Track part typical defect identification and classification method based on point cloud features

Abstract

The invention belongs to the technical field of defect detection, and discloses a point cloud feature-based track part typical defect identification and classification method, which comprises the following steps of: obtaining surface point clouds of track parts such as a steel rail and a turnout through a three-dimensional profile measurement method; registering the obtained point cloud with a standard point cloud on the surface of the track part; identifying defects on the surface of the part by analyzing differences of features such as spatial positions, normal vectors and curvatures of all points between the registered point cloud and the standard point cloud, and obtaining the point cloud of a defect area; and classifying the defects based on a partial binary tree method, the point cloud shape of the defect area and the curvature of each point, and dividing the defects into three types of cracks, spalling and chipping and depressions. According to the method, accurate identification and classification of cracks, peeling, chipping and sunken defects of the track parts can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to a method for identifying and classifying typical defects of rail components based on point cloud features. Background Art

[0002] The surfaces of rail components such as rails and switches are prone to defects such as cracks, peeling, and dents, which have a serious impact on the service performance of rail components. Efficient identification and classification of surface defects of rail components is of great significance for repairing defects and improving the service performance of rail components. At present, conventional defect recognition methods mainly determine whether there are defects by the spatial position difference between the surface point cloud of the rail component and the surface standard point cloud after registration. This method has very high requirements for the registration accuracy of the point cloud, and is prone to omissions and misjudgments, which affect the recognition efficiency and accuracy. For the classification of defects, machine learning methods such as support vector machines and BP neural networks are often used. These methods have disadvantages such as slow convergence speed and poor model versatility, and it is difficult to meet the requirements of high efficiency and high applicability of defect classification. Therefore, it is urgent to propose a method that can accurately identify defects on the surface of rail components and classify the identified defects efficiently and accurately. Summary of the Invention

[0003] To address the above-mentioned issues, the present invention proposes a method for identifying and classifying typical defects of rail components based on point cloud features. Surface point clouds of rail components such as rails and switches are acquired through a three-dimensional profiling method and registered with a standard point cloud. Surface defects of the components are identified based on the characteristic differences between the two point clouds, resulting in a point cloud of the defective area. Based on the point cloud features of the defective area, the defects are classified into three types: cracks, spalling, and dents. This method achieves high-precision identification and classification of typical defects of rail components. The method comprises the following steps:

[0004] S1, obtains surface point clouds of rails, switches and other track components through 3D profile measurement method;

[0005] S2, registering the acquired point cloud with the standard point cloud of the rail component surface;

[0006] S3, by analyzing the differences in spatial position, normal vector, curvature and other features of each point between the registered point cloud and the standard point cloud, the defects on the component surface are identified and the point cloud of the defective area is obtained;

[0007] S4, based on the partial binary tree method and the point cloud shape of the defect area and the curvature of each point, the defects are classified into three types: cracks, spalling and depressions.

[0008] Furthermore, in S1, the surfaces of rail components such as rails and switches are measured by a high-precision three-dimensional profile measuring instrument to obtain surface point clouds.

[0009] Furthermore, in S2, the acquired point cloud is registered with the standard point cloud of the rail component surface through an iterative closest point algorithm, so that the poses of the two point clouds remain consistent, and the correspondence between each point in the two point clouds is obtained.

[0010] Furthermore, the detailed steps of S3 are:

[0011] S31, traverse each point in the registered point cloud and calculate the Euclidean distance between each point and its corresponding point in the standard point cloud. If the Euclidean distance between a point and its corresponding point is greater than a set threshold, it is determined that there is a defect on the surface of the rail component. Then continue to step S32. If there is no point with a Euclidean distance between its corresponding point and the point greater than the set threshold, it is determined that there is no defect on the surface of the rail component, and the defect identification process is completed.

[0012] S32, traverse each point in the registered point cloud, first calculate the product of the modulus of the difference between the Euclidean distance between each point and its corresponding point in the standard point cloud and the normal vector, for the i-th point in the registered point cloud, the product f i The calculation formula is:

[0013]

[0014] Among them, d i is the Euclidean distance between the i-th point in the registered point cloud and its corresponding point in the standard point cloud, is the normal vector of point i in the registered point cloud, is the normal vector of the corresponding point of the i-th point in the registered point cloud in the standard point cloud; if the f of the i-th point i If it is greater than the set threshold, the point is put into the point set M;

[0015] Then calculate the product of the Euclidean distance between each point and its corresponding point in the standard point cloud and the absolute value of the curvature difference. For the j-th point in the registered point cloud, the product h j The calculation formula is:

[0016] h j =d j ·|k j -K j |

[0017] Among them, d j is the Euclidean distance between the jth point in the registered point cloud and its corresponding point in the standard point cloud, k j is the curvature of the jth point in the registered point cloud, K jis the curvature of the corresponding point of the j-th point in the registered point cloud in the standard point cloud; if the h of the j-th point j If it is greater than the set threshold, the point is put into point set A;

