Weld pattern recognition data processing method and system

Abstract

The invention relates to the technical field of weld pattern recognition, and discloses a weld pattern recognition data processing method and system, and the method comprises the steps: obtaining pre-scanning data of a weld region before main scanning, analyzing the surface interference type and interference degree, and dynamically generating a main scanning strategy according to the actual interference condition. The scanning parameters of the line laser contourgraph and / or the scanning path of the robot are / is adjusted. According to the method, the limitation that in the prior art, due to diversified workpiece surface conditions, a fixed data processing algorithm fails is overcome, the quality and efficiency of automatic welding are remarkably improved, and the welding defect caused by misrecognition or key geometric feature loss is avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of weld pattern recognition, and particularly relates to a weld pattern recognition data processing method and system. BACKGROUND

[0002] In modern industrial production, automated welding technology has become a key to improving efficiency and product quality. In order to ensure the quality of welding, a weld pattern recognition system is usually used to monitor and evaluate the weld in real time. Such a system collects data in the weld area and processes and analyzes it to identify the geometry of the weld and whether there are defects. However, in the actual production environment, the surface of the workpiece is often complex and variable, which brings great challenges to weld pattern recognition.

[0003] On an automated welding production line for large structural parts, the workpiece often goes through a transfer and storage link before entering the welding station, which may form an uneven oxide layer, oil stains, cutting fluid residues or scratches on the surface of the base material in the area to be welded. When a line laser profiler scans such a surface, the interaction between the laser beam and these interferences will change the optical reflection characteristics of the material, resulting in a decrease in the intensity of the reflected signal received by the sensor, scattering, or false reflections. In the original point cloud data, this is manifested as missing data points, a decrease in point cloud intensity values, irregular noise points on the contour line, or even large areas of data voids or high-intensity false points, making the real weld edge unclear.

[0004] Further, due to production batches, supplier differences or different processing methods in the previous process, the surface conditions of the workpieces entering the welding station are significantly inconsistent. For example, heavy oxide scales may completely block laser reflection, forming data voids; oil stains may cause specular reflection, producing false high points. This diverse surface condition makes it difficult for any pre-set, fixed-parameter data processing algorithm to effectively cope with it. Filter parameters set for light rust may not effectively remove the interference caused by heavy oxide scales; while aggressive parameters set to remove heavy oxide scales may incorrectly filter out real, small weld features (such as small misaligned edges) as spatter when welding other material workpieces. This problem of spatter characteristics changing due to changes in workpiece material, causing fixed data processing methods to fail, greatly reduces the accuracy and reliability of weld pattern recognition, ultimately causing the welding robot to fail to obtain correct guidance information, resulting in fluctuations in welding quality and even batch welding defects. SUMMARY

[0005] The application provides a welding seam mode recognition data processing method and system, aiming to solve the problems of reduced welding seam mode recognition accuracy and reliability and key geometric features being easily covered by interference information due to complex and variable workpiece surface conditions in existing automatic welding technology.

[0006] In a first aspect, to solve the above technical problems, the application provides a welding seam mode recognition data processing method, comprising: acquiring pre-scanning data of a welding seam area and analyzing the pre-scanning data to determine the surface interference type and surface interference degree of the welding seam area; generating a main scanning strategy according to the surface interference type and surface interference degree, the main scanning strategy including adjusting the scanning parameters of a line laser profiler and / or the scanning path of a robot; performing main scanning according to the main scanning strategy to acquire welding seam contour data; processing the welding seam contour data to recognize the welding seam mode and extract the welding seam geometric dimensions.

[0007] Preferably, the generating of the main scanning strategy according to the surface interference type and surface interference degree, the main scanning strategy including the adjustment of the scanning parameters of the line laser profiler and / or the scanning path of the robot, comprises: when the surface interference type and the surface interference degree indicate that there is composite optical feature interference in the welding seam micro area, triggering a multiple scanning strategy; performing a first sub-scanning according to the multiple scanning strategy to acquire a first group of welding seam contour point cloud data; performing a second sub-scanning according to the multiple scanning strategy to acquire a second group of welding seam contour point cloud data; fusing the first group of welding seam contour point cloud data and the second group of welding seam contour point cloud data to construct the welding seam contour data.

[0008] Preferably, the processing of the welding seam contour data to recognize the welding seam mode and extract the welding seam geometric dimensions comprises: initially segmenting the welding seam contour data to separate the preliminary contour of the current welding seam area from the adjacent weld; performing local geometric feature analysis on the separated current welding seam area to identify and mark potential feature points; performing feature analysis on the preliminary contour of the adjacent weld to extract residual geometric boundary information; judging the spatial relationship according to the potential feature points and the residual geometric boundary information to identify local overlap; when the local overlap exists, adjusting the potential feature points to determine the true geometric boundary of the current welding seam area; According to the real geometric boundary, a weld seam mode is identified and the weld seam geometric size is extracted.

[0009] Preferably, local geometric feature analysis is performed on the separated current weld seam area, potential feature points are identified and marked, including: Multi-scale local curvature calculation is performed on the point cloud data of the separated current weld seam area, and multi-scale curvature results are obtained; According to the multi-scale curvature results, the discrete potential feature points are preliminarily fused; Intensity fluctuation analysis is performed on the discrete potential feature points, and the feature points affected by microscopic surface unevenness are marked; The marked feature points are subjected to continuity evaluation, and continuous feature segments are clustered; According to the clustering results, the continuous feature segments are fitted, and the potential feature points are identified, including the groove edge and root gap feature points.

[0010] Preferably, feature analysis is performed on the preliminary contour of the adjacent weld, and residual geometric boundary information is extracted, including: Multi-scale geometric feature extraction is performed on the preliminary contour to obtain geometric change information of different scales; According to the geometric change information of different scales, local connectivity analysis is performed to identify and cluster continuous geometric feature segments; According to the continuous geometric feature segments, morphological analysis is performed to distinguish abnormal morphologies caused by welding process fluctuations, uneven cooling or spatter; According to the geometric feature segments of the abnormal morphologies, local reconstruction is performed to complete or smooth irregular, discontinuous or overlapping areas; Based on the results of local reconstruction, the residual geometric boundary information is extracted.

[0011] Preferably, according to the spatial relationship between the potential feature points and the residual geometric boundary information, local overlap is identified, including: A local feature coordinate system of the current weld seam area is constructed; A local boundary coordinate system of the residual geometric boundary information is constructed; The transformation relationship between the local feature coordinate system and the local boundary coordinate system is calculated; The residual geometric boundary information is transformed into the local feature coordinate system; The local curvature variation trend of the current weld seam area and the local curvature variation trend of the residual geometric boundary information are analyzed, and local deformation information caused by workpiece posture deviation is identified and quantified; According to the local deformation information, the position of the potential feature points is corrected; calculating a spatial distance between the potential feature point and the residual geometric boundary information to identify the local overlap.

[0012] Preferably, when the local overlap exists, adjusting the potential feature point to determine the real geometric boundary of the current weld region comprises: performing local deformation field analysis on the potential feature point of the current weld region and the residual geometric boundary information to identify a local deformation type and a local deformation degree; constructing a local deformation compensation function according to the local deformation type and the local deformation degree; performing initial position correction on the potential feature point according to the local deformation compensation function; dynamically adjusting a search region around the corrected potential feature point according to a local geometric shape of the residual geometric boundary information; iteratively optimizing the position of the potential feature point based on a matching degree and a spatial distance of local geometric features in the adjusted search region to distinguish the potential feature point from the residual geometric boundary information and determine the real geometric boundary of the current weld region.

[0013] Preferably, the step of performing local deformation field analysis on the potential feature point of the current weld region and the residual geometric boundary information to identify a local deformation type and a local deformation degree comprises: performing geometric feature extraction on the potential feature point of the current weld region and the residual geometric boundary information to obtain a local curvature and a normal vector; performing weighted processing on the local curvature and the normal vector according to material properties, thickness differences or welding heat input information corresponding to the potential feature point of the current weld region and the residual geometric boundary information to obtain weighted geometric features; calculating a local deformation gradient according to the weighted geometric features to obtain a deformation gradient field; identifying a local deformation type according to the deformation gradient field; quantifying a local deformation degree according to an amplitude and a distribution of the deformation gradient field.

[0014] Preferably, the step of constructing a local deformation compensation function according to the local deformation type and the local deformation degree comprises: when a coupling effect of multiple local deformation types is identified, constructing a deformation component compensation function according to each local deformation type and its local deformation degree; combining the deformation component compensation functions by spatial superposition to form the local deformation compensation function.

[0015] In a second aspect, the present application provides a weld seam pattern recognition data processing system, comprising: a detection end configured to acquire pre-scanning data of a weld seam area and analyze the pre-scanning data to determine a surface interference type and a surface interference degree of the weld seam area; an adjustment end configured to generate a main scanning strategy according to the surface interference type and the surface interference degree, the main scanning strategy comprising adjusting scanning parameters of a line laser profiler and / or a scanning path of a robot; a processing end configured to perform main scanning according to the main scanning strategy to acquire weld seam profile data, and process the weld seam profile data to recognize a weld seam pattern and extract weld seam geometric dimensions.

