Weld inspection method, inspection device and readable storage medium for air pipe flange

Abstract

The application discloses a kind of welding seam detection method, detection device and readable storage medium for air pipe flange, the method includes: laser scanner obtains scanning data, and transmission to processing module;Scan data is converted into point cloud data;The normal vector of the point cloud depth direction of each point is obtained;The included angle of normal vector and reference plane is calculated;Obtain the fitting error between point cloud data and reconstructed curved surface model;If included angle > included angle threshold or fitting error > fitting error threshold, the processing module outputs detection early warning signal;If included angle ≤ included angle threshold and fitting error ≤ fitting error threshold, the processing module does not output detection early warning signal;The point cloud data of welding seam is obtained by laser scanner in the application, the influence on welding structure is avoided;The welding seam is detected by the double judgment of included angle and fitting error, the geometric shape information of welding seam can be quickly and accurately obtained, and the quality of welding seam can be comprehensively and high-precision evaluated.

Description

Technical Field

[0001] This invention relates to the field of duct flange manufacturing technology, specifically to a method, device, and readable storage medium for inspecting weld seams in duct flanges. Background Technology

[0002] Welding is a common process in the production of duct flanges, and weld inspection is essential. The main reasons include:

[0003] The quality of welding directly affects the safety and service life of duct flanges. Welding defects may cause problems such as air leakage and cracking during use, thereby affecting the normal operation and service life of the duct system.

[0004] Duct flanges have diverse applications and need to withstand various environmental factors such as pressure and temperature. Variations in weld quality can affect their load-bearing capacity, and weld defects may cause duct flanges to fail to meet actual requirements during use.

[0005] Methods for inspecting welds in duct flange production include visual inspection, magnetic particle testing, ultrasonic testing, and X-ray inspection. Visual inspection has drawbacks: it can only detect surface defects, not internal defects, and the results are easily influenced by the inspector's subjectivity. Magnetic particle testing can only detect ferromagnetic materials, not non-ferromagnetic materials, requires cleaning and treatment, and can easily cause environmental pollution. Ultrasonic testing requires specialized technology and equipment, demands highly skilled inspectors, is greatly affected by the weld geometry and material properties, and results are prone to error. X-ray inspection also requires specialized technology and equipment, demands highly skilled inspectors, is costly, and poses certain radiation hazards to the environment and personnel. Summary of the Invention

[0006] The technical problem to be solved by the present invention is the existence of welding defects in the current production of duct flanges. The purpose is to provide a method, device and readable storage medium for weld inspection of duct flanges, so as to achieve the purpose of inspecting the welds of duct flanges.

[0007] This invention is achieved through the following technical solution:

[0008] A method for inspecting welds on duct flanges, based on a weld inspection device, the weld inspection device comprising at least four laser scanners, the four laser scanners being fixedly installed at four welding points on the duct flange, and the scanning direction of the laser scanners being aligned with the weld of the duct flange.

[0009] The weld inspection method includes the following steps:

[0010] The first step is for the laser scanner to determine that the duct flange has reached the set detection position, start multiple laser scanners to acquire the scanning data of the welding position of the duct flange, and transmit the scanning data to the processing module.

[0011] The second step is for the processing module to convert the scanned data into point cloud data;

[0012] The third step is to obtain the covariance matrix corresponding to the point cloud data, and after performing singular value decomposition on the covariance matrix, obtain the normal vector of the point cloud depth direction of each point.

[0013] The fourth step is to set a reference plane, calculate the angle between the normal vector and the reference plane, and set an angle threshold.

[0014] The fifth step is to establish the rotation matrix corresponding to the point cloud data and normal vectors, and to reconstruct the 3D model of the weld; to obtain the fitting error between the point cloud data and the reconstructed surface model; and to set the fitting error threshold.

[0015] Step 6: If the included angle is greater than the included angle threshold or the fitting error is greater than the fitting error threshold, the processing module outputs a detection warning signal; if the included angle is less than or equal to the included angle threshold and the fitting error is less than or equal to the fitting error threshold, the processing module does not output a detection warning signal.

