A non-standard welding opening reverse modeling method based on three-dimensional scanning and machine learning
By combining 3D scanning with machine learning, the problems of low reverse model matching and large welding path deviation in the modeling of non-standard weld joints of power plant steel structures were solved. This enabled high-precision reconstruction of non-standard weld joints and real-time closed-loop control of the welding process, thereby improving welding quality and connection reliability.
Patent Information
- Application Number
- CN202611139604.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-30
- Publication Date
- 2026-08-25
AI Technical Summary
Existing technologies for modeling non-standard weld joints in power plant steel structures suffer from problems such as low matching degree between the reverse model and the actual morphology, large deviation between the welding path and the edge, and lack of closed-loop control in the welding process, resulting in insufficient welding quality and connection reliability.
By combining 3D scanning with machine learning, a point cloud model of the weld area is generated through coordinate registration, the structure type label is identified, the feature deviation is calculated, the reverse 3D model is corrected, the feature points of the welding path are determined and the points are supplemented and corrected, and the deviation is compared and dynamically adjusted in combination with real-time welding data to achieve real-time closed-loop control of the welding process.
It improves the spatial accuracy and geometric fidelity of reverse modeling of non-standard weld joints, ensures the fit between the welding path and the weld edge, enhances the welding quality and the reliability of steel structure connections, and achieves stability and precision in the welding process.
Smart Images
Figure CN122636740A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of automatic welding modeling technology, and specifically to a reverse modeling method for non-standard weld joints based on 3D scanning and machine learning. Background Technology
[0002] During the manufacturing and installation of power plant steel structures, non-standard weld joints exhibit individual differences in their bevel geometry due to factors such as on-site processing conditions, assembly errors, and welding deformation. This makes it difficult to directly apply standard weld templates for unified welding path planning and posture control. Therefore, how to acquire the geometric information of non-standard weld joints through 3D scanning and combine it with machine learning to achieve intelligent identification of weld joint structure types and reverse 3D model reconstruction has become a technical challenge for improving the automation level and consistency of welding quality in power plant steel structures.
[0003] The existing non-standard weld joint modeling and welding application technologies have the following technical defects: 1. Existing technologies often directly call standard weld joint templates to generate 3D models, or simply stitch together multi-view scan point clouds to complete the reconstruction, without adjusting the standard templates based on the measured bevel geometry features. This results in low matching degree between the non-standard weld joint reverse model and the actual shape, and insufficient geometric accuracy, which makes it impossible to provide accurate geometric benchmarks for subsequent welding path planning, affecting the dimensional accuracy and forming quality of welding construction.
[0004] 2. Existing technologies often use fixed-step, equidistant sampling to generate path feature points when generating welding paths. This does not consider the degree of fit between the path points and the weld edge, resulting in a large deviation between the welding path and the actual weld edge, and incomplete path coverage in local weld areas. This leads to defects such as missed welds and weld misalignment during welding operations, reducing the quality of welding and the structural reliability of steel structure connections.
[0005] 3. Existing technologies do not compare and dynamically adjust real-time welding data during the welding process, and lack closed-loop control and regulation of the welding process. As a result, the deviation between the actual operating trajectory of the welding equipment and the planned path cannot be identified and compensated in a timely manner, causing the cumulative error of the welding posture angle to increase continuously, affecting the stability of the weld pool and the quality of weld formation. Summary of the Invention
[0006] This invention aims to overcome the deficiencies in the existing technology and provide a reverse modeling method for non-standard weld joints based on 3D scanning and machine learning. This method enables accurate registration of multi-view point cloud data, high-precision reconstruction of the reverse 3D model of non-standard weld joints, and real-time deviation closed-loop control of the welding process, thereby improving the accuracy of welding path planning and the stability of the welding execution process for non-standard weld joints in power plant steel structures.
[0007] The technical solution adopted by this invention to solve its technical problem is as follows: This invention provides a method for reverse modeling of non-standard weld joints based on 3D scanning and machine learning, including: S1. Based on the three-dimensional point cloud data of the non-standard weld joint area in the power station steel structure, coordinate registration is used to generate the weld joint area point cloud model. According to the geometric features of the weld joint bevel in the weld joint area point cloud model, the structural type label of the weld joint area is identified.
[0008] S2. Retrieve the standard weld template labeled with the structure type, calculate the feature deviation based on the geometric features of the weld bevel, and obtain the reverse three-dimensional model of the weld area based on the feature deviation.
[0009] S3. Based on the weld root trajectory line in the reverse 3D model, determine the welding path feature points, calculate the edge deviation at the welding path feature points, determine whether the welding path feature points cover the weld area, and when there are uncovered weld areas, perform point correction to generate the welding path coordinate sequence.
