A method for measuring the surface morphology of parts using laser-directed energy processes

Abstract

This invention discloses a method for measuring the surface morphology of a part during a laser-directed energy process. The method includes the following steps: acquiring height information of the part surface and reconstructing a point cloud; extracting high-density distribution regions of the point cloud height values; fitting a reference plane based on an improved RANSAC algorithm; removing the interior points of the reference plane as background point clouds; projecting the point cloud onto the XOY plane to generate a mask image; retaining the largest connected components in the mask image and removing noisy point clouds; calculating the transformation matrix T1 from the reference plane to the XOY plane; aligning the printing reference plane of the point cloud with the XOY plane based on the transformation matrix T1; extracting the first principal direction of the point cloud; calculating the transformation matrix T2 from the first principal direction to the printing direction, ensuring that the direction of the point cloud aligns with the deposition direction of the current printing layer, thus obtaining the surface morphology of the part; and extracting the geometric features of the thin-walled part using a random sampling statistical method. This invention can improve the accuracy and reliability of the part surface morphology.

Description

Technical Field

[0001] This invention relates to the field of surface morphology measurement technology for parts, and specifically to a method for measuring the surface morphology of parts using a laser-directed energy process. Background Technology

[0002] Laser-directed energy deposition (LDED), also known as laser near-net-shape forming, is a metal additive manufacturing technology. It utilizes a high-power laser to melt metal powder and deposit it along a pre-defined printing path onto a substrate surface, forming parts layer by layer. Compared to traditional manufacturing processes, LPBF offers advantages such as high design freedom, short product cycles, and high material utilization. However, during the LDED forming process, problems such as large dimensional deviations and poor surface quality exist, severely reducing the forming capacity and quality of LDED. Surface quality issues, in particular, can significantly affect the printing quality of subsequent layers and the fatigue life of parts, even causing printing failures. Such failures result in substantial waste of manpower and materials. A line laser scanning camera can quickly measure the height information of the part surface and, combined with the laser line scanning speed and resolution, obtain a point cloud of the part surface. Then, through point cloud segmentation, noise reduction, and pose adjustment, the final surface morphology of the part is obtained. Compared to traditional surface quality inspection methods, such as 3D profilometers and optical microscopes, this method can accelerate the measurement of part surface morphology during and after printing and provides a wider field of view. It can improve the monitoring capabilities and post-characterization level of the LDED process, and promote the development of intelligent manufacturing.

[0003] Chen et al. mounted a line laser scanning camera at the end of a robotic arm to acquire the surface morphology of the parts during printing and segmented the substrate point cloud and target point cloud using the RANSAC algorithm. The RANSAC-based point cloud segmentation algorithm has a slow convergence speed and time-consuming iterations. Binega et al., during scanning, fitted the point cloud data of the scanned substrate surface to obtain a straight line and calculated the angle alpha between the line and the XOY plane. Binega et al. calculated alpha for each scanning contour to correct the point cloud pose. However, applying different transformation angles to each scanning contour causes surface morphology distortion. Furthermore, this method ignores the influence of substrate deformation. Experiments show that the influence of substrate deformation is much greater than the influence of camera vibration during the scanning process. Both Chen et al. and Binega et al. neglected the problem of the camera laser line not being perpendicular to the printing direction due to camera mounting errors. Therefore, it is necessary to correct the alignment of the point cloud direction and the printing direction. Summary of the Invention

[0004] To overcome the defects and shortcomings of existing technologies, this invention provides a method for measuring the surface morphology of parts using laser-directed energy processes. This invention is based on acquiring the height information of the part surface using a laser line scanning camera, and combining the scanning speed and resolution of the laser line to generate a point cloud of the part surface. Through point cloud segmentation, noise reduction, and pose adjustment, the surface morphology of the part is finally obtained. This invention can improve the accuracy and reliability of the part surface morphology.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] This invention provides a method for measuring the surface morphology of a part during a laser-directed energy process, comprising the following steps:

[0007] The height information of the part surface is collected by a line laser scanning camera, and the point cloud of the part surface is reconstructed by combining the resolution and scanning speed of the line laser.

[0008] Extract high-density distribution areas of point cloud height values;

[0009] The optimal reference plane is obtained by fitting based on the improved RANSAC algorithm;

[0010] The interior points of the reference plane are removed as background point clouds. The interior points of the reference plane are caused by the substrate surface.

