, determining calibration parameters of a robot according to the actual welding seam image and the theoretical welding seam path, determining a final welding path according to the calibration parameters, determining a final control instruction set according to the final welding path and the material and thickness of the workpiece, and controlling the robot to perform automatic welding according to the final control instruction set. The application can improve welding precision, does not require manual intervention, can greatly improve welding efficiency, and can adapt to complex and variable welding scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine control, and in particular to an industrial robot automatic welding method and a control device. BACKGROUND

[0002] Industrial robot automatic welding technology has irreplaceable important significance for improving production efficiency and ensuring product quality. With the rapid development of intelligent manufacturing, welding robots are widely used in steel structure, automobile manufacturing, shipbuilding and other fields, and their automation level directly affects the competitiveness of industrial production.

[0003] In related technologies, robot welding methods mostly rely on online teaching programming or fixed tool positioning, which not only requires a lot of preliminary work, but also requires high precision of workpiece placement. Once the workpiece position deviates or the model changes, the robot often cannot adapt flexibly and can only mechanically execute the preset program, resulting in poor welding precision and low welding efficiency, especially in complex and variable steel structure welding scenarios. SUMMARY

[0004] The embodiments of the present application provide an industrial robot automatic welding method and a control device to solve the problem of poor welding precision and low welding efficiency of existing robots in complex and variable welding scenarios.

[0005] In a first aspect, the embodiments of the present application provide an industrial robot automatic welding method, comprising:

[0006] Obtain three-dimensional point cloud data of the workpiece, and adjust the pre-constructed BIM model of the workpiece according to the three-dimensional point cloud data, so that the adjusted BIM model is aligned with the actual workpiece position;

[0007] Obtain the theoretical weld path from the adjusted BIM model;

[0008] Obtain the actual weld image of the workpiece;

[0009] Determine the calibration parameters of the robot according to the actual weld image and the theoretical weld path, and determine the final welding path according to the calibration parameters;

[0010] Determine the final control instruction set according to the final welding path and the material and thickness of the workpiece, and control the robot to perform automatic welding according to the final control instruction set.

[0011] In a possible implementation, obtaining the theoretical weld path from the adjusted BIM model comprises:

[0012] Extracting weld features from the adjusted BIM model to determine the theoretical weld position information;

[0013] The spatial alignment method is used to correct the theoretical weld position information, and corrected theoretical weld position information is obtained.

[0014] Based on the corrected theoretical weld position information, a theoretical initial weld path is generated.

[0015] The theoretical initial weld path is verified for path integrity, optimized to adapt to the actual workpiece surface undulation, and optimized to adapt to the robot operation requirements, and a theoretical weld path is obtained.

[0016] In one possible implementation, the actual weld image of the workpiece is obtained, including:

[0017] Based on the scanning device, a scanning data stream of the workpiece is obtained, and the scanning data stream is preprocessed based on the vision system to obtain preliminary scanning data.

[0018] Based on the preliminary scanning data, the workpiece surface features are extracted, the dynamic change trend of the workpiece surface features is determined, and the dynamic imaging range containing the weld is determined according to the dynamic change trend.

[0019] Based on the dynamic imaging range and the optical features, the weld area is separated and obtained;

[0020] The weld area is subjected to image enhancement and noise processing to obtain a stable weld image.

[0021] According to the stable weld image, the weld contour extraction is performed to generate an image containing the weld contour.

[0022] The image containing the weld contour is subjected to contour verification, the image generation parameters affected by environmental changes are adjusted, and the actual weld image is generated.

[0023] In one possible implementation, according to the actual weld image and the theoretical weld path, the calibration parameters of the robot are determined, including:

[0024] The weld edge position in the actual weld image is identified to obtain edge distribution data, and the edge distribution data is subjected to coordinate conversion to obtain an actual weld path.

[0025] The offset amount of the actual weld path and the theoretical weld path is calculated to obtain a preliminary deviation value.

[0026] For the actual area with a preliminary deviation value greater than a preset deviation threshold, interpolation adjustment is performed to obtain adjusted edge distribution data.

[0027] Based on the adjusted edge distribution data, the least square method is used to determine the calibration parameters of the robot.

[0028] In one possible implementation, the final welding path is determined according to the calibration parameters, including:

[0029] According to the calibration parameters, the pose data of the welding head of the robot is corrected to obtain corrected pose data;

[0030] According to the corrected pose data, the smooth movement trajectory is adjusted and smoothed to obtain smoothed trajectory data;

[0031] According to the smoothed trajectory data, an optimized welding path is generated, and feature distribution data of the optimized welding path is extracted;

[0032] If the feature distribution data does not match the preset feature threshold, the optimized welding path is adjusted based on the least square method to obtain adjusted path parameters;

[0033] According to the adjusted path parameters, a candidate control instruction set is generated, and the smoothed trajectory data is optimized according to the candidate control instruction set to obtain stable trajectory data;

[0034] According to the stable trajectory data, a final welding path is generated.

[0035] In a possible implementation, according to the final welding path and the material and thickness of the workpiece, a final control instruction set is determined, including:

[0036] The key point coordinates in the final welding path are extracted, and coordinate feature data of the key point coordinates are determined;

[0037] Based on a preset process parameter database, initial process parameters are determined according to the coordinate feature data and the material and thickness of the workpiece;

[0038] The initial process parameters are filtered to obtain smoothed process parameters;

[0039] If the smoothed process parameters exceed a preset process parameter threshold, the smoothed process parameters are corrected to obtain corrected process parameters;

[0040] The stability feature of the corrected process parameters is determined, and the final welding path is adjusted according to the stability feature and the key point coordinates to obtain an adjusted welding path;

[0041] According to the adjusted welding path and the corrected process parameters, a final control instruction set is generated.

[0042] In a possible implementation, a pre-constructed BIM model of a workpiece is adjusted according to three-dimensional point cloud data, including:

[0043] According to the three-dimensional point cloud data, optimized point cloud data and a target coordinate system reflecting the actual pose of the workpiece are determined;

[0044] Based on the target coordinate system, deviation data of the optimized point cloud data and the BIM model is determined, and a position transformation matrix is determined according to the deviation data;

[0045] According to the position transformation matrix, the BIM model is adjusted to obtain an adjusted BIM model.

[0046] In a possible implementation, the target coordinate system reflecting the actual posture of the workpiece is determined according to the three-dimensional point cloud data, including:

[0047] The three-dimensional point cloud data is subjected to noise removal to obtain effective point cloud data of the surface of the workpiece;

[0048] The effective point cloud data is subjected to clustering segmentation to determine position information of the workpiece in the effective point cloud data;

[0049] Based on the position information of the workpiece, boundary points of the contour feature in the effective point cloud data are determined;

[0050] Based on the boundary points, a preliminary coordinate system reflecting the actual posture of the workpiece is determined;

[0051] Based on the preliminary coordinate system, coordinate conversion is performed on the effective point cloud data to obtain optimized point cloud data;

[0052] The optimized point cloud data is subjected to contour feature verification, and the preliminary coordinate system that passes the verification is taken as the target coordinate system.

[0053] In a possible implementation, after the robot is controlled to perform automatic welding, the method further includes:

[0054] Feedback data of the robot is obtained;

[0055] Based on the feedback data, it is determined whether the welding quality of the robot is qualified;

[0056] If the welding quality of the robot is not qualified, the final control instruction set is updated according to the feedback data, and the robot is controlled to perform automatic welding according to the updated final control instruction set.

[0057] In a second aspect, an embodiment of the present application provides a control device, including a memory and a processor, the memory stores a computer program, and the processor implements the industrial robot automatic welding method in the first aspect or any possible implementation manner of the first aspect when executing the computer program.

[0058] In the embodiment of the present application, the BIM model of the workpiece pre-constructed is adjusted based on the actual three-dimensional point cloud data of the workpiece, so that the adjusted BIM model is completely aligned with the actual workpiece point cloud in spatial position, the welding deviation caused by the mispositioning of the BIM model can be avoided, the welding precision can be improved, then the theoretical welding seam path is obtained from the adjusted BIM model, and the actual welding seam image of the workpiece is obtained, and then the difference between the theoretical welding seam path and the actual welding seam path can be determined according to the actual welding seam image and the theoretical welding seam path, so that the calibration parameters of the robot can be determined, and the final welding path is determined according to the calibration parameters, finally, the final control instruction set is determined according to the final welding path and the material and thickness of the workpiece, and the robot is controlled to automatically weld according to the final control instruction set, the final control instruction set of the robot can be automatically adjusted based on the difference between the theory and the actual, the welding precision can be improved, manual intervention is not required, the welding efficiency can be greatly improved, and the welding scene can be adapted to complex and variable welding scenes. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 is an implementation flowchart of the industrial robot automatic welding method provided by the embodiment of the present application;

[0060] Figure 2 is a structural schematic diagram of the industrial robot automatic welding device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0061] The embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0062] Reference is made to Figure 1 which shows the implementation flowchart of the industrial robot automatic welding method provided by the embodiment of the present application, the execution subject of the method can be a control device, for example, it can be a controller of an industrial robot, or it can be other control devices, etc., which is not specifically limited here.