[0010] S4. Determine the welding posture angle based on the bevel edge contour line at the feature point in the welding path coordinate sequence, and generate equipment welding instructions in combination with the welding path coordinate sequence.
[0011] S5. Acquire real-time welding data during the equipment welding process, compare the data deviation with the equipment welding instructions, and determine the equipment welding data adjustment instructions.
[0012] Compared with the prior art, the present invention has the following beneficial effects: (1) Based on the three-dimensional point cloud data of the non-standard weld joint area in the power station steel structure, the present invention generates the weld joint area point cloud model through coordinate registration, realizes the high-precision fusion of multi-view point cloud data and the unification of spatial coordinates, ensures the complete restoration of the three-dimensional shape of the weld joint area, and improves the spatial accuracy and geometric fidelity of the reverse modeling of non-standard weld joints.
[0013] (2) This invention extracts the geometric features of the weld groove, identifies the structural type label of the weld area, retrieves the standard weld template of the structural type label, calculates the feature deviation based on the geometric features of the weld groove, and obtains the reverse three-dimensional model of the weld area based on the feature deviation, thereby realizing the accurate reverse reconstruction of the three-dimensional shape of the non-standard weld, increasing the matching degree between the standard template and the non-standard actual weld, and ensuring the dimensional accuracy of welding construction from the source.
[0014] (3) Based on the weld root trajectory line in the reverse three-dimensional model, the present invention determines the welding path feature points, calculates the edge deviation at the welding path feature points, determines whether the welding path feature points cover the weld area, and performs point correction when there is an uncovered weld area, generates a welding path coordinate sequence, effectively improves the fit between the welding path and the actual edge of the weld, avoids welding defects such as missed welding and weld deviation, ensures the complete coverage of the weld area by the welding trajectory, and improves the welding forming quality and the reliability of the steel structure connection.
[0015] (4) The present invention determines the welding posture angle based on the bevel edge contour line at the feature point in the welding path coordinate sequence, generates equipment welding instructions in combination with the welding path coordinate sequence, and compares the data deviation with the real-time welding data in the equipment welding process to determine the equipment welding data adjustment instructions, thereby realizing real-time deviation monitoring and dynamic compensation adjustment in the welding execution process, effectively correcting the trajectory deviation and posture deviation in the construction process, and improving the closed-loop accuracy of welding process control and the stability of weld formation quality. Attached Figure Description
[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a schematic diagram of the method steps of the present invention.
[0018] Figure 2 This is a schematic diagram of the process for obtaining the reverse three-dimensional model of the weld area in this invention.
[0019] Figure 3 This is a schematic diagram of the steps for determining the equipment welding data adjustment instruction in this invention. Detailed Implementation
[0020] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps set forth in these embodiments do not limit the scope of the invention. Furthermore, it should be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale.
[0021] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use. Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and apparatus should be considered part of the specification.
[0022] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.
[0023] Please see Figure 1 As shown, this invention provides a method for reverse modeling of non-standard weld joints based on 3D scanning and machine learning, including: S1. Based on the three-dimensional point cloud data of the non-standard weld joint area in the power station steel structure, coordinate registration is used to generate the weld joint area point cloud model. According to the geometric features of the weld joint bevel in the weld joint area point cloud model, the structural type label of the weld joint area is identified.
[0024] S2. Retrieve the standard weld template labeled with the structure type, calculate the feature deviation based on the geometric features of the weld bevel, and obtain the reverse three-dimensional model of the weld area based on the feature deviation.
[0025] S3. Based on the weld root trajectory line in the reverse 3D model, determine the welding path feature points, calculate the edge deviation at the welding path feature points, determine whether the welding path feature points cover the weld area, and when there are uncovered weld areas, perform point correction to generate the welding path coordinate sequence.
[0026] S4. Determine the welding posture angle based on the bevel edge contour line at the feature point in the welding path coordinate sequence, and generate equipment welding instructions in combination with the welding path coordinate sequence.
[0027] S5. Acquire real-time welding data during the equipment welding process, compare the data deviation with the equipment welding instructions, and determine the equipment welding data adjustment instructions.
[0028] Considering the complex spatial morphology of non-standard weld joint areas in the power plant's steel structure, single-view 3D scanning is insufficient to fully capture the overall information of the weld joint area. Furthermore, coordinate system differences exist between scan data from different perspectives. Without effective coordinate registration and data fusion, the point cloud model of the weld joint area will suffer from missing data or spatial misalignment, affecting the accuracy of subsequent extraction of bevel geometric features and the accuracy of structural type label identification. Therefore, preprocessing of multi-view point cloud data, geometric feature point matching, and coordinate transformation registration are necessary to identify the structural type label of the weld joint area.
