Demonstration-free welding method and system for thick-wall pipe groove

Abstract

The invention relates to the technical field of industrial welding robots and automatic welding control, in particular to a thick-wall pipe groove teaching-free welding method and system. The method comprises the steps that a global vision unit and an assembled thick-wall pipe groove area are obtained, and the actual pose of a workpiece and the geometric center line vector of a welding seam are obtained through subarea scanning, point cloud registration and model registration; extracting geometric characteristic parameters of the cross section of the welding seam, matching the groove form code and binding preset welding layer channel information to form a welding seam attribute set; then path planning, multi-layer multi-channel decomposition and welding program generation are executed, and a welding program and a theoretical track are obtained; and finally, correcting subsequent interpolation points on line, scanning the welding seam again under the triggering condition, and updating the welding seam attribute set. According to the method, the welding seam attribute set capable of driving multi-layer and multi-pass welding is constructed, so that the variable cross-section welding problem caused by non-uniform assembly of the thick-wall pipe is effectively solved, and the welding self-adaptive capacity and the welding quality are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of industrial welding robots and automatic welding control technology, and in particular to a method and system for teaching-free welding of thick-walled pipe beveling. Background Technology

[0002] In the field of industrial welding robots and automatic welding control technology, existing solutions for the bevel region of thick-walled pipes typically employ a technical approach involving manual teaching programming, model library retrieval, weld area point cloud acquisition, bevel centerline fitting, automatic welding trajectory generation, and weld error data tracking. This approach suffers from limitations such as unstable correspondence between the weld area point cloud and the actual workpiece pose, disconnection between the geometric feature parameters of the weld cross-section and the preset welding layer information, and difficulty in writing weld error data back to subsequent trajectory processing links. Existing methods often rely on 3D models of the workpiece, program templates, or single-scan results to generate welding programs and theoretical trajectories. In the assembled thick-walled pipe bevel region, inconsistencies easily arise between the root gap width, bevel angle, blunt edge thickness, and base material thickness and the actual working conditions. Furthermore, mismatches exist between multi-layer, multi-pass decomposition and process parameter index code retrieval, making it difficult to achieve continuous multi-layer, multi-pass welding.

[0003] For the joint processing of weld cross-sectional geometric feature parameters, groove form code, process parameter index code, preset welding layer information, and weld attribute set updates, existing technologies generally lack a unified link between point cloud registration, feature extraction algorithm processing, attribute binding, path planning algorithm processing, online correction of subsequent interpolation points, and re-scanning of the weld. This makes it difficult to form a consistent process in the weld area groove region for obtaining weld area point cloud, determining the actual workpiece pose, constructing weld attribute set, generating automatic welding trajectory, generating welding program and theoretical trajectory, collecting weld error data, and updating weld attribute set. As a result, the subsequent interpolation point adjustment and weld attribute set update are disconnected in the multi-layer, multi-pass welding process, making it difficult to continuously connect welding operations, process switching, and subsequent updates. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for teachless welding of thick-walled pipe beveling, comprising:

[0005] S100: Acquire the global vision unit, industrial camera, laser line structured light, and assembled thick-walled pipe bevel area; perform partitioned scanning, point cloud registration, and model-point cloud registration processing to obtain the actual workpiece pose and weld geometric centerline vector; the global vision unit includes an industrial camera, laser line structured light, mounting bracket, control interface, and acquisition position separately set from the automatic welding module; the industrial camera is used to receive reflected light information from the surface of the thick-walled pipe bevel area; the laser line structured light is used to form a continuous light band contour on the surface of the thick-walled pipe bevel area; the assembled thick-walled pipe bevel area includes the bevel opening to be welded, the base material surfaces on both sides of the bevel, and the positioning reference area adjacent to the bevel;

[0006] S200. Based on the actual pose of the workpiece and the geometric centerline vector of the weld, perform geometric feature parameter extraction of the weld cross section, groove form code matching and preset welding layer information binding processing to obtain a set of weld attributes.

[0007] S300. Based on the weld attribute set, perform path planning algorithm, multi-layer multi-pass decomposition and welding program generation processing to obtain welding program and theoretical trajectory;

[0008] S400. Based on the welding program and theoretical trajectory, perform welding operations using the automatic welding module, correct subsequent interpolation points online, and rescan the weld seam to obtain an updated set of weld seam attributes.

[0009] Furthermore, the partition scan process includes:

[0010] The bevel region of the thick-walled pipe is divided into several adjacent scanning sections according to the length of the bevel region and the field of view. Three-dimensional contour and RGB images are acquired for each scanning section, and then processed by a stitching algorithm to form a continuous point cloud of the weld region.

[0011] Furthermore, the point cloud registration process and the model-point cloud registration process include:

[0012] Downsampling is performed on the point cloud of the weld area to unify the spacing, filtering is performed to remove discrete noise points, and edge extraction based on RGB image is performed to determine the bevel boundary;

[0013] The transformation matrix is ​​obtained and the model is registered with the point cloud to obtain the actual pose of the workpiece and the geometric centerline vector of the weld continuously extracted along the bevel opening area.

[0014] Furthermore, the process of extracting and processing the geometric feature parameters of the weld cross-section includes:

[0015] Multiple local sections are cut along the geometric centerline vector of the weld at preset intervals;

[0016] Extract the root gap width, bevel angle, blunt edge thickness, and base material thickness for each section;

[0017] The segments are then merged according to the similarity of their parameters to obtain one or more parameter segments.

[0018] Furthermore, the process of matching bevel form codes and binding preset welding layer information includes:

[0019] The bevel form code matching process includes: matching the four types of parameters of each parameter segment with the process database to generate the bevel form code and process parameter index code for each segment;

[0020] The preset welding layer information binding process includes: binding the groove form code, process parameter index code, preset welding layer information and weld geometric centerline vector of the same parameter segment at the segment level to form a layer record chain as a weld attribute set; wherein, when a certain segment does not match the preset welding layer information, the information of the adjacent segment is copied to form a temporary binding or marked as a segment to be updated.

[0021] Furthermore, the path planning algorithm process includes:

[0022] According to the segment identifier order in the weld attribute set, the starting coordinates, ending coordinates, weld geometric center line vector, groove form code, process parameter index code and preset welding layer information of each segment are read. Then, four-dimensional coupled calculation of layer thickness, swing width, speed and current is performed in each segment to generate an automatic welding trajectory that changes with the segment.

[0023] Furthermore, the process of multi-layer, multi-pass decomposition and welding procedure generation includes:

[0024] The multi-layer, multi-pass decomposition process includes: decomposing the automatic welding trajectory into independent weld track trajectories according to segments, layers, and passes, and generating corresponding multi-axis motion interpolation points;

[0025] The welding program generation process includes: binding the multi-axis motion interpolation points with the welding current, arc voltage, welding speed and oscillation amplitude corresponding to the process parameter index codes point by point to generate the welding program, and simultaneously generating the theoretical trajectory corresponding to each layer and each pass.

[0026] Furthermore, the welding operation process of the automated welding module includes:

[0027] Based on the welding procedure, the vision guidance module is invoked to synchronously collect weld error data, forming a continuous record of the current position deviation, direction deviation, and weld boundary deviation.

[0028] Furthermore, the process of online correction of subsequent interpolation points and re-scanning of the weld seam includes:

[0029] The online correction of subsequent interpolation points includes: performing position feedback and feedforward compensation based on weld error data to generate corrected subsequent interpolation points;

[0030] The rescanning of the weld seam process includes: when the current layer welding is completed, the current pass welding is completed, the marker for the section to be updated exists, or the continuous abnormal acquisition marker reaches a preset number of trigger conditions, the global vision unit or vision guidance module is invoked to rescan the current section, update the geometric feature parameters of the weld seam cross section of the current section, regenerate the bevel form code and process parameter index code, and rebind them with the preset welding layer information to obtain the updated weld seam attribute set, and input the updated weld seam attribute set back to the path planning algorithm for processing.

[0031] Furthermore, a teach-free welding system for beveling thick-walled pipes includes: a panoramic scanning module, a point cloud registration and coordinate transformation module, a cross-sectional parameter extraction module, an attribute binding module, a path planning and interpolation generation module, a welding program generation module, a welding operation and error acquisition module, and an online correction and attribute update module; the modules are connected in sequence to implement the method described in any of the above-mentioned methods.

[0032] The key innovations of this invention include:

[0033] (1) Based on the point cloud and red-green-blue image of the weld area, downsampling, filtering, edge extraction, point cloud data to world coordinate system conversion model establishment, transformation matrix acquisition, model and point cloud registration, point cloud registration and feature extraction algorithm processing, as well as root gap width, bevel angle, blunt edge thickness and base material thickness extraction processing, to form a continuous data link of workpiece actual pose, weld geometric center line vector, weld cross-sectional geometric feature parameters, bevel form code and process parameter index code.

[0034] (2) Based on the groove form code and process parameter index code, perform preset welding layer information binding and weld geometric center line vector binding processing to obtain a weld attribute set, so that the weld attribute set simultaneously includes groove form code, process parameter index code, preset welding layer information and weld geometric center line vector.

[0035] (3) Based on the weld error data, perform position feedback, feedforward compensation and online correction of subsequent interpolation points, and perform weld rescanning, weld cross-sectional geometric feature parameter update and weld attribute set update based on the subsequent interpolation points to obtain the updated weld attribute set, and input the updated weld attribute set back to the path planning algorithm processing link.

[0036] The following are its main beneficial effects:

[0037] (1) Around the bevel area of ​​thick-walled pipe, the present invention sequentially associates the weld area point cloud, the actual pose of the workpiece, the geometric center line vector of the weld, the geometric feature parameters of the weld cross section, the bevel form code and the process parameter index code. The problem of unstable correspondence between the weld area point cloud and the actual pose of the workpiece in the prior art is addressed in a targeted manner, and the subsequent path planning algorithm processing and process parameter index code calling have a unified input basis.

