Control method of welding type coupling equipment

Abstract

The invention provides a control method of welding type coupling equipment, which comprises the following steps: comparing extracted contour and center line data according to a preset welding spot size and a preset fusion width threshold value, and judging whether the size of a welding spot is abnormal or the fusion width of the welding spot is abnormal or not; if the welding spot is abnormal, adopting an image segmentation method to identify an abnormal region in the image, and calculating the area and geometric features of the abnormal region through a pixel statistical method to obtain the size and deformation degree of the welding spot; the corrected welding instruction data are obtained, and energy, focal length and welding gun angle parameters in an instruction are extracted to serve as reference values of welding instruction analysis; by analyzing an energy distribution curve in the welding process and combining focal length data collected in real time and welding gun angle changes, whether welding energy and focal length are matched with reference values of welding instruction analysis or not is judged; if the welding energy and the focal length are not matched with the reference value analyzed by the welding instruction, the control parameters of the welding equipment are adjusted according to the deviation value, and the correction amount of the welding gun path is recalculated.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a control method for welding coupling equipment. Background Art

[0002] In the field of precision welding, the quality of solder joints is a key factor in determining product performance and reliability. With the increasing requirements for welding accuracy in the electronics, medical and other industries, traditional welding quality inspection methods have been unable to meet the needs. For a long time, the industry has mainly relied on manual visual inspection or sampling destructive testing to evaluate the quality of solder joints. This method is not only inefficient, but also consumes a lot of manpower and material resources. It is also impossible to achieve comprehensive and real-time detection of each solder joint, resulting in potential quality risks that are difficult to be discovered in time. In order to improve the control level of welding quality, people have tried to introduce automated detection technology. However, without contacting the solder joint, accurately obtaining the geometric feature information of the solder joint itself faces many challenges, such as the small size and complex shape of the solder joint, as well as interference factors such as spatter and smoke generated during the welding process. Even if the image information of the solder joint can be obtained, how to accurately extract the key parameters such as the contour, size, and weld width of the solder joint from it also has problems such as complex algorithms and poor robustness. In addition, the tiny deformation and position deviation of the solder joint are often difficult to observe with the naked eye, and traditional methods are difficult to achieve automatic recognition and quantitative analysis of these tiny defects. More importantly, during the welding process, parameters such as energy input, focal length adjustment, and welding gun posture will have a significant impact on the quality of the weld, and traditional technology is difficult to monitor the changes in these parameters in real time, and it is even more difficult to establish an effective correlation model between them and the quality of the weld, resulting in a lack of scientific basis for welding process control. Therefore, how to automatically adjust the welding parameters in a timely manner when abnormalities are found to ensure the stability of the welding quality, and how to quickly and accurately evaluate the quality of the weld after welding is completed, and provide data support for subsequent process optimization, are still key issues that need to be urgently solved in the field of precision welding. Summary of the invention

[0003] The present invention provides a control method for a welding coupling device, which mainly includes:

[0004] The continuous image sequence of the welding instruction execution process is collected by a high-speed camera, the contour information of the welding spot is extracted by edge detection algorithm, and the center line of the welding spot is generated by combining morphological processing method.

[0005] According to the preset weld spot size and weld width threshold, the extracted contour and centerline data are compared to determine whether the weld spot has abnormal size or abnormal weld width;

[0006] If there is an abnormality, the image segmentation method is used to identify the abnormal area in the image, and the area and geometric features of the abnormal area are calculated by the pixel statistics method to obtain the size and deformation degree of the solder joint;

[0007] According to the preset welding torch path parameters and combined with the extracted coordinates of the center line of the solder joint, calculate the deviation value between the center line of the solder joint and the preset path to obtain the offset of the solder joint position. Then, combine the size and deformation degree of the solder joint to correct the welding instruction;

[0008] Obtain the data of the welding instruction after correction, and extract the energy, focal length, and welding torch angle parameters in the instruction as the reference values for welding instruction analysis;

[0009] By analyzing the energy distribution curve during the welding process and combining the real-time collected focal length data and the change of the welding torch angle, judge whether the welding energy and focal length match the reference values analyzed from the welding instruction;

[0010] If the welding energy and focal length do not match the reference values analyzed from the welding instruction, adjust the control parameters of the welding equipment according to the deviation value and recalculate the correction amount of the welding torch path;

[0011] Input the adjusted control parameters into the control system of the welding equipment, monitor the changes of energy and focal length during the welding process in real time, verify the stability of the solder joint contour and center line by continuously collecting welding process images, and judge whether the welding quality meets the preset standard.

[0012] Further, the continuous image sequence of the welding instruction execution process is collected by a high-speed camera, the contour information of the solder joint is extracted by using an edge detection algorithm, and the center line of the solder joint is generated by combining a morphological processing method, including: collecting a first image sequence of the movement trajectory of the welding head during the execution of the welding instruction by an industrial-grade high-speed imaging device according to a preset acquisition frequency, extracting first grayscale image data from the first image sequence, obtaining an optimal grayscale threshold according to the statistical analysis of the image histogram, and performing binary processing on the first grayscale image data to obtain first binary image data. The first binary image data is denoised by using a Gaussian filter to obtain second binary image data, and the Sobel operator is used to perform edge detection on the second binary image data to obtain first edge contour data. The density distribution of the contour pixel points is calculated according to the first edge contour data, and the breakpoints of the edge contour are connected by using a density clustering method to obtain second edge contour data. A morphological closing operation is performed on the second edge contour data to obtain closed contour line data. Based on the closed contour line data, a morphological thinning algorithm based on distance transformation is used to extract a first skeleton data, and the first skeleton data is normalized in width according to a preset skeleton width threshold to obtain second skeleton data. The cubic spline interpolation algorithm is used to perform node continuity processing on the second skeleton data to obtain first center line data, and the first center line data is smoothed by a curve according to a preset curvature threshold to obtain second center line data. A morphological erosion operator is used to thin the second center line data to obtain solder joint contour center line data, and the integrity of the center line data is verified by detecting the connectivity of the center line pixel points.

[0013] Further, comparing the extracted contour and centerline data with the preset solder joint size and fusion width threshold to determine whether there are size anomalies or fusion width anomalies in the solder joint, including: reading the standard solder joint size parameters from the preset parameter database, extracting the actual solder joint size parameters by feature extraction of the first solder joint contour data, and calculating the deviation value between the actual size and the standard size by using the curve matching method to obtain the first anomaly feature data. Using the region growing algorithm to extract the fusion width feature region from the first solder joint contour data to obtain the first fusion width data, using the distance transformation method to statistically analyze the first fusion width data to obtain the fusion width mean value, reading the standard fusion width range parameters from the preset parameter database, and calculating the second anomaly feature data according to the fusion width mean value and the standard range. Using the equidistant sampling method to sample the points of the first solder joint contour data to obtain the first sampling point sequence, calculating the local curvature of the first sampling point sequence to obtain the first curvature distribution data, and calculating the roundness parameter by least square fitting to obtain the third anomaly feature data. Extracting the local extreme points from the first curvature distribution data to obtain the second sampling point sequence, calculating the spatial distribution period of the second sampling point sequence to obtain the first period data, and calculating the frequency spectrum components of the first period data by using the fast Fourier transform to obtain the fourth anomaly feature data. Combining the first anomaly feature data to the fourth anomaly feature data to form a solder joint comprehensive anomaly feature vector, reading the anomaly discrimination threshold from the preset parameter database, and performing multi-dimensional threshold comparison according to the comprehensive anomaly feature vector and the anomaly discrimination threshold to obtain the solder joint anomaly type identifier. Writing the solder joint anomaly type identifier and the corresponding anomaly feature data into the anomaly detection result database, and establishing a solder joint image index in the anomaly detection result database to associatively store the corresponding first solder joint contour data.

