Welding seam size measurement and defect detection method based on point cloud features

Abstract

The invention discloses a weld size measurement and defect detection method based on base point cloud features. The method comprises the following steps of weld point cloud data collection, point cloud processing, equidistant slice projection, weld section model fitting, feature point extraction, size measurement and defect calculation. The method provided by the invention can measure the size and defect of the welding seam in real time, and is high in accuracy and strong in applicability. According to the method provided by the invention, welding seam size measurement and defect detection are simultaneously realized in a manner of extracting the welding seam point cloud features, the welding seam forming condition and the quality state are directly reflected, and the problems of low efficiency, large data fluctuation, non-objective judgment and the like of traditional manual welding seam forming quality detection are solved.

Description

Technical Field

[0001] The present invention relates to the field of weld detection, and in particular to a weld size measurement and defect detection method based on point cloud features. Background Art

[0002] As one of the most important processing technologies in modern manufacturing, welding plays an important role in mechanical engineering, construction, shipbuilding and automobile manufacturing. In the whole process, the detection of weld formation quality is an indispensable link. The weld formation quality detection is mainly divided into two parts: one is the weld appearance detection, including the detection of weld width and excess height. The second is the detection of weld surface defects such as undercut and misalignment. The weld formation quality detection is the premise of ensuring welding quality. For example, when undercut appears on the weld surface, the inside may be accompanied by the situation of incomplete penetration. The main way of traditional weld formation quality detection is manual detection by humans, which is very inefficient and the data is prone to fluctuations. In addition, due to the different skills and experience of the inspectors, manual detection cannot objectively judge the welding quality. my country's weld quality detection technology is developing in the direction of intelligence, which requires the ability to accurately characterize and evaluate the weld formation results. It can be seen that relying solely on manual welding quality detection can no longer meet the requirements of modern welding technology, and it is of great practical significance to realize automatic detection of weld quality.

[0003] At present, with the in-depth research and wide application of weld formation quality detection technology, a variety of different methods have been developed to detect welding formation quality, including ultrasonic detection technology, X-ray detection technology and visual detection technology. Although ultrasonic detection is widely used due to its flexibility and low cost, it lacks visual records and requires highly skilled operators to identify defects. Although X-ray detection is very effective, it is costly and can cause harm to the human body. Visual detection has the characteristics of non-contact, high precision and good adaptability, and can obtain richer welding environment information. It not only solves the shortcomings of manual detection, but also avoids the shortcomings and complex processes of other detection methods, such as high cost and the need for highly skilled personnel to operate. In recent years, visual detection has been applied to weld formation quality detection, weld appearance detection and weld surface defect identification. Weld appearance detection is generally based on active vision, usually using line structured light scanning method to extract weld information, and then fit the weld contour and extract feature points to achieve the measurement of weld excess height and weld width. Weld surface defect detection is generally based on passive vision. This method mainly simulates artificial angles to detect weld surface defects. It uses image acquisition, feature extraction, and machine learning technologies to detect information such as defect location and type. However, it may have disadvantages such as difficulty in handling deformation and non-standard defects, requiring a large amount of labeled data, and consuming a lot of computing resources.

[0004] Although there have been many studies on weld formation quality inspection, it is rare to measure the weld formation size and detect weld appearance defects at the same time. Instead, a combination of active vision and passive vision is used to achieve weld appearance inspection. The accuracy and real-time performance of weld inspection are still under study. In response to the above problems, the present invention adopts a three-dimensional visual detection method and proposes a weld size measurement and defect detection method based on point cloud features to achieve automatic and efficient detection of weld morphology. Summary of the invention

[0005] The present invention provides a weld size measurement and defect detection method based on point cloud features to solve the problems of low efficiency, large data fluctuation and subjective judgment in traditional manual weld formation quality detection, and realize simultaneous weld size measurement and defect detection.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A method for weld size measurement and defect detection based on point cloud features, characterized in that it comprises the following steps:

