Vehicle body special-shaped space curve welding seam tracking method and system

Abstract

The invention discloses a vehicle body special-shaped space curve welding seam tracking method and system. The method comprises the steps that target workpiece image information is collected through a preset scanning track, a welding seam point position center line is marked, and then a welding seam data set is formed; enhancing the weld data set, and setting a reflective area as an inverted funnel-shaped highlight distribution structure for data enhancement of reflective characteristics to obtain an initial weld data set; constructing a welding seam point location recognition model to recognize the initial welding seam data set to obtain initial welding seam point locations, filtering the initial welding seam point locations to remove outliers, and generating a reference welding seam point location sequence by using a curve fitting algorithm; constructing a curve welding seam tracking model; during welding, real-time data of a welding seam are synchronously collected, the actual pose of the welding seam is recognized through a curve welding seam tracking model, and a control instruction is generated in combination with a reference welding seam point position sequence to track welding in real time. According to the method, high-precision real-time identification and tracking of the complex space curve welding seam can be realized on the premise that the production takt is not increased.

Description

Technical Field

[0001] This invention relates to the field of automated welding technology, and in particular to a method and system for tracking weld seams on irregularly shaped spatial curves of automotive bodies. Background Technology

[0002] In the automotive manufacturing industry, automated welding is a core component ensuring the quality and efficiency of vehicle body assembly. Currently, the mainstream welding method in the industry is still the teach-and-playback model, which involves manually teaching and recording the welding torch's movement trajectory, and then repeating this trajectory in mass production. While this method is suitable for standardized and highly repetitive production scenarios, it has significant limitations in practical applications: due to factors such as workpiece assembly errors, insufficient positioning accuracy of tooling fixtures, and thermal deformation of parts, the pre-stored teaching trajectory often cannot accurately match the actual weld position, easily leading to quality defects such as welding misalignment, poor weld formation, burn-through, or incomplete penetration. This results in most teach-and-play welding stations requiring additional manual re-welding, severely restricting the quality stability and production efficiency improvement of welding automation.

[0003] To address the aforementioned issues, introducing machine vision technology into the welding process to achieve real-time perception of the weld's true position and trajectory correction has become a core development direction for intelligent welding technology. By acquiring images of the weld area using laser area array cameras, line scan cameras, or structured light sensors, and combining image processing or deep learning algorithms to extract the weld boundary, gap morphology, and centerline position, trajectory deviations caused by tooling positioning errors, workpiece splicing deviations, etc., can be corrected in real time. This can even achieve complete reconstruction of the welding trajectory, significantly improving welding adaptability and robustness.

[0004] However, existing machine vision-based solutions still have many shortcomings in addressing the need for identification and tracking of irregularly shaped spatial curve welds on automobile bodies. These shortcomings can be mainly categorized into three types, each with its own limitations:

[0005] Line laser scanning followed by welding requires an additional pre-scanning process, which is difficult to meet the high-speed requirements of automobile production lines. Moreover, the weld seams of the car body are mostly lap joints, and the steel plate thickness is usually less than 1mm. The geometric features are not obvious, and line laser scanning is easily affected by reflection, obstruction and changes in surface morphology. The weld seam feature extraction is unstable and the misidentification rate is high.

[0006] Laser area array camera full-area scanning solution: Although it can cover the weld seam area at one time, the point cloud spatial resolution and measurement accuracy of the large field of view camera are difficult to balance, which cannot meet the high-precision positioning requirements; at the same time, the camera has poor attitude flexibility and is prone to point cloud loss or distortion due to reflection and occlusion under complex vehicle body structure, which affects the integrity and stability of weld seam recognition.

[0007] Line scan camera scanning while welding solution: Although it does not require additional production cycle time and is suitable for high cycle time requirements, existing solutions are mostly designed for long straight weld seams or regular weld seams, and rely on fixed geometric models or local feature assumptions. When facing irregular spatial curve weld seams on the vehicle body (large curvature changes and frequent direction switching), problems such as recognition interruption and tracking drift are likely to occur, making it difficult to meet the requirements of high-precision tracking.

