A welding cut point identification and welding trajectory automatic fitting method and a storage medium

Abstract

The application discloses a welding cut point recognition and welding track automatic fitting method and a storage medium, and belongs to the technical field of battery manufacturing. Three-dimensional point cloud data of each cross-section contour curve in a welding area of current incoming materials is obtained; welding cut points are extracted according to the three-dimensional point cloud data of each cross-section contour curve, including: R-angle accumulated lines are obtained according to the three-dimensional point cloud data of the cross-section contour curve corresponding to the current incoming materials; welding cut points are obtained according to the R-angle accumulated lines; welding track adaptive fitting is carried out according to the welding cut points corresponding to each cross-section contour curve, so that the welding track corresponding to the current incoming materials is obtained. Through three-dimensional visual imaging and a contour curve cut point recognition algorithm, the complex three-dimensional curvature characteristics of the contact area of the irregular steel shell, the cell shell and the shell cover are adapted; based on the R-angle accumulated curve cut point detection algorithm, the local mutation of the profile caused by the micron-level concave-convex surface is overcome, micro-noise is effectively filtered out, the precise detection of the welding cut point is realized, and the welding track with high precision and high stability is obtained.

Description

Technical Field

[0001] This invention relates to the field of battery manufacturing technology, specifically to a method for identifying welding cut points, automatically fitting welding trajectories, and a storage medium. Background Technology

[0002] Currently, the welding trajectory guidance for steel-cased battery cells mainly adopts a two-dimensional vision solution, which defines the welding path by a fixed distance. However, when key dimensions such as the R-angle and edge position of the casing or housing change, the welding trajectory cannot be adaptively adjusted, resulting in deviations in the welding trajectory.

[0003] Furthermore, the transition area between the flange plane and the radius (R-angle) of the battery cell casing is affected by stamping, surface roughness, and manufacturing tolerances, resulting in a non-ideal curved surface with unevenness at the micron level, rather than a standard plane or a regular arc. Even after filtering and denoising, the scanned laser point cloud cannot eliminate the contour jitter and local distortion caused by the microscopic morphology, and the resulting cross-sectional contour curve is not a completely smooth curve.

[0004] However, existing methods for identifying welding tangent points (FTP points) rely on ideal geometric assumptions. They obtain the tangent point or intersection point by fitting the straight line containing the flange plane and the radius arc as the welding reference point, ignoring the actual microscopic morphology and point cloud contour fluctuations of the steel-shelled battery cell. Furthermore, since irregularly shaped steel-shelled battery cells often have complex geometric shapes, existing methods cannot stably and accurately locate the actual welding tangent point, which in turn affects the generation and fitting accuracy of the welding trajectory. This leads to a misalignment between the generated welding trajectory and the actual weld seam interface of the battery cell, making it difficult to meet the high-precision and high-stability welding process requirements for large-scale steel-shelled battery cells.

[0005] Therefore, there is an urgent need to design a method that can accurately identify welding cutting points and adaptively fit welding trajectories with high precision and high stability. Summary of the Invention

[0006] The main technical problem solved by this invention is how to achieve accurate identification of welding cutting points and adaptive fitting of welding trajectories with high precision and high stability.

[0007] According to the first aspect, one embodiment provides an automatic fitting method for the welding trajectory of an irregularly shaped steel-cased battery cell, comprising:

[0008] Obtain the 3D point cloud data of the profile curve of each section in the welding area of ​​the current incoming material;

[0009] Extracting welding tangent points based on the three-dimensional point cloud data of each cross-sectional profile curve includes: obtaining the R-angle accumulation line based on the three-dimensional point cloud data of the cross-sectional profile curve corresponding to the current incoming material; and obtaining the welding tangent point based on the R-angle accumulation line.

[0010] The welding trajectory is adaptively fitted based on the welding tangent point corresponding to each cross-sectional profile curve to obtain the welding trajectory corresponding to the current incoming material.

[0011] According to a second aspect, one embodiment provides a welding cut point identification method, comprising:

[0012] Obtain the 3D point cloud data of the cross-sectional profile curve in the welding area of ​​the current incoming material;

[0013] Welding tangent points are extracted based on the 3D point cloud data of the cross-sectional profile curve corresponding to the current incoming material, including:

[0014] The R-angle line is obtained based on the three-dimensional point cloud data of the cross-sectional profile curve corresponding to the current incoming material. Each profile point in the cross-sectional profile curve corresponds to a first mapping point in the R-angle line, and each first mapping point corresponds to an index value.

[0015] The R-angle accumulation line is obtained based on the coordinates of each first mapping point in the R-angle line, and each first mapping point in the R-angle line corresponds to a second mapping point in the R-angle accumulation line.

[0016] The R-angle accumulation curve is obtained based on the R-angle accumulation curve, and each second mapping point in the R-angle accumulation curve corresponds to a third mapping point in the R-angle accumulation curve;

[0017] Multiple curvature waves are obtained based on the R-angle cumulative curvature line. Each curvature wave corresponds to a triplet. The triplet is included on the R-angle cumulative curvature line, where the starting position of the curvature wave corresponds to the index value of the third mapping point, the ending position corresponds to the index value of the third mapping point, and the peak position corresponds to the index value of the third mapping point.

[0018] The welding cut point corresponding to the current incoming material is obtained based on the triplet corresponding to each curvature wave.

[0019] According to a third aspect, one embodiment provides a computer-readable storage medium including a program that can be executed by a processor to implement the above-described automatic welding trajectory fitting method and welding cut point identification method for irregularly shaped steel-cased battery cells.

[0020] According to the above embodiments, a welding tangent identification method and automatic welding trajectory fitting method and storage medium combine three-dimensional visual imaging with a contour curve tangent identification algorithm. This method can adapt to the complex three-dimensional curvature characteristics of the contact area of ​​the casing and cover of irregularly shaped steel shell battery cells. It solves the shortcomings of existing two-dimensional visual guided welding methods that cannot capture three-dimensional spatial information and cannot accurately identify the spatial position of feature points. This embodiment is based on a curve tangent detection algorithm of R-angle accumulation. It constructs an R-angle accumulation line based on the cross-sectional contour curve to extract the welding tangent. By first smoothing the cross-sectional contour curve, then constructing the R-angle accumulation line, R-angle accumulation curvature line and curvature wave, and finally determining the welding tangent position by generating segmented approximation straight lines based on adjacent curvature waves, it overcomes the problem of local contour abrupt changes caused by the micron-level concave and convex surfaces of the flange. While preserving macroscopic geometric features, it effectively filters out microscopic noise and achieves accurate detection of the R-angle tangent point (the contact point of the casing and cover), thereby obtaining higher welding trajectory consistency. Attached Figure Description

[0021] Figure 1 A schematic diagram of a closed-loop optimization device for welding irregularly shaped steel-shell battery cells based on 3D vision;

[0022] Figure 2 This is a schematic diagram of a 3D scanning device.

[0023] Figure 3 This is a flowchart of the processor in a closed-loop optimization device for welding irregularly shaped steel shell battery cells based on 3D vision.

[0024] Figure 4 This is a schematic diagram of the cross-sectional profile curve;

[0025] Figure 5 A schematic diagram of the marking points on the base plate of the steel-cased battery cell clamping fixture;

[0026] Figure 6 The flowchart shows the closed-loop optimization method for welding irregularly shaped steel shell battery cells based on 3D vision.

[0027] Figure 7 This is a flowchart of a welding trajectory offset prediction method;

[0028] Figure 8 A flowchart illustrating an automatic fitting method for the welding trajectory of an irregularly shaped steel-shell battery cell;

[0029] Figure 9 This is a flowchart of a welding cut point identification method;

[0030] Figure 10 This is a flowchart of a curve tangent point detection method based on the accumulation of R-angle. Detailed Implementation

[0031] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0032] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0033] The serial numbers assigned to components in this document, such as "first" and "second," are used only to distinguish the described objects and have no sequential or technical meaning. The terms "connection" and "linkage" used in this invention, unless otherwise specified, include both direct and indirect connections (linkages).

[0034] With increasingly stringent global safety requirements for electronic devices, particularly the EU's stricter safety standards for batteries in portable electronic devices, battery cells in 3C electronic devices are gradually shifting from pouch cells to steel-cased cells. Steel-cased cells, due to their superior sealing performance and excellent mechanical protection, have gradually become the preferred battery type for high-precision electronic devices such as smartphones and tablets. However, the casing and cover of steel-cased cells typically employ irregular shapes, which places higher demands on the welding process.

[0035] In the manufacturing process of steel-cased battery cells, the welding of the casing and the cover is a critical step. After welding, subsequent processes such as electrolyte injection and formation are required. Therefore, the accuracy of the welding trajectory directly affects the final quality and safety performance of the battery cell. Currently, the welding trajectory guidance for steel-cased battery cells mainly adopts a two-dimensional vision guidance scheme, which determines the welding path by setting a fixed offset (e.g., a few millimeters from the edge of the cover). The core problem is that it cannot capture the three-dimensional spatial information of the steel-cased battery cell, and can only perform simple feature point extraction and fixed distance calculation based on planar images, resulting in the following significant defects:

[0036] (1) Poor adaptability to batch size tolerance: Since the shell cover and shell (i.e. incoming material) of different batches are usually produced by different manufacturers and using different stamping machines and molds, their size has certain tolerance fluctuations. If the same set of fixed welding trajectory parameters is used across batches, the welding path cannot be adapted to the actual contour and position deviation of the new batch of parts, which can easily cause welding defects such as weld offset, weld deviation, false welding, and undercut. At the same time, the consistency of welding trajectory size deteriorates, causing batch defects and failing to meet the mass production quality requirements.

[0037] Furthermore, due to the complex three-dimensional curvature characteristics of the contact area between the cover and the shell of the irregularly shaped steel shell battery cell (such as the R-corner area), the geometric feature matching method based on two-dimensional vision cannot accurately identify the spatial position of these feature points (welding tangents), resulting in a large error in the extracted feature points. Therefore, when key dimensions such as the R-corner (arc angle) and edge position of the cover or shell change, the two-dimensional vision guidance based on a fixed distance cannot adaptively adjust the welding trajectory, which will also cause deviations in the welding trajectory and a decrease in welding consistency.

[0038] (2) The risks caused by trajectory deviation are high: If the welding trajectory deviates too much outward (outward deviation of the trajectory), the size of the cut cell will exceed the assembly accuracy requirements of the mobile phone battery compartment, which will result in the cell not being able to be properly installed in the device; if the welding trajectory deviates too much inward (inward deviation of the trajectory), the laser may directly penetrate the cell, damage its internal structure, and cause safety hazards such as short circuit and leakage.

[0039] (3) Low conversion efficiency: When the outline shape of the steel shell battery cell changes (e.g., from rectangle to hexagon), the traditional two-dimensional vision guidance solution requires manual teaching, parameter adjustment and batch verification. The entire conversion process is cumbersome and time-consuming, and cannot achieve "one-click conversion", which seriously affects production efficiency and restricts the rapid iteration capability of products.

[0040] (4) Lack of closed-loop control mechanism: Existing technologies generally adopt the offline mode of "welding-sampling inspection-adjustment", which lacks the ability to predict dimensional deviations. It is impossible to predict the dimensional deviation trend of different batches of incoming materials based on historical welding data, so as to achieve forward-looking optimization of welding parameters. Therefore, it is impossible to achieve real-time closed-loop optimization of welding trajectory, which not only easily leads to defective products flowing out, but also makes the process stability poor.

[0041] Furthermore, existing methods typically use straight lines or simple circular arcs to fit welding trajectories when performing trajectory fitting, which cannot adapt to the complex geometry of irregularly shaped steel-cased battery cells, resulting in low fitting accuracy of welding trajectories. Therefore, existing two-dimensional vision-guided solutions have significant shortcomings in terms of size adaptability, welding accuracy, changeover efficiency, and process control, making it difficult to meet the high reliability and high safety manufacturing requirements of steel-cased battery cells.

[0042] Due to the irregular shape of steel-cased battery cells, the dimensional tolerance issues of different batches of incoming materials, and the need for rapid changeover, relying solely on traditional two-dimensional vision guidance technology can no longer meet the requirements for high-quality and high-efficiency steel-cased battery cell welding production. Therefore, developing an intelligent welding trajectory guidance method that can adaptively fit the welding trajectory, optimize welding parameters in real time through closed loop, and support rapid changeover has become an urgent need in the field of steel-cased battery cell manufacturing.