[0038] (2) Based on the set of weld attributes, the present invention puts the preset welding layer information, groove form code, process parameter index code and weld geometric center line vector into the same processing object. The problem of the disconnect between the geometric feature parameters of the weld cross section and the preset welding layer information in the prior art is addressed in a targeted manner, and a continuous correspondence is formed between the automatic welding trajectory, multi-layer and multi-pass decomposition and welding program generation.

[0039] (3) Based on the updated weld attribute set, the present invention connects weld error data, online correction of subsequent interpolation points, rescanning of weld, updating of weld cross-sectional geometric feature parameters and updating of weld attribute set into the same closed loop link. The problem that weld error data is difficult to write back to the subsequent trajectory processing link in the prior art is addressed in a targeted manner. The subsequent automatic welding trajectory and theoretical trajectory can be regenerated according to the current welding state.

[0040] (4) Regarding the operation link of multi-layer and multi-pass welding, the present invention uses a path planning algorithm driven by weld attribute set to process, automatically generate welding trajectory, generate multi-axis motion interpolation points, generate welding program and generate theoretical trajectory. The processing method of relying on workpiece three-dimensional model, program template or single scan result to complete welding program and theoretical trajectory generation in the prior art is supplemented. The process switching and trajectory switching in the bevel area of ​​thick-walled pipe have a unified basis.

[0041] (5) Regarding the welding operation process of the automatic welding module, the present invention connects the sequence of weld error data acquisition and processing, position feedback, feedforward compensation, online correction of subsequent interpolation point processing and updated weld attribute set feedback. The problem of difficulty in continuous connection between welding operation, process parameter index code call and subsequent update in the prior art is addressed in a targeted manner, and the integrity of the link for multi-layer and multi-pass welding in the bevel area of ​​thick-walled pipe is maintained. Attached Figure Description

[0042] Figure 1 A flowchart illustrating a method for teaching-free beveling of thick-walled pipes provided in this application embodiment;

[0043] Figure 2 This is a structural block diagram of a teach-free welding system for beveling thick-walled pipes, provided in an embodiment of this application. Detailed Implementation

[0044] Example 1: Refer to Figure 1 This is a flowchart illustrating a method for teach-free beveling of thick-walled pipes according to an embodiment of the present invention. The process may include at least steps S100-S400:

[0045] S100: Acquire global vision unit, industrial camera, laser line structured light and assembled thick-walled pipe bevel area, perform partition scanning, point cloud registration and model and point cloud registration processing to obtain the actual pose of the workpiece and the geometric centerline vector of the weld.

[0046] S200. Based on the actual pose of the workpiece and the geometric centerline vector of the weld, perform geometric feature parameter extraction of the weld cross section, groove form code matching and preset welding layer information binding processing to obtain a set of weld attributes.

[0047] S300. Based on the weld attribute set, perform path planning algorithm, multi-layer multi-pass decomposition and welding program generation processing to obtain welding program and theoretical trajectory;

[0048] S400. Based on the welding program and theoretical trajectory, perform welding operations using the automatic welding module, correct subsequent interpolation points online, and rescan the weld seam to obtain an updated set of weld seam attributes.

[0049] Step S100 includes at least steps S110-S130:

[0050] S110: Acquire global vision unit, industrial camera, laser line structured light and assembled thick-walled pipe bevel area, perform partition scanning, stitching algorithm and red green blue (RGB) image acquisition and processing to obtain weld area point cloud and RGB image.

[0051] Specifically, the global vision unit is the acquisition and execution unit of this step. The global vision unit includes at least an industrial camera, a laser line structured light, a mounting bracket, a control interface, and an acquisition position set separately from the automatic welding module. The industrial camera is used to receive reflected light information from the surface of the bevel region of the thick-walled pipe. The laser line structured light is used to form a continuous light band outline on the surface of the bevel region of the thick-walled pipe. The assembled thick-walled pipe bevel region is the on-site input object of this step. The thick-walled pipe bevel region includes at least the bevel opening to be welded, the base material surfaces on both sides of the bevel, and the positioning reference area adjacent to the bevel.

[0052] In practice, the assembled thick-walled tube is first fixed on a rotating support or positioning clamping structure. Then, the global vision unit scans the adjacent areas of the thick-walled tube in the circumferential and axial directions according to a preset partition scanning sequence. The partition scanning does not cover the entire outer surface of the thick-walled tube at once, but is divided into several adjacent scanning segments according to the length of the bevel area, the width of the bevel area, and the field of view of the industrial camera. The current scanning segment is collected first, and then the next scanning segment is switched until the complete bevel area is covered.

[0053] The stitching algorithm is used to perform positional correspondence and continuation processing on the three-dimensional contour information and image boundary information of adjacent scanning sections. During stitching, the contour boundaries of each scanning section are read first according to the scanning order, and then the contours are stitched according to the overlapping areas of adjacent scanning sections, so that the bevel area of ​​the thick-walled pipe forms a continuous weld area point cloud in the same acquisition round.

[0054] The RGB image acquisition and processing are carried out simultaneously with the partitioned scanning. In each scanning segment, the industrial camera acquires the surface image of the bevel area during the laser line structured light projection and records the surface color information corresponding to the point cloud of the weld area. This information is used to identify the bevel boundary, the base material boundary, and the occluded area during subsequent edge extraction.

[0055] Understandably, when there is reflection, oxide scale or local contamination on the surface of the thick-walled tube, this step does not change the basic composition of the global vision unit and industrial camera. Instead, during the partitioned scanning process, the exposure order of the industrial camera and the projection order of the laser line structured light are adjusted to re-acquire the current scanning segment. The re-acquisition results are recorded separately from the previous results, and the original acquisition order is retained to facilitate the subsequent steps to filter abnormal scanning segments.

[0056] Furthermore, after the partitioned scanning is completed, the three-dimensional contour information of each scanned segment is processed by a stitching algorithm to generate a complete weld area point cloud, and the surface images corresponding to each scanned segment are organized into RGB images according to the time correspondence. The weld area point cloud and RGB image are the output products of this step and are recorded as the output field name "weld area point cloud and RGB image". They are directly input into "weld area point cloud and RGB image" in S120, and also serve as the pre-entry input for the point cloud organization step in the main step of S100 when the acquisition stage is transferred.

[0057] S120. Based on the weld area point cloud and RGB image, perform downsampling, filtering and edge extraction processing to obtain the weld area point cloud.

[0058] Specifically, in this step, the weld area point cloud and RGB image output in the previous section are used as the sole input source, wherein the weld area point cloud corresponds to the three-dimensional contour set of the thick-walled pipe bevel area, and the RGB image corresponds to the surface image set under the same scanning cycle.

[0059] The downsampling is a point cloud processing operation. The downsampling does not delete the entire scanned segment, but rather retains and unifies the high-density repeated points, redundant points outside the edge, and adjacent overly dense points in the point cloud of the weld area, so that the point spacing in the same bevel area remains basically consistent, which facilitates the subsequent registration of the model and the point cloud.

[0060] The filtering is a point cloud purification action. The filtering at least targets discrete noise points formed by surface reflection, flash, oil stains and local occlusion. The processing order is to first identify isolated points, then identify jump points that are not vertically discontinuous with adjacent points, and then retain continuous points located near the bevel boundary and the base material boundary, so that the point cloud in the weld area retains the true contour while removing abnormal scattered points.

[0061] The edge extraction is an image-assisted recognition action. In this process, the gray-scale change area at the bevel opening and the turning area of ​​the base material surface are first identified from the RGB image. Then, the boundary position in the image is matched with the contour change position in the weld area point cloud to obtain the edge information corresponding to the bevel opening, the two side edges of the bevel and the adjacent base material surface. This edge information is then written back to the weld area point cloud so that the weld area point cloud obtained in subsequent steps not only contains the point set, but also the bevel boundary position that has been processed by edge extraction.

[0062] Furthermore, during the continuous execution of downsampling, filtering, and edge extraction, the trigger condition is that S110 completes the stitching algorithm processing and forms a complete weld seam area point cloud and RGB image. During the operation, if the number of points in a certain scanning segment is insufficient after filtering, or the correspondence between the image edge and the point cloud contour is interrupted, the scanning segment is marked as an abnormal scanning segment, and the resampling results retained in S110 are called to cover the original abnormal scanning segment, and then downsampling, filtering, and edge extraction processing are performed again.

[0063] Understandably, the output of this step is still recorded as the weld area point cloud, but this weld area point cloud is no longer the original acquisition result in S110, but the processed result after completing the unification of point spacing, screening of abnormal scattered points and correspondence of bevel boundaries.

[0064] The processing result is recorded as the output field name "Weld Region Point Cloud", which is directly input into "Weld Region Point Cloud" of S130, and also serves as the basic data for subsequent point cloud data-world coordinate system transformation model establishment, transformation matrix calculation and model-point cloud registration processing.

[0065] S130. Based on the point cloud of the weld area, establish a point cloud data-world coordinate system transformation model, obtain the transformation matrix, and perform model-point cloud registration processing to obtain the actual pose of the workpiece and the geometric centerline vector of the weld.

[0066] Specifically, this step uses the weld area point cloud output in the previous section as input. The point cloud data-world coordinate system transformation model is a correspondence model that transforms the weld area point cloud from the industrial camera acquisition position to the on-site spatial position of the thick-walled pipe. The world coordinate system is a unified coordinate reference used by the global vision unit, the thick-walled pipe positioning and clamping structure, and the automatic welding module. The transformation matrix is ​​coordinate transformation data that describes the positional and directional relationships between the weld area point cloud at the acquisition position and the world coordinate system.

[0067] During implementation, the installation position of the global vision unit, the acquisition position of the industrial camera, and the clamping position of the thick-walled tube positioning and clamping structure are first read. Then, the edge position, bevel opening position, and adjacent area position of the base material in the point cloud of the weld area are mapped to the world coordinate system to form a point cloud data-world coordinate system transformation model.