[0063] The above industrial robot automatic welding method is described in detail as follows:

[0064] In S101, three-dimensional point cloud data of a workpiece is obtained, and a pre-constructed BIM (Building Information Modeling, building information model) model of the workpiece is adjusted according to the three-dimensional point cloud data, so that the adjusted BIM model is aligned with the actual workpiece position.

[0065] The pre-constructed BIM model of the workpiece can be understood as a theoretical model or a design model of the workpiece, and can store three-dimensional design data of the workpiece, such as geometric shape, welding seam position, material and thickness properties, etc.

[0066] In the embodiments of the present application, the workpiece can be scanned by a laser scanner to collect three-dimensional point cloud data, which is the basis for obtaining the spatial information of the workpiece surface. The three-dimensional point cloud data can be obtained by three-dimensional point cloud collection. The three-dimensional point cloud data is the actual point cloud data of the workpiece. For example, in automobile manufacturing, the door weld is scanned, and the laser scanner captures the reflected light points on the workpiece surface at a speed of tens of thousands of points per second, forming dense three-dimensional point cloud data. These data contain information such as height, width and depth of the workpiece. The advantage of this method is non-contact measurement, which avoids potential damage to the workpiece caused by physical contact and improves efficiency.

[0067] The BIM model can be adjusted based on the three-dimensional point cloud data of the workpiece to obtain an adjusted BIM model. The adjusted BIM model is completely aligned with the spatial position of the actual workpiece point cloud data. The embodiments of the present application can avoid welding deviation caused by mispositioning of the BIM model and improve welding accuracy.

[0068] In S102, the theoretical weld path is obtained from the adjusted BIM model.

[0069] In the embodiments of the present application, the weld of the workpiece in the adjusted BIM model is referred to as the theoretical weld, and the corresponding weld path is referred to as the theoretical weld path. The theoretical weld path can be expressed in coordinate form from the starting point to the end point of the weld, or in other forms, which are not limited here.

[0070] The above-mentioned theoretical weld path can include the path of at least one theoretical weld of the workpiece.

[0071] In S103, the actual weld image of the workpiece is obtained.

[0072] In the embodiments of the present application, the actual weld image of the workpiece can be obtained based on a scanning device and a vision system. For example, the actual weld image of the workpiece can be obtained in real time.

[0073] In S104, the calibration parameters of the robot are determined according to the actual weld image and the theoretical weld path, and the final welding path is determined according to the calibration parameters.

[0074] In the embodiments of the present application, the actual weld path of the workpiece can be determined according to the actual weld image, and based on the difference between the actual weld path and the theoretical weld path, the calibration parameters of the robot can be determined. The calibration parameters of the robot can be used to adjust the pose of the welding head of the robot. The final welding path can be determined according to the calibration parameters of the robot. The final welding path can be understood as the path based on which the robot welds the weld of the workpiece.

[0075] In S105, the final control instruction set is determined according to the final welding path and the material and thickness of the workpiece, and the robot is controlled to perform automatic welding according to the final control instruction set.

[0076] The final control instruction set can include the path and corresponding process parameters for welding the workpiece.

[0077] The embodiments of the present application can obtain the material and thickness of the workpiece, generate the final control instruction set of the robot according to the material and thickness of the workpiece and the final welding path determined in the foregoing steps, and control the robot to perform automatic welding on the workpiece according to the final control instruction set, thereby improving the welding precision and efficiency.

[0078] The embodiments of the present application adjust the pre-constructed BIM model of the workpiece based on the actual three-dimensional point cloud data of the workpiece, so that the adjusted BIM model is completely aligned with the actual workpiece point cloud in spatial position, which can avoid welding deviation caused by mispositioning of the BIM model and improve welding precision. Then, the theoretical weld path is obtained from the adjusted BIM model, and the actual weld image of the workpiece is obtained. Furthermore, the difference between the theoretical weld path and the actual weld path can be determined according to the actual weld image and the theoretical weld path, so that the calibration parameters of the robot can be determined, and the final welding path can be determined according to the calibration parameters. Finally, the final control instruction set is determined according to the final welding path and the material and thickness of the workpiece, and the robot is controlled to perform automatic welding according to the final control instruction set. Based on the difference between the theory and the actual, the final control instruction set of the robot can be automatically adjusted, the welding precision can be improved, manual intervention can be avoided, the welding efficiency can be greatly improved, and the welding scene can be adapted to complex and variable conditions.

[0079] The above embodiments introduce the overall implementation process of the industrial robot automatic welding method, and the following will introduce each step in detail.

[0080] In some embodiments, in S101, the above adjustment of the pre-constructed BIM model of the workpiece according to the three-dimensional point cloud data can include:

[0081] According to the three-dimensional point cloud data, the optimized point cloud data and the target coordinate system reflecting the actual posture of the workpiece are determined;

[0082] Based on the target coordinate system, the deviation data of the optimized point cloud data and the BIM model is determined, and the position transformation matrix is determined according to the deviation data;

[0083] According to the position transformation matrix, the BIM model is adjusted to obtain the adjusted BIM model.

[0084] The optimized point cloud data is point cloud data after denoising, clustering segmentation, contour recognition and coordinate conversion on the three-dimensional point cloud data.

[0085] The target coordinate system is a coordinate system that can reflect the actual posture of the workpiece. Exemplarily, the origin of the target coordinate system can be the barycenter of the point cloud data of the workpiece body, the X-axis can be a feature vector of the main direction of the workpiece body, the Y-axis can be a feature vector of the secondary direction of the workpiece body, and the Z-axis can be an axis perpendicular to the XY plane; or the origin of the target coordinate system can be the barycenter of the point cloud data of the workpiece body, the X-axis can be an axis parallel to the fitting straight line of the weld of the workpiece, the Y-axis can be an axis perpendicular to the X-axis, and the Z-axis can be an axis perpendicular to the XY plane; or the origin of the target coordinate system can be the starting point of the weld of the workpiece, the X-axis can be the fitting straight line direction of the weld of the workpiece, the Y-axis can be an axis perpendicular to the X-axis, and the Z-axis can be an axis perpendicular to the XY plane.

[0086] The optimized point cloud data can also be understood as point cloud data converted into the target coordinate system.

[0087] In the embodiments of the present application, based on the target coordinate system, a registration algorithm can be used to preliminarily match the optimized point cloud data with the BIM model to determine the deviation data. The core of the registration algorithm is to align the optimized point cloud data with the BIM model.

[0088] Exemplarily, in the welding workpiece scenario, assuming that the optimized point cloud data comes from the door weld area, and the BIM model is the design blueprint of the door, the barycenter or center of the two can be preliminarily aligned through coarse registration, for example, the geometric center of the optimized point cloud data is coincided with the center of the BIM model, and then the key feature points are used for fine adjustment.

[0089] Wherein, due to the uneven density of the point cloud, there can be about 5mm deviation in the initial alignment during coarse registration, which needs to be further analyzed. After preliminary matching (coarse registration), the determination of the deviation data can be realized by comparing the key features of the optimized point cloud data and the BIM model. In one possible implementation, the starting point and the ending point of the weld can be selected as the key features, and the deviation data can be the deviation data of the key features, for example, the deviation data of the starting point and the ending point of the weld.

[0090] Assuming that in the target coordinate system, the position coordinates of the start and end points of the weld in the optimized point cloud data are (102 mm, 52 mm, 10 mm) and (150 mm, 53 mm, 10 mm), and in the theoretical coordinate system of the BIM model, the position coordinates of the start and end points of the weld corresponding to the BIM model are (100 mm, 50 mm, 10 mm) and (148 mm, 51 mm, 10 mm). By comparison, the deviation data is 2 mm in the X direction, 2 mm in the Y direction, and 0 mm in the Z direction. This method is more intuitive and can quickly locate the problem area, providing a basis for subsequent adjustment.

[0091] The deviation data not only includes the size of the deviation data, but also includes the direction of the deviation data. Through the deviation data, the translation and rotation requirements of the optimized point cloud data relative to the BIM model can be determined. For example, if the optimized point cloud data is shifted to the right and up as a whole, it is calculated that it needs to be translated 2 mm along the X axis and 2 mm along the Y axis. It can be understood that this geometric calculation is based on spatial vector analysis, and the position transformation matrix essentially describes how the optimized point cloud data is "moved" to the ideal position. Illustratively, assuming that a 3-degree clockwise rotation deviation is also detected in the optimized point cloud data, the position transformation matrix needs to contain both rotation and translation information. This way ensures the comprehensiveness of the transformation. Determining the final coordinate transformation result according to the position transformation matrix is an embodiment of applying theoretical calculation to actual data. In one possible implementation, after the position transformation matrix is applied to the point cloud, the coordinates of the start point of the weld are adjusted from (102 mm, 52 mm, 10 mm) to (100 mm, 50 mm, 10 mm), which is completely consistent with the BIM model.

