Aluminum alloy auxiliary frame welding accurate positioning and quality control method and system

Abstract

The invention provides an aluminum alloy auxiliary frame welding accurate positioning and quality control method and system. The method comprises the steps that S4, welding gun track control parameters are updated through a dynamic adjustment instruction, laser tracking data are fused to optimize the welding path angle in real time, and the accurate movement track of each section of welding seam is determined; s5, acquiring real-time thermal expansion data during welding execution, monitoring temperature field distribution of a welding seam area by adopting an optical measurement technology, and judging whether thermal deformation adjustment exceeds a preset threshold value or not; s7, micron-sized welding spot positioning operation is executed according to the stable positioning state, the dynamic path optimization result is fused to generate a final welding instruction, and consistent welding seam quality data is output; and S8, after weld joint quality data are obtained, weld joint deviation and geometric consistency are detected through a laser tracking technology, whether welding quality meets design requirements or not is judged, and a quality verification report is generated. According to the method, the precision and consistency of intelligent welding of the aluminum alloy auxiliary frame are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for precise positioning and quality control of welding of an aluminum alloy subframe. Background Art

[0002] In the automotive industry, aluminum alloy subframes are critical load-bearing components, and the quality of their welding is directly related to vehicle safety, lightweighting, and durability. As the automotive industry moves toward lightweighting and high performance, aluminum alloy, due to its excellent strength-to-weight ratio and corrosion resistance, has become the material of choice for subframe construction. However, the precision and consistency of the welding process are crucial for ensuring structural integrity and stable performance. Research on efficient and precise welding equipment and its use is particularly important. This field is not only at the forefront of technological innovation but also at the core of driving the upgrading of the automotive industry.

[0003] Currently, welding aluminum alloy subframes relies heavily on traditional equipment and manual labor. While this method has met production needs to a certain extent, its limitations are becoming increasingly apparent. Traditional methods suffer from significant deficiencies in positioning accuracy, trajectory control, and thermal deformation mitigation, often leading to large weld deviations and inconsistent quality. Accumulated errors can impact overall performance, especially in complex structural areas. These shortcomings not only reduce production efficiency but also make it difficult to meet the demanding high precision and consistency requirements of modern vehicles.

[0004] The core challenges facing aluminum alloy subframe welding primarily focus on several key technical factors: precise control of the welding gun position and weld trajectory; regulating the material's thermal expansion and deformation at high temperatures; and ensuring stable positioning of complex structural components. Due to the temperature sensitivity of aluminum alloys, uneven heat input during welding can lead to microscopic deformation, and existing equipment often lacks the ability to adjust in real time. Furthermore, the subframe's complex geometry makes it difficult for traditional positioning systems to achieve multi-dimensional coordination, resulting in weld point offsets. These unresolved technical challenges make it difficult to break through the micron-level bottleneck in welding accuracy, which in turn affects the subframe's mechanical properties and manufacturing consistency.

[0005] Therefore, the key to improving the quality consistency of aluminum alloy subframe welding is how to achieve micron-level precision positioning of the welding gun on the designed weld point through the coordinated operation of multi-dimensional positioning and high-precision robotic arms, integrating optical measurement, laser tracking, and force feedback technologies, and adjusting the welding path and angle in real time based on thermal expansion and deformation. Solving this problem will directly promote a leap in welding equipment performance and meet the automotive industry's urgent demand for high-precision manufacturing. Summary of the Invention

[0006] In order to solve the problems raised in the above background technology, the first aspect of the present invention provides a method for precise positioning and quality control of aluminum alloy subframe welding, the method comprising: S1, obtains the 3D geometric model and designed weld point coordinates of the aluminum alloy subframe, scans the workpiece surface using optical measurement technology, generates real-time surface point cloud data, and determines the initial welding gun trajectory and multi-dimensional positioning benchmark; S2 calculates the curvature change of the complex structure of the workpiece and the coordinates of key positioning points based on the point cloud data. Laser tracking technology is used to monitor the spatial deviation between the welding gun and the designed welding point to obtain the micron-level positioning error distribution; S3 adjusts the motion parameters of the high-precision robotic arm based on the error distribution, combines force feedback technology to detect minute deformations on the workpiece surface during welding, determines thermal deformation trends, and generates dynamic adjustment instructions; S4 updates the welding gun trajectory control parameters through dynamic adjustment instructions, integrates laser tracking data to optimize the welding path angle in real time, and determines the precise motion trajectory of each weld; S5, obtains real-time thermal expansion data during welding, uses optical measurement technology to monitor the temperature field distribution in the weld area, and determines whether the thermal deformation adjustment exceeds the thermal deformation threshold; S6, if the thermal deformation adjustment exceeds the thermal deformation threshold, the compensation vector is calculated through force feedback technology, and the posture control data of the high-precision manipulator is updated to obtain a stable complex structure positioning state; S7, performs micron-level welding point positioning operations based on the stable positioning state, integrates the dynamic path optimization results to generate the final welding instructions, and outputs consistent weld quality data; S8, after obtaining the weld quality data, uses laser tracking technology to detect weld deviation and geometric consistency, determines whether the welding quality meets the design requirements, and generates a quality verification report; S9, based on the deviation data in the quality verification report, adjusts the real-time adjustment capability parameters, optimizes the welding gun trajectory control accuracy of subsequent welding tasks, and outputs the improved process parameter set.

[0007] Optionally, step S1, obtaining a three-dimensional geometric model of the aluminum alloy subframe and the coordinates of the designed welding points, scanning the workpiece surface using optical measurement technology, generating real-time surface point cloud data, and determining an initial welding gun trajectory and a multi-dimensional positioning reference, includes: Step S11, obtaining a three-dimensional geometric model of the aluminum alloy subframe and extracting coordinate data of key feature points; Step S12, determining the welding position coordinates according to the characteristic point distribution of the geometric model; Step S13, scanning the workpiece surface to generate point cloud data containing three-dimensional coordinate information; Step S14: Compare and analyze the point cloud data obtained by scanning with the geometric model, and calculate the deviation value of each point; Step S15: Set a point deviation threshold. If the point cloud data exceeds the point deviation threshold, use the K-means algorithm to perform cluster analysis on the abnormal points to generate an optimized point cloud set. Step S16, replanning the welding gun motion trajectory based on the optimized point cloud data to ensure that the welding position is consistent with the designed position; Step S17, fusing the re-planned welding gun trajectory with the original coordinate data to generate a complete welding path program; Step S18: import the welding path program into the industrial robot control system to complete the welding operation.

[0008] Optionally, the step S12, determining the welding position coordinates according to the characteristic point distribution of the geometric model, includes: Step S121, loading the geometric model into CATIA software, and extracting the coordinate data of the model edge points and surface points; Step S122, dividing the regions according to the curvature variation of adjacent points, and outputting the coordinate text file of the partitioned point set; Step S123, performing Z-score normalization processing on the point set coordinates, clustering based on Euclidean distance, determining the optimal number of clusters by the silhouette coefficient, and outputting the grouped point set; Step S124, calculating the standard deviation of the spacing between each group of points. If it exceeds the standard deviation threshold, the point set is exported as an STL file; Step S125, importing the STL file into Geomagic software, using the curvature optimization tool to adjust the point set distribution, and saving the adjusted coordinate data in ASC format; Step S126, importing the ASC file into RobotStudio, setting the robot tool center coordinate system, and calling the CollisionCheck function to generate a collision-free welding path; Step S127: Load the path program through the ABB controller to execute the digital output of the welding coordinate points.

[0009] Optionally, step S2 calculates the curvature change of the complex structure of the workpiece and the coordinates of key positioning points based on the point cloud data, uses laser tracking technology to monitor the spatial deviation between the welding gun and the designed welding point, and obtains the micron-level positioning error distribution, including: Step S21, obtaining the surface features of the complex structure of the workpiece through point cloud data, and calculating the curvature change and the coordinates of the positioning points; Step S22, using laser tracking technology to monitor the spatial relationship between the welding gun position and the designed welding point to determine the spatial deviation; Step S23, performing statistical analysis on the spatial deviation distribution to obtain micron-level error distribution characteristics; Step S24: if the error distribution exceeds the error distribution threshold, the coordinates of the positioning points are adjusted by an interpolation algorithm to obtain optimized coordinate data; Step S25, updating the curvature change trend of the complex structure according to the optimized coordinate data, and determining the surface feature adjustment direction; Step S26, obtaining the adjusted curvature change trend, correcting the welding gun position through geometric mapping technology, and determining the final deviation control result; Step S27 : Compare the final deviation control result with the designed solder joint to obtain verification data of micron-level positioning error distribution.

[0010] Optionally, the step S21, obtaining the surface features of the complex structure of the workpiece through point cloud data, and calculating the curvature change and the coordinates of the positioning points, includes: Step S211, registering the acquired point cloud data using the ICP algorithm, extracting vertex coordinates and normal vectors as initial morphological features; Step S212, constructing a topological relationship based on adjacent triangles, and calculating the average curvature value of each vertex using a discrete curvature estimation method; Step S213, mapping the curvature values to vertex coordinates to generate a curvature distribution map; Step S214, setting a change threshold on the curvature distribution map, and marking abnormal areas where the curvature gradient exceeds the change threshold; Step S215: For the vertices in the abnormal area, the coordinates are recalculated using a thin plate spline interpolation algorithm, with the interpolation weight being proportional to the curvature gradient amplitude; Step S216, the updated vertex coordinates are input as new morphological features; Step S217, recalculating the triangle normal vector based on the updated vertex coordinates, and comparing the change in the normal vector angle before and after the adjustment; Step S218: When the average angle change in the local area exceeds the angle change threshold, the point cloud of the area is optimized twice using the moving least squares method; Step S219: Finally, a set of key point coordinates that satisfy curvature continuity is output.

[0011] Optionally, the step S22, using laser tracking technology to monitor the spatial relationship between the welding gun position and the designed welding point to determine the spatial deviation, includes: Step S221, collecting three-dimensional coordinate data of the welding gun position and the designed welding point by laser tracking technology to generate an initial coordinate set; Step S222, calculating the Euclidean distance between the welding gun coordinates and the designed welding point to obtain the spatial deviation value of each sampling point; Step S223: If the deviation value of the Euclidean distance is greater than the Euclidean distance threshold (mm), the welding gun coordinates are interpolated and adjusted using the bicubic interpolation method, and the adjusted coordinate set is output; Step S224, recalculating the deviation value based on the adjusted coordinates to generate an updated deviation distribution map; Step S225 , extracting the X / Y / Z axial deviation components from the deviation distribution map as feature vectors, clustering using the K-means algorithm of scikit-learn, setting the number of clusters to 3, and outputting the classification results of high / medium / low deviation areas; Step S226: For high deviation areas, increase the laser sampling frequency to 200 Hz; Step S227: The low-to-medium deviation region maintains the original 100 Hz sampling frequency and generates an optimized coordinate set; Step S228, based on the optimized coordinate set, calculate the root mean square value of the deviation between the final position of the welding gun and the designed weld point. If it is less than the preset value, the process is terminated, otherwise it is iterated again from the coordinate acquisition step.