[0068] The transformation matrix is ​​then obtained by calculating the current position offset and direction offset based on the extracted bevel boundary position in the weld area point cloud and the corresponding reference position in the workpiece 3D model. The results are then organized into a unified transformation matrix.

[0069] In this step, the 3D model of the workpiece is used as the input of the theoretical model. The theoretical model does not directly replace the measured results, but serves as a reference for the registration of the model and the point cloud. The registration of the model and the point cloud is a process of matching the position of the 3D model of the workpiece with the point cloud of the weld area according to the same coordinate reference and the boundary correspondence. During the process, a coarse correspondence is first performed, and then a fine correspondence is performed based on the bevel boundary, the base material surface and the adjacent positioning reference area, so as to finally obtain the actual position of the assembled thick-walled tube in the current field.

[0070] The actual pose of the workpiece includes the position and orientation of the workpiece in the world coordinate system. The geometric centerline vector of the weld is a geometric direction description formed by continuously extracting the center position and connecting them sequentially along the opening area of ​​the thick-walled pipe bevel, provided that the actual pose of the workpiece has been determined.

[0071] Furthermore, when there are local misalignment areas between the weld area point cloud and the workpiece 3D model, this step does not directly discard the area, but first determines whether the area belongs to assembly gap, misalignment, or partial occlusion; for areas belonging to assembly gap and misalignment, the actual measured contour is retained to obtain the actual pose of the workpiece; for areas belonging to partial occlusion, the corresponding scan segment result of S110 is called back to fill the points before continuing the registration.

[0072] The output workpiece actual pose and weld geometric centerline vector are both derived from the measured weld area point cloud after assembly, and do not directly use the theoretical position in the workpiece 3D model.

[0073] Finally, this step records the actual pose of the workpiece and the geometric centerline vector of the weld as the output field name "actual pose of the workpiece and geometric centerline vector of the weld", and provides it as input to "actual pose of the workpiece and geometric centerline vector of the weld" in S210. At the same time, this output also forms the prerequisite for the extraction of geometric feature parameters of the weld cross section in the main step of S200.

[0074] Summary of the technical effects of this step: This step further refines the weld area point cloud, after downsampling, filtering, and edge extraction, into the actual workpiece pose and weld geometric centerline vector consistent with the world coordinate system. It no longer remains at the original scan level, nor does it directly use the theoretical model position. Compared to methods that only perform bevel positioning or only fit the bevel centerline, this step adds continuous processing of the point cloud data-world coordinate system transformation model, transformation matrix, and model-point cloud registration. This ensures that the input obtained in subsequent S210 already includes the measured positional relationship and weld geometric direction relationship in the assembled state.

[0075] Step S200 includes at least steps S210-S230:

[0076] S210. Obtain the actual pose of the workpiece and the geometric centerline vector of the weld, perform point cloud registration and feature extraction algorithm processing, and obtain the geometric feature parameters of the weld cross-section.

[0077] Specifically, the actual pose of the workpiece comes from the output of S130, representing the current position and orientation of the assembled thick-walled tube in the world coordinate system. The geometric centerline vector of the weld also comes from the output of S130, representing the center direction extending along the bevel region of the thick-walled tube.

[0078] The point cloud registration and feature extraction algorithm is executed by the point cloud processing module in this step. The point cloud processing module includes a registration processing unit, a section extraction unit, a contour recognition unit, and a parameter recording unit. The registration processing unit first reads the actual pose of the workpiece, then reads the point cloud of the weld region corresponding to the actual pose of the workpiece, and rearranges the weld region point cloud in a direction consistent with the geometric centerline vector of the weld. After this processing, the subsequent section extraction unit can extract points segment by segment along the geometric centerline vector of the weld, instead of directly calculating on the original scattered points.

[0079] In practice, the section extraction unit sequentially extracts multiple local sections along the geometric centerline vector of the weld at preset intervals. Each local section covers the bevel opening area, the base material surfaces on both sides of the bevel, and the adjacent area at the bottom of the bevel. The preset interval can be set as a fixed interval based on the diameter of the thick-walled pipe and the length of the bevel, or it can be set as a variable interval based on the curvature change position; the fixed interval is the core processing method, while the variable interval is the preferred extension method.

[0080] The contour recognition unit makes a continuity judgment on the point cloud contour in each local section, first identifying the left boundary of the bevel, then the right boundary of the bevel, and then the bottom turning area of ​​the bevel and the extended area of ​​the base material surface.

[0081] If a local section has occlusion or local defects, the corresponding RGB image is called for boundary verification. If the boundary is still identifiable in the RGB image, the current local section is retained and recorded as a valid section. If the boundary is not identifiable in the RGB image, the local section is marked as a section to be scanned, and transitional filling is performed along the boundary trend of the two adjacent valid sections. At the same time, the mark to be scanned is retained for use when scanning the weld again later.

[0082] Furthermore, the parameter recording unit uniformly performs feature extraction algorithm processing on all effective cross sections. The feature extraction algorithm is not an abstract model call, but rather extracts geometric quantities item by item around the contour boundary position, bottom turning position and base material surface position of the local cross section, forming cross section parameter records segment by segment, and then sums them along the geometric center line vector of the weld to obtain the geometric feature parameters of the weld cross section corresponding to the bevel of the entire thick-walled pipe.

[0083] The geometric characteristic parameters of the weld cross section include at least the changes in the opening width, bottom position, boundary angle, and distance from the base material surface of each local section, and these records are uniformly organized into the output field name "geometric characteristic parameters of weld cross section".

[0084] This output field is directly input into the "geometric feature parameters of the weld cross section" of S220 in this main step, and also serves as the pre-entry for code matching when transitioning from measured geometry to code matching in the S200 main step.

[0085] Understandably, in the actual scenario of thick-walled pipes, if the workpiece has misaligned edges, ellipticity deviation, or uneven assembly gaps, this step does not use the standard bevel position in the three-dimensional model of the workpiece, but extracts it segment by segment according to the measured local cross-section, so that the geometric feature parameters of the weld cross-section reflect the actual bevel state after assembly.

[0086] S220. Based on the geometric feature parameters of the weld cross-section, the root gap width, bevel angle, blunt edge thickness and base metal thickness are extracted to obtain the bevel form code and process parameter index code.

[0087] Specifically, the geometric characteristic parameters of the weld cross-section are derived from the output of S210. The root gap width, bevel angle, blunt edge thickness, and base metal thickness are the minimum set parameters of this step. The root gap width represents the actual distance between the two sides of the bottom of the bevel, the bevel angle represents the unfolding angle of the two sides of the bevel relative to the surface of the base metal, the blunt edge thickness represents the thickness of the unopened straight edge at the bottom of the bevel, and the base metal thickness represents the actual thickness of the base metal on both sides of the bevel from the surface to the bottom.

[0088] In practice, the parameter extraction unit first reads the boundary and bottom positions of the local sections one by one from the geometric feature parameters of the weld cross section. Then, it calculates the root gap width within the same local section and reads the angle information between the left and right boundaries relative to the base material surface to form a bevel angle record. After that, it identifies the straight edge section at the bottom of the bevel to form a blunt edge thickness record. Finally, it forms a base material thickness record from the position of the base material surface to the position of the bottom of the bevel.

[0089] The above four types of records are all saved sequentially according to the order of the geometric centerline vector of the weld, and are merged into sections within the parameter processing unit.

[0090] The segmentation is not a simple averaging, but rather a process of first determining which local cross-sectional parameters are similar and which local cross-sectional parameters change significantly within the entire circumference of the thick-walled pipe, then dividing the cross-sections with similar parameters into the same segment, and recording the cross-sections with significant changes separately.

[0091] After this processing, one or more parameter sections can be obtained in the same thick-walled pipe bevel area. Each parameter section corresponds to a set of root gap width, bevel angle, blunt edge thickness and base material thickness.

[0092] Furthermore, the code matching unit reads the four types of records for each parameter segment and matches them with the bevel form code matching table in the process database to first form a bevel form code. Then, based on the bevel form code and the four types of records for each parameter segment, it reads the process parameter index code in the process database.

[0093] The bevel type code is used to identify the type of bevel for the current thick-walled pipe, and the process parameter index code is used to call the welding current, arc voltage, welding speed and oscillation amplitude in subsequent steps.

[0094] If four types of records in a certain parameter segment fall into two adjacent code intervals at the same time, the set of codes that is closer to the root gap width and blunt edge thickness will be retained first, and the other set of codes will be recorded as spare codes. The spare codes will not be output in this step, but will be stored in the recording unit for use in the update processing after scanning the weld again.

[0095] Understandably, in a workable engineering embodiment, for a section of thick-walled conveying pipe that has been assembled and clamped, the point cloud processing module cuts multiple local sections in the circumferential direction. In one set of parameter sections, it identifies a state where the root gap width is large, the bevel angle changes little, the blunt edge thickness is continuous, and the base material thickness is stable. Based on this, the code matching unit gives a set of bevel form codes and corresponding process parameter index codes. In another set of parameter sections, it identifies a state where the root gap width is reduced and the bevel angle changes slightly, and then gives another set of bevel form codes and process parameter index codes.

[0096] In this way, a set of codes arranged in sections can be formed within a thick-walled pipe bevel area, instead of just a single code.

[0097] Finally, this step records the bevel form code and process parameter index code as the output field name "Bevel Form Code and Process Parameter Index Code", and inputs this output field into "Bevel Form Code and Process Parameter Index Code" in S230. At the same time, it retains the correspondence with the geometric feature parameters of the weld cross-section, so that it can be called in the subsequent S430 when updating the geometric feature parameters of the weld cross-section and updating the weld attribute set.

[0098] S230. Based on the bevel form code and process parameter index code, perform preset welding layer information binding and weld geometric centerline vector binding processing to obtain a weld attribute set.