[0043] This invention provides a 3D vision-based approach for guiding and intelligently optimizing the welding trajectory of irregularly shaped steel-shell battery cells. It is applicable to the welding manufacturing process of various irregularly shaped steel-shell battery cells. By combining 3D point cloud processing with a contour curve tangent point detection algorithm, it accurately captures welding tangent points (FTP points). Furthermore, it calibrates the camera coordinate system (corresponding to the laser coordinate system) of the 3D vision sensor (such as a laser profilometer) with the coordinate system corresponding to the welding head (corresponding to the welding coordinate system) using a fixture coordinate system, achieving adaptive planning of the welding trajectory. Further, it uses a deep learning mechanism to provide trajectory offset data feedback, achieving closed-loop trajectory control without the need for repeated manual adjustments to trajectory offset parameters. Based on the welding results of historical products and welding parameters during the welding process, it predicts the trajectory offset data of the current product, thereby correcting the welding trajectory and welding parameters of the current product. This solves the problem of inconsistent welding trajectories caused by the dimensional tolerances of the casing and shell in existing methods, improving the accuracy and safety of battery cell manufacturing. Simultaneously, it greatly improves the efficiency of product changeovers (such as when the contour shape of an irregularly shaped battery cell changes), achieving the goal of "one-click changeover."

[0044] The welding trajectory guidance method in this invention embodiment only requires one device to produce a small batch of any batch of materials (one product per welding station) for trial production. It can quickly and accurately synchronize the guidance trajectory parameters of multiple welding stations and different guiding welding equipment. It can achieve truly efficient, stable and reliable welding trajectory guidance for products of different batches and different contour shapes. It completely solves a series of problems such as long production time when changing batches (from different stamping dies) and changing shapes (changes in contour shape) on the production site, as well as poor process capability caused by welding trajectory deviation and poor consistency of welding trajectory dimensions of welding stations or production equipment. Ultimately, it ensures the consistency and high reliability of battery cell size.

[0045] To ensure compatibility with batch dimensional tolerances of different incoming materials and achieve adaptive and precise fitting of the welding trajectory corresponding to the shell and cover, this invention calibrates the coordinate transformation relationship between the laser coordinate system of a 3D vision scanning device (such as a laser profilometer) and the welding coordinate system corresponding to the welding head used for welding the current incoming material, based on the clamping fixture reference. This ensures precise matching between the welding trajectory and the contour of the irregularly shaped battery cell (incoming material), thereby improving the consistency of battery cell welding dimensions between batches and solving the problem of welding trajectory deviation caused by batch dimensional tolerances in traditional 2D vision calibration and welding guidance methods. Furthermore, after changing to different batches, there is no need for manual parameter switching or trajectory offset parameter adjustment for optimization, which significantly shortens changeover time and greatly improves production efficiency. The specific calibration process is as follows:

[0046] Multiple marker points are set on each side of the base plate of the steel-cased battery cell clamping fixture, as shown in the schematic diagram. Figure 5 As shown, a planar coordinate system (i.e., fixture coordinate system) is established on the base plate of the steel-shell battery cell clamping fixture using a high-precision two-dimensional instrument (such as an image measuring instrument). The center of all marked points and the corresponding coordinates of the obtained center in the planar coordinate system are obtained. Each marked point is measured repeatedly (e.g., 10 times), and the average coordinate of the obtained measurement results is calculated and used as the fixture coordinate P0 corresponding to the marked point in the fixture coordinate system. Then, a three-dimensional vision scanning device is used to scan the steel-shell battery cell clamping fixture to obtain an image with marked points. Similarly, the center of each marked point in the obtained image and the image coordinates of the obtained center in the image coordinate system are obtained. Based on the image coordinates of each marked point and the fixture coordinates, the coordinate transformation relationship (i.e., the first coordinate transformation relationship) between the laser coordinate system (image coordinate system) corresponding to the three-dimensional vision scanning device and the fixture coordinate system is calculated.

[0047] Then, calibration paper is placed inside the steel-shell battery cell clamping fixture. The welding head is near the marked point of the steel-shell battery cell clamping fixture, and the calibration paper is burned to punch a circular through hole. The coordinates P1 (i.e., welding coordinates) of the center position of the hole in the welding coordinate system are recorded. Then, the marked point on the steel-shell battery cell clamping fixture and the through hole on the calibration paper are scanned by a three-dimensional vision scanning device to obtain the fixture coordinates P2 of all through holes in the fixture coordinate system. Then, the coordinate transformation relationship between the welding coordinate system of the welding head and the fixture coordinate system (i.e., the second coordinate transformation relationship) is obtained by using the welding coordinates corresponding to each through hole and the fixture coordinates.

[0048] Finally, the calibration accuracy is verified by rescanning the marked points on the steel-shell battery cell clamping fixture and the through holes on the calibration paper using a 3D vision scanning device. Based on the first and second coordinate transformation relationships, the welding coordinates P3 corresponding to each through hole in the welding coordinate system in the image acquired by the 3D vision scanning device are calculated. It is then determined whether the deviation between the welding coordinates P3 of each through hole obtained through coordinate transformation and the welding coordinates P2 of each through hole in the welding coordinate system during drilling is within the allowable error range to determine whether the calibration accuracy meets the calibration requirements. If it is within the allowable error range, it means that the calibration result meets the calibration requirements and the calibration ends; otherwise, it means that the calibration result does not meet the calibration requirements and the calibration process needs to be checked and recalibrated.

[0049] Please refer to Figure 1 Some embodiments provide a closed-loop optimization device for welding irregularly shaped steel-shell battery cells based on three-dimensional vision, including a laser-guided welding unit 100, a transmission unit 110, an optical detection unit 120, and a processor 130; wherein:

[0050] The laser-guided welding unit 100 is used to perform laser-guided welding on the incoming material. In this embodiment, the devices included in the laser-guided welding unit 100 from top to bottom are, in order, a welding head (such as a galvanometer laser welding head), a steel shell battery cell clamping fixture, a three-dimensional vision scanning device (such as a laser profilometer), and a motion control module (XYRZ motion control module).

[0051] The welding head is used to weld the incoming material after determining the welding trajectory (target welding trajectory); the steel shell battery cell clamping fixture is used to clamp the incoming material; the three-dimensional vision scanning device is used to scan the point cloud data of the welding area of ​​the incoming material; and the motion control module is used to drive the three-dimensional vision scanning device to move according to the scanning trajectory.

[0052] Specifically, the welding head is located on top of the steel-shell battery cell cover and is used to perform welding above the cover in the current incoming material according to the target welding trajectory. The steel-shell battery cell clamping fixture can push the current incoming material to the reference position and clamp it, used to hold the current incoming material (shell and cover). The flange planes on each side of the shell contact the fixture. The 3D vision scanning device scans the welding area of ​​the current incoming material from below. The welding area of ​​the current incoming material is the welding R-angle area corresponding to each edge of the incoming material, that is, the area formed by the flange plane and the shell wall in each edge of the incoming material. A schematic diagram of the scan is shown below. Figure 2As shown in the figure, the field of view of the three-dimensional vision scanning device represents the size of the spatial range captured by the three-dimensional vision scanning device. The number of three-dimensional vision scanning devices can be one or more. In this embodiment, two three-dimensional vision scanning devices are set up, which are located on the motion control module. The motion control module drives the three-dimensional vision scanning devices to scan the welding area of ​​the current incoming material according to the scanning trajectory.

[0053] The conveying unit 110 is used to convey the product obtained after welding the current incoming material in the laser-guided welding unit 100 to the optical inspection unit 120; for example, the product (battery cell product) obtained after welding can be conveyed to the optical inspection unit 120 by a conveyor belt.

[0054] The optical inspection unit 120 is used to perform optical inspection on the product (i.e., battery cell product) obtained after the current incoming material is welded. In this embodiment, the devices included in the optical inspection unit 110 from top to bottom are a two-dimensional imaging module, a steel shell battery cell clamping fixture, a three-dimensional vision scanning device (such as a laser profilometer) and a motion control module (XYRZ motion control module).

[0055] The two-dimensional imaging module is used to acquire images of the top of the product, obtaining a top image of the product, which includes the welding trajectory image of the battery cell casing. The steel-cased battery cell clamping fixture is used to clamp the product, and the flange planes on each side of the product casing are in contact with the fixture. The three-dimensional vision scanning device is used to scan the welding area of ​​the product using point cloud data. The number of three-dimensional vision scanning devices can be set according to requirements. In this embodiment, the optical detection unit 110 is equipped with two three-dimensional vision scanning devices. The motion control module is used to drive the three-dimensional vision scanning devices to move according to the scanning trajectory to scan the welding area of ​​the product.

[0056] For ease of distinction, in this embodiment, the motion control module in the laser welding guidance unit 100 is referred to as the first motion control module, and the three-dimensional vision scanning device in this unit is referred to as the first scanning device. Based on the scanning data of the first scanning device, the welding cutting point is identified, and then the welding trajectory (initial welding trajectory) is fitted, and the initial offset data is obtained. The motion control module in the optical detection unit 120 is referred to as the second motion control module, and the three-dimensional vision scanning device in this unit is referred to as the second scanning device. Based on the scanning data of the second scanning device, the position of the shell wall is determined, and the distance from the weld to the shell wall is calculated in combination with the two-dimensional imaging module.

[0057] And, processor 130, its processing flow is as follows Figure 3 As shown, it includes:

[0058] Step S131: Obtain the three-dimensional point cloud data of the profile curve of each section in the welding area of ​​the current incoming material.

[0059] The contour map corresponding to the current incoming material is obtained. Based on the coordinates of each edge of the current incoming material in the contour map, the scanning trajectory of the three-dimensional vision scanning device is obtained. In this embodiment, the three-dimensional vision scanning device (i.e., the first scanning device) is driven by the first motion control module to perform point cloud data scanning on the welding area of ​​the current incoming material, and the three-dimensional point cloud data corresponding to each cross-sectional contour curve in the welding area of ​​the current incoming material is obtained. When the incoming material is changed, the contour map corresponding to the steel shell cell to be welded is automatically loaded, the contour shape of the current incoming material is automatically identified, and the scanning trajectory corresponding to the three-dimensional vision scanning device is quickly generated.

[0060] For example, the first motion control module acquires the outline CAD drawing of the current incoming material. Based on the pre-calibrated CAD coordinate system and the motion control coordinate system of the 3D vision scanning device (such as a laser profilometer, structured light camera, or other 3D vision sensors), the first motion control module parses the outline coordinate information in the outline CAD drawing corresponding to the current incoming material and converts it into a motion control trajectory in the motion control coordinate system of the first scanning device. This automatically plans the scanning trajectory (or motion control trajectory) corresponding to the 3D vision scanning device. Based on the obtained scanning trajectory, the first motion control module is controlled to scan the outer outline of the current incoming material, thereby obtaining the 3D point cloud data corresponding to each cross-sectional outline curve in the welding area of ​​the current incoming material.

[0061] The cross-sectional profile curve obtained in this embodiment is as follows: Figure 4 As shown in the figure, the green area corresponds to the steel-shell battery cell clamping fixture, and the circular protrusions in this area correspond to the marked points on the fixture; the blue area corresponds to the flange plane, and the brown area corresponds to the shell wall of the incoming material. The curve in the figure represents the depth change from the shell to the flange plane and then to the steel-shell battery cell clamping fixture. To ensure the accuracy of the identification of the welding cut point (corresponding to the black dot in the figure), this embodiment further sets a schematic point (corresponding to the green dot in the figure) in the detection area where the welding cut point is located, so as to ensure that the final obtained welding cut point is within the detection area, that is, the identified welding cut point should be between the first green dot and the second green dot from left to right.

[0062] Step S132: Obtain the initial welding trajectory corresponding to the current incoming material based on the three-dimensional point cloud data of each cross-sectional profile curve corresponding to the current incoming material.

[0063] Because the housing is made of stainless steel, which is a reflective metal surface, the laser source of the laser profilometer illuminates the scanned area, resulting in abnormal reflections and stray light. This leads to abnormal data in the point cloud reconstruction (i.e., point cloud data that does not belong to the surface contour of the housing). Therefore, this embodiment requires filtering the acquired 3D point cloud data first to improve the accuracy of subsequent cross-sectional contour curve processing.

[0064] In this embodiment, the welding tangent point corresponding to each cross-sectional contour curve is determined based on the three-dimensional point cloud data corresponding to each cross-sectional contour curve; and the initial welding trajectory corresponding to the current incoming material is obtained based on the welding tangent point corresponding to each cross-sectional contour curve.