[0012] Optionally, step S7, performing micron-level weld point positioning operations according to the stable positioning state, integrating dynamic path optimization results to generate final welding instructions, and outputting consistent weld quality data, includes: Step S71, obtaining stable positioning state data through sensors, and obtaining positioning coordinates using Kalman filtering; Step S72: For the positioning coordinates, the dynamic path information is integrated using a weighted average method to obtain a path optimization result; Step S73, generating a welding instruction sequence using a rule-based method based on the path optimization result and outputting an instruction set; Step S74, obtaining welding execution parameters from the instruction set, performing simulation using finite element analysis, and determining consistency quality indicators; Step S75, using principal component analysis to extract weld quality characteristics based on the consistency quality index to obtain a quality assessment value; Step S76, comparing the quality evaluation value to the quality index threshold, if it exceeds the threshold, adjusting the path optimization parameters using the gradient descent method and regenerating the instruction set; Step S77: Use the weighted update method to update the data output content to obtain the final weld quality data.

[0013] Optionally, step S74, obtaining welding execution parameters from the instruction set, performing simulation using finite element analysis, and determining consistency quality indicators, includes: Step S741, parsing current, voltage, and speed parameters from the welding equipment control instruction set to generate a structured parameter table; Step S742: Building a three-dimensional welding geometry model based on the parameter table and importing the thermophysical parameters in the material property library; Step S743, using the finite element method to solve the Fourier heat conduction equation and calculate the temperature field distribution data of the weld area; Step S744: Input the temperature field data as initial conditions into the ABAQUS explicit dynamics module and output the Mises stress tensor of the weld area. Step S745, using the PCA tool of scikit-learn to reduce the dimension of the stress tensor, retaining the first three principal components whose cumulative contribution rates exceed the contribution rate threshold; Step S746: Divide the principal component data into a training set and a test set at a ratio of 7 to 3, build a three-layer fully connected network in TensorFlow, and use the ReLU activation function to train the quality prediction model; Step S747: When the predicted crack sensitivity coefficient exceeds the first sensitivity coefficient threshold, the genetic algorithm module of the DEAP library is called to perform iterative optimization with the parameter combination as the chromosome and the predicted value as the fitness; Step S748: Finally output the parameter combination that makes the crack sensitivity coefficient lower than the second sensitivity coefficient threshold.

[0014] Optionally, the first sensitivity coefficient threshold is 35, and the second sensitivity coefficient threshold is 28.

[0015] A second aspect of the present invention provides an aluminum alloy subframe welding precision positioning and quality control system, which uses the above-mentioned method to perform welding precision positioning and quality control on the aluminum alloy subframe, and the system includes: The geometric model acquisition module is used to obtain the 3D geometric model of the aluminum alloy subframe and the coordinates of the designed weld points. It uses optical measurement technology to scan the workpiece surface, generate real-time surface point cloud data, and determine the initial welding gun trajectory and multi-dimensional positioning benchmark; The point cloud processing module is used to calculate the curvature changes and key positioning point coordinates of the complex structure of the workpiece based on the point cloud data. Laser tracking technology is used to monitor the spatial deviation between the welding gun and the designed weld point to obtain the micron-level positioning error distribution; The error monitoring module is used to adjust the motion parameters of the high-precision robotic arm according to the error distribution. It combines force feedback technology to detect small deformations on the workpiece surface during welding, determine the thermal deformation trend, and generate dynamic adjustment instructions. The motion control module is used to update the welding gun trajectory control parameters through dynamic adjustment instructions, integrate laser tracking data to optimize the welding path angle in real time, and determine the precise motion trajectory of each weld; The trajectory optimization module is used to obtain real-time thermal expansion data during welding, monitor the temperature field distribution in the weld area using optical measurement technology, and determine whether the thermal deformation adjustment exceeds the thermal deformation threshold; The thermal deformation monitoring module is used to calculate the compensation vector through force feedback technology if the thermal deformation adjustment exceeds the thermal deformation threshold, update the posture control data of the high-precision manipulator, and obtain a stable complex structure positioning state; The compensation correction module is used to perform micron-level weld point positioning operations based on the stable positioning state, integrate the dynamic path optimization results to generate the final welding instructions, and output consistent weld quality data; The welding execution module is used to obtain weld quality data, detect weld deviation and geometric consistency through laser tracking technology, determine whether the welding quality meets the design requirements, and generate a quality verification report; The quality assessment module is used to adjust the real-time adjustment capability parameters based on the deviation data in the quality verification report, optimize the welding gun trajectory control accuracy of subsequent welding tasks, and output the improved process parameter set.

[0016] The technical solution provided by the embodiment of the present invention has the following beneficial effects: The present invention provides a method and system for precise positioning and quality control of aluminum alloy subframe welding. The system uses optical measurement to obtain three-dimensional point cloud data of the workpiece, combines laser tracking technology to monitor welding gun position deviation, and achieves micron-level weld point positioning. During the welding process, force feedback technology is used to detect tiny deformations on the workpiece surface, and the welding gun trajectory is dynamically adjusted based on real-time thermal expansion data.

[0017] The present invention integrates a variety of advanced sensing technologies to construct an intelligent welding control system that can accurately sense the spatial state changes of complex structure workpieces and optimize welding parameters in real time.

[0018] By continuously monitoring weld quality and optimizing process parameters, the present invention significantly improves the accuracy and consistency of intelligent welding of aluminum alloy subframes, providing an effective solution for high-quality intelligent welding production. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flow chart of a method for precise positioning and quality control of aluminum alloy subframe welding.

[0020] Figure 2 The figure is a schematic structural diagram of an aluminum alloy subframe welding precision positioning and quality control system of the present invention. DETAILED DESCRIPTION

[0021] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments derived by those skilled in the art based on the embodiments in this specification without creative effort shall fall within the scope of protection of this specification.

[0022] like Figure 1 As shown, the first aspect of the present invention provides a method for precise positioning and quality control of aluminum alloy subframe welding, which specifically includes: S1 obtains the 3D geometric model and designed weld point coordinates of the aluminum alloy subframe, scans the workpiece surface using optical measurement technology, generates real-time surface point cloud data, and determines the initial welding gun trajectory and multi-dimensional positioning benchmark.

[0023] Optionally, this step also includes: Step S11: Use CAD software to obtain a three-dimensional geometric model of the aluminum alloy subframe and extract coordinate data of key feature points.

[0024] Step S12: determining the welding position coordinates in CATIA software according to the distribution of characteristic points of the geometric model.

[0025] Step S13: Use the GOMInspect optical measurement device to scan the workpiece surface and generate point cloud data containing three-dimensional coordinate information.

[0026] In step S14, the point cloud data obtained by scanning is imported into Geomagic Control X software, compared and analyzed with the three-dimensional geometric model in CAD, and the deviation value of each point is calculated.

[0027] Step S15: Set a point deviation threshold. If the point cloud data exceeds the point deviation threshold, use the K-means algorithm in the Python Scikit-learn library to perform cluster analysis on the abnormal points to generate an optimized point cloud set.

[0028] Step S16: Replan the welding gun motion trajectory in the RobotStudio software based on the optimized point cloud data to ensure that the welding position is consistent with the designed position.

[0029] In step S17, the re-planned welding gun trajectory is integrated with the original coordinate data to generate a complete welding path program in RobotStudio.

[0030] Step S18: import the welding path program into the industrial robot control system to complete the welding operation.

[0031] Specifically, when using CAD software to obtain the three-dimensional geometric model of the aluminum alloy subframe, the design drawings can be imported through tools such as SolidWorks or AutoCAD to generate an accurate three-dimensional entity.

[0032] For example, a typical aluminum alloy subframe may consist of crossbeams, longitudinal beams, and connection nodes. The coordinate data for key feature points, such as the center of the connection hole and the edge corners, can be directly extracted from the model with an accuracy of up to 0.01mm. This data lays the foundation for subsequent steps, ensuring the accuracy of geometric information. When determining the coordinates of welding positions in CATIA software, the welds can be located manually or through automatic recognition based on the distribution of feature points.

[0033] For example, the weld position at the intersection of the crossbeam and the longitudinal beam can be set as a straight line trajectory 5 mm away from the feature point.

[0034] It should be noted that CATIA's modular design capabilities can effectively simulate welding conditions, ensuring that position coordinates are consistent with actual process requirements, thereby improving design efficiency. When scanning the workpiece surface using the GOMInspect optical measurement device, high-density point cloud data can be obtained through blue light scanning technology.

[0035] For example, a surface scan of a subframe might generate 500,000 3D coordinate points with a resolution of 0.05mm. This high-precision scan fully captures even the smallest surface deformations, providing a reliable basis for subsequent deviation analysis. Importing the point cloud data into Geomagic Control X software and comparing it with the CAD model provides a visual display of the deviation distribution.

[0036] For example, comparing the point cloud data at a certain connection node with the design model revealed a maximum deviation of 0.2 mm, concentrated in the edge area. This comparative analysis can quickly pinpoint manufacturing errors and help optimize process parameters.

[0037] When setting the point deviation threshold and using the K-means algorithm to cluster outliers, the point deviation threshold can be set to 0.15 mm.

[0038] In one possible implementation, if the deviation in a certain area exceeds the point deviation threshold, the K-means algorithm can classify the outliers into three categories, corresponding to factors such as machining defects and measurement noise. The optimized point cloud collection eliminates invalid data and retains 95% of valid points, improving data quality.

[0039] When replanning the welding gun motion trajectory in RobotStudio based on the optimized point cloud data, the path of the welding gun along the weld can be simulated.

[0040] For example, for nodes with large deviations, the welding gun angle is adjusted to 45 degrees and the welding speed is reduced to 10mm / second to ensure welding accuracy. This dynamic adjustment can significantly reduce the welding defect rate.

[0041] When the re-planned welding gun trajectory is integrated with the original coordinate data, RobotStudio can generate a smooth welding path program.

[0042] For example, the path length after fusion may be optimized from 500mm to 480mm, reducing redundant movements.

[0043] Specifically, this optimization can improve the robot's operating efficiency and reduce energy consumption.

[0044] When the welding path program is imported into the industrial robot control system to complete the work, it can be executed by ABB or KUKA robots.

[0045] For example, a subframe welding task, which may involve 20 welds, has seen its total welding time reduced from 15 minutes to 12 minutes.

[0046] Understandably, this automated operation improves production consistency and reduces errors caused by human intervention.

[0047] Preferably, the entire process achieves seamless integration of design and manufacturing through digital means.