[0099] Specifically, the groove form code and process parameter index code are from the output of S220, the weld geometric centerline vector is from the output of S130, the preset welding layer information is provided by the path planning pre-configuration module, and the preset welding layer information includes at least the number of layers, the number of welds per layer, the welding sequence, the layer thickness, the width of the weld, and the recording method of the start and end positions of each layer.

[0100] The preset welding layer information is not a fixed template copy, nor is it simply offset according to the number of layers. Instead, it corresponds to the bevel form code and process parameter index code segment by segment.

[0101] Specifically, the binding process first involves the attribute binding module reading the bevel form code and process parameter index code, then reading the weld geometric centerline vector segment corresponding to each code segment, and finally putting the code information, centerline information and preset welding layer information within the same segment into the same recording unit to form a segment-level binding record.

[0102] Furthermore, the attribute binding module organizes the execution order of the segment-level binding records. First, it arranges all segments according to the starting coordinates to the ending coordinates of the weld geometric centerline vector. Then, it maps the number of layers, the number of weld beads, and the welding sequence in each segment to the corresponding process parameter index codes to form a complete layer record chain.

[0103] The resulting layer record chain is neither a simple geometric record nor a simple process record, but a combined record that simultaneously includes bevel form code, process parameter index code, preset welding layer information, and weld geometric centerline vector. This combined record is the weld attribute set used in this invention.

[0104] Understandably, the weld attribute set includes segment identifier, start coordinates, end coordinates, weld geometric centerline vector, groove form code, process parameter index code, and preset welding layer information.

[0105] Its input comes from the measured results of the previous steps, and its output is a path planning algorithm for subsequent steps. Therefore, during the operation, if a certain code segment does not match the corresponding preset welding layer information, the attribute binding module first reads the preset welding layer information of the adjacent segments for comparison; when the number of layers and the number of welds of the adjacent segments are the same, the preset welding layer information is copied and marked as temporary binding; when the difference between the adjacent segments is too large, the segment is recorded as a segment to be updated, and re-bound after the weld is scanned again in S430.

[0106] Furthermore, in actual engineering implementation, for the same thick-walled pipe circumferential weld, a larger swing width and more weld beads can be bound in the section with a wider bevel opening, and a smaller swing width and fewer weld beads can be bound in the section with a narrower bevel opening. At the same time, they are still connected along the uniform weld geometric center line vector sequence. In this way, when S310 reads the weld attribute set later, what is obtained is not a simple center line, but a set of layer record chains that have been organized by section.

[0107] Finally, this step records the weld attribute set as the output field name "weld attribute set", which can be directly called by "weld attribute set" in S310. At the same time, the weld attribute set also serves as the original binding basis when S430 performs weld attribute set update processing.

[0108] Summary of the technical effects of this step: This step further organizes the bevel form code, process parameter index code, and weld geometric centerline vector obtained in the previous steps into a weld attribute set, so that the object read by the subsequent path planning algorithm changes from a simple centerline to a layer record chain. Compared with the processing method of generating a trajectory only according to the bevel centerline or automatically offsetting only according to the number of layers, this step adds segment-level binding and sequential organization processing, so that the measured geometric state of the thick-walled pipe bevel, the process parameter record, and the preset welding layer information correspond in the same output field.

[0109] In one specific embodiment, based on the actual pose of the workpiece and the geometric centerline vector of the weld, the geometric feature parameters of the weld cross section are extracted, the groove form code is matched, and the preset welding layer information is bound to obtain a set of weld attributes.

[0110] First, the point cloud processing module reads the actual pose of the workpiece and the geometric centerline vector of the weld from the output of S130, and reads the point cloud of the weld area after processing by S120. The registration processing unit rearranges the point cloud of the weld area in the same direction as the geometric centerline vector of the weld, so that subsequent section extraction can extract points segment by segment along the centerline. The section extraction unit sequentially extracts multiple local sections along the geometric centerline vector of the weld at preset intervals. Each local section covers the bevel opening area, the base material surface on both sides of the bevel, and the adjacent area at the bottom of the bevel. The contour recognition unit performs continuity judgment on the point cloud contour in each local section, and identifies the left boundary, right boundary, bottom turning area, and extended area of ​​the base material surface of the bevel. If a local section has occlusion or local defects, the RGB image corresponding to the local section is called for boundary verification; if the boundary is still not identifiable in the RGB image, the local section is marked as a section to be scanned, and transitional completion is performed along the boundary trend of the two adjacent effective sections, while the mark to be scanned is retained for subsequent use by S430. The parameter recording unit performs a feature extraction algorithm on all valid cross-sections, extracting geometric quantities item by item around the contour boundary position, bottom turning position, and base material surface position of each local cross-section. To quantify the degree of contour deviation of each local cross-section, formula ① is introduced to calculate the first... Profile discontinuity of each effective section :

[0111]

[0112] in, : No. The profile discontinuity of an effective section is calculated by formula ①; the larger the value, the greater the deviation between the point cloud profile at that section and the theoretically fitted profile.

[0113] : Index number of the effective section, with a value range of The cross section number determined as valid by the contour recognition unit in S210 is derived from the cross section number in S210.

[0114] : No. The total arc length of the effective cross section is obtained by the cross section extraction unit by accumulating the points along both sides of the bevel boundary from the point cloud of the weld area.

[0115] The summation symbol indicates that the expression following it is summed from the lower bound to the upper bound.

[0116] The summation loop variable represents the index of the discrete points on the cross-section, and its value range is... arrive ;

[0117] : No. The number of discrete points participating in the calculation on each effective cross section is derived from the number of effective points in the point cloud data of that cross section.

[0118] Euclidean norm (L2 norm), used to calculate the length of a vector; here it is used to calculate the Euclidean distance between the measured point location and the theoretical fitted location.

[0119] : No. On the effective cross section, the first The position vector of each measured contour point, including three-dimensional coordinates The data originates from the point cloud data of the weld area of ​​this cross section.

[0120] : No. On the effective cross section, the first The theoretical fitted position vector of each point is obtained by linear interpolation through the boundary trends of two adjacent effective sections, and is used as a reference benchmark.

[0121] Formula ① addresses how to quantitatively assess the degree of contour anomaly caused by occlusion, deformation, or noise in each local section, so that high-confidence sections can be prioritized for retention during subsequent segment merging.

[0122] Simple numerical example: Suppose a local cross-section of a bevel of a thick-walled pipe, with the total arc length of the profile... Millimeters, number of discrete points involved in the calculation The average Euclidean distance between each measured point and the theoretical fitting point is 0.15 mm. The value is measured in millimeters. If this value is less than a preset threshold (e.g., 0.3 mm), the cross-section is considered valid and has a good profile. Next, the parameter recording unit summarizes the changes in opening width, bottom position, boundary angle, and distance from the base material surface for each cross-section along the geometric centerline vector of the weld, forming the geometric characteristic parameters of the weld cross-section. This output field is denoted as "Geometric Characteristic Parameters of Weld Cross-section" and is directly input into the "Geometric Characteristic Parameters of Weld Cross-section" field of S220.

[0123] Furthermore, taking into account the aforementioned geometric characteristic parameters of the weld cross-section, S220 first uses the parameter extraction unit to read the boundary and bottom positions of each local section, calculating the root gap width, bevel angle, blunt edge thickness, and base metal thickness. All four types of records are saved sequentially according to the order of the weld geometric centerline vector and then merged into sections within the parameter processing unit. During section merging, it determines which local section parameters are similar and which parameters change significantly within the entire circumference of the thick-walled pipe. Sections with similar parameters are grouped into the same section, while sections with significant changes are recorded separately. To quantify the parameter similarity between adjacent sections, formula ② is introduced to calculate the... The first cross section and the second Comprehensive parameter similarity of each cross section :

[0124]

[0125] in, : The first in the vector order along the geometric centerline of the weld The first cross section and the second The comprehensive parameter similarity between the cross sections, with a value range of [value range missing]. The closer the value is to 1, the more similar the two are in terms of root clearance and bevel angle, which is calculated by formula ②.

[0126] : The sequential index number of the cross section, with a value range of , which is derived from the section numbering arranged vectorally along the geometric center line of the weld in S210;

[0127] : Natural exponential function, with The base is used to map the negative sum of squares to... interval;

[0128] : No. The root gap width of each section is derived from the corresponding item in the geometric characteristic parameters of the weld cross section output by S210;

[0129] : No. The root gap width of each section originates from the same source as above;

[0130] The normalized scaling factor for the root gap width is set based on the typical variation range in the process database, for example, 2 mm; it is used to control the sensitivity of similarity to differences in root gap.

[0131] : No. The bevel angle of each section is derived from the corresponding item in the geometric characteristic parameters of the weld cross-section output by S210;

[0132] : No. The bevel angle of each section is derived from the above.

[0133] The normalized scaling factor for the bevel angle is set based on the typical variation range in the process database, for example, 10 degrees; it is used to control the sensitivity of similarity to differences in bevel angle.

[0134] Formula ② solves the problem of how to quantitatively determine whether two adjacent local sections should be classified into the same parameter segment, avoiding unreasonable segment boundaries caused by simple averaging or artificially setting thresholds.

[0135] Simple numerical example: the width of the root gap between two adjacent cross sections. millimeters millimeters, bevel angle Spend, Degree, take millimeters The degree is then calculated. They were determined to be highly similar and were classified into the same segment.

[0136] After the segment merging is completed, the code matching unit reads the four types of records for each parameter segment and matches them with the bevel form code matching table in the process database. First, a bevel form code is generated. Then, based on the bevel form code and the four types of records for each parameter segment, the process parameter index code in the process database is read. If the four types of records for a certain parameter segment fall into two adjacent code intervals simultaneously, the code set that is closer to the root gap width and blunt edge thickness is retained first, and the other code set is recorded as a spare code. Finally, the bevel form codes and process parameter index codes arranged by segment are obtained. The output field is named "Bevel Form Code and Process Parameter Index Code," and is input into the "Bevel Form Code and Process Parameter Index Code" field in S230.