[0092] The position transformation matrix of the embodiments of the present application not only can correct the position, but also can maintain the geometric characteristics of the point cloud, such as the length and width of the weld are not stretched or compressed. Illustratively, if the original point cloud is inclined due to the scanning angle, it can be restored to the horizontal state in the design after transformation. This adjustment provides a high-precision basis for subsequent workpiece positioning.

[0093] It should be noted that the above description is described by adjusting the optimized point cloud data to align with the BIM model, but the position transformation matrix of the embodiments of the present application is a position transformation matrix including translation and rotation, which is subsequently applied to the BIM model, so that the BIM model is aligned with the optimized point cloud data, and the two are accurately aligned in the same coordinate system, so that the theoretical weld path extracted from the model can directly guide the actual welding.

[0094] It should be noted that the selection of the registration algorithm affects the matching efficiency. For example, if a key feature-based registration is used, it is suitable for scenarios where linear features such as welds are obvious; if the workpiece surface is complex, it can be extended to a global registration scheme. Preferably, a combination of coarse registration and fine registration can improve processing speed while ensuring accuracy.

[0095] The embodiment of the present application adjusts the BIM model according to the position transformation matrix to obtain an adjusted BIM model, so that the adjusted BIM model is aligned with the actual workpiece position.

[0096] Exemplarily, the BIM model and the optimized point cloud data can be aligned. Assuming that the deviation between the BIM model and the optimized point cloud data is 5 cm in the X-axis, 3 cm in the Y-axis, and 0 cm in the Z-axis, after adjustment by the position transformation matrix, the coordinates of each part of the BIM model can be consistent with the actual workpiece position. By adjusting the BIM model, the accuracy of subsequent feature extraction can be ensured.

[0097] Preferably, during the adjustment of the BIM model, a spatial grid division can be introduced to divide the BIM model into multiple sub-regions for separate processing, thereby avoiding cumulative errors caused by excessive overall deviation.

[0098] In some embodiments, the above determination of the optimized point cloud data and the target coordinate system reflecting the actual posture of the workpiece according to the three-dimensional point cloud data comprises:

[0099] Noise removal is performed on the three-dimensional point cloud data to obtain effective point cloud data of the workpiece surface;

[0100] The effective point cloud data is clustered and segmented to determine the position information of the workpiece in the effective point cloud data;

[0101] Based on the position information of the workpiece, boundary points of the contour feature in the effective point cloud data are determined;

[0102] Based on the boundary points, a preliminary coordinate system reflecting the actual posture of the workpiece is determined;

[0103] Based on the preliminary coordinate system, coordinate transformation is performed on the effective point cloud data to obtain the optimized point cloud data;

[0104] The optimized point cloud data is verified for contour features, and the preliminary coordinate system that passes the verification is taken as the target coordinate system.

[0105] The three-dimensional point cloud data contains not only the height, width and depth of the workpiece, but also environmental noise. The collected three-dimensional point cloud data can be irregularly distributed and needs to be further processed to extract useful information. Specifically, the three-dimensional point cloud data can be separated from the workpiece surface to remove noise and obtain effective point cloud data, wherein filtering technology can be used. In one possible implementation, a statistical filter can be used to remove outliers (noise). For example, assuming that the average distance of a neighborhood of a point in the three-dimensional point cloud data is 0.5 mm, and the distance of a noise point exceeds 1.5 mm, the noise point is removed. After separation, the effective point cloud data of the workpiece surface is more clearly distributed, for example, the point cloud of the weld area presents a strip-shaped feature. Noise removal can effectively reduce interference in subsequent analysis and improve data quality.

[0106] The data extraction method is used to cluster and segment the effective point cloud data, group the effective point cloud data according to spatial distance, and separate the workpiece body and the weld to determine the position information of the workpiece in the effective point cloud data, wherein the position information of the workpiece can include the position information of the weld and / or the position information of the workpiece body.

[0107] In one possible implementation, the Euclidean clustering method can be used to group the effective point cloud data according to spatial distance, for example, into two clusters of weld and workpiece body. The position information of the weld can be determined by the geometric center of the cluster corresponding to the weld, for example, (100 mm, 50 mm, 10 mm). The position information of the workpiece body can be determined by the geometric center of the cluster corresponding to the workpiece body. If there is more than one weld, each weld clustering can be a cluster, and each weld corresponds to a position information.

[0108] The method for determining the position information of the workpiece provided by the embodiments of the present application is intuitive and efficient, and can provide a reliable position reference for subsequent feature extraction.

[0109] The embodiment of the present application determines the boundary points of the contour feature in the effective point cloud data based on the position information of the workpiece, which is the key to point cloud analysis. In a possible implementation, the above-mentioned boundary points can be determined by the method of normal vector analysis. Exemplarily, the normal vector direction of the boundary points of the contour feature (such as the weld boundary points) changes greatly, while the normal vector direction of the flat area is relatively stable. A normal vector angle threshold is set, and based on the position information of the weld, if the angle between the normal vector of a certain point of the weld and the normal vector of at least one target adjacent point of the certain point is greater than the normal vector angle threshold, the certain point is determined as the boundary point of the weld; or if the maximum angle among the angles between the normal vector of a certain point and the normal vectors of each target adjacent point of the certain point is greater than the normal vector angle threshold, the certain point is determined as the boundary point of the weld. The target adjacent point of a certain point of the weld is the point adjacent to the certain point on the contour feature of the weld. Similarly, based on the position information of the workpiece body, the same method can be used to determine the boundary points of the contour of the workpiece body.

[0110] The normal vector angle threshold can be set according to actual needs, for example, it can be 30 degrees. The first preset number can be set according to actual needs, which is not limited here.

[0111] This way can accurately outline the weld contour, which is helpful for subsequent spatial positioning.

[0112] Based on the above-mentioned determined boundary points, the spatial coordinates can be generated by geometric calculation, and the preliminary coordinate system reflecting the actual posture of the workpiece can be determined based on the spatial coordinates, which can also be called coordinate system framework. Exemplarily, the least square method can be used to fit the boundary points of the weld to obtain a straight line direction vector, such as (1, 0, 0), which can be used as the X-axis direction, and at the same time, the center of the workpiece is used as the origin to establish the preliminary coordinate system. The preliminary coordinate system reflects the actual posture of the workpiece, provides a reference framework for point cloud adjustment, and reduces the deviation caused by the scanning angle.

[0113] Through the preliminary coordinate system, the distribution of the effective point cloud data can be adjusted to obtain the optimized point cloud data, that is, the effective point cloud data is converted into the preliminary coordinate system to obtain the optimized point cloud data in the preliminary coordinate system. This process can be realized by rigid transformation. Exemplarily, if the workpiece is inclined by 15 degrees, the effective point cloud data can be aligned to the standard coordinate system (i.e. the preliminary coordinate system) by a rotation matrix. The distribution of the optimized point cloud data is more regular, and the weld feature changes from side view to front view, which is convenient for subsequent verification. This adjustment can significantly improve the visualization effect and analysis accuracy of the point cloud.

[0114] The contour features are verified through the optimized point cloud data to determine the final spatial coordinate system, i.e., the target coordinate system. The iterative closest point algorithm can be used for verification. Exemplarily, the optimized point cloud data is compared with the standard model (also referred to as the ideal model) of the workpiece for contour features. If the deviation is less than 0.1 millimeter, the preliminary coordinate system is determined as the target coordinate system, otherwise, the foregoing step can be returned to determine the preliminary coordinate system again.

[0115] The above verification ensures the accuracy of the weld position and provides a reliable basis for subsequent robot path planning. The overall process is progressive, which not only improves the usability of point cloud data, but also brings higher precision and stability to automatic welding.

[0116] In some embodiments, S102 can include:

[0117] The weld feature is extracted from the adjusted BIM model to determine the theoretical weld position information;

[0118] The spatial alignment method is used to correct the theoretical weld position information to obtain the corrected theoretical weld position information;

[0119] Based on the corrected theoretical weld position information, the theoretical initial weld path is generated;

[0120] The theoretical initial weld path is verified for path integrity, optimized to adapt to the actual workpiece surface undulations, and optimized to adapt to the robot operation requirements to obtain the theoretical weld path.

