A welding method, system, apparatus, and storage medium based on coordinate system registration

Abstract

Some embodiments of the specification provide a welding method based on coordinate system registration, the method comprising: acquiring, by a sensor, a sequence of spatial coordinate points of a teaching welding path and teaching point cloud data of a standard workpiece; generating, by a curve fitting algorithm, a standard welding path based on the sequence of spatial coordinate points; determining a reference point based on the teaching point cloud data and the sequence of spatial coordinate points; constructing a welding database based on a first workpiece identifier of the standard workpiece, the reference point, and the standard welding path; acquiring a second workpiece identifier and coarse scanning point cloud data of a workpiece to be welded; acquiring, from the welding database, the teaching point cloud data, the reference point, and the standard welding path corresponding to the second workpiece identifier; determining a first registration transformation based on the teaching point cloud data and the coarse scanning point cloud data; determining a scanning range based on the reference point and the first registration transformation; acquiring fine scanning point cloud data based on the scanning range; and determining a welding instruction according to the fine scanning point cloud data and the standard welding path.

Description

Technical Field

[0001] This specification relates to the field of welding technology, and in particular to a welding method, system, apparatus and storage medium based on coordinate system registration. Background Technology

[0002] In the manufacturing of machinery in industries such as shipbuilding, water conservancy, papermaking, transportation, and coal mining, the types and combinations of products are often complex and diverse, resulting in consistently high demands on workers' technical skills and manpower. Furthermore, intelligent welding systems should not only prioritize ease of one-click start or high welding pass rates in offline simulation, but should also maximize both operational convenience and welding pass rates. PCS replacement applications do not require importing product models, nor do they require lengthy offline simulations, and they have lower requirements for tooling.

[0003] While small and medium-sized assemblies require high precision in material cutting and assembly, their weld seams are complex and diverse. Offline simulation functions need to process the model, and path simulation takes a long time. This not only requires time to process the product model and has a long path simulation cycle, but also demands high professional skills from operators, making it difficult to adapt to small-batch production scenarios. At the same time, existing model import technologies also have significant shortcomings: model alignment operations are highly dependent on manual work, with low automation, resulting in difficulties in effectively guaranteeing both operational efficiency and alignment accuracy.

[0004] Therefore, there is an urgent need to provide a welding method, system, device, and storage medium based on coordinate system registration to achieve automation and high precision in welding operations. Summary of the Invention

[0005] One embodiment of this specification provides a welding method based on coordinate system registration. The method includes: acquiring a spatial coordinate point sequence and teaching point cloud data of a teaching welding path for a standard workpiece using a sensor; generating a standard welding path based on the spatial coordinate point sequence using a curve fitting algorithm; determining a reference point based on the teaching point cloud data and the spatial coordinate point sequence; constructing a welding database based on a first workpiece identifier of the standard workpiece, the reference point, and the standard welding path; acquiring a second workpiece identifier and coarse scan point cloud data of the workpiece to be welded; acquiring teaching point cloud data, the reference point, and the standard welding path corresponding to the second workpiece identifier from the welding database; determining a first registration transformation based on the teaching point cloud data and the coarse scan point cloud data; determining a scanning range based on the reference point and the first registration transformation; acquiring fine scan point cloud data based on the scanning range; and determining a welding command based on the fine scan point cloud data and the standard welding path.

[0006] One embodiment of this specification provides a welding system based on coordinate system registration. The system includes a first acquisition module, a generation module, a reference point determination module, a database construction module, a second acquisition module, a database access module, a transformation determination module, a fine scanning module, and an instruction determination module. The first acquisition module is configured to acquire a spatial coordinate point sequence and teaching point cloud data of a teaching welding path for a standard workpiece using sensors. The generation module is configured to generate a standard welding path based on the spatial coordinate point sequence using a curve fitting algorithm. The reference point determination module is configured to determine reference points based on the teaching point cloud data and the spatial coordinate point sequence. The database construction module is... The system is configured to construct a welding database based on the first workpiece identifier of the standard workpiece, the reference point, and the standard welding path; the second acquisition module is configured to acquire the second workpiece identifier of the workpiece to be welded and the coarse scan point cloud data of the workpiece to be welded; the database access module is configured to acquire the teaching point cloud data corresponding to the second workpiece identifier, the reference point, and the standard welding path from the welding database; the fine scanning module is configured to determine the scanning range based on the reference point and the first registration transformation, and acquire the fine scan point cloud data based on the scanning range; and the instruction determination module is configured to determine the welding instruction based on the fine scan point cloud data and the standard welding path.

[0007] One embodiment of this specification provides a welding apparatus based on coordinate system registration, the apparatus including at least one processor and at least one memory; the at least one memory is used to store computer instructions; the at least one processor is used to execute at least a portion of the computer instructions to implement a welding method based on coordinate system registration.

[0008] This specification provides one or more embodiments of a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions from the storage medium, the computer executes a welding method based on coordinate system registration. Attached Figure Description

[0009] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0010] Figure 1 This is a block diagram of a coordinate-system-registered welding system according to some embodiments of this specification;

[0011] Figure 2 This is an exemplary flowchart of a welding method based on coordinate system registration, as shown in some embodiments of this specification.

[0012] Figure 3 This is an exemplary flowchart illustrating the determination of a reference point according to some embodiments of this specification;

[0013] Figure 4 This is an exemplary flowchart illustrating the determination of the first registration transformation according to some embodiments of this specification;

[0014] Figure 5 This is another exemplary flowchart illustrating the determination of the first registration transformation according to some embodiments of this specification. Detailed Implementation

[0015] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Figure 1 This is an exemplary block diagram of a coordinate-system-registered welding system according to some embodiments of this specification. In some embodiments, the coordinate-system-registered welding system 100 (hereinafter referred to as system 100) may include a first acquisition module 110, a generation module 120, a reference point determination module 130, a database construction module 140, a second acquisition module 150, a database access module 160, a transformation determination module 170, a fine scanning module 180, and an instruction determination module 190.

[0016] In some embodiments, the first acquisition module 110 is configured to acquire the spatial coordinate point sequence and teaching point cloud data of the teaching welding path of a standard workpiece through a sensor.

[0017] In some embodiments, the generation module 120 is configured to generate a standard welding path based on a sequence of spatial coordinate points using a curve fitting algorithm.

[0018] In some embodiments, the reference point determination module 130 is configured to determine a reference point based on the teaching point cloud data and the spatial coordinate point sequence.

[0019] In some embodiments, the reference point determination module 130 is further configured to: determine multiple key points and their corresponding point features based on the teaching point cloud data; determine a first candidate key point based on the multiple key points and their corresponding point features; and determine a reference point based on the spatial coordinate point sequence and the first candidate key point.

[0020] In some embodiments, the reference point determination module 130 is further configured to: divide the spatial coordinate point sequence according to a preset cutting window to obtain multiple sub-welding segments; and determine a reference point based on the first candidate key point and the multiple sub-welding segments.

[0021] In some embodiments, the database construction module 140 is configured to construct a welding database based on a first workpiece identifier, a reference point, and a standard welding path of a standard workpiece.

[0022] In some embodiments, the second acquisition module 150 is configured to acquire a second workpiece identifier of the workpiece to be welded and coarse scan point cloud data of the workpiece to be welded.

[0023] In some embodiments, the database access module 160 is configured to obtain teaching point cloud data, reference points, and standard welding paths corresponding to the second workpiece identifier from the welding database.

[0024] In some embodiments, the transformation determination module 170 is configured to determine a first registration transformation based on the taught point cloud data and the coarse scan point cloud data.