[0048] For example, from CAD modeling to robotic welding, data sharing at every step avoids information gaps. This integrated approach significantly improves the manufacturing precision and efficiency of the subframe, providing technical support for subsequent mass production.

[0049] Optionally, the step S12 of determining the welding position coordinates in CATIA software according to the distribution of characteristic points of the geometric model may further include: Step S121 , loading the geometric model into CATIA software, and extracting the coordinate data of the model edge points and surface points.

[0050] Step S122 , dividing the regions according to the curvature variation of adjacent points, and outputting a text file of the point set coordinates after the partitioning.

[0051] In step S123 , the coordinates of the point set are subjected to Z-score normalization processing, clustering is performed based on the Euclidean distance using the K-means algorithm, the optimal number of clusters is determined by the silhouette coefficient, and the grouped point set is output.

[0052] Step S124 , calculating the standard deviation of the spacing between each group of points. If it exceeds the preset 2mm spacing standard deviation threshold, the point set is exported as an STL format file.

[0053] Step S125 , importing the STL file into Geomagic software, using the curvature optimization tool to adjust the point set distribution, and saving the adjusted coordinate data in ASC format.

[0054] Step S126: import the ASC file into RobotStudio, set the robot tool center coordinate system, and call the CollisionCheck function to generate a collision-free welding path.

[0055] Step S127: Load the path program through the ABB controller to execute the digital output of the welding coordinate points.

[0056] Specifically, when loading the geometric model in CATIA software, the three-dimensional data of the aluminum alloy subframe can be obtained by importing IGES or STEP format files.

[0057] In one possible implementation, users can use the "Digital Shape Editor" module to extract edge and surface point coordinates. This extraction process relies on the topology of the model, and the software automatically identifies boundary lines and surface transition areas.

[0058] For example, if the subframe includes an arc transition surface, the curvature threshold can be set to 0.5, and segmentation can begin at the edge points with the largest curvature changes, generating a point set text file containing XYZ coordinates. This method effectively preserves the key geometric features of the model.

[0059] Specifically, dividing regions according to the curvature variation of adjacent points can be understood as a partitioning strategy based on geometric characteristics.

[0060] In one embodiment, if the curvature variation in a certain area exceeds 30%, the area is divided into a sub-area and a coordinate file of the partitioned area is output.

[0061] Preferably, the coordinate file format can be selected as TXT to facilitate subsequent processing. This partitioning method helps to decompose complex surfaces into more manageable units.

[0062] The principle of Z-score standardization of point set coordinates is to convert the data into a distribution with a mean of 0 and a standard deviation of 1.

[0063] For example, assuming that the Z-axis coordinate range of the original point set is 50-70 mm, after standardization, the data distribution is more even, which is convenient for subsequent cluster analysis.

[0064] It should be noted that standardization can reduce calculation bias caused by dimensional differences. When using the K-means algorithm for clustering based on Euclidean distance, the optimal number of clusters can be determined through multiple iterations.

[0065] In a possible implementation, the k value is initially set to 3-6, and the silhouette coefficient is calculated.

[0066] For example, when k = 4, the silhouette coefficient reaches 0.7, which is higher than 0.65 when k = 5, so k = 4 is selected. This method can reasonably group the point set and improve the accuracy of subsequent path planning.

[0067] When calculating the standard deviation of the spacing between points in each group, if it exceeds the spacing standard deviation threshold of 2 mm, it means that the points are unevenly distributed.

[0068] For example, in the subframe weld area, if the standard deviation of the point spacing in a group is 2.5mm, it requires optimization and adjustment. This inspection ensures the quality of the point set and prevents deviations in the weld path. After exporting the point set as an STL file, when importing it into Geomagic software, the "Curvature Optimization" tool can be used to adjust the point distribution.

[0069] Specifically, if the point density in a certain area is too high, thinning can be used to adjust the point spacing to around 1mm and save it in ASC format. This optimization can improve data smoothness and reduce jitter during robot movement.

[0070] In one embodiment, when importing the ASC file into RobotStudio and setting the tool center point coordinate system, the welding gun tip can be defined as the TCP origin, and the offset can be set to X=0, Y=0, and Z=10 mm.

[0071] In one embodiment, when the CollisionCheck function is called, the software automatically detects potential collision points between the welding gun and the workpiece and adjusts the path, significantly improving the safety of the path.

[0072] After the path program is loaded through the ABB controller, the digital output of the welding coordinate points can be executed, which can be understood as converting the virtual path into actual robot instructions.

[0073] For example, the controller can output a data stream of 50 coordinate points per second to ensure a continuous and stable welding process. This digital output method can improve work efficiency and consistency.

[0074] S2 calculates the curvature changes and key positioning point coordinates of the complex structure of the workpiece based on the point cloud data, and uses laser tracking technology to monitor the spatial deviation between the welding gun and the designed weld point to obtain the micron-level positioning error distribution.

[0075] Optionally, this step also includes: Step S21 , obtaining the surface features of the complex structure of the workpiece through point cloud data, and calculating the curvature change and the coordinates of the positioning points.

[0076] Step S22: Using laser tracking technology to monitor the spatial relationship between the welding gun position and the designed welding point to determine the spatial deviation.

[0077] Step S23 , performing statistical analysis on the spatial deviation distribution to obtain micron-level error distribution characteristics.

[0078] Step S24: If the error distribution exceeds a preset error distribution threshold, the coordinates of the positioning points are adjusted by an interpolation algorithm to obtain optimized coordinate data.

[0079] Step S25 , updating the curvature change trend of the complex structure according to the optimized coordinate data, and determining the surface feature adjustment direction.

[0080] Step S26: Obtain the adjusted curvature change trend, correct the welding gun position through geometric mapping technology, and determine the final deviation control result.

[0081] Step S27 : Compare the final deviation control result with the designed solder joint to obtain verification data of micron-level positioning error distribution.

[0082] Specifically, when obtaining the surface features of a complex workpiece structure through point cloud data, the dense three-dimensional coordinate information in the point cloud can be used to identify curvature changes.

[0083] For example, the surface of a crossmember of an aluminum alloy subframe may contain flat areas and curved transition areas. By analyzing the distance and angle changes between adjacent points, the curvature value can be calculated.

[0084] For example, the curvature in flat areas is close to zero, while the curvature in transition areas may reach 0.1mm^-1. The coordinates of the positioning points can be selected at locations with sudden changes in curvature, such as the intersection of a beam and a connection node. The coordinate accuracy can be controlled within 0.05mm. When using laser tracking technology to monitor the spatial relationship between the welding gun position and the designed weld point, the spatial posture of the welding gun can be captured in real time through the laser transmitter and receiver.

[0085] Specifically, the laser tracker can collect position data 1,000 times per second with an accuracy of 0.01mm.

[0086] In one possible implementation, if the designed weld point coordinates are (100, 50, 20) mm, and the actual welding gun position is (100.05, 50.02, 20.03) mm, the spatial deviations are 0.05 mm, 0.02 mm, and 0.03 mm, respectively. This high-frequency monitoring can promptly reflect welding gun offset. When performing statistical analysis on the spatial deviation distribution, multiple measurement data can be collected to identify error patterns.

[0087] For example, after monitoring 100 solder joints, it was found that the deviations were mainly concentrated in the range of 0.02mm to 0.08mm, with an average value of 0.05mm, showing a normal distribution characteristic.

[0088] It should be noted that this statistical result can help determine whether the deviation is a systematic error or a random fluctuation, providing a basis for subsequent adjustments. If the error distribution exceeds the preset error distribution threshold, such as 0.06mm, the positioning point coordinates are adjusted through interpolation algorithm.

[0089] Preferably, a cubic spline interpolation method may be used to generate a smooth curve according to the coordinates of surrounding points.

[0090] In one embodiment, if the deviation of a weld point is 0.08 mm, the new coordinates after interpolation may be adjusted from (100.05, 50.02, 20.03) mm to (100.02, 50.01, 20.01) mm. This method effectively smooths the error transition. When the curvature trend is updated based on the optimized coordinate data, the geometric characteristics of the surface features can be recalculated.

[0091] For example, the adjusted positioning points may reduce the curvature of the transition zone from 0.1mm^-1 to 0.09mm^-1, indicating that the surface is flattening.

[0092] It is understandable that this trend judgment can guide the subsequent fine-tuning of the welding gun posture. After obtaining the adjusted curvature change trend, the welding gun position is corrected through geometric mapping technology.

[0093] For example, if a reduction in curvature indicates a flatter surface, the welding gun can be adjusted to a vertical position and lowered by 0.5 mm to conform to the surface.

[0094] In one possible implementation, the welding gun position is mapped from (100.02, 50.01, 20.01) mm to (100.02, 50.01, 19.51) mm, ensuring that the deviation is controlled within 0.02 mm. This correction can improve welding consistency. When the final deviation control results are compared with the designed weld point, verification data can be generated.

[0095] For example, after comparing 100 solder joints, it was found that 90% of the positioning errors were distributed between 0.01mm and 0.03mm, and the maximum positioning error was 0.04mm.

[0096] For example, this micron-level precision verification can confirm the effectiveness of process adjustments and provide guarantees for high-quality manufacturing.

[0097] Optionally, the step S21 of acquiring the surface features of the complex structure of the workpiece through point cloud data, calculating the curvature change and the coordinates of the positioning points, further includes: Step S211 : aligning the acquired point cloud data through the ICP algorithm, extracting vertex coordinates and normal vectors as initial morphological features.

[0098] Step S212: constructing a topological relationship based on adjacent triangles, and calculating the average curvature value of each vertex using a discrete curvature estimation method.

[0099] Step S213 : Mapping the curvature values to vertex coordinates to generate a curvature distribution map.

[0100] Step S214: Set a change threshold of 0.5 mm^-1 on the curvature distribution map, and mark abnormal areas where the curvature gradient exceeds the change threshold.

[0101] Step S215: For the vertices in the abnormal area, the coordinates are recalculated using a thin plate spline interpolation algorithm, and the interpolation weight is proportional to the curvature gradient amplitude.

[0102] Step S216: The updated vertex coordinates are input as new morphological features.

[0103] Step S217 , recalculating the triangle normal vector according to the updated vertex coordinates, and comparing the change in the normal vector angle before and after the adjustment.

[0104] Step S218: When the average angle change in the local area exceeds the angle change threshold of 15 degrees, the point cloud in the area is optimized twice using the moving least squares method.

[0105] Step S219: Finally, a set of key point coordinates that satisfy curvature continuity is output.

[0106] Specifically, the point cloud data is registered through the ICP algorithm in order to align the collected discrete point cloud with the reference model and extract the vertex coordinates and normal vectors as the initial morphological features.