[0011] The point cloud is projected onto the XOY plane to generate a mask image;

[0012] The mask image retains the largest connected component, and noisy point clouds are removed to obtain denoised point clouds;

[0013] Calculate the transformation matrix T1 from the reference plane to the XOY plane;

[0014] The transformation matrix T1 is applied to the denoised point cloud to make the printing reference plane of the point cloud coincide with the XOY plane and correct the height value of the point cloud.

[0015] Extract the first principal direction of the point cloud;

[0016] Calculate the transformation matrix T2 from the first principal direction to the printing direction and apply it to the point cloud so that the direction of the point cloud is consistent with the deposition direction of the current printing layer, and obtain the final pose-corrected point cloud as the surface morphology of the part.

[0017] Geometric features of thin-walled structures were extracted using random sampling statistical methods.

[0018] As a preferred technical solution, the high-density distribution area of ​​point cloud height values ​​is extracted, specifically including:

[0019] Histogram statistics are performed on the point cloud height values ​​to divide the data into intervals. The number of data points in each interval is obtained, and the data intervals with more than a threshold number of data points are extracted as the high-density distribution areas of the point cloud height values.

[0020] As a preferred technical solution, the optimal reference plane is obtained by fitting based on the improved RANSAC algorithm, specifically including:

[0021] Set the number of iterations and the distance threshold. In the high-density distribution area of ​​the point cloud height value, randomly select 3 points to fit a plane, calculate the normal vector between the three points, obtain the normal vector by cross product, and calculate the distance from each point in the point cloud to the plane.

[0022] Points whose distance is less than the set distance threshold are recorded as inliers. Through iterative iteration, the optimal fitting plane is obtained as the reference plane.

[0023] As a preferred technical solution, the number of iterations and the distance threshold are set, specifically including:

[0024] The number of iterations and the distance threshold were set through preliminary experiments. In the preliminary experiments, the number of iterations was determined by the relationship between the number of interior points and the number of iterations, so that the fitting reached the convergence condition. The background point cloud was labeled, the height distribution characteristics of the background point cloud were extracted, and the distance threshold was determined by the 3σ principle.

[0025] As a preferred technical solution, three points are randomly selected to fit a plane, and the plane equation is expressed as:

[0026] ax + by + cz + d = 0

[0027] Where a, b, and c are the components of the plane normal vector, and d is the constant term of the plane equation;

[0028] The normal vector between the three points p1, p2, and p3 is calculated as follows:

[0029] n1 = p2 - p1

[0030] n2 = p3 - p1

[0031] The cross product yields the normal vector: n = n1 × n2;

[0032] The distance from each point in the point cloud to the plane is calculated as follows:

[0033]

[0034] Where distance represents the distance from each point in the point cloud to the plane.

[0035] As a preferred technical solution, the first principal direction of the point cloud is extracted, specifically including:

[0036] Perform maximum principal component analysis on the point cloud to obtain the eigenvector corresponding to the largest eigenvalue, which is used as the first principal direction, i.e., the direction of the point cloud in the current layer.

[0037] As a preferred technical solution, maximum principal component analysis is performed on the point cloud to obtain the eigenvector corresponding to the largest eigenvalue, which is used as the first principal direction. Specifically, this includes:

[0038] Centralized processing of point clouds:

[0039] p i =(P i -μ)

[0040] Where μ represents the centroid of the point cloud, P i pi represents the original coordinates of the i-th point in the point cloud, and pi represents the coordinates of the i-th point after centering.

[0041] For a centralized point cloud, calculate the covariance matrix:

[0042]

[0043] Where N represents the number of points in the point cloud;

[0044] Calculate the eigenvalues ​​and eigenvectors of the covariance matrix:

[0045] Cv=λv

[0046] Where C represents the covariance matrix, v represents the eigenvector of the covariance matrix, and λ represents the eigenvalue of the covariance matrix;

[0047] The eigenvector corresponding to the largest eigenvalue is the first principal direction of the point cloud.

[0048] As a preferred technical solution, the geometric features of thin-walled structures are extracted using a random sampling statistical method, specifically including:

[0049] Set the number of samplings and distance thresholds. Within 1 / 5 to 4 / 5 of the height of the point cloud, randomly select two planes and calculate the distance from the point cloud to the two planes. Points with a distance less than the distance threshold are considered interior points of the plane. Use the least squares method to fit the interior points in the upper and lower half-planes to obtain the upper and lower boundary lines. Calculate the wall thickness and center of the thin wall in the current plane to obtain a set of wall thickness variation rate and center variation rate. Obtain multiple sets of wall thickness variation rate and center variation rate through multiple sampling experiments. Then, use histogram statistics to take the average value of the wall thickness variation rate and center variation rate in the maximum distribution interval as the final wall thickness variation rate and center variation rate.