[0008] In summary, current technologies still cannot meet the welding requirements of irregular spatial curve welds on automobile bodies in terms of cycle time adaptability, adaptability to complex working conditions, and recognition robustness. Summary of the Invention

[0009] In view of the above-mentioned shortcomings in the current field of automated welding technology, the present invention provides a method and system for tracking irregular spatial curve welds in vehicle bodies, which can achieve high-precision and robust real-time identification and tracking of complex spatial curve welds without increasing the production cycle.

[0010] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0011] A method for tracking weld seams on irregularly shaped spatial curves of a vehicle body, the method comprising the following steps:

[0012] After acquiring image information of the target workpiece by pre-set scanning trajectory and marking the center line of the weld seam, a weld seam dataset is formed.

[0013] The weld dataset is enhanced by random brightness, random contrast, and rotation, and the reflective properties are enhanced by setting the reflective area as an inverted funnel-shaped high-brightness distribution structure to obtain the initial weld dataset.

[0014] A weld seam location identification model is constructed to identify the initial weld seam dataset to obtain the initial weld seam locations. The initial weld seam locations are then filtered to remove outliers, and a curve fitting algorithm is used to generate a reference weld seam location sequence.

[0015] A training dataset was constructed based on welding interference scenario samples, and a curve weld seam tracking model was obtained by training a generative adversarial network (GAN).

[0016] During the welding process, real-time weld data is collected simultaneously. Based on the real-time weld data, the actual position and orientation of the weld are identified through a curve weld tracking model. Control commands are then generated in conjunction with the reference weld point sequence to track the welding in real time.

[0017] According to one aspect of the present invention, the data enhancement of reflective properties by setting the reflective area as an inverted funnel-shaped high-brightness distribution structure includes: when performing data enhancement of reflective properties, setting the reflective area as an inverted funnel-shaped high-brightness distribution structure, the center of which passes through the weld point; wherein the principal axis direction of the inverted funnel-shaped high-brightness area forms a certain angle with the laser centerline, the angle being randomly selected within a preset angle range to simulate the uncertainty of the reflective direction caused by changes in workpiece posture, surface normal, and illumination conditions during actual welding.

[0018] According to one aspect of the present invention, the specific steps of data enhancement of reflective properties by setting the reflective area into an inverted funnel-shaped high-brightness distribution structure include:

[0019] By fitting a straight line to the laser centerline and rotating it, a linear transformation can be performed to obtain the axis of enhanced reflection.

[0020] Add reflectivity enhancement using the following formula:

[0021]

[0022] in, This represents the maximum brightness value of the reflective area. and For the point on the center line of the reflection, and Control the size and intensity of the reflective area; the values ​​should be set according to the scene.

[0023] According to one aspect of the present invention, the step of filtering out outliers from the initial weld points and generating a reference weld point sequence using a curve fitting algorithm includes: pooling all weld points and performing radius filtering to remove outliers; obtaining a smooth curve trajectory through third-order B-spline fitting, with a smoothing factor of 0.5-1; and sampling at a preset sampling interval to obtain the final reference weld point sequence, forming a reference welding trajectory.

[0024] According to one aspect of the present invention, the method for tracking weld seams on irregularly shaped spatial curves of a vehicle body further includes: establishing a weld seam coordinate system, specifically including:

[0025] The direction of the normal vector of the fitted workpiece plane is defined as the Z-axis;

[0026] The direction from the current weld point to the next weld point is defined as the X-axis;

[0027] The Y-axis is obtained by cross product of the Z-axis and the X-axis.

[0028] According to one aspect of the invention, defining the normal direction of the fitted workpiece plane as the Z-axis includes:

[0029] Centerline point cloud is obtained by scanning using a weld seam location recognition model;

[0030] The ROI region is divided in the point cloud based on the obtained reference weld point sequence;

[0031] Within the ROI region, use RANSAC to extract the workpiece plane and obtain the workpiece plane normal vector. Set the filtering conditions so that the direction of the normal vector is the Z-axis.

[0032] According to one aspect of the present invention, the construction of the weld point identification model to identify the initial weld points from the initial weld dataset includes: using UET to identify weld points and weld centerlines, converting the weld points and laser centerlines into heatmap trend diagrams, and enabling the model to learn their features.

[0033] According to one aspect of the present invention, the curve weld tracking model is obtained by training an adversarial generative network (GAN); wherein the generator is used to simulate welding arc light and spatter interference, and the discriminator adopts a U-Net structure to identify the weld position under interference conditions.