[0016] Specifically, in the second step, the methods for converting the scanned data into point cloud data include:

[0017] A1. Obtain the three-dimensional coordinate information of each point in the scanned data, and save the three-dimensional coordinate information as discretized three-dimensional coordinate point cloud data;

[0018] A2. Denoise the 3D coordinate point cloud data to obtain denoised point cloud data;

[0019] A2-1. Randomly sample the 3D coordinate point cloud data into a fixed-size subset;

[0020] A2-2. Use a Gaussian mixture model to model each subset and fit the true distribution in the point cloud;

[0021] A2-3. Calculate the likelihood probability for each point. Where p(x) represents the likelihood probability of point x, ω i ,μ i ,Σ i The distribution represents the weights, mean, and covariance matrix of the Gaussian mixture model, φ(x|μ i ,Σ i Let represent the probability density function of a multivariate Gaussian distribution; set a probability threshold p', and if p(x) < p', then delete the corresponding point x;

[0022] A3. Reconstruct the denoised point cloud data to obtain a reconstructed 3D model;

[0023] A4. Smooth the reconstructed 3D model and obtain the final point cloud data.

[0024] Specifically, in step A4, the method for smoothing the reconstructed 3D model includes:

[0025] A4-1. Establish the triangle network data structure corresponding to the reconstructed 3D model, and calculate the neighborhood of each vertex;

[0026] A4-2. Calculate the Laplacian matrix for each vertex i, and obtain the neighborhood points of point i through the Laplacian matrix;

[0027] A4-3. Constructing the smoothing equation: in Let n represent the new position of point i after the (k+1)th iteration. i Let w represent the number of neighbors of point i, N(i) represent the set of neighbors of point i, and w ij This represents the weight value between point i and point j. This indicates the position of point j after the k-th iteration;

[0028] A4-4. Set the iteration smoothness or number of iterations, and repeat A4-2 and A4-3 until a smooth 3D model is obtained;

[0029] In step A4, the methods for obtaining the final point cloud data include:

[0030] A4-5. Establish the triangle network data structure corresponding to the smoothed 3D model and perform sampling to obtain sampling points;

[0031] A4-6. Reconstruct the surface from the sampling points to generate a surface model;

[0032] A4-7. Obtain the point cloud data on the surface model as the final point cloud data.

[0033] Specifically, in the third step, the covariance matrix corresponding to the point cloud data is: Where n is the number of points in the point cloud, x i Let i be the position vector of the i-th point. The center of the point cloud;

[0034] Perform singular value decomposition on the covariance matrix to obtain eigenvectors and eigenvalues: M = USV T Where U is the eigenvector matrix, S is the eigenvalue diagonal matrix, and V... T The eigenvector matrix;

[0035] Obtain the minimum eigenvalue of the covariance matrix and use the eigenvector corresponding to the minimum eigenvalue as the normal vector n of the point cloud depth direction.

[0036] Specifically, the method for calculating the angle between the normal vector n and the reference plane is as follows: Project the normal vector n onto the reference plane to obtain the projection vector u of the normal vector n on the reference plane, and calculate the angle between the projection vector u and the normal vector n:

[0037] Specifically, in step five, the methods for reconstructing the 3D model include:

[0038] B1. Obtain the normal vector of each point and calculate the eigenvector of the normal vector to obtain the three principal direction vectors;

[0039] B2. Use the three principal direction vectors as column vectors to form a 3×3 matrix to obtain the rotation matrix;

[0040] B3. Multiply the point cloud data with the rotation matrix to perform coordinate transformation and obtain the point cloud data after coordinate transformation.

[0041] B4. Reconstruct the three-dimensional model from the point cloud data after coordinate transformation.

[0042] Specifically, in step five, the methods for obtaining the fitting error include:

[0043] B5. Perform surface fitting on the reconstructed 3D model to obtain the second surface model;

[0044] B6. Obtain the projection points of each point in the point cloud data onto the quadratic surface model: P t =P-(P-P0)·n·n T Where P is a point in the point cloud data, P0 is the point on the second surface model that is closest to point P, and n is the normal vector of point P;

[0045] B7. Calculate the distance between a point in the point cloud data and its projection point: L = ||PP t ||2;

[0046] B8. Obtain the fitting error.

[0047] A weld inspection device for duct flanges includes: a laser scanner and a processing module. Four laser scanners are respectively fixedly installed at four welding points of the duct flange, and the scanning direction of the laser scanners is aligned with the weld of the duct flange. The laser scanners are located downstream of a welding robot, and the data output terminal of the laser scanners is electrically connected to the data input terminal of the processing module.

[0048] Specifically, the data processing module includes:

[0049] The conversion module is used to convert scanned data into point cloud data;

[0050] The normal vector acquisition module is used to obtain the covariance matrix corresponding to the point cloud data, perform singular value decomposition on the covariance matrix, and obtain the normal vector of the point cloud depth direction of each point.

[0051] The angle calculation module is used to set a reference plane, calculate the angle between the normal vector and the reference plane, and set the angle threshold.