[0016] The weld seam pattern recognition data processing method and system disclosed in the present application can dynamically generate a main scanning strategy according to actual interference conditions by acquiring pre-scanning data of a weld seam area before main scanning and analyzing the surface interference type and the interference degree, the main scanning strategy comprising adjusting scanning parameters of a line laser profiler and / or a scanning path of a robot. Such an adaptive scanning strategy can effectively cope with complex and variable interferences such as uneven oxide layers, oil stains and scratches on a workpiece surface, avoiding problems such as missing data points, reduced intensity values, noise points or data voids under strong interference in traditional fixed parameter scanning, thereby acquiring high-quality weld seam profile data. Subsequently, the acquired weld seam profile data is finely processed to recognize a weld seam pattern and extract geometric dimensions, ensuring the accuracy and reliability of weld seam pattern recognition under complex surface conditions. The method overcomes the limitations of fixed data processing algorithms due to the diversification of workpiece surface conditions in the prior art, significantly improves the quality and efficiency of automatic welding, and avoids welding defects caused by misrecognition or loss of key geometric features. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a weld seam pattern recognition data processing method flowchart provided by an embodiment of the present application; Figure 2 is another weld seam pattern recognition data processing method flowchart provided by an embodiment of the present application; Figure 3 is another weld seam pattern recognition data processing method flowchart provided by an embodiment of the present application; Figure 4 is a weld seam pattern recognition data processing system structure diagram provided by an embodiment of the present application. DETAILED DESCRIPTION

[0018] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of the present application.

[0019] With reference to Figure 1 The method comprises the following steps: S1, obtaining pre-scanning data of a weld area, and analyzing the pre-scanning data to determine the surface interference type and the surface interference degree of the weld area; S2, generating a main scanning strategy according to the surface interference type and the surface interference degree, the main scanning strategy comprising adjusting the scanning parameters of a line laser profiler and / or the scanning path of a robot; S3, performing main scanning according to the main scanning strategy to obtain weld contour data; S4, processing the weld contour data to identify the weld mode and extract the weld geometric dimensions.

[0020] The present application effectively improves the accuracy and integrity of weld contour data collection under complex workpiece surface conditions by pre-evaluating surface interference and dynamically adjusting the scanning strategy, thereby improving the reliability of weld mode identification and ensuring the welding quality.

[0021] In order to better understand the weld mode identification data processing method proposed in the present application, the key terms and implementation environment involved therein will be described in detail as follows.

[0022] The "pre-scanning data of the weld area" refers to the original data about the surface condition of the weld area obtained by preliminary scanning before formal weld contour data collection. These data usually contain information such as surface reflection intensity and local geometric undulation, which are used to evaluate the surface interference condition. The "surface interference type" refers to specific interference factors that affect the accuracy of laser scanning, such as oxidation layer, oil stain, scratch, spatter, etc. The "surface interference degree" refers to the severity of the influence of the above interference factors on the quality of laser scanning data, such as slight, moderate or severe.

[0023] The "main scanning strategy" is a scheme dynamically generated according to the pre-scanning results, which is used to guide the subsequent accurate scanning. The core of the main scanning strategy is to adjust the "laser profilometer scanning parameters" (such as laser power, exposure time, scanning speed, etc.) and / or the "robot scanning path" (such as scanning angle, scanning distance, scanning speed, etc.) to optimize the data acquisition effect. The "weld contour data" is the accurate point cloud data obtained after the main scanning, which is used to describe the geometry of the weld. The "weld pattern recognition" refers to judging the type of the weld (such as V groove, U groove, fillet weld, etc.) and whether there is a defect (such as misalignment, incomplete penetration, burn-through, etc.) according to the weld contour data. The "weld geometry size" refers to the key size parameters of the weld, such as groove width, depth, root gap, etc.

[0024] The implementation environment of the present application is generally an automated welding production line, which includes a welding robot, a line laser profilometer, a data processing unit and a corresponding control system. The line laser profilometer is usually installed at the end of the robot and is used to non-contact acquisition of workpiece surface data. The data processing unit is responsible for receiving, analyzing and processing scanning data, and outputting recognition results and geometry size information to guide the welding robot to perform accurate welding.

[0025] The weld pattern recognition data processing method proposed in the present application is characterized by a series of refined steps, which effectively cope with complex and variable workpiece surface conditions, thereby realizing high-precision weld pattern recognition and geometry size extraction.

[0026] Firstly, the method includes obtaining pre-scanning data of the weld area, and analyzing the pre-scanning data to judge the surface interference type and the surface interference degree of the weld area. In actual operation, the acquisition and analysis of pre-scanning data can be realized in various ways. For example, a low-power laser beam can be used to quickly scan the weld area to obtain preliminary point cloud data and reflection intensity map. Subsequently, image processing algorithms are used to analyze the reflection intensity map to identify high reflection areas (which may indicate oil stains or specular reflection), low reflection areas (which may indicate oxidation layer or rough surface) and irregular texture (which may indicate scratches or splashes). As another implementation, multi-spectral imaging technology can be used to more accurately identify different types of surface interference by analyzing the reflection characteristics of different wavelengths of laser. For example, some wavelengths are sensitive to oil stains, while other wavelengths are sensitive to oxidation layers. By fusing and analyzing these multi-spectral data, the type of surface interference can be determined, such as oil stains, oxidation layers or scratches. At the same time, by quantifying the reflection intensity change, texture roughness or spectral response difference, the degree of surface interference can be evaluated, such as slight, moderate or severe.

[0027] Secondly, a main scanning strategy is generated based on the surface interference type and degree, which includes adjusting the scanning parameters of the line laser profiler and / or the scanning path of the robot. For example, when the pre-scan result shows that there is a moderate oxide layer interference in the weld area, a main scanning strategy can be generated, which includes increasing the laser power of the line laser profiler by 20% and extending the exposure time by 10% to ensure that the laser can penetrate the oxide layer and obtain sufficient reflection signals. In addition, the scanning path of the robot can also be adjusted, for example, the scanning angle is adjusted from perpendicular to the weld direction to 15 degrees, to avoid mirror reflection or shadow effect. As another implementation, when the pre-scan result indicates that there is local oil stain interference, the main scanning strategy can include local encryption scanning in the oil stain area, i.e. doubling the scanning line density in the oil stain area to obtain denser point cloud data, thereby improving the robustness of subsequent data processing. At the same time, the scanning speed of the line laser profiler can be adjusted, and the scanning speed in the oil stain area can be appropriately reduced to increase the action time of the laser on the surface and improve the signal acquisition quality.

[0028] Thirdly, the main scanning is performed according to the main scanning strategy to obtain the weld contour data. After the main scanning strategy is generated, the welding robot will drive the line laser profiler to accurately scan the weld area according to the strategy. For example, if the main scanning strategy requires increasing the laser power and adjusting the scanning angle, the line laser profiler will scan at the set high power and inclined angle. In this way, high-quality weld contour data can be obtained, which is usually represented in the form of point cloud, accurately reflecting the geometric shape of the weld.

[0029] Finally, the weld contour data is processed to identify the weld mode and extract the weld geometric dimensions. After obtaining the weld contour data, a series of processing is needed. For example, a geometric feature-based segmentation algorithm can be used to separate the weld area from the background. Subsequently, feature point extraction is performed on the separated weld area, such as identifying key points such as groove edge and root gap. By analyzing the spatial relationship and geometric shape of these feature points, the weld mode can be identified, such as V-type or U-type groove weld. At the same time, according to the coordinate information of these feature points, the geometric dimensions of the weld can be calculated, such as groove width, depth, and root face height. As a preferred embodiment, machine learning algorithms can be used to process the weld contour data. First, a deep learning model is used to perform semantic segmentation on the point cloud data to distinguish between weld area, base material area and potential defect area. Subsequently, a classification model is trained to identify different weld modes, such as by analyzing the shape features of the weld cross-section to determine the weld type. At the same time, a regression model is used to extract the geometric dimensions of the weld, such as by fitting the weld edge curve to calculate the groove angle and width.

[0030] In summary, the present application realizes adaptive processing of the surface conditions of complex workpieces by introducing pre-scanning, surface interference judgment and dynamic main scanning strategy generation mechanism. This method fundamentally improves the quality of weld contour data acquisition, and further significantly improves the accuracy and reliability of weld pattern recognition, effectively solves the problem of weld pattern recognition failure caused by inconsistency of workpiece surface in the prior art, and ensures the quality and efficiency of automatic welding.

[0031] In some embodiments of the present application described above, the main scanning strategy is generated according to the surface interference type and the surface interference degree to adjust the scanning parameters of the line laser profiler and / or the scanning path of the robot, so as to obtain the weld contour data. However, in the actual welding environment, there may be complex composite optical characteristic interference in the weld microscopic area, such as high reflection, strong absorption, scattering or multiple reflection, etc. These interferences may make it difficult for a single scanning strategy to comprehensively and accurately obtain high-quality weld contour data, which in turn affects the accuracy and robustness of subsequent weld pattern recognition and geometric dimension extraction.