[0025] In some embodiments, the welding database also includes a first workpiece profile corresponding to a first workpiece identifier, the first workpiece profile being generated based on the teaching point cloud data; the transformation determination module is further configured to: determine a second workpiece profile based on the coarse scan point cloud data when the amount of data in the teaching point cloud data and / or the amount of data in the coarse scan point cloud data is greater than a third threshold; and determine a first registration transformation based on the second workpiece profile and the first workpiece profile.

[0026] In some embodiments, the welding database further includes a teaching coarse matching point set; the transformation determination module is further configured to: in response to the data volume of the teaching point cloud data and / or the data volume of the coarse scan point cloud data being greater than a fourth threshold, determine the actual coarse matching point set of the workpiece to be welded based on the teaching coarse matching point set and the coarse scan point cloud data; and determine a first registration transformation based on the teaching coarse matching point set and the actual coarse matching point set; wherein the fourth threshold is greater than a third threshold.

[0027] In some embodiments, the fine scanning module 180 is configured to determine the scanning range based on a reference point and a first registration transformation, and to acquire fine scanning point cloud data based on the scanning range.

[0028] In some embodiments, the instruction determination module 190 is configured to determine welding instructions based on fine scan point cloud data and standard welding paths.

[0029] It should be noted that the above description of the coordinate system-based welding system is for ease of description only and should not be construed as limiting this specification to the embodiments described. It is understood that those skilled in the art, after understanding the principles of this system, may arbitrarily combine the various modules or construct subsystems connected to other modules without departing from these principles. For example, the modules may share a single storage module, or each module may have its own independent storage module.

[0030] Figure 2This is an exemplary flowchart of a coordinate-system-register-based welding method according to some embodiments of this specification. In some embodiments, process 200 may be executed by a coordinate-system-register-based welding system 100 or a processor. Figure 2 As shown, process 200 may include the following steps 210-2010:

[0031] Step 210: Use sensors to acquire the spatial coordinate point sequence and teaching point cloud data of the teaching welding path of the standard workpiece.

[0032] In some embodiments, the sensor may include a 3D vision sensor (such as a structured light scanner, a laser scanner, etc.) or other devices for acquiring spatial data or location information.

[0033] A standard workpiece refers to a workpiece with a preset standard size and shape. For example, a standard workpiece is a sample workpiece that is the same model as the part to be welded and has a preset size and shape. The part to be welded refers to the actual physical workpiece or component that needs to be welded. In some embodiments, the standard workpiece may be preset by those skilled in the art based on experience.

[0034] The taught welding path refers to the ideal welding trajectory that the welding torch tip should follow when welding a standard workpiece during the teaching phase. For example, the taught welding path is a path formed by a series of spatial points recorded by manually guiding the robotic welding torch along the weld seam of a standard workpiece. The teaching phase refers to the process by which the operator "demonstrates" the welding path and parameters to the robot or welding system. The sequence of spatial coordinate points can be three-dimensional coordinate points recorded by the robotic welding torch tip during the taught welding process.

[0035] In some embodiments, the system 100 can control a robot to carry a welding torch along a preset weld seam, while simultaneously recording a sequence of trajectory points at the end of the welding torch via a robot motion axis encoder, serving as a sequence of spatial coordinate points for teaching the welding path of a standard workpiece. In some embodiments, the system 100 can also acquire the spatial coordinate point sequence in other ways, such as through an infrared tracking system or a vision-based tracking system.

[0036] Teaching point cloud data refers to a set of three-dimensional points used to represent the geometric features of a standard workpiece surface during the teaching phase. In some embodiments, the system 100 can generate a set of three-dimensional points after performing a fine scan of the standard workpiece using a high-precision 3D vision sensor, which serves as the teaching point cloud data. The teaching point cloud data can be used as a reference template for subsequent registration.

[0037] Step 220: Based on the spatial coordinate point sequence, a standard welding path is generated using a curve fitting algorithm.

[0038] Curve fitting algorithms can include B-spline fitting, Bézier curve fitting, polynomial fitting, etc. Curve fitting algorithms can be used to convert a sequence of spatial coordinate points of a taught welding path into a smooth curve, thus obtaining a standard welding path.

[0039] A standard welding path is a continuous spatial curve defined in a standard workpiece coordinate system, used to describe the ideal trajectory of the welding torch tip. The standard welding path can be used as a reference for subsequent welding operations.

[0040] Step 230: Determine the reference point based on the teaching point cloud data and the spatial coordinate point sequence.

[0041] A reference point is a set of feature points selected from the taught point cloud data for precise positioning of the weld. For example, reference points may include hole centers, edge corners, or other repeatably identifiable feature points on a standard workpiece. Reference points can be used to precisely position the workpiece to be welded during subsequent registration. In some embodiments, all reference points are combined to surround the weld.

[0042] In some embodiments, the system 100 may randomly select multiple points uniformly distributed on the teaching welding path from the teaching point cloud data as reference points.

[0043] In some embodiments, the system 100 may further determine multiple key points and their corresponding point features based on the taught point cloud data; determine a first candidate key point based on the multiple key points and their corresponding point features; and determine a reference point based on the spatial coordinate point sequence and the first candidate key point. For further explanation of this part, please see below. Figure 3 The explanation in the document.

[0044] Step 240: Construct a welding database based on the first workpiece identifier, reference point, and standard welding path of the standard workpiece.

[0045] The first workpiece identifier refers to the code or string that uniquely identifies the model or type of a standard workpiece.

[0046] In some embodiments, the system 100 can use the first workpiece identifier of the standard workpiece as the first workpiece identifier by scanning the workpiece barcode on the standard workpiece, reading the RFID tag on the workpiece to be welded, or obtaining the first workpiece identifier directly specified by the user.

[0047] The welding database can be used to store various types of data related to welding tasks. In some embodiments, the system 100 can associate and store the first workpiece identifier of each type of standard workpiece, its corresponding teaching point cloud data, reference points, and standard welding paths to construct the welding database.

[0048] Step 250: Obtain the second workpiece identifier and coarse scan point cloud data of the part to be welded.

[0049] The second workpiece identifier refers to a unique code or string that identifies the model or type of the workpiece to be welded.

[0050] In some embodiments, the second workpiece identifier belongs to one of a plurality of first workpiece identifiers stored in the welding database.

[0051] In some embodiments, the second workpiece identifier is obtained in a similar manner to the first workpiece identifier.

[0052] Coarse scan point cloud data refers to the set of three-dimensional points obtained by coarsely scanning the workpiece to be welded. Coarse scanning refers to a rapid, relatively low-resolution scan. Furthermore, coarse scan point cloud data is mainly used for the preliminary positioning of the welding location on the workpiece.

[0053] In some embodiments, the system 100 may use the set of three-dimensional points generated after a rapid scan of the workpiece to be welded at the workstation by a fixedly installed 3D vision sensor (e.g., a structured light scanner or a laser scanner) as coarse scan point cloud data.

[0054] Step 260: Obtain the teaching point cloud data, reference point, and standard welding path corresponding to the second workpiece identifier from the welding database.

[0055] In some embodiments, the system 100 may search the welding database based on the second workpiece identifier to obtain a first workpiece identifier that matches the second workpiece identifier in the welding database. Then, the teaching point cloud data, reference points, and standard welding paths corresponding to the first workpiece identifier are determined as the teaching point cloud data, reference points, and standard welding paths corresponding to the second workpiece identifier.

[0056] Step 270: Determine the first registration transformation based on the teaching point cloud data and the coarse scan point cloud data.

[0057] The first registration transformation refers to the transformation used to convert coarse scan point cloud data from the original acquisition coordinate system (e.g., the scanner coordinate system) to the workpiece coordinate system where the teaching point cloud data is located. The first registration transformation can be a rigid body transformation matrix that includes rotation and translation.