[0107] For example, when processing point clouds on the surface of a welded workpiece, assuming the workpiece is a thin, curved plate, the collected point cloud may deviate from the design model due to acquisition angle or equipment vibration. The ICP algorithm iteratively calculates the optimal match between points and adjusts the actual point cloud to a position consistent with the design model.

[0108] Preferably, the extracted vertex coordinates can be key inflection points on the surface, and the normal vector reflects the orientation of the local surface, providing basic data for subsequent analysis.

[0109] In one possible implementation, a topological relationship is constructed based on adjacent triangles in order to describe the geometric connections between vertices in the point cloud.

[0110] Specifically, the point cloud can be triangulated to form a triangular mesh consisting of connected vertices.

[0111] For example, for a complex curved workpiece, a triangular facet can be composed of three adjacent vertices. By calculating the number of adjacent facets and the angle relationship of each vertex, a complete topological network is constructed. On this basis, the discrete curvature estimation method is used to calculate the average curvature value.

[0112] For example, the curvature of a vertex can be approximated by the difference in the normal vectors of the surrounding triangles. The curvature value reflects the degree of surface curvature. This method can effectively identify flat areas and high curvature areas on the workpiece surface, providing a basis for subsequent optimization.

[0113] For example, mapping the curvature values to vertex coordinates to generate a curvature distribution map can intuitively display the morphological changes of the workpiece surface.

[0114] In one embodiment, if the curvature value of a certain area on the workpiece surface increases sharply from 0.2mm^-1 to 0.8mm^-1, indicating that there may be a sharp corner or defect, a change threshold of 0.5mm^-1 can be set to mark the area with abnormal curvature gradient.

[0115] For example, an area where the curvature changes from 0.3mm^-1 to 0.7mm^-1 is highlighted. This marking helps quickly locate key points that require adjustment. For vertices in abnormal areas, the coordinates are recalculated using the thin plate spline interpolation algorithm, which is a method for smoothing the surface adjustment.

[0116] It can be understood that the interpolation weight is proportional to the curvature gradient amplitude, which can ensure that the adjusted coordinates are closer to the actual design requirements.

[0117] For example, if the curvature gradient of a region is 0.6 mm^-1, a higher weight can be assigned during interpolation, resulting in a smoother vertex position. This approach can effectively reduce surface abrupt changes and improve the continuity of point cloud data.

[0118] It should be noted that after the updated vertex coordinates are input as new morphological features, the normal vectors of the triangles are recalculated to verify the adjustment effect.

[0119] For example, the normal vector angle of a certain area may be 20 degrees before adjustment, and become 10 degrees after adjustment, indicating that the surface is smoother.

[0120] Specifically, when the average angle change in the local area exceeds the set angle change threshold of 15 degrees, it means that the adjustment range is large and may affect the geometric accuracy of the workpiece surface. In this case, the moving least squares method is used for secondary optimization.

[0121] For example, in areas where the angle varies by up to 18 degrees, a smooth surface is fitted using a local point set, further reducing the normal vector variation to within 12 degrees. This secondary optimization improves the stability of key point coordinates.

[0122] In one embodiment, the key point coordinate set finally outputted must satisfy curvature continuity.

[0123] For example, the adjusted curvature distribution map shows that the curvature gradient in all areas is less than 0.5 mm^-1, indicating that the surface features have been optimized to the design requirements. This continuity ensures the reliability of positioning points in subsequent processes and helps improve processing consistency.

[0124] Optionally, the step S22, wherein the laser tracking technology is used to monitor the spatial relationship between the welding gun position and the designed welding point to determine the spatial deviation, further includes: Step S221 , collecting three-dimensional coordinate data of the welding gun position and the designed welding point through laser tracking technology to generate an initial coordinate set.

[0125] Step S222 , calculating the Euclidean distance between the welding gun coordinates and the designed welding point, and obtaining the spatial deviation value of each sampling point.

[0126] Step S223: If the deviation value of the Euclidean distance is greater than the preset Euclidean distance threshold of 2 mm, the welding gun coordinates are interpolated and adjusted using the bicubic interpolation method, and the adjusted coordinate set is output.

[0127] Step S224 , recalculating the deviation value based on the adjusted coordinates to generate an updated deviation distribution map.

[0128] Step S225 , extracting the X / Y / Z axial deviation components from the deviation distribution map as feature vectors, clustering using the K-means algorithm of scikit-learn, setting the number of clusters to 3, and outputting the high / medium / low deviation area classification results.

[0129] Step S226: For high deviation areas, increase the laser sampling frequency to 200 Hz; In step S227 , the low-to-medium deviation region maintains the original 100 Hz sampling frequency and generates an optimized coordinate set.

[0130] Step S228, based on the optimized coordinate set, calculate the root mean square value of the deviation between the final position of the welding gun and the designed weld point. If it is less than the preset value of 1mm, the process is terminated, otherwise it is iterated again from the coordinate acquisition step.

[0131] Specifically, the process of collecting the three-dimensional coordinate data of the welding gun position and the designed welding point through laser tracking technology and generating an initial coordinate set can be understood as the real-time capture of spatial position using the laser ranging principle.

[0132] For example, when welding a workpiece, a laser tracker determines the coordinates of the welding gun tip by emitting a laser beam and receiving the reflected signal. For example, suppose the welding gun position obtained in a certain acquisition is (100, 50, 20) mm, while the designed weld point is (102, 51, 19) mm, forming an initial data pair. This method is suitable for high-precision scenarios, especially for workpieces with complex curved surfaces, ensuring that the data reflects the actual position. Calculating the Euclidean distance between the welding gun coordinates and the designed weld point to obtain the spatial deviation value is a straightforward way to measure error.

[0133] Specifically, if the welding gun position is (100, 50, 20) mm and the designed welding point is (102, 51, 19) mm, the deviation of the Euclidean distance is about 2.45 mm according to the distance formula between the two points.

[0134] For example, if the Euclidean distance threshold is set to 2mm, any deviation from the threshold indicates that the welding gun position needs adjustment. This deviation analysis can quickly pinpoint problem areas. If the Euclidean distance deviation exceeds 2mm, bicubic interpolation is used to adjust the welding gun coordinates, a technique for smoothing spatial data.

[0135] In one possible implementation, assuming the coordinates of the surrounding sampling points are (99, 49, 21) and (101, 50, 19) mm, bicubic interpolation, through weighted calculation, might adjust the welding gun coordinates to (101.5, 50.5, 19.5) mm. This method leverages neighborhood information to ensure that the adjusted coordinates are closer to the design requirements, improving consistency. Recalculating the deviation values based on the adjusted coordinates and generating a deviation distribution plot provides a visual representation of the error distribution.

[0136] For example, after adjustment, the deviation between the coordinates (101.5, 50.5, 19.5) mm and the designed welding point (102, 51, 19) mm is reduced to 0.71 mm.

[0137] Preferably, the deviation distribution map uses color to distinguish areas, such as green for deviations less than 1mm and red for deviations greater than 2mm, to help quickly identify problem areas. Extracting the X / Y / Z axial deviation components from the deviation distribution map as feature vectors and clustering them using the K-means algorithm is an intelligent classification method.

[0138] Specifically, assuming that the X-axis deviation of a certain area is 0.5mm, the Y-axis is 0.3mm, and the Z-axis is 0.2mm, it may be classified into the low deviation group after clustering.

[0139] In one embodiment, when the number of clusters is set to 3, high deviation areas may be concentrated at the corners of the surface with large Z-axis variations. This classification provides a basis for subsequent optimization.

[0140] Increasing the laser sampling frequency to 200 Hz for high deviation areas is a dynamic adjustment strategy.

[0141] For example, if the deviation in a certain area reaches 2.5mm, increasing the frequency can capture more subtle position changes, such as recording coordinates every 5 milliseconds. For areas with medium and low deviations, maintaining 100Hz is sufficient. For flat areas with a deviation of 0.8mm, maintaining the original frequency saves resources. This hierarchical sampling method increases data density. Calculating the root mean square value of the deviation based on the optimized coordinate set is a comprehensive evaluation method.

[0142] For example, if the deviations of the multiple points after adjustment are 0.7, 0.8, and 0.6 mm, respectively, and the root mean square value is approximately 0.7 mm, which is less than 1 mm, the process is terminated.

[0143] It should be noted that if the RMS value is still 1.2 mm, iterative data collection is required to gradually approach the design standard. This iteration ensures the final accuracy.

[0144] In one example, after iteration, the welding gun position was adjusted from (100, 50, 20) mm to (101.8, 50.9, 19.2) mm, with the deviation reduced to 0.3 mm. This gradual optimization approach effectively improves the reliability of welding gun positioning and lays the foundation for subsequent welding processes.

[0145] S3 adjusts the motion parameters of the high-precision robotic arm according to the error distribution, combines force feedback technology to detect slight deformations on the workpiece surface during welding, determines the thermal deformation trend and generates dynamic adjustment instructions.

[0146] Optionally, this step also includes: In step S31 , force feedback data of the welding workpiece surface is collected by a force sensor, and an error distribution curve is fitted using a least squares method to obtain an error distribution feature.

[0147] Step S32: According to the error distribution characteristics, a PID controller is used to adjust the motion parameters of the robot arm to determine the initial motion trajectory.

[0148] Step S33: During the movement of the robotic arm, a force sensor is used to detect deformation data of the surface of the welding workpiece in real time to obtain a small deformation value.

[0149] Step S34: input the micro-deformation data into a linear regression model to analyze the thermal deformation trend and determine the direction of deformation change.

[0150] Step S35 , generating dynamic instructions using a fuzzy control algorithm based on the thermal deformation trend to adjust the motion parameters of the robotic arm.

[0151] In step S36, the adjusted parameters are compared with preset parameter thresholds, where the parameter thresholds are set according to welding process requirements, to obtain an optimized motion trajectory.

[0152] Step S37: drive the robotic arm to perform welding operations through the optimized motion trajectory, use a laser displacement sensor to detect welding accuracy, and complete high-precision adjustment.

[0153] Specifically, when force feedback data of the surface of a welding workpiece is collected by a force sensor, a highly sensitive sensor can be used to capture the pressure changes on the workpiece during the welding process.

[0154] For example, when welding an aluminum alloy workpiece, a force sensor might detect an increase in pressure from 5N to 8N at a specific surface point. This change reflects the local stress distribution during welding. Using the least squares method to fit the error distribution curve, a smooth error curve can be generated based on multiple acquisitions of force feedback data.

[0155] For example, after collecting force data 10 times, the least squares method was used to calculate the trend line of the error distribution, resulting in an error mean of 0.2 N and a standard deviation of 0.05 N. This method clearly reflects the regularity of force feedback. Based on the error distribution characteristics, when using a PID controller to adjust the motion parameters of the robotic arm, the response speed and stability of the robotic arm can be optimized through the proportional, integral, and differential components.