[0121] The embodiments of the present application can extract the weld feature from the adjusted BIM model to determine the theoretical weld position information. The edge detection technology can be used to identify the contour of the theoretical weld in the adjusted BIM model to obtain the theoretical weld position information. Exemplarily, on a steel member with a length of 2 meters, the weld feature with a width of 1 centimeter and a depth of 0.5 centimeters is detected by the edge detection technology, so that the weld position information can be determined. When extracting the weld feature, the light angle analysis can also be combined to avoid misjudgment caused by surface reflection. This method relies on the accuracy of the adjusted BIM model and can provide a reliable basis for subsequent correction.

[0122] The embodiment of the application can adopt a spatial alignment method to verify or correct the theoretical weld position information to obtain corrected theoretical weld position information. In a possible implementation manner, the theoretical weld position information can be compared with actual weld position information. If the distance between the theoretical weld position information and the actual weld position information is greater than a preset distance, for example, the preset distance can be 2 mm, the theoretical weld position information needs to be corrected to obtain the corrected theoretical weld position information. The corrected theoretical weld position information can be obtained through spatial projection correction. The actual weld position information can be actual weld position information in optimized point cloud data, or actual weld position information determined by scanning the weld position of a workpiece by a laser scanner, and the like.

[0123] In some possible implementation manners, during the correction of the theoretical weld position information, multi-angle data fusion can also be combined to further improve the accuracy.

[0124] By correcting the theoretical weld position information, the reliability of the weld distribution can be significantly improved, laying a foundation for path planning.

[0125] Based on the corrected theoretical weld position information, a theoretical initial weld path can be generated. For example, for a theoretical weld of a curve, a theoretical initial weld path containing 100 nodes can be generated based on the theoretical weld position information of the curve, and adjacent nodes can be spaced 2 cm apart to preliminarily form a continuous welding track. The theoretical initial weld path directly reflects the spatial characteristics of the weld, which is helpful for subsequent optimization. For example, if the weld has a corner, the node density can be increased at the corner to ensure smooth transition of the path.

[0126] The theoretical initial weld path is verified for path integrity. If the theoretical initial weld path does not cover the full length of the theoretical weld, the weld nodes are supplemented by interpolation to form a path covering the full length of the theoretical weld. For example, in a 3-meter-long weld, it is detected that the theoretical initial weld path misses a 5-centimeter region at the end, and the complete path can be formed by supplementing the nodes by interpolation.

[0127] The coordinates of each node in the theoretical initial weld path can be fine-tuned, for example, the Z-axis coordinate of a node is adjusted from 10 cm to 10.2 cm, and the like, to adapt to the actual workpiece surface undulation.

[0128] The running requirements of the robot can also be considered to determine whether the path curvature in the theoretical initial weld seam path exceeds the capability range of the robot. If so, the path curvature in the theoretical initial weld seam path is adjusted so that the path curvature is within the capability range of the robot, thereby effectively avoiding path interruption or redundancy and improving practicability. It is determined whether the theoretical initial weld seam path can ensure that the welding head of the robot can always maintain the optimal angle with the workpiece. If not, the theoretical initial weld seam path is optimized so that the optimized path can ensure that the welding head of the robot can always maintain the optimal angle with the workpiece, thereby improving welding efficiency and quality.

[0129] The path obtained after the theoretical initial weld seam path is optimized is referred to as a theoretical weld seam path. The theoretical weld seam path obtained after optimization not only fits the weld seam distribution but also adapts to the running requirements of the robot and provides accurate theoretical basis for automatic welding, reducing manual intervention.

[0130] In some embodiments, S103 can include:

[0131] Based on the scanning device, a scanning data stream of the workpiece is obtained, and the scanning data stream is preprocessed based on a vision system to obtain preliminary scanning data;

[0132] Based on the preliminary scanning data, workpiece surface features are extracted, a dynamic change trend of the workpiece surface features is determined, and a dynamic imaging range containing a weld seam is determined according to the dynamic change trend;

[0133] Based on the dynamic imaging range and the optical features, a weld seam region is separated and obtained;

[0134] The weld seam region is subjected to image enhancement and noise processing to obtain a stable weld seam image;

[0135] Weld seam contour extraction is performed according to the stable weld seam image to generate an image containing a weld seam contour;

[0136] The image containing the weld seam contour is subjected to contour verification, and image generation parameters affected by environmental changes are adjusted to generate an actual weld seam image.

[0137] The scanning device can be a laser scanner or a high-resolution camera, etc.

[0138] The scanning device can obtain real-time scanning data stream of the workpiece, and the purpose is to capture the original information of the workpiece surface. Exemplarily, it is assumed that a laser scanner scans a steel member with a length of 1 meter at a speed of 100,000 points per second, and a point cloud data stream containing the height, texture, etc. of the surface of the steel member can be generated.

[0139] The visual system can be a real-time image acquisition and processing module integrating hardware and software, mainly used for dynamic capture, analysis and matching of weld features on the workpiece surface. The hardware part usually includes a high-resolution industrial camera (such as a CCD (Charge-Coupled Device) / CMOS (Complementary Metal Oxide Semiconductor) sensor), an anti-interference active light source (such as structured light or laser-assisted illumination), and an optical filter assembly for stable imaging in a complex welding environment (such as a strong arc light, smoke, etc. environment); the software part is based on image processing algorithms (such as filtering, edge enhancement) or deep learning models (such as convolutional neural networks) to extract the weld contour in real time, obtain the actual weld path, and perform high-precision matching with the theoretical weld path of the BIM model, thereby calculating the positional offset of the actual weld. In addition, the visual system needs to be synchronized with the laser scanner and control equipment to ensure consistent data timing.

[0140] Based on the visual system, the preliminary scanning data can be obtained by pre-processing the scanning data stream, which can include: based on the visual system, the scanning data stream is down-sampled and filtered to remove environmental light interference to obtain preliminary scanning data. Down-sampling can reduce the amount of data, for example, the point cloud density can be reduced from 100 points per square centimeter to 20 points per square centimeter, retaining the main features while reducing the burden of subsequent processing. Filtering out environmental light interference can ensure that the data reflects the true surface state of the workpiece. The above preliminary scanning data can also be referred to as preliminary surface distribution.

[0141] The key to extracting the workpiece surface features based on the preliminary scanning data and determining the dynamic change trend of the workpiece surface features lies in identifying key geometric information. Exemplarily, the surface can be judged to be a plane or a curved surface through curvature analysis, if it is a curved surface, it indicates that there may be an edge or a depression at this place, and there may be a weld, for example, on a curved steel plate, the curvature of a certain area is detected to change from 0 to 0.5, indicating that there may be an edge or a depression at this place; it can also track the change of height value in the scanning process to determine whether the workpiece moves due to vibration, for example, the height of a certain point in the preliminary scanning data corresponding to 10 consecutive frames fluctuates from 5 centimeters to 5.2 centimeters, indicating that the workpiece may have a small displacement due to vibration.

[0142] According to the dynamic change trend of the surface features of the workpiece, the dynamic imaging range containing the weld can be further determined. For example, the region with a height value change exceeding a preset height change threshold can be locked as the dynamic imaging range, or the region with a curvature greater than a preset curvature threshold and a height value change exceeding a preset height change threshold can be locked as the dynamic imaging range, which is helpful for subsequent focusing on the weld. The preset height change threshold and the preset curvature threshold can be determined according to actual needs, for example, the preset height change threshold can be 0.1 cm, and the preset curvature threshold can be 0, 0.1 or 0.2, etc.

[0143] For the dynamic imaging range, the weld region and the surrounding region can be distinguished based on the optical features. Specifically, the weld region and the surrounding region can be distinguished based on the reflection value, and the reflection value of the weld region is lower than that of the surrounding region by a certain value. For example, in a weld region with a width of 8 mm, the reflection value is 20% lower than that of the base material, based on which the weld region can be separated.

[0144] Image enhancement can be performed on the weld region to make the weld details more prominent. For example, a contrast stretching algorithm can be used to expand the gray value range from 50-100 to 0-255, etc. If there is noise in the weld region after image enhancement, such as random bright spots caused by dust, a stable image structure can be obtained through Gaussian filtering smoothing processing, that is, a stable weld image is obtained.

[0145] For the stable weld image, edge detection technology can be used to extract the weld contour to generate an image containing the weld contour. For example, in a weld with a length of 50 cm, the boundary point coordinates are detected to change continuously, thereby forming a clear trajectory, that is, the weld contour.

[0146] The image containing the weld contour can be verified by contour verification, which can verify the influence caused by environmental changes based on the contour data, so as to adjust the image generation parameters affected by the environmental changes to generate an actual weld image. For example, if it is found through contour verification that the light intensity decreases by 10%, the exposure parameters can be adjusted to regenerate the image to ensure the integrity of the contour, thereby generating the actual weld image. The above actual weld image can provide a reliable basis for subsequent welding path planning.