[0018] Calculate the Euclidean distance between each point in the point set M and each point in the point set A, and select a point A in the point set A i , if there is a point M in the point set M whose Euclidean distance is less than the set threshold i , then mark A i For similar points, traverse each point in point set A and get the number a1 of similar points in point set A; select a point M in point set M j , if there is a point A in point set A whose Euclidean distance is less than the set threshold j , then mark M j For similar points, traverse each point in the point set M and obtain the number m1 of similar points in the point set M; calculate the similarity value L between the point set M and the point set A based on the number of similar points in the point set M:

[0019]

[0020] Wherein, a is the number of points in point set A, and m is the number of points in point set M. If the similarity value is greater than the set threshold, it indicates that the similarity between the two point sets is sufficiently high, and the point set M with the larger number of points than the point set A is used as the defect recognition result, and the point cloud of the defect area is obtained, completing the defect recognition process. If the similarity value is less than or equal to the set threshold, it indicates that the similarity between the two point sets is insufficient, and an accurate defect recognition result cannot be obtained, and the process proceeds to step S33.

[0021] S33, traverse each point in the registered point cloud, calculate the Euclidean distance between each point and its corresponding point in the standard point cloud, the product of the modulus of the difference between the normal vector and the absolute value of the difference in curvature, and for the kth point in the registered point cloud, the product g k The calculation formula is:

[0022]

[0023] Among them, d k is the Euclidean distance between the kth point in the registered point cloud and its corresponding point in the standard point cloud, is the normal vector of the kth point in the registered point cloud, is the normal vector of the corresponding point of the kth point in the registered point cloud in the standard point cloud, k k is the curvature of the kth point in the registered point cloud, K k is the curvature of the corresponding point of the k-th point in the registered point cloud in the standard point cloud; if the g of the k-th point k If the value is greater than the set threshold, the point is determined to belong to the defect area, and all points belonging to the defect area are screened out to obtain the point cloud of the defect area, completing the defect identification process.

[0024] Furthermore, the detailed steps of S4 are:

[0025] S41, in the first layer of the partial binary tree structure, projecting the point cloud of the defect area onto a horizontal plane tangent to the surface of the rail component and converting it into a two-dimensional point cloud, extracting the outer contour of the two-dimensional point cloud using a convex hull algorithm and drawing the minimum circumscribed rectangle of the outer contour, calculating the ratio of the length to the width of the minimum circumscribed rectangle, and classifying the defect as a crack if the ratio is greater than a set threshold; and classifying the defect as a pit if the ratio is less than or equal to the set threshold;

[0026] S42, in the second layer of the partial binary tree structure, calculate the proportion of points whose curvature is less than the set threshold in the point cloud of the defect area divided into pits. If the value is greater than the set threshold, the pits are further divided into peeling blocks. If the value is less than or equal to the set threshold, the pits are further divided into depressions.

[0027] Compared with the prior art, the above technical solution conceived by the present invention has the following beneficial effects:

[0028] 1. This invention proposes a method for identifying and classifying typical defects of rail components based on point cloud features. This method targets typical defects of rail components, including cracks, spalling, and dents, and identifies defects by comprehensively analyzing the differences in the spatial position, normal vector, and curvature of each point between the registered point cloud and the standard point cloud. This method achieves high accuracy while ensuring efficiency.

[0029] 2. The present invention proposes a method for identifying and classifying typical defects of rail components based on point cloud features. It classifies cracks, spalling and dents based on the partial binary tree method, simplifies complex classification problems through hierarchical classification, and improves the efficiency of defect classification.

[0030] 3. The present invention proposes a method for identifying and classifying typical defects of rail components based on point cloud features. According to the shape characteristics of cracks, spalling and depression defects, the minimum circumscribed rectangle of the outer contour of the two-dimensional point cloud of the defect area is used in a partial binary tree structure to divide the defects into cracks or pits. The curvature of each point in the point cloud of the defect area divided into pits is then used to further divide the pits into spalling or depressions, with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of a method for identifying and classifying typical defects of rail components based on point cloud features of the present invention;

[0032] Figure 2 Schematic diagram of the process of measuring rail tread with a 3D profile measuring instrument;

[0033] Figure 3 This is a schematic diagram of the rail tread point cloud;

[0034] Figure 4 This is a point cloud diagram of the defect area;

[0035] Figure 5 It is a schematic diagram of the outer contour of the two-dimensional point cloud;

[0036] Figure 6 This is a schematic diagram of the first-level classification results of the partial binary tree structure;

[0037] Figure 7 This is a schematic diagram of the second-level classification results of the partial binary tree structure;