[0156] Specifically, if the error distribution indicates a large pressure deviation, the PID controller can increase the proportional gain, adjusting the robot's movement speed from 0.5 mm / s to 0.6 mm / s while reducing overshoot. This determines the initial motion trajectory, ensuring the robot quickly approaches the target position.

[0157] In one possible implementation, if the target pressure is 7 N and the actual pressure is 7.3 N, the robotic arm can move smoothly to the appropriate position after PID adjustment. During the robotic arm's movement, the force sensor can be used to detect the deformation data of the welding workpiece surface in real time, capturing tiny deformation values.

[0158] For example, the surface of a workpiece may deform by 0.03 mm due to heat. This tiny change can be accurately recorded by a force sensor with high-frequency sampling.

[0159] It should be noted that this real-time detection can promptly reflect changes in the workpiece state. When inputting small deformation data into a linear regression model to analyze thermal deformation trends, the direction of change can be determined based on the time series of deformation values.

[0160] In one embodiment, if the deformation increases from 0.01 mm to 0.03 mm, the linear regression model may predict that the deformation will continue to increase, pointing toward the center of the workpiece. Based on the thermal deformation trend, when using a fuzzy control algorithm to generate dynamic instructions, the robot arm motion parameters can be adjusted based on fuzzy rules.

[0161] For example, if the deformation trend indicates that the workpiece is expanding, the fuzzy controller might output a command to lower the robot arm's height by 0.2mm and simultaneously slow its movement speed to 0.4mm / s. This dynamic adjustment adapts to real-time changes in the workpiece. The adjusted parameters are then compared with a preset threshold, which can be set to a deformation of no more than 0.05mm.

[0162] Preferably, if the deformation after adjustment is 0.02 mm, which is less than the parameter threshold, an optimized motion trajectory is generated to ensure the stability of the welding process. When the robot arm is driven to perform the welding operation by the optimized motion trajectory, the laser displacement sensor can be used to detect the welding accuracy.

[0163] For example, after welding a 10cm long weld, the laser displacement sensor measured a maximum weld height deviation of 0.01mm.

[0164] It is understood that this high-precision adjustment can ensure the uniformity and consistency of the weld.

[0165] In one possible implementation, if the deviation of a weld seam is 0.015mm, the deviation can be reduced to 0.005mm through trajectory optimization. This method significantly improves welding quality and provides reliable assurance for subsequent workpiece assembly.

[0166] S4 updates the welding gun trajectory control parameters by dynamically adjusting the instructions, integrates the laser tracking data to optimize the welding path angle in real time, and determines the precise motion trajectory of each weld.

[0167] Specifically, when first using a laser sensor to collect weld position data, it can be understood that the laser sensor measures the geometric characteristics of the weld by emitting a laser beam and receiving a reflected signal.

[0168] For example, a sensor scans a 500mm-long weld at 10mm intervals, collecting 50 raw data points. Each point contains distance and angle information. Because real-world environments may contain light interference or surface reflections, a Kalman filter algorithm is used for noise reduction.

[0169] In one possible implementation, suppose the original distance data for a certain point is 2.05 mm, but noise interference causes it to deviate from the true value. After iterative Kalman filtering, it is corrected to 2.00 mm, resulting in smoothed data that is closer to the actual weld contour. This method has the advantage of effectively filtering out random noise, improving the reliability of subsequent path generation.

[0170] Based on the filtered weld profile information, when the cubic spline curve fitting algorithm is applied to generate the welding path, specifically, features can be extracted from 50 data points to generate a continuous curve.

[0171] Preferably, assuming there is a slight offset of 5mm in the middle of the weld, the cubic spline curve can naturally transition this change, forming a smooth trajectory and avoiding sudden changes when the welding gun moves. This smoothness is crucial to maintaining welding consistency.

[0172] When extracting key control points from the path curve, for example, the weld start, midpoint, and end point can be selected as initial control points, with the welding gun angles calculated as 0°, 45°, and 90°, respectively, and the movement speeds as 10mm / s, 12mm / s, and 8mm / s, respectively. This segmented control facilitates precise adjustment. This is combined with the welding process parameter library to determine process parameters.

[0173] In one embodiment, for steel plate welding, the current of the starting control point can be set to 150A and the voltage to 22V, while the midpoint is adjusted to 160A and 23V due to the increase in speed.

[0174] It should be noted that the parameter library can preset multiple sets of data according to material thickness and weld type to ensure adaptability. When using linear interpolation algorithm to generate dense trajectory point sequence.

[0175] For example, on a 250mm path from the starting point to the midpoint, 100 trajectory points are interpolated, with each point spaced 2.5mm apart, and parameters transition smoothly and proportionally. This dense point sequence improves the continuity of the welding gun's motion. This is also true when calculating the motion angles of each welding gun joint.

[0176] It can be understood that based on the robot kinematic model, the trajectory points are converted into joint angle sequences.

[0177] For example, when a welding gun moves from the starting point to the midpoint, the angle of joint 1 may change from 0° to 30°, and the angle of joint 2 may be adjusted from 10° to 45°.

[0178] When the control instruction sequence is generated and sent to the control system, in one possible implementation, the instructions are issued in real time with a cycle of 50ms to ensure the real-time performance of trajectory tracking.

[0179] For example, the robot can track a 100mm weld within 10 seconds, maintaining a steady rhythm of motion. This real-time performance is especially important for complex welds.

[0180] S5, obtains real-time thermal expansion data during welding execution, uses optical measurement technology to monitor the temperature field distribution in the weld area, and determines whether the thermal deformation adjustment exceeds the thermal deformation threshold.

[0181] Specifically, the thermal expansion data of the weld area is collected at 100 frames per second by a high frame rate infrared camera, and the output grayscale image sequence is 640×480 pixels.

[0182] Understandably, this high frame rate acquisition is able to capture the rapidly changing temperature distribution during the welding process.

[0183] For example, on a 300mm long weld, the camera generates 300 frames of images in 3 seconds, each frame reflecting the thermal state of different parts of the weld.

[0184] For example, if the weld near its starting point becomes brighter due to concentrated heat, the pixel grayscale value may rise from 100 to 200, while areas farther from the heat source remain around 50. This high-resolution data provides rich raw information for subsequent analysis. Gaussian filtering and denoising are performed on the image sequence using the OpenCV library to generate a smoothed temperature field data matrix.

[0185] Specifically, Gaussian filtering reduces random noise points in an image by blurring.

[0186] In one possible implementation, suppose a pixel in a frame of image displays an outlier value of 180 due to environmental interference, while the surrounding pixels have an average value of 120. After filtering, this point is smoothed to 130. This denoising method preserves the overall trend of the temperature field, facilitating subsequent modeling. The temperature field data matrix is fed into a ResNet18 network built with TensorFlow, which outputs the lateral expansion rate and longitudinal displacement gradient.

[0187] It should be noted that ResNet18 extracts the spatial features of the temperature field through deep convolution.

[0188] Optimally, in the middle of a weld, the network might detect a lateral expansion rate of 0.5 mm / s and a longitudinal displacement gradient of 0.2 mm / mm. These parameters reflect the directionality and intensity of thermal expansion, providing a basis for path adjustment. Based on a preset 2 mm thermal deformation threshold, displacement data is monitored in real time, and outliers are flagged.

[0189] For example, if the displacement of a weld point suddenly increases from 1.8mm to 2.3mm at a given moment, the system will record a timestamp of, for example, 2.45 seconds and mark it as an anomaly. This thermal deformation threshold monitoring allows for timely detection of potential problems. A 3x3 mean filter is applied to a 10x10 pixel area surrounding the anomaly to generate an adjusted temperature field distribution matrix.

[0190] In one embodiment, if the grayscale value of an outlier point is 200 and the mean value of the surrounding area is 150, it is adjusted to 160 after filtering. This local smoothing process can reduce the interference of outliers and maintain data consistency. If the adjusted data still exceeds the thermal deformation threshold for five consecutive samplings, the linear regression prediction module is activated.

[0191] Understandably, this mechanism is designed to deal with persistent anomalies.

[0192] For example, if the data for five frames are 2.1mm, 2.2mm, 2.3mm, 2.1mm, and 2.2mm, respectively, this indicates that thermal deformation has not eased and further prediction is needed. Scikit-learn is used to fit the relationship between thermal deformation and time for the past 30 frames and output predictions for the next five frames.

[0193] Specifically, if the past 30 frames show a linear increase in thermal deformation from 1.5mm to 2.0mm, it is predicted that the thermal deformation may reach 2.1mm to 2.3mm in the next five frames. This prediction provides data support for early intervention. The root mean square error (RMS) between the predicted result and the real-time data is calculated. When the error is less than 0.5mm, the thermal deformation adjustment is considered stable and a report is generated.

[0194] For example, if the predicted value is 2.2mm and the real-time value is 2.1mm, the error is 0.1mm, indicating that the adjustment effect is good. This error assessment ensures the reliability of the system and provides an important reference for welding quality control.

[0195] S6, if the thermal deformation adjustment exceeds the thermal deformation threshold, the compensation vector is calculated through force feedback technology, and the posture control data of the high-precision manipulator is updated to obtain a stable complex structure positioning state.

[0196] Optionally, this step also includes: Step S61: If the posture error caused by thermal deformation exceeds the preset thermal deformation threshold of ±0.1 mm, the joint force data is obtained through the six-dimensional force sensor.

[0197] Step S62 , calculating the compensation vector based on the homogeneous matrix, and updating the control parameters of the servo motors of the respective axes of the robotic arm.

[0198] Step S63: drive the robot arm to move according to the updated control parameters, and use a laser tracker to measure the actual end position.

[0199] Step S64: When the deviation between the actual position and the target position is greater than ±0.05 mm, a quadratic compensation vector is calculated using the least squares method.

[0200] In step S65 , the quadratic compensation vector is superimposed on the motor control parameters to stabilize the end of the robotic arm within a repeatability range of ±0.02 mm.

[0201] In step S66, a 5-megapixel CCD camera is used to capture the workpiece contour, and the edge pixel coordinates are extracted using the Canny operator of the Halcon library.

[0202] In step S67 , the edge coordinates are compared with the CAD model to generate a three-dimensional deviation vector, which is input into a pre-trained SVM classifier.

[0203] Step S68: Adjust the joint angle of the end of the robot arm according to the SVM output, and finally achieve a contour tracking error of less than ±0.01mm.

[0204] Specifically, during the welding process, if thermal deformation causes the position of the end of the robotic arm to deviate, the six-dimensional force sensor can sense the force changes at the joint.

[0205] For example, a sensor mounted on a robotic arm joint might detect an increase in axial force from 5 N to 15 N when a workpiece expands due to high temperature, causing the end to deviate from the target position. This data directly reflects the impact of thermal deformation and provides a basis for subsequent adjustments.