[0029] Based on this, the specific implementation details of the structural type label for identifying weld joint areas in this invention include: S11. Extract multi-view 3D point cloud data of non-standard weld joint areas, and perform noise reduction processing on the 3D point cloud data of each view to obtain pre-processed point cloud data of each view.
[0030] It should be noted that the multi-view 3D point cloud data is acquired by multiple 3D laser scanners arranged around the weld area from different azimuth angles. The scanning angle of each 3D laser scanner covers the weld bevel area and the base material areas on both sides, ensuring that there are overlapping feature areas between adjacent viewpoints.
[0031] The noise reduction process employs an outlier removal method based on statistical distribution. It calculates the average distance between each data point in the point cloud data of each viewpoint and its nearest neighbor data point. Data points whose average distance exceeds a preset neighborhood distance threshold are identified as outlier noise points and removed, while retaining the effective geometric point cloud data of the weld joint area.
[0032] Preferably, in one specific embodiment of the present invention, the preset neighborhood distance threshold is determined based on the data point density distribution characteristics of the point cloud data from each viewpoint. For example, the sum of the mean of the average neighborhood distance distribution of the data points and three times the standard deviation is taken as the preset neighborhood distance threshold, i.e., μ+3σ, where μ is the mean of the average neighborhood distance distribution of the data points and σ is the standard deviation of the average neighborhood distance distribution of the data points. The implementer may also adjust the value of this threshold.
[0033] S12. Based on the geometric feature points in the point cloud data from each viewpoint, select the same geometric feature points, and register the point cloud data from different viewpoints in the same coordinate system according to the same geometric feature points, and stitch them together to generate a point cloud model of the weld joint area. The stitching method of the weld joint area point cloud model is to fuse the registered point cloud data from each viewpoint, perform mean fusion processing on the duplicate data points in the overlapping area, and retain the independent data points in the non-overlapping area to generate a point cloud model of the weld joint area.
[0034] It should be noted that the geometric feature points include, but are not limited to, weld edge corners, bevel intersection points, and abrupt changes in texture on the base material surface.
[0035] This invention is based on the three-dimensional point cloud data of non-standard weld joint areas in the steel structure of power plants. It uses coordinate registration to generate a point cloud model of the weld joint area, realizing high-precision fusion of multi-view point cloud data and unification of spatial coordinates, ensuring the complete restoration of the three-dimensional shape of the weld joint area, and improving the spatial accuracy and geometric fidelity of reverse modeling of non-standard weld joints.
[0036] S13. Extract the bevel contour point set, blunt edge contour point set, and root gap contour point set from the weld area point cloud model, and identify the bevel face angle, blunt edge thickness, and root gap from them, and use them as the geometric features of the weld bevel.
[0037] It should be noted that the bevel profile point set is the set of data points located at the junction of the bevel surface and the base material surface in the point cloud model of the weld area. The bevel surface angle is the angle between the bevel surface and the normal vector of the base material surface in the bevel profile point set, which represents the degree of inclination of the bevel surface.
[0038] The blunt edge contour point set is the set of data points located at the junction of the bottom of the bevel surface and the root gap in the point cloud model of the weld area. The blunt edge thickness is the distance between the two blunt edges in the blunt edge contour point set in the direction perpendicular to the weld center line, which represents the thickness of the un-beveled area at the bottom of the bevel.
[0039] The root gap contour point set is the set of data points located at the gap between the two base materials in the point cloud model of the weld area. The root gap is the minimum vertical distance between the two base material mating surfaces in the root gap contour point set, which characterizes the size of the assembly gap at the root of the weld.
[0040] S14. Match and compare the geometric features of the weld bevel with the trained weld structure type recognition model to determine the structure type label.
[0041] Preferably, in a specific embodiment of the present invention, the method for determining the structure type label is as follows: S141. A weld joint structure type recognition model is constructed using the bevel angle, blunt edge thickness, and root gap as the model input layer, and the weld joint structure type label as the model output layer. The weld joint structure type recognition model to be trained employs a classification neural network based on a multilayer perceptron. The model input layer contains three neurons, corresponding to the three input features: bevel angle, blunt edge thickness, and root gap. The output layer neuron data is consistent with the preset total number of weld joint structure types. Implementers can also choose other types of classification models, such as random forests or other deep learning models for point cloud data, according to actual needs.
[0042] S142. Collect historical weld sample data of the power station steel structure, extract the bevel angle, blunt edge thickness and root gap of each historical sample as input features, use the weld structure type label as the output layer, construct a training dataset, and train the weld structure type recognition model to be trained to obtain the trained weld structure type recognition model.
[0043] S143. Input the bevel face angle, blunt edge thickness and root gap from the weld bevel geometry features into the trained weld structure type recognition model, and select the weld structure type with the highest similarity of weld bevel geometry features as the structure type label of the weld area.