[0050] As a preferred technical solution, the formulas for the wall thickness variation rate and the center variation rate are as follows:

[0051] Width1=mean(upperBoundary1-lowerBoundary1)

[0052] center1=mean((upperBoundary1+lowerBoundary1) / 2)

[0053] Width2=mean(upperBoundary2-lowerBoundary2))

[0054] center2=mean((upperBoundary2+lowerBoundary2) / 2)

[0055]

[0056] Wherein, upperBoundary1, upperboundary2, lowerBoundary1, and lowerBoundary2 represent the upper and lower boundary lines of the thin wall in random planes 1 and 2, respectively; Width1 and Width2 represent the average distance between the upper and lower boundary lines of the thin wall in random planes 1 and 2, respectively, serving as the thickness of the thin wall in the current plane; center1 and center2 represent the average value of the center lines of the thin wall in random planes 1 and 2, respectively, serving as the center position of the thin wall in the current plane; Width variationrate represents the rate of change of the thin wall thickness; center variation rate represents the rate of change of the center of the thin wall; and h1 and h2 represent the heights of random planes 1 and 2, respectively.

[0057] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0058] (1) Compared with existing methods for measuring the surface morphology of parts, such as profilometers, the present invention is based on a line laser scanning camera to measure the surface of parts. It has the advantages of large depth of field, wide measurement range, fast measurement speed, small size and easy integration of equipment, and can realize online or offline measurement.

[0059] (2) Compared with existing surface topography measurement and extraction algorithms based on line laser scanning cameras, this invention proposes an improved RANSAC algorithm to accelerate convergence and improve the algorithm’s noise resistance.

[0060] (3) In view of the fact that the pose of point cloud is not considered in the prior art, the present invention proposes a point cloud reference plane (printing plane) pose correction method and a point cloud printing direction pose correction method. This method can improve the accuracy and reliability of the surface morphology of the part.

[0061] (4) The present invention uses random sampling statistical method to extract the geometric features of the parts, which reduces the random error of the geometric feature values. Attached Figure Description

[0062] Figure 1 This is a schematic flowchart of the part surface morphology measurement method for laser directional energy process according to the present invention;

[0063] Figure 2 This is a schematic diagram of the original point cloud on the surface of the part of this invention;

[0064] Figure 3 This is a histogram diagram of the point cloud height values ​​of the present invention;

[0065] Figure 4 This is a schematic diagram of the improved RANSAC fitting process for the reference plane of the present invention;

[0066] Figure 5 This is a schematic diagram of the fitting result of the reference plane of the present invention;

[0067] Figure 6 This is a schematic diagram illustrating the result of removing background point clouds according to the present invention;

[0068] Figure 7 This is a schematic diagram of the point cloud in the XOY plane according to the present invention;

[0069] Figure 8 This is a schematic diagram of the point cloud noise reduction and pose correction process of the present invention;

[0070] Figure 9 This is a schematic diagram showing the result of removing noise point clouds according to the present invention;

[0071] Figure 10 This is a schematic diagram showing the comparison of pose correction before and after the reference plane of the present invention;

[0072] Figure 11 This is a schematic diagram comparing the pose of a single melt channel and a thin-walled point cloud before and after printing direction correction in this invention;

[0073] Figure 12 This is a flowchart of the random sampling statistical method for feature extraction in this invention. Detailed Implementation

[0074] 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 merely illustrative and not intended to limit the invention.

[0075] Example

[0076] like Figure 1As shown, this embodiment provides a method for measuring the surface morphology of a part during a laser-directed energy process, including the following steps:

[0077] S1: During or after processing, a line laser scanning camera is used to acquire the height information of the part surface, and this information is combined with the resolution and scanning speed of the line laser, such as... Figure 2 As shown, the point cloud on the surface of the part is reconstructed;

[0078] In this embodiment, the raw data acquired by the line laser scanning camera is a (3000*n) height matrix. In this embodiment, the line laser scanning camera has a laser line length of 30mm in the scanning plane, containing 3000 points. Therefore, the resolution in the vertical scanning direction is 10 micrometers. The sampling frequency of the line laser scanning camera is 2500Hz, and the scanning speed is 25mm / s; therefore, the resolution in the scanning direction is 10 micrometers. Taking (1,1) of the height matrix as the origin, the (x, y) coordinates of each point in the point cloud are obtained as follows:

