A method and system for extracting parameters of a groove V-type bevel weld of a ship

Abstract

The application discloses a kind of ship slotting V-type groove weld parameter extraction method and system, the method includes: using the statistical filtering method based on percentile to eliminate abnormal points, and based on neighborhood z coordinate range realizes positive side weld point cloud separation;Then for the geometric characteristics of two kinds of structures of blunt edge V-type weld and butt V-type weld, rectangular surface recognition algorithm and boundary detection algorithm are designed respectively, accurate separation and boundary positioning of groove point cloud are realized;Finally, the adaptive extraction of groove angle parameters is completed by fusing RANSAC straight line fitting and endpoint method.The application can be suitable for the above two kinds of structures simultaneously, can provide accurate groove geometric parameters for ship welding robot, effectively improve the automation welding level of ship and marine equipment.

Description

Technical Field

[0001] This invention relates to laser welding technology, specifically to a method and system for extracting parameters of V-groove weld seams in ship grooving. Background Technology

[0002] In recent years, robot-based automated welding technology for ships has been rapidly adopted. To improve welding efficiency, weld parameters are typically extracted before welding to achieve optimal welding process matching. In the shipbuilding industry, butt joints of thick plates often employ slotted V-grooves to enhance weld strength. However, these grooves, due to their blunt edge structure and large depth-to-width ratio, suffer from severe light obstruction within the groove and difficulty in identifying feature points, making automatic parameter extraction particularly challenging. Currently, parameter acquisition still relies heavily on manual methods, which are inefficient and labor-intensive, becoming a bottleneck for achieving efficient automated welding in the shipbuilding industry. Therefore, researching automatic parameter extraction technology for ship V-grooves is a crucial step in realizing automated welding in ships.

[0003] Existing research, including Chinese invention patent application CN116152508A, discloses a method for extracting feature parameters of ship welds. This method identifies weld features based on the total number of contour lines and the minimum contour line distance. It can accurately identify four types of joints: flat joints, corner joints, T-joints, and lap joints, as well as various bevel types such as I-joints, V-joints, Y-joints, open joints, single-sided joints, and double-sided joints. Based on three types of joints, it establishes mathematical models to extract feature parameters such as bevel angle, bevel gap, and plate thickness, offering advantages in terms of wide recognition and high accuracy. However, this method primarily processes the contour lines of the welded part's digital model (STL file). Its input data is an idealized geometric model, fundamentally different from the weld point cloud data collected by laser vision sensors in actual industrial sites. Actual weld point cloud data often exhibits irregularity, non-uniformity, and high noise, making it difficult to directly extract regular contour line information. Furthermore, this method is mainly aimed at conventional bevel structures such as standard V-type and Y-type. For the slotted V-type bevel widely used in the welding of medium and thick plates in ships, especially the complex bevel morphology with blunt edge structure, there is still a lack of specialized boundary identification and parameter extraction methods.

[0004] Therefore, how to develop an automatic point cloud parameter extraction method for slotted V-groove welds that can directly process 3D point cloud data collected by laser vision sensors, while adapting to the boundary feature differences between two typical structures, blunt-edge V-groove and butt-groove V-groove, and achieve high-precision and automated extraction of groove geometric parameters, is a technical problem that urgently needs to be solved in the field of automated ship welding. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a method and system for extracting parameters of slotted V-groove welds in ships, which can solve the technical problems of low accuracy and poor adaptability of parameter detection for slotted V-groove welds in the prior art, and realize unified processing and high-precision parameter extraction for two types of structures: blunt edge V-groove welds and butt V-groove welds.

[0006] Technical solution: The present invention provides a method for extracting parameters of a slotted V-groove weld in shipbuilding, comprising:

[0007] Pre-processing of slotted V-shaped welds: An outlier point is removed from the weld point cloud acquired by laser scanning using a percentile-based statistical filtering method, retaining... Weld point clouds with coordinates within a preset range are considered valid weld point clouds; based on neighborhood... The coordinate range is used to calculate the local height change of each point cloud. Using the local height change of each point cloud, combined with the median threshold method, the effective weld point cloud is separated into the front weld point cloud and the side weld point cloud.

[0008] Extraction of general parameters for slotted V-shaped welds: The plate thickness parameters of the V-shaped welds are extracted from the point cloud of the side welds using a percentile-based statistical filtering method; the bevel angle parameters are extracted by combining the RANSAC linear fitting method and the endpoint method based on the density and distribution characteristics of the weld point cloud.

[0009] Extraction of other parameters for slotted V-welds: Different methods were used to extract parameters for blunt-edge V-welds and butt-weld V-welds; among them, for blunt-edge V-welds, a method based on... The coordinate range stability of the rectangular surface recognition algorithm detects and locates the boundary between the rectangular blunt edge and the V-groove by detecting the first-order differential mutation point. Based on the boundary position, the front weld point cloud is divided into rectangular surface point cloud and V-groove point cloud. For the V-groove point cloud, the groove length and root gap parameters are extracted. For butt V-groove welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is adopted. The effective analysis range of the V-shaped profile is determined by adaptive curvature threshold, and then the groove length parameter is extracted.

[0010] Furthermore, the calculation method for the effective weld point cloud is as follows:

[0011] ;

[0012] In the formula, and They are respectively The 5th and 95th percentiles of the coordinates; The effective weld point cloud obtained after removing outliers; For each point in the weld seam point cloud; For each point in the weld point cloud coordinate.

[0013] Furthermore, the neighborhood-based The coordinate range is used to calculate the local height change of each point cloud. Using this local height change, combined with the median threshold method, the effective weld point cloud is separated into front weld point clouds and side weld point clouds, including:

[0014] Construct a KD-tree spatial index structure and calculate the range of z-coordinates of each point within a given neighborhood radius r. :

[0015] ;

[0016] In the formula, For point The set of neighborhood points within a neighborhood radius r; This reflects the local height changes at each point; For neighborhood points The z-coordinate;

[0017] The median threshold method is used for classification to achieve robust separation of the point clouds of the front and side welds, i.e.:

[0018] ;

[0019] ;

[0020] In the formula, Point cloud for the side weld; for the median; This is the threshold coefficient; Point cloud for the front weld seam; For effective weld seam point cloud.

[0021] Furthermore, the plate thickness parameters of the V-shaped weld are extracted from the point cloud of the side weld using a percentile-based statistical filtering method, including:

[0022] Point cloud of side weld of The coordinates are filtered using a percentile-based statistical method:

[0023] ;

[0024] In the formula, For V-shaped weld seam plate thickness; Dot cloud for side weld seam All of the points Coordinate set; and They are respectively 1st and 99th percentiles.

[0025] Furthermore, the extraction of bevel angle parameters based on weld point cloud density and distribution characteristics, combining the RANSAC linear fitting method and the endpoint method, includes:

[0026] For the bevel region with dense weld point clouds, a linear model is established using the RANSAC linear fitting method:

[0027] ;

[0028] In the formula, This represents the inclination of the bevel face, used to calculate the bevel angle; It is the intercept of the straight line in the direction of bevel width, used to locate the spatial position of the bevel hypotenuse;

[0029] Single bevel angle The angle between the fitted line and the vertical direction is used to represent the total bevel angle. The calculation formula is:

[0030] ;

[0031] ;

[0032] In the formula, The angle between the bevel fitting line and the vertical direction for a grooved V-groove weld. The left bevel angle for a slotted V-groove weld; The right bevel angle for a slotted V-groove weld;

[0033] For bevel areas with sparse or unevenly distributed weld point clouds, the endpoint method is used to directly calculate the single bevel angle, utilizing the coordinate difference between the top and bottom points of the bevel.

[0034] ;

[0035] In the formula, This represents the change in width along the bevel direction; This represents the change along the length of the weld.

[0036] Furthermore, for the blunt-edge V-shaped weld, a method based on... A rectangular surface recognition algorithm with stable coordinate range detects and locates the boundary between the blunt edge of the rectangle and the V-groove by using first-order differential abrupt change points. Based on the boundary position, the point cloud of the front weld is divided into a rectangular surface point cloud and a V-groove point cloud, including:

[0037] along The direction is to perform equidistant layered sampling of the point cloud of the front weld seam, and Direction is divided into Each sampling layer is used to calculate the values ​​of each layer. Coordinate distribution range:

[0038] ;

[0039] In the formula, For the first The range of the y-coordinates of all point clouds within each layered sampling layer; Hierarchical index; For the first Layer start coordinate; For layer spacing; for Coordinates at the Within the layer range Value set; through calculation The first-order difference is used to detect abrupt change points and locate the boundary between the rectangular surface and the V-groove. Based on the boundary location, the point cloud of the front weld is divided into the point cloud of the rectangular surface and the point cloud of the V-groove.