[0058] In some embodiments, the system 100 may employ a point cloud registration algorithm (such as the sample consistency initial registration algorithm) to calculate the transformation matrix from coarse scan point cloud data to teaching point cloud data, and use the transformation matrix as the first registration transformation.

[0059] In some embodiments, the welding database also includes a first workpiece profile corresponding to a first workpiece identifier, the first workpiece profile being generated based on the taught point cloud data; the system 100 may also determine a first registration transformation in response to the amount of data in the taught point cloud data and / or the amount of data in the coarse scan point cloud data exceeding a third threshold or a fourth threshold. For further explanation of this part, please see below. Figure 4 or Figure 5 The relevant explanations are in the text.

[0060] Step 280: Determine the scanning range based on the reference point and the first registration transformation.

[0061] A scan range refers to a local area defined in three-dimensional space, designated for more detailed or high-resolution scanning operations. For example, a scan range is a local three-dimensional region defined in the coarse scan point cloud data after initial registration, including reference points. The fine scanning equipment will perform a high-precision scan only within this scan range, and will be able to scan the reference points within the workpiece to be welded.

[0062] In some embodiments, the system 100 may utilize a first registration transformation to convert the coordinates of reference points in the welding database to the current scanner coordinate system, and predefine a three-dimensional boundary (such as a cube) as the scanning range, centered on the reference points in the current scanner coordinate system. In some embodiments, the predefinement of the three-dimensional boundary may be predefined by those skilled in the art based on experience.

[0063] Step 290: Based on the scanning range, acquire fine-scan point cloud data.

[0064] Fine-scan point cloud data refers to point cloud data obtained by performing a fine scan within the scanning range. For example, fine scanning includes high-resolution, locally fine scanning. In some embodiments, the system 100 can control a 3D sensor to perform a fine scan of the scanning range to generate fine-scan point cloud data.

[0065] Step 2010: Determine the welding instructions based on the fine scan point cloud data and the standard welding path.

[0066] Welding instructions are control commands or data sequences used to drive welding equipment to perform specific welding operations. For example, welding instructions include parameters such as the motion trajectory coordinates of the robot's welding torch, welding speed, current, and voltage. Welding instructions can be directly executed by the robot controller to complete the welding task.

[0067] In some embodiments, the system 100 may employ a point cloud registration algorithm (such as the sample consistency initial registration algorithm) to calculate a transformation matrix that registers the fine scan point cloud data and the reference point, and perform coordinate transformation on the standard welding path based on this transformation matrix to generate welding instructions that can be directly executed in actual space.

[0068] In some embodiments, the system 100 can determine the error deviation based on the fine scan point cloud data and the reference point; determine an alarm command in response to the error deviation being greater than or equal to a preset error threshold; and determine a second registration transformation based on the fine scan point cloud data and the reference point in response to the error deviation being less than the preset error threshold, and determine a welding command based on the second registration transformation and the standard welding path.

[0069] Error deviation refers to the difference between the measured position of a reference point and the theoretical position of a reference point on a workpiece to be welded. For example, error deviation can be the Euclidean distance between the measured reference point and the theoretical reference point, or the root mean square error of overall registration.

[0070] In some embodiments, the system 100 can re-identify or match the measured reference point positions (i.e., actual three-dimensional coordinates) of each reference point in the fine-scan point cloud data; calculate the Euclidean distance between each measured reference point position and the theoretical reference point position mapped through the first registration transformation, as an error deviation. Alternatively, it can calculate the root mean square error (RMSE) of the overall registration, as an error deviation.

[0071] A preset error threshold refers to a preset maximum allowable positioning deviation value. For example, the preset error threshold can be a specific distance value (such as millimeters) or a value representing the transformation residual. In some embodiments, the preset error threshold can be preset by those skilled in the art based on experience. An alarm command refers to a control signal used to trigger a system alarm. In some embodiments, an alarm command can trigger an audible and visual alarm, send a message notification, or display an error message on the user interface.

[0072] In some embodiments, the system 100 can compare the calculated error deviation with a preset error threshold. In response to the error deviation being greater than or equal to the preset error threshold, the system 100 determines that an alarm is required and generates an alarm command.

[0073] The second registration transformation refers to the rigid body transformation matrix that characterizes the matching relationship between the reference points of the workpiece to be welded and the standard workpiece. The second registration transformation is mainly used to register weld seams to guide the robot in welding operations.

[0074] In some embodiments, the system 100 can compare the calculated error deviation with a preset error threshold. In response to an error deviation less than the error threshold, it uses the precisely identified reference points in the finely scanned point cloud data and the reference points in the welding database to calculate a second registration transformation using a precise point cloud registration algorithm (such as the iterative nearest point algorithm). The iterative nearest point algorithm can repeatedly iterate to find the optimal rigid body transformation matrix between point sets.

[0075] In some embodiments, the system 100 can also generate welding instructions that are ultimately accurately corresponding in actual space by applying a second registration transformation to each point of the standard welding path, thereby accurately transforming it from the theoretical coordinate system to the actual workpiece coordinate system of the workpiece to be welded.

[0076] In some embodiments described in this specification, error deviations are determined based on finely scanned point cloud data and reference points. By comparing these error deviations with a preset error threshold, the system can flexibly select between triggering an alarm or refining the standard welding path. When the error is large, an alarm is triggered promptly to avoid invalid or dangerous welding operations. When the error is within acceptable limits, a second registration transformation is determined to accurately map the standard welding path onto the actual workpiece, significantly improving the accuracy of welding commands and ensuring welding quality. This adaptive decision-making mechanism effectively avoids welding defects caused by initial positioning errors, reduces scrap rates, and improves production efficiency and automation levels.

[0077] In some embodiments of this specification, by establishing a complete "coarse-to-fine" welding positioning and path generation process, by constructing a welding database containing high-precision template data, and by combining preliminary registration with local fine scanning, this method can effectively address the placement deviation of the workpiece to be welded, and accurately reproduce the standard welding path generated in the teaching stage into the current actual workpiece coordinate system of the workpiece to be welded, thereby achieving automation and high precision in welding operations.

[0078] Figure 3 This is an exemplary flowchart illustrating the determination of reference points according to some embodiments of this specification. In some embodiments, process 300 may be executed by a coordinate-system-registered welding system 100 or a processor. Figure 3 As shown, process 300 may include the following steps 310-330.

[0079] Step 310: Based on the teaching point cloud data, determine multiple key points and their corresponding point features.

[0080] Keypoints are points with significant geometric features extracted from teaching point cloud data. For example, keypoints include plane intersections (corner points), points on the axis of a cylinder, and the center of a sphere.

[0081] In some embodiments, the system 100 can use feature extraction algorithms (such as corner detection algorithms based on curvature changes, geometric shape fitting algorithms based on random sampling consistency, etc.) to identify multiple key points in the teaching point cloud data. For example, the system 100 can use a corner detection algorithm based on curvature changes to identify points in the teaching point cloud data with significant curvature changes as corner key points.

[0082] In some embodiments, determining multiple key points can also be achieved through other point cloud processing algorithms, such as edge point detection based on normal features.

[0083] Point features refer to information describing the type and local geometric attributes of key points. For example, point features include feature types (such as corner points, hole centers) and local geometric descriptors.

[0084] In some embodiments, the system 100 may also label each identified key point with its type and local geometric attributes as point features corresponding to the key point. For example, if a key point is the center point of a hole, the point features of the key point labeled by the system 100 may include a "hole center" type identifier, as well as local geometric information describing the radius, depth, etc. of the hole.

[0085] In some embodiments, the point features of key points annotated by the system 100 can also be determined in other ways, such as using a Local Surface Features Histogram (LSFH) to describe the local geometric properties of the points.