[0079] x i =0.010*i;

[0080] y i =0.010*j;

[0081] S2: Perform histogram statistics on the point cloud height values ​​to extract the high-density distribution areas of the point cloud height values;

[0082] like Figure 3 As shown, histogram statistics are performed on the point cloud height values ​​to automatically divide the data intervals and obtain the number of data points in each data interval. The 75th percentile of the number of data points is taken as the threshold, and the data intervals with more data points than the threshold are extracted as the high-density distribution areas of the point cloud height values.

[0083] S3: As Figure 4 As shown, an improved RANSAC fitting is performed on the printed reference plane to obtain the optimal reference plane.

[0084] In this embodiment, three points are randomly selected in a high-density distribution area to solve the plane equation, determine the distance between the complete point cloud and the plane, and calculate the number of interior points, such as... Figure 5 As shown, the optimal reference plane equation is obtained through iteration;

[0085] In this embodiment, an improved RANSAC fitting is performed on the printed reference plane to obtain the optimal reference plane. First, the number of iterations and a distance threshold are set to ensure that the fitting converges and meets the fitting accuracy requirements within the specified number of iterations. Then, within a high-density distribution region of point cloud height values, three points are randomly selected to fit a plane. The plane equation can be expressed as:

[0086] ax + by + cz + d = 0

[0087] Where a, b, and c are the components of the plane normal vector, and d is the constant term of the plane equation;

[0088] Calculate the normal vector between the three points:

[0089] n1 = p2 - p1

[0090] n2 = p3 - p1

[0091] The cross product yields the normal vector: n = n1 × n2;

[0092] The distance from each point in the point cloud to the plane is calculated as follows:

[0093]

[0094] Points with a distance less than a pre-set distance threshold are designated as inliers. The number of inliers is counted; a higher number of inliers indicates a better fit. Through iterative iteration, the optimal fitting plane is obtained as the baseline plane. The pre-set number of iterations and distance threshold are determined through preliminary experiments. In the preliminary experiments, the number of iterations is determined by analyzing the relationship between the number of inliers and the number of iterations, ensuring the fit reaches convergence. Also in the preliminary experiments, the height distribution characteristics of the background point cloud are extracted by manually labeling it, and the distance threshold is determined using the 3σ principle.

[0095] S4: As Figure 6 As shown, the interior points of the reference plane are removed as background point clouds. The interior points of the reference plane are caused by the substrate surface. In this embodiment, the background point cloud is removed to avoid the background point cloud interfering with the extraction of the first principal direction of the target point cloud.

[0096] S5: As Figure 7 As shown, the point cloud is projected onto the XOY plane to generate a mask image;

[0097] S6: As Figure 8 As shown, the mask image preserves the largest connected components, removes the noise point cloud caused by spatter adhering to the substrate surface, and then applies it to the point cloud, as follows: Figure 9 As shown, the point cloud after noise reduction is obtained;

[0098] S7: Calculate the transformation matrix T1 from the reference plane to the XOY plane;

[0099] S8: The transformation matrix T1 is applied to the denoised point cloud to make the printing reference plane of the point cloud coincide with the XOY plane and correct the height value of the point cloud.

[0100] like Figure 10As shown in the figure, (a) is the front view of the point cloud without the reference plane being corrected. As can be seen from the red dashed line marked in the front view, the point cloud as a whole has a pitch angle in the front view, and the z coordinate of the point cloud cannot correctly reflect the height value of the point cloud. (b) is the front view of the point cloud after the reference plane is corrected. The bottom of the point cloud coincides with the horizontal line, indicating that the reference plane of the point cloud coincides with the XOY plane.

[0101] S9: Extract the first principal direction of the point cloud;

[0102] Due to installation errors in the line laser scanning camera, the laser line cannot be perfectly perpendicular to the printing direction. The scanned point cloud direction will have a certain angle deviating from the actual printing direction. In this embodiment, maximum principal component analysis (PCA) is performed on the point cloud to obtain the eigenvector corresponding to the largest eigenvalue, which is used as the first principal direction, i.e., the direction of the current layer of the point cloud. Specifically, this includes:

[0103] First, the point cloud is centrally processed: p i =(P i -μ), where μ represents the centroid of the point cloud, P i p represents the original coordinates of the i-th point in the point cloud. i This represents the coordinates of the i-th point after centralization.