[0040] Furthermore, the extraction of bevel length and root gap parameters from the V-shaped bevel point cloud includes:

[0041] Along the V-shaped area Direction is divided into Each point cloud is sampled in three equally spaced layers. The K-means clustering algorithm is applied to each layer to identify the left and right edge point groups and calculate the geometric center of each layer.

[0042] ;

[0043] In the formula, For the first The left edge point group within each sampling layer Coordinate mean; For the first The right edge point group within each sampling layer Coordinate mean; For the first Within each sampling layer, the left and right edge point groups are The center position in the direction;

[0044] right Coordinates of the center point Least squares linear regression was performed to fit the centerline equation; based on the positional relationship of each point relative to the centerline, the V-shaped bevel point cloud was separated into the left bevel point cloud. Clouds at the right slope Left slope point cloud And right slope point cloud The extraction of bevel parameters is based on the localization of inner edge feature points;

[0045] For the left slope, the strategy for selecting the top inner point is: in Within the preset percentage of points with the smallest coordinates, first eliminate those... Outliers whose direction exceeds a preset multiple of the standard deviation are identified. Then, the top few points closest to the center of the V-groove are extracted, and the average of these points is used as the top inner point. The bottom inner point selection strategy is as follows: first... Within the preset distance range between the coordinates and the bottom of the V-shape, select the nearest preset percentage point group, and then extract... The top few points with the largest coordinate values ​​are selected, and the average of these points is taken as the bottom inner point.

[0046] For the right bevel, the selection of the top inner point and bottom inner point corresponds to that of the left bevel, and both are selected as follows: The first few points with the smallest coordinate values;

[0047] bevel length Defined as the Euclidean distance between the inner points of the top and bottom, i.e.:

[0048]

[0049] In the formula, ( , ( ) represents the coordinates of the top inner point; , () represents the coordinates of the bottom inner point;

[0050] Root gap By detecting the left and right edge points of the bottom region of the V-shape The coordinate difference is obtained, that is:

[0051] ;

[0052] In the formula, The inner side of the bottom of the right bevel coordinate; The inner side of the bottom of the left bevel coordinate.

[0053] Furthermore, for butt V-shaped welds, a V-shaped boundary detection algorithm based on second-order difference curvature analysis is used. An adaptive curvature threshold is used to determine the effective analysis range of the V-shaped profile, thereby extracting the bevel length parameter, including:

[0054] The V-shaped profile of the butt V-weld is extracted using the following method:

[0055] Anomalies were applied to the point cloud of the front weld. Coordinate filtering, applying the filtered weld point cloud to... Sort the coordinates in ascending order. After sorting, in each... Extract the location from the coordinates. The point with the smallest coordinates forms the upper boundary envelope; a continuity constraint mechanism is introduced, and when extracting candidate boundary points layer by layer along the sampling direction, the coordinate offset between adjacent layers is calculated. :

[0056] ;

[0057] In the formula, For the first The coordinates of the layer boundary point in the y-direction; For the first The coordinates of the layer boundary point in the y-direction;

[0058] For continuity constraint jump threshold; when At that time, the judgment of the first Layer boundary points are identified as anomalous jump points and are removed. Boundary gaps resulting from the removal operation are filled using linear interpolation.

[0059] ;

[0060] In the formula, For the first Interpolation compensation coordinates for missing boundary points in the layer; and The first Layer and first Coordinates of the effective boundary points of the layer.

[0061] Furthermore, for butt V-shaped welds, a V-shaped boundary detection algorithm based on second-order difference curvature analysis is used. An adaptive curvature threshold is used to determine the effective analysis range of the V-shaped profile, thereby extracting the bevel length parameter, including:

[0062] After smoothing the envelope using a moving average filter, the second-order difference approximate curvature in the filter... The calculation formula is:

[0063]

[0064] In the formula, For layer spacing; For the first Layer envelope in The value of direction; For the first Layer envelope in The value of direction; For the first Layer envelope in The value of direction;

[0065] Design an adaptive curvature threshold :

[0066] ;

[0067] In the formula, The median of the non-zero curvature values; It is the maximum value of the absolute value of curvature; and These are the median coefficient and the maximum coefficient, respectively.

[0068] The boundary detection algorithm searches outwards from the peak curvature location. When three or more consecutive sampling points with curvature below the adaptive curvature threshold are detected, the algorithm continues to perform the boundary detection. When the boundary of the V-shaped region is reached, the left and right boundaries determined here limit the effective analysis range of the V-shaped profile.

[0069] Within the effective range of the V-shaped profile, Maximum coordinate point Using coordinates as boundaries, the contour points are separated into left and right parts; edge point extraction employs a layered extreme value method, along... The direction of the slope is used to stratify the point cloud; each layer is taken from the left slope. The maximum coordinate point is taken from each layer at the right-side bevel. The point with the minimum coordinates;

[0070] bevel length Defined as the edge point sequence and The Euclidean distance between the extreme points of the coordinate system is:

[0071]

[0072] In the formula, ( , ( ) represents the coordinates of the top extreme point; , ( ) represents the coordinates of the bottom extreme point.

[0073] Based on the same inventive concept, the present invention provides a parameter extraction system for ship slotted V-groove welds, comprising:

[0074] The slotted V-shaped weld preprocessing module is used to remove outliers from the weld point cloud acquired by laser scanning using a percentile-based statistical filtering method, retaining weld point clouds with z-coordinates within a preset range to form effective weld point clouds; it calculates the local height change of each point cloud based on the range of neighboring z-coordinates, and uses the local height change of each point cloud, combined with the median threshold method, to separate the effective weld point cloud into front weld point cloud and side weld point cloud;

[0075] A general parameter extraction module for slotted V-shaped welds is used to extract the plate thickness parameters of V-shaped welds from the side weld point cloud using a percentile-based statistical filtering method; and to extract the bevel angle parameters by combining the RANSAC linear fitting method and the endpoint method based on the weld point cloud density and distribution characteristics.

[0076] The module for extracting other parameters of slotted V-welds is used to extract parameters for blunt-edge V-welds and butt-weld V-welds using different methods; among them, for blunt-edge V-welds, a method based on... The coordinate range stability of the rectangular surface recognition algorithm detects and locates the boundary between the rectangular blunt edge and the V-groove by detecting the first-order differential mutation point. Based on the boundary position, the front weld point cloud is divided into rectangular surface point cloud and V-groove point cloud. For the V-groove point cloud, the groove length and root gap parameters are extracted. For butt V-groove welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is adopted. The effective analysis range of the V-shaped profile is determined by adaptive curvature threshold, and then the groove length parameter is extracted.

[0077] Beneficial effects: Compared with the prior art, the significant technical effects of the present invention are as follows:

[0078] (1) Propose a neighborhood-based approach A point cloud separation method for front and side welds with coordinate range was developed, which, combined with a robust median thresholding method, achieved... and Automatic point cloud separation. Optimal parameter configuration was determined through joint sensitivity analysis, achieving a separation consistency of 88.7%, effectively reducing the need for manual intervention and laying a reliable data foundation for subsequent feature extraction.

[0079] (2) For blunt-edge V-shaped welds, a method based on... A rectangular surface recognition algorithm with stable coordinate range. This algorithm is suitable for rectangular blunt-edge regions where the y-coordinate range is constant, while for V-shaped bevel regions... The coordinate range exhibits an increasing geometric difference. The first-order differential abrupt change point detection was used to accurately locate the boundary position, thereby controlling the identification error within 2.80 mm, which meets the positioning accuracy requirements of the ship welding site.

[0080] (3) For butt V-shaped welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is proposed. This algorithm takes into account the characteristics that the absolute value of curvature in the V-shaped region is large, while the curvature of the base material plane region is close to zero. It determines the effective analysis range of the V-shaped profile through adaptive threshold, realizes accurate positioning of smooth transition boundary, and improves the detection robustness under complex bevel morphology;

[0081] (4) General parameter extraction process, including robust plate thickness calculation based on percentile method and adaptive bevel angle calculation by integrating RANSAC and endpoint method, improves the algorithm's adaptability to different weld structures and weld point cloud densities, and enhances the practicality and promotion value of the engineering site.

[0082] (5) Experimental verification shows that the MAE of the bevel angle, bevel length, root gap, and plate thickness of the blunt-edge V-weld are 1.52°, 2.76 mm, 0.22 mm, and 0.75 mm, respectively; and the MAE of the bevel angle, bevel length, and plate thickness of the butt-weld V-weld are 0.41°, 0.96 mm, and 0.76 mm, respectively. Compared with the two existing representative weld parameter extraction algorithms, the detection accuracy of the blunt-edge V-weld by the algorithm of this invention is improved by 79.2% and 74.5%, respectively, and the detection accuracy of the butt-weld V-weld is improved by 78.5% and 44.1%, respectively. The above accuracy indicators can meet the technical specifications for weld bevel parameter detection in the shipbuilding welding industry, and can provide accurate and reliable data support for welding robot path planning and adaptive adjustment of process parameters. This is conducive to improving the operation efficiency and welding quality of automated shipbuilding welding, and has significant engineering application value. Attached Figure Description

[0083] Figure 1 This is a schematic diagram of the laser scanning system for small-group stand-up welds in ships and the geometric features of the welds disclosed in the embodiments of the present invention;

[0084] Figure 2 This is a flowchart illustrating a method for extracting parameters of a slotted V-groove weld in a ship, as disclosed in an embodiment of the present invention.