[0008] S1. Collect weld point cloud data: Use a binocular structured light camera to collect three-dimensional point cloud data of the weld;

[0009] S2, point cloud processing: processing the three-dimensional point cloud data in step S1, downsampling by voxel filtering method, removing abnormal points by statistical filtering method, and correcting the main direction of the point cloud by principal component analysis method;

[0010] S3, isometric slicing projection: isometrically slicing the weld point cloud processed in step S2 in the main direction of the weld, extracting the three-dimensional point cloud information on each slice, and projecting it onto a two-dimensional plane to obtain two-dimensional scattered point coordinates;

[0011] S4, weld cross-section model fitting: fitting the two-dimensional scattered points in step S3 to establish a weld cross-section model;

[0012] S5, feature point extraction: for the weld cross-section model in step S4, extract the weld left feature point P according to the detection criteria l , weld right feature point P r , the two lowest points P on the left and right of the weld contour u1 , P u2 and the highest point P of the weld profile t ;

[0013] S6. Dimension measurement and defect calculation: perform difference calculation on the characteristic point coordinates obtained in step S5 to obtain the weld height, weld width, weld undercut and weld misalignment value of each section to reflect the weld formation size and defects.

[0014] Furthermore, the binocular structured light camera in step S1 is installed on the conveyor belt rack, and the binocular structured light camera collects 3D point clouds in a single time within 2 seconds, and the measurement accuracy is not less than 0.2 mm.

[0015] Furthermore, in the point cloud processing process described in step S2: when downsampling is performed using the voxel filtering method, the point closest to the center of gravity in the voxel grid of the original point cloud data is used to replace the center of gravity; when the statistical filtering method is used to remove abnormal points, the abnormal points are filtered out based on the number of neighboring points and the standard deviation threshold calculation; when the principal component analysis method is used to correct the main direction of the point cloud, the main normal vector direction of the point cloud is calculated to determine the center line of the weld, and the horizontal plane of the point cloud is corrected to the length direction of the weld, so that the cross-section plane is perpendicular to the weld surface.

[0016] Furthermore, in the isometric slicing projection process described in step S3: the isometric slicing adopts the minimum directed bounding box method to plan the direction and distance of the slice; the projection mapping selects the centroid point of the point cloud slice to form a projection surface, and establishes a local coordinate system on the projection surface, and then projects each point in the point cloud onto the projection surface to obtain a two-dimensional data point.

[0017] Furthermore, the weld cross-section model fitting in step S4 uses a cubic B-spline smooth curve to fit the weld portion, and uses a random sampling consistency algorithm to fit the base material portion.

[0018] Furthermore, the inflection points on both sides of the weld of the fitting curve in the feature extraction in step S5 are P u1 , P u2 , the highest point of the fitting curve is P t The intersection of the fitting straight line and the fitting curve on both sides of the weld is P l and P r .

[0019] Further, the dimension measurement and defect calculation in step S6 include: calculating the absolute value of the difference between the horizontal coordinate of the left characteristic point of the weld and the horizontal coordinate of the right characteristic point of the weld to obtain the weld width; calculating the difference between the vertical coordinate of the highest point of the weld curve profile and the vertical coordinate of the lowest point of the weld profile to obtain the weld height; calculating the absolute value of the difference between the vertical coordinate of the left lowest point of the weld profile curve and the vertical coordinate of the left characteristic point of the weld to obtain the left side bite value of the weld; calculating the absolute value of the difference between the vertical coordinate of the right lowest point of the weld profile curve and the vertical coordinate of the right characteristic point of the weld to obtain the right side bite value of the weld; calculating the absolute value of the difference between the vertical coordinate of the left characteristic point of the weld and the vertical coordinate of the right characteristic point of the weld to obtain the weld misalignment value.