[0032] To this end, with reference to Figure 2 the present application further proposes that the above S2 comprises: when the surface interference type and the surface interference degree indicate that there is composite optical characteristic interference in the weld microscopic area, a multiple scanning strategy is triggered; S21, according to the multiple scanning strategy, a first sub-scanning is performed to obtain a first group of weld contour point cloud data; S22, according to the multiple scanning strategy, a second sub-scanning is performed to obtain a second group of weld contour point cloud data; S23, the first group of weld contour point cloud data and the second group of weld contour point cloud data are fused to construct the weld contour data.

[0033] Specifically, when the pre-scanning data analysis result indicates that there is composite optical characteristic interference in the weld microscopic area, for example, due to the uneven roughness of the weld surface, the optical characteristic difference of the oxide layer, the spatter or the different material area, the reflection, absorption or scattering characteristics of the laser beam in different areas are complex and changeable, at this time the traditional single scanning strategy may not be able to effectively cope with it. Therefore, the system is configured to trigger a multiple scanning strategy, which aims to obtain more comprehensive and reliable weld contour information through multiple scanning.

[0034] The first sub-scan of the multi-scan strategy is performed to obtain a first set of weld contour point cloud data. This scan can employ a specific set of scan parameters, such as adjusting the laser power, exposure time, frame rate, or laser incidence angle of the line laser profiler, to optimize the capture of a certain type of optical interference. For example, for highly reflective areas, the laser power can be reduced or the incidence angle adjusted to avoid saturation; for strongly absorbing areas, the laser power can be increased or the exposure time extended to enhance the signal.

[0035] Subsequently, a second sub-scan is performed according to the multi-scan strategy to obtain a second set of weld contour point cloud data. The parameter settings for this scan are different from the first sub-scan, aiming to capture information in areas that may have been missed or poorly captured in the first scan, or to obtain supplementary data from different perspectives, lighting conditions, or wavelengths. For example, the wavelength, polarization state of the laser can be changed, or the incidence angle and scan path can be further adjusted to address different types of composite optical characteristic interference.

[0036] Finally, the first set of weld contour point cloud data and the second set of weld contour point cloud data are fused to construct the final weld contour data. The fusion process can employ various techniques such as point cloud registration, data denoising, feature extraction and merging, etc., aiming to integrate the complementary information obtained from the two scans to form a more complete, accurate and robust three-dimensional representation of the weld contour. The purpose is to overcome the limitations of single scan data acquisition in complex optical environments and provide high-quality input data for subsequent weld pattern recognition and geometric dimension extraction.

[0037] The scheme of the present application effectively addresses the composite optical characteristic interference in the microscopic area of the weld by introducing a multi-scan strategy. When such complex interference is identified by pre-scan analysis, the system no longer relies on a single scan parameter or path, but performs at least two sub-scans with different parameters or paths to capture the geometric information of the weld surface from multiple dimensions or angles. The first sub-scan may focus on obtaining contour data under a certain optical characteristic, while the second sub-scan may target other optical characteristics or supplement the deficiencies of the first scan. By fusing the two sets of point cloud data, the blind areas, noise or data missing that may exist in single scan can be effectively made up, thereby constructing more complete, accurate and robust weld contour data. The essence of this strategy is to utilize information redundancy and complementarity to improve the reliability of data acquisition in harsh optical environments.

[0038] By the technical solution, the welding seam profile data acquisition quality in the case of composite optical characteristic interference (such as high reflection, strong absorption, scattering or multiple reflection) in the welding seam micro area can be significantly improved. Compared with only using a single scanning strategy, the multiple scanning strategy combined with data fusion can effectively overcome the limitations of single scanning in a complex optical environment, reduce data missing, noise and distortion, and thus obtain more complete and more accurate welding seam three-dimensional profile data. Thus, high-quality and high-robustness input is provided for subsequent welding seam pattern recognition and welding seam geometric size extraction, and the accuracy and reliability of the welding seam detection are significantly improved, and the method is especially suitable for the precision manufacturing field with extremely high welding quality requirements.

[0039] In some preferred embodiments, the following is described by a specific example. It is assumed that when detecting a stainless steel welding seam, the pre-scanning data analysis finds that there is a local oxidation layer and a small amount of spatter on the welding seam surface, and these areas exhibit composite optical characteristic interference of high reflection and scattering. In this case, the system is configured to trigger a multiple scanning strategy.

[0040] Specifically, the first sub-scanning can use a lower laser power and a longer exposure time, and adjust the incident angle of the line laser profiler to be more suitable for capturing the welding seam geometric features below the oxidation layer, while avoiding laser saturation in the high reflection area. Thus, a first set of welding seam profile point cloud data is obtained, which may have partial missing or noise in the spatter area.

[0041] Subsequently, the second sub-scanning is performed, which can use different laser incident angles and a higher frame rate to more quickly scan the spatter area, and can reduce the influence of scattering effects by adjusting the wavelength or polarization state of the laser. Thus, a second set of welding seam profile point cloud data is obtained, which may capture the profile of the spatter area more clearly, but capture the details below the oxidation layer less than the first scanning.

[0042] Finally, the first set of welding seam profile point cloud data and the second set of welding seam profile point cloud data are fused. The fusion process can include registering the two sets of point cloud data, and then integrating the effective information in the two sets of data by weighted averaging or confidence-based fusion algorithm, removing noise points, and filling in missing areas. For example, in the spatter area, the data of the second scanning can be preferentially used; and below the oxidation layer, the data of the first scanning is preferentially used. In this way, the final welding seam profile data will be a more complete and accurate three-dimensional profile that integrates the advantages of the two sets of scanning, thereby providing a reliable basis for subsequent welding seam pattern recognition and geometric size extraction.

[0043] Specifically, referring to Figure 3 The above S4 includes: S41, initial segmentation of the weld contour data, separating the current weld region from the preliminary contour of the adjacent weld pass; S42, local geometric feature analysis of the separated current weld region, identifying and marking potential feature points; S43, feature analysis of the preliminary contour of the adjacent weld pass, extracting residual geometric boundary information; S44, spatial relationship judgment based on the potential feature points and the residual geometric boundary information, identifying local overlap; S45, when there is local overlap, adjusting the potential feature points to determine the true geometric boundary of the current weld region; S46, identifying the weld mode and extracting the weld geometric dimensions based on the true geometric boundary.

[0044] Wherein, the initial segmentation of the weld contour data aims to effectively separate the target weld region from the overall scanning data and preliminarily identify the possible adjacent weld pass contour. This step provides a clear data boundary for subsequent refined analysis, avoiding interference from irrelevant background information. Specifically, by applying clustering algorithms based on geometric features (such as curvature, normal vector change) or intensity information (such as laser reflection intensity), the preliminary definition of the weld region can be achieved.

[0045] Further, the local geometric feature analysis of the separated current weld region aims to identify and mark potential feature points that are crucial for weld mode and geometric dimension extraction. These potential feature points usually correspond to positions where the geometric morphology of the weld contour changes significantly, such as the edge of the groove, the root of the weld, or the weld toe. By accurately analyzing local curvature, gradient, or shape factor, these key points can be effectively located.

[0046] At the same time, the feature analysis of the preliminary contour of the adjacent weld pass aims to extract its residual geometric boundary information. In complex welding scenarios such as multi-pass welding or repair welding, the current weld region may be close to or even overlap with the completed adjacent weld pass in space. Extracting the geometric boundary information of these adjacent structures is crucial for subsequent judgment of their influence on the identification of the current weld.

[0047] On this basis, spatial relationship judgment is performed based on the potential feature points and the residual geometric boundary information to identify local overlap. By calculating the spatial distance, relative position, or geometric similarity between the potential feature points and the residual geometric boundary of the adjacent weld pass, it can be judged whether there is a local overlap phenomenon caused by factors such as welding process fluctuations, workpiece deformation, or scanning errors.

[0048] When there is the local overlap, the potential feature points need to be adjusted to determine the real geometric boundary of the current weld region. This means that the preliminarily identified potential feature points may be disturbed by the adjacent structure and need to be corrected according to the overlap situation and the geometric information of the adjacent structure, so as to eliminate the disturbance and ensure that the determined real geometric boundary of the current weld region is accurate.

[0049] Finally, according to the real geometric boundary, the weld mode can be identified and the weld geometric size can be extracted. Once the accurate real geometric boundary is obtained, the current weld mode can be identified based on these boundary information combined with the preset weld mode definition (for example, V groove, U groove, lap welding, etc.). At the same time, the geometric size of the weld, such as weld width, depth, groove angle, excess height, penetration, etc., can be calculated and extracted according to the key points on the real geometric boundary.