[0085] Figure 3 This is a schematic diagram of the point cloud geometric features of the blunt-edge V-shaped weld disclosed in an embodiment of the present invention;

[0086] Figure 4 This is a schematic diagram of the point cloud geometric features of the butt joint V-shaped weld disclosed in an embodiment of the present invention;

[0087] Figure 5 This is a schematic diagram of the laser point cloud acquisition system disclosed in an embodiment of the present invention;

[0088] Figure 6 This is a schematic diagram of the actual object and point cloud data of the blunt-edge V-shaped weld sample D1 disclosed in the embodiment of the present invention;

[0089] Figure 7 This is a schematic diagram of the physical sample K1 of the butt joint V-shaped weld disclosed in the embodiment of the present invention and its point cloud data;

[0090] Figure 8 These are separation consistency thermograms under different parameter combinations disclosed in the embodiments of the present invention;

[0091] Figure 9 This is a schematic diagram illustrating the influence of the curvature threshold coefficient on parameter detection accuracy as disclosed in an embodiment of the present invention;

[0092] Figure 10 This is a schematic diagram illustrating the measurement accuracy under different point cloud densities disclosed in the embodiments of the present invention;

[0093] Figure 11 This is a schematic diagram of the rectangular surface recognition process disclosed in an embodiment of the present invention;

[0094] Figure 12 This is a schematic diagram of the curvature boundary detection process disclosed in an embodiment of the present invention;

[0095] Figure 13 This is a diagram showing the absolute error distribution of various parameters of the blunt-edge V-shaped weld disclosed in an embodiment of the present invention;

[0096] Figure 14 This is a thermogram showing the relative errors of various parameters of the blunt-edge V-shaped weld disclosed in an embodiment of the present invention.

[0097] Figure 15 This is a diagram showing the absolute error distribution of various parameters of the butt joint V-shaped weld disclosed in an embodiment of the present invention;

[0098] Figure 16 This is a thermogram showing the relative errors of various parameters of the butt joint V-shaped weld disclosed in an embodiment of the present invention;

[0099] Figure 17 This is a comparison chart of the absolute errors of three algorithms for blunt-edge V-shaped welds disclosed in the embodiments of the present invention;

[0100] Figure 18 This is a comparison chart of the absolute errors of three algorithms for butt joint V-shaped welds disclosed in the embodiments of the present invention;

[0101] Figure 19 This is a comparison chart of the MAE distribution of the three algorithms disclosed in the embodiments of the present invention;

[0102] Figure 20 This is a schematic diagram of a ship slotted V-groove weld parameter extraction system disclosed in an embodiment of the present invention. Detailed Implementation

[0103] The technical solution of the present invention will now be described in detail with reference to specific embodiments and accompanying drawings.

[0104] Example 1

[0105] like Figure 1 As shown in the figure, the established The coordinate system serves as a spatial reference frame for laser scanning and weld seam point cloud processing. For example... Figure 1 Figure (a) in the diagram is a schematic diagram of a gantry-type dual-arm welding robot in operation; as shown in Figure (a). Figure 1 Figure (b) in the diagram is a schematic diagram of robot scanning based on a laser vision sensor; as shown in Figure (b). Figure 1 Figure (c) in the diagram is a schematic diagram of the blunt-edge V-shaped weld structure and its basic parameters; as shown in Figure (c). Figure 1 Figure (d) shows a schematic diagram of the structure and basic parameters of a butt weld. Based on the geometric differences in the V-groove design of ship grooves, this type of weld can be divided into two categories: blunt-edge V-welds and butt V-welds. For example... Figure 1 As shown in Figure (c), the typical characteristic of a blunt-edge V-weld is that a rectangular blunt-edge structure connects below the V-groove, and the cross-section is a combination of V-shape and rectangle; as shown in Figure (c), the blunt-edge V-weld is characterized by a rectangular blunt-edge structure below the V-groove. Figure 1 As shown in Figure (d), the V-groove of the butt V-weld is directly formed on the surface of the plate, with both ends forming a smooth transition with the base material. Although the two types of welds differ in local structural morphology, both feature a grooved V-groove as their core geometric characteristic.

[0106] like Figure 2 As shown, the present invention provides a method for extracting parameters of a slotted V-groove weld in shipbuilding, comprising the following steps:

[0107] S1. Pre-processing of slotted V-shaped welds: Anomalies are removed from the weld point cloud obtained by laser scanning using a percentile-based statistical filtering method, retaining... Weld point clouds with coordinates within a preset range are considered valid weld point clouds; based on neighborhood... The coordinate range is used to calculate the local height change of each point cloud. Using the local height change of each point cloud, combined with the median threshold method, the effective weld point cloud is separated into the front weld point cloud and the side weld point cloud.

[0108] The preprocessing includes two steps: removing abnormal points and separating the point cloud of the weld seam on the front and side sides. It is applicable to both blunt-edge V-type welds and butt V-type welds.

[0109] The specific implementation process of step S1 is as follows:

[0110] S1.1 Outlier Removal: An outlier is removed from the weld point cloud obtained by laser scanning using a percentile-based statistical filtering method. Weld point clouds with z-coordinates within a preset range are retained to form an effective weld point cloud.

[0111] During laser scanning, the original weld point cloud inevitably contains a small number of extreme outliers due to factors such as material surface reflection characteristics, local occlusion, and sensor noise. To reduce the interference of outliers on subsequent geometric parameter extraction, this embodiment of the invention introduces a percentile-based statistical filtering method. Based on the percentile interval of the z-coordinate of the slotted V-shaped weld point cloud, the effective weld point cloud is obtained after removing outliers. The criterion is as follows:

[0112] ;

[0113] In the formula, and They are respectively The 5th and 95th percentiles of the coordinates; The weld point cloud obtained after removing outliers; For each point in the weld seam point cloud; For each point in the weld point cloud coordinate.

[0114] This method retains the middle 90% of the data distribution, effectively suppressing extreme noise while maximizing the preservation of effective measurement information in the weld area.

[0115] S1.2, Point cloud separation of front and side welds: based on neighborhood The coordinate range is used to calculate the local height change of each point cloud. Using this local height change, combined with the median threshold method, the effective weld point cloud is separated into front weld point cloud and side weld point cloud, as detailed below:

[0116] like Figure 1 As shown in Figure (b), during the weld scanning process, the laser vision sensor illuminates the weld area vertically from above, and the acquired weld point cloud data includes the weld point cloud on the front side of the weld. Point cloud of side weld The former is used to extract parameters such as bevel angle, bevel length, and root gap (for blunt-edge V-welds only), while the latter is used for plate thickness extraction. (Combined) Figure 1 The coordinate system defined in Figure (c) shows a significant difference in the spatial geometric distribution between the weld face and side: the face region along... The z-axis changes gradually, while the lateral region changes drastically along the z-axis. Based on these geometric characteristics, this embodiment of the invention analyzes the local neighborhood of the point cloud. Automatic separation of weld point clouds from the front and side surfaces is performed based on the degree of coordinate change. First, a KD-tree spatial index structure is constructed, and then the position of each point within a given neighborhood radius r is calculated. Coordinate range :

[0117] ;

[0118] In the formula, For point The set of neighborhood points within a neighborhood radius r; This reflects the local height changes at each point; For neighborhood points of coordinate

[0119] Because the lateral region is approximately vertically distributed, the z-coordinate in its neighborhood varies considerably, i.e., the corresponding... The values ​​are relatively high; while the positive area is approximately horizontally distributed, and its neighborhood... The coordinate changes are gradual, that is... The value is relatively small. Based on the above characteristics, this embodiment of the invention uses the median threshold method for classification to achieve robust separation of the point clouds of the front and side welds, that is:

[0120] ;

[0121] ;

[0122] In the formula, Point cloud for the side weld; for the median; This is the threshold coefficient; Point cloud for the front weld seam; For effective weld seam point cloud.

[0123] S2. Extraction of general parameters for slotted V-shaped welds: The plate thickness parameters of the V-shaped welds are extracted from the point cloud of the side welds using a percentile-based statistical filtering method; the bevel angle parameters are extracted by combining the RANSAC linear fitting method and the endpoint method based on the density and distribution characteristics of the weld point cloud.