[0044] The similarity of weld bevel geometric features is calculated using the cosine similarity formula, which is used to characterize the degree of similarity between the measured weld bevel geometric features and the corresponding bevel geometric features of different weld structure types.
[0045] This invention extracts bevel contour point sets, blunt edge contour point sets, and root gap contour point sets from the weld area point cloud model to identify the geometric features of the weld bevel. It then performs item-by-item matching and comparison with the historical feature database corresponding to the preset weld structure type, thereby achieving intelligent identification and labeling of the weld structure type. This effectively improves the accuracy of non-standard weld structure type determination and provides a type basis for subsequent retrieval of standard weld templates and deviation correction.
[0046] Considering that the standard weld template is an idealized parametric model, which deviates from the actual geometric dimensions of non-standard welds on site, directly applying the standard template to generate a reverse model would result in a mismatch between the model and the actual weld shape, failing to provide an accurate geometric benchmark for subsequent welding path planning. Therefore, it is necessary to calculate the characteristic deviation and compensate for and correct the standard weld template to generate a reverse 3D model of the non-standard weld.
[0047] Based on this, such as Figure 2 As shown, the specific implementation of calculating the characteristic deviation in this invention includes: S21. Extract the standard weld template corresponding to the structure type label from the preset standard weld template library, and extract the standard bevel angle, standard blunt edge thickness and standard root gap of the standard weld template.
[0048] It should be noted that the preset standard weld template library is a pre-established three-dimensional parametric weld model database. Each structural type label corresponds to a standard weld template, which includes standard bevel angles, standard blunt edge thicknesses, standard root gaps, and other standard bevel geometric parameters. The standard bevel angle is the design value of the bevel angle for that structural type under ideal machining conditions, the standard blunt edge thickness is the design value of the blunt edge thickness for that structural type under ideal machining conditions, and the standard root gap is the design value of the root gap for that structural type under ideal assembly conditions.
[0049] S22. Calculate the difference between the bevel face angle, blunt edge thickness, and root gap in the weld bevel geometry and the standard bevel face angle, standard blunt edge thickness, and standard root gap, respectively, to obtain the bevel geometry deviation. A positive bevel geometry deviation indicates that the measured bevel geometry parameters are greater than the standard values, and the actual bevel depth or gap of the weld is too deep; a negative bevel geometry deviation indicates that the measured bevel geometry parameters are less than the standard values, and the actual bevel depth or gap of the weld is too shallow.
[0050] Based on this, the specific implementation of obtaining the reverse three-dimensional model of the weld area in this invention includes: S23. Determine the bevel angle compensation value, blunt edge thickness compensation value, and root gap compensation value based on the bevel geometric feature deviation.
[0051] Preferably, the bevel angle compensation value is the product of the bevel angle deviation and the preset angle compensation coefficient; the blunt edge thickness compensation value is the product of the blunt edge thickness deviation and the preset thickness compensation coefficient; and the root gap compensation value is the product of the root gap deviation and the preset gap compensation coefficient. In this invention, the preset angle compensation coefficient, preset thickness compensation coefficient, and preset gap compensation coefficient are all set to 1, achieving equal correction of geometric dimensions. Implementers can also adjust the values of each compensation coefficient according to actual conditions.
[0052] S24. The standard bevel angle, standard blunt edge thickness, and standard root gap of the standard weld template are superimposed with their corresponding compensation values to obtain the compensated bevel angle, blunt edge thickness, and root gap.
[0053] S25. Based on the compensated bevel angle, blunt edge thickness and root gap, the standard weld template is corrected to generate a reverse three-dimensional model of the weld area.
[0054] It should be noted that the modification of the standard weld template involves substituting the compensated bevel angle, blunt edge thickness, and root gap into the three-dimensional parametric modeling equation of the standard weld template, regenerating a reverse three-dimensional model of the weld area that matches the actual shape of the non-standard weld, and creating a three-dimensional digital twin model that characterizes the actual bevel geometry of the non-standard weld, providing a precise geometric reference for subsequent welding path planning.
[0055] This invention extracts the geometric features of the weld bevel, identifies the structural type label of the weld area, retrieves the standard weld template of the structural type label, calculates the feature deviation based on the geometric features of the weld bevel, and obtains a reverse three-dimensional model of the weld area based on the feature deviation. This achieves accurate reverse reconstruction of the three-dimensional morphology of non-standard welds, increases the matching degree between the standard template and the actual non-standard welds, and ensures the dimensional accuracy of welding construction from the source.