[0147] In some possible implementations, in the process of generating the actual weld image, if it is found that the weld region appears blurred due to surface rust or the like, the recognition accuracy can be enhanced in combination with multispectral data. Thus, the image quality can be ensured effectively in complex environments.

[0148] In some embodiments, in the above S104, the calibration parameters of the robot are determined according to the actual weld image and the theoretical weld path, including:

[0149] The edge distribution data is obtained by identifying the weld edge position in the actual weld image, and the actual weld path is obtained by coordinate conversion on the edge distribution data.

[0150] The offset between the actual weld path and the theoretical weld path is calculated to obtain a preliminary deviation value.

[0151] For the actual area with a preliminary deviation value greater than a preset deviation threshold, interpolation adjustment is performed to obtain adjusted edge distribution data.

[0152] Based on the adjusted edge distribution data, the calibration parameters of the robot are determined by using the least square method.

[0153] In the embodiments of the present application, the first convolutional neural network model can be used to capture the detailed information in the actual weld image, identify the weld edge position in the actual weld image, and obtain the edge distribution data. The first convolutional neural network model can extract the texture and shape features of the weld edge through multiple convolution operations to obtain the edge distribution data. For example, in an image containing a 50 cm long weld, the first convolutional neural network model can identify the gray scale change pattern of the edge region and output an edge probability distribution map, i.e., the edge distribution data. This method can effectively deal with the interference caused by uneven illumination in the image and obtain more accurate edge distribution data.

[0154] The edge distribution data is converted into coordinate points by coordinate conversion, so as to obtain the actual weld path represented by coordinates. The coordinate points can be pixel coordinate points or coordinate points in a target coordinate system, which can be determined according to actual requirements. For example, it is assumed that in the coordinate points corresponding to the edge distribution data of a weld, the horizontal coordinates of a certain continuous boundary point change from 100 pixels to 120 pixels, which indicates that there may be a slight bend at this position.

[0155] The offset between the actual weld path and the theoretical weld path is determined by comparing each coordinate point in the actual weld path with the corresponding coordinate point in the theoretical weld path, and a preliminary deviation value is obtained. For example, if the coordinate of a point in the theoretical weld path is (100 pixels, 50 pixels, 10 pixels), and the coordinate of the corresponding point in the actual weld path is (102 pixels, 51 pixels, 10 pixels), then the offset is a horizontal offset of 2 pixels, a vertical offset of 1 pixel, and a depth offset of 0 pixel, and the preliminary deviation value includes the offset. The quantification of the preliminary deviation value helps subsequent adjustment.

[0156] It should be noted that if the offset between each coordinate point in the actual weld seam path and the corresponding coordinate point in the theoretical weld seam path is the same, the preliminary deviation value can include the offset; if the offsets between each coordinate point in the actual weld seam path and the corresponding coordinate point in the theoretical weld seam path are not completely the same, the preliminary deviation value can include the correspondence between the coordinate points and the offsets, which can be each coordinate point in the actual weld seam path or each coordinate point in the theoretical weld seam path.

[0157] For the actual region where the preliminary deviation value is greater than the preset deviation threshold, the preliminary deviation value greater than the preset deviation threshold specifically means that the horizontal offset, the longitudinal offset or the depth offset included in the offset of the preliminary deviation value is greater than or equal to the preset deviation threshold. The above-mentioned actual region refers to the region corresponding to the preliminary deviation value greater than the preset deviation threshold in the actual weld seam path. The preset deviation threshold can be set according to actual needs, for example, it can be 3 pixels, etc.

[0158] For the above-mentioned actual region, it needs to be corrected. The preferred method is to adjust the edge position by using the interpolation method to obtain the adjusted edge distribution data. Exemplarily, in a weld seam with a length of 10 cm, the horizontal offset in the preliminary deviation value of a certain region is detected to be 3 pixels, and the edge point coordinates are smoothly adjusted from (104, 50, 10) to (101, 50, 10) by linear interpolation to generate the adjusted edge distribution data. Exemplarily, the function corresponding to the linear interpolation can be: x(k)=104-k, y(k)=50, z(k)=10, k=0, 1, 2, 3, k is the index of the point, x(k) is the coordinate of the X axis, y(k) is the coordinate of the Y axis, and z(k) is the coordinate of the Z axis; when k=0, the coordinate is (104, 50, 10), when k=1, the coordinate is (103, 50, 10), when k=2, the coordinate is (102, 50, 10), and when k=3, the coordinate is (101, 50, 10); that is, the points (103, 50, 10), (102, 50, 10) and (101, 50, 10) are added after (104, 50, 10), so that the edge point coordinates are smoothly adjusted from (104, 50, 10) to (101, 50, 10). This processing method can maintain the continuity of the edge.

[0159] Based on the adjusted edge distribution data, the calibration parameters of the robot can be determined by using the least square method to optimize the adjustment process. Specifically, the actual weld path can be re-determined based on the adjusted edge distribution data, and the least square method can be used to fit the relationship between the actual weld path and the theoretical weld path to calculate the adjustment coefficient, which can be used as the calibration parameters of the robot. The calibration parameters can include at least one of a scaling factor, a translation amount (lateral and / or longitudinal translation amount), and a rotation angle. For example, in a 20 cm long weld, the scaling factor is determined to be 0.95 by the least square method, indicating that the actual weld path needs to be slightly scaled to fit the theoretical design, and the corresponding robot needs to adjust its corresponding parameters to fit the adjustment when welding; if the lateral translation amount is determined to be 2 mm by the least square method, the robot needs to move 2 mm and / or rotate the corresponding angle when welding to align the actual weld; if the rotation angle is determined to be 5 degrees by the least square method, the robot needs to rotate 5 degrees and / or move the corresponding distance when welding to align the actual weld; and so on. The above method can significantly improve the calibration accuracy.

[0160] In some possible implementations, the coordinate transformation can be performed on the actual weld image by using the above adjustment coefficient. During the coordinate transformation, each pixel point in the image can be re-mapped according to the coefficient.

[0161] In a possible implementation, the resolution of an original image is 1920x1080, and the adjusted actual weld path is generated after calibration, in which the weld trajectory is closer to the theoretical expectation. When extracting the boundary information from the calibrated actual weld path, it can be determined whether the path is complete by detecting continuity. For example, if the distance between two boundary points suddenly jumps from 1 pixel to 5 pixels, it may indicate that there is a break at this point, which needs to be further checked. When generating a complete weld trajectory (actual weld path) according to the stable path structure, the slight jitter can be eliminated by smoothing filtering.

[0162] In a possible implementation, after applying the mean filter on a 30 cm long weld, the fluctuation between the trajectory points is reduced from 0.2 cm to 0.05 cm, forming a smooth final actual weld path. This smoothing process can provide a more stable reference for subsequent processes.

[0163] It should be noted that if the weld edge is blurred due to surface roughness, the matching accuracy can be enhanced by combining multiple image data. For example, 5 consecutive images are collected, and the reliability of the distribution data is improved by averaging the edge position. This method is particularly useful in dynamic environments and can effectively cope with challenges caused by workpiece displacement or changes in lighting.

[0164] In some embodiments, in S104, the final welding path is determined according to the calibration parameters, including:

[0165] According to the calibration parameters, the pose data of the welding head of the robot is corrected to obtain corrected pose data;

[0166] According to the corrected pose data, the movement trajectory is adjusted and smoothed to obtain smoothed trajectory data;

[0167] According to the smoothed trajectory data, an optimized welding path is generated, and feature distribution data of the optimized welding path is extracted;

[0168] If the feature distribution data does not match the preset feature threshold, the optimized welding path is adjusted based on the least square method to obtain adjusted path parameters;

[0169] According to the adjusted path parameters, a candidate control instruction set is generated, and the smoothed trajectory data is optimized according to the candidate control instruction set to obtain stable trajectory data;

[0170] According to the stable trajectory data, a final welding path is generated.

[0171] According to the calibration parameters determined in the foregoing, the pose data of the welding head of the robot can be corrected, so that the corrected pose data of the welding head is aligned with the actual weld start point. For example, in a straight line weld with a length of 40 cm, if a transverse offset of 2 mm is detected, the welding head angle can be rotated, for example, adjusted by 5 degrees, so that it is aligned with the actual weld. Such adjustment depends on the accurate input of the calibration parameters, which can ensure that the welding head pose is consistent with the actual demand.

[0172] According to the corrected pose data, an interpolation method can be used to adjust and smooth the movement trajectory of the welding head to obtain smoothed trajectory data. For example, in a curved weld, if the data point spacing is 5 cm, intermediate points can be generated by cubic spline interpolation to change the trajectory from a discrete point set to a continuous curve. For example, when a segment of the trajectory is adjusted from the starting point coordinate 10 cm to 15 cm, interpolation can fill in smooth transition points to avoid abrupt changes.