[0124] like Figure 1 As shown in Figure (c), the general parameters to be extracted for V-groove welds include: V-groove weld plate thickness. Single bevel angle Total bevel angle Although the two types of welds differ in structural morphology, the calculation principles for plate thickness and bevel angle are the same, and a universal parameter extraction algorithm can be designed.

[0125] The specific implementation process of step S2 is as follows:

[0126] S2.1 Plate Thickness Extraction: The plate thickness parameters of the V-shaped weld are extracted from the point cloud of the side weld using a percentile-based statistical filtering method. Details are as follows:

[0127] Because laser scanning data generally exhibits measurement fluctuations due to edge flypoints and surface roughness, directly using the extreme value method to calculate the plate thickness would lead to significant errors. Therefore, in this embodiment of the invention, the side weld point cloud is analyzed. of The coordinates are filtered using a percentile-based statistical method:

[0128] ;

[0129] In the formula, For V-shaped weld seam plate thickness; Dot cloud for side weld seam All of the points Coordinate set; and They are respectively 1st and 99th percentiles.

[0130] This method eliminates 1% of the extreme points at both ends of the distribution, effectively suppressing noise interference while preserving the true information of the side of the plate to the maximum extent, thus obtaining more accurate plate thickness calculation results.

[0131] S2.2, Groove Angle Extraction: Based on the weld point cloud density and distribution characteristics, the groove angle parameters are extracted by combining the RANSAC linear fitting method and the endpoint method. Details are as follows:

[0132] This invention employs two methods for calculating bevel angles: the RANSAC linear fitting method and the endpoint method. The appropriate method is selected based on the weld point cloud density and distribution characteristics. Specifically, the following operations are included:

[0133] For bevel regions with dense weld point clouds, the RANSAC linear fitting method is used. This method iteratively fits the edge point sequence to establish a linear model. The RANSAC algorithm uses the default configuration of the sklearn library. Adaptive residual thresholding (automatically calculated based on the median of the absolute residual deviation) is used for bevel point determination. The linear model is established as follows:

[0134] ;

[0135] In the formula, This represents the inclination of the bevel face, used to calculate the bevel angle; It is the intercept of the straight line in the direction of bevel width, used to locate the spatial position of the bevel hypotenuse;

[0136] In this embodiment of the invention, the single bevel angle The angle between the fitted line and the vertical direction is used to represent the total bevel angle. The calculation formula is:

[0137] ;

[0138] ;

[0139] In the formula, The angle between the bevel fitting line and the vertical direction for a grooved V-groove weld. For the left bevel angle of the slotted V-groove weld, The right bevel angle for a slotted V-groove weld.

[0140] The RANSAC algorithm effectively eliminates the influence of outliers on the fitting results through random sampling consistency testing.

[0141] For bevel areas with sparse or unevenly distributed weld point clouds, the endpoint method is used to directly calculate the single bevel angle, that is, the single bevel angle is calculated using the coordinate difference between the top and bottom points of the bevel:

[0142] ;

[0143] In the formula, This represents the change in width along the bevel direction; This represents the change along the length of the weld.

[0144] The endpoint rule avoids the problem of error accumulation caused by insufficient samples during the line fitting process.

[0145] S3. Extraction of other parameters for slotted V-welds: Different methods are used to extract parameters for blunt-edge V-welds and butt-weld V-welds; among them, for blunt-edge V-welds, a method based on... The coordinate range stability of the rectangular surface recognition algorithm detects and locates the boundary between the rectangular blunt edge and the V-groove by detecting the first-order differential mutation point. Based on the boundary position, the front weld point cloud is divided into rectangular surface point cloud and V-groove point cloud. For the V-groove point cloud, the groove length and root gap parameters are extracted. For butt V-groove welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is adopted. The effective analysis range of the V-shaped profile is determined by adaptive curvature threshold, and then the groove length parameter is extracted.

[0146] Please see Figure 3 In addition to the general parameters mentioned above, other parameters need to be extracted for blunt-edge V-welds: left bevel length. Length of right bevel Root gap Before extracting parameters, it is necessary to first complete the rectangular surface P of the weld. rect Identification and stripping.

[0147] The specific implementation process of step S3 is as follows:

[0148] S3.1 Rectangular Surface Recognition and Peeling: For blunt-edged V-shaped welds, a method based on... A rectangular surface recognition algorithm with stable coordinate range is used to detect and locate the boundary between the blunt edge of the rectangle and the V-groove through first-order differential abrupt change point detection. Based on the boundary position, the point cloud of the front weld is divided into a rectangular surface point cloud and a V-groove point cloud. Details are as follows:

[0149] In order to effectively and accurately identify and peel off rectangular surfaces The embodiments of the present invention propose a method based on Stability detection within the coordinate range. Along The direction is to perform equidistant layer sampling of the point cloud of the front weld, that is... Direction is divided into Each sampling layer is used to calculate the values ​​of each layer. The criterion for determining the range of coordinate distribution is:

[0150] ;

[0151] In the formula, For the first Within each hierarchical sampling layer, all point clouds The range of coordinates; Hierarchical index; For the first Layer start coordinate; For layer spacing; for Coordinates at the Within the layer range Value set. In the region with blunt edges of the rectangle, because the rectangular surface is... direction The range is stable, therefore The value remains relatively constant; however, in the V-groove region, the value increases due to the inclination of the groove. The range has increased, therefore With hierarchical index (Right now The coordinates are monotonically increasing. This is achieved through calculation. The first-order difference and abrupt change point detection can locate the boundary between the rectangular surface and the V-groove; based on the boundary position, the point cloud of the front weld is divided into the point cloud of the rectangular surface and the point cloud of the V-groove.

[0152] S3.2, Left-right separation based on multi-layer clustering: For the point cloud of a V-shaped bevel, extract the bevel length and root gap parameters. Details are as follows:

[0153] For the point cloud of the V-shaped bevel after rectangular surface peeling, a multi-layer centerline fitting method is used to achieve left-right separation. The V-shaped region is then separated along... Direction is divided into Equally spaced sampling layers (in this embodiment of the invention) The K-means clustering algorithm is applied to each layer of the point cloud (the number of cluster centers in this embodiment of the invention is...). Identify the point groups on the left and right edges and calculate the geometric center of each layer:

[0154] ;

[0155] In the formula, Hierarchical index; For the first The left edge point group within each sampling layer Coordinate mean; For the first The right edge point group within each sampling layer Coordinate mean; For the first Within each sampling layer, the left and right edge point groups are The center position in the direction, that is, the center point of the V-shaped bevel of this layer. coordinate.

[0156] Through the Coordinates of the center point Least squares linear regression is used to fit the centerline equation; based on the positional relationship of each point relative to the centerline, the V-shaped bevel point cloud can be separated into the left bevel point cloud. Clouds at the right slope ;

[0157] S3.3, Extraction of bevel geometric parameters: Point cloud of left bevel And right slope point cloud The bevel parameter extraction is based on the precise localization of inner edge feature points.

[0158] For the left slope, the strategy for selecting the top inner point is: in Within the 5% of points with the smallest coordinates, first eliminate them. Outliers with directions exceeding 3 standard deviations were then extracted. The three points with the largest coordinate values, i.e., the three points closest to the center of the V-groove, are selected, and the average of these three points is taken as the top inner point. The strategy for selecting the bottom inner point is as follows: first, in... Within a 5mm range of the coordinates and the bottom of the V-shape, select the 30% closest point cluster, and then extract... The three points with the largest coordinate values ​​are selected, and the average of these three points is taken as the bottom inner point.

[0159] For the right bevel, the selection of the top inner point and bottom inner point corresponds to that of the left bevel, and both are selected as follows: The three points with the smallest coordinate values; the above strategy effectively suppresses the impact of single-point noise through multi-point statistics;

[0160] bevel length Defined as the Euclidean distance between the inner points of the top and bottom, i.e.:

[0161]

[0162] In the formula, ( , ( ) represents the coordinates of the top inner point; , () represents the coordinates of the bottom inner point;

[0163] Root gap By detecting the left and right edge points of the bottom region of the V-shape The coordinate difference is obtained, that is:

[0164] ;

[0165] In the formula, The inner side of the bottom of the right bevel coordinate; The inner side of the bottom of the left bevel coordinate.

[0166] S3.3 For butt V-shaped welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is adopted. An adaptive curvature threshold is used to determine the effective analysis range of the V-shaped profile, and then the bevel length parameter is extracted. Details are as follows:

[0167] Please see Figure 4 In addition to the general parameters, the following other parameters also need to be extracted for butt V-welds: left bevel length Length of right bevel To extract these parameters, the V-shaped contour must first be extracted.