[0014] Further, if there is an abnormality, an image segmentation method is used to identify the abnormal area in the image, and the area and geometric features of the abnormal area are calculated by a pixel statistics method to obtain the size and deformation degree of the solder joint, including: extracting the solder joint image to be processed from the image database, performing region annotation on the solder joint image according to the preset abnormal judgment result to obtain the first abnormal marked image, and using the watershed segmentation algorithm to extract the boundary of the first abnormal marked image to obtain the first region boundary data. Curve fitting is performed on the first region boundary data to obtain the first abnormal contour curve, the chain code encoding method is used to extract the sequence of contour feature points to obtain the first feature point data, and the perimeter and area of the abnormal area are calculated according to the first feature point data to obtain the first geometric parameter data. The centroid coordinates and the main axis direction of the abnormal area are calculated according to the first geometric parameter data to obtain the first shape feature data, and the gray scale statistics of the abnormal area in the first abnormal marked image are performed to obtain the first gray scale histogram data. Multilevel threshold segmentation is performed on the abnormal area according to the first gray scale histogram data to obtain the second abnormal area image, and the region growing method is used to mark the segmented area to obtain the first density distribution map. The centroid position and boundary point coordinates of the region are extracted from the first density distribution map to obtain the second shape feature data, and the major axis and minor axis parameters of the abnormal area are calculated by the minimum circumscribed rectangle method to obtain the first size data. The symmetry parameter of the second shape feature data is calculated by the region projection method to obtain the first deformation parameter, and the first geometric parameter data, the first size data and the first deformation parameter are combined to form an abnormal area feature vector. An index relationship is established between the abnormal area feature vector and the corresponding first abnormal marked image, and written into the abnormal feature database to complete the storage of the abnormal area features.

[0015] Further, based on the preset welding torch path parameters and combined with the extracted coordinates of the center line of the solder joint, calculate the deviation value between the center line of the solder joint and the preset path to obtain the solder joint position offset. Then, combine the size and deformation degree of the solder joint to correct the welding instruction, including: reading the preset welding path coordinate sequence from the path parameter database, generating the first path curve by using the cubic spline interpolation method, performing curve fitting on the coordinate data of the center line of the solder joint to obtain the second path curve, and calculating the distance between corresponding points of the two curves by the least square method to obtain the first deviation data. Perform interval sampling on the first deviation data to calculate the local deviation mean to obtain the first offset data, use the sliding window method to smooth the first offset data to obtain the second offset data, and perform segmented marking on the second offset data according to the preset deviation threshold to obtain the path offset interval data. Read the solder joint size abnormal data and deformation degree data from the quality detection database, calculate the coordinate correction coefficient according to the preset compensation rule to obtain the first correction parameter, and perform normalization processing on the first correction parameter to obtain the second correction parameter. Combine the path offset interval data and the second correction parameter to calculate the path compensation value to obtain the first path compensation data, and use the segmented interpolation method to generate the path compensation curve to obtain the second path compensation data. Read the standard welding parameters from the welding process database to obtain the first instruction parameter, and generate the position correction amount according to the second path compensation data to obtain the second instruction parameter. Combine the first instruction parameter and the second instruction parameter to obtain the corrected instruction sequence, and use the kinematic constraint verification method to perform feasibility verification on the corrected instruction sequence to obtain the qualified instruction sequence. Write the qualified instruction sequence into the welding control database, establish an instruction index record, and associate and store the corresponding path compensation data and correction parameters.

[0016] Further, obtaining the corrected welding instruction data and extracting the energy, focal length, and torch angle parameters in the instruction as the reference values for welding instruction analysis includes: reading the corrected welding instruction sequence from the instruction database, obtaining the instruction timestamp and instruction type identifier using the instruction parsing method to get the first instruction data, classifying and filtering the first instruction data according to the instruction type identifier to obtain the welding parameter instruction set. Sorting the welding parameter instruction set according to the timestamp to get the first parameter time series data, separating the energy parameter, focal length parameter, and angle parameter through the parameter extraction algorithm to get the first parameter array. Resampling the first parameter array using a fixed time window to get the second parameter array, and normalizing the second parameter array according to the parameter unit and range to get the first standard parameter sequence. Calculating the moving mean and standard deviation of the first standard parameter sequence to get the first statistical parameter data, and extracting the parameter fluctuation amplitude and frequency characteristics using the wavelet decomposition method to get the first fluctuation characteristic data. Calculating the parameter stable interval according to the first statistical parameter data and the first fluctuation characteristic data to get the first interval data, and segmenting the first interval data through the adaptive threshold method to get the second interval data. Reading the standard parameter range from the process parameter database to get the parameter limit data, and calculating the characteristic of each parameter interval according to the second interval data to get the first feature vector. Performing weighted fusion on the first feature vector and the parameter limit data to get the reference parameter vector, reconstructing the reference parameter vector according to the instruction format to get the welding instruction reference value. Writing the welding instruction reference value into the instruction reference database, establishing a reference value index record, and associating and storing the corresponding statistical parameter data and fluctuation characteristic data.

[0017] Further, judging whether the welding energy and focal length match the reference values analyzed by the welding instruction by analyzing the energy distribution curve during the welding process and combining the focal length data and the change of the welding torch angle collected in real time includes: reading the real-time energy sampling data during the welding process from the sensor database to obtain the first raw data, verifying the validity of the first raw data according to the sampling timestamp to obtain the first valid data, and performing smoothing processing on the first valid data by using the moving average method to obtain the first curve data. Collecting the focal length data and the angle data by using the welding equipment sensor to obtain the second raw data, removing the outliers from the second raw data by using the data screening rule to obtain the second valid data, and performing piecewise least squares fitting on the second valid data to obtain the second curve data. Reading the welding reference parameters from the reference value database to obtain the reference parameter vector, normalizing the three groups of parameters of energy, focal length, and angle in the reference parameter vector to obtain the standard reference data, and calculating the parameter fluctuation range according to the standard reference data to obtain the parameter threshold data. Performing time series alignment on the first curve data and the second curve data by using a fixed time window to obtain the first parameter sequence, and unifying the sampling frequency by using the data resampling method to obtain the second parameter sequence. Calculating the curve eigenvalue of the second parameter sequence to obtain the characteristic parameter data, and calculating the matching degree between the characteristic parameter data and the standard reference data by using the curve correlation algorithm to obtain the parameter similarity data. Performing hierarchical determination on the parameter similarity data according to the parameter threshold data to obtain the parameter matching result, reading the determination rule from the process database to obtain the rule data, and obtaining the parameter matching identifier by using the multi-dimensional threshold judgment method. Writing the parameter matching identifier and the corresponding parameter similarity data into the result database, establishing a result index record, and associating and storing the corresponding characteristic parameter data and the determination rule data.

[0018] Further, if the welding energy and focal length do not match the reference values analyzed in the welding instruction, adjust the control parameters of the welding equipment according to the deviation value, and recalculate the correction amount of the welding torch path, including: reading the welding parameter deviation data from the parameter analysis database, calculating the energy deviation value and the focal length deviation value using an adaptive mapping algorithm to obtain the first deviation data, generating a compensation coefficient according to the welding process rules to obtain the first compensation parameter, and performing numerical optimization on the first compensation parameter to obtain the corrected control parameter. Collect the welding position image using a binocular vision sensor to obtain the first position data, read the molten pool standard position parameters from the process database to obtain the reference position data, and calculate the position difference through an image registration method to obtain the first offset data. Convert the first offset data to the welding torch coordinate system using a coordinate transformation matrix to obtain the second offset data, calculate the welding torch position compensation coefficient according to the corrected control parameter to obtain the first compensation vector, and perform normalization processing on the first compensation vector to obtain the welding torch position compensation amount. Calculate the path adjustment parameter according to the welding torch position compensation amount to obtain the third offset data, perform smoothing processing on the third offset data using the cubic spline interpolation method to obtain the first path correction amount, and perform coordinate calibration on the first path correction amount to obtain the second path correction amount. Map the second path correction amount to the joint space using the Jacobian matrix method to obtain the first path compensation data, and perform trajectory optimization according to the kinematic constraints of the robotic arm to obtain the second path compensation data. Perform reachability verification on the second path compensation data to obtain the path verification result, verify the compensated trajectory using a motion simulation method to obtain the trajectory verification data, and generate a compensation instruction according to the verification result to obtain the qualified path data. Write the qualified path data into the control parameter database, establish a path index record, and associatively store the corresponding deviation data and compensation parameter data.

