Aero-engine blade defect identification and positioning method based on point cloud processing
By using point cloud processing technology, combined with SAC-IA and ICP algorithms, the non-overlapping regions of aero-engine blades are extracted and classified using the minimum bounding box, solving the accuracy problem of blade defect identification and localization in existing technologies and achieving high-precision defect identification and localization.
Patent Information
- Application Number
- CN202511034663.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-25
- Publication Date
- 2025-11-11
AI Technical Summary
Existing technologies cannot accurately quantify the crack depth and wear volume of aero-engine blades, and traditional methods produce blurry images in oily environments and cannot achieve high-precision defect identification and location.
A point cloud-based approach was adopted, in which blade point cloud data was acquired by 3D scanning. After preprocessing, coarse and fine registration was performed using the SAC-IA and ICP algorithms to extract non-overlapping regions. Defects were then classified and located by the aspect ratio and depth ratio of the minimum bounding box.
It achieves high-precision identification and location of blade defects, with a micro-defect detection rate of 98.5% and a defect location error of ≤±0.05mm. The defect identification accuracy has been improved to the sub-millimeter level, providing a reliable basis for repair.
Smart Images
Figure CN120931586A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of aero-engine blade defect identification, specifically relating to a method for identifying and locating aero-engine blade defects based on point cloud processing. Background Technology
[0002] As a core hot-end component, the structural integrity of aero-engine blades directly determines the engine's safety performance. Under high-speed rotation and extreme temperature conditions, blades are prone to defects such as cracks, wear, and deformation. Traditional coordinate measuring machines (CMMs) are limited by single-point sampling, industrial endoscopes produce blurred images in oily environments, and thermal expansion of materials caused by high temperatures introduces systematic biases into optical measurements. Existing methods such as eddy current testing and fluorescence penetrant testing can only achieve qualitative judgment of defects and cannot accurately quantify key maintenance parameters such as crack depth and wear volume. Summary of the Invention
[0003] The purpose of this invention is to overcome the shortcomings of the prior art and propose a method for identifying and locating defects in aero-engine blades based on point cloud processing. This method achieves defect identification and location through point cloud registration and non-overlapping region analysis, providing a reliable, accurate, and precise method for identifying and locating defects in aero-engine blades.
[0004] To achieve the above objectives, the present invention proposes the following technical solution: A method for defect identification and localization of aero-engine blades based on point cloud processing determines the type of defect based on the aspect ratio of the minimum bounding box and locates the defect based on the vertex coordinates of the minimum bounding box, including the following steps: S1) Use a 3D scanner to perform 3D scanning on the surface of normal and defective blades of the aero-engine, convert the blade surface data into discrete point cloud data, perform point cloud data preprocessing, and obtain preprocessed point cloud data of normal and defective blades. S2) The SAC-IA algorithm is used to perform coarse registration on the preprocessed point cloud data, and then the ICP algorithm is used for fine matching to extract the non-overlapping parts of the two point clouds, i.e. the defect parts of the blade. S3) The point cloud data of the defective parts of the blade are processed by establishing the minimum bounding box, the side length ratio of the minimum bounding box is calculated to classify the defects, the vertex coordinates of the minimum bounding box of the defect are calculated, and the defects are located.
[0005] Specifically, the point cloud processing-based aero-engine blade defect identification and localization method of the present invention places normal and defective blades of the aero-engine on the fixtures of a rotating worktable, so that the blades can rotate along the axial direction, and the three-dimensional scanner is fixed at a certain angle above the blades to scan.
[0006] More specifically, the scanning parameters of the 3D scanner are preferably set as follows: single-image measurement accuracy ≤ 0.02 mm, single-scan time ≤ 0.6 s, and rotation speed ≤ 4 rad / min.
[0007] Furthermore, the point cloud data preprocessing in step S1) specifically includes: S1-1) Perform pass-through filtering on the point cloud data of the normal and defective blades in the X, Y, and Z axes to obtain the point cloud data of the normal and defective blades after pass-through filtering. S1-2) The normal and defective blade point cloud data after the pass-through filtering are downsampled by the VoxelGrid filter to obtain the downsampled normal and defective blade point cloud data. S1-3) Radius filtering is performed on the downsampled point cloud data of normal and defective blades to obtain preprocessed point cloud data of normal and defective blades.