[0137] Furthermore, S230 receives the bevel form code and process parameter index code from S220, and the weld geometric centerline vector from S130. Simultaneously, the path planning pre-configuration module provides preset welding layer information (including at least the number of layers, the number of weld passes per layer, the welding sequence, layer thickness, weld width, and the recording method for the start and end positions of each layer). The attribute binding module first reads the bevel form code and process parameter index code, then reads the weld geometric centerline vector segment corresponding to each code segment. It then places the code information, centerline information, and preset welding layer information within the same segment into the same recording unit, forming a segment-level binding record. Subsequently, the segment-level binding records are sequentially organized. First, all segments are arranged from the start coordinates to the end coordinates of the weld geometric centerline vector. Then, the number of layers, the number of weld passes, and the welding sequence within each segment are mapped one-to-one with the corresponding process parameter index code, forming a complete layer record chain. During the binding process, if a code segment does not match the corresponding preset weld layer information, the attribute binding module first reads the preset weld layer information of adjacent segments for comparison. When the number of layers and welds in adjacent segments are the same, the preset weld layer information is copied and marked as temporary binding. When the difference between adjacent segments is too large, the segment is recorded as a segment to be updated and re-bound after the weld is scanned again in S430. To quantify the matching degree between the current segment and the preset weld layer information of adjacent segments, formula ③ is introduced to calculate the fit. :

[0138]

[0139] in, : No. The compatibility degree between the preset welding layer information of each segment and its adjacent segments, with a value range of [value range missing]. The closer it is to 1, the more suitable it is to directly copy the layer number and width information of adjacent sections, which is calculated by formula ③.

[0140] The index number of the segment, with a value range of: , which comes from the parameter segment number formed after the segment merging in S220;

[0141] : No. The theoretical number of layers required for each section is derived from the bevel form code and process parameter index code of that section in combination with the process database.

[0142] The layer number in the preset welding layer information of adjacent sections is directly read from the preset welding layer information of adjacent sections; whereby... It is an abbreviation for "adjacent," meaning adjacent.

[0143] : Absolute value symbol, used to calculate the magnitude of the difference;

[0144] : No. The theoretical basis for each section needs to be broadened, and it is derived from the bevel form code and process parameter index code of that section in combination with the process database.

[0145] The width of the preset welding layer information of the adjacent section is directly read from the preset welding layer information of the adjacent section.

[0146] Formula ③ solves the problem of how to quantitatively determine whether to directly copy the information of adjacent sections or mark them as sections to be updated when a certain section lacks preset welding layer information.

[0147] Simple numerical example: Theoretically, a certain section requires a certain number of layers. Preset number of layers in adjacent sections The theory needs to be broadened Millimeters, pre-set width of adjacent sections millimeters, then If the value is greater than the preset threshold of 0.8, it is determined that temporary binding is possible. If the compatibility is too low (e.g., ... If the segment is marked as pending update, then it is considered a segment to be updated.

[0148] After binding is completed, the attribute binding module outputs a set of weld attributes, which includes segment identifier, start coordinates, end coordinates, weld geometric centerline vector, groove type code, process parameter index code, and preset welding layer information.

[0149] The output field is named "Weld Attribute Set", which can be directly called by the "Weld Attribute Set" of S310, and also serves as the original binding basis when S430 performs weld attribute set update processing.

[0150] Engineering Example: A thick-walled conveying pipe with an outer diameter of 800 mm and a wall thickness of 40 mm has a circumferential root gap of 5 mm at 0° and 2 mm at 180° after assembly. 120 local sections are cut along the circumference using S210. Formula ① is used to calculate the discontinuity of the profile of each section, and 3 sections are discarded due to oxide scale obstruction. The invalid cross-sections are counted in millimeters, and the remaining 117 valid cross-sections are merged into sections. Formula ② calculates the similarity between adjacent cross-sections, assigning the 0°60° region (root gap 54.8 mm, bevel angle 3837 degrees) to section A, the 60°120° region (root gap 4.83 mm, bevel angle 3735 degrees) to section B, and the 120°180° region (root gap 32 mm, bevel angle 35~33 degrees) to section C. The code matching unit matches the V-shaped large gap bevel code P01 and process index I05 (14 mm width, 5 layers) for section A, the V-shaped medium gap code P02 and process index I03 (11 mm width, 4 layers) for section B, and the V-shaped small gap code P03 and process index I01 (8 mm width, 3 layers) for section C. S230 binds the codes, centerline vectors, and preset layer information of each segment to form a weld attribute set. At the boundary between segment B and segment C, formula ③ calculates the fit. It was confirmed that no marker was needed for updating. Finally, the set was read by S310 to generate an automatic welding trajectory that varied with the circumference.

[0151] This section summarizes the technical effects: Formulas ① to ③ enable quantitative evaluation of local cross-sectional profile quality, quantitative comparison of similarity of adjacent cross-sectional parameters, and adaptive binding decision-making when layer information is missing. This allows the weld attribute set to reflect the true non-uniform assembly state of the thick-walled pipe in the circumferential direction and provides a traceable segment-level data foundation for subsequent closed-loop updates.

[0152] Step S300 includes at least steps S310-S330:

[0153] S310. Obtain the set of weld seam attributes, perform path planning algorithm processing, and obtain the automatic welding trajectory.

[0154] Specifically, the weld attribute set comes from the output of S230, and includes segment identifier, start coordinates, end coordinates, weld geometric centerline vector, bevel form code, process parameter index code, and preset welding layer information. The path planning algorithm is executed by the path planning module in the industrial control computer, which is an execution unit that reads the weld attribute set and generates an automatic welding trajectory.

[0155] During operation, the system first reads the starting and ending coordinates of each segment in the order of segment identifiers, then organizes the connection order of each segment according to the geometric centerline vector of the weld, and then reads the bevel form code, process parameter index code and preset welding layer information corresponding to each segment.

[0156] Specifically, within each segment, the path planning module first determines the number of layers and welds in the preset welding layer information, then reads the corresponding relationship between the root gap width, bevel angle, blunt edge thickness and base material thickness within the same segment, and then performs a four-dimensional coupled calculation of layer thickness, swing width, speed and current.

[0157] The layer thickness refers to the thickness arrangement of the current layer in the bevel opening direction, the swing width refers to the swing amplitude of the current weld bead in the transverse direction, the speed refers to the running speed of the current weld bead along the geometric center line vector of the weld, and the current refers to the welding current corresponding to the current weld bead.

[0158] The four-dimensional coupling operation does not directly generate the trajectory after reading the number of layers, nor does it simply expand outward along the geometric center line vector of the weld. Instead, it first determines the order of the bottom layer, the filling layer and the cover layer within the section, and then simultaneously maps the position of the weld bead in each layer to the width, speed and current to form a continuous trajectory segment for each layer and each bead.

[0159] Furthermore, when there are multiple parameter segments within the same thick-walled pipe bevel area, the path planning module generates trajectory segments segment by segment according to the segment boundaries, and rereads the bevel form code and process parameter index code of the next parameter segment at the connection of adjacent segments, and recalculates the four-dimensional coupled calculation results of layer thickness, swing width, speed and current of the next parameter segment, so that the automatic welding trajectory changes with the segment.

[0160] Understandably, in engineering implementation, for a thick-walled conveying pipe that has been assembled and clamped, if the root gap width on one side of the circumference is greater than that on the other side, the path planning module will provide a larger number of weld passes in that side segment and a smaller number of weld passes in the relatively narrower segment, and output them in the order of the geometric center line vector of the same weld. If a segment is marked as a segment to be updated, the path planning module will only retain the segment identifier, start coordinates, and end coordinates of that segment, and will not output the complete trajectory segment. It will re-enter this step after S430 is updated.

[0161] Finally, this step organizes the continuous trajectory segments corresponding to all sections into the output field name "Automatic Welding Trajectory", and inputs the Automatic Welding Trajectory into "Automatic Welding Trajectory" in S320; at the same time, the Automatic Welding Trajectory also serves as the prerequisite for the generation of theoretical trajectory and online correction of subsequent interpolation points in the main step of S400, and is called again by this step after the updated weld attribute set is output in S430.

[0162] S320. Based on the automatic welding trajectory, perform multi-layer multi-pass decomposition and multi-axis motion interpolation point generation processing to obtain multi-axis motion interpolation points.

[0163] Specifically, the automatic welding trajectory comes from the output of S310, the multi-layer multi-pass decomposition is executed by the decomposition processing unit in the industrial control computer, and the multi-axis motion interpolation point generation is executed by the interpolation processing unit in the robot controller. At the start of processing, the decomposition processing unit first reads the trajectory segments of each layer according to the segment order in the automatic welding trajectory, and then, according to the layer number order and weld bead number order in the preset welding layer information, it decomposes each trajectory segment in each layer into an independent weld bead trajectory.

[0164] The multi-layer, multi-pass decomposition does not involve uniformly cutting off the entire automatic welding trajectory, but rather unfolding it continuously by segment, layer, and pass, so that each weld pass trajectory has a clear segment identifier, start coordinates, end coordinates, and welding torch posture. The welding torch posture is a spatial orientation record of the automatic welding module on the current weld pass, which includes at least the forward direction consistent with the geometric center line vector of the weld and the pointing relationship corresponding to the bevel opening direction.

[0165] The interpolation processing unit reads the starting coordinates, ending coordinates, welding torch posture, oscillation amplitude, and welding speed of each weld path, and generates multi-axis motion interpolation points segment by segment by combining the motion relationship between the automatic welding module, linear guide rail, and single-axis positioner.

[0166] Specifically, in sections near the thick-walled pipe where the circumferential direction changes significantly, a single-axis positioner undertakes the circumferential following motion, an automatic welding module handles the welding torch posture adjustment, and a linear guide rail handles the axial displacement adjustment of the welding torch; in sections near the thick-walled pipe where the axial direction changes less, the automatic welding module and the linear guide rail jointly complete the trajectory following.