[0086] Step 320: Determine the first candidate key point based on multiple key points and their corresponding point features.

[0087] The first candidate critical point refers to a subset of critical points that, after initial screening, meet the basic conditions to become a benchmark point. For example, critical points in this subset may include points located on the taught welding path within a preset tolerance range, or points with specific geometric shapes (such as right angles or the center of a circular arc) that are close to the welding area.

[0088] In some embodiments, the system 100 can determine first candidate key points by filtering according to preset filtering rules based on the point features of each key point and its preliminary spatial relationship with the taught welding path. In some embodiments, the preset filtering rules may be to retain only key points whose point features are "corner points" or "hole centers" and are located within the preset tolerance range of the taught welding path. The preset tolerance range may be preset by those skilled in the art based on experience.

[0089] In some embodiments, the system 100 may also evaluate the distance score, visibility score, and feature quality score of each of the plurality of keypoints; and determine a first candidate keypoint based on the distance score, visibility score, and feature quality score of each keypoint.

[0090] The distance score is a score used to assess the proximity between a key point and a preset target path (e.g., a taught welding path). In some embodiments, for each key point, the system 100 may calculate the shortest Euclidean distance from the key point to all points on the taught welding path and compare this shortest Euclidean distance with a maximum effective distance threshold. The closer the shortest Euclidean distance is to the maximum effective distance threshold, the closer the distance score is to 1. The maximum effective distance threshold can be preset by those skilled in the art based on experience, for example, 5 mm.

[0091] A visibility score is a score used to assess whether a key point can be effectively observed by a sensing device (e.g., a 3D sensor). In some embodiments, the system 100 can determine the visibility score by simulating light projection from the sensing device's viewpoint to the key point and determining whether the key point is occluded. If the occlusion level is below a preset occlusion threshold, the visibility score of the key point is determined to be 1; otherwise, the visibility score is determined to be 0. The preset occlusion threshold can be preset by those skilled in the art based on experience.

[0092] Feature quality score refers to the score used to evaluate the inherent quality or reliability of the features associated with a key point.

[0093] In some embodiments, for each keypoint, the system 100 can evaluate the feature quality score based on the local point cloud neighborhood and fitting parameters used when extracting the keypoint, thereby determining the feature quality score of the keypoint. The local point cloud neighborhood refers to the set of all neighboring points within a certain range around the keypoint. The local point cloud neighborhood can be preset by those skilled in the art based on experience.

[0094] For example, for the fitted cylindrical feature, the system 100 can calculate the standard deviation of the distance from the key point to the fitted cylindrical surface; if the standard deviation is less than or equal to the preset maximum allowable residual, the feature quality score of the key point is determined to be 1, otherwise the feature quality score of the key point is determined to be 0. In some embodiments, the preset maximum allowable residual can be preset by those skilled in the art based on experience.

[0095] For example, for a key point that is a corner point, the system 100 can evaluate the local point cloud density of the key point; if the density is higher than a preset density threshold, the feature quality score of the key point is assigned a value of 1, otherwise the feature quality score of the key point is assigned a value of 0. The preset density threshold can be preset by those skilled in the art based on experience.

[0096] In some embodiments, the system 100 may identify all keypoints that simultaneously satisfy a distance score, a visibility score, and a feature quality score of 1 as first candidate keypoints.

[0097] Some embodiments in this specification comprehensively evaluate three specific and quantifiable technical indicators: distance score ensures the key point is related to the weld; visibility score ensures the key point can be reliably scanned from a fixed viewpoint; and feature quality score ensures the features extracted from the point cloud are accurate and reliable. This triple filtering of multiple key points effectively filters out key points that do not meet the spatial, perceptual, and feature reliability requirements, ensuring that the selected first-choice key point possesses high availability and stability.

[0098] In some embodiments, the system 100 may further evaluate the detectability score, stability score, and relevance score of each first candidate keypoint when the number of first candidate keypoints exceeds a first threshold; determine a first comprehensive score for each first candidate keypoint based on the detectability score, stability score, and relevance score of each first candidate keypoint; and update the first candidate keypoints based on the first comprehensive score and a second threshold.

[0099] The first threshold refers to a preset value used to judge and filter the number of first candidate keywords. For example, the first threshold could be 20.

[0100] The detectability score is a numerical value used to quantify how easily a first-choice keypoint can be detected. The detectability score reflects the uniqueness or feature strength of the first-choice keypoint itself. The detectability score ranges from 0 to 1.

[0101] In some embodiments, the system 100 can determine the base score and information entropy corresponding to the feature type of the first candidate keypoint (e.g., corner point, center of a circular hole, etc.) by looking up a feature type score table; and determine the detectability score as the value obtained by adding the base score and the additional score corresponding to the information entropy of the first candidate keypoint and then normalizing the result. In some embodiments, the higher the information entropy (i.e., the more unique the first candidate keypoint), the higher the additional score.

[0102] The feature type lookup table contains the feature type of the first candidate key point, along with its corresponding reference base score and reference information entropy. This feature type lookup table can be pre-set by someone skilled in the art based on prior knowledge.

[0103] For example, if the feature type of the first candidate keypoint is a corner point, the base score can be set to 0.7 and the information entropy can be set to 2; if the feature type of the first candidate keypoint is the center of a circular hole, the base score can be set to 0.9 and the information entropy can be set to 3.

[0104] The stability score is a numerical value used to quantify the robustness or stability of the first candidate keypoint and its local neighborhood in the face of small perturbations. The stability score reflects the flatness or resistance to deformation within the local neighborhood of the first candidate keypoint. In some embodiments, the system 100 may calculate the variance of the point cloud normal vectors in the local neighborhood of the first candidate keypoint; determine the stability score of the first candidate keypoint based on the variance; and normalize the stability score of the first candidate keypoint to 0-1.

[0105] In some embodiments, the system 100 may respond that a smaller variance indicates a flatter, more rigid, and less susceptible-to-small deformation local neighborhood, and a higher stability score for the first candidate key point.

[0106] The relevance score refers to the distance relationship between a first candidate keypoint and a preset operation path (e.g., a taught welding path). In some embodiments, the system 100 can calculate the shortest distance from the first candidate keypoint to the taught welding path; and based on the shortest distance, map the shortest distance to a score between 0 and 1 using a decay function (e.g., a Gaussian function), thus determining the relevance score. This decay function can be designed to have the highest score at an ideal distance, while the score is lower when the distance is too close or too far.

[0107] The first comprehensive score refers to the numerical value of a comprehensive quantitative evaluation of the first candidate key point based on multiple evaluation indicators (such as detectability score, stability score, and relevance score).

[0108] In some embodiments, for each candidate keypoint, the system 100 can perform a weighted summation of the detectability score, stability score, and relevance score of the first candidate keypoint to determine a first comprehensive score for the first candidate keypoint. In some embodiments, the weight coefficients corresponding to the detectability score, stability score, and relevance score can be preset by those skilled in the art based on experience, so that the sum of the weight coefficients corresponding to the detectability score, stability score, and relevance score is 1.

[0109] The second threshold refers to a preset value used to judge and filter the first comprehensive score. In some embodiments, the second threshold can be preset by those skilled in the art based on experience. For example, the second threshold can be 0.8.

[0110] In some embodiments, the system 100 may retain only the first candidate keypoints whose first comprehensive score is higher than the second threshold, thereby obtaining the updated first candidate keypoints.