[0008] Further, in step S2), extracting the non-overlapping portions of the two point clouds, i.e., extracting the non-overlapping portions of the normal and defective point clouds based on the finely matched leaf point cloud data, includes the following sub-steps: S2-1) Set each discrete point in the normal leaf point cloud as a point set. Each point Let be the coordinates of each discrete point in space. Similarly, let the point cloud of the defective blade be denoted as . ; S2-2) For finely matched point cloud data, the point set of a normal point cloud Point set of point cloud with defective leaf Each point in the diagram corresponds one-to-one, through calculation. , The Euclidean distance between points is used to extract points in non-overlapping regions; (S2-3) Repeat the above steps until all non-overlapping points in the two point sets are extracted, thus obtaining the non-overlapping parts of the normal and defective point clouds.
[0009] Furthermore, in step S2-2), the formula for calculating the Euclidean distance is: .
[0010] Further, in step S3), the point cloud data of the defective parts of the blade is processed to establish a minimum bounding box, the side length ratio of the minimum bounding box is calculated to classify the defects, and the vertex coordinates of the minimum bounding box of the defect are calculated to locate the defect. This includes the following sub-steps: S3-1) Calculate the point cloud mass C and construct the covariance matrix. Calculate the eigenvalues of the covariance matrix. With feature vectors ; S3-2) Construct a minimum bounding box coordinate system and transform the point cloud coordinates to the minimum bounding box coordinate system; S3-3) Calculate the extent of the minimum bounding box on each axis to determine the dimensions of each side of the minimum bounding box. ; S3-4) Take the depth direction of the defect on the blade surface as the height h, and the other two sides as the length l and width w. Given the blade thickness H, when h / H≥1, it is a through defect. When h / H<1, compare the ratio of length to width k. Given a threshold z, when k>z, it is a scratch defect. When k<z, it is a pit defect. S3-5) Calculate the eight vertices of the minimum bounding box as the coordinates of the defect location to locate the defect point cloud.
[0011] Further, in step S3-1), the point cloud mass C is: covariance matrix for: eigenvalues for eigenvectors for: .
[0012] Furthermore, in step S3-2), the point cloud centroid is used as the coordinate system of the minimum bounding box. With the origin, As the X, Y, and Z axis basis vectors of the coordinate system, the minimum bounding box coordinate system is: .
[0013] Furthermore, in step S3-3), the dimensions of each side... for , , ,in These are the coordinates in the minimum bounding box coordinate system.
[0014] Further, in steps S3-5), the eight vertices of the minimum bounding box in the minimum bounding box coordinate system are used as the defect location coordinates, specifically: ; ; ; ; ; ; ; .
[0015] Furthermore, in steps S3-5), the eight vertices of the minimum bounding box in the minimum bounding box coordinate system, used as the defect location coordinates, can be converted into the coordinates of the eight vertices in the original coordinate system using a transformation formula. The transformation formula is as follows: ;
[0016] The coordinates of the eight vertices in the original coordinate system are: ; ; ; ; ; ; ; .
[0017] The beneficial effects of this invention are as follows: This invention effectively removes noise and improves data quality by preprocessing the original point cloud data; it makes the data more reliable and improves the accuracy of the model by removing outliers; it reduces the number of point clouds by downsampling while preserving the main information of the data as much as possible, which not only reduces the computation time and storage requirements, but also makes subsequent point cloud registration more efficient; and it uses Gaussian filtering to smooth out noisy points and minimize the impact of noise, thereby obtaining more realistic and smooth point cloud data.
[0018] This invention achieves a micro-defect (≥0.1mm) detection rate of 98.5% through point cloud registration (SAC-IA coarse registration + ICP fine registration) and non-overlapping region extraction; by accurately separating non-overlapping point clouds based on Euclidean distance in three-dimensional space, it eliminates the interference of blade assembly error, and the defect positioning error is ≤±0.05mm, significantly improving the defect identification accuracy to sub-millimeter level detection capability.
[0019] This invention achieves an accuracy rate of 99.2% for through-hole defects (h > H) and classifies scratch defects (k > z) and pit defects (k < z) by using the minimum bounding box aspect ratio k = l / w and depth ratio h / H. The adaptive threshold mechanism reduces the false positive rate, thus doubling the efficiency of intelligent defect classification and realizing the quantification of bounding box geometric features.