[0034] According to one aspect of the present invention, the step of synchronously acquiring real-time weld data during the welding process, identifying the actual position and orientation of the weld through a curve weld tracking model based on the real-time weld data, and generating control commands to track the welding in real time in conjunction with the reference weld point sequence, specifically includes:

[0035] Set up a sliding window queue to cache the pre-identified weld points;

[0036] The weld seam points within the window are smoothed.

[0037] Find the reference point that is closest to the smoothed weld point from the reference weld point sequence, obtain its corresponding reference welding posture and smooth it, and generate control commands.

[0038] A vehicle body irregular spatial curve weld seam tracking system, implemented based on the above-mentioned vehicle body irregular spatial curve weld seam tracking method, includes:

[0039] The scanning and acquisition module collects image information of the target workpiece through a preset scanning trajectory and marks the center line of the weld seam to form a weld seam dataset.

[0040] The data augmentation module enhances the weld dataset by random brightness, random contrast, and rotation, and enhances the reflective properties by setting the reflective area as an inverted funnel-shaped high-brightness distribution structure to obtain the initial weld dataset.

[0041] The identification module constructs a weld seam location identification model to identify the initial weld seam dataset to obtain the initial weld seam locations, filters the initial weld seam locations to remove outliers, and uses a curve fitting algorithm to generate a reference weld seam location sequence.

[0042] The model building module constructs a training dataset based on welding interference scenario samples and uses a generative adversarial network (GAN) to train a curve weld seam tracking model.

[0043] The tracking module synchronously collects real-time weld data during the welding process. Based on the real-time weld data, it identifies the actual position and orientation of the weld through a curve weld tracking model, and generates control commands to track the welding in real time by combining the reference weld point sequence.

[0044] The advantages of this invention are as follows: Only a single scan is performed when acquiring weld reference information, resulting in no additional cycle time loss. Simultaneously, the position and attitude sequence of the reference weld is constructed, providing a precise benchmark for online weld tracking. Combined with a simultaneous scanning and welding operation mode, the weld reference information is dynamically fused with real-time pose, enabling real-time correction of positional offsets caused by assembly errors and workpiece deformation, achieving millimeter-level precise tracking of spatial curve welds. Furthermore, by simulating reflective noise to train the model and resisting reflective interference from thin steel plates, and relying on a GAN model to combat the impact of arc spatter, the weld misidentification rate is significantly reduced, effectively improving the system's robustness under complex working conditions. Ultimately, this achieves stable control over the welding quality of irregularly shaped welds on the vehicle body, precisely adapting to the high-cycle production demands of the automotive industry. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a schematic flowchart of a method for tracking weld seams on irregularly shaped spatial curves of a vehicle body, as described in this invention.

[0047] Figure 2 This is a schematic diagram of the structure of a vehicle body irregular spatial curve weld seam tracking system according to the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1

[0050] like Figure 1 As shown, a method for tracking weld seams on irregularly shaped spatial curves of a vehicle body includes the following steps:

[0051] Step S1: Collect image information of the target workpiece through a preset scanning trajectory and mark the center line of the weld seam to form a weld seam dataset;

[0052] Step S1 involves acquiring image information of the target workpiece through a preset scanning trajectory and marking the center line of the weld seam to form a weld seam dataset. This is the offline pre-scanning stage. First, the workpiece is pre-positioned, that is, the weld seam is scanned through the preset scanning trajectory. The purpose of pre-positioning is to obtain the approximate location of the weld seam by scanning the entire weld seam and to obtain the point cloud information of the workpiece.

[0053] Specifically, it includes:

[0054] Step S11: Pre-scanning of the workpiece. First, set the scanning trajectory according to the position of the weld on the workpiece. When setting the scanning trajectory, the posture can be adjusted separately to reduce or eliminate the effect of reflection.

[0055] Step S12: Acquire workpiece image information using a preset scanning trajectory, manually annotate the data, and establish an initial database. The annotation information includes the weld point location and the laser centerline. This forms an initial database covering different working conditions, ensuring the accuracy and completeness of the annotation information.

[0056] The weld area of ​​the target workpiece is pre-scanned using a preset scanning trajectory to obtain image sequences and point cloud data.