[0019] Further, input the adjusted control parameters into the welding equipment control system, and monitor the energy and focal length changes during the welding process in real time. By continuously collecting welding process images, verify the stability of the solder joint contour and center line, and determine whether the welding quality meets the preset standard, including: write the corrected welding control parameters into the welding equipment controller to obtain the first control instruction, collect the energy data and focal length data during the welding process from the equipment sensor to obtain the first monitoring data, generate the second monitoring data through data validity verification, and calculate the parameter change trend using the moving average method to obtain the first trend data. Continuously collect the solder joint forming images using a high-speed industrial camera to obtain the first image sequence, screen the effective images according to the image quality evaluation index to obtain the second image sequence, and enhance the contrast of the second image sequence using the histogram equalization method to obtain the third image sequence. Use the watershed segmentation algorithm to extract the solder joint contour data from the third image sequence to obtain the first contour sequence, calculate the solder joint center line data according to the least squares method to obtain the first center line sequence, and calculate the contour shape parameter and center line offset parameter using the continuity detection method to obtain the first stability data. Perform time-domain analysis on the first trend data to obtain the parameter fluctuation index, perform statistical analysis on the first stability data to obtain the shape stability index, and combine the two sets of indexes to generate the quality feature vector to obtain the first feature data. Read the preset quality parameters from the quality standard database to obtain the standard feature data, calculate the feature matching degree through the multi-dimensional threshold comparison method to obtain the first matching data, and determine the welding quality according to the matching rule to obtain the quality determination result. Generate parameter adjustment suggestions according to the quality determination result to obtain the first adjustment data, and calculate the control parameter correction amount using the feedback compensation algorithm to obtain the second adjustment data. Write the quality determination result and the second adjustment data into the quality monitoring database, establish a quality record index, and associate and store the corresponding monitoring data and feature data.

[0020] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0021] The present invention discloses a control method for a welded coupling device. Collect welding process images through a high-speed camera, and use edge detection and morphological processing to extract the solder joint contour and center line. Judge the solder joint abnormality according to the preset threshold, use image segmentation to identify the abnormal area and calculate its features. Combine the welding torch path parameters, calculate the solder joint position offset amount and correct the welding instruction. By analyzing the energy distribution curve, focal length and welding torch angle changes, judge whether they match the reference values. If they do not match, adjust the control parameters and recalculate the correction amount. Finally, monitor the welding process in real time, verify the stability of the solder joint contour and center line, and judge the welding quality. The present invention realizes the automatic monitoring, abnormal detection and real-time correction of the welding process, and improves the welding quality and efficiency. Description of the Drawings

[0022] Figure 1It is a flowchart of a control method for a welded coupling device of the present invention.

[0023] Figure 2 It is a schematic diagram of a control method for a welded coupling device of the present invention. Specific embodiments

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and detailedly described in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention.

[0025] Such as Figure 1-2 , a control method for a welded coupling device in this embodiment may specifically include:

[0026] S101. Collect a continuous image sequence during the execution of the welding instruction through a high-speed imaging device, extract the solder joint contour features by using edge detection technology, and generate the solder joint center line data in combination with a morphological algorithm, specifically including collecting the image sequence of the welding head movement trajectory at a preset frequency, obtaining the initial contour data by using gray-scale processing and binarization methods, generating a complete contour line after noise reduction and edge detection, and then generating the center line data through a thinning algorithm.

[0027] In the embodiment of the present invention, after the welding device is started, the high-speed imaging device collects the welding process images at a preset frequency to capture the key features of the solder joint formation. The collected image sequence is processed through multiple steps to generate the solder joint center line. The specific hardware model is not limited here, and a suitable imaging device can be selected according to the actual scenario.

[0028] S1011. Collect the initial image sequence of the welding head movement trajectory through an industrial-grade high-speed imaging device at a preset acquisition frequency, extract the gray-scale image data from the initial image sequence, determine the optimal binarization threshold by using gray-scale histogram analysis, and perform binarization processing on the gray-scale image data to generate the initial binarized image data.

[0029] In the embodiment of the present invention, the acquisition frequency can be set to 1000 to 5000 frames per second to ensure recording of the molten pool dynamics and splash details. The gray-scale image is stored in an 8-bit format, and the gray-scale value range is 0 to 255. Through histogram statistics, the threshold with the largest between-class variance is selected, usually between 120 and 180, to distinguish the molten pool from the background area.

[0030] S1012. Perform noise reduction processing on the initial binarized image data by using a Gaussian filter to generate the noise-reduced binarized image data, perform edge detection on the noise-reduced binarized image data through a Sobel operator to generate the initial edge contour data, calculate the pixel point density distribution according to the initial edge contour data, and perform break point connection processing to generate the complete edge contour data.

[0031] In the embodiment of the present invention, the size of the Gaussian filter kernel can be selected as 5×5 or 7×7, and the standard deviation is set to 1.5 to 2.0 to smooth the noise and retain the edge features. The Sobel operator calculates the gradient field through horizontal and vertical templates, and the gradient threshold is set to 30 to mark the edge points. When density clustering connects the break points, the clustering radius is 10 pixels and the minimum density is 5 pixel points to ensure the continuity of the contour.

[0032] S1013. Perform morphological closing operation on the complete edge contour data to generate closed contour line data, use the morphological thinning algorithm based on distance transformation to extract the skeleton of the closed contour line data to generate initial skeleton data, perform node continuity processing on the initial skeleton data through the cubic spline interpolation algorithm, and generate the final solder joint center line data by combining curvature adjustment.

[0033] In the embodiment of the present invention, the closing operation uses a circular structural element with a radius of 3 pixels to fill the gaps. The distance transformation threshold is set to one-third of the contour width to extract the skeleton. The control point spacing of the cubic spline interpolation is 20 pixels, and the curvature threshold is 0.15 for smoothing processing. Finally, the center line width is thinned to 1 pixel through the erosion operator, and its connectivity is verified, and the maximum break point gap does not exceed 2 pixels.

[0034] In the embodiment of the present invention, the movement speed of the welding head is usually 50 to 200 millimeters per second, and the displacement between image frames is less than 3 pixels to ensure the sequence continuity. The accuracy of the extracted center line reaches the sub-pixel level, and the average width is normalized to 1 pixel to meet the requirements of solder joint trajectory analysis. The Sobel operator can also provide the gradient direction for judging the weld width distribution and improving the comprehensiveness of the analysis.

[0035] It can be understood that the specific values of the parameters in this step can be adjusted by technicians according to the actual application scenario to optimize the image processing effect. Through multi-level image processing, the accuracy of the solder joint contour and center line data is ensured, laying a foundation for subsequent quality judgment.

[0036] S102. According to the preset solder joint size and weld width standard value, compare the solder joint contour and center line data extracted through image processing to determine whether there are size abnormalities or weld width abnormalities in the solder joints. Specifically, read the reference data from the standard parameter library, extract the actual solder joint features and calculate the deviation, generate the abnormality type identifier by combining multi-dimensional feature analysis, and store the results to support subsequent optimization.