[0050] The scheme of the present application effectively solves the problem of inaccurate weld mode recognition and large error in geometric size extraction caused by factors such as adjacent structure disturbance and surface unevenness in complex weld scene through multi-stage and refined processing of the obtained weld contour data. Specifically, first, the target weld region is separated from the surrounding environment through initial segmentation to provide a clear data basis for subsequent analysis. Then, local geometric feature analysis is performed on the current weld region to accurately identify key potential feature points, which are the core of defining the weld shape and size. At the same time, by analyzing the features of the preliminary contour of the adjacent weld, the residual geometric boundary information is extracted, so that the system can perceive the possible external disturbance.

[0051] Further, by judging the spatial relationship between the potential feature points and the residual geometric boundary information, the local overlap phenomenon can be identified and quantified, which is crucial for distinguishing real weld features from external disturbances. When detecting local overlap, the scheme can intelligently adjust the potential feature points to eliminate the disturbance and ensure that the determined real geometric boundary of the current weld region is accurate.

[0052] Through the above technical scheme, the present application can significantly improve the accuracy of weld mode recognition and the precision of weld geometric size extraction. Especially in the scene of multi-pass welding, repair welding or complex background disturbance, the scheme effectively avoids the misrecognition and measurement error caused by external disturbance in the traditional method through refined segmentation, feature point recognition, adjacent structure analysis and overlap judgment and correction mechanism. Therefore, the reliability of the weld detection result is ensured, which provides a solid data basis for subsequent weld quality evaluation and process control, thereby improving the intelligent level and production efficiency of the automatic welding system.

[0053] Specifically, the step of performing local geometric feature analysis on the separated current weld region to identify and label potential feature points can be further refined, which includes: performing multi-scale local curvature calculation on the point cloud data of the separated current weld region to obtain multi-scale curvature results; According to the multi-scale curvature results, the discrete potential feature points are preliminarily fused; the intensity fluctuation analysis is performed on the discrete potential feature points, and the feature points affected by the micro surface unevenness are labeled; Performing continuity assessment on the labeled feature points, clustering continuous feature segments; According to the clustering results, the continuous feature segments are fitted, and the potential feature points are identified, including the groove edge and root gap feature points.

[0054] Among them, the multi-scale local curvature calculation on the point cloud data of the separated current weld region aims to obtain the geometric change information under different spatial scales. Specifically, various curvature calculation methods such as Gaussian curvature, average curvature or principal curvature can be used, and the calculation is performed under different neighborhood radius, thereby obtaining multi-scale curvature results. The purpose is to capture various geometric features in the weld region from fine texture to macro shape, while enhancing the robustness to noise.

[0055] Further, according to the multi-scale curvature results, the discrete potential feature points are preliminarily fused. Specifically, based on the curvature peaks or valleys calculated under different scales, a preliminary feature point set can be identified. Subsequently, through spatial proximity or feature similarity criteria, these discrete feature points are preliminarily merged and de-duplicated to form a more concise and representative potential feature point set.

[0056] On this basis, the intensity fluctuation analysis is performed on the discrete potential feature points, and the feature points affected by the micro surface unevenness are labeled. In practical applications, during the scanning process of the line laser profiler, the reflection intensity of the laser beam on the micro surface of the weld (such as weld slag, oxide layer, spatter, etc.) will fluctuate. By analyzing the laser reflection intensity data corresponding to these potential feature points, abnormal intensity values caused by surface unevenness can be identified, and these affected feature points can be labeled to distinguish them from real geometric feature points.

[0057] Subsequently, continuity evaluation is performed on the labeled feature points, and continuous feature segments are clustered. Specifically, the geometric continuity between the labeled feature points can be evaluated based on indicators such as spatial distance, normal vector consistency, or curvature variation trend. Through a clustering algorithm, adjacent feature points with good continuity are merged into continuous feature segments, for example, continuous point sets constituting the root gap or the edge of the groove are identified.

[0058] Finally, the continuous feature segments are fitted according to the clustering results, and the potential feature points are identified. For example, linear fitting, curve fitting, or spline fitting methods can be used to geometrically model the continuous feature segments obtained by clustering. Through fitting, the precise position and geometric shape of key potential feature points such as the edge of the groove and the root gap can be accurately determined, thereby providing accurate input for subsequent weld pattern recognition and geometric dimension extraction.

[0059] The scheme of the present application can comprehensively capture the geometric features of the weld area at different levels of detail by introducing multi-scale local curvature calculation, effectively dealing with the complex and variable geometric shapes of the weld surface. At the same time, combined with strength fluctuation analysis, it can effectively identify and exclude false feature points caused by microscopic surface inhomogeneity (such as slag, oxide layer), avoiding interference with subsequent identification. Through continuity evaluation and clustering of feature points, discrete point cloud data can be organized into meaningful geometric feature segments, and through fitting operation, key potential feature points such as the edge of the groove and the root gap can be accurately located and identified.

[0060] Through the above technical scheme, the accuracy and robustness of identifying potential feature points in complex weld environments can be significantly improved. Multi-scale analysis ensures comprehensive perception of features of different sizes, strength fluctuation analysis effectively suppresses surface interference, and continuity evaluation and fitting ensure the accuracy of feature point positioning. Thus, a solid foundation is laid for subsequent weld pattern recognition and geometric dimension extraction, improving the reliability and accuracy of the overall data processing method.

[0061] In some embodiments of the present application described above, when processing the weld contour data, preliminary contour of adjacent welds needs to be analyzed to extract residual geometric boundary information. However, in the actual welding process, due to factors such as welding process fluctuations, uneven cooling, or slag splashing, the preliminary contour of adjacent welds may exhibit irregular, discontinuous, or locally overlapping abnormal shapes, making it difficult to accurately and completely extract residual geometric boundary information through direct feature analysis, thereby affecting the accuracy of subsequent weld pattern recognition.

[0062] To this end, the present application further proposes that the step of analyzing the preliminary contour of the adjacent welds to extract residual geometric boundary information includes: performing multi-scale geometric feature extraction on the preliminary contour to obtain geometric variation information at different scales; performing local connectivity analysis based on the geometric variation information at different scales to identify and cluster continuous geometric feature segments; performing morphological analysis based on the continuous geometric feature segments to distinguish abnormal morphologies caused by welding process fluctuations, uneven cooling, or spatter; performing local reconstruction based on the geometric feature segments of abnormal morphologies to complete or smooth irregular, discontinuous, or overlapping regions; extracting the residual geometric boundary information based on the results of local reconstruction.

[0063] Specifically, performing multi-scale geometric feature extraction on the preliminary contour means processing the point cloud data of the preliminary contour using different scale analysis windows or filters to obtain geometric variation information at different spatial resolutions. For example, methods such as Gaussian smoothing, curvature calculation, or normal vector estimation can be used to capture fine local features at small scales and overall macroscopic morphology at large scales, with the goal of comprehensively understanding the geometric properties of the preliminary contour.

[0064] Among them, performing local connectivity analysis based on the geometric variation information at different scales to identify and cluster continuous geometric feature segments can be understood as grouping points belonging to the same continuous geometric structure based on spatial proximity and geometric feature similarity between points. For example, a density-based clustering algorithm (such as DBSCAN) or region growing algorithm can be used to group adjacent points with similar curvature, normal vector, or height variation into the same feature segment, with the goal of organizing discrete point cloud data into meaningful geometric units.

[0065] In practical applications, performing morphological analysis based on the continuous geometric feature segments to distinguish abnormal morphologies caused by welding process fluctuations, uneven cooling, or spatter specifically refers to evaluating the shape, size, smoothness, etc. of the clustered geometric feature segments. For example, the convexity, concavity, length, average curvature, or deviation from the ideal curve of the feature segment can be calculated to determine whether it meets the characteristics of a normal weld contour, thereby identifying abnormal morphologies caused by welding defects (such as weld bumps, depressions, and spatter attachment), with the goal of accurately identifying and locating contour distortions caused by non-ideal welding conditions.

[0066] Further, according to the abnormal morphological geometric feature segment, local reconstruction is performed to complete or smooth irregular, discontinuous or overlapping areas, which means that geometric modeling or data interpolation technology is used to correct the abnormal areas identified in the morphological analysis. For example, for discontinuous areas, the missing part can be completed by spline interpolation or curve fitting; for irregular or rough areas, noise can be eliminated by local smoothing algorithm (such as moving average, Gaussian filter); for locally overlapping areas, redundancy can be eliminated by geometric clipping or fusion algorithm, the purpose of which is to restore or construct a more true, smooth and continuous adjacent bead contour.

[0067] Finally, based on the results of local reconstruction, the residual geometric boundary information is extracted, which means that the key geometric points or line segments representing the true boundary of adjacent beads are accurately identified and extracted from the preliminary contour after correction and optimization. For example, the edge points, turning points or specific curvature points of the reconstructed contour can be identified as residual geometric boundaries, the purpose of which is to provide accurate and reliable input for subsequent spatial relationship judgment.