[0065] In some embodiments, the welding tangent point identification corresponding to each cross-sectional profile curve can be obtained by fitting the tangent point or intersection point of the straight line containing the flange plane and the radius arc. For example, based on the curvature of each point in the three-dimensional point cloud data corresponding to the cross-sectional profile curve, the set of straight line segments and the set of circular arc segments in the cross-sectional profile curve are selected. For the set of straight line segments, the straight line equation is fitted using a straight line fitting algorithm, such as the least squares method or the RANSAC algorithm, to obtain the straight line corresponding to the shell wall and the straight line corresponding to the flange plane. For the set of circular arc segments, the circle is fitted using the least squares method to obtain the center coordinates and radius of the circle corresponding to each point. According to the distance between the center coordinates of each point and the straight line corresponding to the flange plane, theoretically, if the distance is equal to the radius of the fitted circle corresponding to the point, then the point is considered to be the welding tangent point corresponding to the cross-sectional profile curve. However, in reality, if the straight line corresponding to the flange plane and the fitted circle are not perfectly tangent, then the method is to obtain the intersection point of the straight line corresponding to the flange plane and the fitted circle, and then obtain the corresponding welding tangent point.

[0066] Considering that the flange plane and the R-angle region present uneven curved surfaces at the microscopic (micrometer level), the corresponding cross-sectional profile curve is not completely smooth even after abnormal data filtering. Therefore, this embodiment uses a curve tangent detection method based on R-angle accumulation to identify the welding tangent points in each cross-sectional profile curve. First, the cross-sectional profile curve is smoothed, then an R-angle accumulation line, an R-angle accumulation line curvature line, and a curvature wave are constructed. Finally, the welding tangent point position is obtained by finding the intersection point of adjacent segmented approximation lines. The specific implementation method is explained in the welding tangent point identification method below, and will not be repeated here.

[0067] After obtaining the welding tangent points corresponding to each cross-sectional contour curve, the welding tangent points are transformed into the welding coordinate system by using the coordinate transformation relationship between the laser coordinate system of the pre-calibrated three-dimensional vision scanning device and the welding coordinate system corresponding to the welding head used to weld the current incoming material. Then, the Bezier curve fitting algorithm is used to fit the trajectory of the coordinate-transformed welding tangent points to obtain the initial welding trajectory of the current incoming material.

[0068] Step S133: Obtain the initial offset data corresponding to the current incoming material, and / or obtain the predicted offset data corresponding to the current incoming material.

[0069] Initial offset data is used to characterize the deviation corresponding to the initial welding trajectory of the current incoming material. In this embodiment, the initial welding trajectory corresponding to the current incoming material includes the initial welding trajectory corresponding to each edge of the current incoming material, and each edge of the current incoming material can be determined according to its corresponding contour map. Based on the offset of the initial welding trajectory corresponding to each edge of the incoming material relative to the shell wall of the corresponding edge, the initial offset data corresponding to the current incoming material is obtained, specifically:

[0070] The shell wall position corresponding to each incoming material edge is determined based on the three-dimensional point cloud data corresponding to each cross-sectional profile curve. For example, the set of straight line segments in the cross-sectional profile curve is filtered out based on the curvature of each point in the three-dimensional point cloud data corresponding to the cross-sectional profile curve. At this time, the set of straight line segments includes the set of points corresponding to the flange plane and the set of points corresponding to the shell wall. Based on the distribution characteristics of the flange plane and the shell wall, the set of points corresponding to the shell wall in the cross-sectional profile curve can be directly determined, thereby determining the position of the shell wall.

[0071] Then, based on the initial welding trajectory corresponding to each edge of the incoming material and the position of the shell wall, the initial offset data corresponding to the current incoming material is obtained.

[0072] For example, for any incoming material edge: obtain the shell wall position corresponding to all cross-sectional profile curves corresponding to the incoming material edge. For example, the point with the largest vertical coordinate in the set of points corresponding to the shell wall represents the shell wall position corresponding to the cross-sectional profile curve, and perform straight line fitting. By comparing the obtained shell wall straight line with the initial welding trajectory corresponding to the incoming material edge in the initial welding trajectory, the offset of the initial welding trajectory corresponding to the incoming material edge relative to the shell wall straight line is obtained, and then the initial offset data corresponding to the incoming material edge is obtained.

[0073] The initial offset data in this embodiment includes one or more of the following: initial offset direction, initial offset distance, and initial rotation angle corresponding to each edge of the incoming material; wherein, the initial offset direction corresponding to each edge of the incoming material includes inward offset or outward offset of the welding trajectory, for example, 1 or -1 can be used to represent the initial offset direction; the initial offset distance corresponding to each edge of the incoming material represents the distance by which the initial welding trajectory corresponding to the incoming material edge deviates from the shell wall; the initial rotation angle corresponding to each edge of the incoming material represents the rotation angle of the initial welding trajectory corresponding to the incoming material edge relative to the shell wall, for example, -1 or 1 can be used to characterize whether the rotation angle is clockwise or counterclockwise.

[0074] The predicted offset data is the predicted deviation obtained by predicting the trajectory offset of the current incoming material using a pre-trained prediction model. In this embodiment, the predicted offset data includes one or more of the following: predicted offset direction, predicted offset distance, and predicted rotation angle for each edge of the current incoming material. The predicted offset direction represents the possible offset direction of the predicted welding trajectory corresponding to the edge of the incoming material relative to the shell wall; the predicted offset distance represents the distance of the predicted welding trajectory corresponding to the edge of the incoming material offset from the shell wall; and the predicted rotation angle represents the rotation angle of the predicted welding trajectory corresponding to the edge of the incoming material relative to the straight line of the shell wall.

[0075] It should be noted that since different batches of incoming materials may come from different manufacturers or different molds from the same manufacturer, there are certain dimensional tolerances between different batches. To improve the adaptability to the dimensional tolerances of the corresponding batches of shells and covers, this embodiment uses a pre-trained prediction model to predict the dimensional deviation trend of the current incoming material based on the welding parameters used in the welding process of the previous incoming material and the welding dimensions after welding, in order to further improve welding accuracy and welding consistency and achieve forward-looking optimization of welding parameters. The prediction model includes three parts: a multi-head attention module, a spatial feature extraction module, and a temporal feature extraction module. Its specific implementation is described in the following section on a welding trajectory offset prediction method, and will not be repeated here.

[0076] Step S134: Correct the initial welding trajectory based on the initial offset data and / or the predicted offset data corresponding to the current incoming material to obtain the target welding trajectory corresponding to the current incoming material.

[0077] This embodiment determines whether the current incoming material meets the preset compensation conditions based on the initial offset data and / or the predicted offset data corresponding to the current incoming material. Since each incoming material edge generates a set of initial offset data and a set of predicted offset data, the deviation vector corresponding to the current incoming material is obtained by comparing the deviation between the corresponding parameters of the initial offset data and the predicted offset data, and it is determined whether the initial offset trajectory needs to be corrected. Specifically, the deviation vector between the initial offset data and the predicted offset data is calculated; it is determined whether the obtained deviation vector is less than or equal to a preset threshold. If so, it is determined that the current incoming material meets the preset compensation conditions.

[0078] In this embodiment, for any incoming material edge of the current incoming material: the difference between each type of initial offset data corresponding to the current incoming material at the incoming material edge and the predicted offset data of the current incoming material at the incoming material edge of the same type are obtained respectively; if the absolute value of the difference corresponding to each type of initial offset data is less than or equal to its corresponding preset threshold, it is considered that the current incoming material at the incoming material edge meets the preset compensation condition; otherwise, it is considered that the current incoming material at the incoming material edge does not meet the preset compensation condition.

[0079] For example, the initial offset direction and the predicted offset direction in the initial offset data should be the same, that is, the preset threshold corresponding to the offset direction is 0; the distance deviation between the initial offset distance and the predicted offset distance, and the angle deviation between the initial rotation angle and the predicted rotation angle can be set according to actual needs. For example, the preset threshold corresponding to the offset distance can be set to 0.5μm, and the preset threshold corresponding to the rotation angle can be set to 5°.

[0080] If the current incoming material meets the preset compensation conditions at all incoming material edges, it is considered to meet the preset compensation conditions. At this time, the initial welding trajectory is corrected to obtain the target welding trajectory corresponding to the current incoming material. Specifically, the predicted offset data or initial offset data corresponding to the current incoming material can be used as offset compensation data. The initial welding trajectory corresponding to the current incoming material is corrected based on the offset compensation data to obtain the target welding trajectory corresponding to the current incoming material.

[0081] For example, the initial offset data is used as offset compensation data. The initial welding trajectory corresponding to each incoming material edge is adjusted as a whole by using the initial offset data (initial offset direction, initial offset distance, and initial rotation angle) corresponding to each incoming material edge. This achieves secondary correction of the initial welding trajectory, thereby obtaining the target welding trajectory corresponding to the current incoming material. Then, the obtained target welding trajectory is sent to the welding head. Based on the coordinate transformation relationship between the laser coordinate system of the three-dimensional vision scanning device and the welding coordinate system of the welding head, the target welding trajectory is transformed into the welding coordinate system. The welding head then welds the current incoming material according to the target welding trajectory after coordinate transformation.

[0082] If one of the incoming material edges does not meet the preset compensation conditions, it indicates that there may be an abnormality in the incoming material or that the incoming material is not properly clamped. Therefore, it is considered that the incoming material does not meet the preset compensation conditions, and an alarm needs to be triggered to prompt the operator to handle the situation.

[0083] Step S135: Obtain the dimensional measurement data of the current incoming material after welding according to the target welding trajectory.

[0084] The product (i.e., the battery cell product) obtained after welding the incoming material is transmitted to the optical inspection unit 120, where the product's dimensions are measured to obtain dimensional measurement data. This data includes the weld dimensions, which include one or more of the following: weld distance between opposite sides, weld width, weld angle, radius of curvature, and distance from the inner side of the weld corresponding to each edge to the shell wall. Specifically, the weld distance between opposite sides refers to the distance between the center lines of two welds on a set of opposite sides of the shell; the weld width refers to the width of the weld covering the shell surface; the weld angle refers to the angle between the center lines (or edge lines) of two intersecting or adjacent welds; the radius of curvature refers to the radius of curvature of the smooth transition arc at the junction of the weld surface and the base material (shell or cover); and the distance from the inner side of the weld corresponding to each edge to the shell wall refers to the shortest vertical distance from the inner edge of the weld on the cover (the side closest to the shell center) to the outer surface of the main shell wall.

[0085] In this embodiment, the dimensional measurement data of the incoming material after welding according to the target welding trajectory is obtained through the following method:

[0086] The top image of the product is obtained through a two-dimensional imaging module. Based on the outline of the incoming material, the scanning trajectory of the three-dimensional vision scanning device (second scanning device) is obtained. The second motion control module drives the second scanning device to scan the welding area of ​​the product, and the point cloud data corresponding to the welding area of ​​the product is obtained. According to the top image of the product and the point cloud data corresponding to the welding area of ​​the product, the dimensional measurement data is obtained. Among them, the shell wall position corresponding to each edge of the product is determined according to the obtained point cloud data, and the distance from the inner side of the weld to the shell wall of each edge is determined in combination with the top image.

[0087] In addition, the processor 130 will also perform defect detection on the weld seams based on the top image of the product, including whether the weld seam color is abnormal (e.g., yellowing of the weld seam is an appearance defect) and defects on the shell surface (such as dirt and scratches). Based on the product's size measurement data, combined with the upper and lower limits of the specifications of each monitoring item, it will determine whether the product's welding trajectory belongs to a defective product. If the product is a defective product, it needs to be rejected to prevent defective products from flowing out.

[0088] Step S136: Based on the dimensional measurement data, predict the predicted offset data of the next incoming material using the prediction model.

[0089] Historical compensation data is acquired. In this embodiment, the historical compensation data includes the dimensional measurement data of the product obtained after the current incoming material is welded, the initial offset data of the product, the reference process capability index set of the product, and the batch dimensional tolerance of the current incoming material. The reference process capability index set (CPK data) is predicted by a prediction model based on the dimensional measurement data of the product obtained after the previous incoming material is welded. Each detection item of the dimensional measurement data corresponds to a reference process capability index, and the reference process capability indices corresponding to all detection items constitute the reference process capability index set.

[0090] It should be noted that the reference process capability index set corresponding to the current product is actually a set of predicted results of the process capability index of the product obtained after the previous material was welded. In this embodiment, the predicted results of the process capability index set corresponding to the previous material are used as the reference process capability index set corresponding to the current material. Therefore, it can be considered that the reference process capability index set corresponding to the current material is predicted by the prediction model based on the dimensional measurement data of the product obtained after the previous material was welded.