[0111] Some embodiments in this specification employ a sophisticated evaluation of first candidate keypoints by introducing detectability scores, stability scores, and relevance scores. Based on these scores, a first comprehensive score is determined, and then the first candidate keypoints are updated through a second threshold screening. This effectively improves the quality of the first candidate keypoints, ensuring that the retained first candidate keypoints are easier to detect, more stable, and more relevant to the task objective, significantly enhancing the accuracy and robustness of subsequent task processing.

[0112] Step 330: Determine the reference point based on the spatial coordinate point sequence and the first candidate key point.

[0113] In some embodiments, the system 100 may randomly select several points from the first candidate key points along the taught welding path, such that the selected points are evenly distributed on the taught welding path, and these selected points are the reference points.

[0114] In some embodiments, the system 100 can also divide the spatial coordinate point sequence according to a preset cutting window to obtain multiple sub-welding segments; and determine a reference point based on the first candidate key point and multiple sub-welding segments.

[0115] A preset cutting window refers to a length value or range used to divide a sequence or path according to a preset length. For example, a preset cutting window can be a fixed length value (Length of window, L_window) used to divide a sequence of spatial coordinate points.

[0116] A sub-welding segment refers to multiple parts or segments obtained by dividing the taught welding path according to preset conditions. For example, a sub-welding segment is multiple parts obtained by dividing the spatial coordinate point sequence according to a preset cutting window.

[0117] In some embodiments, the system 100 can divide the spatial coordinate point sequence from the starting point along the length of the taught welding path to obtain multiple sub-welding segments.

[0118] In some embodiments, when the total length of the path containing the spatial coordinate point sequence is not divisible by L_window, the length of the last sub-welding segment may be less than L_window.

[0119] In some embodiments, the system 100 can also divide the spatial coordinate point sequence based on feature density to obtain multiple sub-welding segments.

[0120] Feature density refers to the spatial distribution density of the first candidate keypoints in the region along both sides of the standard welding path in the teaching point cloud data. High feature density indicates that the region has abundant keypoints, while low feature density indicates that the region has sparse keypoints.

[0121] In some embodiments, the system 100 may set a sliding analysis window along a standard welding path, and at each window position, count the number of first candidate key points falling within it, thereby obtaining the feature density at that location. In some embodiments, the length of the sliding analysis window may be initialized based on the total length of the standard welding path and the expected number of sub-welding segments, and may remain fixed or dynamically adjusted during the analysis process.

[0122] In some embodiments, the system 100 can divide the spatial coordinate point sequence based on the acquired feature density to obtain multiple sub-welding segments. For example, the system 100 can divide relatively long sub-welding segments in regions with high feature density (indicating abundant first candidate key points) and relatively short sub-welding segments in regions with low feature density (indicating sparse first candidate key points). This non-uniform division strategy aims to ensure that each relatively short sub-welding segment contains at least one suitable candidate point to choose from, thereby improving the accuracy and robustness of subsequent welding planning.

[0123] In some embodiments of this specification, by non-uniformly dividing the spatial coordinate point sequence based on feature density, the problem of traditional equal-length cutting windows potentially fragmenting continuous feature regions or making it difficult to select reference points in feature-sparse areas is avoided. This method can intelligently adjust the length of sub-welding segments according to the actual distribution of the first candidate key points in the teaching point cloud, adaptively dividing the segment so that the first candidate key point resources within each sub-welding segment match its length, ensuring that each sub-welding segment contains sufficient and suitable key points, significantly improving adaptability to complex workpieces.

[0124] In some embodiments, for each sub-welding segment, the system 100 may select, from a set of first candidate key points spatially adjacent to the sub-welding segment, the one with the highest first comprehensive score or multiple first candidate key points that meet a scoring threshold as a reference point serving the sub-welding segment; the set of reference points selected from all sub-welding segments shall be used as the final determined reference point. The scoring threshold shall be set by system default or determined based on experience.

[0125] In some embodiments, spatial proximity can be determined by setting a preset distance threshold, whereby any first candidate key point located within a preset distance range of the sub-welding segment is considered spatially nearby. The preset distance threshold can be preset by those skilled in the art based on experience.

[0126] In some embodiments, system 100 may also determine the reference point in other ways.

[0127] In some embodiments, system 100 may determine the reference point using the following steps 331-333:

[0128] Step 331: Determine the influence radius of each of the multiple sub-welding segments.

[0129] The radius of influence refers to the spatial range of influence set for a sub-welding segment, used to determine whether a first candidate key point is associated with that sub-welding segment. In some embodiments, when a first candidate key point falls within the radius of influence of a sub-welding segment, it is considered that the first candidate key point can be selected by that sub-welding segment.

[0130] In some embodiments, the system 100 can adaptively adjust the radius of influence of each sub-weld segment according to its length. For example, the longer the sub-weld segment, the larger the radius of influence that the system 100 can preset, and vice versa. Another example is that the system 100 can set the radius of influence to a preset proportion of the sub-weld segment length.

[0131] In some embodiments, the system 100 can adaptively adjust the radius of influence of each sub-weld segment based on its curvature. For example, the smaller the curvature of a sub-weld segment (i.e., the straighter it is), the larger the radius of influence that the system 100 can preset; conversely, the larger the curvature (i.e., the more curved it is), the smaller the radius of influence that the system 100 can preset. Alternatively, the system 100 can calculate the radius of influence based on the maximum curvature of the sub-weld segment using an inverse proportional relationship or a lookup table.

[0132] In some embodiments, the radius of influence can also be determined by comprehensively considering the length and curvature of the sub-welded segment, or by other means such as according to actual welding process requirements.

[0133] Step 332: Based on the first candidate key point, multiple sub-welding segments, and the radius of influence, construct the candidate point-segment relationship table.

[0134] The candidate point-segment relationship table is a data structure used to record the first candidate key points that can be selected for each sub-welding segment. For example, this relationship table records information such as the first candidate key points that can be selected for sub-welding segment S1 are {K1,K2}, the first candidate key points that can be selected for sub-welding segment S2 are {K2,K3}, ..., and the first candidate key points that can be selected for sub-welding segment Sn are {Kn-1,Kn}, etc.

[0135] In some embodiments, for each first candidate key point, the system 100 can calculate the nearest point from the first candidate key point to the entire standard welding path; based on the nearest point, determine whether the nearest point falls within the influence radius of a certain sub-welding segment; and in response to falling within the influence radius of a certain sub-welding segment, register the first candidate key point in the candidate list of first candidate key points of the sub-welding segment and store it in the candidate point-segment relationship table. For example, if the nearest point P1 of a first candidate key point K1 is located within the influence radius of sub-welding segment S1, then K1 can be used as a candidate point of S1.

[0136] In some embodiments, a first candidate key point may appear simultaneously in the candidate point-segment relationship table of multiple adjacent extension segments.

[0137] Step 333: Determine the benchmark point based on the candidate point-road segment relationship table.

[0138] In some embodiments, the system 100 may process each sub-welded segment sequentially, starting from the first sub-welded segment, based on a candidate point-segment relationship table. For the current sub-welded segment, a first candidate key point is selected from its candidate point-segment relationship table as a reference point; if, among the candidate first candidate key points corresponding to the sub-welded segment, a first candidate key point has already been selected as a reference point by a previous sub-welded segment, then, among the remaining candidate first candidate key points, the one with the highest comprehensive score is selected as the reference point for the current sub-welded segment; and so on, until all sub-welded segments have determined reference points.

[0139] In some embodiments of this specification, by first determining the influence radius of each sub-welding segment, multiple sub-welding segments can share high-quality first candidate key points, effectively solving the problem of difficulty in selecting good reference points on both sides when the first candidate key point happens to be located at the partition boundary. Subsequently, a candidate point-segment relationship table is constructed based on the influence radius, clearly establishing the potential association between the first candidate key point and the sub-welding segment. Finally, by determining the reference point based on this candidate point-segment relationship table, it can be ensured that each sub-welding segment can obtain a suitable reference point, and high-quality first candidate key points can be prioritized, thereby improving the robustness, accuracy, and efficiency of reference point selection, and providing a more stable and reliable reference for subsequent welding operations.