[0206] As you can see, the six-dimensional force sensor comprehensively captures the dynamic state of the robotic arm by measuring the six components of force and torque. Based on this force data, a homogeneous matrix can be used to calculate the compensation vector and update the robotic arm control parameters.

[0207] Specifically, the matrix describes the transformation relationship from the current position of the end point to the target position.

[0208] In one possible implementation, if the end point deviates by 0.15 mm along the X-axis, the compensation vector calculated by the matrix may indicate that the motor needs to be adjusted in the opposite direction by 0.12 mm. This method ensures the accuracy of the adjustment direction.

[0209] Preferably, the updated parameters are sent to the servo motor in real time to improve response speed. The robot arm then moves according to the new parameters, and the laser tracker is used to verify the end position.

[0210] It should be noted that the laser tracker accurately measures the end coordinates by emitting a light beam and receiving a reflected signal.

[0211] For example, if the target position is X = 100mm and the actual measured value is 100.06mm, the deviation is 0.06mm, exceeding the ±0.05mm tolerance. This high-precision measurement provides reliable data for further optimization. If the deviation exceeds the tolerance, the least squares method can be used to calculate the quadratic compensation vector.

[0212] In one embodiment, if multiple measurements show a deviation fluctuating between 0.05mm and 0.07mm, the algorithm generates a 0.03mm compensation value by fitting the data. This approach reduces error accumulation and makes the system more stable. Secondary compensation is then added to the motor parameters, ultimately bringing the end-point accuracy to within ±0.02mm, effectively improving repeatability. To further verify the workpiece's condition, a 5-megapixel CCD camera captures the workpiece's contour, and the Halcon library's Canny operator extracts edge information.

[0213] Specifically, the camera captures the weld edge with high resolution, and the Canny operator outputs clear pixel coordinates by detecting grayscale gradients.

[0214] For example, the coordinates of a point on the weld edge might be (320, 240), reflecting the true contour location. This edge extraction lays the foundation for subsequent comparison. After comparing the edge coordinates with the CAD model, the resulting 3D deviation vector is input into the SVM classifier.

[0215] It is understandable that the deviation vector may show that a certain area deviates from the design value by 0.02 mm, and the SVM determines whether it is within the acceptable range based on the training data.

[0216] In one possible implementation, the classifier output indicates that the deviation needs to be adjusted, and the joint angle needs to be fine-tuned by 0.1 degrees. This intelligent classification improves decision-making efficiency. Finally, based on the SVM results, the joint angle of the end arm is adjusted to keep the contour tracking error within ±0.01mm.

[0217] For example, if a section of the weld is slightly convex due to heat, the tip can be adjusted to accurately follow the contour change. This high-precision tracking significantly improves welding consistency and ensures product quality.

[0218] S7, performs micron-level weld point positioning operations based on the stable positioning state, integrates the dynamic path optimization results to generate the final welding instructions, and outputs consistent weld quality data.

[0219] Optionally, this step also includes: Step S71: Acquire stable positioning state data through sensors, and obtain positioning coordinates using Kalman filtering.

[0220] Step S72: For the positioning coordinates, the weighted average method is used to fuse the dynamic path information to obtain the path optimization result.

[0221] Step S73: Generate a welding instruction sequence based on the path optimization result using a rule-based method and output an instruction set.

[0222] Step S74: Obtain welding execution parameters from the instruction set, perform simulation using finite element analysis, and determine consistency quality indicators.

[0223] In step S75, principal component analysis is used to extract weld quality characteristics based on the consistency quality index to obtain a quality assessment value.

[0224] Step S76: Compare the quality evaluation value to the quality index threshold. If it exceeds the threshold, adjust the path optimization parameters using the gradient descent method and regenerate the instruction set.

[0225] Step S77: Use the weighted update method to update the data output content to obtain the final weld quality data.

[0226] Specifically, in welding robot systems, the positioning state data acquired by sensors often contains noise and interference. Kalman filtering, as a recursive estimation algorithm, can effectively handle these uncertainties.

[0227] For example, if the laser sensor measures the robot's end position as (100.2, 200.3, 300.1) mm, the Kalman filter might output a more accurate estimate of (100.15, 200.25, 300.05) mm. This filtering not only improves positioning accuracy but also allows for prediction of the next position. After obtaining a precise positioning, a weighted average method is used to integrate dynamic path information.

[0228] Specifically, this method comprehensively considers multiple factors such as trajectory smoothness and obstacle avoidance requirements.

[0229] In one possible implementation, if the original path points are A(10, 20, 30) and B(15, 25, 35), after accounting for workpiece thermal deformation, the optimized results might be adjusted to A'(10.2, 20.1, 30.3) and B'(15.1, 25.2, 35.2). This optimization ensures real-time adaptability of the welding path. Based on the path optimization results, the system uses a rule-based approach to generate welding instructions.

[0230] It should be noted that these rules cover the adjustment logic of parameters such as welding speed and current.

[0231] For example, when the path angle is greater than 30°, the welding speed is automatically reduced by 20% to ensure weld quality. This rule-based approach makes instruction generation more flexible and controllable. To verify the feasibility of the instructions, the system uses finite element analysis simulation.

[0232] It is understood that this analysis can predict the thermal stress distribution and deformation.

[0233] For example, a simulation might reveal a temperature of 1200°C and a stress concentration of approximately 20 MPa at a certain point. Based on this data, the system can evaluate and adjust parameters before actual welding, effectively preventing defects. During quality assessment, principal component analysis (PCA) is used to extract key features.

[0234] Optimally, PCA may identify weld width and depth ratio as the two most significant features from the multidimensional data. This dimensionality reduction not only simplifies subsequent analysis but also reveals key factors influencing quality. If the quality assessment value exceeds the quality index threshold, gradient descent is used to adjust the optimization parameters.

[0235] In one example, if the weld width is found to be 5% too wide, the algorithm might recommend reducing the welding current by 2A and increasing the travel speed by 0.5mm / s. This fine-tuning ensures continuous optimization of the welding process. Finally, a weighted update method is used to generate the final quality data.

[0236] Specifically, the method considers both historical data and new measurement results.

[0237] For example, if the historical average weld strength is 500 MPa and the new measurement is 520 MPa, the updated assessment value may be 510 MPa. This method preserves long-term trends while reflecting the latest changes, providing a reliable basis for quality control.

[0238] Optionally, the step S74 of obtaining welding execution parameters from the instruction set, performing simulation using finite element analysis, and determining consistency quality indicators may also include: Step S741: parse the current, voltage, and speed parameters from the welding equipment control instruction set to generate a structured parameter table.

[0239] Step S742: Create a three-dimensional welding geometry model in ANSYS TwinBuilder based on the parameter table, and import the thermophysical parameters in the material property library.

[0240] In step S743, the finite element method is used to solve the Fourier heat conduction equation to calculate the temperature field distribution data of the weld area, and the time step is set to 1 second.

[0241] In step S744, the temperature field data is input as initial conditions into the ABAQUS explicit dynamics module, and the Mises stress tensor of the weld area is output.

[0242] In step S745 , the stress tensor is reduced in dimension using the PCA tool of scikit-learn, and the first three principal components whose cumulative contribution rates exceed a contribution rate threshold of 95% are retained.

[0243] In step S746, the principal component data is divided into a training set and a test set at a ratio of 7 to 3, a three-layer fully connected network is constructed in TensorFlow, and the quality prediction model is trained using the ReLU activation function.

[0244] Step S747: When the predicted crack sensitivity coefficient exceeds the first sensitivity coefficient threshold of 35, the genetic algorithm module of the DEAP library is called to perform iterative optimization with the parameter combination as the chromosome and the predicted value as the fitness.

[0245] Step S748: finally outputting a parameter combination that makes the crack sensitivity coefficient lower than the second sensitivity coefficient threshold 28.

[0246] Specifically, a welding equipment control instruction set usually includes a variety of parameter information.

[0247] For example, a typical instruction might include data such as current 200A, voltage 28V, and welding speed 5mm / s. After parsing these instructions, a structured parameter table can be generated to facilitate subsequent modeling and analysis. ANSYS TwinBuilder provides powerful support for building 3D welding geometry models.

[0248] Specifically, an accurate model of the weld and its surrounding area can be constructed based on the actual dimensions of the workpiece.

[0249] For example, for a T-joint, you might need to model a 200mm x 100mm x 10mm plate and a 3mm wide weld along the joint line. The thermophysical parameters in the material property library have a significant impact on the simulation results.

[0250] It's important to note that these parameters typically include density, specific heat capacity, thermal conductivity, and other parameters, and they often vary with temperature. For example, the thermal conductivity of mild steel at room temperature is approximately 50 W / (m·K), but at 1000°C it can drop to around 30 W / (m·K). Solving the Fourier heat conduction equation using the finite element method is a key step in simulating temperature field distributions.

[0251] One possible implementation involves dividing the weld area into tens of thousands of mesh elements and calculating heat transfer once a second. This approach effectively captures the transient thermal effects of welding. By inputting the temperature field data into the ABAQUS Explicit Dynamics module, detailed stress distribution information can be obtained.

[0252] Preferably, these data may include stress states at different locations and at different times.

[0253] For example, a peak stress of 800 MPa may be observed at the center of the weld, while the stress in areas away from the weld may drop to below 100 MPa. Using PCA for dimensionality reduction is an effective method for dealing with high-dimensional stress data.

[0254] It is understandable that this method can extract the most representative features in the data.

[0255] In one embodiment, if the original data contains hundreds of stress components, PCA may simplify them into three main features, such as maximum principal stress, stress gradient, and stress concentration factor. When training a quality prediction model, the selection of a fully connected network structure is crucial.

[0256] For example, a possible network structure is: 3 neurons in the input layer (corresponding to the three principal components), 20 and 10 neurons in the hidden layer, and 1 neuron in the output layer (to predict the crack sensitivity coefficient). This structure can capture the complex nonlinear relationships in the data while avoiding the risk of overfitting.

[0257] S8, after obtaining the weld quality data, uses laser tracking technology to detect weld deviation and geometric consistency, determines whether the welding quality meets the design requirements, and generates a quality verification report.

[0258] Step S81: Use a 3D scanner to obtain 3D point cloud data of the weld.

[0259] In step S82 , the acquired three-dimensional point cloud data of the weld is registered with the pre-established ideal model through the ICP algorithm to obtain aligned three-dimensional point cloud data.

[0260] Step S83: Calculate the deviation value of each weld point based on the registration result, and use Matplotlib to generate a weld deviation distribution diagram.

[0261] Step S84: Calculate the maximum deviation value and the average deviation value on the deviation distribution graph using Numpy.

[0262] In step S85, the least square method is used to perform curve fitting on the weld profile, and the error between the fitting curve and the actual profile is used to evaluate the geometric consistency.

[0263] Step S86: Determine whether the weld quality is qualified based on the preset error threshold and consistency standard.