[0020] The present invention has the following beneficial effects:

[0021] The present invention provides a method for weld size measurement and defect detection based on point cloud features, comprising the following steps: collecting weld point cloud data, point cloud processing, isometric slice projection, weld cross-section model fitting, feature point extraction, size measurement and defect calculation. The method proposed by the present invention can measure weld size and defects in real time, and has the following beneficial effects compared with the prior art:

[0022] (1) The present invention provides a weld size measurement and defect detection method based on point cloud features, which solves the problems of low efficiency and poor stability of traditional manual detection methods. By extracting weld point cloud features, size measurement and defect detection are performed, deformation and non-standard defects can be processed, and a large amount of labeled data is not required and computing resources are not consumed.

[0023] (2) The present invention provides a weld size measurement and defect detection method based on point cloud features, which can simultaneously realize weld size measurement and defect detection, and directly reflect the weld formation and quality status;

[0024] (3) The present invention provides a weld size measurement and defect detection method based on point cloud features, which processes the collected weld point cloud with complete geometric information, including downsampling, filtering and main direction correction, thereby enhancing the versatility and extensibility of the method, reducing the difficulty of the fitting algorithm, and improving the overall computing efficiency of the system;

[0025] (4) The present invention provides a weld size measurement and defect detection method based on point cloud features. Based on the weld cross-sectional features, the weld area and the base material area are fitted respectively, and feature points are extracted for calculation, thereby improving the overall extraction accuracy and completeness.

[0026] (5) The present invention provides a weld size measurement and defect detection method based on point cloud features. It can perform detection according to the change curve of the relevant indicators of each weld section, quickly identify possible problems in the weld, and take timely measures to correct or adjust them, thereby ensuring that the welding quality meets the requirements and improving work efficiency.

[0027] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0029] Figure 1 It is a flowchart of the steps of a weld size measurement and defect detection method based on point cloud features provided by an embodiment of the present invention;

[0030] Figure 2 is a front diagram of a point cloud processing method provided by an embodiment of the present invention;

[0031] Figure 3 This is a diagram showing a point cloud processing method according to an embodiment of the present invention;

[0032] Figure 4 It is a weld cross-section fitting and feature point extraction diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0034] Figure 1 It is a step flow chart of a weld size measurement and defect detection method based on point cloud features provided by an embodiment of the present invention, including collecting weld point cloud data, point cloud processing, isometric slice projection, weld cross-section model fitting, feature point extraction, size measurement and defect calculation. Figure 2 is a front diagram of a point cloud processing method provided by an embodiment of the present invention; Figure 3 This is a diagram showing a point cloud processing method according to an embodiment of the present invention; Figure 4 It is a weld cross-section fitting and feature point extraction diagram provided by an embodiment of the present invention.

[0035] like Figure 1 As shown, a weld size measurement and defect detection method based on point cloud features in this embodiment includes the following steps:

[0036] S1. Collecting weld point cloud data: using a binocular structured light camera to collect weld point cloud data;

[0037] Among them, in this embodiment, a binocular structured light camera is used to collect weld point cloud images. The binocular structured light camera is installed on the conveyor belt rack with an installation height of 120 cm. The binocular structured light camera takes no more than 2 seconds to collect 3D point clouds at a time, and the measurement accuracy is not less than 0.2 mm.

[0038] S2, point cloud processing: processing the point cloud data in step S1, downsampling by using voxel filtering method, removing abnormal points by using statistical filtering method, and correcting the main direction of the point cloud by using principal component analysis method;

[0039] Among them, in the point cloud processing process: when the voxel filtering method is used for downsampling, the point closest to the center of gravity in the voxel grid of the original point cloud data is used to replace the center of gravity; when the statistical filtering method is used to remove abnormal points, the abnormal points are filtered out based on the number of neighboring points and the standard deviation threshold calculation; when the principal component analysis method is used to correct the main direction of the point cloud, the main normal vector direction of the point cloud is calculated to determine the center line of the weld, and the horizontal plane of the point cloud is corrected to the length direction of the weld, so that the cross-section plane is perpendicular to the weld surface.