[0140] In some embodiments of this specification, by dividing the spatial coordinate point sequence into multiple sub-welding segments according to a preset cutting window, and determining a reference point for each sub-welding segment based on a first candidate key point, fine-grained management and reference point allocation of long welding paths are achieved, ensuring effective coverage of the entire welding path. This partitioned coverage strategy significantly improves the accuracy and reliability of reference point determination, avoiding the problem of the entire path failing to be effectively guided due to the failure of a single reference point, and is particularly suitable for scenarios with complex or ultra-long welding paths.

[0141] In some embodiments of this specification, potential key points and their features are automatically discovered from the teaching point cloud data using algorithms, and optimized based on the teaching welding path to select the final reference point. This automates and standardizes the reference point selection process, and makes it highly relevant to subsequent welding tasks.

[0142] Figure 4 This is an exemplary flowchart illustrating the determination of a first registration transformation according to some embodiments of this specification. In some embodiments, process 400 may be executed by a coordinate-system-based registration welding system 100 or a processor. Figure 4 As shown, process 400 may include the following steps 410-420.

[0143] In some embodiments, the welding database also includes a first workpiece profile corresponding to the first workpiece identifier. The first workpiece profile is generated based on the teaching point cloud data. When the amount of data in the teaching point cloud data and / or the amount of data in the coarse scan point cloud data are greater than a third threshold, the system 100 may perform the following steps 410-420.

[0144] The first workpiece profile refers to the geometric model that represents the shape and main structure of the standard workpiece corresponding to the first workpiece identifier. For example, the first workpiece profile can be a triangular mesh model of the standard workpiece or a simplified point cloud.

[0145] In some embodiments, when constructing the database, the system 100 can perform denoising (e.g., statistical filtering) and downsampling (e.g., voxel mesh method) on the taught point cloud data, and then use a geometric feature extraction algorithm (e.g., polygon mesh reconstruction algorithm) to convert the discrete point cloud into a triangular mesh model as the first workpiece contour. In some embodiments, the system can perform uniform downsampling or extract based on key points (e.g., corner points, edge points) on the taught point cloud data to obtain a sparse but representative point set that can represent the overall shape of the workpiece as the first workpiece contour in a simplified point cloud form.

[0146] In some embodiments, when the amount of data in the teaching point cloud data and / or the amount of data in the coarse scan point cloud data is greater than a third threshold, the system 100 may perform the following steps 410-420.

[0147] Data volume refers to the number of data points or data elements contained in a dataset. For example, the data volume of teaching point cloud data can indicate the number of 3D points in the teaching point cloud data, and the data volume of coarse scan point cloud data can indicate the number of 3D points in the coarse scan point cloud data. In some embodiments, the system 100 can directly count the number of points in the teaching point cloud data as the data volume of the teaching point cloud data, or directly count the number of points in the coarse scan point cloud data as the data volume of the coarse scan point cloud data. The data volume in this specification can be determined in various ways. For example, in addition to counting the number of points, it can also be the file size of the data. The third threshold refers to a preset numerical limit for the number of point clouds. In some embodiments, the third threshold can be preset by those skilled in the art based on experience.

[0148] Step 410: Determine the contour of the second workpiece based on the coarse scan point cloud data.

[0149] The second workpiece profile refers to the geometric model extracted from the coarse scan point cloud data of the workpiece to be welded, corresponding to the first workpiece profile. For example, the second workpiece profile can be a triangular mesh model or a simplified point cloud of the workpiece to be welded.

[0150] In some embodiments, the method for determining the second workpiece contour is similar to the method for generating the first workpiece contour, except that the processing object changes from the taught point cloud data to the coarse scan point cloud data, which will not be described in detail here.

[0151] Step 420: Determine the first registration transformation based on the second workpiece contour and the first workpiece contour.

[0152] In some embodiments, system 100 may use a 3D model registration algorithm (such as a feature-based normal alignment algorithm or a variant of the Iterative Closest Point (ICP) algorithm) to calculate the rigid body transformation between the second workpiece contour and the first workpiece contour, and use this transformation as the first registration transformation. The rigid body transformation includes rotation and translation. This rigid body transformation aligns the second workpiece contour to the position and orientation of the first workpiece contour.

[0153] Some embodiments in this specification introduce a contour registration strategy based on a data volume threshold. When the point cloud data volume is large, registration is performed using a simplified workpiece contour instead of the original high-density point cloud. This significantly reduces the data scale and complexity, improves the running speed of the registration algorithm and the robustness of the initial matching, and provides a more reliable preliminary positioning for determining the subsequent fine scanning range.

[0154] Figure 5 This is another exemplary flowchart illustrating the determination of a first registration transformation according to some embodiments of this specification. In some embodiments, process 500 may be executed by a coordinate-system-based registration welding system 100 or a processor. Figure 5As shown, process 500 may include the following steps 510-520.

[0155] In some embodiments, the welding database also includes a set of teaching coarse matching points. In response to the fact that the amount of data of the teaching point cloud data and / or the amount of data of the coarse scan point cloud data are greater than a fourth threshold, the system 100 may also perform the following steps 510-520.

[0156] The fourth threshold is a preset critical value. It can be used to determine whether the amount of data in the current teaching point cloud data and / or the amount of data in the coarse scan point cloud data is large enough to warrant a more efficient matching strategy. In some embodiments, the fourth threshold is greater than the aforementioned third threshold.

[0157] In some embodiments, the system first acquires the data volume of the teaching point cloud data and the data volume of the coarse scan point cloud data. For methods on acquiring the data volume of the teaching point cloud data and the coarse scan point cloud data, please refer to [link to documentation]. Figure 4 The relevant explanations are in the text.

[0158] A coarse-matching point set refers to a set of points selected from the teaching point cloud data that represent unique and easily detectable features on a standard workpiece. For example, the coarse-matching point set may include specific corner points, small facet centers, etc., of the standard workpiece. In some embodiments, the system 100 may randomly select multiple points from the teaching point cloud data as the coarse-matching point set. For example, the system 100 may interactively select highly discriminative and easily identifiable feature points, such as specific corner points of a standard workpiece, from the teaching point cloud data as the coarse-matching point set.

[0159] In some embodiments, system 100 may employ a feature point detection algorithm (such as ISS or Harris3D) to identify key points in the teaching point cloud data as a teaching coarse matching point set. In some embodiments, the system may also employ other methods to determine the teaching coarse matching point set.

[0160] In some embodiments, the system 100 may also employ steps a1-a4 to determine the teaching coarse matching point set.

[0161] Step a1: Determine multiple structure sets based on the teaching point cloud data.

[0162] A structure set refers to a collection of multiple spatially distributed instances with geometric features (such as cylinders, cuboids, etc.) that collectively and uniquely determine the six-degree-of-freedom pose of a target object (such as a standard workpiece) in space. For example, a structure set may include combinations of geometric feature instances of a standard workpiece, such as planes, cylinders, or long sides.

[0163] In some embodiments, the system 100 can extract multiple geometric feature instances from the teaching point cloud data and filter them according to the principle of "large size and easy detection" to select the top N (e.g., N=20) geometric feature instances; and use heuristic rules to generate multiple structure sets from the top N (e.g., N=20) geometric feature instances.

[0164] In some embodiments, the system 100 can identify features such as planes, cylinders, and long sides in a point cloud based on shape recognition or surface fitting algorithms, and select the top N (e.g., N=20) geometric feature instances. In some embodiments, the heuristic rules may include a first rule, a second rule, and a third rule.