[0037] In the embodiment of the present invention, this step aims to identify potential defects of the solder joints by comparing the standard and actual parameters. The standard parameters are generated based on a large number of sample statistics, and the actual features are extracted through image analysis. After comparing the two, a multi-dimensional feature vector is formed for abnormality determination. This step does not limit the performance of specific computing devices, and appropriate hardware can be selected according to actual needs.

[0038] S1021. In the embodiment of the present invention, the reference data of the solder joint size is read from the preset standard parameter library, and the collected solder joint contour data is processed by using the contour feature extraction technology to generate the actual size parameters. The deviation value between the actual size and the standard size is calculated by the curve matching method and used as the preliminary feature data of size abnormality.

[0039] In the embodiment of the present invention, the standard parameter library stores the reference values of the solder joint diameter and depth. The diameter range is usually 3.0 mm to 5.0 mm, and the depth is 1.0 mm to 2.0 mm. The contour data is represented by 512 two-dimensional coordinate points. The actual contour and the standard contour are matched by the least squares method to calculate the deviation value. If the deviation exceeds 0.3 mm, it is regarded as the basis for the preliminary judgment of size abnormality. The curve matching process ensures the accuracy through iterative optimization, and the deviation calculation considers the overall distribution of the contour rather than a single local abnormality.

[0040] S1022. In the embodiment of the present invention, the region growing algorithm is used to extract the molten width feature region from the solder joint contour data to generate the initial molten width data. The distance transformation method is used to perform statistical analysis on the initial molten width data to calculate the average molten width. The reference range of the molten width is read from the standard parameter library, and the feature data of molten width abnormality is generated by comparing with the average molten width.

[0041] The region growing algorithm uses the center of the solder joint as the seed point and expands according to the gray similarity. The threshold is set to 20 to ensure the accurate segmentation of the molten width region. The distance transformation samples 32 points along the circumferential direction to calculate the average molten width. The normal range is 2.5 mm to 3.5 mm, and the standard deviation is less than 0.2 mm. If the average value exceeds the range or the deviation is too large, the feature data of molten width abnormality is generated. The statistical reliability is improved by uniform sampling to avoid local noise interference.

[0042] S1023. In the embodiment of the present invention, the equidistant sampling method is used to extract the point sequence from the solder joint contour data to generate the sampling point data. The local curvature distribution of the sampling point data is calculated by the three-point method and the roundness parameter is fitted to generate the shape abnormality feature data. At the same time, the curvature extreme points are extracted and the periodic distribution feature is calculated to generate the edge abnormality feature data.

[0043] The equidistant sampling interval is set to 0.1 mm, and the curvature is calculated with a distance of 0.3 mm between every three adjacent points to reflect the smoothness of the contour. The circle is fitted by the least squares method, and the root mean square deviation is calculated. The normal roundness deviation is less than 0.15 mm. The curvature extreme value is extracted through a 1.0 mm sliding window, and the periodic distribution is analyzed by the fast Fourier transform. The main frequency of the normal solder joint is lower than 8 waves per circle, and the amplitude ratio is less than 0.3. High-frequency fluctuations indicate irregular edges or cracks, and the two features together constitute the basis for abnormality analysis.

[0044] In the embodiment of the present invention, the dimensional anomaly feature data, the weld width anomaly feature data, the shape anomaly feature data, and the edge anomaly feature data are integrated to form a comprehensive solder joint anomaly feature vector. The multi-dimensional anomaly discrimination threshold is read from the standard parameter library, and the solder joint anomaly type is judged and an identifier is generated through the multi-dimensional threshold comparison method. At the same time, the anomaly type identifier and related feature data are stored in the detection result database and associated with the original contour data.

[0045] The comprehensive feature vector includes six-dimensional data: maximum size deviation, average weld width deviation, standard deviation of weld width, roundness deviation, number of cycles, and maximum ripple amplitude. The multi-dimensional threshold comparison checks item by item. If any dimension exceeds the range, the corresponding anomaly type is determined, such as size deviation, insufficient weld width, irregular shape, or edge ripple. The detection result database records the feature values and image indexes, and the response time is controlled within 50 milliseconds, supporting real-time monitoring and traceability. The storage process optimizes the retrieval efficiency through hash indexing to ensure data correlation.

[0046] In the embodiment of the present invention, this step improves the detection accuracy through multi-dimensional analysis. The diversity of anomaly types reflects different causes of solder joint defects, such as process parameter imbalance or equipment jitter. The automated design of feature extraction and threshold comparison reduces manual intervention and ensures the objectivity of the results. The subsequent process optimization can directly call the database data to analyze the anomaly distribution law to adjust the welding parameters.

[0047] For the actual application scenario, the solder joint contour analysis can also be extended to the detection of the dynamic characteristics of the molten pool. The accuracy can be further improved by increasing the sampling density or adjusting the algorithm parameters. For example, the number of sampling points is increased to 1024 or the region growing threshold is optimized to 15 to meet the requirements of high-speed welding or complex weld seams scenarios. This expandable design provides flexibility for different welding environments. The high sampling density can capture small deformations, and the threshold optimization enhances the anti-interference ability. The adjusted method significantly improves the sensitivity to anomaly features on the premise of keeping the core logic unchanged and is applicable to high-precision fields such as electronic devices or medical equipment.

[0048] S103. If an anomaly is detected in the solder joint, the anomaly area is identified through image segmentation technology and its boundary is extracted. The area and geometric characteristics of the anomaly area are calculated using the pixel statistics method to determine the size and deformation degree of the solder joint. Specifically, the feature vector of the anomaly area is generated through region annotation, watershed segmentation, and multi-dimensional feature analysis and stored in the database.

[0049] In the embodiment of the present invention, this step aims to conduct a refined analysis of the abnormal solder joint to quantify its defect degree. The identification of the anomaly area is based on image segmentation technology, and the extraction of geometric features is completed through multi-dimensional calculations. The results provide data basis for subsequent process adjustment.

[0050] Obtain the solder joint image to be analyzed from the image database, perform region annotation on it according to the preset abnormal judgment result to generate an initial abnormal marked image, use the watershed segmentation algorithm to extract the boundary of the initial abnormal marked image to generate abnormal region boundary data, and calculate the boundary features through curve fitting and chain code encoding methods to obtain geometric parameters.

[0051] The initial abnormal marked image is generated through a preset abnormal type identifier. The watershed segmentation algorithm constructs a watershed line based on the gray gradient. The gradient threshold is set to 50 to 80 to distinguish high-gradient regions such as cracks, and the gradient of the normal region is lower than 20. When the image resolution is 40 pixels per millimeter, defects above 0.1 millimeter can be detected. After the boundary data is smoothed by curve fitting, 8-direction chain code encoding is used to track the contour point by point to calculate the perimeter and area. The frequency of change of the chain code direction of normal solder joints is 0.2 to 0.3 per perimeter, while that of the abnormal region is up to more than 0.5, reflecting the difference in contour complexity.

[0052] S1031. In the embodiment of the present invention, calculate the centroid coordinates and the main axis direction according to the abnormal region boundary data to generate preliminary shape feature data, perform gray scale statistics on the initial abnormal marked image to generate gray scale histogram data, and extract the density distribution features of the abnormal region through multi-level threshold segmentation and region growing methods for further refined analysis.

[0053] The centroid coordinates are calculated through the average value of the boundary points, and the main axis direction is determined by principal component analysis to form preliminary shape features. The gray scale histogram statistically analyzes the pixel distribution of the abnormal region. The normal solder joints show a single-peak distribution, with the peak value between 180 and 220, and secondary peaks appear in the defect regions such as slag inclusions between 100 and 140. Three thresholds of 120, 160, and 200 are set for multi-level threshold segmentation to divide four gray scale levels. Region growing uses the gray centroid as the seed, the gray difference threshold is 15, and the 8-neighborhood is used to determine spatial adjacency to generate a density distribution map to distinguish the characteristics of cracks with an aspect ratio greater than 5 or pores approximately circular.