[0173] According to the smoothed trajectory data, an optimized welding path can be generated, and a second convolutional neural network model can be used to extract feature distribution data of the optimized welding path. For example, the curvature change and key turning points can be extracted as feature distribution data. For example, in a weld, an arc feature with a radius of 3 cm is detected, which can be used as feature distribution data. The feature distribution data extracted here can provide a basis for subsequent optimization.

[0174] The preset feature threshold can be determined based on the theoretical weld path, and different feature distribution data can correspond to different preset feature thresholds. The feature distribution data not matching the preset feature threshold can be understood as the difference between the feature distribution data and the preset feature threshold being greater than a certain value. If the feature distribution data does not match the preset feature threshold, the optimized welding path is adjusted based on the least square method to obtain adjusted path parameters, so that the adjusted path parameters are consistent with or have a small difference from the theoretical weld path. For example, in a 30 cm long weld, if a certain segment of feature points in the feature distribution data deviates from the corresponding preset feature threshold by more than 2 mm, the actual points (optimized welding path) and the theoretical weld path can be fitted by the least square method to obtain adjustment parameters, which adjust the optimized welding path to obtain adjusted path parameters. The adjusted path parameters can reduce the deviation to within 0.5 mm. This method improves the matching degree of the optimized welding path through data optimization.

[0175] According to the adjusted path parameters, a candidate control instruction set for the robot can be generated, which can be a control instruction set obtained by optimizing the initial control instruction set according to the adjusted path parameters. For example, in a complex weld, the initial control instruction set can include 100 control instructions, and after adjustment, 120 control instructions are generated through parameter optimization, which are more delicate and have stronger adaptability. The update of such an instruction set directly affects the welding precision.

[0176] When the smoothed trajectory data is optimized according to the candidate control instruction set, the filtering method can further stabilize the data to obtain stable trajectory data. For example, in a 50 cm long weld, Gaussian filtering is applied to the smoothed trajectory data to remove 0.3 mm fluctuations caused by noise, making the trajectory smoother. This processing provides a reliable basis for subsequent execution. Finally, the final welding path can be generated according to the stable trajectory data.

[0177] It should be noted that if the distance between certain trajectory points suddenly increases from 1 mm to 4 mm, it can be judged as a potential breakpoint, and intermediate points need to be supplemented. For example, in a 25 cm long weld, the integrity of the final welding path is significantly improved after supplementing the points, providing a stable reference for the welding process.

[0178] In the embodiments of the present application, these steps are gradually advanced, from pose adjustment to final path generation, forming a set of rigorous logic chains. The optimization of each link provides support for the next step, such as smoothing the trajectory to reduce the risk of welding head jitter, and feature extraction to enhance the adaptability of the path. This method can effectively improve the stability and consistency of the welding process in practical applications, especially in complex trajectory scenarios.

[0179] In a possible implementation, if the weld surface condition is variable, multiple sets of trajectory data can be combined to enhance stability. For example, the final path is generated by taking the average of the trajectory distribution collected for three consecutive times. This extended scheme further enriches the robustness of the scheme and ensures the applicability in dynamic environment.

[0180] In some embodiments, in S105, according to the final welding path and the material and thickness of the workpiece, the final control instruction set is determined, including:

[0181] The key point coordinates in the final welding path are extracted, and coordinate feature data of the key point coordinates are determined;

[0182] Based on the preset process parameter database, the initial process parameters are determined according to the coordinate feature data and the material and thickness of the workpiece;

[0183] The initial process parameters are filtered to obtain smoothed process parameters;

[0184] If the smoothed process parameters exceed the preset process parameter threshold, the smoothed process parameters are corrected to obtain corrected process parameters;

[0185] The stability feature of the corrected process parameters is determined, and the final welding path is adjusted according to the stability feature and the key point coordinates to obtain an adjusted welding path;

[0186] The final control instruction set is generated according to the adjusted welding path and the corrected process parameters.

[0187] When the key point coordinates in the final welding path are extracted, the geometric characteristics of the final welding path can be analyzed. For example, in a straight line weld with a length of 30 cm, suppose a key point coordinate is set every 5 cm, 6 key point coordinates can be extracted. The distribution of these key point coordinates reflects the linear feature of the final welding path. If the final welding path is a curve, the turning points can be selected according to the curvature change, for example, the vertex coordinates of an arc region with a radius of 4 cm can be used as key point coordinates.

[0188] When the coordinate feature data of the key point coordinates is determined, statistical methods can be used to calculate the point spacing and angle change data, and then the coordinate feature data is determined. The coordinate feature data can be stable or unstable, etc. For example, in a 20 cm long weld, if the average point spacing is 3 cm and the angle change is less than 10 degrees, it indicates that the distribution is uniform, and the coordinate feature data can be marked as stable, otherwise it is unstable.

[0189] The preset process parameter database can store the correspondence between the coordinate feature data, the workpiece material, the workpiece thickness, and the process parameters of the robot. Thus, based on the preset process parameter database, the coordinate feature data of the key point coordinates and the process parameters corresponding to the material and thickness of the workpiece can be obtained as initial process parameters. The initial process parameters can include current parameters and speed parameters, etc. For example, if the coordinate feature data is stable, the material is stainless steel, and the thickness is 3 mm, the preset process parameter database can return the corresponding initial process parameters, such as the initial process parameters can be a current range of 100-150 amperes, a speed range of 20-30 cm / min, etc. For a steel material with a thickness of 5 mm, the database can recommend that the current be increased to 180 amperes and the speed be reduced to 15 cm / min to meet the heat input requirement.

[0190] The initial process parameters are filtered to process data fluctuations, and smoothed process parameters can be obtained. For example, in a 40 cm long weld, if the current data of a certain section fluctuates between 120-130 amperes, it can be stabilized at 125 amperes after filtering, reducing the instability of the robot during execution.

[0191] The preset process parameter threshold can be an upper limit of the process parameter, or a normal range of the process parameter. If the smoothed process parameter exceeds the preset process parameter threshold, the smoothed process parameter is corrected to obtain a corrected process parameter, and the corrected process parameter does not exceed the preset process parameter threshold. The least squares method can be used to correct the smoothed process parameter. For example, assuming that the upper limit of the current is 140 amperes, if the current process parameter exceeds 140 amperes, it needs to be corrected to be less than 140 amperes; assuming that the speed is 32 cm / min and the corresponding threshold is 28 cm / min, it is adjusted to 26 cm / min by the least squares method to ensure that the parameter is reasonable and does not exceed the limit.

[0192] The stability feature of the corrected process parameter can be determined by focusing on the continuity of the data. The stability feature can be high stability, low stability, or instability, etc. For example, in a 25 cm long weld, if the current distribution changes smoothly from beginning to end, the stability feature can be marked as high stability, if the points of the current distribution that are not smooth are greater than 0 and less than a second preset number, the stability feature can be low stability, and if the points of the current distribution that are not smooth are more than the second preset number, the stability feature can be instability. The second preset number is greater than 0 and can be set according to actual needs, which is not limited here, for example, it can be 2 or 3, etc. The stability feature can also be determined according to the speed distribution.

[0193] For the stability features of low stability and instability, the final welding path can be adjusted based on the key point coordinates to obtain an adjusted welding path, so that the stability feature finally determined based on the adjusted welding path is high stability.

[0194] According to the adjusted welding path and the corrected process parameters determined through the adjusted welding path, a final control instruction set can be generated. The instruction density in the final control instruction set can be large to refine the control of the robot. For example, the control instruction can be refined from one control instruction per 5 centimeters to one control instruction per 3 centimeters. The final control instruction set can bind the corrected process parameters to the adjusted welding path, and each control instruction can include corresponding path coordinates and process parameters.

[0195] The control instructions in the final control instruction set can be arranged in chronological order. For example, in a 50-centimeter-long weld, the final control instruction set can include 150 control instructions covering all actions from the starting point to the ending point to obtain stable execution data. This method forms a complete process from key point coordinate extraction to parameter adjustment and then to instruction generation, ensuring the reliability and adaptability of the welding process.

[0196] In some possible implementations, controlling the robot to perform automatic welding according to the final control instruction set can include: generating a robot execution program according to the final control instruction set, and transmitting the program to the corresponding robot through a general interface protocol, so as to control the robot to perform automatic welding according to the program.

[0197] For example, the robot execution program is a program recognizable by the robot, and the robot execution program containing starting and ending points and process parameters can be generated through a preset template.

[0198] The program transmission is completed through a general interface protocol, such as TCP / IP. The program can be transmitted to the corresponding robot in the form of a data packet. It should be noted that the protocol needs to ensure that there is no packet loss during transmission, and the robot returns an acknowledgement signal after receiving the program, for example, a “data complete” status code.