[0167] The multi-axis motion interpolation points are used to transcribe each weld trajectory into a point sequence that can be directly read by the automatic welding module, linear guide, and single-axis positioner. This point sequence not only records the spatial position, but also the switching positions of the welding torch posture, oscillation amplitude, and welding speed.

[0168] Furthermore, when a weld bead trajectory crosses the boundary of an adjacent parameter segment, the interpolation processing unit re-establishes the next set of multi-axis motion interpolation points at the segment boundary, without directly using the previous set of point sequences, so that the motion points of adjacent parameter segments correspond to their respective weld attribute sets.

[0169] Understandably, in real-world scenarios, for a thick-walled pipe with a large outer diameter and wall thickness, the automatic welding module moves along the linear guide rail, the single-axis positioner drives the workpiece to rotate, the multi-axis motion interpolation points are output sequentially according to the weld sequence, and the robot controller drives each execution component to move synchronously according to the point sequence, completing one layer before switching to the next.

[0170] If there are sections to be updated in the current weld bead segment, the interpolation processing unit pauses the expansion of the subsequent point sequence of that segment, only outputs the multi-axis motion interpolation points of the currently completed segment, and leaves the sections to be updated to be replanned after S430.

[0171] Finally, this step records the point sequence as the output field name "multi-axis motion interpolation point" and inputs the multi-axis motion interpolation point into the "multi-axis motion interpolation point" of S330; at the same time, this output field also constitutes the motion basis when S410 performs welding operations of the automatic welding module.

[0172] S330. Based on the multi-axis motion interpolation points, process parameter index code calling, welding program generation, and theoretical trajectory generation are performed to obtain the welding program and theoretical trajectory.

[0173] Specifically, the multi-axis motion interpolation point comes from the output of S320. Although the process parameter index code comes from S220, it already corresponds to each weld track when S320 generates the multi-axis motion interpolation point. Therefore, when reading the multi-axis motion interpolation point in this step, the corresponding process parameter index code can be read simultaneously.

[0174] During operation, the industrial control computer first reads the segment sequence, layer sequence, and track sequence of the multi-axis motion interpolation points, and then calls the corresponding process parameter index code according to the point sequence to retrieve the welding current, arc voltage, welding speed, and oscillation amplitude corresponding to the code from the process database.

[0175] The robot controller binds the multi-axis motion interpolation points to the welding current, arc voltage, welding speed, and oscillation amplitude point by point, generating a welding program that can be directly executed by the automatic welding module.

[0176] The welding program includes robot motion instructions, welding instructions, and arc tracking instructions. The robot motion instructions correspond to the displacement sequence of the multi-axis motion interpolation points and the welding torch posture sequence. The welding instructions correspond to the switching sequence of welding current, arc voltage, and oscillation amplitude. The arc tracking instructions correspond to the tracking and calling sequence of the welding operation stage of the automatic welding module in S410.

[0177] The theoretical trajectory is jointly generated by the path planning module and the robot controller. Specifically, it is formed by continuously connecting the multi-axis motion interpolation points in the world coordinate system to create a set of reference trajectories for each layer and pass. This theoretical trajectory is not a single centerline, but a set of reference trajectories corresponding to each pass of the welding program. After the S410 collects the weld error data, it is directly compared with this theoretical trajectory.

[0178] Furthermore, in the implementation of the project, for the circumferential weld of a section of thick-walled conveying pipe, the industrial control computer first generates a welding program in the order of the first pass of the first layer to the last pass of the last layer, and then simultaneously generates the corresponding theoretical trajectory; after the automatic welding module reads the welding program, it enters the welding operation, and the vision guidance module collects weld error data according to the arc tracking command during the welding process. S410 and S420 perform position feedback, feedforward compensation and online correction of subsequent interpolation points around the theoretical trajectory.

[0179] If S430 outputs an updated weld attribute set, this step does not directly modify the current welding program. Instead, it waits for S310 and S320 to regenerate the automatic welding trajectory and multi-axis motion interpolation points based on the updated weld attribute set before calling the process parameter index code again to generate a new welding program and a new theoretical trajectory.

[0180] Finally, this step records the program results and reference trajectory results as a unified output field name "Welding Program and Theoretical Trajectory", and inputs this output field into "Welding Program and Theoretical Trajectory" in S410; at the same time, the theoretical trajectory serves as the comparison benchmark for S420 to perform online correction of subsequent interpolation point processing.

[0181] Summary of the technical effects of this step: This step expands the weld attribute set through two levels—automatic welding trajectory and multi-axis motion interpolation points—and further organizes it into a welding program and a theoretical trajectory. This ensures that the differences in sections, layers, and process parameters of thick-walled pipe bevels are preserved simultaneously at the program level and the reference trajectory level. Compared to processing methods that only generate a single-pass trajectory or only call the program according to a fixed number of layers, this step places the process parameter index code call, welding program generation, and theoretical trajectory generation in the same link. Subsequent steps S410 to S430 can directly perform welding operations, online corrections, and re-scan updates around the same reference object.

[0182] Step S400 includes at least steps S410-S430:

[0183] S410. Obtain the welding program and theoretical trajectory, perform welding operations and weld error data acquisition and processing using the automatic welding module, and obtain weld error data.

[0184] Specifically, the welding program and theoretical trajectory are derived from the output of S330. The welding program includes robot motion instructions, welding instructions, and arc tracking instructions. The theoretical trajectory is a set of reference trajectories corresponding to each layer and each pass.

[0185] The automatic welding module is the main body for this step. The automatic welding module includes at least a welding torch, a robot body, a linear guide rail, a single-axis positioner, and a control interface. The welding torch is used to perform the welding operation of the current weld bead. The robot body is used to drive the welding torch to move along the multi-axis motion interpolation point. The linear guide rail is used to complete the axial displacement. The single-axis positioner is used to complete the circumferential direction following.

[0186] The acquisition and processing of weld error data is completed collaboratively by the vision guidance module and the robot controller. The vision guidance module is a laser vision sensor fixedly installed at the end of the welding torch. Its acquisition direction is consistent with the current weld bead's forward direction, and its acquisition area covers the bevel boundary, the edge of the molten pool, and the adjacent area of ​​the already welded bead in front of the current weld bead.

[0187] During operation, the robot controller first calls the robot motion instructions according to the layer and pass order in the welding program, and then calls the welding instructions to complete the switching of welding current, arc voltage, welding speed and oscillation amplitude. At the same time, it calls the arc tracking instruction to make the vision guidance module enter the synchronous acquisition state.

[0188] The synchronous acquisition state refers to the vision guidance module moving together with the automatic welding module, continuously reading the micro-contour acquisition results of the weld during the current weld bead operation, and mapping the acquisition results to the adjacent trajectory positions in the current theoretical trajectory in chronological order.

[0189] Furthermore, during the welding operation, the vision guidance module first reads the contour image in front of the current weld bead, then reads the boundary positions on both sides of the current weld bead, and then matches the image boundary positions with the laser contour positions to form a record of the actual weld center, weld bead boundary positions, and local contour changes.

[0190] If the welding arc light is strong, High Dynamic Range (HDR) imaging is performed within the current acquisition cycle, followed by morphological filtering and edge extraction. If the welding arc light is weak, edge extraction is performed directly. The robot controller correlates the above acquisition results with the reference position in the current theoretical trajectory segment by segment, forming a continuous record of the current position deviation, direction deviation, and weld boundary deviation. This continuous record is the weld error data in this step.

[0191] Understandably, when the circumferential weld of the thick-walled pipe enters the section boundary position, the vision guidance module does not change the acquisition method, but continues to acquire data along the current welding procedure sequence; when the robot controller records the weld error data, it adds the current section identifier and the current weld sequence, so that the subsequent S420 can perform position feedback and feedforward compensation by section, layer, and weld.

[0192] If obstruction, splashing, or localized strong reflection occurs during a certain acquisition cycle, the robot controller first retains the continuous record of the previous acquisition cycle, and then adds an abnormal acquisition mark to the current acquisition cycle without directly interrupting the welding operation; when the abnormal acquisition mark appears consecutively for a preset number of times, the current weld error data is retained and the current section is recorded as a section to be updated for S430 to call.

[0193] Finally, this step organizes the continuous records into an output field named "Weld Error Data" and inputs the weld error data into "Weld Error Data" in S420. At the same time, the weld error data also maintains a correspondence with the current welding program, the current theoretical trajectory, and the current section identifier, so that S430 can check the differences before and after the update after scanning the weld again.

[0194] S420. Based on the weld error data, perform position feedback, feedforward compensation, and online correction of subsequent interpolation points to obtain subsequent interpolation points.

[0195] Specifically, the weld error data comes from the output of S410, the position feedback is the real-time correction action performed by the robot controller based on the current position difference between the actual weld center, weld boundary position and theoretical trajectory collected at the moment, and the feedforward compensation is the processing action of the robot controller to adjust adjacent points that have not yet been executed in advance based on the deviation change trend in multiple consecutive collection cycles.

[0196] During implementation, the robot controller first reads the weld error data according to the current segment identifier, then reads the corresponding position in the theoretical trajectory according to the current layer order and the current weld order, and then judges whether the current position deviation, direction deviation and weld boundary deviation have reached the preset threshold.

[0197] The preset threshold is the trigger condition for online correction of subsequent interpolation points, and the preset threshold includes at least a lateral offset threshold and an angular deviation threshold.

[0198] If both the current position deviation and direction deviation are below the preset threshold, the robot controller retains the order of the current multi-axis motion interpolation points and only writes the weld error data of this acquisition cycle into the data traceability record; if one of them reaches the preset threshold, it enters the position feedback processing; if multiple consecutive acquisition cycles are offset in the same direction, it continues to enter the feedforward compensation processing after the position feedback processing.