[0054] S1032. In the embodiment of the present invention, use the minimum circumscribed rectangle method to calculate the major axis and minor axis parameters of the abnormal region from the density distribution map to generate size feature data, analyze the symmetry of the preliminary shape features through the region projection method to generate deformation parameters, and integrate the geometric parameters, size features, and deformation parameters to form an abnormal region feature vector and store it associated.

[0055] The minimum bounding rectangle is calculated through the convex hull of boundary points. The major axis direction reflects the crack propagation trend. The aspect ratio of normal solder joints ranges from 0.8 to 1.2, while that of cracks reaches 3 to 5. Area projections generate curves along the horizontal and vertical directions. Normal solder joints exhibit a bell-shaped distribution, while abnormal areas are skewed or bimodal. The deformation parameters quantify symmetry accordingly. The eigenvector consists of six-dimensional components including area ratio, perimeter ratio, shape factor, mean gray value, standard deviation, and projection symmetry degree, which is associated with the original image through a hash index and stored in a database with the retrieval time controlled within 5 milliseconds.

[0056] In the embodiment of the present invention, the high-precision boundary extraction by watershed segmentation is applicable to complex defect recognition, and chain code encoding and gray-scale statistics enhance the sensitivity to minute deformations. The multi-dimensional design of the eigenvector not only quantifies the defect type, for example, adjusting the welding torch angle or energy input based on the crack aspect ratio.

[0057] In the embodiment of the present invention, according to the requirements of the production site, the analysis accuracy can be further improved by increasing the gray-scale threshold levels or optimizing the projection direction. For example, increasing the threshold to five levels or introducing oblique projection to adapt to the detection requirements of diverse welding defects, thereby improving the robustness and applicability of the method. The flexible design enables the method to be extended to different welding scenarios. The five-level threshold can refine the gray-scale levels, and the oblique projection captures non-orthogonal deformation features. By adjusting parameters, the method can more accurately identify various defects such as pores, slag inclusions, or cracks while maintaining the core logic, enhancing the reliability of real-time monitoring.

[0058] S104. Calculate the deviation value between the preset welding torch path parameters and the extracted solder joint centerline coordinates to determine the position offset, and generate a corrected welding instruction by combining the solder joint size and deformation data. Specifically, path optimization is achieved through steps such as path fitting, deviation smoothing, and kinematic verification and stored in the control database.

[0059] In the embodiment of the present invention, this step analyzes the difference between the actual solder joint position and the preset path, dynamically adjusts the welding torch movement trajectory, and ensures welding accuracy. The path parameters are preset based on process requirements, and deviation calculation and correction rely on real-time data analysis to finally generate an executable instruction.

[0060] S1041. In the embodiment of the present invention, extract the welding path coordinate sequence from the preset path database and generate a smooth first path curve using the cubic spline interpolation method. Curve fitting is performed on the solder joint centerline coordinates to generate a second path curve, and the distance between corresponding points on the two curves is calculated by the least squares method to generate initial deviation data.

[0061] The path coordinate sequence is stored as discrete points. The spacing between the straight weld points is about 2 mm, and the curve welds are encrypted to 1 mm. Cubic spline interpolation adopts natural boundary conditions, with the first-order derivative of the endpoints being zero to ensure smooth movement of the welding torch. During fitting, the number of points on the center line of the weld points is sampled at 5 times the path length. For example, 500 interpolation points are generated for a 100-mm weld. The least squares method calculates the deviation in the vertical direction, and the average value of the normal deviation is controlled within 0.2 mm, reflecting the matching degree between the weld points and the path.

[0062] S1042. In the embodiment of the present invention, interval sampling is performed on the initial deviation data to calculate the local deviation mean value to generate preliminary offset data. The preliminary offset data is smoothed by the sliding window method to generate smoothed offset data, and the smoothed offset data is segmented and marked according to a preset deviation threshold to determine the path offset interval.

[0063] Interval sampling divides the path into several segments, and the average deviation value is taken for each segment. The sliding window is set to a width of 15 points, and the weight decreases with the distance from the center point. For example, Gaussian weighted smoothing is used to affect the noise. If the offset of a certain segment exceeds 0.5 mm, it is marked as an offset interval. The segmented processing not only retains local features but also avoids overcorrection caused by excessive global deviation, ensuring the pertinence of the adjustment.

[0064] In the embodiment of the present invention, the abnormal solder joint size and deformation data are read from the quality inspection database. The coordinate correction coefficient is calculated according to the deviation direction and degree and normalized to generate the final correction parameter. The path compensation value is calculated in combination with the offset interval data and a smooth path compensation curve is generated by segmented interpolation.

[0065] The abnormal size includes a diameter deviation exceeding 15% or uneven width. The deformation direction determines the compensation direction. For example, an elliptical solder joint is compensated along the short axis. The correction coefficient is positively correlated with the deviation and does not exceed 30% of the standard diameter at most. The segmented interpolation extends 50 mm at both ends of the offset interval for transition, and a cubic polynomial is used to ensure first-order continuity, generating a smooth compensation curve to optimize the welding torch trajectory.

[0066] In the embodiment of the present invention, the position correction amount is generated according to the path compensation curve and combined with the standard welding parameters to generate a correction instruction sequence. The feasibility of the instruction sequence is checked by the kinematic constraint verification method to generate the final qualified instruction sequence, and the qualified instruction sequence and related parameters are stored in the welding control database to support subsequent traceability.

[0067] The standard parameters include a speed of 5 to 15 millimeters per second, a voltage of 18 to 24 volts, and a current of 80 to 120 amperes. When the position is corrected, the speed is adjusted inversely. Kinematic verification ensures that the acceleration of the welding torch is less than 1000 millimeters per square second, the angular velocity is less than 90 degrees per second, and the change in the attitude angle is less than 3 degrees. Qualified instructions are stored with a sampling period of 4 milliseconds, including position, attitude, and process data, and the compensation parameters and the original path are associated through weld number indexing.

[0068] S1043. In the embodiment of the present invention, for complex weld scenarios, the correction accuracy can be improved by increasing the interpolation point density or adjusting the window width. For example, the sampling points can be increased to 8 times the path length or the window can be extended to 20 points to meet the high-precision welding requirements with a curvature radius less than 10 millimeters, thereby improving the flexibility and robustness of path adjustment. The optimized design enhances the adaptability of the method to irregular welds. High-density sampling captures small deviations, and window extension smooths complex fluctuations. Through these adjustments, the correction instructions can more accurately match the actual solder joint state, improving the welding quality and the operating efficiency of the equipment.

[0069] S105. Obtain the corrected welding instruction data and extract the energy, focal length, and welding torch angle parameters therein as the analysis reference values. Specifically, generate a standardized reference parameter vector through instruction parsing, resampling, and feature analysis and store it in the database for subsequent use.

[0070] In the embodiment of the present invention, this step aims to extract key parameters from complex instruction sequences and establish a stable reference value system. The extraction process combines time series analysis and fluctuation feature judgment to ensure that the parameters reflect the actual welding state. This method has no special requirements for the instruction generation device and can be flexibly adapted to different control systems.

[0071] S1051. In the embodiment of the present invention, read the corrected welding instruction sequence from the instruction database and generate initial instruction data by parsing to obtain the timestamp and type identifier. Filter out the instructions related to welding parameters according to the type identifier and sort them by timestamp to generate time-series parameter data. Then, use the parameter extraction algorithm to separate the energy, focal length, and angle parameters to form a preliminary parameter array.

[0072] The instruction sequence is recorded with millisecond-level timestamps and includes three types of instructions: motion, welding, and process. Among them, the welding parameter instructions account for about one-third. The parsing process identifies the type identifier of each frame of instructions and filters out the instructions containing voltage, current, focal length, and angle. After sorting, about 2000 frames of data are generated for a 200-millimeter weld, with a sampling period of 10 milliseconds. The parameter extraction algorithm decomposes the instruction fields frame by frame to generate an array containing three types of parameters for subsequent processing.