[0068] The scheme of the present application can comprehensively capture the details and overall structure of the preliminary contour of adjacent beads by introducing multi-scale geometric feature extraction, avoiding feature omission or misjudgment caused by single-scale analysis. On this basis, local connectivity analysis organizes discrete point cloud data into continuous geometric feature segments, laying a foundation for subsequent morphological analysis. It is precisely because of the morphological analysis of these continuous feature segments that the system can effectively distinguish abnormal morphologies caused by welding process fluctuations, uneven cooling or spatter, thereby accurately positioning the problem area. Subsequently, through targeted local reconstruction, irregular, discontinuous or overlapping areas are completed or smoothed, effectively eliminating the interference of these abnormal morphologies on boundary information extraction. Through this series of logically progressive steps, the scheme of the present application ensures that even in complex and variable actual welding environments, the residual geometric boundary information can be accurately and robustly extracted from the disturbed preliminary contour, providing a high-quality data basis for subsequent weld mode recognition.

[0069] Through the above technical scheme, the present application can effectively deal with complex situations such as irregularity, discontinuity or local overlap of the preliminary contour of adjacent beads caused by welding process fluctuations, uneven cooling or spatter. Compared with simple feature analysis, the present application significantly improves the accuracy and robustness of residual geometric boundary information extraction through the combined application of multi-scale feature extraction, local connectivity analysis, morphological analysis and local reconstruction. This not only avoids errors in weld mode recognition caused by inaccurate boundary information, but also ensures the reliability of weld geometry size extraction in complex industrial environments, thereby improving the adaptability and precision of the entire weld mode recognition data processing method.

[0070] In some preferred embodiments, the following is described by a specific example. Assume that in the acquired weld contour data, the preliminary contour of the adjacent bead presents multiple discrete protrusions in the local area due to spatter, and the contour line of the partial area presents slight wavy irregularities due to uneven cooling.

[0071] Firstly, multi-scale geometric feature extraction is performed on the preliminary contour. At a small scale, the sharp protrusions caused by spatter and the wavy irregularities can be identified; at a large scale, the overall smooth trend of the bead can be captured.

[0072] Secondly, local connectivity analysis is performed according to the geometric variation information at different scales. For example, points belonging to the same spatter or the same wave segment are clustered into continuous geometric feature segments.

[0073] Then, morphological analysis is performed on the continuous geometric feature segments. By calculating the curvature, height and area of the protrusion segments, they can be identified as abnormal morphologies caused by spatter; by analyzing the frequency and amplitude of the wave segments, they can be identified as irregular morphologies caused by uneven cooling.

[0074] Then, local reconstruction is performed according to the identified geometric feature segments of abnormal morphology. For the protrusions caused by spatter, local smoothing or removing abnormal points can be used to process them to make them smoothly transition with the surrounding contour; for the wavy irregularities, spline curve fitting can be used for smoothing to eliminate unnecessary fluctuations.

[0075] Finally, based on the smooth and continuous contour after local reconstruction, the residual geometric boundary information of the adjacent bead, such as its outermost edge line, is accurately extracted, thereby providing accurate input for subsequent weld pattern recognition.

[0076] In some embodiments of the present application described above, by performing initial segmentation, local geometric feature analysis, feature analysis of the preliminary contour of the adjacent bead on the weld contour data, and judging the spatial relationship between the potential feature points and the residual geometric boundary information to identify the local overlap. However, in actual application, slight deviation of the workpiece posture may cause local deformation, thereby affecting the accuracy of spatial relationship judgment, making the identification of local overlap not accurate enough, thereby affecting the reliability of subsequent weld pattern recognition and geometric dimension extraction.

[0077] To this end, the present application further proposes that the step of judging the spatial relationship between the potential feature points and the residual geometric boundary information to identify the local overlap comprises: constructing a local feature coordinate system of the current weld area; constructing a local boundary coordinate system of the residual geometric boundary information; calculating a transformation relationship between the local feature coordinate system and the local boundary coordinate system; transforming the residual geometric boundary information into the local feature coordinate system; analyzing local curvature variation trends of the current weld region and the residual geometric boundary information, identifying and quantifying local deformation information caused by workpiece posture deviation; correcting positions of the potential feature points according to the local deformation information; calculating spatial distances between the potential feature points and the residual geometric boundary information to identify the local overlap.

[0078] Specifically, a local feature coordinate system of the current weld region is constructed, aiming to provide a unified reference frame for potential feature points in the current weld region. This coordinate system can be defined based on the geometric center, main direction or specific feature points of the current weld region, for example, the center line of the weld can be taken as the X axis, the direction perpendicular to the center line and in the weld plane as the Y axis, and the direction perpendicular to the weld plane as the Z axis. Among them, the local boundary coordinate system of the residual geometric boundary information is constructed, which aims to provide an independent reference frame for the residual geometric boundary information extracted from the preliminary contour of the adjacent weld. This coordinate system can be established based on the geometric characteristics of the residual geometric boundary, such as its barycenter or main extension direction.

[0079] In practical applications, calculating the transformation relationship between the local feature coordinate system and the local boundary coordinate system means determining the rotation and translation matrix between the two coordinate systems through coordinate transformation algorithms, such as least squares method based on corresponding point matching or iterative closest point (ICP) algorithm. The purpose is to unify the geometric information in different reference systems into the same coordinate system for comparison and analysis. Further, transforming the residual geometric boundary information into the local feature coordinate system means using the transformation relationship calculated above to convert all point coordinates of the residual geometric boundary information from its local boundary coordinate system to the local feature coordinate system of the current weld region. Thus, the potential feature points and the residual geometric boundary information of the current weld region can be directly compared in the same spatial framework.

[0080] Further, analyzing the local curvature variation trend of the current weld seam region and the local curvature variation trend of the residual geometric boundary information, identifying and quantifying the local deformation information caused by the workpiece posture deviation, means that the local curvature of the two geometric entities is analyzed in detail, such as calculating the Gaussian curvature, the average curvature or the principal curvature, and the variation patterns are compared. When the workpiece posture deviates, its projection in three-dimensional space will be deformed, causing an identifiable change in local curvature. By comparing these curvature changes, the deformation area can be identified, and the deformation degree can be quantified by deformation gradient, deformation vector, etc. On this basis, according to the local deformation information, the position of the potential feature point is corrected, which means that the position of the potential feature point in the local feature coordinate system is adjusted compensatorily using the quantified local deformation information. For example, if it is identified that there is a stretching deformation in a certain direction in a certain area, the potential feature points in that area will be fine-tuned in the opposite direction to offset the positional error caused by the deformation.

[0081] Finally, calculating the spatial distance between the potential feature point and the residual geometric boundary information to identify local overlap means that after the position of the potential feature point is corrected, the Euclidean distance from each potential feature point to the nearest residual geometric boundary information point is calculated. When the spatial distance is less than a predetermined threshold, it is considered that there is a local overlap between the potential feature point and the residual geometric boundary information. The threshold value can be set according to the actual welding process and the measurement accuracy requirement.

[0082] The scheme of the present application effectively solves the influence of workpiece posture deviation on the accuracy of spatial relationship judgment by introducing local coordinate system construction, coordinate transformation, and identification and quantification of local deformation information. First, local coordinate systems are constructed for the current weld seam region and the residual geometric boundary information, laying the foundation for subsequent accurate alignment. Second, by calculating and applying the coordinate transformation relationship, the residual geometric boundary information is accurately converted to the local feature coordinate system of the current weld seam region, ensuring that the two are compared in the same reference system. More importantly, by analyzing the local curvature variation trend to identify and quantify the local deformation information caused by the workpiece posture deviation, the system can perceive and understand the geometric deformation of the actual workpiece. Therefore, according to these deformation information, the position of the potential feature point is corrected, which can effectively eliminate or weaken the positional error caused by the workpiece posture deviation, making the position of the potential feature point closer to its true position. Finally, on the basis of the corrected potential feature point, the spatial distance between the potential feature point and the residual geometric boundary information is calculated, which can more accurately and reliably identify local overlap, avoiding false positives caused by deformation.

[0083] By the technical solution, the accuracy and robustness of the local overlap recognition in the weld seam mode recognition data processing method can be improved. Especially in the complex actual welding environment with the workpiece posture deviation, the recognition and correction mechanism of the local deformation information is introduced, and the spatial relationship judgment error caused by the workpiece deformation in the traditional method is effectively overcome. This makes the position of the potential feature point more accurately determined, thereby providing a solid foundation for the accurate recognition of the subsequent weld seam mode and the accurate extraction of the weld seam geometric size, and improving the reliability and applicability of the entire weld seam detection system.