[0199] Specifically, during position feedback processing, the robot controller first maps the current weld error data to the point sequence of the current weld bead, and then maps the current position deviation to the current position correction amount of the automatic welding module, linear guide, and single-axis positioner, so that the welding torch posture and the current point position are synchronously adjusted within the current weld bead. During feedforward compensation processing, the robot controller continues to read the adjacent points after the current weld bead that have not yet been processed, and rewrites the deviation change trend in multiple consecutive acquisition cycles into the correction amount of the next point sequence, and writes this correction amount into the point sequence that has not yet been processed. The resulting corrected point sequence is the result of online correction of subsequent interpolation point processing.

[0200] Furthermore, the online correction of subsequent interpolation points does not rewrite all points uniformly, but rewrites them segment by segment according to the current section, the current layer, and the current weld bead; for point sequences that have been executed, only their original records and weld error data records are retained, and no backwriting is performed; for point sequences that have not been executed, corrections are performed within the section according to the current position deviation and direction deviation.

[0201] If the current weld bead is close to the boundary of the parameter segment, the robot controller will only correct the points before the boundary in the current segment, retain the original order of the points after the boundary, and record the points after the boundary as points to be scanned again.

[0202] Understandably, in actual engineering scenarios, when a certain circumferential section of a thick-walled pipe is welded in the filler layer, the weld boundary position shifts to one side due to the accumulation of heat input. After the vision guidance module continuously collects the deviation in the same direction, the robot controller first provides position feedback on the current weld position, then performs feedforward compensation on the remaining points of the weld and the front points of the next weld, and then outputs the corrected subsequent interpolation points.

[0203] If a sharp change occurs in multiple consecutive acquisition cycles, the robot controller retains the completed portion of the current weld, pauses further correction of the points after the boundary, and records the current segment as the segment to be updated, handing it over to S430 to perform another scan of the weld.

[0204] Finally, this step records the corrected point sequence as the output field name "Subsequent Interpolation Points" and inputs the subsequent interpolation points into "Subsequent Interpolation Points" in S430; at the same time, the subsequent interpolation points continue to be called in real time by the automatic welding module during the current welding operation, forming the basis for online correction execution of the current layer or the current pass.

[0205] S430. Based on the subsequent interpolation points, perform a second scan of the weld, update the geometric feature parameters of the weld cross-section, and update the weld attribute set to obtain the updated weld attribute set.

[0206] Specifically, the subsequent interpolation points come from the output of S420, and the rescanning of the weld seam is a scanning process performed on the current segment by the global vision unit or vision guidance module after completing one or one layer corresponding to the current preset welding layer information.

[0207] The triggering conditions for re-scanning the weld include at least one of the following: the current layer welding is completed, the current pass welding is completed, the section to be updated is marked, or the continuous abnormal acquisition mark reaches a preset number.

[0208] At the start of processing, the industrial control computer first reads the current segment identifier, current layer sequence, and current weld sequence corresponding to the subsequent interpolation points, and then determines whether the triggering conditions for re-scanning the weld are met. When the triggering conditions are met, the execution of the welding program for the next layer or the next pass is paused, the global vision unit is called to re-execute the partition scan and RGB image acquisition of the current segment, or the vision guidance module is called to perform a local scan near the current position, and the re-acquisition results are organized into a new weld area point cloud and RGB image.

[0209] Subsequently, the point cloud processing module performs the same point cloud registration and feature extraction algorithm as S210 on the new weld area point cloud and RGB image to obtain the geometric feature parameters of the weld cross section of the current segment.

[0210] The update of the geometric feature parameters of the weld cross section is not a uniform replacement of the entire circumference, but a local replacement based on the current segment. That is, the original records of the unchanged segments are retained, and only the root gap width, bevel angle, blunt edge thickness and base metal thickness records corresponding to the current segment are updated.

[0211] The updated weld cross-sectional geometric feature parameters are then input into the code matching process to regenerate the groove form code and process parameter index code for the current section. The new groove form code, the new process parameter index code, the preset welding layer information corresponding to the current section, and the weld geometric centerline vector are then rebound to complete the weld attribute set update process.

[0212] Furthermore, in the weld attribute set update process, if the parameter difference between the current segment and the adjacent segment is reduced after rescanning, the original segment boundary remains unchanged, and only the records inside the current segment are replaced; if the parameter difference between the current segment and the adjacent segment is increased after rescanning, the segment boundary is redefined, and the new segment identifier is written into the updated weld attribute set.

[0213] The updated weld attribute set can also synchronously generate welding quality information, which includes at least the weld defect probability, quality confidence level, and OK or NG result corresponding to weld width, reinforcement height, and porosity.

[0214] Understandably, in a complete and operable engineering embodiment, for a thick-walled conveying pipe in the transition stage before entering the filling layer after the bottom layer is completed, if S420 has recorded that there is a continuous offset in the same direction in the current circumferential section, then S430 triggers a rescan of the weld seam; the point cloud of the weld seam area after rescanning shows that the bevel opening of the section has changed locally, and the point cloud processing module updates the root gap width and bevel angle accordingly, the code matching process re-provides the process parameter index code, and the attribute binding module generates a new set of weld seam attributes.

[0215] The updated weld attribute set does not directly generate the automatic welding trajectory in this step. Instead, it is recorded as the output field name "updated weld attribute set" and fed back to the "weld attribute set" of S310. S310, S320 and S330 then re-execute the path planning algorithm, multi-layer multi-pass decomposition, multi-axis motion interpolation point generation, process parameter index code call, welding program generation and theoretical trajectory generation processing.

[0216] In this way, the operating chain consisting of S410 to S430 and the operating chain consisting of S310 to S330 form a closed-loop update relationship at the current segment.

[0217] In summary, this step extends the online correction of subsequent interpolation points to include rescanning the weld, updating the geometric features of the weld cross-section, and updating the weld attribute set. This ensures that subsequent welding of thick-walled pipe bevels no longer depends on the initial scan results. Compared to methods that only perform real-time correction during welding without writing back the layer information, this step re-integrates the re-sampling results into the weld attribute set, allowing subsequent automatic welding trajectories, welding programs, and theoretical trajectories to be regenerated around the updated current segment.

[0218] Example 2: Figure 2 A structural block diagram of a teach-free welding system for beveling thick-walled pipes according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include:

[0219] The panoramic scanning module 01 is used to acquire data from the global vision unit, industrial camera, laser line structured light, and the assembled thick-walled pipe bevel area. It performs partitioned scanning, a stitching algorithm, and red-green-blue image acquisition and processing to obtain a weld area point cloud and red-green-blue images. Specifically, the panoramic scanning module receives the on-site status information of the global vision unit, industrial camera, laser line structured light, and the assembled thick-walled pipe bevel area. It drives the industrial camera and laser line structured light to move synchronously according to the partitioned scanning sequence, first acquiring the surface contour of each scanning segment, then acquiring the corresponding red-green-blue images, and performing a stitching algorithm on adjacent scanning segments to form a weld area point cloud and red-green-blue images covering the thick-walled pipe bevel area. The weld area point cloud and red-green-blue images are sent as output objects of this module to the point cloud registration and coordinate transformation module for use as the weld area point cloud and red-green-blue images. Simultaneously, the original acquisition records are retained within this module for comparison and reading during subsequent weld scans.

[0220] The point cloud registration and coordinate transformation module 02, connected to the panoramic scanning module, is used to perform downsampling, filtering, edge extraction, point cloud data to world coordinate system transformation model establishment, transformation matrix calculation, and model-point cloud registration processing based on the weld area point cloud and red-green-blue images, to obtain the actual workpiece pose and weld geometric centerline vector. Specifically, the point cloud registration and coordinate transformation module receives the weld area point cloud and red-green-blue images output by the panoramic scanning module, first performs downsampling, filtering, and edge extraction on the weld area point cloud, then establishes a point cloud data to world coordinate system transformation model according to the installation relationship between the global visual unit and the bevel area of ​​the thick-walled pipe, and performs transformation matrix calculation and model-point cloud registration processing on the weld area point cloud. After processing, the point cloud registration and coordinate transformation module outputs the actual workpiece pose and weld geometric centerline vector, and sends the actual workpiece pose and weld geometric centerline vector to the cross-sectional parameter extraction module as input objects, while registering the current scanning round and registration status for use by the online correction and attribute update module.

[0221] The cross-sectional parameter extraction module 03, connected to the point cloud registration and coordinate transformation module, is used to acquire the actual pose of the workpiece and the geometric centerline vector of the weld, perform point cloud registration and feature extraction algorithm processing, and extract the root gap width, bevel angle, blunt edge thickness, and base material thickness to obtain the weld cross-sectional geometric feature parameters, bevel form code, and process parameter index code. Specifically, the cross-sectional parameter extraction module receives the actual pose of the workpiece and the geometric centerline vector of the weld from the point cloud registration and coordinate transformation module, retrieves the corresponding weld area point cloud, segments local cross-sections along the weld geometric centerline vector, and performs point cloud registration and feature extraction algorithms on each local cross-section to form the weld cross-sectional geometric feature parameters. The cross-sectional parameter extraction module then extracts the root gap width, bevel angle, blunt edge thickness, and base material thickness from the weld cross-sectional geometric feature parameters, and organizes them by segment to obtain the bevel form code and process parameter index code. The geometric feature parameters of the weld cross-section, the groove form code, and the process parameter index code are sent as output objects to the attribute binding module. The groove form code and the process parameter index code are used as inputs for binding processing. The geometric feature parameters of the weld cross-section are stored in this module for the online correction and attribute update module to perform update processing.

[0222] The attribute binding module 04, connected to the cross-sectional parameter extraction module, is used to perform preset welding layer information binding and weld geometric centerline vector binding processing based on the bevel form code and process parameter index code to obtain a weld attribute set. Specifically, the attribute binding module receives the bevel form code and process parameter index code output by the cross-sectional parameter extraction module, and retrieves the weld geometric centerline vector and preset welding layer information, performing binding processing in segment order. The binding processing includes segment identifier writing, starting coordinate and ending coordinate correspondence, weld geometric centerline vector association, preset welding layer information association, and process parameter index code association. After processing, a weld attribute set is generated. The weld attribute set is provided as an output object to the path planning and interpolation generation module for use as the weld attribute set. At the same time, this module registers the association record between the current segment and the current layer for the online correction and attribute update module to update.