[0056] Considering that the reverse 3D model represents the 3D geometry of the weld area, the welding path needs to be distributed along the weld root trajectory line to ensure the integrity and uniformity of the weld metal filling. If fixed-step, equidistant sampling is used to generate path feature points without considering the degree of fit between the path points and the weld edge, the deviation between the welding path and the actual weld edge will be insufficient, resulting in blind spots in the path coverage of local weld areas. Edge deviation calculation and point correction are necessary to ensure complete coverage of the weld area by the welding path.
[0057] Based on this, the specific implementation details of generating the welding path coordinate sequence in this invention include: S31. Extract the weld root trajectory line from the reverse 3D model of the weld area, and perform equidistant sampling along the weld root trajectory line at a preset step size to determine the feature points of the welding path.
[0058] S32. Extract the bevel edge contour line at each welding path feature point, calculate the shortest distance between the bevel edge contour line and the corresponding weld edge contour of the reverse 3D model, and use it as the edge deviation.
[0059] It should be noted that the bevel edge contour line is the cross-sectional contour line perpendicular to the weld root trajectory line at the feature point of the welding path, representing the spatial shape of the interface between the weld and the base material at that feature point. The weld edge contour is the interface contour line between the weld bevel surface and the base material surface in the reverse 3D model.
[0060] The shortest distance is the minimum Euclidean distance between each data point on the bevel edge contour line and the nearest data point on the weld edge contour, representing the proximity of the weld toe position at the welding path feature point to the weld edge.
[0061] S33. If the edge deviation is less than or equal to the preset deviation threshold, the feature point of the welding path is determined to cover the weld area and is recorded as a valid feature point; otherwise, the feature point of the welding path is determined not to cover the weld area. The preset deviation threshold is determined based on the allowable weld toe deviation in the welding quality acceptance standard. For example, a preset ratio (such as 0.8) of the upper limit of weld toe deviation in the welding quality acceptance standard is taken as the preset deviation threshold. The smaller the edge deviation, the closer the welding path is to the weld edge.
[0062] S34. Insert supplementary feature points into the adjacent areas of the welding path feature points where there are uncovered weld joint areas, calculate the edge deviation at the supplementary feature points and perform iterative verification until the edge deviation at the supplementary feature points is less than or equal to the preset deviation threshold.
[0063] S35. Summarize all valid feature points and supplementary feature points to generate a welding path coordinate sequence.
[0064] Preferably, the adjacent region is the trajectory segment region between the previous and next effective feature points of the welding path feature points along the weld root trajectory line direction in the uncovered weld area. A new welding path feature point is inserted at the midpoint of the adjacent region, and the edge deviation at the supplementary feature point is calculated. If the edge deviation at the supplementary feature point is still greater than the preset deviation threshold, secondary supplementary feature points are inserted in the region between the supplementary feature point and its adjacent effective feature points for iterative verification until the edge deviation at all supplementary feature points meets the preset deviation threshold requirement.
[0065] This invention determines the welding path feature points based on the weld root trajectory line in the reverse 3D model, calculates the edge deviation at the welding path feature points, determines whether the welding path feature points cover the weld area, and performs point correction when there is an uncovered weld area, generating a welding path coordinate sequence. This effectively improves the fit between the welding path and the actual edge of the weld, avoids welding defects such as missed welds and weld deviations, ensures complete coverage of the weld area by the welding trajectory, and improves the welding forming quality and the reliability of steel structure connections.
[0066] Considering that the welding path coordinate sequence only represents the spatial movement trajectory of the welding torch, the welding posture angle directly affects the relative positional relationship between the welding wire and the weld bevel surface, thus affecting the molten pool morphology and weld formation quality. If the welding posture angle is not determined based on the local geometry of the bevel edge contour, it will lead to a mismatch between the welding torch posture and the actual weld bevel surface, causing molten pool deviation or undercut defects. Therefore, it is necessary to determine the welding torch travel direction angle and tilt angle through bevel edge contour analysis to generate equipment welding commands.
[0067] Based on this, the specific implementation details of generating equipment welding instructions in this invention include: S41. Extract adjacent feature points from the welding path coordinate sequence, determine the welding torch travel direction vector based on the spatial coordinates of the adjacent feature points, and calculate the angle between the welding torch travel direction vector and the reference direction vector to obtain the welding torch travel direction angle.
[0068] The reference direction vector is a preset reference direction unit vector, such as the projection direction of the weld centerline vector onto the horizontal reference plane.
[0069] S42. Extract the bevel edge contour line at each feature point in the welding path coordinate sequence, fit a local tangent plane based on the bevel edge contour line, determine the normal vector of the local tangent plane, and take the angle between the normal vector and the welding torch travel direction vector as the welding torch tilt angle.
[0070] It should be noted that the local tangent plane is centered on the feature point, takes the set of contour points within a defined range around it, and obtains the fitting plane by fitting with the least squares method. Its normal vector is perpendicular to the local tangent plane.