[0057] The scanning is achieved through a line scan camera. During the acquisition process, the frame rate of the image sequence is matched with the sampling frequency of the point cloud data to ensure that the two-dimensional visual data and the three-dimensional spatial data of the same weld position correspond one-to-one.

[0058] This pre-scanning process is performed only once during the entire production cycle, without adding any extra cycle time cost to mass production. The following steps are required before scanning the target workpiece:

[0059] 1) First, based on the 3D model of the target workpiece, the weld seam design path, and the structural characteristics of the irregular spatial curves of the vehicle body, a suitable scanning trajectory is preset. The scanning trajectory must completely cover the entire weld seam area to ensure no scanning blind spots.

[0060] 2) Considering the characteristics of the steel plate thickness of the vehicle body weld seam being less than 1mm, the indistinct geometric features, and the tendency to generate reflections, the posture of the scanning equipment (line scan camera) was adjusted separately. By adjusting parameters such as the angle and distance between the camera and the weld seam area, the interference of metal surface reflections on the scanning data was reduced or eliminated.

[0061] Step S2: Enhance the weld dataset by random brightness, random contrast and rotation, and enhance the reflective properties by setting the reflective area as an inverted funnel-shaped high-brightness distribution structure to obtain the initial weld dataset;

[0062] Dataset augmentation:

[0063] The dataset is expanded by randomizing image brightness, randomly changing contrast, and randomly rotating images to improve the model's adaptability to changes in ambient lighting and image angle deviations.

[0064] To enhance reflective noise, when performing data enhancement on reflective characteristics, the reflective area is set as an inverted funnel-shaped high-brightness distribution structure, with its center passing through the weld point; the main axis of the inverted funnel-shaped high-brightness area forms a certain angle with the laser centerline, and the angle is randomly selected within a preset angle range to simulate the uncertainty of reflective direction caused by changes in workpiece posture, surface normal, and illumination conditions during actual welding.

[0065] To address the issue of reflective interference from vehicle body welds, an enhancement method simulating actual reflective characteristics is employed: In the training image, an inverted funnel-shaped bright area is added along a straight line passing through the weld point. The brightness of this bright area gradually decreases from the center to the edge, and its principal axis forms a random angle within a preset range with the laser centerline, simulating the uncertainty in reflective direction caused by changes in workpiece posture, surface normal, and lighting conditions. The reflective intensity and range are controlled by the following formula:

[0066]

[0067] in, This represents the maximum brightness value of the reflective area. , Points on the center line of the reflective surface; and Control the size and intensity of the reflective area. The value range is 2-15, which can be flexibly selected for different scenes. When the reflection is weak or the area is narrow, use 2-5 to enhance local contrast; when the reflection is strong or the area is wide, use 10-15 to make the reflective area smoother and more continuous.

[0068] By randomly setting the center position, orientation angle, and brightness distribution of the reflective area, the generated reflective noise has diversity in shape, direction, and intensity, thereby effectively expanding the distribution range of training samples and improving the adaptability of deep learning models to different reflective conditions.

[0069] Step S3: Construct a weld seam location identification model to identify the initial weld seam dataset and obtain the initial weld seam locations. Filter the initial weld seam locations to remove outliers and use a curve fitting algorithm to generate a reference weld seam location sequence.

[0070] Model training: A weld seam location recognition model is constructed using a U-Net structure. The labeled weld seam locations and the laser centerline are converted into a heatmap trend map as the model input. Through feature learning, the model learns the visual feature patterns of the weld seam locations. After training, the image sequence obtained from the pre-scan is input into the model, and the model outputs the coordinates of the weld seam locations corresponding to each image, which are then integrated to form an initial set of weld seam locations.

[0071] The initial weld points are subjected to radius filtering to remove outliers. The filtering radius is preset according to the weld density to ensure that isolated interference points are eliminated. A third-order B-spline curve fitting algorithm is used to fit the screened weld points. The smoothing factor is set to 0.5-1. After fitting, sampling is performed at a sampling interval of 3-5mm to generate a continuous and smooth reference weld point sequence to avoid curve distortion.

[0072] For the reference weld point sequence, a weld coordinate system is established by fitting the workpiece plane based on its neighboring point cloud data, thus obtaining a reference welding posture sequence, i.e., a reference welding trajectory. Specifically, this includes:

[0073] 1) Point cloud preprocessing: The laser centerline obtained by scanning is converted into point cloud data through hand-eye calibration parameters and camera intrinsic parameters. The ROI (region of interest) with preset boundaries is delineated in the point cloud according to the reference weld point sequence.