[0040] S3, equidistant slice projection: equidistantly slice the weld point cloud processed in step S2 in the main direction of the weld, extract the point cloud information on each slice, and project it onto a two-dimensional plane to obtain two-dimensional scattered point coordinates;

[0041] Among them, in the isometric slicing projection process: isometric slicing uses the minimum directed bounding box method to plan the direction and distance of the slice; projection mapping selects the centroid point of the point cloud slice to form the projection surface, and establishes a local coordinate system on the projection surface, and then projects each point in the point cloud onto the projection surface to obtain a two-dimensional data point.

[0042] S4, weld cross-section model fitting: fitting the two-dimensional scattered points in step S3 to establish a weld cross-section model;

[0043] Among them, the weld section model is fitted by cubic B-spline smooth curve to fit the weld part, and the random sampling consistency algorithm is used to fit the parent material part.

[0044] S5, feature point extraction: for the weld cross-section model in step S4, extract the weld left feature point P according to the detection criteria l 、Weld right feature point P r , the two lowest points P on the left and right of the weld contour u1 , P u2 and the highest point P of the weld profile t

[0045] Among them, the inflection points on both sides of the weld of the fitting curve in feature extraction are P u1 , P u2 , the highest point of the fitting curve is P t The intersection of the fitting straight line and the fitting curve on both sides of the weld is P l and P r .

[0046] S6. Dimension measurement and defect calculation: perform difference calculation on the characteristic point coordinates obtained in step S5 to obtain the weld height, weld width, weld undercut and weld misalignment value of each section to reflect the weld formation size and defects.

[0047] Among them, the size measurement and defect calculation include: calculating the absolute value of the difference between the horizontal coordinate of the left characteristic point of the weld and the horizontal coordinate of the right characteristic point of the weld to obtain the weld width; calculating the difference between the vertical coordinate of the highest point of the weld curve contour and the vertical coordinate of the lowest point of the weld contour to obtain the weld height; calculating the absolute value of the difference between the vertical coordinate of the left lowest point of the weld contour curve and the vertical coordinate of the left characteristic point of the weld to obtain the left bite value of the weld; calculating the absolute value of the difference between the vertical coordinate of the right lowest point of the weld contour curve and the vertical coordinate of the right characteristic point of the weld to obtain the right bite value of the weld; calculating the absolute value of the difference between the vertical coordinate of the left characteristic point of the weld and the vertical coordinate of the right characteristic point of the weld to obtain the weld misalignment value.

[0048] Example 1

[0049] S1. Collecting weld point cloud data: A binocular structured light camera is used to collect weld point cloud data; the binocular structured light camera is installed on a conveyor belt rack at a height of 120 cm; the binocular structured light camera collects 3D point cloud data in a single time in no more than 2 seconds, and the measurement accuracy is not less than 0.2 mm.

[0050] S2, point cloud processing: the point cloud data in step S1 is processed, the voxel filtering method is used for downsampling, the statistical filtering method is used for removing abnormal points, and the principal component analysis method is used to correct the main direction of the point cloud; in the point cloud processing process: when the voxel filtering method is used for downsampling, the side length of the voxel cube is set to 0.5, and the point closest to the center of gravity in the voxel grid of the original point cloud data is used to replace the center of gravity; when the statistical filtering method is used to remove abnormal points, the abnormal points are filtered out based on the number of neighboring points and the standard deviation threshold, and the number of domain points is set to 50 and the standard deviation threshold is set to 1; when the principal component analysis method is used to correct the main direction of the point cloud, the main normal direction of the point cloud is calculated to determine the center line of the weld, and the horizontal plane of the point cloud is corrected to the length direction of the weld, so that the section plane is perpendicular to the weld surface.

[0051] S3, equidistant slicing projection: the weld point cloud processed in step S2 is equidistantly sliced ​​in the main direction of the weld, the point cloud information on each slice is extracted, and the projection mapping is performed on a two-dimensional plane to obtain two-dimensional scattered point coordinates; in the equidistant slicing projection process: the equidistant slicing uses the minimum directed bounding box method to plan the direction and distance of the slice; the projection mapping selects the centroid point of the point cloud slice to form a projection surface, and establishes a local coordinate system on the projection surface, and then projects each point in the point cloud onto the projection surface to obtain a two-dimensional data point.