[0052] The fitting error calculation module is used to establish the rotation matrix corresponding to the point cloud data and the normal vector, and to reconstruct the 3D model of the weld; to obtain the fitting error between the point cloud data and the reconstructed surface model; and to set the fitting error threshold.

[0053] The judgment module is used to judge the included angle and the fitting error. If the included angle is greater than the included angle threshold or the fitting error is greater than the fitting error threshold, the processing module outputs a detection warning signal; if the included angle is less than or equal to the included angle threshold and the fitting error is less than or equal to the fitting error threshold, the processing module does not output a detection warning signal.

[0054] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in any of the preceding claims.

[0055] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0056] This invention acquires point cloud data of the weld seam using a laser scanner and evaluates the weld seam based on the point cloud data, without causing any physical damage to the weld seam and avoiding any impact on the welded structure.

[0057] By introducing covariance matrix and singular value decomposition to obtain the normal vector, the included angle is calculated. Then, a rotation matrix is ​​used for coordinate transformation and three-dimensional reconstruction to obtain the fitting error. By using the dual judgment of included angle and fitting error to detect the weld, the geometric shape information of the weld can be obtained quickly and accurately, and the weld can be comprehensively and accurately evaluated. Attached Figure Description

[0058] The accompanying drawings illustrate exemplary embodiments of the present invention and, together with the description thereof, serve to explain the principles of the invention. These drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, but do not constitute a limitation on the embodiments of the present invention.

[0059] Figure 1 This is a schematic flowchart of a weld inspection method for duct flanges according to the present invention.

[0060] Figure 2 This is a schematic diagram of a weld inspection device for duct flanges according to the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0062] It should also be noted that, for ease of description, only the parts relevant to the present invention are shown in the accompanying drawings.

[0063] Where there is no conflict, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0064] Example 1

[0065] This implementation provides, for example Figure 2 The device shown is for inspecting weld seams of duct flanges, comprising: a laser scanner and a processing module. Four laser scanners are fixedly installed at the four welding points of the duct flange, and the scanning direction of the laser scanners is aligned with the weld seam of the duct flange. The laser scanners are located downstream of the welding robot, and the data output end of the laser scanners is electrically connected to the data input end of the processing module.

[0066] A laser scanner is a measuring device capable of quickly and accurately acquiring point cloud data of a three-dimensional object's surface. It determines the three-dimensional coordinates of the object's surface by emitting a laser beam and receiving the signal reflected back from the reflected laser. The working principle of a laser scanner is to use a laser to generate a high-energy laser beam, which is focused and scanned by a set of lenses and scanning mirrors. The laser beam scans along horizontal and vertical directions to form a scanning plane, which is then projected onto the surface of the object being measured. The laser beam reflected back from the object's surface is received by a receiver and converted into an electrical signal. After processing, the three-dimensional coordinate information of that point can be obtained.

[0067] In laser scanner measurements, the laser beam is reflected and scattered upon reaching the surface of an object, resulting in reflected and scattered light. Reflected light refers to the light rays that are reflected back at the same angle according to the law of reflection after reaching the object's surface. Scattered light refers to the light rays that are scattered back at various angles due to the roughness and inhomogeneity of the object's surface.

[0068] Therefore, in actual laser scanning, point cloud data of an object's surface can be obtained by collecting reflected and scattered light. Point cloud data is a collection of discrete point coordinates, where the coordinates of each point represent the three-dimensional coordinates of the object's surface at that location.

[0069] In this application, by placing the laser scanner downstream of the welding robot, after the welding robot welds the duct flange, it moves to the position of the laser scanner by means of a robotic arm or conveyor belt, and then the morphology of the weld can be collected by the laser scanner.

[0070] The processing module includes multiple different internal logic modules. Alternatively, it can consist of multiple different processors forming a processing cluster, which serves as the processing module. The data processing module includes:

[0071] The conversion module is used to convert scanned data into point cloud data;

[0072] The normal vector acquisition module is used to obtain the covariance matrix corresponding to the point cloud data, perform singular value decomposition on the covariance matrix, and obtain the normal vector of the point cloud depth direction of each point.

[0073] The angle calculation module is used to set a reference plane, calculate the angle between the normal vector and the reference plane, and set the angle threshold.

[0074] The fitting error calculation module is used to establish the rotation matrix corresponding to the point cloud data and the normal vector, and to reconstruct the 3D model of the weld; to obtain the fitting error between the point cloud data and the reconstructed surface model; and to set the fitting error threshold.