[0168] S3.3.1 V-shaped profile extraction under continuity constraints: The V-shaped profile of the butt weld is extracted using the following method:

[0169] First, anomalies were applied to the point cloud of the front weld seam. Coordinate filtering, applying the filtered weld point cloud to... Sort the coordinates in ascending order. After sorting, in each... Extract the location from the coordinates. The point with the smallest coordinates forms the upper boundary envelope. Since weld edge detection is susceptible to local noise interference, leading to abrupt changes or breaks in results, this embodiment of the invention introduces a continuity constraint mechanism to improve the robustness of boundary extraction. When extracting candidate boundary points layer by layer along the sampling direction, the coordinate offset between adjacent layers is calculated. :

[0170] ;

[0171] In the formula, For the first Layer boundary points at The coordinates of the direction; For the first Layer boundary points at The coordinates of the direction.

[0172] To effectively identify abnormal boundary points caused by noise interference, embodiments of the present invention define... The threshold for continuity constraint transition is used. Based on the weld geometry characteristics, in this embodiment of the invention... mm. This threshold is approximately 1 / 3 to 1 / 5 of the V-groove width, effectively distinguishing normal boundary variations from noise jumps. When the coordinate offset of adjacent layer boundary points... At that time, the judgment of the first Layer boundary points are identified as anomalous jump points and are removed. Boundary gaps resulting from the removal operation are filled using linear interpolation.

[0173]

[0174] In the formula, For the first Interpolation compensation coordinates for missing boundary points in the layer; and The first Layer and first Coordinates of the effective boundary points of the layer.

[0175] This constraint strategy significantly improves the smoothness and completeness of the boundary extraction results.

[0176] S3.3.2 Boundary Detection Based on Second-Order Difference Curvature: The V-shaped region of a butt weld has a smooth transition with the base metal plane, making detection algorithms based on geometric abrupt changes unsuitable. Therefore, this embodiment of the invention uses a moving average filter with a window length of 3 to smooth the envelope, suppressing high-frequency noise while preserving the local variation trend of boundary features. After smoothing the envelope using the moving average filter, the second-order difference approximate curvature in the filter... The calculation formula is:

[0177]

[0178] In the formula, This represents the change along the length of the weld. For the first Layer envelope in The value of direction; For the first Layer envelope in The value of direction; For the first Layer envelope in The value of direction; The absolute value is relatively large in the V-groove region due to the drastic change in geometric shape, while it approaches zero in the planar region of the base material. Based on these characteristics, an adaptive curvature threshold is designed. :

[0179] ;

[0180] In the formula, The median of the non-zero curvature values; It represents the maximum absolute value of the curvature. and These are the median coefficient and the maximum coefficient, respectively.

[0181] The boundary detection algorithm searches outwards from the peak curvature location. When three or more consecutive sampling points with curvature below the adaptive curvature threshold are detected, the algorithm continues to perform the boundary detection. When the boundary of the V-shaped region is reached, the left and right boundaries determined here can limit the effective analysis range of the V-shaped profile.

[0182] S3.3.3, Bevel Separation and Angle Extraction: Within the effective range of the V-shaped contour, using... The point with the maximum coordinates (i.e., the bottom of the V-shaped valley) Using coordinates as boundaries, the contour points are separated into left and right parts. Edge point extraction employs a layered extreme value method, i.e., along... The direction of the slope is used to stratify the cloud layers. Layers are taken from the left slope. The maximum coordinate point is taken from each layer at the right-side bevel. The point with the minimum coordinates.

[0183] bevel length Defined as the edge point sequence and The Euclidean distance between the extreme points of the coordinate system is:

[0184]

[0185] In the formula, ( , ( ) represents the coordinates of the top extreme point; , ( ) represents the coordinates of the bottom extreme point.

[0186] The technical solution of this invention first employs a percentile-based statistical filtering method to eliminate outliers, and then uses the neighborhood z-coordinate range to separate the point cloud of the front and side welds. Next, for the geometric features of two types of structures—blunt-edge V-welds and butt-weld V-welds—rectangular surface recognition and boundary detection algorithms are designed respectively to achieve accurate separation of the bevel point cloud and boundary localization. Finally, RANSAC linear fitting and endpoint methods are integrated to adaptively extract the bevel angle parameters. This invention is applicable to both types of structures, providing accurate bevel geometric parameters for marine welding robots and effectively improving the level of automated welding in ships and marine engineering equipment.

[0187] To verify the effectiveness of the method of the present invention, a laser point cloud acquisition system was built for experimental verification.

[0188] Experimental Design

[0189] Please see Figure 5 The laser point cloud acquisition system includes a laser vision sensor 2 for scanning the weldment 1. The laser vision sensor 2 is mounted on the collaborative robot 3, and then the upper computer 4 performs the acquisition and processing of the weldment point cloud.

[0190] To verify the applicability of the algorithm of this invention under different structures, five sets of blunt-edged V-shaped samples (D1~D5) and five sets of butt-joint V-shaped samples (K1~K5) were designed, all made of Q235B low-carbon steel. The samples cover different plate thicknesses, bevel angles, bevel lengths, and root gap dimensions. Specific geometric parameters are shown in Tables 1 and 2. Before the experiment, each geometric parameter was measured three times using a contact angle gauge and depth gauge, and the average value was taken as the true value benchmark.

[0191] Table 1 Design parameters of blunt-edge V-shaped weld specimen

[0192]

[0193] Table 2 Design parameters of butt V-weld samples

[0194]

[0195] Please see Figure 6-7 , Figure 6 The document showcases photographs and point cloud acquisition results of a typical sample, D1. Figure 6 Figure (a) in the diagram is a photograph of the actual object D1. Figure 6 Figure (b) is the point cloud map of D1. Figure 7 Showcasing physical photographs and point cloud acquisition results of a typical sample K1. Figure 7 Figure (a) in the figure is a physical image of K1. Figure 7 Figure (b) in the figure is the K1 point cloud map.

[0196] Algorithm sensitivity analysis

[0197] To assess the impact of key parameters on measurement results, this embodiment of the invention conducted sensitivity analysis on five factors: neighborhood radius, threshold coefficient, median coefficient, maximum coefficient, and weld point cloud density. Mean Absolute Error (MAE) and Mean Relative Error (MRE) were used as accuracy evaluation indicators.

[0198]

[0199]

[0200] In the formula, This is a predicted value; The actual value; This represents the sample size. A smaller MAE value indicates higher detection accuracy, while MRE reflects the average proportion of error to the true value.

[0201] (1) Sensitivity of point cloud separation parameters for front and side welds

[0202] Point cloud separation of front and side welds involves the neighborhood radius. With threshold coefficient Two key parameters. To avoid the coupling effects of single-parameter analysis in the algorithm, this embodiment of the invention employs a joint sensitivity analysis method. Figure 8 Heatmaps of separation consistency under different parameter combinations are presented. Experimental results show that separation consistency decreases with increasing neighborhood radius and threshold coefficient. Between 2.0 and 3.0 mm When the value is in the range of 1.0 to 1.3, the separation consistency remains above 87%. Considering both separation performance and computational stability, the algorithm of this invention selects... , .

[0203] (2) Adaptive curvature threshold sensitivity analysis

[0204] The adaptive threshold strategy for detecting the boundary of a butt weld V-shaped weld involves the median coefficient. With the maximum coefficient . Figure 9 The impact of threshold coefficient variation on parameter detection accuracy. Figure 9 Figure (a) in the figure is a schematic diagram of the sensitivity of the median coefficient. Figure 9 Figure (b) in the diagram illustrates the sensitivity of the median coefficient. Experimental results show that the median coefficient... When the bevel angle MAE is in the range of 0.4 to 0.6, it is less than The bevel length MAE is less than 1.05 mm, and the optimal value appears in Maximum value coefficient When the bevel angle MAE is in the range of 0.15 to 0.25, it is less than The bevel length MAE is less than 1.00 mm, and the optimal value appears in Test results show that the adaptive curvature threshold has good robustness to parameter perturbations.

[0205] (3) Sensitivity analysis of weld point cloud density

[0206] To evaluate the impact of weld point cloud density on measurement accuracy and computational efficiency, the mean square error (MAE) of the bevel angle and bevel length was statistically analyzed under different random sampling ratios, and processing time was recorded. Weld point cloud density is defined as: within a fixed region of interest (ROI), the density of the weld point cloud projected onto the target area. Number of points per unit area after plane The ROI is the bounding rectangle of the effective area of ​​the V-groove. Table 3 shows the variation of measurement error under different weld point cloud densities.

[0207] Table 3 Measurement errors under different weld point cloud densities

[0208]

[0209] Please see Figure 10 Experiments show that the weld point cloud density is lower than At this point, the errors in bevel angle and bevel length increase significantly; when the density exceeds this threshold, the rate of error reduction slows down significantly. This is in contrast to unsampled high-density point clouds (…). Compared to 50% downsampling, Under these conditions, the bevel angle and bevel length MAE only increase by [percentage missing]. The thickness is reduced to 0.88 mm, while the processing time is reduced from 2794 ms to 824 ms, a decrease of approximately 70.5%. Therefore, 50% downsampling can be used as the default configuration for the algorithm of this invention to balance the requirements of real-time performance and accuracy.