[0165] The first rule ensures that each set of structures contains at least one plane with an area greater than a preset area among the top N geometric feature instances, thereby stably determining the pose of the target object (e.g., a standard workpiece) in space. The preset area can be predetermined by those skilled in the art based on experience.

[0166] The second rule ensures that each set of structures contains at least one cylinder or long-side feature from the top N geometric feature instances, used to determine the rotational pose of the target object about the Z-axis.

[0167] The third rule can ensure that the remaining geometric feature instances in each structure set are as far away as possible from the geometric feature instances determined by the first and second rules, in order to form a large volume.

[0168] Step a2: Evaluate the detection speed score, discrimination score, and distribution score for each of the multiple structure sets.

[0169] The detection speed score is a metric used to measure the estimated time required to extract geometric feature instances from taught point cloud data and match them to a specific set of structures. In some embodiments, a higher detection speed score indicates a shorter time required to extract and match the set of structures.

[0170] In some embodiments, the system 100 can assign a base extraction time weight to each geometric feature instance type (e.g., a base extraction time weight of 1.0 for a plane, 1.5 for a cylinder, 2.0 for an edge, and 5.0 for a complex freeform surface), and set a reward factor for the feature size (e.g., area of ​​a plane, radius of a cylinder) of each geometric feature instance. If the geometric feature instance has a large feature size, it will receive a higher reward factor.

[0171] In some embodiments, the system 100 may also determine the detection speed score of the corresponding structure set by summing the products of the basic extraction time weight corresponding to the type of all geometric feature instances in each structure set and the reward factor corresponding to its feature size.

[0172] Discrimination score is a metric used to measure the degree of difference between a set of structures and other sets of structures. In some embodiments, a higher discrimination score indicates that the set of structures is less similar to other sets of structures, thereby reducing the probability of mismatches.

[0173] In some embodiments, system 100 can compute a local descriptor for each structure set and calculate the average dissimilarity between that local descriptor and the local descriptors of all other structure sets; the greater the dissimilarity, the higher the score. A local descriptor is a quantized vector or structure used to characterize the data features in a structure set, such as a fast point feature histogram (FPFH).

[0174] The distribution score is an index used to measure the stability of the spatial distribution of geometric feature instances within a set of structures for solving the pose of a target object (e.g., the six-DOF pose of a standard workpiece). In some embodiments, a higher distribution score indicates a better distribution of geometric feature instances within the set of structures, and thus better stability in solving the pose of the target object.

[0175] In some embodiments, system 100 can evaluate the minimum distance between the center point of a structure set and the center points of one or more other selected structure sets. A larger minimum distance results in a higher score, thereby improving the stability of the degree-of-freedom pose solution. For example, if multiple structure sets include A, B, C, and D, and the distribution score of structure set A needs to be evaluated, system 100 can calculate the distances between the center points of A and B, A and C, A and D, A and BC, A and BD, A and CD, and A and BCD, respectively. Then, it selects the minimum distance among these distances and determines the distribution score of structure set A based on a preset distribution score table. The preset distribution score table establishes a correlation between the minimum distance of a structure set and its distribution score. The preset distribution score table can be preset by those skilled in the art based on experience.

[0176] Step a3: Determine the second comprehensive score for each structure set based on its detection speed score, discrimination score, and distribution score.

[0177] The second comprehensive score is an indicator that comprehensively evaluates a structure set based on its detection speed score, discrimination score, and distribution score. The second comprehensive score reflects the overall performance of the structure set in terms of detection efficiency, discriminability, and pose solution stability.

[0178] In some embodiments, the system 100 may use the weighted sum of the detection speed score, discrimination score, and distribution score as the second comprehensive score. In some embodiments, the weights of the detection speed score, discrimination score, and distribution score may be preset by those skilled in the art based on experience, such that the sum of the weights of the detection speed score, discrimination score, and distribution score is 1.

[0179] Step a4: Determine the coarse matching point set for teaching based on the second comprehensive score and the set score threshold.

[0180] The ensemble scoring threshold refers to a preset critical value for the second comprehensive score. In some embodiments, the ensemble scoring threshold can be preset by those skilled in the art based on experience.

[0181] In some embodiments, the system 100 may compare a second comprehensive score of each structure set with a set score threshold, and if the second comprehensive score of a structure set is higher than the set score threshold, then the structure set is identified as a teaching coarse matching point set.

[0182] Some embodiments in this specification improve the quality of the teaching coarse matching point set by comprehensively evaluating the detection speed score, discrimination score, and distribution score of the structure set, and selecting the optimal set based on a second comprehensive score and a set score threshold. This method overcomes the problems of high mismatch rate, unstable pose solution, or low efficiency of traditional schemes in complex scenes, ensuring the accuracy, robustness, and real-time performance of target object pose determination, significantly reducing the computational burden of subsequent fine matching, and improving the overall performance of point cloud registration.

[0183] In some embodiments, for a set of structures that satisfies a second comprehensive score greater than a set score threshold, the system 100 may also determine a subset of key points of the set of structures based on the set of structures, multiple key points associated with the set of structures and their corresponding point features, through a key point prediction model; and determine the determined subset of key points as a set of coarse matching points for teaching.

[0184] For an explanation of key points and point features, please refer to [link / reference]. Figure 3 The relevant explanation in step 310.

[0185] In some embodiments, the key point prediction model is a machine learning model. In some embodiments, the risk prediction model may be a neural network (NN) model or a deep neural network (DNN) model, etc.

[0186] In some embodiments, the input to the keypoint prediction model may include a structure set, multiple keypoints associated with the structure set and their corresponding point features, and the output may include a subset of keypoints from the structure set.

[0187] A keypoint subset refers to a selection of keypoints from one or more keypoints. For example, a keypoint subset might consist of 3-4 keypoints and is used for high-precision matching or attitude estimation.

[0188] In some embodiments, the keypoint prediction model can be trained based on multiple first training samples with first training labels.

[0189] In some embodiments, each group of first training samples may include a set of sample structures, multiple key points associated with the set of sample structures, and their corresponding point features. In some embodiments, the system 100 may simulate a sample standard workpiece under real working conditions (such as adding noise or changing the viewpoint) and obtain the set of sample structures of the sample standard workpiece, as well as the key points and their point features associated with the set of sample structures, to obtain the first training samples.

[0190] In some embodiments, the first training label may be a subset of the actual key points corresponding to the historical structure set. In some embodiments, the system 100 may select the smallest subset that meets preset performance requirements (e.g., matching accuracy less than 0.1 mm and matching success rate greater than 95%) from all possible combinations of key point subsets as the first training label of the sample structure set.

[0191] In some embodiments, the system 100 can input multiple first training samples with first training labels into an initial keypoint prediction model, construct a loss function using the first training labels and the results of the initial keypoint prediction model, and iteratively update the parameters of the initial keypoint prediction model based on the loss function using gradient descent or other methods. When preset conditions are met, the initial risk prediction model training is complete, resulting in a trained keypoint prediction model. These preset conditions may include loss function convergence, the number of iterations reaching a threshold, etc.

[0192] Some embodiments in this specification introduce a keypoint prediction model that intelligently determines a high-precision subset of keypoints as the teaching coarse matching point set based on a second comprehensive score of the structural set and the keypoints and point features of the set. This method can effectively filter out a small number of keypoints that best represent the structural features and are used for precise matching. This significantly reduces the computational complexity and time consumption of the subsequent matching process. At the same time, by selecting high-quality and critical feature points, it effectively improves the accuracy and robustness of coarse matching, thereby improving the overall recognition and localization performance of the system in complex environments.