[0020] This invention preprocesses point cloud data of normal and defective blades by scanning them, effectively improving the accuracy and efficiency of obtaining the original point cloud data, reducing the influence of noise points, and obtaining more realistic data. By registering the preprocessed normal and defective point cloud data, the non-overlapping portions of the two point clouds are extracted to obtain point cloud data of the defective parts of the blade, removing the influence of the main body point cloud and making subsequent calculations of the defect size more accurate. Finally, by establishing minimum bounding boxes for the non-overlapping point cloud data and calculating the side length ratio of the minimum bounding boxes to classify defects, and calculating the vertex coordinates of the defect minimum bounding boxes, the defects are located.
[0021] Using the above scheme, the present invention can reliably, accurately, and computationally precisely identify and locate defects in aero-engine blades, significantly improving the blade defect identification accuracy to sub-millimeter level, thus providing a reliable basis for blade repair. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 This is a flowchart illustrating the method for identifying and locating defects in aero-engine blades based on point cloud processing, as described in an embodiment of the present invention.
[0024] Figure 2 This is a schematic diagram of the preprocessing point cloud in an embodiment of the present invention.
[0025] Figure 3 This is a schematic diagram of the defective portion extracted in an embodiment of the present invention.
[0026] Figure 4 This is a schematic diagram of defect classification in an embodiment of the present invention.
[0027] Figure 5 This is a schematic diagram of defect location in an embodiment of the present invention. Detailed Implementation
[0028] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present 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 of the present invention and are not intended to limit the present invention.
[0029] It should be noted that when a component is referred to as being "fixed to" or "set on" another component, it can be directly on or indirectly on that other component. When a component is referred to as being "connected to" another component, it can be directly connected to or indirectly connected to that other component.
[0030] It should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.
[0031] 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 one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0032] The flowchart of the point cloud-based aero-engine blade defect identification and localization method in this embodiment is as follows: Figure 1 As shown, it includes the following steps: 1. Use a 3D scanner to perform 3D scanning of the surfaces of normal and defective blades of an aero-engine, and convert the blade surface data into discrete point cloud data. Point cloud data was scanned using a 3D scanner, which was a blue light 3D scanner with binocular cameras on both sides and a grating sensor in the middle. The near-end field of view was 145×120~340mm, the measurement point distance was 0.11mm, the single-frame measurement accuracy was 0.02mm, and the single scan time was 0.6s. Turn on the rotary table, set the rotation speed to 4 rad / min, and turn on the blue light 3D scanner to scan the aero-engine blades for 4 minutes to obtain the 3D point cloud data of the blades; point cloud data of 1 normal blade and 3 defective blades were collected respectively.
[0033] 2. Perform point cloud data preprocessing to obtain preprocessed point cloud data for normal and defective blades. according to Figure 2 The flowchart for obtaining preprocessed point cloud data includes the following steps: The point cloud data of the normal and torsion blades are subjected to pass-through filtering in the X, Y, and Z axes to remove outliers, resulting in pass-through filtered point cloud data of the normal and defective blades. The point cloud data of normal and torsional blades after direct filtering were downsampled using a VoxelGrid filter. The voxel grid size was set to 0.7 mm to obtain the downsampled point cloud data of normal and defective blades. Radius filtering is performed on the downsampled point cloud data of normal and defective blades to obtain the preprocessed point cloud data of the normal and defective blades.
[0034] 3. Register the preprocessed normal and defective point cloud data to extract the defective parts of the blades. The preprocessed point cloud data of the normal and torsional blades were coarsely matched according to the SAC-IA algorithm. The minimum sample distance was set to 0.3, the maximum corresponding distance to 16, the maximum number of iterations to 100, the number of samples used in each iteration to 4, and the randomness of feature correspondence to 10, so as to obtain coarsely matched blade point cloud data. The coarsely matched blade point cloud data is finely matched using the ICP algorithm. The maximum number of iterations is set to 100, the maximum corresponding point distance is 3.0, the transformation matrix convergence threshold is 10⁻⁸, the mean square error convergence threshold is 10⁻⁶, and the RANSAC outlier threshold is 1.2 to obtain finely matched blade point cloud data.