[0210] Key algorithm module verification

[0211] (1) Verification of rectangular face recognition effect

[0212] Rectangular surface identification is a crucial preliminary step in calculating the parameters of blunt-edge V-shaped welds. Figure 11 The process of recognizing the rectangular surface of sample D1. Figure 11 Figure (a) in the middle is Schematic diagram of change curves and mutation point detection. Figure 11 Figure (b) shows a schematic diagram of the separation result between the rectangular surface and the V-shaped region. For example... Figure 11 As shown in Figure (a), coordinate range The curve showing the change in x-coordinate reveals the rectangular blunt-edged region (X > -106.3 mm). The value remains relatively stable, but after entering the V-shaped bevel area... It exhibits a monotonically increasing trend. A mutation point was identified at X = -106.3 mm using first-order difference detection. Figure 11 Figure (b) shows the visualization result of the weld point cloud after separation, where the rectangular surface point cloud and the V-shaped groove point cloud are effectively separated.

[0213] Table 4 shows the rectangular surface recognition results for five groups of blunt-edged V-shaped samples. The identified abrupt change points for each sample deviate slightly from the actual top position of the blunt edge, with absolute errors ranging from 1.75 to 2.80 mm. Since the subsequent bevel region segmentation algorithm has a certain tolerance for initial deviations in the rectangular surface position, and the magnitude of the deviation is much smaller than the overall bevel size, this recognition accuracy meets practical requirements.

[0214] Table 4. Statistics on Rectangular Surface Recognition Position Errors

[0215]

[0216] (2) Verification of curvature boundary detection effect

[0217] Since the V-shaped area of ​​the butt V-weld transitions smoothly with the plane of the base material, the effective boundary needs to be determined through curvature analysis. Figure 12 The curvature boundary detection process for sample K1. Figure 12 Figure (a) in the figure is a schematic diagram of envelope curvature distribution and boundary detection. Figure 12 Figure (b) in the figure is a schematic diagram of the effective V-shaped region extraction results. Figure 12 As shown in Figure (a), the absolute value of curvature in the V-groove region is relatively large, while the curvature of the base material planar region approaches zero. Based on the adaptive curvature threshold, the left boundary position X = -151.1 mm and the right boundary position X = -117.9 mm were detected. Figure 12 In Figure (b), the effective analysis area is compared before and after boundary detection, where the gray area represents the removed part of the parent material plane.

[0218] Parameter extraction accuracy analysis

[0219] (1) Extraction results of blunt-edge V-shaped weld

[0220] Parameter extraction experiments were conducted on 5 groups of blunt-edge V-shaped weld samples, and the results are shown in Table 5.

[0221] Table 5. Summary of Parameter Extraction Accuracy for Blunt-edge V-shaped Welds

[0222]

[0223] As shown in the table, the MAE of the bevel angle is 1.52° and the MRE is 1.42%; the MAE of the bevel length is 2.76 mm and the MRE is 4.46%; the MAE of the root gap is 0.22 mm and the MRE is 2.53%; and the MAE of the plate thickness is 0.75 mm and the MRE is 9.80%.

[0224] Figure 13 The absolute error distribution of each parameter across the five sets of samples is shown. Figure 13Figure (a) shows the absolute error distribution of the bevel angle of the blunt-edge V-shaped weld sample across five groups of samples. Figure 13 Figure (b) shows the absolute error distribution of the bevel length of the blunt-edge V-shaped weld sample across five groups of samples. Figure 13 Figure (c) shows the absolute error distribution of root gap and plate thickness for the blunt-edge V-shaped weld samples across five groups of samples. Regarding the bevel angle, the angle error calculated using the endpoint method remained within 2° for most samples, with only samples D1 and D4 showing slightly larger errors. This was because local burrs on the bevel edges of these two samples affected the positioning accuracy of the top and bottom inner feature points. The bevel length error mainly ranged from 1 to 4 mm, originating from the aforementioned selection process of inner edge feature points. When the bevel edge point cloud density was uneven, the percentile-based feature point selection strategy introduced a certain positioning deviation. The root gap extraction accuracy was the most stable, with absolute errors all within 0.3 mm, thanks to the high point cloud density in the V-shaped bottom region and the good reliability of calculating the y-coordinate difference between the left and right edge points. The plate thickness parameter was calculated independently based on the percentile robust statistical method and the side weld point cloud. Although the absolute error was small, the relative error was relatively high, mainly because the side weld point cloud was located in the laser scanning edge region. The limited field of view of the sensor results in sparse weld point clouds, and the boundary effect has a certain impact on the measurement accuracy.

[0225] Figure 14 The relative error distribution of each parameter across different samples is presented in the form of a heatmap. The relative errors of the bevel angle and root gap are generally low across all samples, indicating that the endpoint method for angle calculation and the root gap detection method have high accuracy and stability. The relative error of the bevel length reaches 8.41% on sample D5, which is related to the relatively small design value of the bevel length for this sample. The relative errors of the plate thickness parameter are relatively high on samples D3 and D4, at 15.88% and 16.56% respectively, further confirming the limiting effect of the quality of the side weld point cloud acquisition on the accuracy of plate thickness measurement.

[0226] (2) Extraction results of butt V-shaped weld

[0227] Table 6 shows the parameter extraction results of 5 sets of butt joint V-shaped weld samples.

[0228] Table 6 Summary of Parameter Extraction Accuracy for Butt V-Welds

[0229]

[0230] As shown in the table, the MAE for the bevel angle is 0.41° and the MRE is 0.74%; the MAE for the bevel length is 0.96 mm and the MRE is 6.92%; and the MAE for the plate thickness is 0.76 mm and the MRE is 7.92%. Compared with the blunt-edge V-weld, the bevel angle extraction accuracy of the butt V-weld is significantly improved, mainly due to the RANSAC robust linear fitting method, which can more effectively utilize the bevel edge point cloud information, thereby reducing the impact of single-point noise on the angle calculation.

[0231] Figure 15 The absolute error distribution of each parameter across the five sets of samples is shown. Figure 15 Figure (a) shows the absolute error distribution of the bevel angle of the butt weld sample across five groups of samples. Figure 15 Figure (b) shows the absolute error distribution of the groove length of the butt joint V-weld sample across five groups of samples. Figure 15 Figure (c) shows the absolute error distribution of plate thickness for the V-shaped weld samples across five groups of samples. Regarding the bevel angle, the angle error calculated by the RANSAC linear fitting method remained within 1.0° for all samples, with the angle errors for samples K1 and K2 controlled within 0.5°, verifying the robust suppression capability of the RANSAC algorithm for edge outliers. The bevel length error was mainly concentrated within 1.2 mm; this parameter calculation depends on the V-shaped region boundary position determined by curvature boundary detection. When there are slight undulations on the base material surface, the boundary determination based on the adaptive curvature threshold may produce a shift of 1-2 sampling points. The absolute error distribution of the plate thickness parameter was relatively uniform, verifying the versatility of the percentile method under different plate thickness specifications.

[0232] Figure 16 The relative error distribution of each parameter across different samples is presented in heatmap form. The relative error of the bevel angle is within 1.7% across all samples and exhibits the most uniform distribution, indicating that the RANSAC fitting method possesses excellent accuracy and stability. The relative error of the bevel length reaches its maximum on sample K5, with 13.11% on the left and 9.28% on the right, which is directly related to the minimum designed bevel length value for sample K5. The relative error of the plate thickness parameter reaches 10.76% on sample K2, mainly due to slight surface oxidation on the side of this sample, which affects the laser reflection quality.

[0233] Comparative experiment

[0234] To comprehensively evaluate the effectiveness of the weld parameter extraction algorithm of this invention, two representative groove detection algorithms were selected for comparison. Algorithm 1 [1] K-means clustering was used to segment the bevel point cloud, combined with MSAC algorithm for line fitting; Algorithm 2 [2]FFT frequency domain filtering was used for preprocessing, and feature points were located using first-order derivative analysis. To ensure fairness in the comparison, the three algorithms used the same laser weld point cloud data and ground truth parameters.

[0235] Attached comparative literature

[0236] [1] Wang Xueni, Peng Fei, Wang Zhong, et al. A method for detecting the size of V-groove weldment based on line laser scanning [J]. Shipbuilding Engineering, 2024, 46(02): 99-105.

[0237] [2] Wang W, Shi Y, Li C, et al. Feature information extraction method for narrow gap U-type groove based on laser vision[J]. Journal ofManufacturing Processes, 2023, 104: 257-272.