[0084] In some preferred embodiments, the following is described by a specific example. Assuming that when detecting a T-shaped joint weld, due to the fixture installation or the workpiece itself manufacturing tolerance, the workpiece has a slight posture tilt. After obtaining the weld contour data and performing the initial segmentation, the system recognizes the potential groove edge feature points of the current weld region and the residual geometric boundary information of the adjacent weld. In order to accurately determine whether there is local overlap between them, first, the system will construct a local feature coordinate system based on the center line of the current weld region. At the same time, the geometric center of the residual geometric boundary of the adjacent weld is taken as the reference to construct a local boundary coordinate system. Then, the rotation and translation matrix between the two coordinate systems is calculated by the iterative closest point (ICP) algorithm, and the residual geometric boundary information is accurately transformed to the local feature coordinate system of the current weld region. After that, the system will analyze the local curvature variation of the groove edge and the residual geometric boundary of the current weld region. For example, if it is found that the curvature variation of a certain region deviates from the standard model, and the deviation pattern is consistent with the deformation pattern caused by the workpiece tilt, the system will identify and quantify the local deformation information of the region, such as a slight twist or stretch. According to the quantified deformation information, the system will correct the position of the affected potential groove edge feature points, for example, fine-tune them in the opposite direction of the deformation. Finally, after the potential groove edge feature points are corrected, the system will calculate the spatial distance between each corrected groove edge point and the nearest residual geometric boundary point. If the distance between a groove edge point and the residual geometric boundary is less than the preset 0.5 millimeter threshold, the system will accurately recognize that there is local overlap in the region. In this way, even if the workpiece has a posture deviation, the method of the present application can accurately recognize the local overlap, avoid the misjudgment caused by the deformation, and ensure the accuracy of the subsequent weld seam mode recognition.

[0085] In some embodiments of the present application, when there is local overlap, the potential feature points need to be adjusted to determine the true geometric boundary of the current weld region. However, in actual welding detection process, due to the comprehensive influence of various factors such as welding heat input, material property difference, workpiece assembly error or measurement noise, the weld region may exhibit complex local deformation. These deformations may cause the spatial relationship between potential feature points and residual geometric boundary information to become complex and nonlinear, and if only based on simple geometric relationship or fixed rules for adjustment, the true geometric boundary may not be accurately distinguished and determined, thereby affecting the accuracy and robustness of subsequent weld mode recognition and geometric dimension extraction. In view of this, the present application further proposes a more refined method for adjusting potential feature points and determining the true geometric boundary of the current weld region, which significantly improves the accuracy and reliability of boundary determination by in-depth analysis and compensation of local deformation.

[0086] In view of this, the present application further proposes a step of adjusting potential feature points and determining the true geometric boundary of the current weld region when there is local overlap, which specifically includes: Performing local deformation field analysis on the potential feature points and residual geometric boundary information of the current weld region, identifying the local deformation type and local deformation degree; According to the local deformation type and local deformation degree, a local deformation compensation function is constructed; According to the local deformation compensation function, the initial position of the potential feature point is corrected; Around the corrected potential feature point, the search region is dynamically adjusted according to the local geometric shape of the residual geometric boundary information; In the adjusted search region, the position of the potential feature point is iteratively optimized based on the matching degree and spatial distance of local geometric features, the potential feature point is distinguished from the residual geometric boundary information, and the true geometric boundary of the current weld region is determined.

[0087] Specifically, the local deformation field analysis refers to extracting geometric features such as local curvature and normal vector from the potential feature points and residual geometric boundary information of the current weld region, and combining the material properties, thickness difference or welding heat input information corresponding to the potential feature points and residual geometric boundary information of the current weld region to perform weighted processing on the local curvature and normal vector to obtain weighted geometric features. By calculating the local deformation gradient, the deformation gradient field can be obtained, and then the local deformation type such as stretching, compression, bending or twisting can be identified according to the deformation gradient field, and its degree can be quantified according to the amplitude and distribution of the deformation gradient field. The purpose is to comprehensively understand and quantify the complex deformation of the weld region, and to provide data basis for subsequent accurate compensation.

[0088] Wherein, the local deformation compensation function is constructed according to the local deformation type and the local deformation degree, which can be understood as establishing a mathematical model or mapping relationship according to the identified deformation characteristics, for offsetting or correcting the influence of deformation. For example, when the coupling effect of multiple local deformation types is identified, a deformation component compensation function can be constructed for each deformation type and its degree, and these component functions are combined through spatial superposition to form a comprehensive local deformation compensation function. The purpose is to provide a quantifiable correction mechanism to eliminate or weaken the influence of deformation on the feature point position.

[0089] In practical applications, the initial position correction of the potential feature point according to the local deformation compensation function refers to applying the compensation function to the original coordinates of the potential feature point to obtain a preliminary corrected position. This correction aims to move the feature point affected by deformation to a region closer to its true position, laying a foundation for subsequent refined search and optimization.

[0090] Further, around the corrected potential feature point, the search area is dynamically adjusted according to the local geometric shape of the residual geometric boundary information, aiming to adapt to the complex boundary shape under different deformation and overlap conditions. For example, if the residual geometric boundary information presents a large curvature change or irregularity in a certain local area, the search area can be correspondingly expanded to ensure that the true boundary is not missed; on the contrary, if the boundary shape is relatively smooth, the search area can be reduced to improve search efficiency and accuracy.

[0091] Thus, within the adjusted search area, the position of the potential feature point is iteratively optimized based on the matching degree of local geometric features and spatial distance, and the potential feature point is distinguished from the residual geometric boundary information, to finally determine the true geometric boundary of the current weld area. Specifically, the geometric feature similarity (such as curvature matching, normal vector consistency) and spatial distance between the potential feature point and the residual geometric boundary information in the search area can be calculated, and the Iterative Closest Point (ICP) algorithm or energy minimization-based method can be used for optimization. Through multiple iterations, the best matching position is gradually converged, so as to accurately distinguish the true boundary of the current weld area and exclude the interference caused by overlap.

[0092] The scheme of the present application can deeply understand and quantify the complex deformation of the potential feature points and residual geometric boundary information in the weld area by introducing local deformation field analysis. By identifying the type and degree of deformation, a local deformation compensation function can be constructed to correct the initial position of the potential feature points, effectively offsetting the positional deviation caused by deformation. On this basis, combined with the dynamic adjustment of the search area according to the local geometric shape of the residual geometric boundary information, it ensures that the potential position of the true boundary can still be effectively covered in the complex deformation environment. Finally, through the iterative optimization process based on the matching degree of local geometric features and spatial distance, the true geometric boundary of the current weld area can be accurately distinguished and determined, overcoming the limitations of traditional methods in dealing with complex local overlap and deformation.

[0093] Through the above technical scheme, the present application can significantly improve the accuracy and robustness of weld geometric boundary determination in the presence of local overlap. Compared with the method of relying only on simple spatial relationship adjustment, the present application effectively solves the interference of complex deformation on boundary identification caused by factors such as welding heat input, material property difference or workpiece posture deviation, etc. Thus, the weld mode can be more accurately identified and the weld geometric size can be extracted, providing a more reliable data basis for subsequent weld quality evaluation and control, thereby improving the adaptability and reliability of the entire weld mode recognition data processing method.

[0094] In some preferred embodiments, the following is described by a specific example. Assuming that when detecting a multi-layer multi-pass weld, due to the cooling shrinkage of the previous pass and the heat input of the subsequent pass, there is complex local overlap and deformation between the potential feature points (i.e. potential feature points) of the gap between the root and the bevel edge of the current weld area to be identified and the residual geometric boundary information of the adjacent completed pass.

[0095] Firstly, the system obtains the pre-scanning data of the current weld area and judges that there is complex optical characteristic interference and moderate surface unevenness on the surface. Based on this, the main scanning strategy is generated, including adjusting the exposure time, laser power of the line laser profiler and the scanning speed of the robot to obtain high-quality weld contour data.

[0096] When processing the weld contour data, firstly, the data is initially segmented to separate the preliminary contour of the current weld area and the adjacent pass. Then, the point cloud data of the current weld area is subjected to multi-scale local curvature calculation, and combined with intensity fluctuation analysis, the potential feature points such as the gap between the root and the bevel edge are identified. At the same time, the preliminary contour of the adjacent pass is subjected to multi-scale geometric feature extraction and local connectivity analysis, and morphological analysis and local reconstruction are performed to extract the residual geometric boundary information.

[0097] The scheme of the present application is triggered when the system identifies local overlap based on spatial relationship judgment of potential feature points and residual geometric boundary information. Specifically: 1. Perform local deformation field analysis on the potential feature points and residual geometric boundary information of the current weld region. For example, by calculating the local Gaussian curvature and average curvature of each point, and combining with the elastic modulus and thermal expansion coefficient of the material for weighting, the weighted geometric features are obtained. Based on these features, the deformation gradient field is calculated to identify the existence of coupled deformation of stretching and bending in the local region, and to quantify the degree of deformation.

[0098] 2. According to the identified stretching and bending deformation types and their degrees, respectively, construct a stretching compensation function (e.g., based on a linear or nonlinear model) and a bending compensation function (e.g., based on a polynomial fitting). Subsequently, by means of spatial superposition, the two component compensation functions are combined into a comprehensive local deformation compensation function.

[0099] 3. Apply the constructed local deformation compensation function to the original coordinates of the potential feature points to correct the initial positions of the potential feature points. For example, if a certain potential feature point is offset outward due to local stretching, the compensation function will move it inward by a calculated distance.