[0075] The judgment module is used to judge the included angle and the fitting error. If the included angle is greater than the included angle threshold or the fitting error is greater than the fitting error threshold, the processing module outputs a detection warning signal; if the included angle is less than or equal to the included angle threshold and the fitting error is less than or equal to the fitting error threshold, the processing module does not output a detection warning signal.

[0076] Example 2

[0077] This embodiment provides a method for inspecting weld seams on duct flanges. Based on the weld seam inspection device for duct flanges in Embodiment 1, this method utilizes a laser scanner to acquire point cloud data, obtains normal vectors through singular value decomposition, and establishes a rotation matrix corresponding to the point cloud data and normal vectors to reconstruct a three-dimensional model, thereby achieving the inspection of duct flange weld seams. By setting included angle thresholds and fitting error thresholds, it is possible to determine whether the weld seam meets the requirements and output an early warning signal when necessary.

[0078] like Figure 1 As shown, the weld inspection method includes the following steps:

[0079] The first step is for the laser scanner to determine that the duct flange has reached the set detection position, start multiple laser scanners to acquire the scanning data of the welding position of the duct flange, and transmit the scanning data to the processing module.

[0080] In this step, the first step is to determine the detection position of the duct flange. This can be achieved by setting a trigger or sensor to detect the flange's position. When the flange reaches the designated location, the trigger or sensor sends a signal to the laser scanner, which then begins scanning the welding position.

[0081] Next, multiple laser scanners need to be activated to scan the four welding points of the duct flange. These laser scanners can work simultaneously or scan in turn at certain time intervals. The laser scanners obtain the three-dimensional spatial information of the duct flange by sending laser beams and recording the time and position information of the reflected light.

[0082] Finally, the scanned data is transmitted to the processing module for processing. The processing module can fuse data acquired by multiple laser scanners to obtain more accurate point cloud data and provide input data for subsequent steps. In this step, it is necessary to ensure the accuracy and integrity of the scanned data to guarantee the accuracy and reliability of the subsequent data processing and analysis results.

[0083] The second step is for the processing module to convert the scanned data into point cloud data;

[0084] Point cloud data is a collection of points in three-dimensional space, each with its own position coordinates and other attribute information. In weld inspection, point cloud data consists of a series of points acquired by a laser scanner. The coordinates of these points can represent the surface shape of the duct flange, thus reflecting the morphological information of the weld.

[0085] Before converting point cloud data, the scanned data needs to be processed and filtered. This processing includes noise removal, data smoothing, and outlier removal. These processes improve the quality of the point cloud data and reduce noise and outliers.

[0086] Next, the processed scan data needs to be converted into point cloud data. The conversion process mainly involves converting information such as point coordinates and reflection intensity in the scan data into information such as point coordinates and normal vectors in the point cloud data.

[0087] Finally, point cloud data can be used for subsequent data processing and analysis, such as obtaining weld shape information, calculating normal vectors, and reconstructing 3D models. These analysis results can be used to determine whether the weld meets requirements and to output corresponding detection and warning signals.

[0088] The third step is to obtain the covariance matrix corresponding to the point cloud data, and after performing singular value decomposition on the covariance matrix, obtain the normal vector of the point cloud depth direction of each point.

[0089] In point cloud data, each point has its own position coordinates and other attribute information. The analysis and processing of point cloud data requires the calculation and derivation of this attribute information. Among these, the normal vector is one of the important attribute information in point cloud data, which can be used to characterize the curvature and surface normal information of the point cloud data. In weld inspection, the calculation of the normal vector can help determine the location and shape information of the weld.

[0090] Normal vectors in point cloud data can be obtained by calculating the covariance matrix of each point and its surrounding points. Singular value decomposition (SVD) is a linear algebraic method that decomposes a matrix into three parts: left singular vectors, singular values, and right singular vectors. By decomposing the covariance matrix, the normal vector information in the depth direction of the point cloud can be obtained.

[0091] The fourth step is to set a reference plane, calculate the angle between the normal vector and the reference plane, and set an angle threshold.

[0092] A plane parallel to the welding plane is designated as the reference plane (the distance between the reference plane and the welding plane represents the standard weld thickness). The angle between the normal vector and the reference plane is calculated. The smaller the angle, the closer the normal vector is to the reference plane, indicating that a point in the point cloud data is more likely to belong to the reference plane. To determine whether a point in the point cloud data belongs to the reference plane, an angle threshold needs to be set. If the angle between the normal vector and the reference plane is less than the set threshold, it is classified as a weld point that meets the set standard thickness; if the angle is greater than the set threshold, it is classified as a weld point that does not meet the standard thickness. The threshold setting needs to be adjusted according to the actual situation to ensure the accuracy and reliability of the detection results.