[0238] Figure 17 The absolute errors of the three algorithms were compared on five sets of blunt-edge V-shaped weld samples. Figure 17 Figure (a) shows a schematic diagram comparing the absolute error of the left bevel angle of the blunt-edge V-shaped weld sample. Figure 17 Figure (b) shows a schematic diagram comparing the absolute error of the right bevel angle of the blunt-edge V-weld sample. Figure 17 Figure (c) shows a schematic diagram comparing the absolute error of the left bevel length of the blunt-edge V-shaped weld sample. Figure 17 Figure (d) shows a schematic diagram comparing the absolute error of the right bevel length of the blunt-edge V-shaped weld sample. Figure 17 Figure (e) shows a schematic diagram comparing the absolute error of the gap in a blunt-edge V-shaped weld sample. Figure 17 Figure (f) shows a comparison of the absolute error of the plate thickness of the blunt-edge V-shaped weld sample. As can be seen from the sub-figures, the absolute error of the algorithm of this invention is lower than that of the two comparative algorithms on the vast majority of samples. Regarding the bevel angle detection, the error levels of the three algorithms are all controlled within... Within the range, the algorithm of this invention performed the most stably on all samples; in terms of bevel length detection, Algorithm 1 showed a serious deviation on sample D3 because the K-means clustering segmentation boundary deviated from the actual position; in terms of root gap detection, the error of the algorithm of this invention was almost close to zero, while the errors of Algorithm 1 and Algorithm 2 were significantly higher and fluctuated more; in terms of plate thickness detection, the error curve of the algorithm of this invention was the lowest and most stable overall.

[0239] Figure 18 The absolute errors of the three algorithms were compared on five sets of butt joint V-weld samples. Figure 18Figure (a) shows a schematic diagram comparing the absolute error of the left bevel angle of the butt joint V-weld sample. Figure 18 Figure (b) shows a schematic diagram comparing the absolute error of the right bevel angle of the docking V-shaped sample. Figure 18 Figure (c) shows a schematic diagram comparing the absolute error of the left bevel length of the butt joint V-weld sample. Figure 18 Figure (d) shows a schematic diagram comparing the absolute error of the right bevel length of the butt joint V-weld sample. Figure 18 Figure (e) shows a schematic diagram comparing the absolute error of plate thickness in the butt joint V-weld samples. Regarding bevel angle detection, the algorithm of this invention has a particularly significant advantage, with the error curve remaining at an extremely low level across all samples. Regarding bevel length detection, Algorithm 1 shows significantly higher errors than the other two algorithms on multiple samples. Regarding plate thickness detection, the error curve of the algorithm of this invention is the most stable.

[0240] Figure 19 The MAE distributions of the three algorithms are presented in bar chart format. Figure 19 Figure (a) shows a comparison of MAE parameters for various parameters of the blunt-edge V-shaped weld. Figure 19 Figure (b) in the table shows a comparison of MAE parameters for various parameters of the butt joint V-shaped weld. Table 7 compares the comprehensive performance of the three algorithms on the two types of welds.

[0241] Table 7 Comparison of the overall performance of the three algorithms

[0242]

[0243] The comparison algorithms have the following limitations: Algorithm 1 uses K-means clustering to segment the bevel, which is difficult to adapt to the combination structure of V-groove and rectangular blunt edge in blunt edge V-groove welds. When the depth of the V-groove is shallow, the clustering segmentation accuracy drops significantly. Algorithm 2 was originally designed for narrow gap U-grooves, but its fixed proportion boundary expansion strategy is not adaptable to V-grooves and cannot identify root gap parameters. Both algorithms lack a continuity constraint mechanism for outliers, and weld point cloud noise directly affects the accuracy of subsequent parameter calculations.

[0244] The algorithm of this invention effectively overcomes the above limitations through targeted design: it uses y-coordinate range stability detection to achieve automatic identification of rectangular surfaces; it uses curvature boundary detection to accurately locate the boundary of V-shaped regions; it uses multi-layer centerline fitting to replace simple clustering to achieve robust left and right separation; and it introduces a continuity constraint mechanism to suppress noise jump interference.

[0245] Example 2

[0246] Please see Figure 20 , Figure 20This is a schematic diagram of a ship grooved V-groove weld parameter extraction system disclosed in an embodiment of the present invention. This system can achieve efficient and high-precision extraction of grooved V-groove weld parameters in a ship midship assembly welding production line, specifically including:

[0247] The slotted V-shaped weld preprocessing module is used to remove outliers from the weld point cloud acquired by laser scanning using a percentile-based statistical filtering method, retaining weld point clouds with z-coordinates within a preset range to form effective weld point clouds; it calculates the local height change of each point cloud based on the range of neighboring z-coordinates, and uses the local height change of each point cloud, combined with the median threshold method, to separate the effective weld point cloud into front weld point cloud and side weld point cloud;

[0248] A general parameter extraction module for slotted V-shaped welds is used to extract the plate thickness parameters of V-shaped welds from the side weld point cloud using a percentile-based statistical filtering method; and to extract the bevel angle parameters by combining the RANSAC linear fitting method and the endpoint method based on the weld point cloud density and distribution characteristics.

[0249] The module for extracting other parameters of slotted V-welds is used to extract parameters for blunt-edge V-welds and butt-weld V-welds using different methods. For blunt-edge V-welds, a rectangular surface recognition algorithm based on the stability of the y-coordinate range is used. The boundary between the rectangular blunt edge and the V-groove is located by detecting first-order differential mutation points. Based on the boundary position, the front weld point cloud is divided into a rectangular surface point cloud and a V-groove point cloud. For the V-groove point cloud, the groove length and root gap parameters are extracted. For butt-weld V-welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is used. The effective analysis range of the V-shaped profile is determined by an adaptive curvature threshold, and then the groove length parameter is extracted.

[0250] In one optional implementation, the method for extracting parameters of a slotted V-groove weld in a ship includes: a) removing outliers from the weld point cloud obtained by laser scanning using a percentile-based statistical filtering method; separating the weld point cloud into a front weld point cloud and a side weld point cloud; b) extracting the plate thickness parameters and groove angle parameters of the V-groove weld; c) for blunt-edge V-groove welds, dividing the front weld point cloud into a rectangular surface point cloud and a V-groove point cloud; extracting the groove length and root gap parameters for the V-groove point cloud; and for butt-welded V-groove welds, determining the effective analysis range of the V-groove profile, and then extracting the groove length parameter.

Claims

1. A method for extracting parameters of a slotted V-groove weld in shipbuilding, characterized in that, include: Pre-processing of slotted V-shaped welds: An outlier point is removed from the weld point cloud acquired by laser scanning using a percentile-based statistical filtering method, retaining... Weld point clouds with coordinates within a preset range are considered valid weld point clouds; based on neighborhood... The local height change of each point cloud is calculated using the coordinate range. Using the local height change of each point cloud, combined with the median threshold method, the effective weld point cloud is separated into the front weld point cloud and the side weld point cloud. Extraction of general parameters for slotted V-shaped welds: The plate thickness parameters of the V-shaped welds are extracted from the point cloud of the side welds using a percentile-based statistical filtering method; the bevel angle parameters are extracted by combining the RANSAC linear fitting method and the endpoint method based on the density and distribution characteristics of the weld point cloud. Extraction of other parameters for slotted V-welds: Different methods were used to extract parameters for blunt-edge V-welds and butt-weld V-welds; among them, for blunt-edge V-welds, a method based on... The coordinate range stability of the rectangular surface recognition algorithm detects and locates the boundary between the blunt edge of the rectangle and the V-groove by detecting the first-order differential mutation point. Based on the boundary position, the front weld point cloud is divided into rectangular surface point cloud and V-groove point cloud. For the V-groove point cloud, the groove length and root gap parameters are extracted. For butt V-groove welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is adopted. The effective analysis range of the V-shaped profile is determined by adaptive curvature threshold, and then the groove length parameter is extracted.

2. The method for extracting parameters of a ship slotted V-groove weld according to claim 1, characterized in that, The calculation method for the effective weld point cloud is as follows: ； In the formula, and They are respectively The 5th and 95th percentiles of the coordinates; The effective weld point cloud obtained after removing outliers; For each point in the weld seam point cloud; For each point in the weld point cloud coordinate.

3. The method for extracting parameters of a ship slotted V-groove weld according to claim 1, characterized in that, The neighborhood-based The coordinate range is used to calculate the local height change of each point cloud. Using this local height change, combined with the median threshold method, the effective weld point cloud is separated into front weld point clouds and side weld point clouds, including: Construct a KD-tree spatial index structure and calculate the value of each point within a given neighborhood radius r. Coordinate range : ； In the formula, For point The set of neighborhood points within a neighborhood radius r; This reflects the local height changes at each point; For neighborhood points of coordinate; The median threshold method is used for classification to achieve robust separation of the point clouds of the front and side welds, i.e.: ； ； In the formula, Point cloud for the side weld; for the median; This is the threshold coefficient; Point cloud for the front weld seam; For effective weld seam point cloud.