[0073] In the embodiment of the present invention, the preliminary parameter array is resampled using a fixed time window to generate resampled parameter data, which is normalized according to the unit and range of each parameter to generate a standardized parameter sequence. The stability of the parameters is judged by calculating statistical features and analyzing the fluctuation characteristics through wavelet decomposition, and a reference parameter vector is generated.

[0074] The resampling window width is set to 50 milliseconds, and the overlap rate is 50% to ensure data continuity. The voltage range of 18 to 24 volts, the focal length of 12 to 15 millimeters, and the angle of plus or minus 15 degrees are respectively normalized to 0 to 1. The sliding mean and standard deviation are calculated to obtain the steady state level, and three-layer wavelet decomposition is used to extract high, medium, and low frequency fluctuations. The high frequency amplitude limit is 5%, and the medium frequency limit is 10%. If the voltage deviation exceeds 0.5 volts or the focal length fluctuation exceeds 0.5 millimeters, it is marked as unstable. The final vector is generated based on the principle of minimum fluctuation and reconstructed into an instruction format.

[0075] S1052. In the embodiment of the present invention, the generated reference parameter vector is weighted and fused with the process parameter limit values to generate the final welding instruction reference value, which is written into the instruction reference database. At the same time, an index is established to associatively store the corresponding statistical parameters and fluctuation characteristic data to support fast retrieval and process optimization.

[0076] During fusion, the mean deviation weight is 0.4, the standard deviation is 0.3, and the fluctuation characteristic is 0.3. Intervals with smaller fluctuations are preferentially selected. The reference value record is 32 bytes, and the characteristic data is 128 bytes. The hash index ensures efficient retrieval. The database supports retrieval by weld type or time, adapts to the parameter configuration requirements of batch tasks, and improves welding consistency.

[0077] In the embodiment of the present invention, for dynamic welding scenarios, the window width can be adjusted to 30 milliseconds or the wavelet decomposition level can be increased to 5 layers to further improve the accuracy of parameter extraction and anti-interference ability, so as to meet the parameter analysis requirements under complex conditions such as arc fluctuations or workpiece deformations.

[0078] This optimized design enhances the adaptability of the method to non-steady processes. The short window captures rapid changes, and the multi-layer decomposition refines the frequency characteristics. Through these adjustments, the reference value can more accurately reflect the process state and provide more reliable data support for real-time control and quality improvement.

[0079] S106. By analyzing the energy distribution curve during the welding process and combining the real-time collected focal length data and the change of the welding torch angle, it is judged whether these parameters match the welding instruction reference value. Specifically, a parameter matching identifier is generated through data verification, curve fitting, and correlation analysis, and the result is stored to support process monitoring.

[0080] In the embodiments of the present invention, this step aims to evaluate the execution effect of welding parameters in real time to ensure consistency with the expected benchmark. The energy distribution reflects the welding stability, and the focal length and angle affect the weld formation. The quality control accuracy is improved through multi-dimensional comparative analysis. This method is applicable to welding systems with various sensor configurations.

[0081] S1061. In the embodiments of the present invention, the energy sampling data during the welding process is extracted from the sensor database and verified for validity according to the time stamp to generate valid energy data. The sliding average method is used to smooth the valid energy data to generate an energy distribution curve. At the same time, the focal length and angle data are collected by the sensor, and after removing the outliers, the piecewise least squares fitting is used to generate the focal length and angle curves.

[0082] The energy data is sampled at 1000 Hz, with a range of 800 to 3000 W. During verification, the numerical range and continuity are checked. If there are more than 5 consecutive outliers, that section is removed. The sliding average uses a 21-point window to smooth the high-frequency noise. The focal length data is sampled at 200 Hz with a range of 10 to 20 mm, and the angle data is sampled at 500 Hz with a range of plus or minus 30 degrees. The outliers are removed according to 3 times the standard deviation. The fitting is performed in segments of 50 ms, and a quadratic polynomial is used to ensure the curve continuity and reflect the parameter trend.

[0083] In the embodiments of the present invention, the welding reference parameters are read from the reference value database, and their energy, focal length, and angle values are normalized to generate standard reference data. Through a fixed time window, the energy distribution curve and the focal length and angle curves are aligned in time series and resampled to generate a parameter sequence with a unified frequency. Then, the curve characteristic values are calculated and correlated with the standard reference data to evaluate the matching degree.

[0084] The reference value normalization uses the maximum-minimum method to map to 0 to 1. The energy reference is 85% to 95% of the rated power, and the focal length reference, such as 15 mm, matches the 5-mm steel plate. The time series alignment window is 100 ms, with 100 points for energy, 20 points for focal length, and 50 points for angle, and resampled to 100 Hz. The characteristic values include the mean, standard deviation, peak factor, and waveform factor. The cosine similarity is used to calculate the matching degree. A value higher than 0.95 indicates a high match, 0.9 to 0.95 indicates a medium match, and a value lower than 0.9 indicates a mismatch.

[0085] S1062. In the embodiments of the present invention, the threshold data is generated according to the process parameter fluctuation range, and multi-dimensional threshold judgment is performed in combination with the matching degree result to generate a parameter matching identifier. The matching identifier and the similarity data are written into the result database, and the characteristic parameters and the determination rules are associated and stored for subsequent traceability and optimization.

[0086] The threshold is set for energy fluctuation at plus or minus 5%, focal length at plus or minus 0.5 mm, and angle at plus or minus 2 degrees. Weighted voting is used to judge the match, with an energy weight of 0.5, a focal length weight of 0.3, and an angle weight of 0.2. The result database records a 16-byte timestamp, a 32-byte parameter value, and a 16-byte identifier, and is packed every 5 minutes to support fast retrieval. The match analysis improves the standard deviation of the weld reinforcement by 40% and the width consistency by 35%.

[0087] In the embodiment of the present invention, for a high-dynamic welding scenario, the real-time analysis can be enhanced by shortening the window to 50 milliseconds or increasing the sampling frequency to 2000 Hz, so as to more accurately capture the influence of energy mutation or angle fine-tuning, further optimize the reliability of parameter matching and process adaptability, and improve the response ability to transient changes. High-frequency sampling refines the data granularity, and the short window focuses on local features, ensuring that the parameter matching results are closer to the actual working conditions and providing technical support for high-quality welds.

[0088] S107. If it is detected that the welding energy and focal length do not match the welding instruction reference values, the equipment control parameters are adjusted according to the deviation value, and the welding torch path correction amount is recalculated. Specifically, the deviation is calculated through adaptive mapping, the binocular vision positioning offset and kinematic mapping are used to generate optimized path data and store it in the database to improve the welding accuracy. By adjusting the parameters and path in real-time feedback, the influence of the deviation on the weld quality is solved. The adaptive algorithm ensures that the adjustment matches the degree of deviation, and the visual positioning and kinematic optimization ensure the accuracy and executability of the path correction. This method is applicable to various welding processes and robot systems.

[0089] S1071. In the embodiment of the present invention, the welding parameter deviation data is extracted from the parameter analysis database, and the adaptive mapping algorithm is used to calculate the deviation values of the energy and focal length to generate initial deviation data. The compensation coefficient is generated according to the process rules and optimized to obtain the corrected control parameters. At the same time, the binocular vision sensor is used to collect the welding position image, and the position difference is calculated through image registration to generate the initial offset data.

[0090] The deviation data includes the difference between the real-time values of the energy and focal length and the reference values. The adaptive mapping linearly compensates within plus or minus 5% of the deviation, and non-linearly adjusts when it exceeds. For example, a 50-watt deviation corresponds to a coefficient of 0.8 at a power of 2000 watts, and 100 watts corresponds to 1.2. The binocular vision resolution is 1280×1024, the field of view is 40×32 mm, the registration accuracy of the phase correlation method is 0.1 mm, and the deviation vector between the molten pool center and the reference point is calculated to provide a basis for subsequent correction.