[0052] S4, weld cross-section model fitting: fit the two-dimensional scattered points described in step S3 to establish a weld cross-section model; the weld cross-section model fitting adopts a cubic B-spline smooth curve to fit the weld part, and adopts a random sampling consistency algorithm to fit the parent material part to establish the weld cross-section model.

[0053] S5, feature point extraction: for the weld cross-section model in step S4, extract the weld left feature point P according to the detection criteria l , weld right feature point P r , the two lowest points P on the left and right of the weld contour u1 , P u2 and the highest point P of the weld profile t ; The inflection points on both sides of the weld of the fitting curve in the feature extraction are P u1 , P u2 , the highest point of the fitting curve is P t The intersection of the fitting straight line and the fitting curve on both sides of the weld is P l and P r .

[0054] S6, size measurement and defect calculation: perform difference calculation on the feature point coordinates obtained in step S5 to obtain the weld height, weld width, weld undercut and weld misalignment value of each section to reflect the weld formation size and defects; the size measurement and defect calculation include: calculating the absolute value of the difference between the horizontal coordinate of the left feature point of the weld and the horizontal coordinate of the right feature point of the weld to obtain the weld width; calculating the difference between the vertical coordinate of the highest point of the weld curve profile and the vertical coordinate of the lowest point of the weld profile to obtain the weld height; calculating the absolute value of the difference between the vertical coordinate of the left lowest point of the weld profile curve and the vertical coordinate of the left feature point of the weld to obtain the left undercut value of the weld; calculating the absolute value of the difference between the vertical coordinate of the right lowest point of the weld profile curve and the vertical coordinate of the right feature point of the weld to obtain the right undercut value of the weld; calculating the absolute value of the difference between the vertical coordinate of the left feature point of the weld and the vertical coordinate of the right feature point of the weld to obtain the weld misalignment value.

[0055] The weld size measurement and defect detection method based on point cloud features is described in detail below.

[0056] In step S2, the voxel filtering method is used to downsample to reduce the density of the point cloud, reduce data complexity, and retain the main structural information. The point cloud is traversed to obtain the maximum value x in the three coordinate systems of X, Y, and Z. max ,y max 、z max , minimum value x min ,y min 、z min , subtract the minimum value from the maximum value to get the side length of the bounding box. Set the side length of the voxel cube to r, so that the three directions of the X, Y, and Z coordinate axes are cut into M, N, and L parts, so a total of SUM = M*N*L voxel cubes are obtained:

[0057]

[0058] Calculate the centroid of each voxel cube, keep only the centroid point cloud of each voxel cube, and remove other point clouds. ijk is the center of mass of the voxel cube, p i Point cloud information, k is the number of point cloud points:

[0059]

[0060] After downsampling, according to the results, the point closest to the sampling point is selected as the final downsampling point according to the k-nearest neighbor search algorithm, and the center of gravity is replaced by the point closest to the center of gravity in the original point cloud data. Calculate the Euclidean distance ρ from the point in the voxel to the center of gravity c:

[0061]

[0062] Where (x c ,y c ,z c ) and (x i ,y i ,z i ) represent the centroid and voxel interior point coordinates respectively.

[0063] The statistical filtering method is used to remove abnormal points. Assuming that the coordinates of the nth point in the point cloud are P n (x n ,y n ,z n ), then point to any point P m (x m ,y m ,z m ) is Si. μ and σ are the average and standard deviation of the distance between each point and any other point. We assume that the multiple of the standard deviation is std, so when the average distance of n points near a point is within the standard range (μ-σ·std, μ+σ·std), the point is retained. If it is not within the range, it is defined as outlier filtering.