4. The method for extracting parameters of a slotted V-groove weld in shipbuilding according to claim 1, characterized in that, The point cloud of the side weld seam is used to extract the plate thickness parameters of the V-shaped weld seam using a percentile-based statistical filtering method, including: Point cloud of side weld of The coordinates are filtered using a percentile-based statistical method: ； In the formula, For V-shaped weld seam plate thickness; Dot cloud for side welds All of the points Coordinate set; and They are respectively 1st and 99th percentiles.

5. The method for extracting parameters of a slotted V-groove weld in shipbuilding according to claim 1, characterized in that, The extraction of bevel angle parameters based on weld point cloud density and distribution characteristics, combining RANSAC linear fitting and endpoint methods, includes: For the bevel region with dense weld point clouds, a linear model is established using the RANSAC linear fitting method: ； In the formula, This represents the inclination of the bevel face, used to calculate the bevel angle; It is the intercept of the straight line in the direction of bevel width, used to locate the spatial position of the bevel hypotenuse; Single bevel angle The angle between the fitted line and the vertical direction is used to represent the total bevel angle. The calculation formula is: ； ； In the formula, The angle between the bevel fitting line and the vertical direction for a grooved V-groove weld. The left bevel angle for a slotted V-groove weld; The right bevel angle for a slotted V-groove weld; For bevel areas with sparse or unevenly distributed weld point clouds, the endpoint method is used to directly calculate the single bevel angle, utilizing the coordinate difference between the top and bottom points of the bevel. ； In the formula, This represents the change in width along the bevel direction; This represents the change along the length of the weld.

6. The method for extracting parameters of a slotted V-groove weld in shipbuilding according to claim 1, characterized in that, For the blunt-edge V-shaped weld, the method based on A rectangular surface recognition algorithm with stable coordinate range detects and locates the boundary between the blunt edge of the rectangle and the V-groove by using first-order differential abrupt change points. Based on the boundary position, the point cloud of the front weld is divided into a rectangular surface point cloud and a V-groove point cloud, including: along The direction is to perform equidistant layered sampling of the point cloud of the front weld seam, and Direction is divided into Each sampling layer is used to calculate the values ​​of each layer. Coordinate distribution range: ； In the formula, For the first Within each hierarchical sampling layer, all point clouds The range of coordinates; Hierarchical index; For the first Layer start coordinate; For layer spacing; for Coordinates at the Within the layer range Value set; through calculation The first-order difference is used to detect abrupt change points and locate the boundary between the rectangular surface and the V-groove. Based on the boundary location, the point cloud of the front weld is divided into the point cloud of the rectangular surface and the point cloud of the V-groove.

7. The method for extracting parameters of a slotted V-groove weld in shipbuilding according to claim 1, characterized in that, The extraction of bevel length and root gap parameters from the point cloud of the V-shaped bevel includes: Along the V-shaped area Direction is divided into Each point cloud is sampled in three equally spaced layers. The K-means clustering algorithm is applied to each layer to identify the left and right edge point groups and calculate the geometric center of each layer. ； In the formula, For the first The left edge point group within each sampling layer Coordinate mean; For the first The right edge point group within each sampling layer Coordinate mean; For the first Within each sampling layer, the left and right edge point groups are The center position in the direction; right Coordinates of the center point Least squares linear regression was performed to fit the centerline equation; based on the positional relationship of each point relative to the centerline, the V-shaped bevel point cloud was separated into the left bevel point cloud. Clouds at the right slope Left slope point cloud And right slope point cloud The extraction of bevel parameters is based on the localization of inner edge feature points; For the left slope, the strategy for selecting the top inner point is: in Within the preset percentage of points with the smallest coordinates, first eliminate those... Outliers whose direction exceeds a preset multiple of the standard deviation are identified. Then, the top few points closest to the center of the V-groove are extracted, and the average of these points is used as the top inner point. The bottom inner point selection strategy is as follows: first... Within the preset distance range between the coordinates and the bottom of the V-shape, select the nearest preset percentage point group, and then extract... The top few points with the largest coordinate values ​​are selected, and the average of these points is taken as the bottom inner point. For the right bevel, the selection of the top inner point and bottom inner point corresponds to that of the left bevel, and both are selected as follows: The first few points with the smallest coordinate values; bevel length Defined as the Euclidean distance between the inner points of the top and bottom, i.e.: In the formula, ( , ( ) represents the coordinates of the top inner point; , () represents the coordinates of the bottom inner point; Root gap By detecting the left and right edge points of the bottom region of the V-shape The coordinate difference is obtained, that is: ； In the formula, The inner side of the bottom of the right bevel coordinate; The inner side of the bottom of the left bevel coordinate.

8. The method for extracting parameters of a slotted V-groove weld in shipbuilding according to claim 1, characterized in that, For butt V-shaped welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is used. An adaptive curvature threshold is used to determine the effective analysis range of the V-shaped profile, and then the bevel length parameter is extracted, including: The V-shaped profile of the butt V-weld is extracted using the following method: Anomalies were applied to the point cloud of the front weld. Coordinate filtering, applying the filtered weld point cloud to... Sort the coordinates in ascending order. After sorting, in each... Extract the location from the coordinates. The point with the smallest coordinates forms the upper boundary envelope; a continuity constraint mechanism is introduced, and when extracting candidate boundary points layer by layer along the sampling direction, the coordinate offset between adjacent layers is calculated. : ； In the formula, For the first Layer boundary points at The coordinates of the direction; For the first Layer boundary points at The coordinates of the direction; For continuity constraint jump threshold; when At that time, the judgment of the first Layer boundary points are identified as anomalous jump points and are removed. Boundary gaps resulting from the removal operation are filled using linear interpolation. ； In the formula, For the first Interpolation compensation coordinates for missing boundary points in the layer; and The first Layer and first Coordinates of the effective boundary points of the layer.

9. The method for extracting parameters of a slotted V-groove weld in shipbuilding according to claim 1, characterized in that, For butt V-shaped welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is used. An adaptive curvature threshold is used to determine the effective analysis range of the V-shaped profile, and then the bevel length parameter is extracted, including: After smoothing the envelope using a moving average filter, the second-order difference approximate curvature in the filter... The calculation formula is: In the formula, This represents the change along the length of the weld. For the first Layer envelope in The value of direction; For the first Layer envelope in The value of direction; For the first Layer envelope in The value of direction; Design an adaptive curvature threshold : ； In the formula, The median of the non-zero curvature values; It is the maximum value of the absolute value of curvature; and These are the median coefficient and the maximum coefficient, respectively. The boundary detection algorithm searches outwards from the peak curvature location. When three or more consecutive sampling points with curvature below the adaptive curvature threshold are detected, the algorithm continues to perform the boundary detection. When the boundary of the V-shaped region is reached, the left and right boundaries determined here limit the effective analysis range of the V-shaped profile. Within the effective range of the V-shaped profile, Maximum coordinate point Using coordinates as boundaries, the contour points are separated into left and right parts; edge point extraction employs a layered extreme value method, along... The direction of the slope is used to stratify the point cloud; each layer is taken from the left slope. The maximum coordinate point is taken from each layer at the right-side bevel. The point with the minimum coordinates; bevel length Defined as the edge point sequence and The Euclidean distance between the extreme points of the coordinate system is: ； In the formula, ( , ( ) represents the coordinates of the top extreme point; , ( ) represents the coordinates of the bottom extreme point.

10. A parameter extraction system for V-groove weld seams in shipbuilding, characterized in that, include: The slotted V-shaped weld preprocessing module is used to remove outliers from the weld point cloud acquired by laser scanning using a percentile-based statistical filtering method, retaining... Weld point clouds with coordinates within a preset range are considered valid weld point clouds; based on neighborhood... The local height change of each point cloud is calculated using the coordinate range. Using the local height change of each point cloud, combined with the median threshold method, the effective weld point cloud is separated into the front weld point cloud and the side weld point cloud. A general parameter extraction module for slotted V-shaped welds is used to extract the plate thickness parameters of V-shaped welds from the side weld point cloud using a percentile-based statistical filtering method; and to extract the bevel angle parameters by combining the RANSAC linear fitting method and the endpoint method based on the weld point cloud density and distribution characteristics. The module for extracting other parameters of slotted V-welds is used to extract parameters for blunt-edge V-welds and butt-weld V-welds using different methods; among them, for blunt-edge V-welds, a method based on... The coordinate range stability of the rectangular surface recognition algorithm detects and locates the boundary between the blunt edge of the rectangle and the V-groove by detecting the first-order differential mutation point. Based on the boundary position, the front weld point cloud is divided into rectangular surface point cloud and V-groove point cloud. For the V-groove point cloud, the groove length and root gap parameters are extracted. For butt V-groove welds, a V-shaped boundary detection algorithm based on second-order differential curvature analysis is adopted. The effective analysis range of the V-shaped profile is determined by adaptive curvature threshold, and then the groove length parameter is extracted.

Images