[0104] For a centralized point cloud, calculate the covariance matrix:

[0105]

[0106] Where N represents the total number of points in the point cloud;

[0107] Eigenvalue decomposition: The eigenvalues ​​and eigenvectors of the covariance matrix are calculated using the following formula:

[0108] Cv=λv

[0109] Where C represents the covariance matrix, v represents the eigenvector of the covariance matrix, and λ represents the eigenvalue of the covariance matrix;

[0110] The eigenvector corresponding to the largest eigenvalue is the first principal direction of the point cloud;

[0111] S10: Calculate the transformation matrix T2 from the first principal direction to the printing direction and apply it to the point cloud so that the direction of the point cloud is consistent with the printing direction of the current deposition layer, and obtain the final pose-corrected point cloud as the surface morphology of the part.

[0112] like Figure 11As shown in the figure, (a) and (c) are side views of the point cloud without printing direction correction for the single-melt point cloud and the thin-walled point cloud, respectively. As can be seen from the figure, the direction of the point cloud is tilted at an angle to the printing direction. (b) and (d) are side views of the point cloud after printing direction correction for the single-melt point cloud and the thin-walled point cloud, respectively. In the side view, the boundary line of the point cloud is clear.

[0113] S11: As Figure 12 As shown, the geometric features of thin-walled structures are extracted using a random sampling statistical method.

[0114] In this embodiment, the number of samplings (preferably 1000) and the distance threshold (preferably 0.02) are first set. The larger the number of samplings, the better the noise resistance of the result, but the longer the calculation time. The smaller the distance threshold, the higher the accuracy, but the lower the noise resistance. Within 1 / 5 to 4 / 5 of the height of the point cloud, two planes (random planes 1 and 2) are randomly selected, and the distance from the point cloud to the two planes is calculated. Points with a distance less than the distance threshold are the interior points of the plane. The interior points in the upper and lower half-planes are fitted using the least squares method to obtain the upper and lower boundary lines. Then, the wall thickness and center of the thin wall in the current plane are calculated, and a set of wall thickness change rate and center change rate are obtained. Through multiple sampling experiments, multiple sets of wall thickness change rate and center change rate are obtained. Then, through histogram statistics, the average value of the wall thickness change rate and center change rate in the maximum distribution interval is taken as the final wall thickness change rate and center change rate.

[0115] The formulas for the wall thickness variation rate and the center variation rate are as follows:

[0116] Width1=mean(upperBoundary1-lowerBoundary1)

[0117] center1=mean((upperBoundary1+lowerBoundary1) / 2)

[0118] Width2=mean(upperBoundary2-lowerBoundary2)

[0119] center2=mean((upperBoundary2+lowerBoundary2) / 2)

[0120]

[0121] Wherein, upperBoundary1, upperboundary2, lowerBoundary1, and lowerBoundary2 represent the upper and lower boundary lines of the thin wall in random planes 1 and 2, respectively; Width1 and Width2 represent the average distance between the upper and lower boundary lines of the thin wall in random planes 1 and 2, respectively, serving as the thickness of the thin wall in the current plane; center1 and center2 represent the average value of the center lines of the thin wall in random planes 1 and 2, respectively, serving as the center position of the thin wall in the current plane; Width variationrate represents the rate of change of the thin wall thickness; center variation rate represents the rate of change of the center of the thin wall; and h1 and h2 represent the heights of random planes 1 and 2, respectively.