[0064] After the point cloud is preprocessed, a data analysis method based on principal component analysis is used to establish a covariance matrix, and the covariance matrix is ​​eigendecomposed to obtain eigenvectors and eigenvalues. The eigenvector corresponding to the maximum eigenvalue of the matrix is ​​the main direction, which can express the density of the points in the point cloud data or the tendency or trend of distribution in space. The eigenvector corresponding to the maximum eigenvalue is used as the main normal direction, and the three point cloud eigenvectors are placed as the three world coordinate directions at the center of the corrected weld point cloud, and the main direction is rotated to the Z axis. The eigendecomposition formula is as follows:

[0065] λv=Rυ

[0066] Among them, λ and υ are the eigenvalue and eigenvector of the 3*3 correlation coefficient matrix R respectively.

[0067] In step S3, in order to extract the weld feature information, the processed weld point cloud data is sliced ​​using a plane perpendicular to the main direction of the point cloud. The minimum directed bounding box of the weld point cloud is calculated according to the coordinate range of the weld point cloud data, and the 8 vertices of its bounding box cube are obtained. The direction and distance of the equally spaced slices are planned using these vertex information. In this embodiment, the corrected Z axis is used as the cutting axis, and the slice interval is determined to be δ until Z max =Z min +δ·n, n+1 groups of slice sections can be obtained. In order to better detect the weld features, a plane is constructed with the centroid of the point cloud slice as the center of the plane, and a local coordinate system is established with the plane unit normal vector as the main direction, and the obtained point cloud tangent plane data is converted from three-dimensional space to two-dimensional plane.

[0068] In step S4, in order to achieve two-dimensional scattered point fitting, a weld cross-section model is established, and a cubic B-spline curve is used to fit the weld area, which can better smooth the fitting data and has good local approximation performance. The overall equation is as follows:

[0069] P(t)=P0×F 0,3 (t)+P1×F 1,3 (t)+P2×F 2,3 (t)+P3×F 3,3 (t)

[0070] Where P i is the characteristic point of the control curve, F i,k (t) is the K-order B-spline basis function.

[0071] In order to fit the weld mother plate area, the curve is divided into two parts, the left and right inflection points of the weld curve are used as the dividing points, and the two parts are fitted with the random sampling consistency algorithm to find the two mother plate lines of the weld. The random sampling consistency algorithm randomly selects data points for model fitting, and then eliminates outliers through iterative optimization to finally obtain the optimal fitting straight line model.

[0072] In step S5, the left characteristic point P of the weld is extracted according to the detection criteria. l , weld right feature point P r , the two lowest points P on the left and right of the weld contour u1 , P u2 and the highest point P of the weld profile t The inflection point of the cubic B-spline fitting curve is considered to be P u1 , P u2 , the highest point is P t The intersection of the random sampling consensus algorithm fitting straight line and the curve on both sides is considered to be P l and P r.

[0073] In step S6, the weld size calculation and defect detection can be performed according to the determined feature points, as follows: the absolute value of the difference between the horizontal coordinate of the left feature point of the weld and the horizontal coordinate of the right feature point of the weld is calculated to obtain the weld width; the difference between the vertical coordinate of the highest point of the weld curve contour and the vertical coordinate of the lowest point of the weld contour is calculated to obtain the weld height; the absolute value of the difference between the vertical coordinate of the left lowest point of the weld contour curve and the vertical coordinate of the left feature point of the weld is calculated to obtain the left bite value of the weld; the absolute value of the difference between the vertical coordinate of the right lowest point of the weld contour curve and the vertical coordinate of the right feature point of the weld is calculated to obtain the right bite value of the weld; the absolute value of the difference between the vertical coordinate of the left feature point of the weld and the vertical coordinate of the right feature point of the weld is calculated to obtain the weld misalignment value.

[0074] The results of the weld measurements in this example are shown in Table 1.