[0091] In the embodiment of the present invention, the initial offset data is converted into the welding torch coordinate system through a coordinate transformation matrix to generate welding torch offset data. The position compensation coefficient is calculated according to the correction control parameter and normalized to generate a compensation vector. Then, the offset data is smoothed by cubic spline interpolation to generate a path correction amount, which is mapped to the joint space through the Jacobian matrix to generate path compensation data.

[0092] The coordinate transformation is implemented by a 4×4 matrix. The origin of the welding torch coordinate system is at the nozzle end, and the z-axis is along the axis. The compensation coefficient is adjusted with the speed, which is 1.2 times the deviation at 10 millimeters per second and increases to 1.5 at 20 millimeters per second. The control point spacing of the cubic spline interpolation is 10 millimeters to ensure a smooth path. The inverse solution of the Jacobian matrix converts the end pose into joint angles to generate compensation data and optimize the path continuity.

[0093] In the embodiment of the present invention, kinematic constraint verification and trajectory optimization are performed on the path compensation data to generate executable path data. The trajectory reachability is verified through motion simulation to generate a verification result, and qualified path data is generated accordingly. Finally, the qualified path data, related deviations, and compensation parameters are written into the control parameter database and indexed for traceability.

[0094] The constraints include joint angles of plus or minus 170 degrees, angular velocity of 120 degrees per second, and acceleration of 200 degrees per square second. Gradient descent is used to optimize smoothness and time. The simulation step size is 0.001 second to check for singular positions and workspace boundaries. The database records a 64-byte path, 32-byte deviation, and compensation data, supports multi-condition retrieval, and improves the accuracy to 0.1 millimeter.

[0095] In the embodiment of the present invention, for complex welds or high-speed welding, the correction accuracy can be further improved by encrypting the registration points to 16 points or shortening the interpolation spacing to 5 millimeters, so as to adapt to the dynamic changes of the molten pool or the requirements of pose adjustment, and ensure the consistency of weld formation and process stability.

[0096] This optimization enhances the adaptability of the method. Encrypted registration improves the positioning accuracy, and short-spacing interpolation refines the path adjustment, enabling the system to handle high-dynamic scenarios and significantly improving the quality consistency.

[0097] S108. Input the adjusted control parameters into the welding equipment control system and monitor the changes in energy and focal length in real time. At the same time, verify the stability of the contour and center line by continuously collecting the solder joint images to determine whether the welding quality meets the preset standards. Specifically, generate a quality determination result through sensor data analysis and image processing and propose adjustment suggestions.

[0098] In the embodiment of the present invention, this step ensures the stability of the welding process and the controllability of the solder joint quality through real-time monitoring and image analysis. Parameter monitoring provides dynamic feedback, and image verification quantifies the forming characteristics. The combination of the two improves the reliability of quality evaluation. This method can be adapted to different welding equipment and workpiece types.

[0099] S1081. In an embodiment of the present invention, the corrected control parameters are written into the controller to generate execution instructions, and energy and focal length data during the welding process are collected through device sensors to generate initial monitoring data. Outliers are removed through validity verification to generate valid monitoring data, and then the moving average method is used to analyze the parameter change trend to generate trend data. At the same time, a high-speed industrial camera is used to continuously collect solder joint forming images, and after screening and enhancement, contour and centerline data are extracted.

[0100] The energy data is sampled at 1000 Hz, the focal length is 200 Hz, and the 3-sigma principle is used for verification to remove outliers. A 25-point sliding window is used for smoothing. Under the 2000-watt process, an energy fluctuation of plus or minus 100 W and a focal length fluctuation of plus or minus 0.5 mm are normal. The camera frame rate is 200 frames per second, the exposure is 0.5 ms. After screening for clarity and contrast, histogram equalization is used to enhance the grayscale to 0 to 255. The watershed algorithm is used to extract the contour based on the gradient, and the least squares method is used to fit the centerline.

[0101] In an embodiment of the present invention, shape parameters and centerline offset parameters are calculated for the contour data to generate stability data, and statistical analysis is performed in combination with the trend data to generate a quality feature vector. Standard feature data is read from the quality standard database, and the matching degree is calculated through multi-dimensional threshold comparison to generate a quality determination result. At the same time, parameter adjustment suggestions are generated according to the result, and the correction amount is calculated to optimize the control parameters.

[0102] The shape parameters include roundness, symmetry, and smoothness. The offset parameters include average and maximum offsets. The fitting error of a normal solder joint is less than 0.2 mm. The trend data extracts the mean, standard deviation, peak factor, and waveform factor. The feature vector is 8-dimensional, and the weighted Euclidean distance is used to measure the matching degree. Proportional-integral control is used for adjustment, with an energy step size of 20 W and a focal length of 0.1 mm, and a compensation period of 100 ms to ensure real-time response.

[0103] S1082. In an embodiment of the present invention, the quality determination result and adjustment data are written into the quality monitoring database, and an index is established to associate and store the monitoring data and feature data for traceability. At the same time, parameter adaptive adjustment is realized through a closed-loop feedback mechanism to improve the welding stability and consistency.

[0104] The database records 64-byte quality data and 128-byte process parameters, and supports sub-millisecond retrieval. The closed-loop feedback significantly reduces the size standard deviation by 40% and the centerline offset by 50%, increases the qualified rate by 15 percentage points, and reduces rework.

[0105] In an embodiment of the present invention, for high-precision or complex workpieces, the quality analysis can be further refined by increasing the camera frame rate to 500 frames per second or increasing the feature vector dimension to 12 dimensions, so as to more accurately capture micro defects and optimize the parameter adjustment effect to meet strict process requirements.

[0106] This optimization improves the ability to detect subtle changes. High frame rates reveal dynamic details, and multi-dimensional features enhance discrimination accuracy, ensuring high-quality weld formation.

[0107] For those skilled in the art, it is obvious that the present application is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present application, the present application can be implemented in other specific forms. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present application is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be construed as limiting the claims involved.

Claims

1. A control method for a welding coupling device, characterized in that: The method comprises: The continuous image sequence of the welding instruction execution process is collected by a high-speed camera, the contour information of the welding spot is extracted by edge detection algorithm, and the center line of the welding spot is generated by combining morphological processing method. According to the preset weld spot size and weld width threshold, the extracted contour and centerline data are compared to determine whether the weld spot has abnormal size or abnormal weld width; If there is an abnormality, the image segmentation method is used to identify the abnormal area in the image, and the area and geometric features of the abnormal area are calculated by the pixel statistics method to obtain the size and deformation degree of the solder joint; According to the preset welding gun path parameters, combined with the extracted weld center line coordinates, the deviation value between the weld center line and the preset path is calculated to obtain the weld position offset, and the welding instruction is corrected in combination with the size and deformation degree of the weld; Obtain the welding instruction data after execution correction, extract the energy, focal length and welding gun angle parameters in the instruction as the reference value for welding instruction analysis; By analyzing the energy distribution curve during welding, combined with the real-time collected focal length data and welding gun angle changes, it is determined whether the welding energy and focal length match the reference values ​​analyzed by the welding instruction; If the welding energy and focal length do not match the reference values ​​analyzed by the welding instruction, the control parameters of the welding equipment are adjusted according to the deviation values ​​and the correction amount of the welding gun path is recalculated; The adjusted control parameters are input into the welding equipment control system to monitor the energy and focal length changes during the welding process in real time. By continuously collecting images of the welding process, the stability of the weld contour and center line is verified to determine whether the welding quality meets the preset standards.