[0091] The processor 130 in this embodiment is also used to receive the laser process parameters of the laser-guided welding unit 100 during the welding process of the current incoming material. For example, the welding process parameters include laser power, pulse frequency, welding speed, defocusing amount, shielding gas flow rate, etc.

[0092] Modal decomposition is performed on historical compensation data to obtain multiple trajectory deviation modal components. In this embodiment, the historical compensation data is decomposed into several stationary modal components through modal decomposition, so that the center frequency of each component is independent, there is no aliasing, and the characteristics are clear, eliminating noise and modal aliasing. Thus, only the effective components are input into the prediction model to improve the prediction accuracy of the model.

[0093] Input all trajectory deviation modal components of the current incoming material and welding process parameters into the prediction model to obtain the predicted offset data corresponding to the next incoming material output by the prediction model.

[0094] It should be noted that after inputting all trajectory deviation modal components of the current incoming material and welding process parameters into the prediction model, the output of the prediction model also includes welding process optimization parameters and a reference process capability index set for the next incoming material. The obtained welding process optimization parameters are used to optimize the welding process parameters in the welding process of the next incoming material. Then, the predicted offset data of the next incoming material and the welding process optimization parameters are fed back to the laser-guided welding unit 100.

[0095] For example, assuming the current incoming material is the i-th material in the corresponding batch, and the i-th material yields the i-th product after welding, then the historical compensation data corresponding to the i-th product, including the dimensional measurement data, offset compensation data (i.e., initial offset data), reference process capability index set, and batch dimensional tolerance of the batch to which the i-th material belongs, is used. After modal decomposition of the historical compensation data corresponding to the i-th product, the resulting trajectory deviation modal components and the welding process parameters of the i-th product during the welding process are input into the instrument. In the prediction model, the next incoming material (or product) is predicted to obtain the predicted offset data, reference process capability index set, and welding process optimization parameters corresponding to the (i+1)th incoming material. Subsequently, in the laser-guided welding unit 100, the trajectory of the (i+1)th incoming material is corrected based on the predicted offset data and the initial offset data corresponding to the (i+1)th incoming material, and welding is performed on the (i+1)th incoming material based on the welding process optimization parameters corresponding to the (i+1)th incoming material. In the optical detection unit 120, the next incoming material (i.e., the (i+2)th incoming material) is predicted based on the reference process capability index set corresponding to the (i+1)th incoming material.

[0096] Specifically, if the current incoming material is the first material in a batch, the reference process capability index set corresponding to the current incoming material is the reference process capability index set output by the prediction model corresponding to the last material in the previous batch; if the current incoming material is the first material in the first batch, its corresponding reference process capability index set is a dataset composed of CPK base values. For example, the CPK base value can be set to 1.33.

[0097] It should be noted that in this embodiment, the initial offset data corresponding to the current product is input into the prediction model as offset compensation data. If the offset compensation data is set as the predicted offset data corresponding to the product, then the predicted offset data corresponding to the product needs to be input into the prediction model.

[0098] This embodiment is the first to combine 3D vision with AI model closed-loop optimization for steel-shell battery cell welding. A 3D vision scanning device acquires the 3D spatial information of the shell and casing, and a predictive model is used to form a complete closed-loop optimization system. This enables adaptive adjustment of the welding trajectory and welding parameters, solving the problem that existing methods cannot handle the complex geometric features of irregular contours. By analyzing data such as process capability index, the dimensional deviation trends of different batches of incoming materials are predicted, enabling forward-looking optimization of welding parameters. This improves the CPK data value of the battery cell from less than 1.0 to over 1.67, solving the problem of poor welding trajectory consistency caused by dimensional tolerances in multiple batches of incoming materials. Furthermore, this embodiment reduces the risk of laser penetration to 0% by precisely controlling the distance from the welding trajectory to the casing wall, while ensuring that the battery cell dimensions meet assembly requirements, thus improving welding quality and safety. Furthermore, this embodiment matches the welding trajectory with the edge contour of the incoming material and fine-tunes the welding trajectory based on the prediction results of the prediction model. There is no need for manual teaching and parameter adjustment. Only a small number of samples need to be produced for trial production to complete the changeover, which shortens the changeover time by more than 90%. This achieves one-click changeover and greatly improves the flexible production capacity of the production line. It is suitable for the welding manufacturing process of various irregular steel shell battery cells.

[0099] Please refer to Figure 6 Some embodiments provide a closed-loop optimization method for welding irregularly shaped steel shell battery cells based on three-dimensional vision, including the following steps:

[0100] Step S200: Obtain the three-dimensional point cloud data of the profile curve of each section in the welding area of ​​the current incoming material.

[0101] Step S210: Obtain the initial welding trajectory corresponding to the current incoming material based on the three-dimensional point cloud data of each cross-sectional contour curve corresponding to the current incoming material.

[0102] Step S220: Obtain the initial offset data corresponding to the current incoming material, and / or obtain the predicted offset data corresponding to the current incoming material; wherein the initial offset data is used to characterize the deviation corresponding to the initial welding trajectory of the current incoming material; the predicted offset data is the predicted deviation obtained by predicting the trajectory offset of the current incoming material through a pre-trained prediction model.

[0103] Step S230: Correct the initial welding trajectory based on the initial offset data and / or the predicted offset data corresponding to the current incoming material to obtain the target welding trajectory corresponding to the current incoming material.

[0104] Step S240: Obtain the dimensional measurement data of the current incoming material after welding according to the target welding trajectory.

[0105] Step S250: Based on the dimensional measurement data, predict the predicted offset data of the next incoming material using the prediction model.

[0106] It should be noted that the various method steps in this embodiment correspond to the processing flow of the processor in the above-mentioned closed-loop optimization device for welding irregular steel shell battery cells based on three-dimensional vision. The specific implementation methods have been described in the above embodiments and will not be repeated here.

[0107] Please refer to Figure 7 Some embodiments provide a welding trajectory offset prediction method, including the following steps:

[0108] Since existing technologies can only handle single-batch size deviations and cannot optimize parameters for multiple batches, nor can they predict the size deviation trends of different batches of incoming materials based on historical product welding data, they cannot achieve forward-looking optimization of welding parameters. This embodiment addresses the problem of poor consistency in welding trajectories caused by size tolerances of multiple batches of incoming materials by analyzing the size measurement data and process capability index related data of different batches of battery cell products (including defective products), and generating a training dataset by combining the welding process parameters of the battery cell products during the welding process. The preset prediction model is then trained based on each set of training data in the training dataset.

[0109] Each training dataset corresponds to a product. For example, the j-th training dataset includes historical compensation data for the j-th product (including the product's dimensional measurement data, welding trajectory deviation data, reference process capability index set, and batch dimensional tolerance of the incoming material) and welding process parameters during the welding process. The welding trajectory deviation data consists of one or more of the following: the offset direction, offset distance, and rotation angle between the actual welding trajectory corresponding to each edge of the product and the shell wall corresponding to that edge. The reference process capability index set for the product is obtained based on the dimensional measurement data of historical products. Specifically, the process capability index corresponding to each inspection item is calculated based on the dimensional measurement data of multiple historical products. The reference process capability index set for the product is obtained based on the process capability indices corresponding to all inspection items.

[0110] The prediction model used in this embodiment includes a multi-head attention module, a spatial feature extraction module, and a temporal feature extraction module, which are used to reconstruct each input trajectory deviation mode component and output the reconstructed trajectory deviation mode components and welding process optimization parameters; based on the reconstructed trajectory deviation mode components, the predicted offset data of the next product and the reference process capability index set are obtained.

[0111] The multi-head attention mechanism module is used to extract the correlation between trajectory deviation mode components and welding process parameters. Specifically, the multi-head attention mechanism module performs cross-processing on trajectory deviation mode components and welding process parameters to mine feature correlations. It establishes a deep connection between weakly correlated features and offset errors by leveraging the relationship between strongly correlated features and offset errors, avoiding reduced prediction accuracy due to data waste. For example, the multi-head attention mechanism module can perform feature extraction using multi-scale convolution, nonlinear activation, and pooling layers, and achieve multi-head attention feature fusion through channel attention and positional attention. Alternatively, it can generate query, key, and value matrices through fully connected layers, dividing the resulting matrices into multiple parts along the feature dimensions, with the number of parts representing the number of attention heads (or attention branches). Each computational branch independently performs scaled dot product attention on its respective query submatrix, key submatrix, and value submatrix. The results of all attention branches are concatenated and fused through linear transformation to obtain the output of the multi-head attention mechanism module. The module outputs spatial features from the multi-head attention mechanism module. For example, a convolutional neural network can be used to extract spatial features from the output of the multi-head attention mechanism module. Multi-scale one-dimensional convolutions slide along the variable dimension to extract local spatial relationships between different feature combinations. A non-linear activation function introduces expressive power, pooling layers compress the variable dimension and expand the spatial receptive field, 1×1 convolutions adjust the number of channels and fuse cross-channel features, and finally, global average pooling aggregates all variable information to complete spatial feature extraction. The temporal feature extraction module extracts time-series trends from the output of the spatial feature extraction module. For example, a long short-term memory neural network can be used to extract temporal features from the output of the spatial feature extraction module. Multi-scale convolutions extract local change patterns across different time spans, a non-linear activation function introduces expressive power, pooling layers compress the temporal resolution and expand the receptive field, and finally, a long short-term memory network models long-range temporal dependencies to complete temporal feature extraction.

[0112] During the training process of the prediction model, the training data in the training dataset are first input into the prediction model in sequence. The prediction model is then trained for the first time based on the actual welding results corresponding to the training data and the obtained prediction results. After the initial training, the error between the prediction result and the actual welding result corresponding to each set of training data is obtained and used as the new label for that set of training data, thereby generating a new training dataset. The new training dataset is then input into the prediction model again, and the prediction model is retrained based on the prediction results of each set of training data at this time and the corresponding new label, resulting in a fully trained prediction model.

[0113] Based on the trained prediction model, the trajectory offset data and welding process parameters are predicted. While ensuring the consistency of welding trajectories for different batches of incoming materials, real-time closed-loop optimization of the welding trajectory and forward-looking optimization of welding parameters are achieved. The specific process is as follows:

[0114] Step S300: Obtain historical compensation data, including the current product's size measurement data, the current product's offset compensation data, the reference process capability index set corresponding to the current product, and the batch size tolerance of the incoming material corresponding to the current product. The reference process capability index set is predicted by a pre-trained prediction model based on the size measurement data corresponding to the previous product.

[0115] Step S310: Obtain the welding process parameters of the current product during the welding process.

[0116] Step S320: Perform modal decomposition on the historical compensation data to obtain multiple trajectory deviation modal components.

[0117] It should be noted that before training the prediction model, it is necessary to first determine the globally optimal parameter combination through key parameter optimization, and determine the optimal number of decomposition layers (i.e., the optimal number of modes) and the penalty factor before training the prediction model. In the actual detection process, the historical compensation data corresponding to the current product can be directly decomposed into stationary mode components based on the optimal mode data to obtain the corresponding number of trajectory deviation mode components.

[0118] Specifically, the historical compensation data is adaptively decomposed into modal components with the number of optimal modes by mode decomposition (VMD), so that the center frequencies of each component are independent, non-overlapping and have clear characteristics. The trend term, periodic term and random noise are separated, and only the effective components (corresponding to the trajectory deviation modal components) are retained and input into the prediction model.

[0119] It should be noted that, in this embodiment, when optimizing key parameters, the corresponding optimization objects include the number of heads in the multi-head attention mechanism (i.e., the number of parallel independent attention branches), the number of convolution kernels in the spatial feature extraction module (such as the number of convolution kernels in a CNN), the pooling size, the hidden layer dimension in the temporal feature extraction module (such as the hidden layer dimension in a long short-term memory neural network), as well as the time step and learning rate. Specifically, firstly, the population is initialized in the solution space of the optimization objects, with the fitness process capability index CPK being greater than or equal to 1.67 as the fitness optimization target. The quality of the initial population is improved by using the opposition search operator, and the global exploration and local development are balanced by the nonlinear convergence operator. Iterative optimization is then performed until convergence, resulting in the globally optimal parameter combination.

[0120] Modality number optimization aims to find the number of decomposition layers with no modal aliasing and minimal reconstruction error. Penalty factor optimization controls the convergence accuracy and smoothness of the decomposition, avoiding over- or under-decomposition. Specifically, during modality number optimization, permutation entropy is used as the fitness function to evaluate the stability of the decomposed signal. The search iterates through the range of decomposition layers (typically 2 to 12). The stopping criterion is: optimization stops when all modes are free of aliasing, permutation entropy is minimized, and reconstruction error is less than a preset error threshold (e.g., 0.005 mm). The modality number at this point is the optimal modality number.