[0075] Table 1 Dimension measurement and defect calculation results

[0076]

[0077]

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A weld size measurement and defect detection method based on point cloud features, characterized in that: The following steps are involved: S1. Collecting weld point cloud data: using a binocular structured light camera to collect weld point cloud data; S2, point cloud processing: processing the point cloud data in step S1, downsampling by using voxel filtering method, removing abnormal points by using statistical filtering method, and correcting the main direction of the point cloud by using principal component analysis method; S3, equidistant slice projection: equidistantly slice the weld point cloud processed in step S2 in the main direction of the weld, extract the point cloud information on each slice, and project it onto a two-dimensional plane to obtain two-dimensional scattered point coordinates; S4, weld cross-section model fitting: fitting the two-dimensional scattered points in step S3 to establish a weld cross-section model; S5, feature point extraction: for the weld cross-section model in step S4, extract the weld left feature point P according to the detection criteria l 、Weld right feature point P r , the two lowest points P on the left and right of the weld contour u1 , P u2 and the highest point P of the weld profile t ; S6. Dimension measurement and defect calculation: perform difference calculation on the characteristic point coordinates obtained in step S5 to obtain the weld height, weld width, weld undercut and weld misalignment value of each section to reflect the weld formation size and defects.

2. A weld size measurement and defect detection method based on point cloud features as claimed in claim 1, characterized in that: The binocular structured light camera described in step S1 is installed on the conveyor belt rack. The binocular structured light camera collects 3D point clouds in a single time within 2 seconds, and the measurement accuracy is not less than 0.2 mm.

3. A weld size measurement and defect detection method based on point cloud features as claimed in claim 1, characterized in that: In the point cloud processing process described in step S2: when downsampling is performed using the voxel filtering method, the point closest to the center of gravity in the voxel grid of the original point cloud data is used to replace the center of gravity; when the statistical filtering method is used to remove abnormal points, the abnormal points are filtered out based on the number of neighboring points and the standard deviation threshold calculation; when the principal component analysis method is used to correct the main direction of the point cloud, the main normal vector direction of the point cloud is calculated to determine the center line of the weld, and the horizontal plane of the point cloud is corrected to the length direction of the weld, so that the section plane is perpendicular to the weld surface.

4. A weld size measurement and defect detection method based on point cloud features as claimed in claim 1, characterized in that: In the isometric slicing projection process described in step S3: the isometric slicing uses the minimum directed bounding box method to plan the direction and distance of the slice; the projection mapping selects the centroid point of the point cloud slice to form a projection surface, and establishes a local coordinate system on the projection surface, and then projects each point in the point cloud onto the projection surface to obtain a two-dimensional data point.

5. The weld size measurement and defect detection method based on point cloud features according to claim 1, characterized in that: The weld cross-section model fitting in step S4 adopts a cubic B-spline smooth curve to fit the weld portion, and adopts a random sampling consistency algorithm to fit the base material portion.

6. A weld size measurement and defect detection method based on point cloud features as claimed in claim 1, characterized in that: The inflection point of the fitting curve in the feature extraction in step S5 is P u1 , P u2 , the highest point of the fitting curve is P t , the intersection point of the fitting straight line and the curve on both sides is P l and P r .

7. A weld size measurement and defect detection method based on point cloud features as claimed in claim 1, characterized in that: The dimension measurement and defect calculation in step S6 include: calculating the absolute value of the difference between the horizontal coordinate of the left characteristic point of the weld and the horizontal coordinate of the right characteristic point of the weld to obtain the weld width; calculating the difference between the vertical coordinate of the highest point of the weld curve profile and the vertical coordinate of the lowest point of the weld profile to obtain the weld height; calculating the absolute value of the difference between the vertical coordinate of the left lowest point of the weld profile curve and the vertical coordinate of the left characteristic point of the weld to obtain the left side bite value of the weld; calculating the absolute value of the difference between the vertical coordinate of the right lowest point of the weld profile curve and the vertical coordinate of the right characteristic point of the weld to obtain the right side bite value of the weld; calculating the absolute value of the difference between the vertical coordinate of the left characteristic point of the weld and the vertical coordinate of the right characteristic point of the weld to obtain the weld misalignment value.

Images