[0199] In some embodiments, after the robot is controlled to perform automatic welding in S105, the method further includes:

[0200] obtaining feedback data of the robot;

[0201] determining whether the welding quality of the robot is qualified based on the feedback data;

[0202] If the welding quality of the robot is not qualified, the final control instruction set is updated according to the feedback data, and the robot is controlled to perform automatic welding according to the updated final control instruction set.

[0203] During the automatic welding process, the robot can collect real-time current and speed values of the robot through sensors as feedback data, which is fed back to the control device. For example, during the welding process, the feedback data of a certain section fluctuates between 118-122 amperes, and the speed is stable at 24-26 cm / min.

[0204] In some possible implementations, the above determining whether the welding quality of the robot is qualified based on the feedback data can include:

[0205] determining whether the feedback data has abnormal fluctuations, and if so, determining a fluctuation range distribution;

[0206] if the proportion of abnormal fluctuations in the fluctuation range distribution exceeds a preset proportion, adjusting the feedback data by a filtering algorithm to obtain smoothed feedback data;

[0207] determining whether the welding quality of the robot is qualified based on the smoothed feedback data.

[0208] For example, whether there is abnormal fluctuation can be determined according to whether the fluctuation range of the feedback data exceeds the normal fluctuation range. For example, if the fluctuation range of the current in a certain section of the feedback data exceeds the normal fluctuation range of ±5 amperes, for example, the current is 115-125 amperes, it is determined that the feedback data in this section has abnormal fluctuation. Whether there is abnormal fluctuation can also be determined by speed fluctuation, which is not described again.

[0209] If there is abnormal fluctuation, a statistical method can be used to draw a fluctuation range distribution. If the proportion of abnormal fluctuations in the fluctuation range distribution exceeds a preset proportion, a filtering algorithm, such as a moving average method, is used to adjust the feedback data to obtain smoothed feedback data. The preset proportion can be set according to actual needs, such as 8%, etc.

[0210] For example, if 90% of the current fluctuations of the points in the fluctuation range distribution are within ±5 amperes, and 10% of the current fluctuations of the points exceed ±5 amperes, the moving average method is used to adjust the current data that fluctuates beyond ±5 amperes to within ±5 amperes to improve data flow consistency.

[0211] Then, based on the smoothed feedback data, it is determined whether the welding quality of the robot is qualified, i.e., whether it meets the standard.

[0212] If the feedback data does not have abnormal data, the welding quality of the robot is directly determined based on the feedback data.

[0213] When determining whether the welding quality of a robot is acceptable, it is necessary to pay attention to the matching degree of current and speed, comparing them with the speed and current in the preset process parameter database to determine whether the deviation between the two is within a certain proportion. For example, if the speed is 25 cm / min, the current is stable at 121 amperes, and the deviation from the reference value in the preset process parameter database is less than 5%, then the welding quality can be determined to be acceptable; otherwise, the welding quality is determined to be unacceptable.

[0214] Feedback data indicating substandard welding quality can be marked as requiring optimization, facilitating subsequent adjustments.

[0215] If the robot's welding quality is unsatisfactory, the final control instruction set can be updated based on feedback data to ensure that the updated final control instruction set controls the robot to achieve satisfactory welding quality during automatic welding. Specifically, the final control instruction set can be adjusted based on the deviation between the two, for example, increasing the speed from 25 cm / min to 27 cm / min, etc.

[0216] The updated final control instruction set is sent to the corresponding robot using a transmission protocol, and the robot returns a status feedback that "parameters have been updated".

[0217] The robot's feedback data on current and speed is retrieved again. If this feedback data shows no abnormal fluctuations, the robot is considered to have good compatibility and can continue to operate stably for a long time using the updated final control instruction set. This method improves welding consistency by optimizing parameters through feedback.

[0218] In one possible implementation, for complex weld seams such as curved paths, feedback data may show localized speed unevenness, for example, a drop in speed from 25 cm / min to 22 cm / min in one section. After filtering, smoothing, and updating the program, the speed distribution tends to be more uniform, reducing frequent pauses during equipment adjustments. Understandably, this method enhances the robot's adaptability to dynamic processes and effectively ensures execution accuracy.

[0219] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0220] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0221] Figure 2 A schematic diagram of the structure of the automatic welding device for industrial robots provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below:

[0222] likeFigure 2 The industrial robot automatic welding device 2 shown in the figure includes an alignment module 21, a theoretical weld path acquisition module 22, an actual weld image acquisition module 23, a final welding path determination module 24, and a welding module 25.

[0223] The alignment module 21 is configured to obtain three-dimensional point cloud data of the workpiece, and adjust a pre-constructed BIM model of the workpiece according to the three-dimensional point cloud data, so that the adjusted BIM model is aligned with the actual workpiece position.

[0224] The theoretical weld path acquisition module 22 is configured to acquire a theoretical weld path from the adjusted BIM model.

[0225] The actual weld image acquisition module 23 is configured to acquire an actual weld image of the workpiece.

[0226] The final welding path determination module 24 is configured to determine calibration parameters of the robot according to the actual weld image and the theoretical weld path, and determine a final welding path according to the calibration parameters.

[0227] The welding module 25 is configured to determine a final control instruction set according to the final welding path and the material and thickness of the workpiece, and control the robot to perform automatic welding according to the final control instruction set.

[0228] In a possible implementation, the theoretical weld path acquisition module 22 is specifically configured to:

[0229] extract weld features from the adjusted BIM model to determine theoretical weld position information;

[0230] correct the theoretical weld position information by using a spatial alignment method to obtain corrected theoretical weld position information;

[0231] generate a theoretical initial weld path based on the corrected theoretical weld position information;

[0232] perform path integrity verification and optimization, optimization to adapt to actual workpiece surface undulations, and optimization to adapt to robot operation requirements on the theoretical initial weld path to obtain a theoretical weld path.

[0233] In a possible implementation, the actual weld image acquisition module 23 is specifically configured to:

[0234] acquire a scanning data stream of the workpiece based on a scanning device, and pre-process the scanning data stream based on a vision system to obtain preliminary scanning data;

[0235] extract workpiece surface features based on the preliminary scanning data, determine a dynamic change trend of the workpiece surface features, and determine a dynamic imaging range containing a weld seam according to the dynamic change trend.

[0236] Separating the weld region based on dynamic imaging range and optical characteristics;

[0237] Image enhancement and noise processing are performed on the weld region to obtain a stable weld image;

[0238] According to the stable weld image, the weld contour is extracted to generate an image containing the weld contour;

[0239] The image containing the weld contour is verified, and the image generation parameters affected by environmental changes are adjusted to generate an actual weld image.

[0240] In a possible implementation, in the final welding path determination module 24, according to the actual weld image and the theoretical weld path, the calibration parameters of the robot are determined, including:

[0241] The weld edge position in the actual weld image is identified to obtain edge distribution data, and the edge distribution data is coordinate-converted to obtain an actual weld path;

[0242] The offset of the actual weld path and the theoretical weld path is calculated to obtain a preliminary deviation value;

[0243] For the actual region whose preliminary deviation value is greater than a preset deviation threshold, interpolation adjustment is performed to obtain adjusted edge distribution data;

[0244] Based on the adjusted edge distribution data, the least square method is used to determine the calibration parameters of the robot.

[0245] In a possible implementation, in the final welding path determination module 24, the final welding path is determined according to the calibration parameters, including:

[0246] According to the calibration parameters, the pose data of the welding head of the robot is corrected to obtain corrected pose data;

[0247] According to the corrected pose data, the smooth movement trajectory is adjusted and smoothed to obtain smoothed trajectory data;

[0248] According to the smoothed trajectory data, an optimized welding path is generated, and feature distribution data of the optimized welding path is extracted;

[0249] If the feature distribution data does not match a preset feature threshold, the optimized welding path is adjusted based on the least square method to obtain adjusted path parameters;

[0250] According to the adjusted path parameters, a candidate control instruction set is generated, and the smoothed trajectory data is optimized according to the candidate control instruction set to obtain stable trajectory data;

[0251] The final welding path is generated according to the stable trajectory data.

[0252] In a possible implementation, in the welding module 25, the final control instruction set is determined according to the final welding path and the material and thickness of the workpiece, and includes:

[0253] The key point coordinates in the final welding path are extracted, and coordinate feature data of the key point coordinates are determined;

[0254] Based on the preset process parameter database, the initial process parameters are determined according to the coordinate feature data and the material and thickness of the workpiece;

[0255] The initial process parameters are filtered to obtain smoothed process parameters;

[0256] If the smoothed process parameters exceed the preset process parameter threshold, the smoothed process parameters are corrected to obtain corrected process parameters;

[0257] The stability feature of the corrected process parameters is determined, and the final welding path is adjusted according to the stability feature and the key point coordinates to obtain an adjusted welding path;

[0258] The final control instruction set is generated according to the adjusted welding path and the corrected process parameters.