[0074] 2) Workpiece plane fitting: The RANSAC algorithm is used to fit the workpiece plane within the ROI. The fitting threshold is set to 2mm and the number of fitting iterations is 1000 to ensure the plane fitting accuracy.

[0075] 3) Weld coordinate system establishment: The direction of the normal vector of the fitting plane is defined as the Z-axis. To prevent the normal vector from being reversed, a filtering condition is set to ensure that the normal vector is negative on the Z-axis in the camera coordinate system, that is, pointing in the direction of the camera. The direction from the current weld point to the next weld point is defined as the X-axis. The Y-axis is obtained by the cross product of the Z-axis and the X-axis, forming a single-point independent weld coordinate system.

[0076] 4) Attitude calculation and smoothing: Set a fixed welding angle in each weld coordinate system, automatically match the scanning angle and generate the initial welding attitude; smooth the initial welding attitude through quaternion interpolation to obtain a reference welding attitude sequence that corresponds one-to-one with the reference weld point sequence.

[0077] Step S4: Construct a training dataset based on welding interference scene samples, and use a Generative Adversarial Network (GAN) to train a curve weld tracking model;

[0078] The specific construction and training logic of the curve weld seam tracking model is as follows:

[0079] 1) Dataset construction: The weld seam images collected before arc initiation are used as the base samples, and the images containing arc light and spatter interference during the arc initiation stage are used as supplementary samples. The weld seam point locations and interference areas (arc light and spatter) are labeled separately.

[0080] 2) Data augmentation: Random brightness and contrast adjustments, rotation, and funnel-shaped reflection noise enhancement are applied to the dataset to cover various complex optical interference scenarios in the welding process;

[0081] 3) Adversarial training: The generator of GAN is used to simulate welding arc light, spatter and reflection interference. The discriminator adopts U-Net structure. Through adversarial training between the generator and the discriminator, the discriminator is able to accurately distinguish weld features from interference features in the interference environment and finally output the actual position and pose of the weld.

[0082] Step S5: During the welding process, real-time weld data is collected synchronously. Based on the real-time weld data, the actual position and orientation of the weld are identified through the curve weld tracking model. Control commands are generated in conjunction with the reference weld point sequence to track the welding in real time.

[0083] Step S5 is the online welding stage, which uses a line scan camera in a simultaneous scanning and welding mode to synchronously acquire real-time weld data (i.e., real-time image sequences) during the welding process. During acquisition, a real-time scanning path is planned based on step S1 to ensure that the scanning range always covers the current welding area; the known laser position of the weld sensor (i.e., the position of the laser point relative to the robot flange) is used to achieve spatial synchronization between scanning and welding actions without increasing the production cycle time.

[0084] The acquired real-time image sequence of the weld is input into the curve weld tracking model to identify the actual position and orientation of the weld.

[0085] Based on the reference weld point sequence, reference welding posture sequence, and the identified actual weld pose, control commands are generated to drive the robot to adjust the welding pose in real time, completing weld tracking. Specifically, this includes:

[0086] Step S51: Set up a sliding window queue to cache the pre-identified weld points;

[0087] Set up a sliding window queue to cache the actual weld locations identified in advance due to the advance distance between the welding torch and the sensor. The window size is dynamically adjusted according to the advance distance.

[0088] Step S52: Smooth the weld points within the window;

[0089] The actual weld locations within the window are smoothed and filtered to eliminate random errors from real-time data acquisition.

[0090] Step S53: Find the reference point that is closest to the smoothed weld point from the reference weld point sequence, obtain its corresponding reference welding posture and smooth it, and generate control commands.

[0091] The system finds the reference point that is closest to the actual point of the smoothed weld from the reference weld point sequence and extracts the reference welding posture corresponding to the reference point. The extracted reference welding posture is fused with the actual weld posture, and the posture is smoothed by quaternion interpolation to generate continuous and stable welding control commands. These commands are then sent to the robot actuator to adjust the welding position and posture in real time, thereby achieving high-precision tracking of irregular spatial curve welds on the vehicle body.