[0121] Step S330: Input all trajectory deviation modal components and welding process parameters of the current product into the prediction model to obtain the predicted offset data, reference process capability index set and welding process optimization parameters of the next product.

[0122] Among them, the predicted offset data of the next product is used to correct the welding trajectory of the next product, and the welding process optimization parameters corresponding to the next product are used to optimize the welding process parameters during the welding process of the next product.

[0123] It should be noted that the method steps of predicting trajectory offset data (corresponding to predicted offset data) and welding process parameters (welding process optimization parameters) based on the pre-trained prediction model in this embodiment correspond to step S136 in the processor processing flow of the above-mentioned three-dimensional vision-based closed-loop optimization device for welding irregular steel shell battery cells. The specific implementation method has been described in the above embodiments and will not be repeated here.

[0124] This embodiment achieves full utilization of information and obtains more accurate prediction results through hierarchical progressive prediction, enabling forward-looking optimization of welding parameters and further improving welding accuracy and consistency. The process capability index value is increased from less than 1.0 to greater than or equal to 1.67. By predicting the process capability index in real time through the prediction model, quality control is transformed from passive post-event inspection to proactive pre-event prevention. By instantly capturing minute fluctuations in the production process, early warnings are issued before dimensional deviations exceed specification limits, thereby effectively avoiding the generation of batches of non-conforming products, significantly reducing rework and scrap costs, and ensuring cell size consistency and high reliability.

[0125] Please refer to Figure 8 Some embodiments provide an automatic fitting method for the welding trajectory of irregularly shaped steel-cased battery cells, including:

[0126] Due to the influence of stamping, surface roughness, and manufacturing tolerances, the transition area between the flange plane and the R-angle of the battery cell casing exhibits a non-ideal curved surface with unevenness at the micron level. Existing welding cut point identification methods rely on ideal geometric assumptions and cannot accurately locate the welding cut point, thus affecting the generation and fitting accuracy of the welding trajectory. This embodiment achieves accurate detection of the R-angle cut point (contact point of the casing and cover) of the casing by combining three-dimensional visual imaging with a contour curve cut point identification algorithm and a curve cut point detection algorithm based on R-angle accumulation, thereby obtaining higher welding trajectory consistency.

[0127] Step S400: Obtain the three-dimensional point cloud data of the profile curve of each section in the welding area of ​​the current incoming material.

[0128] The contour map corresponding to the current incoming material is obtained. Based on the coordinates of each edge of the current incoming material in the contour map, the scanning trajectory of the three-dimensional vision scanning device is obtained. In this embodiment, the three-dimensional vision scanning device (i.e., the first scanning device) is driven by the first motion control module to perform point cloud data scanning on the welding area of ​​the current incoming material, and the three-dimensional point cloud data corresponding to each cross-sectional contour curve in the welding area of ​​the current incoming material is obtained. When the incoming material is changed, the contour map corresponding to the steel shell cell to be welded is automatically loaded, the contour shape of the current incoming material is automatically identified, and the scanning trajectory corresponding to the three-dimensional vision scanning device is quickly generated.

[0129] For example, the first motion control module acquires the outline CAD drawing of the current incoming material. Based on the pre-calibrated CAD coordinate system and the motion control coordinate system of the 3D vision scanning device (such as a laser profilometer, structured light camera, or other 3D vision sensors), the first motion control module parses the outline coordinate information in the outline CAD drawing corresponding to the current incoming material and converts it into a motion control trajectory in the motion control coordinate system of the first scanning device, thereby automatically planning the scanning trajectory corresponding to the 3D vision scanning device. Based on the obtained scanning trajectory, the first motion control module is controlled to scan the outer outline of the current incoming material, thereby obtaining the 3D point cloud data corresponding to each cross-sectional outline curve in the welding area of ​​the current incoming material.

[0130] Step S410: Extract welding tangents based on the three-dimensional point cloud data corresponding to each cross-sectional profile curve, including: obtaining the R-angle accumulation line based on the three-dimensional point cloud data of the cross-sectional profile curve corresponding to the current incoming material; and obtaining the welding tangents based on the R-angle accumulation line.

[0131] This embodiment constructs an R-angle accumulation line based on the cross-sectional profile curve to extract the welding tangent point. First, the cross-sectional profile curve is smoothed. Then, an R-angle accumulation line, an R-angle accumulation curvature line, and a curvature wave are constructed. Finally, the intersection point is determined by generating piecewise approximation lines based on adjacent curvature waves, thereby determining the welding tangent point location. The welding tangent point identification process in this embodiment specifically includes the following steps:

[0132] First, an initial contour curve point sequence is obtained based on the 3D point cloud data of the cross-sectional contour curve; the sequence of all contour points in the cross-sectional contour curve is denoted as the initial contour curve point sequence; where the i-th contour point in this initial contour curve point sequence is denoted as Q. i , i=0,1,…,n, where n represents the total number of contour points in the initial contour curve point series.

[0133] Next, obtain the preset discrete point sequence; based on the discrete point sequence and the initial contour curve point sequence, obtain the smooth contour curve point sequence, where each contour point in the initial contour curve point sequence corresponds to a smooth point in the smooth contour curve point sequence.

[0134] Since the initial contour curve point sequence in this embodiment is an integer discrete representation of the original cross-sectional contour curve and is affected by imaging noise, the initial contour curve point sequence is not smooth. In order to eliminate the influence on subsequent FTP point identification, the initial contour curve point sequence needs to be smoothed first. However, since the real microscopic morphology noise and point cloud measurement noise are mixed at the micrometer scale, traditional filtering methods can only remove random abnormal noise points and cannot eliminate the contour unsmoothness caused by microscopic morphology.

[0135] This embodiment smooths the initial contour curve point series using a preset discrete point series. The initial values ​​of each point in the discrete point series can be set to the coordinates of the corresponding contour points in the initial contour point series. In this case, the number of points in the discrete point series corresponds to the number of contour points in the initial contour curve point series. Then, an objective function is obtained based on the discrete point series and the initial contour curve point series. This objective function includes the energy model of the discrete point series and the smoothing deviation degree. Specifically, the energy model of the discrete point series is obtained based on the chord length between adjacent contour points in the initial contour curve point series; the smoothing deviation degree corresponding to the current discrete point series is obtained based on the coordinates of each point in the discrete point series and the coordinates of each contour point in the initial contour curve point series. By minimizing the objective function, while ensuring that the final contour curve is as smooth as possible, it closely approximates the initial contour curve data, resulting in the final smoothing result (i.e., a smoothed contour curve point series).

[0136] Specifically, for any local interval formed by three adjacent points in a discrete point sequence: based on the known chord length between corresponding contour points in the local interval of the initial contour curve point sequence, the piecewise slope of each pair of adjacent points in the local interval is calculated. The difference between the two obtained piecewise slopes is used to assess the degree of abrupt change in the local slope of the initial contour curve point sequence. Scale normalization is then performed in conjunction with the total length of the local interval to eliminate scale interference caused by inconsistent interval lengths. This yields the local bending energy corresponding to the local interval, which characterizes the smoothness of the discrete point sequence within that local interval. The smoother the curve within the local interval, the smaller the corresponding local bending energy; conversely, the more severe the bends and abrupt changes in the curve within the local interval, the larger the corresponding local bending energy. The local bending energies corresponding to all local intervals constitute the energy model of the current discrete point sequence, used to constrain the entire contour curve to be as smooth as possible and reduce abrupt changes. The energy model of the current discrete point sequence can then be expressed as:

[0137]

[0138] Where F represents the energy model of the discrete point sequence, α represents the energy coefficient; n represents the total number of points in the discrete point sequence; i represents the i-th point in the discrete point sequence, i=0,1,…,n; P i P represents the coordinates of the i-th point in a discrete point sequence; i+1 P represents the coordinates of the (i+1)th point in a discrete point sequence; i－1 Represents the coordinates of the (i-1)th point in a discrete point sequence; l i This represents the chord length between the i-th point and the (i-1)-th point in the discrete point sequence. This chord length is equal to the chord length between the i-th contour point and the (i-1)-th contour point in the initial contour curve point sequence; l i+1 It represents the chord length between the (i+1)th point and the ith point in the discrete point sequence.

[0139] It should be noted that, due to the significant differences in the R-angle morphology of stamping from different dies, the location and degree of imaging noise vary. In this embodiment, the energy coefficient α is an adjustable hyperparameter. Its value can be adaptively adjusted according to the magnitude of curve noise and the sharpness of the contour curve. Parameter optimization relies on engineering experience and actual scenarios to increase compatibility and adaptability for field applications. For example, the selectable value range for the energy coefficient is 0.1 to 2.0. When the contour curve noise is low and contour details need to be preserved, the energy coefficient can be set to 0.1 to 0.5. Conversely, when the contour curve noise is high and strong smoothing and noise reduction are needed to suppress curvature fluctuations, the corresponding energy coefficient can be set to 0.5 to 2.0.

[0140] To prevent the smoothed contour curve from deviating excessively from the original data (i.e., the initial contour curve point series), this embodiment further incorporates the degree to which points in the discrete point series deviate from their corresponding contour points in the initial contour curve point series (i.e., the smoothing deviation degree) to obtain the objective function for smoothing the initial contour curve point series. This embodiment obtains the smoothing deviation degree corresponding to the current discrete point series based on the coordinates of each point in the discrete point series and the coordinates of each contour point in the initial contour curve point series.

[0141] Specifically, the deviation degree of each point is obtained based on the coordinates of each point in the discrete point column and the coordinates of the corresponding contour point in the initial contour curve point column; the reference weight of each point in the discrete point column is obtained; and the smoothing deviation degree of the current discrete point column is obtained based on the deviation degree of each point in the discrete point column and the reference weight.

[0142] It should be noted that the reference weight corresponding to each point in the discrete point column can be set according to the degree of outlier of that point in the initial contour curve point column. The greater the degree of outlier of a point, the more likely the contour point (original point) corresponding to that point in the initial contour curve point column is a noise point and the less reliable it is. A smaller reference weight needs to be set. Conversely, if the degree of outlier of a point is smaller, the more likely the original point corresponding to that point is a true contour point and the more reliable it is. A larger reference weight needs to be set. By relaxing the position constraints on unreliable contour points, the smoothing effect is strengthened simultaneously, thereby weakening the interference of abnormal measurement points on the fitted curve.

[0143] For example, for any contour point in the initial contour curve point sequence: obtain the local window generated by the contour point and its multiple neighboring contour points, and obtain the mean center point corresponding to the local window. Calculate the distance of the contour point from the mean center point within the corresponding local window. The smaller the distance value, the more reliable the corresponding point. When the obtained distance value is less than a preset distance threshold, reverse normalization is performed within a preset larger range (e.g., 1.0~2.0), thereby transforming the smaller distance value corresponding to the contour point into a larger outlier. Conversely, reverse normalization is performed within a preset smaller range (e.g., 0.1~0.5), transforming the larger distance value corresponding to the contour point into a smaller outlier. This process is repeated to obtain the outlier degree corresponding to each contour point in the initial contour curve point sequence. Finally, the obtained outlier degree is used as the reference weight for the corresponding point in the discrete point sequence.

[0144] The objective function obtained based on the energy model of the discrete point series and the degree of smooth deviation can be expressed as:

[0145]

[0146] Where M represents the objective function, Q iβ represents the coordinates of the i-th contour point in the initial contour curve point sequence. i This represents the reference weight corresponding to the i-th point in the discrete point column. F represents the degree of smooth deviation of the discrete point series; F represents the energy model of the discrete point series.

[0147] By constructing an energy model by replacing the chord lengths of corresponding points in the discrete point sequence with the chord lengths of adjacent contour points in the original initial contour curve point sequence, and then combining this with the degree of smooth deviation to construct the objective function, the problem of minimizing the objective function is transformed into a quadratic programming problem. To minimize the objective function, let:

[0148]

[0149] By solving the linear equation, a smoothed discrete point sequence can be obtained. This discrete point sequence is the smoothed contour curve point sequence, and each smoothed point in the obtained smoothed contour point sequence corresponds to a contour point in the initial contour point sequence.

[0150] Then, the R-angle line is obtained based on the smooth contour curve point sequence. At this time, each smooth point in the smooth contour curve point sequence corresponds to a first mapping point in the R-angle line.