[0093] The fifth step is to establish the rotation matrix corresponding to the point cloud data and normal vectors, and to reconstruct the 3D model of the weld; to obtain the fitting error between the point cloud data and the reconstructed surface model; and to set the fitting error threshold.

[0094] Before reconstructing the 3D model, a rotation matrix corresponding to the point cloud data and normal vectors needs to be established. The rotation matrix rotates the points in the point cloud data along the direction of the normal vector, ensuring that the plane containing the weld is perpendicular to the coordinate system containing the point cloud data. This eliminates errors caused by the tilt of the point cloud data in the coordinate system, thereby improving the accuracy and precision of weld inspection.

[0095] After reconstructing the 3D model, it is necessary to calculate the fitting error between the point cloud data and the reconstructed surface model. The fitting error represents the degree of difference between the point cloud data and the surface model, and is one of the indicators for evaluating the reconstruction accuracy.

[0096] Step 6: If the included angle is greater than the included angle threshold or the fitting error is greater than the fitting error threshold, the processing module outputs a detection warning signal; if the included angle is less than or equal to the included angle threshold and the fitting error is less than or equal to the fitting error threshold, the processing module does not output a detection warning signal.

[0097] Based on a set threshold, it can be determined whether a point in the point cloud data belongs to a reference plane, and a corresponding detection warning signal is output. If the angle between the normal vector and the reference plane is greater than the set threshold, or the fitting error between the point cloud data and the reconstructed surface model is greater than the set threshold, the weld is considered unacceptable and needs to be repaired or re-welded, and a corresponding detection warning signal is output. If the angle between the normal vector and the reference plane is less than the set threshold and the fitting error between the point cloud data and the reconstructed surface model is less than the set threshold, the weld is considered acceptable and does not require repair or re-welding, and no detection warning signal is output.

[0098] Warning signals can be detected using sound, light, or other methods to facilitate timely detection and handling of problems.

[0099] Example 3

[0100] This embodiment provides a detailed description of the second step, in which the specific method for converting scanned data into point cloud data includes:

[0101] A1. Obtain the three-dimensional coordinate information of each point in the scanned data, and save the three-dimensional coordinate information as discretized three-dimensional coordinate point cloud data;

[0102] During the scanning process, the laser scanner records the position coordinates of each point in three-dimensional space. This position information can be represented as three-dimensional coordinates. By acquiring the three-dimensional coordinates of each point in the scan data and saving it as discretized three-dimensional point cloud data, the system can achieve this. Discretized point cloud data typically consists of a large number of points, each with its own position coordinates and other attribute information.

[0103] A2. Denoise the 3D coordinate point cloud data to obtain denoised point cloud data;

[0104] During point cloud data processing, some noise points will appear, which usually interfere with the analysis and processing of point cloud data and cause errors. Therefore, before proceeding with subsequent point cloud data analysis and processing, it is necessary to perform noise reduction processing. The purpose of noise reduction processing is to remove as many noise points as possible and retain useful point cloud data information.

[0105] A2-1. Randomly sample 3D coordinate point cloud data into a fixed-size subset. Subsampling is often used in denoising to reduce the amount and complexity of point cloud data. Subsampling typically involves randomly sampling point cloud data into a fixed-size subset. This reduces the amount of data, speeds up the algorithm, and makes the processing results more stable.

[0106] A2-2. A Gaussian Mixture Model (GMM) is used to model each subset and fit the true distribution in the point cloud. A GMM is a commonly used statistical model that decomposes a complex distribution into a weighted sum of multiple Gaussian distributions, where each Gaussian distribution corresponds to a cluster. In point cloud denoising, a GMM can be used to model each subset and fit the true distribution in the point cloud. This approach better describes the characteristics and distribution of the point cloud data, facilitating subsequent processing and analysis.

[0107] A2-3. Calculate the likelihood probability for each point. Where p(x) represents the likelihood probability of point x, ω i ,μ i ,Σ i The distribution represents the weights, mean, and covariance matrix of the Gaussian mixture model, φ(x|μ i ,Σ i Let represent the probability density function of a multivariate Gaussian distribution; a probability threshold p' is set, and if p(x) < p', the corresponding point x is deleted; the likelihood probability is the probability that a point comes from a certain Gaussian distribution, which can be calculated using the probability density function of a multivariate Gaussian distribution. Setting a probability threshold is to retain the true information in the point cloud data and delete points that may be noise. If the likelihood probability of a point is lower than the set probability threshold, it is considered a noise point and should be deleted from the point cloud data.