[0071] S43. Combine the welding torch travel direction angle and welding torch tilt angle into a welding posture angle, associate the welding path coordinate sequence with the welding posture angle at the corresponding feature point, and generate equipment welding instructions.
[0072] This invention determines the welding torch travel direction vector and welding torch travel direction angle by extracting adjacent feature points from the welding path coordinate sequence, determines the welding torch tilt angle by fitting a local tangent plane based on the bevel edge contour line, and generates equipment welding commands by associating the welding posture angle with the welding path coordinate sequence. This achieves precise planning of the welding posture and adaptive adjustment of the welding torch posture, effectively ensuring the matching degree between the welding wire and the weld bevel surface, and improving the weld formation quality and welding process stability.
[0073] Considering the influence of mechanical transmission errors and external disturbances on welding equipment during the execution of welding commands, deviations may exist between the actual welding trajectory and the planned path. Without real-time monitoring and dynamic compensation of these deviations in the welding data, the cumulative error of the welding posture angle will continue to increase, affecting the stability of the weld pool. Therefore, it is necessary to compare the deviation between real-time welding data and the equipment's welding commands to generate dynamic adjustment commands, thereby achieving closed-loop control of the welding process.
[0074] Based on this, such as Figure 3 As shown, the specific implementation details of determining the equipment welding data adjustment instruction in this invention include: S51. Extract real-time welding data during the welding process of the equipment, including real-time welding position coordinates, real-time welding torch travel direction angle, and real-time welding torch tilt angle.
[0075] It should be noted that the real-time welding position coordinates are acquired in real time by displacement sensors installed on the welding equipment, representing the actual position of the welding torch tip in three-dimensional space. The real-time welding torch travel direction angle is acquired in real time by angle encoders installed on the welding equipment, representing the deflection angle of the actual movement direction of the welding torch relative to the reference direction. The real-time welding torch tilt angle is acquired in real time by attitude sensors installed on the welding equipment, representing the actual tilt relationship between the actual axis of the welding torch and the normal vector of the bevel surface at the bevel edge contour line.
[0076] S52. Compare the real-time welding data with the welding posture angle at the corresponding feature point in the equipment welding command to obtain the welding data deviation amount and deviation change rate.
[0077] Preferably, in a specific embodiment of the present invention, the method for obtaining the welding data deviation and the rate of change of deviation is as follows: S521. Based on the position coordinates of all feature points in the welding path coordinate sequence, perform spatial vector difference calculation between the real-time welding position coordinates during the equipment welding process and the position coordinates of all feature points to obtain the position deviation vector, and select the feature point with the smallest position deviation vector as the feature point corresponding to the real-time welding position coordinates.
[0078] S522. Calculate the angle difference between the real-time welding torch travel direction angle and the real-time welding torch tilt angle and the welding torch travel direction angle and the welding torch tilt angle of the corresponding feature point to obtain the travel direction deviation and tilt angle deviation. Use the travel direction deviation and tilt angle deviation as the welding data deviation.
[0079] S523. Extract the welding data deviation between adjacent feature points, calculate the rate of change of the amplitude of the welding data deviation between adjacent feature points, and use it as the rate of change of welding data deviation.
[0080] The amplitude change rate is the ratio of the absolute value of the difference between the welding data deviation of the next feature point and the welding data deviation of the current feature point to the path length between the two feature points, representing the spatial variation of the welding data deviation along the welding path.
[0081] S53. Generate corresponding equipment welding data adjustment instructions based on the welding data deviation and the deviation change rate.
[0082] Preferably, in a specific embodiment of the present invention, the step of generating corresponding equipment welding data adjustment instructions based on the welding data deviation and the deviation change rate includes: S531. If the welding data deviation exceeds the corresponding preset deviation threshold, the welding data compensation amount is determined based on the welding data deviation, and an equipment welding data adjustment instruction is generated.
[0083] It should be noted that the preset deviation threshold is determined based on the allowable deviation of the welding trajectory in the welding quality acceptance standard, including the travel direction deviation threshold and the tilt angle deviation threshold.
[0084] The welding data compensation amount is the welding data deviation amount, and the compensation direction is opposite to the deviation direction, which is used to offset the deviation between the real-time welding data and the planned welding command.
[0085] S532. If the rate of change of welding data deviation exceeds the preset rate of change threshold, the direction of compensation adjustment is determined based on the positive or negative polarity of the rate of change of deviation, and the step size of compensation adjustment is determined in combination with the magnitude of the rate of change of deviation, and a compensation adjustment command is generated.
[0086] It should be noted that the preset rate of change threshold is determined based on the stability requirements of the welding process, and represents the maximum allowable rate of change in the spatial variation of welding data deviation.