[0264] Step S87: If the deviation exceeds the error threshold or the consistency does not meet the standard, it is marked as an unqualified weld.

[0265] Step S88: Based on the quality assessment results and related data, a weld quality verification report is automatically generated using Python.

[0266] Specifically, using a 3D scanner to obtain 3D point cloud data of welds is an important starting point for weld quality analysis.

[0267] For example, in a welding robot system, a 3D scanner scans the surface of a workpiece using laser or structured light to generate a 3D point cloud containing millions of points.

[0268] For example, a 50mm weld might generate approximately 100,000 points of data, recording the weld's height, width, and depth. This method fully captures the weld's geometric characteristics, providing a foundation for subsequent analysis. Aligning the point cloud data with the ideal model using the ICP algorithm is a key step.

[0269] It should be noted that the ICP algorithm iteratively calculates the distance between point clouds and continuously adjusts the position and posture to achieve the best alignment between the two.

[0270] In one possible implementation, assuming the ideal model is a smooth, straight weld, while the actual point cloud is slightly offset due to welding vibration, ICP registration can translate and rotate the entire point cloud to align it with the model. This alignment ensures the accuracy of subsequent deviation calculations. Calculating deviation values based on the registration results and generating a distribution plot provides a visual representation of weld quality.

[0271] Specifically, the distance between each point and the ideal model is recorded as a deviation value, and a color heat map can be drawn using Matplotlib.

[0272] For example, the deviation near the edge of a weld might reach 0.5 mm, while the center area might only be 0.1 mm. A heat map displays the deviation distribution through color depth. This visualization facilitates quick identification of problem areas. Using Numpy to calculate the maximum and average deviation values on the deviation distribution plot is the core of quantitative analysis.

[0273] For example, the maximum deviation of a weld might be 0.6mm and the average deviation might be 0.15mm. These values directly reflect the overall consistency of the weld.

[0274] Preferably, if the maximum deviation is concentrated in a certain section, further analysis can be conducted to determine whether it is caused by equipment vibration or material inhomogeneity. Using the least squares method to curve fit the weld profile is an effective means of evaluating geometric consistency.

[0275] In one embodiment, assuming a weld point cloud exhibits slight undulations, the least squares method can generate a smooth curve with an error within 0.05 mm. This fitting not only quantifies the degree of contour deviation but also provides a basis for quality assessment. Determining weld quality based on error thresholds and consistency criteria is the decision-making process.

[0276] Understandably, if the maximum error threshold is set at 0.4mm, and a weld's deviation at a certain point reaches 0.5mm, it will be deemed unqualified. This standardized judgment logic enhances the objectivity of the assessment. If the deviation exceeds the error threshold or the consistency does not meet the standard, marking the weld as unqualified is necessary.

[0277] For example, if a weld's width exceeds the specified tolerance due to thermal deformation, the system will automatically mark and record the exact location. This marking facilitates subsequent rework or adjustment of process parameters. Generating a verification report based on the quality assessment results concludes the process.

[0278] In one embodiment, Python can integrate the deviation distribution graph, maximum deviation value, and acceptance judgment to generate a PDF report with a timestamp and part number. This automated report not only improves efficiency but also provides a reliable basis for quality traceability.

[0279] S9, based on the deviation data in the quality verification report, adjusts the real-time adjustment capability parameters, optimizes the welding gun trajectory control accuracy of subsequent welding tasks, and outputs the improved process parameter set.

[0280] Specifically, the current process parameter set and the corresponding quality verification result data are obtained, and the Pearson correlation coefficient is used to calculate the matching value between the parameter set and the quality result; if the matching value is lower than the preset threshold, the parameter optimization process is triggered; the linear regression method in the scikit-learn library is used to construct a mathematical model of parameters and quality indicators, and the degree of influence of each parameter on the quality is determined according to the coefficients of the regression model; for parameters with a greater degree of influence, the batch gradient descent method is used to adjust the parameter value; the scikit-learn library is used to perform quality prediction on the adjusted parameter set to verify the optimization effect.

[0281] Specifically, after obtaining the deviation data from the welding process, the range of outliers can be determined by the mean variance analysis method.

[0282] For example, on an automated welding production line, weld deviation data may be caused by welding torch vibration or thermal expansion and contraction of the material. Assuming the deviation data for a particular weld section exhibits a normal distribution, mean-variance analysis can calculate the mean and standard deviation of the data to determine a reasonable range of abnormalities. For example, points exceeding twice the standard deviation are considered abnormal. This method can quickly screen potential problem points and lay the foundation for subsequent analysis.

[0283] It is understandable that this statistical method is simple and efficient, and is particularly suitable for real-time monitoring scenarios. After the outlier range is determined, extracting the influencing factors with correlation coefficients greater than the correlation coefficient threshold of 5 is a key step.

[0284] Specifically, influencing factors might include torch speed, current intensity, or workpiece temperature. Suppose analysis reveals a positive correlation between torch speed and deviation, with a correlation coefficient of 6, while temperature and deviation are negatively correlated, with a correlation coefficient of -5.5. This suggests that if the speed is too high, it should be reduced, and if the temperature is too low, it should be increased. This determination of positive and negative correlations provides clear guidance for adjustment.

[0285] It should be noted that setting the correlation coefficient threshold to 5 is an empirical value that can effectively filter out minor interference items. When adjusting the welding capability parameters based on the influencing factors, a linear adjustment method can be used to update the control model.

[0286] In one possible implementation, if welding gun speed is confirmed to be the primary influencing factor, it can be adjusted linearly from 50 mm / s to 45 mm / s while observing the deviation. This adjustment, based on real-time data feedback, can gradually approach the optimal parameter state.

[0287] Preferably, linear adjustment avoids instability caused by large changes and improves the robustness of the model. The least squares method is a practical tool when calculating the welding gun trajectory based on the updated control model.

[0288] For example, the ideal trajectory of a weld seam is a straight line, but the actual trajectory deviates slightly due to vibration. The least squares method can be used to fit the trajectory and generate a smooth path that is closest to the actual point cloud. This method effectively smooths out trajectory fluctuations and ensures the accuracy of the welding gun.

[0289] In one embodiment, the adjusted trajectory instruction can reduce the weld width deviation from 0.3 mm to within 0.1 mm. After optimizing the welding task through trajectory control instructions, determining the process parameter set is the core of the process.

[0290] For example, a process parameter set may include a voltage of 25V, a current of 150A, and a speed of 45mm / s. The output of this parameter set is based on the previous deviation analysis and trajectory optimization, and can directly guide the operation of the welding equipment.

[0291] It can be understood that the rationality of the parameter set directly determines the quality stability of the weld.

[0292] When calculating the matching degree between the process parameter set and the quality verification results, if it is lower than the preset matching degree threshold, it can be further optimized through regression analysis.

[0293] For example, if the matching threshold is 90%, but the actual calculated result is only 85%, the regression analysis can re-evaluate the weight of each parameter and find that the high current is the main problem, and then adjust it from 150A to 140A.

[0294] In one embodiment, the adjusted parameter set increased the matching degree to 92%, effectively improving welding consistency. This iterative optimization method provides data support for process improvement.

[0295] like Figure 2 As shown, the second aspect of the present invention provides an aluminum alloy subframe welding precise positioning and quality control system, which uses the above-mentioned method to accurately position and control the quality of aluminum alloy subframe welding. The system mainly includes: The geometric model acquisition module is used to obtain the 3D geometric model of the aluminum alloy subframe and the coordinates of the designed weld points. It uses optical measurement technology to scan the workpiece surface, generate real-time surface point cloud data, and determine the initial welding gun trajectory and multi-dimensional positioning benchmark; The point cloud processing module is used to calculate the curvature changes and key positioning point coordinates of the complex structure of the workpiece based on the point cloud data. Laser tracking technology is used to monitor the spatial deviation between the welding gun and the designed weld point to obtain the micron-level positioning error distribution; The error monitoring module is used to adjust the motion parameters of the high-precision robotic arm according to the error distribution. It combines force feedback technology to detect small deformations on the workpiece surface during welding, determine the thermal deformation trend, and generate dynamic adjustment instructions. The motion control module is used to update the welding gun trajectory control parameters through dynamic adjustment instructions, integrate laser tracking data to optimize the welding path angle in real time, and determine the precise motion trajectory of each weld; The trajectory optimization module is used to obtain real-time thermal expansion data during welding, monitor the temperature field distribution in the weld area using optical measurement technology, and determine whether the thermal deformation adjustment exceeds the thermal deformation threshold; The thermal deformation monitoring module is used to calculate the compensation vector through force feedback technology if the thermal deformation adjustment exceeds the thermal deformation threshold, update the posture control data of the high-precision manipulator, and obtain a stable complex structure positioning state; The compensation correction module is used to perform micron-level weld point positioning operations based on the stable positioning state, integrate the dynamic path optimization results to generate the final welding instructions, and output consistent weld quality data; The welding execution module is used to obtain weld quality data, detect weld deviation and geometric consistency through laser tracking technology, determine whether the welding quality meets the design requirements, and generate a quality verification report; The quality assessment module is used to adjust the real-time adjustment capability parameters based on the deviation data in the quality verification report, optimize the welding gun trajectory control accuracy of subsequent welding tasks, and output the improved process parameter set.

[0296] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for precise positioning and quality control of aluminum alloy subframe welding, characterized in that: The method comprises: S1, obtains the 3D geometric model and designed weld point coordinates of the aluminum alloy subframe, scans the workpiece surface using optical measurement technology, generates real-time surface point cloud data, and determines the initial welding gun trajectory and multi-dimensional positioning benchmark; S2 calculates the curvature change of the complex structure of the workpiece and the coordinates of key positioning points based on the point cloud data. Laser tracking technology is used to monitor the spatial deviation between the welding gun and the designed welding point to obtain the micron-level positioning error distribution; S3 adjusts the motion parameters of the high-precision robotic arm based on the error distribution, combines force feedback technology to detect minute deformations on the workpiece surface during welding, determines thermal deformation trends, and generates dynamic adjustment instructions; S4 updates the welding gun trajectory control parameters through dynamic adjustment instructions, integrates laser tracking data to optimize the welding path angle in real time, and determines the precise motion trajectory of each weld; S5, obtains real-time thermal expansion data during welding, uses optical measurement technology to monitor the temperature field distribution in the weld area, and determines whether the thermal deformation adjustment exceeds the thermal deformation threshold; S6, if the thermal deformation adjustment exceeds the thermal deformation threshold, the compensation vector is calculated through force feedback technology, and the posture control data of the high-precision manipulator is updated to obtain a stable complex structure positioning state; S7, performs micron-level welding point positioning operations based on the stable positioning state, integrates the dynamic path optimization results to generate the final welding instructions, and outputs consistent weld quality data; S8, after obtaining the weld quality data, uses laser tracking technology to detect weld deviation and geometric consistency, determines whether the welding quality meets the design requirements, and generates a quality verification report; S9, based on the deviation data in the quality verification report, adjusts the real-time adjustment capability parameters, optimizes the welding gun trajectory control accuracy of subsequent welding tasks, and outputs the improved process parameter set.