[0035] Based on the finely matched blade point cloud data, the Euclidean distance between corresponding points in the two point clouds after fine matching is calculated. Point sets whose Euclidean distance exceeds a preset distance threshold are selected as defect parts, and defect parts of three point clouds are extracted respectively: 1) Set each discrete point in the point cloud of a normal leaf as a point set. Each point Let be the coordinates of each discrete point in space. Similarly, let the point cloud of the defective blade be denoted as . ; 2) For finely matched point cloud data, the point set of a normal point cloud Point set of point cloud with defective leaf Each point in the equation corresponds one-to-one. The Euclidean threshold is set to 0.1, and calculations are performed... , The Euclidean distance between two points is used to extract points in non-overlapping regions; 3) Repeat the above steps until all non-overlapping points in the two point sets are extracted, thus obtaining the defect locations of the three blades respectively, such as... Figure 3 As shown.
[0036] Fourth, construct the minimum bounding box from the point cloud data of the defect location, calculate the side length ratio to classify the defect, and calculate the vertex coordinates to locate the defect. exist Figure 3 Based on this, a minimum bounding box processing method is established. The ratio of the side lengths is calculated to classify defects, and the vertex coordinates are calculated to locate the defects. This includes: 3-1) Calculate the covariance matrix T for the three defect point cloud particles C respectively, and calculate the eigenvalues of the covariance matrix. With feature vectors .
[0037] 3-2) Construct its minimum bounding box coordinate system and transform the point cloud coordinates to the minimum bounding box coordinate system.
[0038] 3-3) Calculate the range of its minimum bounding box on each axis to determine the dimensions of each side of the minimum bounding box. In this embodiment, from left to right, the dimensions of each side of the minimum bounding box of the first defect are: The dimensions of the minimum bounding box of the second defect are: The minimum bounding box dimensions of the third defect are... .
[0039] 3-4) Taking the depth of the defect on the blade surface as the height h, and the other two sides as the length l and width w, given the blade thickness H, when h / H≥1, it is a through defect. When h / H<1, compare the length to width ratio k. Given a threshold z, when k>z, it is a scratch defect, and when k<z, it is a pit defect. Since the length-to-width ratio k of scratch defects is mainly distributed in the range of k > 4, while the length-to-width ratio k of pit defects is mainly distributed in the range of k < 3, the given threshold is 3≤z≤4.
[0040] Defect classification as follows Figure 4 As shown, the blade thickness H at the first defect location is 4.3 mm, and the measured height h of the through-thickness defect is 4.3061. Since h / H≥1, the first defect is a through-thickness defect. The heights h of the second and third defects are 2.2179 and 2.6800, respectively, both less than the blade thickness H, and therefore not through-thickness defects. Further analysis of the aspect ratio is needed. Given a threshold z of 3.5, the aspect ratio k of the second defect is 4.7896, indicating that k>z, and the second defect is a scratch defect. The aspect ratio k of the third defect is 2.1493, indicating that k<z, and the third defect is a pit defect.
[0041] 3-5) Calculate the eight vertices of the minimum bounding box as the defect location coordinates to achieve defect point cloud localization. Defect localization is as follows: Figure 5 As shown.
[0042] The coordinates of the eight vertices of the first defect, i.e., the through defect, are: ; ; ; ; ; ; ; .
[0043] The coordinates of the eight vertices of the second defect, namely the scratch defect, are: ; ; ; ; ; ; ; .
[0044] The coordinates of the eight vertices of the third defect, namely the pit defect, are: ; ; ; ; ; ; ; .
[0045] As can be seen from this embodiment, the present invention achieves high-precision identification and positioning of defects in aero-engine blades through point cloud registration and non-overlapping region analysis, providing a reliable basis for blade repair.
[0046] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for defect identification and localization of aero-engine blades based on point cloud processing, characterized in that: This is used to determine the type of defect based on the aspect ratio of the minimum bounding box, and to locate the defect based on the vertex coordinates of the minimum bounding box, including: S1) Use a 3D scanner to perform 3D scanning on the surface of normal and defective blades of the aero-engine, convert the blade surface data into discrete point cloud data, perform point cloud data preprocessing, and obtain preprocessed point cloud data of normal and defective blades. S2) The SAC-IA algorithm is used to perform coarse registration on the preprocessed point cloud data, and then the ICP algorithm is used for fine matching to extract the non-overlapping parts of the two point clouds, i.e. the defect parts of the blade. S3) The point cloud data of the defective parts of the blade are processed by establishing the minimum bounding box, the side length ratio of the minimum bounding box is calculated to classify the defects, the vertex coordinates of the minimum bounding box of the defect are calculated, and the defects are located.