[0223] The path planning and interpolation generation module 05, connected to the attribute binding module, is used to acquire the weld attribute set, perform path planning algorithm processing, multi-layer multi-pass decomposition, and multi-axis motion interpolation point generation processing to obtain the automatic welding trajectory and multi-axis motion interpolation points. Specifically, the path planning and interpolation generation module receives the weld attribute set from the attribute binding module, reads the bevel form code, process parameter index code, preset welding layer information, start coordinates, end coordinates, and weld geometric centerline vector from the weld attribute set by segment, layer, and pass, and performs path planning algorithm processing on each segment to generate the automatic welding trajectory. Subsequently, the path planning and interpolation generation module performs multi-layer multi-pass decomposition on the automatic welding trajectory and generates multi-axis motion interpolation points according to the motion relationship between the automatic welding module, linear guide rail, and single-axis positioner. The automatic welding trajectory and multi-axis motion interpolation points are sent as output objects to the welding program generation module. The multi-axis motion interpolation points are used as program generation inputs, and the automatic welding trajectory is retained as a trajectory recording object for the online correction and attribute update module to recall when writing back the updated weld attribute set.

[0224] The welding program generation module 06, connected to the path planning and interpolation generation module, is used to perform process parameter index code invocation, welding program generation, and theoretical trajectory generation based on the multi-axis motion interpolation points to obtain the welding program and theoretical trajectory. Specifically, the welding program generation module receives the multi-axis motion interpolation points output by the path planning and interpolation generation module, and calls the corresponding process parameter index codes according to the segment order, layer order, and pass order of the multi-axis motion interpolation points, thereby generating welding current, arc voltage, welding speed, and oscillation amplitude records corresponding to each point. The welding program generation module then combines the multi-axis motion interpolation points with the parameter records corresponding to the process parameter index codes to form the welding program, and simultaneously organizes the reference trajectory in the same order to form the theoretical trajectory. The welding program and theoretical trajectory are sent as output objects to the welding operation and error acquisition module as welding operation input and weld error data acquisition comparison input.

[0225] The welding operation and error acquisition module 07, connected to the welding program generation module, is used to acquire the welding program and theoretical trajectory, perform welding operations and weld error data acquisition and processing by the automatic welding module, and obtain weld error data. Specifically, the welding operation and error acquisition module receives the welding program and theoretical trajectory output by the welding program generation module, drives the automatic welding module to execute the welding operation according to the robot motion instructions, welding instructions, and arc tracking instructions in the welding program, and simultaneously reads the contour information acquired by the vision guidance module during the welding process. The welding operation and error acquisition module compares the currently acquired contour with the corresponding position in the theoretical trajectory to form a record of the current position deviation, direction deviation, and weld boundary deviation, and organizes this record into weld error data. The weld error data is sent as an output object to the online correction and attribute update module as an input object for position feedback, feedforward compensation, and subsequent interpolation point processing of online correction, while retaining the operation record corresponding to the current segment, current layer, and current pass.

[0226] The online correction and attribute update module 08 is used to perform position feedback, feedforward compensation, and online correction of subsequent interpolation points based on the weld error data to obtain subsequent interpolation points. Based on these subsequent interpolation points, it performs a second scan of the weld, updates the weld cross-sectional geometric feature parameters, and updates the weld attribute set to obtain an updated weld attribute set. The updated weld attribute set is then output to the path planning and interpolation generation module. Specifically, the online correction and attribute update module receives the weld error data output by the welding operation and error acquisition module, and calls the weld cross-sectional geometric feature parameters retained by the cross-sectional parameter extraction module and the weld attribute set retained by the attribute binding module. It first performs position feedback, feedforward compensation, and online correction of subsequent interpolation points according to the current segment, current layer, and current pass to obtain subsequent interpolation points. When the conditions for re-scanning the weld are met, the online correction and attribute update module calls the panoramic scanning module or the visual guidance module to acquire new weld area point clouds and red-green-blue images. It then performs updates to the weld cross-sectional geometric feature parameters and weld attribute set for the current segment, forming an updated weld attribute set. This updated weld attribute set is output to the path planning and interpolation generation module as input for re-executing the path planning algorithm, multi-layer multi-pass decomposition, and multi-axis motion interpolation point generation. Simultaneously, it completes the corresponding registration of the current segment update record and the previous operation record.

Claims

1. A method for teaching-free beveling of thick-walled pipes, characterized in that, include: S100: Acquire the global vision unit, industrial camera, laser line structured light, and assembled thick-walled pipe bevel area; perform partitioned scanning, point cloud registration processing, and model-point cloud registration processing to obtain the actual workpiece pose and weld geometric centerline vector; the global vision unit includes an industrial camera, laser line structured light, mounting bracket, control interface, and acquisition position separately set from the automatic welding module; the industrial camera is used to receive reflected light information from the surface of the thick-walled pipe bevel area; the laser line structured light is used to form a continuous light band contour on the surface of the thick-walled pipe bevel area; the assembled thick-walled pipe bevel area includes the bevel opening to be welded, the base material surfaces on both sides of the bevel, and the positioning reference area adjacent to the bevel; S200. Based on the actual pose of the workpiece and the geometric centerline vector of the weld, perform geometric feature parameter extraction of the weld cross section, groove form code matching and preset welding layer information binding processing to obtain a set of weld attributes. S300. Based on the weld attribute set, perform path planning algorithm, multi-layer multi-pass decomposition and welding program generation processing to obtain welding program and theoretical trajectory; S400. Based on the welding program and theoretical trajectory, perform welding operations using the automatic welding module, correct subsequent interpolation points online, and rescan the weld seam to obtain an updated set of weld seam attributes.

2. The method according to claim 1, characterized in that, The partition scan process includes: The bevel region of the thick-walled pipe is divided into several adjacent scanning sections according to the length of the bevel region and the field of view. Three-dimensional contour and RGB images are acquired for each scanning section, and then processed by a stitching algorithm to form a continuous point cloud of the weld region.

3. The method according to claim 2, characterized in that, The point cloud registration process and the model-point cloud registration process include: Downsampling is performed on the point cloud of the weld area to unify the spacing, filtering is performed to remove discrete noise points, and edge extraction based on RGB image is performed to determine the bevel boundary; The transformation matrix is ​​obtained and the model is registered with the point cloud to obtain the actual pose of the workpiece and the geometric centerline vector of the weld continuously extracted along the bevel opening area.

4. The method according to claim 3, characterized in that, The process of extracting and processing the geometric feature parameters of the weld cross-section includes: Multiple local sections are cut along the geometric centerline vector of the weld at preset intervals; Extract the root gap width, bevel angle, blunt edge thickness, and base material thickness for each section; The segments are then merged according to the similarity of their parameters to obtain one or more parameter segments.

5. The method according to claim 1, characterized in that, The process of matching bevel form codes and binding preset welding layer information includes: The bevel form code matching process includes: matching the four types of parameters of each parameter segment with the process database to generate the bevel form code and process parameter index code for each segment; The preset welding layer information binding process includes: binding the groove form code, process parameter index code, preset welding layer information and weld geometric centerline vector of the same parameter segment at the segment level to form a layer record chain as a weld attribute set; wherein, when a certain segment does not match the preset welding layer information, the information of the adjacent segment is copied to form a temporary binding or marked as a segment to be updated.

6. The method according to claim 5, characterized in that, The path planning algorithm process includes: According to the segment identifier order in the weld attribute set, the starting coordinates, ending coordinates, weld geometric center line vector, groove form code, process parameter index code and preset welding layer information of each segment are read. Then, four-dimensional coupled calculation of layer thickness, swing width, speed and current is performed in each segment to generate an automatic welding trajectory that changes with the segment.

7. The method according to claim 6, characterized in that, The process of multi-layer, multi-pass decomposition and welding procedure generation includes: The multi-layer, multi-pass decomposition process includes: decomposing the automatic welding trajectory into independent weld track trajectories according to segments, layers, and passes, and generating corresponding multi-axis motion interpolation points; The welding program generation process includes: binding the multi-axis motion interpolation points with the welding current, arc voltage, welding speed and oscillation amplitude corresponding to the process parameter index codes point by point to generate the welding program, and simultaneously generating the theoretical trajectory corresponding to each layer and each pass.

8. The method according to claim 7, characterized in that, The welding process of the automatic welding module includes: Based on the welding procedure, the vision guidance module is invoked to synchronously collect weld error data, forming a continuous record of the current position deviation, direction deviation, and weld boundary deviation.

9. The method according to claim 8, characterized in that, The process of online correction of subsequent interpolation points and rescanning of the weld includes: The online correction of subsequent interpolation points includes: performing position feedback and feedforward compensation based on weld error data to generate corrected subsequent interpolation points; The rescanning of the weld seam process includes: when the current layer welding is completed, the current pass welding is completed, the marker for the section to be updated exists, or the continuous abnormal acquisition marker reaches a preset number of trigger conditions, the global vision unit or vision guidance module is invoked to rescan the current section, update the geometric feature parameters of the weld seam cross section of the current section, regenerate the bevel form code and process parameter index code, and rebind them with the preset welding layer information to obtain the updated weld seam attribute set, and input the updated weld seam attribute set back to the path planning algorithm for processing.

10. A teaching-free welding system for beveling thick-walled pipes, characterized in that, include: The system comprises a panoramic scanning module, a point cloud registration and coordinate transformation module, a cross-sectional parameter extraction module, an attribute binding module, a path planning and interpolation generation module, a welding program generation module, a welding operation and error acquisition module, and an online correction and attribute update module; these modules are connected in sequence to implement the method described in any one of claims 1-9.

Images