[0092] Advantages of implementing this embodiment:

[0093] 1. Improve weld seam identification accuracy;

[0094] By utilizing funnel-shaped reflective enhancement, anti-arc light deep learning models, and point cloud information, high-precision weld seam location and attitude recognition can be achieved under complex lighting and welding interference conditions.

[0095] 2. Improve the real-time performance and stability of online welding;

[0096] By acquiring reference trajectories through offline pre-scanning, and performing real-time window smoothing and attitude smoothing, continuous welding torch movement is ensured to meet the high cycle time requirements of automotive production lines.

[0097] 3. Adaptable to complex spatial curve welds;

[0098] This invention is not only applicable to straight welds, but can also accurately track spatial curved welds and automatically adjust the welding posture to adapt to changes in the workpiece surface.

[0099] 4. Enhance the model's robustness to environmental disturbances;

[0100] By combining reflective noise enhancement and arc spatter countermeasure training, the deep learning model can still stably identify weld seams under different lighting, reflective and arc interference conditions.

[0101] 5. Highly feasible;

[0102] The offline database construction, data augmentation, deep learning model training, point cloud coordinate system establishment, and online real-time tracking form a complete closed loop, which can be directly deployed on industrial welding production lines.

[0103] Example 2

[0104] like Figure 2 As shown, a vehicle body irregular spatial curve weld seam tracking system is implemented based on the vehicle body irregular spatial curve weld seam tracking method described in Embodiment 1, including:

[0105] The scanning acquisition module M1 acquires image information of the target workpiece through a preset scanning trajectory and marks the center line of the weld seam point to form a weld seam dataset.

[0106] The data augmentation module M2 enhances the weld dataset by random brightness, random contrast, and rotation, and enhances the reflective properties by setting the reflective area as an inverted funnel-shaped high-brightness distribution structure to obtain the initial weld dataset.

[0107] The identification module M3 constructs a weld point identification model to identify the initial weld point in the initial weld dataset, filters the initial weld point to remove outliers, and uses a curve fitting algorithm to generate a reference weld point sequence.

[0108] Model building module M4 constructs a training dataset based on welding interference scene samples and uses a generative adversarial network (GAN) to train a curve weld tracking model.

[0109] The tracking module M5 synchronously collects real-time weld data during the welding process. Based on the real-time weld data, it identifies the actual position and orientation of the weld through a curve weld tracking model, and generates control commands to track the welding in real time by combining the reference weld point sequence.

[0110] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for tracking weld seams on irregularly shaped spatial curves of a vehicle body, characterized in that, The method includes the following steps: After acquiring image information of the target workpiece by pre-set scanning trajectory and marking the center line of the weld seam, a weld seam dataset is formed. The weld dataset is enhanced by random brightness, random contrast, and rotation, and the reflective properties are enhanced by setting the reflective area as an inverted funnel-shaped high-brightness distribution structure to obtain the initial weld dataset. A weld seam location identification model is constructed to identify the initial weld seam dataset to obtain the initial weld seam locations. The initial weld seam locations are then filtered to remove outliers, and a curve fitting algorithm is used to generate a reference weld seam location sequence. A training dataset was constructed based on welding interference scenario samples, and a curve weld seam tracking model was obtained by training a generative adversarial network (GAN). During the welding process, real-time weld data is collected simultaneously. Based on the real-time weld data, the actual position and orientation of the weld are identified through a curve weld tracking model. Control commands are then generated in conjunction with the reference weld point sequence to track the welding in real time.

2. The method for tracking weld seams on irregularly shaped spatial curves of a vehicle body according to claim 1, characterized in that, The method of enhancing reflective properties by setting the reflective area as an inverted funnel-shaped high-brightness distribution structure includes: when enhancing reflective properties, the reflective area is set as an inverted funnel-shaped high-brightness distribution structure, with its center passing through the weld point; wherein the main axis of the inverted funnel-shaped high-brightness area forms a certain angle with the laser centerline, and the angle is randomly selected within a preset angle range to simulate the uncertainty of the reflective direction caused by changes in workpiece posture, surface normal, and illumination conditions during the actual welding process.