[0151] This embodiment first obtains the curvature angle corresponding to each smooth point based on the coordinates of each smooth point in the smooth contour curve point array. Specifically, for any smooth point in the smooth contour curve point array, taking the i-th smooth point as an example, the process of obtaining its corresponding curvature angle is as follows:

[0152] Based on the smoothing point and the two smoothing points adjacent to it, the first unit vector and the second unit vector corresponding to the smoothing point are obtained. For example, for the i-th smoothing point, it and the (i-1)-th smoothing point form the first unit vector V of the smoothing point. i It, together with the (i+1)th smoothing point, forms the second unit vector V of that smoothing point. i+1 .

[0153] Obtain the initial unit vector V0. In this embodiment, the initial unit vector is a fixed unit reference vector, which can be uniformly set to (0, 0, 1) on the two-dimensional contour plane. Based on the initial unit vector, the first unit vector corresponding to the smoothing point, and the second unit vector, obtain the bending angle corresponding to the smoothing point. Calculate the bending direction coefficient by combining it with the initial unit vector. The obtained bending direction coefficient is used to determine the turning direction of the contour curve, such as clockwise or counterclockwise. The corresponding value can be set to 1 or -1, thereby adding a positive or negative sign to the bending angle corresponding to the smoothing point. Therefore, the bending angle corresponding to the i-th smoothing point can be expressed as:

[0154]

[0155] Where, θi V represents the curvature angle corresponding to the i-th smooth point; i V represents the first unit vector corresponding to the i-th smoothing point; i+1 h represents the second unit vector corresponding to the i-th smoothing point; i V0 represents the curvature direction coefficient corresponding to the i-th smoothing point; "·" represents the dot product; "×" represents the cross product; V0 represents the initial unit vector.

[0156] Then, based on the index value and curvature angle of each smooth point in the smooth contour curve point list, the coordinates of the first mapping point corresponding to that smooth point are obtained. The R-angle line is then obtained based on all the first mapping points. For example, a new coordinate system is constructed using the index value of the first mapping point as the x-coordinate and the curvature angle of the first mapping point as the y-coordinate. The coordinates of the i-th first mapping point can then be expressed as (i, θ). i The curve formed by all the first mapping points is denoted as the R-angle line.

[0157] It should be noted that since the 0th smoothing point does not have a first unit vector and the nth smoothing point does not have a second unit vector, the curvature angle between the 0th and nth smoothing points can be omitted here. That is, the range of the index number corresponding to the first mapping point is 0 < i < n.

[0158] Further, based on the coordinates of each first mapping point in the R-angle line, the R-angle accumulation line is obtained. At this time, each first mapping point in the R-angle line corresponds to a second mapping point in the R-angle accumulation line.

[0159] To eliminate noise, a Gaussian filter is used to perform one-dimensional Gaussian filtering on the R-angle line, and an R-angle accumulation line is constructed based on this. In this embodiment, the second mapping point corresponding to each first mapping point in the R-angle line is obtained according to the coordinates of the first mapping point, and the R-angle accumulation line is obtained according to all the second mapping points.

[0160] For any first mapping point: the accumulated angle corresponding to the first mapping point is obtained based on the bending angle corresponding to the first mapping point and the accumulated angle corresponding to the previous first mapping point; the second mapping point corresponding to the first mapping point is obtained based on the index value corresponding to the first mapping point and the accumulated angle; when the first mapping point is the first mapping point in the R-angle line, the accumulated angle corresponding to the previous first mapping point is a preset angle.

[0161] For example, for the i-th first mapping point, the corresponding bending angle θ of the first mapping point is... i The accumulated angle f corresponding to the previous first mapping point (i.e., the (i-1)th first mapping point) i－1 Add them together to obtain the accumulated angle f corresponding to the i-th first mapping point. i That is, f i =θi +f i－1 For the first mapping point in the R-angle line, the accumulated angle corresponding to the previous first mapping point is a preset angle θ0. For example, this preset angle can be set to 0, in which case f0 = θ0, and the accumulated angle corresponding to the first first mapping point is f1 = θ1 + f0. And so on, to obtain the accumulated angle corresponding to each first mapping point.

[0162] Then, based on the index value and curvature angle of each first mapping point in the R-angle line, the coordinates of the second mapping point corresponding to that first mapping point are obtained. The R-angle accumulation line is then obtained based on all the second mapping points. For example, the index value of the second mapping point is used as the x-coordinate, and the accumulated angle of the second mapping points is used as the y-coordinate. At this point, the coordinates S of the i-th second mapping point are... i It can be represented as (i, f) i The curve formed by all the second mapping points is denoted as the R-angle accumulation line.

[0163] Next, the R-angle cumulative curvature line is obtained based on the R-angle cumulative line, and each second mapping point in the R-angle cumulative line corresponds to a third mapping point in the R-angle cumulative curvature line.

[0164] To eliminate noise, this embodiment further applies a one-dimensional Gaussian filter to the obtained R-angle accumulation line, and determines the R-angle accumulation curvature line based on the filtered R-angle accumulation line. This embodiment obtains the curvature corresponding to each second mapping point based on its coordinates, and then obtains the third mapping point corresponding to that second mapping point based on its index value and curvature; the R-angle accumulation curvature line is obtained based on all third mapping points.

[0165] Theoretically, the straight sections (such as shell walls and flange planes) contained in the cross-sectional profile curve appear as horizontal straight lines in the R-angle accumulation line, while the arc segments (such as R-angle regions) appear as inclined straight lines. Let the straight lines in the R-angle accumulation line be called R-angle accumulation lines. Then, the tangent points between the straight sections and the arc segments, or between the arc segments themselves, are transformed into the intersection points of the two R-angle accumulation lines. However, due to errors, the ends of the R-angle accumulation lines may bend, forming curves, resulting in no obvious intersection point between adjacent R-angle accumulation lines. The intersection point should be located where the curvature on the R-angle accumulation line is greater (corresponding to one end of the horizontal / inclined straight line). Therefore, this embodiment further constructs an R-angle accumulation curvature line based on the curvature corresponding to each second mapping point in the R-angle accumulation line to determine the theoretical intersection point of the two straight lines in the R-angle accumulation line.

[0166] First, the bending angle corresponding to each second mapping point is obtained based on each second mapping point and its two adjacent second mapping points. Taking the i-th second mapping point as an example, the bending angle of the second mapping point obtained based on the i-th second mapping point and its two adjacent second mapping points (the (i-1)-th and (i+1)-th second mapping points) can be expressed as:

[0167]

[0168] in, S represents the bending angle corresponding to the i-th second mapping point; i S represents the coordinates corresponding to the i-th second mapping point; i－1 S represents the coordinates corresponding to the (i-1)th second mapping point; i+1 This represents the coordinates corresponding to the (i+1)th second mapping point.

[0169] Find the maximum value among all the bending angles corresponding to the second mapping points to obtain the maximum bending angle, and denote this maximum bending angle as φ. max ;

[0170] Based on the bending angle and maximum bending angle corresponding to each second mapping point, the curvature corresponding to that second mapping point is obtained; in this embodiment, the filtering coefficient λ is obtained, and for example, the value of the filtering coefficient can be set to 0.1; based on the filtering coefficient and the maximum bending angle, the bending angle corresponding to each second mapping point is filtered to obtain the curvature corresponding to that second mapping point;

[0171] The filtering threshold is obtained based on the maximum bending angle and the filtering coefficient. In this embodiment, the filtering threshold can be expressed as λφ. max When the bending angle corresponding to a second mapping point is less than or equal to the filtering threshold, it means that the second mapping point does not belong to the end position of the straight line in the R-angle accumulation line, and the curvature corresponding to the second mapping point is set to 0. When the bending angle corresponding to the second mapping point is greater than the filtering threshold, it means that the second mapping point may belong to the end position of the straight line in the R-angle accumulation line, so the bending angle corresponding to the second mapping point is retained, that is, the curvature corresponding to the second mapping point is set to the bending angle of the second mapping point. And so on, to obtain the curvature corresponding to each second mapping point.

[0172] Then, using the index value and curvature of each second mapping point, the coordinates of the corresponding third mapping point are obtained. Based on all third mapping points, the R-angle cumulative curvature line is obtained. For example, using the index value of the third mapping point as the x-coordinate and the curvature angle of the third mapping point as the y-coordinate, the coordinates of the i-th third mapping point can be represented as (i, c). i ), where c iLet represent the curvature corresponding to the i-th third mapping point, and the curve formed by all third mapping points is denoted as the R-angle cumulative curvature line.

[0173] Finally, multiple curvature waves are obtained by accumulating the curvature lines based on the R angle, and the welding tangent point corresponding to the profile curve of the cross section is obtained based on the triplet corresponding to each curvature wave.

[0174] By processing each second mapping point in the R-angle cumulative curve, areas with greater curvature are selected to locate the welding tangent. The resulting R-angle cumulative curvature line can be represented as several curvature waves, each with a tangent point. Each curvature wave corresponds to a triplet, which includes the index values ​​of the third mapping point corresponding to the start, end, and peak positions of the curvature wave on the R-angle cumulative curvature line. Therefore, for the r-th curvature wave, its corresponding triplet can be represented as (Q... r Z r Loc r ).

[0175] Since the FTP point is theoretically the intersection of two adjacent R-angle accumulation lines, a piecewise linear approximation can be made on the R-angle accumulation line using a curvature wave. Then, the intersection of two adjacent approximation line segments is found. Taking the r-th curvature wave as an example, in order to determine the FTP point in this curvature wave, this embodiment uses the least squares method to construct corresponding approximation lines by combining the two curvature waves adjacent to this curvature wave. The approximation line constructed by this curvature wave and the (r-1)-th curvature wave can be expressed as y=k r－ 1x+b r－1 The approximation line constructed by this curvature wave and the (r+1)th curvature wave can be expressed as y=k r＋1 x+b r＋1 Then, find the intersection point of these two approximating lines. The x-coordinate of the intersection point is the position number of the FTP point in the cross-sectional profile curve.

[0176] Based on the curvature wave and the triplet corresponding to its adjacent (r+1)th curvature wave, the first approximation error metric corresponding to the curvature wave is obtained. Taking the first approximation error metric as an example, by obtaining the weight corresponding to each third mapping point in the curvature wave in the first approximation error metric, and combining it with the accumulated angle corresponding to each third mapping point in the curvature wave, the constructed first approximation error metric can be expressed as:

[0177]

[0178]

[0179] Among them, E r Loc represents the first approximation error metric for the current r-th curvature wave;r This represents the index value corresponding to the peak position of the current r-th curvature wave; Loc r+1 The index value represents the position of the (r+1)th curvature wave crest; j represents the j-th third mapping point between the current (r)th curvature wave crest position and the corresponding crest position of the (r+1)th curvature wave; u rj f represents the index value corresponding to the j-th third mapping point in the current r-th curvature wave; rj Let represent the accumulated angle of the second mapping point corresponding to the j-th third mapping point in the current r-th curvature wave; k represents the slope of the first approximation line corresponding to this curvature wave. After minimizing the first approximation error metric, this value is the true slope k corresponding to the first approximation line. r＋1 b represents the intercept of the first approximation line corresponding to the curvature wave. After the first approximation error metric is minimized, this value is the true intercept b corresponding to the first approximation line. r＋1 ;w rj This represents the reference weight of the j-th third mapping point corresponding to the r-th curvature wave; t rj Q represents the relative offset corresponding to the j-th third mapping point in the current r-th curvature wave; r+1 Z represents the index value of the third mapping point corresponding to the starting position of the (r+1)th curvature wave; r This represents the index value of the third mapping point corresponding to the termination position of the current r-th curvature wave.

[0180] Based on the curvature wave and the triplet corresponding to its adjacent (r-1)th curvature wave, the second approximation error metric corresponding to the curvature wave is obtained; based on the curvature wave and the adjacent (r-1)th curvature wave, the weights of each third mapping point in the rth curvature wave in the second approximation error metric are determined; and then, combined with the accumulated angle corresponding to each third mapping point in the curvature wave, the second approximation error metric is constructed.

[0181] To obtain the optimal approximation line, all approximation error measures should be minimized. In this embodiment, the first approximation error measure is minimized by making the partial derivatives of the first approximation error measure with respect to the slope and intercept both zero, and the second approximation error measure is minimized by making the partial derivatives of the second approximation error measure with respect to the slope and intercept both zero, thereby obtaining the first approximation line and the second approximation line. Similarly, each curvature wave is processed to obtain the first approximation line and the second approximation line corresponding to each curvature wave.

[0182] Finally, the welding tangent point corresponding to the cross-sectional profile curve is obtained based on all curvature waves and their corresponding first and second approximation lines. Since the straight lines included in the cross-sectional profile curve (such as the shell wall and flange plane) appear as horizontal straight lines in the R-angle accumulation line, the two approximation lines with a slope of 0 among all curvature waves are the straight lines corresponding to the shell wall and the flange plane. The two curvature waves corresponding to these two approximation lines are the transition positions between the shell wall and the R-angle and the flange plane and the R-angle region, respectively. Thus, the R-angle transition interval (i.e., the detection area where the welding tangent point is located) is determined.