[0087] The polarity of the deviation change rate is as follows: a deviation change rate greater than zero indicates an increasing trend in welding data deviation, while a deviation change rate less than zero indicates a decreasing trend in welding data deviation. When the deviation change rate is greater than zero, adjustments are made in the direction of decreasing deviation; when the deviation change rate is less than zero, the current compensation amount is maintained unchanged.
[0088] The compensation adjustment step size is the magnitude of the deviation change rate, representing the magnitude of the compensation adjustment.
[0089] S533. Summarize the equipment welding data adjustment instructions and compensation adjustment instructions to generate equipment welding data adjustment instructions.
[0090] This invention determines the welding posture angle based on the bevel edge contour line at the feature point in the welding path coordinate sequence, generates equipment welding instructions based on the welding path coordinate sequence, and compares the data deviation with the real-time welding data during the equipment welding process to determine the equipment welding data adjustment instructions. This enables real-time deviation monitoring and dynamic compensation adjustment during the welding execution process, effectively correcting trajectory deviation and posture deviation during construction, and improving the closed-loop accuracy of welding process control and the stability of weld formation quality.
[0091] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.
[0092] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0093] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.
[0094] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
[0095] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for reverse modeling non-standard weld joints based on 3D scanning and machine learning, characterized in that, include: Based on the three-dimensional point cloud data of non-standard weld joint areas in the steel structure of the power station, coordinate registration is used to generate a point cloud model of the weld joint area. According to the geometric features of the weld joint bevel in the point cloud model of the weld joint area, the structural type label of the weld joint area is identified. Retrieve the standard weld template labeled with the structure type, calculate the feature deviation based on the geometric features of the weld bevel, and obtain the reverse three-dimensional model of the weld area based on the feature deviation. Welding path feature points are determined based on the weld root trajectory line in the inverse 3D model. The edge deviation at the welding path feature points is calculated. It is determined whether the welding path feature points cover the weld area. When there is an uncovered weld area, the point is added for correction, and a welding path coordinate sequence is generated. The welding posture angle is determined based on the bevel edge contour line at the feature point in the welding path coordinate sequence, and the equipment welding command is generated by combining the welding path coordinate sequence. Acquire real-time welding data during the equipment welding process, compare the data deviations with the equipment welding instructions, and determine the equipment welding data adjustment instructions.
2. The method for reverse modeling non-standard weld joints based on 3D scanning and machine learning according to claim 1, characterized in that, The method for identifying the structural type label of the weld joint area is as follows: Extract multi-view 3D point cloud data of non-standard weld joint area, and perform noise reduction processing on the 3D point cloud data of each view to obtain pre-processed point cloud data of each view. Based on the geometric feature points in the point cloud data from various perspectives, the same geometric feature points are selected, and the point cloud data from different perspectives are registered in the same coordinate system according to the same geometric feature points, and then stitched together to generate a point cloud model of the weld area. Extract the bevel profile point set, blunt edge profile point set, and root gap profile point set from the point cloud model of the weld area, and identify the bevel face angle, blunt edge thickness, and root gap from them, and use them as the geometric features of the weld bevel. The geometric features of the weld bevel are matched and compared with the trained weld structure type recognition model to determine the structure type label.
3. The method for reverse modeling of non-standard weld joints based on 3D scanning and machine learning according to claim 2, characterized in that, The method for determining the structure type label is as follows: Using the bevel angle, blunt edge thickness, and root gap as the model input layer and the weld structure type label as the model output layer, a weld structure type recognition model to be trained is constructed. Historical weld sample data of power plant steel structure were collected. The bevel angle, blunt edge thickness and root gap of each historical sample were extracted as input features. The weld structure type label was used as the output layer to construct a training dataset. The weld structure type recognition model was trained to obtain the trained weld structure type recognition model. The bevel angle, blunt edge thickness, and root gap from the weld bevel geometry are input into the trained weld structure type recognition model, and the weld structure type with the highest similarity to the weld bevel geometry is selected as the structure type label of the weld region.
4. The method for reverse modeling of non-standard weld joints based on 3D scanning and machine learning according to claim 1, characterized in that, The method for calculating the characteristic deviation is as follows: Extract the standard weld template corresponding to the structure type label from the preset standard weld template library, and extract the standard bevel angle, standard blunt edge thickness and standard root gap of the standard weld template. The deviation of the weld bevel geometry is obtained by calculating the difference between the bevel face angle, blunt edge thickness, and root gap in the weld bevel geometry and the standard bevel face angle, standard blunt edge thickness, and standard root gap, respectively.