[0100] 4. Around the corrected potential feature points, the system dynamically adjusts the search area according to the local geometric shape of the residual geometric boundary information. For example, if the residual geometric boundary presents an S-shaped bending in a certain local area, the search area will be adjusted to a non-rectangular area that can cover the bending shape.

[0101] 5. Within the adjusted search area, the system uses the Iterative Closest Point (ICP) algorithm to iteratively optimize the positions of the potential feature points based on the matching degree of local geometric features (e.g., normal vector angle, curvature similarity) and spatial distance. Through multiple iterations, the positions of the potential feature points will gradually converge, and finally accurately distinguish the true groove edge and root gap of the current weld region, and clearly distinguish them from the residual geometric boundary information of the adjacent weld, thereby determining the true geometric boundary of the current weld region.

[0102] Through the above process, even under complex local overlap and deformation conditions, the scheme of the present application can accurately determine the true geometric boundary of the weld, providing high-precision input data for subsequent weld mode recognition (e.g., identifying as a V-shaped groove weld) and geometric dimension extraction (e.g., groove width, root gap size).

[0103] Specifically, the above steps of local deformation field analysis on the potential feature points and residual geometric boundary information of the current weld region, identifying the local deformation type and the local deformation degree, can be further refined.

[0104] The step of performing local deformation field analysis on the potential feature points of the current weld region and the residual geometric boundary information, identifying the local deformation type and the local deformation degree, includes: Performing geometric feature extraction on the potential feature points of the current weld region and the residual geometric boundary information, obtaining local curvature and normal vector; According to the material properties, thickness differences or welding heat input information corresponding to the potential feature points of the current weld region and the residual geometric boundary information, performing weighted processing on the local curvature and the normal vector to obtain weighted geometric features; According to the weighted geometric features, calculating the local deformation gradient to obtain a deformation gradient field; According to the deformation gradient field, identifying the local deformation type; According to the amplitude and distribution of the deformation gradient field, quantifying the local deformation degree.

[0105] Specifically, geometric feature extraction refers to calculating parameters that describe the local surface shape from point cloud data, such as local curvature and normal vector. Local curvature reflects the degree of surface bending, while the normal vector indicates the direction of the surface. These features are the basis for analyzing surface deformation.

[0106] Among them, the weighted processing of local curvature and normal vector to obtain weighted geometric features means that the extracted geometric features are modified in combination with physical properties related to the weld region. These physical properties can include material properties (such as elastic modulus, yield strength), thickness differences (such as the inconsistency of weld and base material thickness), or welding heat input information (such as welding current, voltage, speed, etc.), all of which can affect the actual deformation response of the weld region. Through weighted processing, the deformation sensitivity of different regions can be more accurately reflected.

[0107] In practical applications, calculating the local deformation gradient to obtain the deformation gradient field means that based on the weighted geometric features, the rate of change of the geometric features in space is calculated through mathematical methods (such as finite difference or gradient operator). The deformation gradient field can intuitively represent the direction and intensity of deformation, providing a basis for subsequent deformation type identification.

[0108] Further, according to the deformation gradient field to identify the local deformation type, it means that by analyzing the pattern and distribution of the deformation gradient field, it is judged whether the deformation belongs to the specific types of stretching, compression, shear, bending, etc. For example, the divergence or vorticity of the gradient field can be used to distinguish different deformation patterns.

[0109] The local deformation degree is quantified according to the amplitude and distribution of the deformation gradient field, which means that the size of the identified deformation type is further calculated. The amplitude can represent the intensity of the deformation, and the distribution can reveal the spread range and concentration degree of the deformation in space, thereby providing accurate quantitative indicators.

[0110] The scheme of the present application solves the problem that it is difficult to accurately identify and quantify the local deformation by only preliminary geometric features in a complex welding environment by performing more fine local deformation field analysis on the potential feature points and residual geometric boundary information of the current weld area. Specifically, first, the local curvature and normal vector are obtained by geometric feature extraction to lay the foundation for deformation analysis. Then, the material properties, thickness differences or welding heat input information are introduced to weight these geometric features, so that the deformation analysis can fully consider the influence of actual physical factors on the weld deformation, thereby improving the representativeness and accuracy of the features. On this basis, the local deformation gradient is calculated to construct a deformation gradient field, which can clearly depict the spatial variation trend and intensity of the deformation, thereby providing a quantitative basis for accurately identifying the local deformation type. Finally, the accurate quantification of the local deformation degree is realized by analyzing the amplitude and distribution of the deformation gradient field, which provides reliable data support for subsequent deformation compensation and determination of the real geometric boundary.

[0111] Through the above technical scheme, the local deformation characteristics of the weld area can be more comprehensively and accurately understood. The geometric features are weighted by combining the material properties, thickness differences or welding heat input information, so that the deformation analysis result is closer to the actual working condition, effectively avoiding misjudgment caused by single geometric feature analysis. By constructing the deformation gradient field and conducting in-depth analysis, not only the specific deformation type can be identified, but also the deformation degree can be accurately quantified, thereby providing more fine and reliable input for subsequent local deformation compensation function construction and position correction of potential feature points, significantly improving the accuracy and robustness of weld mode recognition and geometric dimension extraction.

[0112] In some embodiments of the present application described above, when determining the real geometric boundary of the current weld area, it is necessary to construct a local deformation compensation function according to the local deformation type and the local deformation degree. However, in the actual welding process, the local deformation of the workpiece is often not a single type, but a result of mutual coupling and joint action of multiple deformation types. If only a single deformation type or simple superposition processing is used, the influence of these complex coupled deformations on the weld contour data may not be accurately captured, resulting in insufficient accuracy of the constructed local deformation compensation function, which further affects the correction effect of the potential feature points and the determination of the real geometric boundary. In view of this, the present application further proposes a method for constructing a local deformation compensation function, which aims to more accurately process the coupling influence of multiple local deformation types.

[0113] Specifically, the local deformation compensation function is constructed according to the local deformation type and the local deformation degree, and the construction includes: When the coupling effect of multiple local deformation types is identified, a deformation component compensation function is constructed according to each local deformation type and the local deformation degree thereof; The deformation component compensation functions are combined in a spatial superposition manner to form the local deformation compensation function.

[0114] The coupling effect of multiple local deformation types refers to the occurrence of two or more different types of deformation in the local deformation field of the weld area, and the interaction between them makes the overall deformation effect not simply linearly superimposed. For example, there may be local thermal stress deformation caused by uneven welding heat input, mechanical stress deformation caused by improper clamping of the workpiece, and deformation caused by unevenness of the microstructure of the material, etc. These deformation types may enhance, weaken or change the distribution of each other, thereby forming a complex coupled deformation field.

[0115] The deformation component compensation function refers to a compensation model or function established for each identified local deformation type according to its specific deformation mechanism and degree. The function aims to quantify and correct the geometric deviation caused by the single deformation type. For example, for thermal stress deformation, a compensation function based on temperature field distribution and material thermal expansion coefficient can be constructed; for mechanical stress deformation, a compensation function based on stress distribution and material elastic modulus can be constructed.

[0116] The spatial superposition manner refers to the combination or fusion of each deformation component compensation function constructed for different deformation types in space. This combination is not simply numerical addition, but considers the distribution, interaction and contribution of different deformation components to the overall deformation field, and integrates them through mathematical models (such as vector superposition, weighted average, finite element analysis result fusion, etc.) to form a comprehensive compensation function that can fully reflect all coupled deformation effects. The purpose is to ensure that the final local deformation compensation function can accurately reflect the actual complex deformation situation.

[0117] The solution of the present application can solve the above problems because when the coupling effects of multiple local deformation types are identified, the complex coupling deformation is no longer compensated as a whole, but is first decomposed into multiple independent deformation components. It is because the deformation component compensation function is constructed for each deformation component and its degree that the influence of each deformation type can be accurately quantified and modeled. On this basis, by combining these deformation component compensation functions in a spatial superposition manner, the interaction and superposition effect between different deformation types can be effectively captured, thereby forming a more comprehensive and accurate local deformation compensation function. This decomposition and superposition strategy makes the compensation of complex coupling deformation more refined and accurate, avoiding under-compensation or over-compensation due to simplification.

[0118] Through the above technical solution, the present application can significantly improve the accuracy and robustness of the local deformation compensation function, especially in complex working conditions where multiple local deformation types are coupled in the weld area. Compared with the basic solution that does not distinguish or effectively handle coupling deformation, the present application can more accurately quantify and correct the geometric deviation caused by complex deformation by independently modeling and spatially superimposing different deformation components. This directly leads to more accurate correction of the initial position of the potential feature point, and further enables the real geometric boundary of the current weld area to be more accurately determined, ultimately improving the reliability and accuracy of weld mode recognition and geometric dimension extraction.

[0119] In some preferred embodiments, the following is described by a specific example. Assume that in a certain welding process, through local deformation field analysis of the potential feature points and residual geometric boundary information of the current weld area, two main local deformation types are identified: the first is thermal stress deformation caused by uneven welding heat input, which is manifested as local area bulging or sagging; the second is mechanical stress deformation caused by uneven force of the workpiece in the clamp, which is manifested as slight bending along the weld direction.