[0183] The effective main wave is then determined based on the peak heights of all curvature waves in the obtained R-angle transition interval. In this embodiment, the effective main wave is determined based on the peak heights of all curvature waves and a preset peak threshold. The curvature wave with the first peak ordinate greater than the preset peak threshold is taken as the effective main wave. The empirical value of the peak threshold is 0.5. Then, the welding tangent point corresponding to the effective main wave is obtained based on the first approximation line and the second approximation line corresponding to the effective main wave. That is, the intersection point between the first approximation line and the second approximation line of the effective main wave is obtained. The obtained intersection point is the welding tangent point corresponding to the current cross-sectional profile curve.

[0184] Step S420: Perform adaptive fitting of the welding trajectory based on the welding tangent point to obtain the welding trajectory corresponding to the current incoming material.

[0185] Since the welding cut points are analyzed based on the point cloud data collected by the three-dimensional vision scanning device, after obtaining all the welding cut points corresponding to the current incoming material, it is necessary to convert them to the welding coordinate system to obtain the welding trajectory of the welding head, so as to control the welding head to weld the current incoming material based on the obtained welding trajectory.

[0186] This embodiment first obtains the coordinate transformation relationship between the pre-calibrated laser coordinate system of the three-dimensional vision scanning device and the welding coordinate system corresponding to the welding head used for welding the current incoming material; then, it performs coordinate transformation on the welding tangent point according to the coordinate transformation relationship to obtain the welding trajectory corresponding to the current incoming material; wherein, the coordinate transformation relationship between the laser coordinate system and the welding coordinate system is obtained through the first coordinate transformation relationship and the second coordinate transformation relationship; the first coordinate transformation relationship is the coordinate transformation relationship between the laser coordinate system and the fixture coordinate system corresponding to the steel shell battery cell clamping fixture, and the second coordinate transformation relationship is the coordinate transformation relationship between the fixture coordinate system and the welding coordinate system.

[0187] This embodiment combines three-dimensional visual imaging with a contour curve tangent point recognition algorithm, which can adapt to the complex three-dimensional curvature characteristics of the contact area of ​​irregularly shaped steel shell battery cell housings and covers. It solves the shortcomings of existing two-dimensional vision-guided welding methods that cannot capture three-dimensional spatial information and cannot accurately identify the spatial position of feature points. This embodiment is based on a curve tangent point detection algorithm of R-angle accumulation. It constructs an R-angle accumulation line based on the cross-sectional contour curve to extract the welding tangent point. By first smoothing the cross-sectional contour curve, then constructing the R-angle accumulation line, R-angle accumulation curvature line, and curvature wave, and finally determining the welding tangent point position by generating segmented approximation straight lines based on adjacent curvature waves, it overcomes the problem of local contour abrupt changes caused by the micron-level concave and convex surfaces of the flange. While preserving macroscopic geometric features, it effectively filters out microscopic noise, realizes accurate detection of the R-angle tangent point of the housing cover (the contact point of the housing cover), avoids welding trajectory deviation, and thus obtains higher welding trajectory consistency.

[0188] Please refer to Figure 9 Some embodiments provide a welding cut point identification method, including:

[0189] Step S500: Obtain the three-dimensional point cloud data of the profile curve of each section in the welding area of ​​the current incoming material.

[0190] Step S510: Extract the welding cutting point based on the three-dimensional point cloud data of the profile curve of the current incoming material.

[0191] This embodiment uses a curve tangent detection algorithm based on R-angle accumulation to detect welding tangents. The specific method and steps are as follows: Figure 10 The above includes:

[0192] Step S511: Obtain the R-angle line based on the three-dimensional point cloud data of the profile curve of the current incoming material.

[0193] In this embodiment, each contour point in the cross-sectional contour curve corresponds to a first mapping point in the R-angle line, and each first mapping point corresponds to an index value; this embodiment obtains the initial contour curve point list based on the three-dimensional point cloud data of the cross-sectional contour curve;

[0194] Obtain a preset discrete point list, the number of points in the discrete point list corresponds to the number of contour points in the initial contour curve point list; obtain a smooth contour curve point list based on the discrete point list and the initial contour curve point list, each contour point in the initial contour curve point list corresponds to a smooth point in the smooth contour curve point list.

[0195] Then, the R-angle line is obtained based on the smooth contour curve point list, including: obtaining the curvature angle corresponding to each point based on the coordinates of each smooth point in the smooth contour curve point list; obtaining the coordinates of the first mapping point corresponding to each smooth point based on the index value and curvature angle of each smooth point in the smooth contour curve point list; and obtaining the R-angle line based on all the first mapping points.

[0196] Step S512: Obtain the R-angle accumulation line based on the coordinates of each first mapping point in the R-angle line. Each first mapping point in the R-angle line corresponds to a second mapping point in the R-angle accumulation line.

[0197] For any first mapping point in the R-angle accumulation line: the accumulated angle corresponding to the first mapping point is obtained according to the bending angle corresponding to the first mapping point and the accumulated angle corresponding to the previous first mapping point; the second mapping point corresponding to the first mapping point is obtained according to the index value corresponding to the first mapping point and the accumulated angle; when the first mapping point is the first mapping point in the R-angle accumulation line, the accumulated angle corresponding to the first mapping point is a preset angle.

[0198] Step S513: Obtain the R-angle cumulative curvature line based on the R-angle cumulative line. Each second mapping point in the R-angle cumulative line corresponds to a third mapping point in the R-angle cumulative curvature line.

[0199] In this embodiment, the curvature corresponding to each second mapping point is obtained based on the coordinates of each second mapping point in the R-angle cumulative curve; the third mapping point corresponding to the second mapping point is obtained based on the index value and curvature of the second mapping point; and the R-angle cumulative curvature curve is obtained based on all the third mapping points.

[0200] Step S514: Obtain multiple curvature waves based on the R-angle cumulative curvature line. Each curvature wave corresponds to a triplet. The triplet includes the index value of the third mapping point corresponding to the start position, the index value of the third mapping point corresponding to the end position, and the index value of the third mapping point corresponding to the peak position on the R-angle cumulative curvature line.

[0201] Step S515: Obtain the welding tangent point of the cross-sectional profile curve corresponding to the current incoming material based on the triplet corresponding to each curvature wave.

[0202] For any curvature wave, taking the r-th curvature wave as an example: based on the curvature wave and the triplet corresponding to its adjacent (r+1)-th curvature wave, the first approximation error metric corresponding to the curvature wave is obtained; based on the curvature wave and the triplet corresponding to its adjacent (r-1)-th curvature wave, the second approximation error metric corresponding to the curvature wave is obtained; by minimizing the first and second approximation error metrics respectively, the first and second approximation lines are obtained; and so on, the first and second approximation lines corresponding to each curvature wave are obtained; based on all curvature waves and their corresponding first and second approximation lines, the welding tangent points corresponding to the cross-sectional profile curve are obtained, and then all welding tangent points corresponding to the current incoming material are obtained.

[0203] It should be noted that each step of the welding cut point identification method in this embodiment corresponds to step S410 in the above-mentioned automatic fitting method for welding trajectory of irregular steel shell battery cells. The specific implementation method has been described in the above embodiment and will not be repeated here.

[0204] Some embodiments of the present invention also disclose a storage medium including a program that can be executed by a processor to implement the methods described in any of the embodiments herein.

[0205] In the above embodiments, implementation can be achieved, in whole or in part, by software, hardware, firmware, or any combination thereof. Furthermore, as those skilled in the art will understand, the principles herein can be reflected in a computer program product on a computer-readable storage medium pre-loaded with computer-readable program code. Any tangible, non-transitory computer-readable storage medium may be used, including magnetic storage devices (hard disks, floppy disks, etc.), optical storage devices (CDs, DVDs, Blu-ray discs, etc.), flash memory, and / or the like. These computer program instructions can be loaded onto a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to form a machine, such that instructions executing on the computer or other programmable data processing apparatus can generate means for performing a specified function. These computer program instructions can also be stored in a computer-readable storage medium that can instruct the computer or other programmable data processing apparatus to operate in a particular manner, such that instructions stored in the computer-readable storage medium can form an article of manufacture, including means for implementing the specified function. The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to perform a series of operational steps on the computer or other programmable apparatus to produce a computer-implemented process, such that instructions executing on the computer or other programmable apparatus can provide steps for implementing the specified function.

[0206] This document describes various exemplary embodiments with reference to them. However, those skilled in the art will recognize that changes and modifications can be made to the exemplary embodiments without departing from the scope of this document. For example, various operational steps and components for performing operational steps can be implemented in different ways depending on the specific application or considering any number of cost functions associated with the operation of the system (e.g., one or more steps can be deleted, modified, or combined with other steps).

[0207] While the principles herein have been illustrated in various embodiments, numerous modifications to the structures, arrangements, proportions, elements, materials, and components, particularly suited to specific environments and operational requirements, may be used without departing from the principles and scope of this disclosure. These modifications and other alterations or alterations will be included within the scope of this document. Those skilled in the art will recognize that many changes can be made to the details of the above embodiments without departing from the fundamental principles of the invention.

Claims

1. An automatic fitting method for the welding trajectory of an irregularly shaped steel-shell battery cell, characterized in that, include: Obtain the 3D point cloud data of the profile curve of each section in the welding area of ​​the current incoming material; Extracting welding tangent points based on the three-dimensional point cloud data of each cross-sectional profile curve includes: obtaining the R-angle accumulation line based on the three-dimensional point cloud data of the cross-sectional profile curve corresponding to the current incoming material; and obtaining the welding tangent point based on the R-angle accumulation line. The welding trajectory is adaptively fitted based on the welding tangent point corresponding to each cross-sectional profile curve to obtain the welding trajectory corresponding to the current incoming material.

2. The method as described in claim 1, characterized in that, The acquisition of three-dimensional point cloud data of the profile curve of each cross section in the welding area of ​​the current incoming material includes: Get the outline diagram corresponding to the current incoming material; The scanning trajectory of the three-dimensional vision scanning device is obtained based on the coordinates of each edge of the current incoming material in the contour diagram. After the three-dimensional vision scanning device scans the welding area of ​​the current incoming material according to the scanning trajectory, it obtains the three-dimensional point cloud data corresponding to each cross-sectional contour curve in the welding area of ​​the current incoming material.

3. The method as described in claim 1, characterized in that, The step of obtaining the R-angle accumulation line based on the three-dimensional point cloud data of the cross-sectional contour curve corresponding to the current incoming material includes: The initial contour curve point list is obtained based on the three-dimensional point cloud data of the cross-sectional contour curve; Obtain a preset discrete point list, wherein the number of points in the discrete point list corresponds to the number of contour points in the initial contour curve point list; A smooth contour curve point list is obtained based on the discrete point list and the initial contour curve point list, wherein each contour point in the initial contour curve point list corresponds to a smooth point in the smooth contour curve point list. The R-angle line is obtained based on the smooth contour curve point list, and each smooth point in the smooth contour curve point list corresponds to a first mapping point in the R-angle line; The R-angle accumulation line is obtained based on the coordinates of each first mapping point in the R-angle line, and each first mapping point in the R-angle line corresponds to a second mapping point in the R-angle accumulation line.

4. The method as described in claim 3, characterized in that, The step of obtaining a smooth contour curve point list based on the discrete point list and the initial contour curve point list includes: An objective function is obtained based on the discrete point sequence and the initial contour curve point sequence. The objective function includes the energy model of the discrete point sequence and the smoothing deviation degree. The energy model of the discrete point sequence is obtained based on the chord length between adjacent contour points in the initial contour curve point sequence. The smoothing deviation degree corresponding to the discrete point sequence is obtained based on the coordinates of each point in the discrete point sequence and the coordinates of each contour point in the initial contour curve point sequence. By minimizing the objective function, the discrete point sequence is updated to obtain a smooth contour curve point sequence.

5. The method as described in claim 4, characterized in that, The step of obtaining the energy model of the discrete point sequence based on the chord length between adjacent contour points in the initial contour curve point sequence includes: Where F represents the energy model of the discrete point sequence, α represents the energy coefficient; n represents the total number of points contained in the discrete point sequence; i represents the i-th point in the discrete point sequence, i=0,1,…,n; P i P represents the coordinates of the i-th point in the discrete point sequence; i+1 P represents the coordinates of the (i+1)th point in the discrete point sequence; i－1 Represents the coordinates of the (i-1)th point in the discrete point sequence; l i This represents the chord length corresponding to the i-th point and the (i-1)-th point in the discrete point sequence, and this chord length is equal to the chord length corresponding to the i-th point and the (i-1)-th point in the initial contour curve point sequence; l i+1 This represents the chord length between the (i+1)th point and the ith point in the discrete point sequence.