3. The method for tracking weld seams on irregularly shaped spatial curves of a vehicle body according to claim 2, characterized in that, The specific steps for data enhancement of reflective properties by setting the reflective area as an inverted funnel-shaped high-brightness distribution structure include: By fitting a straight line to the laser centerline and rotating it, a linear transformation can be performed to obtain the axis of enhanced reflection. Add reflectivity enhancement using the following formula: in, This represents the maximum brightness value of the reflective area. and For the point on the center line of the reflection, and Control the size and intensity of the reflective area; the values ​​should be set according to the scene.

4. The method for tracking irregular spatial curve welds on a vehicle body according to claim 1, characterized in that, The step of filtering out outliers from the initial weld points and generating a reference weld point sequence using a curve fitting algorithm includes: pooling all weld points and performing radius filtering to remove outliers; obtaining a smooth curve trajectory through third-order B-spline fitting with a smoothing factor of 0.5-1; and sampling at a preset sampling interval to obtain the final reference weld point sequence, forming a reference welding trajectory.

5. The method for tracking weld seams on irregularly shaped spatial curves of a vehicle body according to claim 1, characterized in that, The method for tracking weld seams on irregularly shaped spatial curves of a vehicle body also includes: establishing a weld seam coordinate system, specifically including: The direction of the normal vector of the fitted workpiece plane is defined as the Z-axis; The direction from the current weld point to the next weld point is defined as the X-axis; The Y-axis is obtained by cross product of the Z-axis and the X-axis.

6. The method for tracking weld seams on irregularly shaped spatial curves of a vehicle body according to claim 5, characterized in that, The step of defining the normal vector direction of the fitted workpiece plane as the Z-axis includes: Centerline point cloud is obtained by scanning using a weld seam location recognition model; The ROI region is divided in the point cloud based on the obtained reference weld point sequence; Within the ROI region, use RANSAC to extract the workpiece plane and obtain the workpiece plane normal vector. Set the filtering conditions so that the direction of the normal vector is the Z-axis.

7. The method for tracking weld seams on irregularly shaped spatial curves of a vehicle body according to claim 1, characterized in that, The construction of the weld seam location recognition model to identify the initial weld seam locations from the initial weld seam dataset includes: using UET to identify weld seam points and weld centerlines, converting weld seam points and laser centerlines into heatmap trend images, and enabling the model to learn their features.

8. The method for tracking weld seams on irregularly shaped spatial curves of a vehicle body according to claim 1, characterized in that, The curve weld tracking model trained using Generative Adversarial Network (GAN) is specifically as follows: the generator is used to simulate welding arc light and spatter interference, and the discriminator adopts a U-Net structure to identify the weld position under interference conditions.

9. The method for tracking weld seams on irregularly shaped spatial curves of a vehicle body according to claim 1, characterized in that, The process of synchronously acquiring real-time weld data during welding, identifying the actual weld position using a curve weld tracking model based on the real-time weld data, and generating control commands to track the welding in real time in conjunction with the reference weld point sequence, specifically includes: Set up a sliding window queue to cache the pre-identified weld points; The weld seam points within the window are smoothed. Find the reference point that is closest to the smoothed weld point from the reference weld point sequence, obtain its corresponding reference welding posture and smooth it, and generate control commands.

10. A tracking system for irregularly shaped spatial curve welds on a vehicle body, characterized in that, The vehicle body irregular spatial curve weld seam tracking system is implemented based on the vehicle body irregular spatial curve weld seam tracking method according to any one of claims 1 to 9, including: The scanning and acquisition module collects image information of the target workpiece through a preset scanning trajectory and marks the center line of the weld seam to form a weld seam dataset. The data enhancement module enhances the weld dataset by random brightness, random contrast, and rotation, and enhances the reflective properties by setting the reflective area as an inverted funnel-shaped high-brightness distribution structure to obtain the initial weld dataset. The identification module constructs a weld seam location identification model to identify the initial weld seam dataset to obtain the initial weld seam locations, filters the initial weld seam locations to remove outliers, and uses a curve fitting algorithm to generate a reference weld seam location sequence. The model building module constructs a training dataset based on welding interference scenario samples and uses a generative adversarial network (GAN) to train a curve weld seam tracking model. The tracking module synchronously collects real-time weld data during the welding process. Based on the real-time weld data, it identifies the actual position and orientation of the weld through a curve weld tracking model, and generates control commands to track the welding in real time by combining the reference weld point sequence.

Images