2. The method according to claim 1, characterized in that Step S1, obtaining a three-dimensional geometric model of the aluminum alloy subframe and the coordinates of the designed welding points, scanning the workpiece surface using optical measurement technology, generating real-time surface point cloud data, and determining the initial welding gun trajectory and multi-dimensional positioning reference, includes: Step S11, obtaining a three-dimensional geometric model of the aluminum alloy subframe and extracting coordinate data of key feature points; Step S12, determining the welding position coordinates according to the characteristic point distribution of the geometric model; Step S13, scanning the workpiece surface to generate point cloud data containing three-dimensional coordinate information; Step S14: Compare and analyze the point cloud data obtained by scanning with the geometric model, and calculate the deviation value of each point; Step S15: Set a point deviation threshold. If the point cloud data exceeds the point deviation threshold, use the K-means algorithm to perform cluster analysis on the abnormal points to generate an optimized point cloud set. Step S16, replanning the welding gun motion trajectory based on the optimized point cloud data to ensure that the welding position is consistent with the designed position; Step S17, fusing the re-planned welding gun trajectory with the original coordinate data to generate a complete welding path program; Step S18: import the welding path program into the industrial robot control system to complete the welding operation.

3. The method according to claim 2, characterized in that The step S12, determining the welding position coordinates according to the characteristic point distribution of the geometric model, includes: Step S121, loading the geometric model into CATIA software, and extracting the coordinate data of the model edge points and surface points; Step S122, dividing the regions according to the curvature variation of adjacent points, and outputting the coordinate text file of the partitioned point set; Step S123, performing Z-score normalization processing on the point set coordinates, clustering based on Euclidean distance, determining the optimal number of clusters by the silhouette coefficient, and outputting the grouped point set; Step S124, calculating the standard deviation of the spacing between each group of points. If it exceeds the standard deviation threshold, the point set is exported as an STL file; Step S125, importing the STL file into Geomagic software, using the curvature optimization tool to adjust the point set distribution, and saving the adjusted coordinate data in ASC format; Step S126, importing the ASC file into RobotStudio, setting the robot tool center coordinate system, and calling the CollisionCheck function to generate a collision-free welding path; Step S127: Load the path program through the ABB controller to execute the digital output of the welding coordinate points.

4. The method according to claim 1, wherein The step S2 calculates the curvature change of the complex structure of the workpiece and the coordinates of key positioning points based on the point cloud data, uses laser tracking technology to monitor the spatial deviation between the welding gun and the designed welding point, and obtains the micron-level positioning error distribution, including: Step S21, obtaining the surface features of the complex structure of the workpiece through point cloud data, and calculating the curvature change and the coordinates of the positioning points; Step S22, using laser tracking technology to monitor the spatial relationship between the welding gun position and the designed welding point to determine the spatial deviation; Step S23, performing statistical analysis on the spatial deviation distribution to obtain micron-level error distribution characteristics; Step S24: if the error distribution exceeds the error distribution threshold, the coordinates of the positioning points are adjusted by an interpolation algorithm to obtain optimized coordinate data; Step S25, updating the curvature change trend of the complex structure according to the optimized coordinate data, and determining the surface feature adjustment direction; Step S26, obtaining the adjusted curvature change trend, correcting the welding gun position through geometric mapping technology, and determining the final deviation control result; Step S27 : Compare the final deviation control result with the designed solder joint to obtain verification data of micron-level positioning error distribution.

5. The method according to claim 4, characterized in that The step S21, obtaining the surface features of the complex structure of the workpiece through point cloud data, and calculating the curvature change and the coordinates of the positioning points, includes: Step S211, registering the acquired point cloud data using the ICP algorithm, extracting vertex coordinates and normal vectors as initial morphological features; Step S212, constructing a topological relationship based on adjacent triangles, and calculating the average curvature value of each vertex using a discrete curvature estimation method; Step S213, mapping the curvature values to vertex coordinates to generate a curvature distribution map; Step S214, setting a change threshold on the curvature distribution map, and marking abnormal areas where the curvature gradient exceeds the change threshold; Step S215: For the vertices in the abnormal area, the coordinates are recalculated using a thin plate spline interpolation algorithm, with the interpolation weight being proportional to the curvature gradient amplitude; Step S216, the updated vertex coordinates are input as new morphological features; Step S217, recalculating the triangle normal vector based on the updated vertex coordinates, and comparing the change in the normal vector angle before and after the adjustment; Step S218: When the average angle change in the local area exceeds the angle change threshold, the point cloud of the area is optimized twice using the moving least squares method; Step S219: Finally, a set of key point coordinates that satisfy curvature continuity is output.

6. The method according to claim 5, characterized in that The step S22, using laser tracking technology to monitor the spatial relationship between the welding gun position and the designed welding point to determine the spatial deviation, includes: Step S221, collecting three-dimensional coordinate data of the welding gun position and the designed welding point by laser tracking technology to generate an initial coordinate set; Step S222, calculating the Euclidean distance between the welding gun coordinates and the designed welding point to obtain the spatial deviation value of each sampling point; Step S223: If the deviation value of the Euclidean distance is greater than the Euclidean distance threshold (mm), the welding gun coordinates are interpolated and adjusted using the bicubic interpolation method, and the adjusted coordinate set is output; Step S224, recalculating the deviation value based on the adjusted coordinates to generate an updated deviation distribution map; Step S225 , extracting the X / Y / Z axial deviation components from the deviation distribution map as feature vectors, clustering using the K-means algorithm of scikit-learn, setting the number of clusters to 3, and outputting the classification results of high / medium / low deviation areas; Step S226: For high deviation areas, increase the laser sampling frequency to 200 Hz; Step S227: The low-to-medium deviation region maintains the original 100 Hz sampling frequency and generates an optimized coordinate set; Step S228, based on the optimized coordinate set, calculate the root mean square value of the deviation between the final position of the welding gun and the designed weld point. If it is less than the preset value, the process is terminated, otherwise it is iterated again from the coordinate acquisition step.

7. The method according to claim 1, characterized in that The step S7, performing micron-level weld spot positioning operations according to the stable positioning state, integrating the dynamic path optimization results to generate the final welding instructions, and outputting consistent weld quality data, includes: Step S71, obtaining stable positioning state data through sensors, and obtaining positioning coordinates using Kalman filtering; Step S72: For the positioning coordinates, the dynamic path information is integrated using a weighted average method to obtain a path optimization result; Step S73, generating a welding instruction sequence using a rule-based method based on the path optimization result and outputting an instruction set; Step S74, obtaining welding execution parameters from the instruction set, performing simulation using finite element analysis, and determining consistency quality indicators; Step S75, using principal component analysis to extract weld quality characteristics based on the consistency quality index to obtain a quality assessment value; Step S76, comparing the quality evaluation value to the quality index threshold, if it exceeds the threshold, adjusting the path optimization parameters using the gradient descent method and regenerating the instruction set; Step S77: Use the weighted update method to update the data output content to obtain the final weld quality data.

8. The method according to claim 7, characterized in that The step S74, obtaining welding execution parameters from the instruction set, performing simulation using finite element analysis, and determining consistency quality indicators, includes: Step S741, parsing current, voltage, and speed parameters from the welding equipment control instruction set to generate a structured parameter table; Step S742: Building a three-dimensional welding geometry model based on the parameter table and importing the thermophysical parameters in the material property library; Step S743, using the finite element method to solve the Fourier heat conduction equation and calculate the temperature field distribution data of the weld area; Step S744: Input the temperature field data as initial conditions into the ABAQUS explicit dynamics module and output the Mises stress tensor of the weld area. Step S745, using the PCA tool of scikit-learn to reduce the dimension of the stress tensor, retaining the first three principal components whose cumulative contribution rates exceed the contribution rate threshold; Step S746: Divide the principal component data into a training set and a test set at a ratio of 7 to 3, build a three-layer fully connected network in TensorFlow, and use the ReLU activation function to train the quality prediction model; Step S747: When the predicted crack sensitivity coefficient exceeds the first sensitivity coefficient threshold, the genetic algorithm module of the DEAP library is called to perform iterative optimization with the parameter combination as the chromosome and the predicted value as the fitness; Step S748: Finally output the parameter combination that makes the crack sensitivity coefficient lower than the second sensitivity coefficient threshold.

9. The method according to claim 8, characterized in that The first sensitivity coefficient threshold is 35, and the second sensitivity coefficient threshold is 28.

10. A precise positioning and quality control system for welding an aluminum alloy subframe, wherein the method according to any one of claims 1 to 9 is used to precisely position and control the welding quality of the aluminum alloy subframe, the system comprising: The geometric model acquisition module is used to obtain the 3D geometric model of the aluminum alloy subframe and the coordinates of the designed weld points. It uses optical measurement technology to scan the workpiece surface, generate real-time surface point cloud data, and determine the initial welding gun trajectory and multi-dimensional positioning benchmark; The point cloud processing module is used to calculate the curvature changes and key positioning point coordinates of the complex structure of the workpiece based on the point cloud data. Laser tracking technology is used to monitor the spatial deviation between the welding gun and the designed weld point to obtain the micron-level positioning error distribution; The error monitoring module is used to adjust the motion parameters of the high-precision robotic arm according to the error distribution. It combines force feedback technology to detect small deformations on the workpiece surface during welding, determine the thermal deformation trend, and generate dynamic adjustment instructions. The motion control module is used to update the welding gun trajectory control parameters through dynamic adjustment instructions, integrate laser tracking data to optimize the welding path angle in real time, and determine the precise motion trajectory of each weld; The trajectory optimization module is used to obtain real-time thermal expansion data during welding, monitor the temperature field distribution in the weld area using optical measurement technology, and determine whether the thermal deformation adjustment exceeds the thermal deformation threshold; The thermal deformation monitoring module is used to calculate the compensation vector through force feedback technology if the thermal deformation adjustment exceeds the thermal deformation threshold, update the posture control data of the high-precision manipulator, and obtain a stable complex structure positioning state; The compensation correction module is used to perform micron-level weld point positioning operations based on the stable positioning state, integrate the dynamic path optimization results to generate the final welding instructions, and output consistent weld quality data; The welding execution module is used to obtain weld quality data, detect weld deviation and geometric consistency through laser tracking technology, determine whether the welding quality meets the design requirements, and generate a quality verification report; The quality assessment module is used to adjust the real-time adjustment capability parameters based on the deviation data in the quality verification report, optimize the welding gun trajectory control accuracy of subsequent welding tasks, and output the improved process parameter set.

Images