[0120] At this time, according to the solution of the present application, first, for thermal stress deformation, a thermal stress deformation component compensation function is constructed according to the identified deformation degree (e.g., bulging height or sagging depth). This function can map the temperature field distribution to geometric deformation based on the thermoelasticity model. Then, for mechanical stress deformation, a mechanical stress deformation component compensation function is constructed according to the identified deformation degree (e.g., bending radius or deviation amount). This function can map external load to geometric deformation based on the structural mechanics model.

[0121] Subsequently, the two deformation component compensation functions are combined by means of spatial superposition. For example, the superposition principle in finite element analysis can be used to vectorially superimpose the geometric correction amounts of the two deformation components in their respective action regions, thereby forming a comprehensive local deformation compensation function. This comprehensive function can reflect the coupled effects of thermal stress and mechanical stress on the weld contour and provide a more accurate overall compensation scheme. In this way, the position correction of the potential feature points will be more accurate, and the final determined real geometric boundary of the current weld region will be closer to the actual situation.

[0122] Referring to Figure 4 The embodiment of the present application provides a kind of weld mode identification data processing system structure diagram, comprising: Detection end, for obtaining the pre-scanning data of weld area, and analyzing the pre-scanning data to judge the surface interference type and surface interference degree of the weld area; Adjustment end, for generating main scanning strategy according to the surface interference type and surface interference degree, the main scanning strategy includes adjusting the scanning parameter of line laser profiler and / or the scanning path of robot; Processing end, with according to the main scanning strategy, executes main scanning, obtains weld contour data;Weld contour data is processed, weld mode is identified and weld geometric dimension is extracted.

[0123] It needs to be explained that the weld mode identification data processing system provided by the embodiment of the present application is used to execute all process steps of the weld mode identification data processing method of the above-mentioned embodiment, and the working principles and beneficial effects of the two are one-to-one correspondence, thus no longer tedious.

[0124] The embodiment of the present application further provides a kind of terminal equipment. The terminal equipment includes: processor, memory and computer program stored in the memory and executable on the processor. The processor executes the computer program to realize the steps in each weld mode identification data processing method embodiment described above, for example Figure 1 The step S1 shown. Alternatively, the processor executes the computer program to realize the functions of each module / unit in each system embodiment described above.

[0125] It should be noted that the system embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e. can be located in one place or distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. In addition, the connection relationship between the modules in the system embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0126] The above specific embodiments further illustrate the purpose, technical scheme and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A weld seam pattern recognition data processing method, characterized in that, The method comprises: acquiring pre-scanning data of a weld area, and analyzing the pre-scanning data to determine a surface interference type and a surface interference degree of the weld area; generating a main scanning strategy according to the surface interference type and the surface interference degree, the main scanning strategy comprising adjusting scanning parameters of a line laser profiler and / or a scanning path of a robot; performing main scanning according to the main scanning strategy to acquire weld contour data; processing the weld contour data to identify a weld mode and extract a weld geometric size.

2. The weld pattern recognition data processing method of claim 1, wherein, The generating of the main scanning strategy according to the surface interference type and the surface interference degree, the main scanning strategy comprising adjusting scanning parameters of a line laser profiler and / or a scanning path of a robot, comprises: when the surface interference type and the surface interference degree indicate that there is composite optical characteristic interference in a microscopic area of the weld, triggering a multiple scanning strategy; performing first-time sub-scanning according to the multiple scanning strategy to acquire a first group of weld contour point cloud data; performing second-time sub-scanning according to the multiple scanning strategy to acquire a second group of weld contour point cloud data; fusing the first group of weld contour point cloud data and the second group of weld contour point cloud data to construct the weld contour data.

3. The method of claim 1, wherein The processing of the weld contour data to identify a weld mode and extract a weld geometric size comprises: initially segmenting the weld contour data to separate a preliminary contour of a current weld area from an adjacent weld; performing local geometric feature analysis on the separated current weld area to identify and mark potential feature points; performing feature analysis on the preliminary contour of the adjacent weld to extract residual geometric boundary information; judging a spatial relationship according to the potential feature points and the residual geometric boundary information to identify local overlap; when the local overlap exists, adjusting the potential feature points to determine a real geometric boundary of the current weld area; identifying a weld mode and extracting the weld geometric size according to the real geometric boundary.

4. A weld pattern recognition data processing method according to claim 3, characterized in that, The performing of local geometric feature analysis on the separated current weld area to identify and mark potential feature points comprises: performing multi-scale local curvature calculation on point cloud data of the separated current weld area to obtain multi-scale curvature results; preliminarily fusing discrete potential feature points according to the multi-scale curvature results; performing intensity fluctuation analysis on the discrete potential feature points to mark feature points affected by microscopic surface unevenness; performing continuity evaluation on the marked feature points to cluster continuous feature segments; fitting the continuous feature segments according to clustering results to identify the potential feature points, the potential feature points comprising a groove edge and a root gap feature point.

5. The method of claim 3, wherein the weld pattern recognition data processing method is characterized by, The performing of feature analysis on the preliminary contour of the adjacent weld to extract residual geometric boundary information comprises: performing multi-scale geometric feature extraction on the preliminary contour to obtain geometric change information of different scales; performing local connectivity analysis according to the geometric change information of different scales to identify and cluster continuous geometric feature segments; performing morphology analysis according to the continuous geometric feature segments to distinguish abnormal morphologies caused by welding process fluctuation, uneven cooling or spatter of welding slag; According to the abnormal morphological geometric feature segment, local reconstruction is performed to complete or smooth irregular, discontinuous or overlapping areas; Based on the results of local reconstruction, the residual geometric boundary information is extracted.

6. The data processing method for weld pattern recognition according to claim 3, wherein, The spatial relationship judgment according to the potential feature points and the residual geometric boundary information includes: A local feature coordinate system of the current weld area is constructed; A local boundary coordinate system of the residual geometric boundary information is constructed; A transformation relationship between the local feature coordinate system and the local boundary coordinate system is calculated; The residual geometric boundary information is transformed into the local feature coordinate system; The local curvature variation trend of the current weld area and the local curvature variation trend of the residual geometric boundary information are analyzed to identify and quantify the local deformation information caused by the workpiece posture deviation; According to the local deformation information, the position of the potential feature point is corrected; The spatial distance between the potential feature point and the residual geometric boundary information is calculated to identify the local overlap.

7. The data processing method for weld pattern recognition according to claim 3, wherein, When the local overlap exists, the potential feature point is adjusted to determine the true geometric boundary of the current weld area, including: Local deformation field analysis is performed on the potential feature points of the current weld area and the residual geometric boundary information to identify the local deformation type and the local deformation degree; According to the local deformation type and the local deformation degree, a local deformation compensation function is constructed; According to the local deformation compensation function, the initial position of the potential feature point is corrected; According to the local geometric morphology of the residual geometric boundary information around the corrected potential feature point, the search area is dynamically adjusted; Based on the matching degree and the spatial distance of the local geometric features in the adjusted search area, the position of the potential feature point is iteratively optimized to distinguish the potential feature point from the residual geometric boundary information and determine the true geometric boundary of the current weld area.

8. A weld seam pattern recognition data processing method according to claim 7, characterized in that, The step of performing local deformation field analysis on the potential feature points of the current weld area and the residual geometric boundary information to identify the local deformation type and the local deformation degree includes: Geometric feature extraction is performed on the potential feature points of the current weld area and the residual geometric boundary information to obtain local curvature and normal vector; According to the material properties, thickness difference or welding heat input information corresponding to the potential feature points of the current weld area and the residual geometric boundary information, the local curvature and the normal vector are weighted to obtain weighted geometric features; According to the weighted geometric features, a local deformation gradient is calculated to obtain a deformation gradient field; According to the deformation gradient field, the local deformation type is identified; According to the amplitude and distribution of the deformation gradient field, the local deformation degree is quantified.

9. The method of data processing for weld pattern recognition according to claim 7, wherein, The construction of the local deformation compensation function according to the local deformation type and the local deformation degree includes: When the coupling effect of multiple local deformation types is identified, a deformation component compensation function is constructed for each local deformation type and its local deformation degree; The deformation component compensation functions are combined by spatial superposition to form the local deformation compensation function.

10. A weld seam pattern recognition data processing system characterized by, The system includes: The detection end is used for acquiring pre-scanning data of the weld area and analyzing the pre-scanning data to determine the surface interference type and the surface interference degree of the weld area. The adjustment end is used for generating a main scanning strategy according to the surface interference type and the surface interference degree, wherein the main scanning strategy comprises adjusting the scanning parameters of the line laser profiler and / or the scanning path of the robot. The processing end is used for performing main scanning according to the main scanning strategy to acquire weld contour data, processing the weld contour data to identify a weld mode and extract weld geometric dimensions.

Images