2. The method according to claim 1, characterized in that The method collects a continuous image sequence of the welding instruction execution process by a high-speed camera, extracts the contour information of the welding spot by an edge detection algorithm, and generates the center line of the welding spot by combining a morphological processing method, including: Collecting a first image sequence of the welding joint motion trajectory according to a preset acquisition frequency, extracting first grayscale image data from the first image sequence, and obtaining an optimal grayscale threshold by using image histogram statistics to obtain first binary image data; Using a Gaussian filter to perform noise reduction processing on the first binarized image data to obtain second binarized image data, and using a Sobel operator to perform edge detection on the second binarized image data to obtain first edge contour data; Calculating the density distribution of contour pixels for the first edge contour data, performing breakpoint connection processing on the edge contour using a density clustering method to obtain second edge contour data, and processing the second edge contour data through a morphological closing operation to obtain closed contour line data; A morphological thinning algorithm based on distance transformation is used to perform skeleton extraction on the closed contour line data to obtain first skeleton data, and a cubic spline interpolation algorithm is used to perform node continuity processing on the first skeleton data to obtain welding point contour centerline data.

3. The method according to claim 1, characterized in that The method of comparing the extracted contour and centerline data based on the preset weld spot size and weld width thresholds to determine whether the weld spot has abnormal size or abnormal weld width includes: Reading standard solder joint size parameters from a preset parameter database, and processing the first solder joint contour data using a contour feature extraction method to obtain actual solder joint size parameters; Extracting the weld width feature region by using a regional growing algorithm according to the first weld spot contour data to obtain weld width data, and performing statistical analysis on the weld width data by using a distance transformation method to obtain a weld width mean value; The first welding spot contour data is processed by an equidistant sampling method to obtain a first sampling point sequence, and first curvature distribution data is obtained by local curvature calculation; If the first abnormal feature data to the fourth abnormal feature data form a comprehensive abnormal feature vector of a solder joint, a multi-dimensional threshold comparison is performed based on the comprehensive abnormal feature vector and the abnormal discrimination threshold to obtain a solder joint abnormal type identifier.

4. The method according to claim 1, characterized in that If an abnormality exists, an image segmentation method is used to identify the abnormal area in the image, and the area and geometric features of the abnormal area are calculated by a pixel statistics method to obtain the size and deformation degree of the solder joint, including: Marking the welding spot image according to the preset abnormality judgment result to obtain a first abnormality marked image; Using a watershed segmentation algorithm to extract the boundary of the first abnormality mark image to obtain first region boundary data, wherein the first region boundary data is used to calculate geometric parameters of the abnormal region; Calculate the coordinates of the center of gravity and the direction of the main axis of the abnormal area according to the first area boundary data to obtain first shape feature data, wherein the first shape feature data is used for density distribution analysis of the abnormal area; A symmetry parameter is calculated for the first shape feature data by a regional projection method to obtain a first deformation parameter, and the first deformation parameter is combined with the first geometric parameter data to form an abnormal region feature vector.

5. The method according to claim 1, characterized in that The method of calculating the deviation between the center line of the weld spot and the preset path according to the preset welding gun path parameters and the extracted weld spot center line coordinates to obtain the weld spot position offset, and correcting the welding instruction in combination with the size and deformation degree of the weld spot includes: Acquire a welding path coordinate sequence from a preset path database, generate a first path curve using a cubic spline interpolation method, and calculate the distance between corresponding points based on the first path curve and a weld centerline coordinate fitting curve to obtain first deviation data; Performing interval sampling according to the first deviation data and calculating the local deviation average value to obtain first offset data, and performing smoothing processing on the first offset data using a sliding window method to obtain path offset interval data; Calculating a coordinate correction coefficient based on the path offset interval data and the abnormal solder joint size data in the quality inspection database to obtain a second correction parameter, and using a piecewise interpolation method to generate a path compensation curve to obtain second path compensation data; A position correction amount is generated according to the second path compensation data to obtain a second instruction parameter, the second instruction parameter is merged with a standard welding parameter to obtain a corrected instruction sequence, and a kinematic constraint verification method is used to perform a feasibility check on the corrected instruction sequence to obtain a qualified instruction sequence.

6. The method according to claim 1, characterized in that The step of obtaining and executing the corrected welding instruction data and extracting the energy, focal length and welding gun angle parameters in the instruction as reference values ​​for welding instruction analysis includes: The welding instruction sequence is read from the instruction database by using the instruction parsing method, and the instruction timestamp and instruction type identification are obtained by using the instruction parsing method to obtain the welding parameter instruction set; Sorting the welding parameter instruction set according to the timestamps, separating the energy parameter, the focal length parameter and the angle parameter by a parameter extraction algorithm to obtain a first parameter array; Resampling the first parameter array through a fixed time window, and normalizing the resampled data according to parameter units and ranges to obtain a first standard parameter sequence; Statistical parameters and fluctuation characteristics are calculated for the first standard parameter sequence. If the statistical parameters and fluctuation characteristics meet the stability interval requirements, a reference parameter vector is generated. The reference parameter vector is reconstructed according to the instruction format to obtain a welding instruction reference value.

7. The method according to claim 1, characterized in that The energy distribution curve during welding is analyzed, and the focal length data and welding gun angle changes collected in real time are combined to determine whether the welding energy and focal length match the reference values ​​analyzed by the welding instruction, including: Reading energy sampling data during welding from a sensor database, verifying the validity of the energy sampling data according to a sampling timestamp, and obtaining first curve data by using a sliding average method; The focal length data and angle data are subjected to outlier elimination, and the second curve data is obtained by piecewise least square fitting; Reading welding reference parameters from a reference value database, and normalizing energy parameters, focal length parameters, and angle parameters in the welding reference parameters to obtain standard reference data; The first curve data and the second curve data are time-series aligned to obtain characteristic parameter data, a curve correlation algorithm is used to calculate the matching degree between the characteristic parameter data and the standard reference data, and a parameter matching identifier is obtained according to the parameter threshold data.

8. The method according to claim 1, characterized in that If the welding energy and focal length do not match the reference values ​​analyzed by the welding instruction, the control parameters of the welding equipment are adjusted according to the deviation value, and the correction amount of the welding gun path is recalculated, including: Read welding parameter deviation data from a parameter analysis database, and calculate energy deviation value and focal length deviation value using an adaptive mapping algorithm according to the welding parameter deviation data to obtain first deviation data; A binocular vision sensor is used to collect a welding position image, and a position difference is calculated according to the welding position image by an image registration method to obtain first offset data; According to the first offset data, a coordinate transformation matrix is ​​used to transform the first offset data into a welding gun coordinate system to obtain second offset data, and a welding gun position compensation coefficient is calculated by using the second offset data to obtain a first compensation vector; A first path correction value is obtained by using a cubic spline interpolation method according to the first compensation vector, and path compensation data is obtained by mapping the first path correction value to a joint space using a Jacobian matrix method.

9. The method according to claim 1, characterized in that: The adjusted control parameters are input into the welding equipment control system, and the energy and focal length changes during the welding process are monitored in real time. By continuously collecting welding process images, the stability of the weld point contour and center line is verified, and whether the welding quality meets the preset standard is determined, including: The energy data and focal length data of the welding process are collected by using a device sensor to obtain first monitoring data, and the second monitoring data is obtained by verifying the validity of the data according to the first monitoring data; According to the second monitoring data, a high-speed industrial camera is used to continuously collect welding spot forming images to obtain a first image sequence, and a watershed segmentation algorithm is used to extract welding spot contour data from the first image sequence to obtain a first contour sequence; Calculating contour shape parameters and centerline offset parameters of the first contour sequence using a continuity detection method to obtain first stability data; A quality feature vector is generated by performing statistical analysis on the first stability data to obtain first feature data, and a quality determination result is obtained by calculating a matching degree between the first feature data and standard feature data using a multi-dimensional threshold comparison method.

Images