[0259] In a possible implementation, in the alignment module 21, the BIM model of the workpiece pre-constructed is adjusted according to the three-dimensional point cloud data, and includes:

[0260] The optimized point cloud data and a target coordinate system reflecting the actual posture of the workpiece are determined according to the three-dimensional point cloud data;

[0261] Based on the target coordinate system, the deviation data of the optimized point cloud data and the BIM model are determined, and a position transformation matrix is determined according to the deviation data;

[0262] The BIM model is adjusted according to the position transformation matrix to obtain an adjusted BIM model.

[0263] In a possible implementation, in the alignment module 21, the optimized point cloud data and a target coordinate system reflecting the actual posture of the workpiece are determined according to the three-dimensional point cloud data, and include:

[0264] The three-dimensional point cloud data is subjected to noise removal to obtain effective point cloud data of the surface of the workpiece;

[0265] The effective point cloud data is subjected to clustering segmentation to determine the position information of the workpiece in the effective point cloud data;

[0266] Based on the position information of the workpiece, the boundary points of the contour feature in the effective point cloud data are determined.

[0267] Determine a preliminary coordinate system reflecting the actual pose of the workpiece based on the boundary points;

[0268] Perform coordinate conversion on the effective point cloud data based on the preliminary coordinate system to obtain optimized point cloud data;

[0269] Verify the contour features of the optimized point cloud data, and take the preliminary coordinate system that passes the verification as a target coordinate system.

[0270] In a possible implementation, after the robot is controlled to perform automatic welding in the welding module 25, the method further includes:

[0271] Obtaining feedback data of the robot;

[0272] Determining whether the welding quality of the robot is qualified based on the feedback data;

[0273] If the welding quality of the robot is not qualified, updating the final control instruction set according to the feedback data, and controlling the robot to perform automatic welding according to the updated final control instruction set.

[0274] The embodiment of the application also provides a control device including a memory and a processor, the memory stores a computer program, and the processor implements the industrial robot automatic welding method in the above method embodiment when executing the computer program. Exemplarily, the control device can be a controller, etc., which is not limited here.

[0275] The embodiment of the application also provides a robot including the above control device.

[0276] The embodiment of the application also provides a computer readable storage medium storing a computer program, and the computer program is executed by a processor to implement the industrial robot automatic welding method in the above method embodiment.

[0277] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in a certain embodiment can be referred to the relevant description of other embodiments. If there is no special description and no logical conflict, the terms and / or descriptions of different embodiments are consistent and can be mutually referenced, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationship.

[0278] The above-described embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. An automatic welding method for industrial robots, characterized in that, include: Acquire the three-dimensional point cloud data of the workpiece, and adjust the pre-built BIM model of the workpiece according to the three-dimensional point cloud data so that the adjusted BIM model is aligned with the actual position of the workpiece. Obtain the theoretical weld path from the adjusted BIM model; Obtain an image of the actual weld seam of the workpiece; Based on the actual weld image and the theoretical weld path, the robot's calibration parameters are determined, and the final welding path is determined based on the calibration parameters. Based on the final welding path and the material and thickness of the workpiece, a final control instruction set is determined, and the robot is controlled to perform automatic welding according to the final control instruction set. The step of obtaining the actual weld image of the workpiece includes: Based on the scanning device, the scanning data stream of the workpiece is acquired, and based on the vision system, the scanning data stream is preprocessed to obtain preliminary scanning data; Based on the preliminary scanning data, the surface features of the workpiece are extracted, the dynamic change trend of the surface features of the workpiece is determined, and the dynamic imaging range including the weld is determined according to the dynamic change trend. Based on the dynamic imaging range and optical characteristics, the weld seam region is separated. Image enhancement and noise reduction are performed on the weld area to obtain a stable weld image; Based on the stable weld image, the weld contour is extracted to generate an image containing the weld contour. The image containing the weld contour is verified, and the image generation parameters affected by environmental changes are adjusted to generate the actual weld image.

2. The automatic welding method for industrial robots according to claim 1, characterized in that, The step of obtaining the theoretical weld path from the adjusted BIM model includes: Weld features are extracted from the adjusted BIM model to determine the theoretical weld location information; The theoretical weld position information is corrected using a spatial alignment method to obtain the corrected theoretical weld position information; Based on the corrected theoretical weld position information, a theoretical initial weld path is generated; The theoretical initial weld path is verified and optimized for path integrity, adapted to the surface undulations of the actual workpiece, and adapted to the operational requirements of the robot, thus obtaining the theoretical weld path.

3. The automatic welding method for industrial robots according to claim 1, characterized in that, The step of determining the robot's calibration parameters based on the actual weld image and the theoretical weld path includes: Identify the weld edge position in the actual weld image, obtain edge distribution data, and perform coordinate transformation on the edge distribution data to obtain the actual weld path; Calculate the offset between the actual weld path and the theoretical weld path to obtain a preliminary deviation value; For actual areas where the initial deviation value is greater than the preset deviation threshold, interpolation adjustment is performed to obtain the adjusted edge distribution data; Based on the adjusted edge distribution data, the calibration parameters of the robot are determined using the least squares method.

4. The automatic welding method for industrial robots according to claim 1, characterized in that, Determining the final welding path based on the calibration parameters includes: Based on the calibration parameters, the posture data of the robot's welding head is corrected to obtain the corrected posture data. Based on the corrected attitude data, the movement trajectory is adjusted and smoothed to obtain smoothed trajectory data; An optimized welding path is generated based on the smoothed trajectory data, and the feature distribution data of the optimized welding path is extracted. If the feature distribution data does not match the preset feature threshold, the optimized welding path is adjusted based on the least squares method to obtain the adjusted path parameters. Based on the adjusted path parameters, a candidate control instruction set is generated, and the smoothed trajectory data is optimized based on the candidate control instruction set to obtain stable trajectory data; The final welding path is generated based on the stable trajectory data.

5. The automatic welding method for industrial robots according to claim 1, characterized in that, The step of determining the final control instruction set based on the final welding path and the material and thickness of the workpiece includes: Extract the coordinates of key points in the final welding path and determine the coordinate feature data of the key point coordinates; Based on a preset process parameter database, initial process parameters are determined according to the coordinate feature data and the material and thickness of the workpiece. The initial process parameters are filtered to obtain smoothed process parameters; If the smoothed process parameters exceed the preset process parameter threshold, the smoothed process parameters are corrected to obtain the corrected process parameters. Determine the stability characteristics of the modified process parameters, and adjust the final welding path based on the stability characteristics and the key point coordinates to obtain the adjusted welding path; Based on the adjusted welding path and the corrected process parameters, a final control instruction set is generated.

6. The automatic welding method for industrial robots according to claim 1, characterized in that, The adjustment of the pre-built BIM model of the workpiece based on the three-dimensional point cloud data includes: Based on the three-dimensional point cloud data, the optimized point cloud data and the target coordinate system reflecting the actual posture of the workpiece are determined; Based on the target coordinate system, the deviation data between the optimized point cloud data and the BIM model is determined, and the position transformation matrix is ​​determined according to the deviation data; The BIM model is adjusted according to the position transformation matrix to obtain the adjusted BIM model.

7. The automatic welding method for industrial robots according to claim 6, characterized in that, The step of determining the optimized point cloud data and the target coordinate system reflecting the actual posture of the workpiece based on the three-dimensional point cloud data includes: Noise removal is performed on the three-dimensional point cloud data to obtain the effective point cloud data of the workpiece surface. Clustering and segmentation are performed on the effective point cloud data to determine the position information of the workpiece in the effective point cloud data; Based on the position information of the workpiece, the boundary points of the contour features in the effective point cloud data are determined; Based on the boundary points, a preliminary coordinate system reflecting the actual posture of the workpiece is determined; Based on the initial coordinate system, coordinate transformation is performed on the effective point cloud data to obtain optimized point cloud data; The optimized point cloud data is subjected to contour feature verification, and the preliminary coordinate system that passes the verification is used as the target coordinate system.

8. The automatic welding method for industrial robots according to any one of claims 1 to 7, characterized in that, After the robot performs automated welding, the process also includes: Obtain the feedback data from the robot; Based on the feedback data, it is determined whether the welding quality of the robot is up to standard; If the welding quality of the robot is unqualified, the final control instruction set is updated according to the feedback data, and the robot is controlled to perform automatic welding according to the updated final control instruction set.

9. A control device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the industrial robot automatic welding method as described in any one of claims 1 to 8.

Images