2. The method for defect identification and localization of aero-engine blades based on point cloud processing according to claim 1, characterized in that: Normal and defective blades of an aero-engine are placed on the fixtures of a rotary table, allowing the blades to rotate axially. The 3D scanner is fixed at a certain angle above the blades to perform scanning.
3. The method for defect identification and localization of aero-engine blades based on point cloud processing according to claim 1, characterized in that: The scanning parameters of the 3D scanner are set as follows: single-image measurement accuracy ≤ 0.02 mm, single-scan time ≤ 0.6 s, and rotation speed ≤ 4 rad / min.
4. The method for defect identification and localization of aero-engine blades based on point cloud processing according to claim 1, characterized in that: The point cloud data preprocessing includes: S1-1) Perform pass-through filtering on the point cloud data of the normal and defective blades in the X, Y, and Z axes to obtain the point cloud data of the normal and defective blades after pass-through filtering. S1-2) The normal and defective blade point cloud data after the pass-through filtering are downsampled by the VoxelGrid filter to obtain the downsampled normal and defective blade point cloud data. S1-3) Radius filtering is performed on the downsampled point cloud data of normal and defective blades to obtain preprocessed point cloud data of normal and defective blades.
5. The method for defect identification and localization of aero-engine blades based on point cloud processing according to claim 1, characterized in that: The extraction of the non-overlapping portions of the two point clouds involves extracting the non-overlapping portions of the normal and defective point clouds based on the finely matched leaf point cloud data, including: S2-1) Set each discrete point in the normal leaf point cloud as a point set. Each point Let be the coordinates of each discrete point in space. Similarly, let the point cloud of the defective blade be denoted as . ; S2-2) For finely matched point cloud data, the point set of a normal point cloud Point set of point cloud with defective leaf Each point in the diagram corresponds one-to-one, through calculation. , The Euclidean distance between two points is used to extract points in non-overlapping regions; (S2-3) Repeat the above steps until all non-overlapping points in the two point sets are extracted, thus obtaining the non-overlapping parts of the normal and defective point clouds.
6. The method for defect identification and localization of aero-engine blades based on point cloud processing according to claim 1, characterized in that: The point cloud data of the defective areas of the blade are processed to establish a minimum bounding box. The ratio of the side length of the minimum bounding box is calculated to classify the defects. The vertex coordinates of the minimum bounding box of the defect are calculated to locate the defect. This includes: S3-1) Calculate the point cloud mass C, construct the covariance matrix T, and calculate the eigenvalues of the covariance matrix. With feature vectors ; S3-2) Construct a minimum bounding box coordinate system and transform the point cloud coordinates to the minimum bounding box coordinate system; S3-3) Calculate the extent of the minimum bounding box on each axis to determine the dimensions of each side of the minimum bounding box. ; S3-4) Take the depth direction of the defect on the blade surface as the height h, and the other two sides as the length l and width w. Given the blade thickness H, when h / H≥1, it is a through defect. When h / H<1, compare the ratio of length to width k. Given a threshold z, when k>z, it is a scratch defect. When k<z, it is a pit defect. S3-5) Calculate the eight vertices of the minimum bounding box as the coordinates of the defect location to locate the defect point cloud.
7. The method for defect identification and localization of aero-engine blades based on point cloud processing according to claim 5, characterized in that: The formula for calculating Euclidean distance is: 。 8. The method for defect identification and localization of aero-engine blades based on point cloud processing according to claim 6, characterized in that: The point cloud mass C is: covariance matrix for: eigenvalues for eigenvectors for: In the minimum bounding box coordinate system, with the centroid of the point cloud as the reference point... With the origin, As the X, Y, and Z axis basis vectors of the coordinate system, the minimum bounding box coordinate system is: Dimensions of each side for , , ,in These are the coordinates in the minimum bounding box coordinate system.
9. The method for defect identification and localization of aero-engine blades based on point cloud processing according to claim 6, characterized in that: The eight vertices of the minimum bounding box in the minimum bounding box coordinate system are used as the coordinates of the defect location, specifically: ； ； ； ； ； ； ； 。 10. The method for identifying and locating defects in aero-engine blades based on point cloud processing according to claim 6, characterized in that: The eight vertices of the minimum bounding box in the minimum bounding box coordinate system are used as the coordinates of the defect location. These coordinates can be converted to the coordinates of the eight vertices in the original coordinate system using a transformation formula: The coordinates of the eight vertices in the original coordinate system are: ； ； ； ； ； ； ； 。