[0108] A3. Reconstruct the denoised point cloud data to obtain a reconstructed 3D model; commonly used 3D model reconstruction methods include mesh-based reconstruction algorithms and voxel-based reconstruction algorithms.

[0109] A4. Smooth the reconstructed 3D model and obtain the final point cloud data.

[0110] The reconstructed 3D surface model may contain some non-smooth regions, which may interfere with subsequent analysis and processing and introduce errors. Therefore, smoothing the reconstructed 3D surface model is necessary.

[0111] In step A4, the reconstructed 3D model needs to be smoothed. Specific methods include:

[0112] A4-1. Establish the triangle network data structure corresponding to the reconstructed 3D model, and calculate the neighborhood of each vertex;

[0113] This process converts the vertices, edges, and faces of a 3D model into a triangular data structure for subsequent processing and analysis. In this step, you need to build the triangular network data structure corresponding to the reconstructed 3D model and calculate the neighborhood of each vertex. The neighborhood of each vertex refers to the set of other vertices adjacent to that vertex.

[0114] A4-2. Calculate the Laplacian matrix for each vertex i, and obtain the neighborhood points of point i through the Laplacian matrix;

[0115] The Laplacian matrix is ​​a matrix that represents the structure of a graph, describing the relationships and interactions between nodes. In smoothing processes, the Laplacian matrix can be used to describe the neighborhood structure and interactions between points in a 3D model. Neighborhood points refer to the set of other points adjacent to point i.

[0116] A4-3. Constructing the smoothing equation: in Let n represent the new position of point i after the (k+1)th iteration. i Let w represent the number of neighbors of point i, N(i) represent the set of neighbors of point i, and w ij This represents the weight value between point i and point j. This indicates the position of point j after the k-th iteration;

[0117] A4-4. Smoothing usually requires multiple iterations to achieve a better smoothing effect. Set the iteration smoothness or number of iterations, and repeat A4-2 and A4-3 until a smooth 3D model is obtained.

[0118] In step A4, the method for obtaining the final point cloud data includes:

[0119] A4-5. Establish the triangle network data structure corresponding to the smoothed 3D model and perform sampling to obtain sampling points;

[0120] After smoothing the 3D model, it needs to be converted into a triangular network data structure and sampled to obtain sampling points. Sampling points are a set of points on the 3D model that can represent its features and shape, and can also serve as input for subsequent surface reconstruction.

[0121] A4-6. Perform surface reconstruction on the sampled points to generate a surface model; surface reconstruction refers to the process of recovering the shape of a surface from point cloud data. Various surface reconstruction algorithms can be used, such as Poisson reconstruction, mesh-based reconstruction, and deep learning-based reconstruction.

[0122] A4-7. Obtain the point cloud data on the surface model as the final point cloud data.

[0123] Example 4

[0124] This embodiment further explains the third step.

[0125] The covariance matrix is ​​a matrix that describes the distribution and correlation of data. It can be used to calculate the features and shape of point cloud data. The covariance matrix corresponding to point cloud data is: Where n is the number of points in the point cloud, x i Let i be the position vector of the i-th point. The center of the point cloud;

[0126] Perform singular value decomposition on the covariance matrix to obtain eigenvectors and eigenvalues: M = USV T Where U is the eigenvector matrix, S is the eigenvalue diagonal matrix, and V... T is the eigenvector matrix; singular value decomposition is a linear algebra operation used to decompose a matrix into the product of three matrices, including an eigenvector matrix, an eigenvalue diagonal matrix, and the transpose of another eigenvector matrix.

[0127] After calculating the eigenvectors and eigenvalues, the main orientation and shape of the point cloud data can be determined based on the magnitude of the eigenvalues. Typically, the depth direction of the point cloud data corresponds to the eigenvector with the smallest eigenvalue. The smallest eigenvalue of the covariance matrix is ​​obtained, and the eigenvector corresponding to the smallest eigenvalue is used as the normal vector *n* of the point cloud's depth direction. This normal vector can be used to calculate the point cloud data's normal vectors and surface normals to describe the shape and features of the point cloud data.

[0128] In the third step, the method for calculating the angle between the normal vector n and the reference plane is as follows: Project the normal vector n onto the reference plane to obtain the projection vector u of the normal vector n on the reference plane, and calculate the angle between the projection vector u and the normal vector n:

[0129] Example 5

[0130] In step five, the methods for reconstructing the 3D model include:

[0131] B1. Obtain the normal vector of each point and calculate the eigenvector of the normal vector to obtain three principal direction vectors; the principal direction vectors can be used to describe the main direction and shape of the point cloud data.