6. The method as described in claim 4, characterized in that, The step of obtaining the smoothing deviation degree corresponding to the discrete point series based on the coordinates of each point in the discrete point series and the coordinates of each contour point in the initial contour curve point series includes: The degree of deviation of a point is obtained by using the coordinates of each point in the discrete point column and the coordinates of the corresponding contour point in the initial contour curve point column. Obtain the reference weight corresponding to each point in the discrete point column; The smoothness deviation of the discrete point column is obtained based on the deviation degree of each point in the discrete point column and the reference weight.

7. The method as described in claim 3, characterized in that, The step of obtaining the R-angle line based on the smooth contour curve point list includes: The curvature angle corresponding to each point is obtained based on the coordinates of each smooth point in the smooth contour curve point list; Based on the index value and curvature angle of each smooth point in the smooth contour curve point column, the coordinates of the first mapping point corresponding to the smooth point are obtained; The R-angle line is obtained based on all the first mapping points.

8. The method as described in claim 7, characterized in that, The step of obtaining the curvature angle corresponding to each point based on the coordinates of each smooth point in the smooth contour curve point list includes: For any smooth point in the smooth contour curve point sequence: obtain the first unit vector and the second unit vector corresponding to the smooth point based on the smooth point and the two smooth points adjacent to the smooth point; Obtain the initial unit vector; based on the initial unit vector, the first unit vector corresponding to the smooth point, and the second unit vector, obtain the curvature angle corresponding to the smooth point.

9. The method as described in claim 8, characterized in that, The step of obtaining the curvature angle corresponding to the smooth point based on the initial unit vector, the first unit vector corresponding to the smooth point, and the second unit vector includes: Where, θ i V represents the curvature angle corresponding to the i-th smooth point; i V represents the first unit vector corresponding to the i-th smoothing point; i+1 h represents the second unit vector corresponding to the i-th smoothing point; i The curve direction coefficient corresponding to the i-th smoothing point is represented by "·"; "×" represents the dot product; and V0 represents the initial unit vector.

10. The method as described in claim 9, characterized in that, The step of obtaining the R-angle accumulation line based on the coordinates of each first mapping point in the R-angle line includes: The second mapping point corresponding to each first mapping point in the R-angle line is obtained based on the coordinates of the first mapping point, and the R-angle accumulation line is obtained based on all the second mapping points. For any first mapping point: the accumulated angle corresponding to the first mapping point is obtained based on the bending angle corresponding to the first mapping point and the accumulated angle corresponding to the previous first mapping point; the second mapping point corresponding to the first mapping point is obtained based on the index value corresponding to the first mapping point and the accumulated angle; when the first mapping point is the first mapping point in the R-angle line, the accumulated angle corresponding to the previous first mapping point is a preset angle.

11. The method as described in claim 10, characterized in that, The step of obtaining the welding tangent point based on the accumulated R-angle line includes: The R-angle accumulation curve is obtained based on the R-angle accumulation curve, and each second mapping point in the R-angle accumulation curve corresponds to a third mapping point in the R-angle accumulation curve; Multiple curvature waves are obtained based on the R-angle cumulative curvature line. Each curvature wave corresponds to a triplet. The triplet is included on the R-angle cumulative curvature line, where the starting position of the curvature wave corresponds to the index value of the third mapping point, the ending position corresponds to the index value of the third mapping point, and the peak position corresponds to the index value of the third mapping point. The welding tangent point is obtained based on the triplet corresponding to each curvature wave.

12. The method as described in claim 11, characterized in that, The step of obtaining the R-angle cumulative curvature line based on the R-angle cumulative line includes: The curvature corresponding to each second mapping point is obtained based on the coordinates of each second mapping point in the R-angle accumulation line; Based on the index value and curvature corresponding to the second mapping point, the third mapping point corresponding to the second mapping point is obtained; The cumulative curvature line of the R-angle is obtained from all the third mapping points.

13. The method as described in claim 12, characterized in that, The step of obtaining the curvature corresponding to each second mapping point based on the coordinates of each second mapping point in the R-angle accumulation line includes: The bending angle corresponding to each second mapping point is obtained based on each second mapping point and its two adjacent second mapping points. Obtain the maximum value among all the bending angles corresponding to the second mapping points to get the maximum bending angle; The curvature corresponding to each second mapping point is obtained based on the bending angle corresponding to each second mapping point and the maximum bending angle.

14. The method as described in claim 13, characterized in that, The step of obtaining the bending angle corresponding to each second mapping point based on each second mapping point and its two adjacent second mapping points includes: in, S represents the bending angle corresponding to the i-th second mapping point; i S represents the coordinates corresponding to the i-th second mapping point; i－1 S represents the coordinates corresponding to the (i-1)th second mapping point; i+1 This represents the coordinates corresponding to the (i+1)th second mapping point.

15. The method as described in claim 13, characterized in that, The step of obtaining the curvature corresponding to the second mapping point based on the bending angle corresponding to each second mapping point and the maximum bending angle includes: Obtain the filtering coefficient; based on the filtering coefficient and the maximum bending angle, filter the bending angle corresponding to each second mapping point to obtain the curvature corresponding to the second mapping point; The filtering threshold is obtained based on the maximum bending angle and the filtering coefficient. When the bending angle corresponding to a second mapping point is less than or equal to the filtering threshold, the curvature of the second mapping point is 0. When the bending angle corresponding to the second mapping point is greater than the filtering threshold, the curvature of the second mapping point is the bending angle of the second mapping point.

16. The method as described in claim 11, characterized in that, The process of obtaining the welding tangent point based on the triplet corresponding to each curvature wave includes: For the r-th curvature wave: based on the curvature wave and the triplet corresponding to its adjacent (r+1)-th curvature wave, the first approximation error metric is obtained; based on the curvature wave and the triplet corresponding to its adjacent (r-1)-th curvature wave, the second approximation error metric is obtained; by minimizing the first approximation error metric and the second approximation error metric respectively, the first approximation line and the second approximation line are obtained; The welding tangent point corresponding to the profile curve of the cross section is obtained by using all curvature waves and their corresponding first and second approximation lines.

17. The method as described in claim 16, characterized in that, The step of obtaining the first approximation error metric corresponding to the curvature wave based on the curvature wave and the triplet corresponding to its adjacent (r+1)th curvature wave includes: Among them, E r Loc represents the first approximation error metric for the current r-th curvature wave; r This represents the index value corresponding to the peak position of the current r-th curvature wave; Loc r+1 This represents the index value corresponding to the crest position of the (r+1)th curvature wave; j represents the j-th third mapping point between the crest position of the current r-th curvature wave and the corresponding crest position of the (r+1)-th curvature wave; u rj f represents the index value corresponding to the j-th third mapping point in the current r-th curvature wave; rj represents the accumulated angle of the second mapping point corresponding to the j-th third mapping point in the current r-th curvature wave; k represents the slope of the first approximation line corresponding to this curvature wave; b represents the intercept of the first approximation line corresponding to this curvature wave; w rj This represents the reference weight of the j-th third mapping point corresponding to the r-th curvature wave; t rj Q represents the relative offset corresponding to the j-th third mapping point in the current r-th curvature wave; r+1 Z represents the index value of the third mapping point corresponding to the starting position of the (r+1)th curvature wave; r This represents the index value of the third mapping point corresponding to the termination position of the current r-th curvature wave.

18. The method as described in claim 2, characterized in that, The step of adaptively fitting the welding trajectory based on the welding tangent point corresponding to each cross-sectional contour curve to obtain the welding trajectory corresponding to the current incoming material includes: The coordinate transformation relationship between the laser coordinate system of the 3D vision scanning device and the welding coordinate system corresponding to the welding head used to weld the current incoming material is obtained; the coordinate transformation of the welding tangent point is performed according to the coordinate transformation relationship to obtain the welding trajectory corresponding to the current incoming material; The coordinate transformation relationship between the laser coordinate system and the welding coordinate system is obtained through a first coordinate transformation relationship and a second coordinate transformation relationship; the first coordinate transformation relationship is the coordinate transformation relationship between the laser coordinate system and the fixture coordinate system corresponding to the steel shell battery cell clamping fixture, and the second coordinate transformation relationship is the coordinate transformation relationship between the fixture coordinate system and the welding coordinate system.

19. A method for identifying welding cut points, characterized in that, include: Obtain the 3D point cloud data of the cross-sectional profile curve in the welding area of ​​the current incoming material; Welding tangent points are extracted based on the 3D point cloud data of the cross-sectional profile curve corresponding to the current incoming material, including: The R-angle line is obtained based on the three-dimensional point cloud data of the cross-sectional profile curve corresponding to the current incoming material. Each profile point in the cross-sectional profile curve corresponds to a first mapping point in the R-angle line, and each first mapping point corresponds to an index value. The R-angle accumulation line is obtained based on the coordinates of each first mapping point in the R-angle line, and each first mapping point in the R-angle line corresponds to a second mapping point in the R-angle accumulation line. The R-angle accumulation curve is obtained based on the R-angle accumulation curve, and each second mapping point in the R-angle accumulation curve corresponds to a third mapping point in the R-angle accumulation curve; Multiple curvature waves are obtained based on the R-angle cumulative curvature line. Each curvature wave corresponds to a triplet. The triplet is included on the R-angle cumulative curvature line, where the starting position of the curvature wave corresponds to the index value of the third mapping point, the ending position corresponds to the index value of the third mapping point, and the peak position corresponds to the index value of the third mapping point. The welding tangent point of the current incoming material's cross-sectional profile curve is obtained based on the triplet corresponding to each curvature wave.

20. The method as described in claim 19, characterized in that, The step of obtaining the R-angle line based on the three-dimensional point cloud data of the cross-sectional contour curve corresponding to the current incoming material includes: The initial contour curve point sequence is obtained based on the three-dimensional point cloud data of the cross-sectional contour curve; Obtain a preset discrete point list, wherein the number of points in the discrete point list corresponds to the number of contour points in the initial contour curve point list; A smooth contour curve point list is obtained based on the discrete point list and the initial contour curve point list, wherein each contour point in the initial contour curve point list corresponds to a smooth point in the smooth contour curve point list. Obtaining the R-angle line based on the smooth contour curve point list includes: obtaining the curvature angle corresponding to each point based on the coordinates of each smooth point in the smooth contour curve point list; obtaining the coordinates of the first mapping point corresponding to each smooth point based on the index value and curvature angle of each smooth point in the smooth contour curve point list; and obtaining the R-angle line based on all the first mapping points.

21. The method as described in claim 20, characterized in that, The step of obtaining the R-angle accumulation line based on the coordinates of each first mapping point in the R-angle line includes: For any first mapping point in the R-angle accumulation line: the accumulated angle corresponding to the first mapping point is obtained according to the bending angle corresponding to the first mapping point and the accumulated angle corresponding to the previous first mapping point; the second mapping point corresponding to the first mapping point is obtained according to the index value corresponding to the first mapping point and the accumulated angle; when the first mapping point is the first mapping point in the R-angle accumulation line, the accumulated angle corresponding to the first mapping point is a preset angle.

22. The method as described in claim 19, characterized in that, The step of obtaining the R-angle cumulative curvature line based on the R-angle cumulative line includes: The curvature corresponding to each second mapping point is obtained based on the coordinates of each second mapping point in the R-angle accumulation line; Based on the index value and curvature corresponding to the second mapping point, the third mapping point corresponding to the second mapping point is obtained; The cumulative curvature line of the R-angle is obtained from all the third mapping points.

23. The method as described in claim 19, characterized in that, The step of obtaining the welding tangent point of the current incoming material's cross-sectional profile curve based on the triplet corresponding to each curvature wave includes: For the r-th curvature wave: based on the curvature wave and the triplet corresponding to its adjacent (r+1)-th curvature wave, the first approximation error metric is obtained; based on the curvature wave and the triplet corresponding to its adjacent (r-1)-th curvature wave, the second approximation error metric is obtained; by minimizing the first approximation error metric and the second approximation error metric respectively, the first approximation line and the second approximation line are obtained; The welding tangent point corresponding to the profile curve of the cross section is obtained by using all curvature waves and their corresponding first and second approximation lines.

24. A storage medium, characterized in that, Includes a program that can be executed by a processor to implement the method as described in any one of claims 1 to 23.

Images