5. The method for reverse modeling non-standard weld joints based on 3D scanning and machine learning according to claim 4, characterized in that, The method for obtaining the reverse 3D model of the weld joint area is as follows: The compensation values for bevel face angle, blunt edge thickness, and root gap are determined based on the deviation of bevel geometric features. The standard bevel angle, standard blunt edge thickness, and standard root gap of the standard weld template are superimposed with their corresponding compensation values to obtain the compensated bevel angle, blunt edge thickness, and root gap. Based on the compensated bevel angle, blunt edge thickness, and root gap, the standard weld template is corrected to generate a reverse three-dimensional model of the weld area.
6. The method for reverse modeling of non-standard weld joints based on 3D scanning and machine learning according to claim 1, characterized in that, The method for generating the welding path coordinate sequence is as follows: Extract the weld root trajectory line from the reverse 3D model of the weld area, and perform equidistant sampling along the weld root trajectory line at a preset step size to determine the feature points of the welding path; Extract the bevel edge contour line at each welding path feature point, calculate the shortest distance between the bevel edge contour line and the corresponding weld edge contour of the reverse 3D model, and use it as the edge deviation. If the edge deviation is less than or equal to the preset deviation threshold, the feature point of the welding path is determined to cover the weld area and is recorded as a valid feature point; otherwise, the feature point of the welding path is determined not to cover the weld area. For welding path feature points with uncovered weld joint areas, insert supplementary feature points in adjacent areas, calculate the edge deviation at the supplementary feature points and perform iterative verification until the edge deviation at the supplementary feature points is less than or equal to the preset deviation threshold. Summarize all valid feature points and supplementary feature points to generate a welding path coordinate sequence.
7. The method for reverse modeling of non-standard weld joints based on 3D scanning and machine learning according to claim 6, characterized in that, The method for generating welding instructions for the equipment is as follows: Extract adjacent feature points from the welding path coordinate sequence, determine the welding torch travel direction vector based on the spatial coordinates of the adjacent feature points, and calculate the angle between the welding torch travel direction vector and the reference direction vector to obtain the welding torch travel direction angle. Extract the bevel edge contour line at each feature point in the welding path coordinate sequence, fit a local tangent plane based on the bevel edge contour line, determine the normal vector of the local tangent plane, and take the angle between the normal vector and the welding torch travel direction vector as the welding torch tilt angle. The welding torch travel direction angle and welding torch tilt angle are combined into welding posture angles. The welding path coordinate sequence is associated with the welding posture angles at the corresponding feature points to generate equipment welding instructions.
8. The method for reverse modeling of non-standard weld joints based on 3D scanning and machine learning according to claim 1, characterized in that, The method for determining the equipment welding data adjustment instruction is as follows: Extract real-time welding data during the welding process of the equipment, including real-time welding position coordinates, real-time welding torch travel direction angle, and real-time welding torch tilt angle; By comparing the real-time welding data with the welding posture angle at the corresponding feature point in the equipment welding command, the amount of welding data deviation and the rate of change of deviation are obtained. Based on the welding data deviation and the rate of change of the deviation, corresponding equipment welding data adjustment instructions are generated.
9. A method for reverse modeling non-standard weld joints based on 3D scanning and machine learning according to claim 8, characterized in that, The method for obtaining the welding data deviation and the rate of change of deviation is as follows: Based on the position coordinates of all feature points in the welding path coordinate sequence, the spatial vector difference calculation is performed between the real-time welding position coordinates of the equipment during the welding process and the position coordinates of all feature points to obtain the position deviation vector. The feature point with the smallest position deviation vector is selected as the feature point corresponding to the real-time welding position coordinates. The angle difference between the real-time welding torch travel direction angle and the real-time welding torch tilt angle and the welding torch travel direction angle and the welding torch tilt angle of the corresponding feature point is calculated to obtain the travel direction deviation and tilt angle deviation. The travel direction deviation and tilt angle deviation are used as the welding data deviation. Extract the welding data deviation between adjacent feature points, calculate the rate of change of the amplitude of the welding data deviation between adjacent feature points, and use it as the welding data deviation rate of change.
10. A method for reverse modeling non-standard weld joints based on 3D scanning and machine learning according to claim 9, characterized in that, The step of generating corresponding equipment welding data adjustment instructions based on welding data deviation and deviation change rate includes: If the welding data deviation exceeds the corresponding preset deviation threshold, the welding data compensation amount is determined based on the welding data deviation, and an equipment welding data adjustment instruction is generated. If the rate of change of welding data deviation exceeds the preset rate of change threshold, the direction of compensation adjustment is determined based on the positive or negative polarity of the rate of change of deviation, and the step size of compensation adjustment is determined in combination with the magnitude of the rate of change of deviation, and a compensation adjustment command is generated. The equipment welding data adjustment instructions and compensation adjustment instructions are summarized to generate the equipment welding data adjustment instructions.