[0132] B2. Use the three principal direction vectors as column vectors to form a 3×3 matrix to obtain the rotation matrix; the rotation matrix describes the rotation transformation of the point cloud data and can be used to rotate the point cloud data from the current coordinate system to another coordinate system.

[0133] B3. Multiply the point cloud data with the rotation matrix to perform coordinate transformation and obtain the point cloud data after coordinate transformation.

[0134] B4. Perform secondary reconstruction of the 3D model from the coordinate-transformed point cloud data. After coordinate transformation, the point cloud data is further processed and analyzed to extract its features and shapes, thereby generating a more accurate and complete 3D model.

[0135] In step five, the methods for obtaining the fitting error include:

[0136] B5. Perform surface fitting on the reconstructed 3D model to obtain the second surface model; surface fitting refers to the process of fitting point cloud data to obtain a continuous surface.

[0137] B6. Obtain the projection points of each point in the point cloud data onto the quadratic surface model: P t =P-(P-P0)·n·n T Where P is a point in the point cloud data, P0 is the point on the second surface model that is closest to point P, and n is the normal vector of point P;

[0138] B7. Calculate the distance between a point in the point cloud data and its projection point: L = ||PP t ||2; The distance here refers to the Euclidean distance between a point in the point cloud data and its projection point on the second surface model.

[0139] B8. Obtain the fitting error. Fitting error refers to the average distance between a point in the point cloud data and its projection point on the second surface model.

[0140] Example 6

[0141] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the methods described above.

[0142] Without loss of generality, computer-readable media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instruction data structures, program modules, or other data. Computer storage media includes RAM, ROM, EPROM, EEPROM, flash memory or other solid-state storage technologies, CD-ROM, DVD or other optical storage, magnetic tape cassettes, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will recognize that computer storage media are not limited to the above-mentioned types. The aforementioned system memories and mass storage devices can be collectively referred to as memory.

[0143] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment / mode or example is included in at least one embodiment / mode or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0144] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0145] Those skilled in the art should understand that the above embodiments are merely for illustrating the present invention and are not intended to limit the scope of the invention. Those skilled in the art can make other changes or modifications based on the above invention, and these changes or modifications still fall within the scope of the present invention.

Claims

1. A method for inspecting weld seams on duct flanges, characterized in that, A weld inspection device includes at least four laser scanners, which are respectively fixed at four welding points of a duct flange, and the scanning direction of the laser scanners is aligned with the weld of the duct flange. The weld inspection method Includes the following steps: The first step is for the laser scanner to determine that the duct flange has reached the set detection position, start multiple laser scanners to acquire the scanning data of the welding position of the duct flange, and transmit the scanning data to the processing module. The second step is for the processing module to convert the scanned data into point cloud data; The third step is to obtain the covariance matrix corresponding to the point cloud data, and after performing singular value decomposition on the covariance matrix, obtain the normal vector of the point cloud depth direction of each point. The fourth step is to set a reference plane, calculate the angle between the normal vector and the reference plane, and set an angle threshold. The fifth step is to establish the rotation matrix corresponding to the point cloud data and normal vectors, and to reconstruct the 3D model of the weld; to obtain the fitting error between the point cloud data and the reconstructed surface model; and to set the fitting error threshold. Step 6: If the included angle is greater than the included angle threshold or the fitting error is greater than the fitting error threshold, the processing module outputs a detection warning signal; if the included angle is less than or equal to the included angle threshold and the fitting error is less than or equal to the fitting error threshold, the processing module does not output a detection warning signal.

2. The method for inspecting weld seams in duct flanges according to claim 1, characterized in that, The specific methods for converting scanned data into point cloud data in the second step include: A1. Obtain the three-dimensional coordinate information of each point in the scanned data, and save the three-dimensional coordinate information as discretized three-dimensional coordinate point cloud data; A2. Denoise the 3D coordinate point cloud data to obtain denoised point cloud data; A2-1. Randomly sample the 3D coordinate point cloud data into a fixed-size subset; A2-2. Use a Gaussian mixture model to model each subset and fit the true distribution in the point cloud; A2-3. Calculate the likelihood probability for each point. ,in Let x represent the likelihood probability of point x. The distribution represents the weights, mean, and covariance matrix of the Gaussian mixture model. Represent the probability density function of a multivariate Gaussian distribution; set a probability threshold. ,like If the corresponding point x is deleted, then delete the point x. A3. Reconstruct the denoised point cloud data to obtain a reconstructed 3D model; A4. Smooth the reconstructed 3D model and obtain the final point cloud data.