[0122] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method of part surface topography measurement for a laser directed energy process, characterized by, The method comprises the following steps: Based on the line laser scanning camera to collect the height information of the part surface, combined with the resolution and scanning speed of the line laser to reconstruct the point cloud of the part surface; Extract the high-density distribution area of the point cloud height value; The improved RANSAC fitting is performed on the printed reference plane to obtain the optimal reference plane, specifically including: Set the iteration number and distance threshold, randomly select 3 points to fit a plane in the high-density distribution area of the point cloud height value, calculate the normal vector between the three points, and cross multiply to obtain the normal vector, and calculate the distance of each point in the point cloud to the plane; The points with a distance less than the set distance threshold are recorded as inliers, and the optimal fitting plane is obtained as the reference plane through iterative loop; Set the iteration number and distance threshold, specifically including: The iteration number and distance threshold are set through pre-experiment, in the pre-experiment, the iteration number is determined through the relationship diagram of the number of inliers with the iteration number, so that the fitting reaches the convergence condition, the background point cloud is marked, the height value distribution characteristics of the background point cloud are extracted, and the distance threshold is determined through the 3σ principle; Randomly select 3 points to fit a plane, and the plane equation is represented as: ； Wherein, a, b, c are the components of the plane normal vector, and d is the constant term of the plane equation; The normal vector between the 3 points is calculated , , ​ ； ； Cross product gives normal vector: ; The distance of each point in the point cloud to the plane is calculated, which is represented as: ； wherein, represents the distance of each point in the point cloud to the plane; Remove the inliers of the reference plane as the background point cloud, and the inliers of the reference plane are obtained based on the substrate surface; Project the point cloud to the XOY plane to generate a mask image; The mask image retains the maximum connected domain, removes the noise point cloud, and obtains the denoised point cloud; Calculate the transformation matrix T1 of the reference plane to the XOY plane; The transformation matrix T1 is applied to the denoised point cloud, so that the printing reference plane of the point cloud coincides with the XOY plane, and the height value of the point cloud is corrected; Extract the first principal direction of the point cloud, specifically including: Perform maximum principal component analysis on the point cloud to obtain the eigenvector corresponding to the maximum eigenvalue as the first principal direction, i.e. the current layer point cloud direction; Calculate the transformation matrix T2 of the first principal direction to the printing direction, and apply it to the point cloud, so that the direction of the point cloud is consistent with the deposition direction of the current printing layer, and obtain the point cloud after final pose correction as the surface topography of the part; Extract the geometric features of the thin wall by random sampling statistical method.

2. The method of part surface topography measurement for a laser directed energy process of claim 1, wherein, Extract the high-density distribution area of the point cloud height value, specifically including: Perform histogram statistics on the point cloud height value, divide the data interval, obtain the number of data points in each data interval, and extract the data interval with the number of data points exceeding the threshold as the high-density distribution area of the point cloud height value.

3. The method of part surface topography measurement for a laser directed energy process of claim 1, wherein, Perform maximum principal component analysis on the point cloud to obtain the eigenvector corresponding to the maximum eigenvalue as the first principal direction, specifically including: Centralize the point cloud: ； wherein, denotes the centroid of the point cloud, denotes the original coordinates of the i-th point in the point cloud, denotes the coordinates of the i-th point after the centering process; For the centralized point cloud, calculate the covariance matrix: ； wherein, denotes the number of points of the point cloud; Calculate the eigenvalues and eigenvectors of the covariance matrix: ； Wherein, C represents the covariance matrix, v represents the eigenvector of the covariance matrix, and λ represents the eigenvalue of the covariance matrix; The eigenvector corresponding to the maximum eigenvalue is the first principal direction of the point cloud.

4. The method of part surface topography measurement for a laser directed energy process of claim 1, wherein, Extract the geometric features of the thin wall by random sampling statistical method, specifically including: Set the sampling times and distance threshold, in the 1 / 5~4 / 5 height of the point cloud, randomly select two planes, calculate the distance from the point cloud to the two planes, the distance less than the distance threshold is the inlier of the plane, fit the inliers in the upper and lower half planes by the least square method to obtain the upper and lower boundary straight lines, calculate the wall thickness and center of the thin wall in the current plane to obtain a group of wall thickness variation rate and center variation rate, through multiple sampling experiments to obtain multiple groups of wall thickness variation rate and center variation rate, and then through histogram statistics, take the average value of the wall thickness variation rate and the center variation rate in the maximum distribution interval as the final wall thickness variation rate and the center variation rate.

5. The method of claim 4, wherein, The wall thickness variation rate and the center variation rate formula are as follows: ； ； ； ； ； ； Wherein, upperBoundary1, upperBoundary2, lowerBoundary1, lowerBoundary2 respectively represent the upper and lower boundary straight lines of the thin wall in the random plane 1, 2, Width1, Width2 respectively represent the average distance between the upper and lower boundary straight lines of the thin wall in the random plane 1, 2, as the thickness of the thin wall in the current plane, center1, center2 respectively represent the average value of the thin wall center straight line in the random plane 1, 2, as the center position of the thin wall in the current plane, Width variation rate represents the variation rate of the wall thickness of the thin wall, center variation rate represents the variation rate of the center of the thin wall, h1, h2 respectively represent the height of the random plane 1, 2.

Images