[0193] Step 510: Based on the teaching coarse matching point set and coarse scan point cloud data, determine the actual coarse matching point set of the part to be welded.

[0194] The part to be welded refers to the workpiece or object that will be welded. For example, the part to be welded could be a car body or structural component. The actual coarse matching point set refers to a set of points selected from the coarse scan point cloud data that represent unique and easily detectable features on the part to be welded. For example, the actual coarse matching point set may include specific corner points, small plane centers, etc., of the part to be welded.

[0195] In some embodiments, the system 100 can detect a set of points in the coarse scan point cloud data that corresponds to features in the taught coarse matching point set, and determine it as the actual coarse matching point set of the workpiece to be welded. For example, the system 100 can use a corner detection algorithm or a plane detection algorithm in the coarse scan point cloud data to identify these features.

[0196] In some embodiments, system 100 may also determine the actual coarse matching point set using a feature descriptor-based matching method. For example, system 100 may compute a local geometric feature descriptor (such as FPFH) for each point in the taught coarse matching point set, and then search for the best-matching point set in the coarse scan point cloud data as the actual coarse matching point set for the workpiece to be welded.

[0197] Step 520: Determine the first registration transformation based on the taught coarse matching point set and the actual coarse matching point set.

[0198] In some embodiments, system 100 may use a feature-based fast registration algorithm (e.g., solving the transformation after feature matching using a FastPoint Feature Histogram (FPFH) descriptor) to compute a first registration transformation between the taught coarse matching point set and the actual coarse matching point set.

[0199] Some embodiments in this specification enable rapid registration when the amount of teaching point cloud data and / or coarse scan point cloud data is large. This is achieved by using a pre-stored set of teaching coarse matching points and the actual set of coarse matching points detected from the coarse scan point cloud data. This method avoids time-consuming registration calculations on the entire large point cloud, significantly improving registration efficiency. It is particularly suitable for complex workpiece scenarios where the amount of data exceeds a fourth threshold (the fourth threshold is greater than the third threshold).

[0200] The basic concepts have been described above. Obviously, for those skilled in the art, the detailed disclosure above is merely illustrative and does not constitute a limitation of this specification. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this specification. Such modifications, improvements, and corrections are suggested in this specification and therefore remain within the spirit and scope of the exemplary embodiments described herein.

[0201] It should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be consistent with the teachings of this specification, rather than as examples or limitations. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A welding method based on coordinate system registration, characterized in that, include: The spatial coordinate point sequence and teaching point cloud data of the teaching welding path of the standard workpiece are obtained through sensors. Based on the spatial coordinate point sequence, a standard welding path is generated using a curve fitting algorithm; Based on the teaching point cloud data, multiple key points and their corresponding point features are determined; Based on the multiple key points and their corresponding point features, a first candidate key point is determined; According to the preset cutting window, the spatial coordinate point sequence is divided to obtain multiple sub-welding segments; Based on the first candidate key point and the plurality of sub-welding segments, a reference point is determined, specifically including: Determine the radius of influence of each of the plurality of sub-welding segments; Based on the first candidate key point, the multiple sub-welding segments, and the influence radius, a candidate point-road segment relationship table is constructed. The benchmark point is determined based on the candidate point-road segment relationship table; A welding database is constructed based on the first workpiece identifier of the standard workpiece, the reference point, and the standard welding path. Acquire the second workpiece identifier and coarse scan point cloud data of the part to be welded; Obtain teaching point cloud data, reference points, and standard welding paths corresponding to the second workpiece identifier from the welding database; Based on the teaching point cloud data and the coarse scan point cloud data, a first registration transformation is determined; Based on the reference point and the first registration transformation, the scanning range is determined; Based on the aforementioned scanning range, acquire fine-scan point cloud data; and Welding instructions are determined based on the fine-scan point cloud data and the standard welding path.

2. The method according to claim 1, characterized in that, The welding database also includes a first workpiece contour corresponding to the first workpiece identifier, the first workpiece contour being generated based on the teaching point cloud data; determining the first registration transformation based on the teaching point cloud data and the coarse scan point cloud data includes: In response to the fact that the amount of data in the teaching point cloud data and / or the amount of data in the coarse scan point cloud data exceeds a third threshold, Based on the coarse scan point cloud data, the contour of the second workpiece is determined; and The first registration transformation is determined based on the second workpiece contour and the first workpiece contour.

3. The method according to claim 2, characterized in that, The welding database also includes a set of coarse matching points for teaching; determining the first registration transformation based on the teaching point cloud data and the coarse scan point cloud data further includes: In response to the fact that the amount of data in the teaching point cloud data and / or the amount of data in the coarse scan point cloud data exceeds a fourth threshold, Based on the teaching coarse matching point set and the coarse scan point cloud data, the actual coarse matching point set of the workpiece to be welded is determined; and Based on the teaching coarse matching point set and the actual coarse matching point set, the first registration transformation is determined; The fourth threshold is greater than the third threshold.

4. A welding system based on coordinate system registration, characterized in that, It includes a first acquisition module, a generation module, a benchmark point determination module, a database construction module, a second acquisition module, a database access module, a transformation determination module, a fine scan module, and an instruction determination module; The first acquisition module is configured to acquire the spatial coordinate point sequence and teaching point cloud data of the teaching welding path of the standard workpiece through a sensor; The generation module is configured to generate a standard welding path based on the spatial coordinate point sequence using a curve fitting algorithm. The reference point determination module is configured as follows: Based on the teaching point cloud data, multiple key points and their corresponding point features are determined; Based on the multiple key points and their corresponding point features, a first candidate key point is determined; According to the preset cutting window, the spatial coordinate point sequence is divided to obtain multiple sub-welding segments; Based on the first candidate key point and the plurality of sub-welding segments, a reference point is determined, specifically including: Determine the radius of influence of each of the plurality of sub-welding segments; Based on the first candidate key point, the multiple sub-welding segments, and the influence radius, a candidate point-road segment relationship table is constructed. The benchmark point is determined based on the candidate point-road segment relationship table; The database construction module is configured to construct a welding database based on the first workpiece identifier of the standard workpiece, the reference point, and the standard welding path. The second acquisition module is configured to acquire the second workpiece identifier of the workpiece to be welded and the coarse scan point cloud data of the workpiece to be welded; The database access module is configured to obtain teaching point cloud data, reference points, and standard welding paths corresponding to the second workpiece identifier from the welding database. The transformation determination module is configured to determine a first registration transformation based on the teaching point cloud data and the coarse scan point cloud data. The fine scanning module is configured to determine the scanning range based on the reference point and the first registration transform, and to acquire fine scanning point cloud data based on the scanning range; and The instruction determination module is configured to determine welding instructions based on the fine scan point cloud data and the standard welding path.

5. The system according to claim 4, characterized in that, The welding database also includes a first workpiece contour corresponding to the first workpiece identifier, the first workpiece contour being generated based on the teaching point cloud data; the transformation determination module is further configured to: When the amount of data in the teaching point cloud data and / or the amount of data in the coarse scan point cloud data exceeds a third threshold, Based on the coarse scan point cloud data, the contour of the second workpiece is determined; and The first registration transformation is determined based on the second workpiece contour and the first workpiece contour.

6. A welding device based on coordinate system registration, characterized in that, The device includes at least one processor and at least one memory; The at least one memory is used to store computer instructions; The at least one processor is configured to execute at least a portion of the computer instructions to implement the coordinate-registration-based welding method as described in any one of claims 1 to 3.

7. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions. When the computer reads the computer instructions in the storage medium, the computer executes the welding method based on coordinate system registration as described in any one of claims 1 to 3.

Images