3. The method for inspecting weld seams in duct flanges according to claim 2, characterized in that, In step A4, the methods for smoothing the reconstructed 3D model include: A4-1. Establish the triangle network data structure corresponding to the reconstructed 3D model, and calculate the neighborhood of each vertex; A4-2. Calculate the Laplacian matrix for each vertex i, and obtain the neighborhood points of point i through the Laplacian matrix; A4-3. Constructing the smoothing equation: ,in This represents the new position of point i after the (k+1)th iteration. This represents the number of neighbors of point i. Let i represent the set of neighborhood points of point i. This represents the weight value between point i and point j. This indicates the position of point j after the k-th iteration; A4-4. Set the iteration smoothness or number of iterations, and repeat A4-2 and A4-3 until a smooth 3D model is obtained; In step A4, the methods for obtaining the final point cloud data include: A4-5. Establish the triangle network data structure corresponding to the smoothed 3D model and perform sampling to obtain sampling points; A4-6. Reconstruct the surface from the sampling points to generate a surface model; A4-7. Obtain the point cloud data on the surface model as the final point cloud data.

4. The method for inspecting weld seams in duct flanges according to claim 1, characterized in that, In the third step, the covariance matrix corresponding to the point cloud data is: Where n is the number of points in the point cloud. Let i be the position vector of the i-th point. The center of the point cloud; Singular value decomposition is performed on the covariance matrix to obtain eigenvectors and eigenvalues: ,in, The eigenvector matrix, It is an eigenvalue diagonal matrix. The eigenvector matrix; Find the minimum eigenvalue of the covariance matrix and use the eigenvector corresponding to the minimum eigenvalue as the normal vector of the point cloud depth direction. .

5. The method for inspecting weld seams in duct flanges according to claim 4, characterized in that, Calculate the normal vector The method for determining the angle with the reference plane is as follows: [The normal vector is then used as a reference.] Project onto the reference plane to obtain the normal vector. Projection vector on the reference plane Calculate the projection vector With normal vector The included angle: .

6. The method for inspecting weld seams in duct flanges according to claim 1, characterized in that, In step five, the methods for reconstructing the 3D model include: B1. Obtain the normal vector of each point and calculate the eigenvector of the normal vector to obtain the three principal direction vectors; B2. Use the three principal direction vectors as column vectors to form a 3×3 matrix to obtain the rotation matrix; B3. Multiply the point cloud data with the rotation matrix to perform coordinate transformation and obtain the point cloud data after coordinate transformation. B4. Reconstruct the three-dimensional model from the point cloud data after coordinate transformation.

7. The method for inspecting weld seams in duct flanges according to claim 6, characterized in that, In step five, the methods for obtaining the fitting error include: B5. Perform surface fitting on the reconstructed 3D model to obtain the second surface model; B6. Obtain the projection points of each point in the point cloud data onto the quadratic surface model: Where P is a point in the point cloud data, It is the point on the second surface model that is closest to point P. It is the normal vector of point P; B7. Calculate the distance between a point in the point cloud data and its projection point: ; B8. Obtain the fitting error. .

8. A weld inspection device for duct flanges, characterized in that, include: The laser scanner and processing module are respectively fixedly installed at the four welding points of the duct flange, and the scanning direction of the laser scanner is aligned with the weld seam of the duct flange. The laser scanner is located downstream of the welding robot, and the data output terminal of the laser scanner is electrically connected to the data input terminal of the processing module. The processing module includes: The conversion module is used to convert scanned data into point cloud data; The normal vector acquisition module is used to obtain the covariance matrix corresponding to the point cloud data, perform singular value decomposition on the covariance matrix, and obtain the normal vector of the point cloud depth direction of each point. The angle calculation module is used to set a reference plane, calculate the angle between the normal vector and the reference plane, and set the angle threshold. The fitting error calculation module is used to establish the rotation matrix corresponding to the point cloud data and the normal vector, and to reconstruct the 3D model of the weld; to obtain the fitting error between the point cloud data and the reconstructed surface model; and to set the fitting error threshold. The judgment module is used to judge the included angle and the fitting error. If the included angle is greater than the included angle threshold or the fitting error is greater than the fitting error threshold, the processing module outputs a detection warning signal; if the included angle is less than or equal to the included angle threshold and the fitting error is less than or equal to the fitting error threshold, the processing module does not output a detection warning signal.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images