[0038] In the figure: 1-3D profile measuring instrument, 2-rail, 3-rail tread, 4-crack, 5-peeling, 6-depression, 7-point cloud of defect area 1, 8-point cloud of defect area 2, 9-point cloud of defect area 3, 10-outer contour of 2D point cloud 1, 11-outer contour of 2D point cloud 2, 12-outer contour of 2D point cloud 3, 13-pit. DETAILED DESCRIPTION

[0039] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments. The exemplary embodiments of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0040] like Figure 1 As shown, the present invention proposes a method for identifying and classifying typical defects of rail components based on point cloud features, comprising the following steps:

[0041] S1, obtaining surface point clouds of rail components such as rail 2 and turnouts through a three-dimensional profile measurement method;

[0042] S2, registering the acquired point cloud with the standard point cloud of the rail component surface;

[0043] S3, by analyzing the differences in spatial position, normal vector, curvature and other features of each point between the registered point cloud and the standard point cloud, the defects on the component surface are identified and the point cloud of the defective area is obtained;

[0044] S4, based on the partial binary tree method and the point cloud shape of the defect area and the curvature of each point, the defects are classified into three types: crack 4, spalling 5 and depression 6.

[0045] In the following embodiments, a rail 2 is selected as a track component, and the specific implementation of the method is described by taking cracks 4, peeling blocks 5 and depressions 6 on the rail tread 3 as examples.

[0046] The rail tread 3 of the rail 2 is measured using a three-dimensional profile measuring instrument 1, such as Figure 2 As shown, obtain the rail tread 3 point cloud, such as Figure 3 shown.

[0047] The obtained rail tread 3 point cloud is aligned with the standard point cloud of rail tread 3 by using the iterative closest point algorithm. The point with the smallest Euclidean distance to each point in the rail tread 3 point cloud is selected in the standard point cloud as the corresponding point to form a corresponding relationship. The point pairs with corresponding relationships are combined into a corresponding point set. The rotation matrix and translation matrix are calculated based on the corresponding point set. The error function value between the corresponding point sets is calculated using the rotation matrix and translation matrix, and it is iterated until the error function value is lower than the set threshold or the maximum number of iterations is reached. The rotation matrix R and translation matrix T at the end of the iteration are used to transform the coordinate system of each point in the rail tread 3 point cloud. The calculation formula is:

[0048] Q n =RP n +T

[0049] Among them, P n is the coordinate of the nth point in the 3-point cloud of the rail tread, Q n is the coordinate of the nth point in the rail tread 3-point cloud after coordinate system transformation; by performing coordinate system transformation on each point in the rail tread 3-point cloud, the registered point cloud is obtained.

[0050] Traverse each point in the registered point cloud and calculate the Euclidean distance between each point and its corresponding point in the standard point cloud. If there is a point in the point cloud whose Euclidean distance to its corresponding point is greater than the set threshold, it is determined that there is a defect in the rail tread 3 and proceed to the next step.

[0051] The points in the registered point cloud are traversed twice. The first traversal calculates the product of the modulus of the difference between the Euclidean distance and the normal vector between each point and its corresponding point in the standard point cloud. The product f of the i-th point is i The calculation formula is:

[0052]

[0053] Among them, d i is the Euclidean distance between the i-th point in the registered point cloud and its corresponding point in the standard point cloud, is the normal vector of point i in the registered point cloud, is the normal vector of the corresponding point of the i-th point in the registered point cloud in the standard point cloud; if the f of the i-th point i If the value is greater than the set threshold, the point is put into the point set M.

[0054] The second traversal calculates the product of the absolute value of the Euclidean distance between each point and its corresponding point in the standard point cloud and the difference in curvature. The product h of the jth pointj The calculation formula is:

[0055] h j =d j ·|k j -K j |

[0056] Among them, d j is the Euclidean distance between the jth point in the registered point cloud and its corresponding point in the standard point cloud, k j is the curvature of the jth point in the registered point cloud, K j is the curvature of the corresponding point of the j-th point in the registered point cloud in the standard point cloud; if the h of the j-th point j If the value is greater than the set threshold, the point is put into point set A.

[0057] Calculate the Euclidean distance between each point in the point set M and each point in the point set A; select a point A in the point set A i , if there is a point M in the point set M whose Euclidean distance is less than the set threshold i , then mark A i For similar points, traverse each point in point set A and get the number a1 of similar points in point set A; select a point M in point set M j , if there is a point A in point set A whose Euclidean distance is less than the set threshold j , then mark M j For similar points, traverse each point in the point set M and obtain the number m1 of similar points in the point set M.

[0058] The calculation formula of the similarity value L between point set M and point set A is:

[0059]

[0060] Where a is the number of points in point set A, and m is the number of points in point set M.

[0061] The similarity value between point set M and point set A is less than the set threshold, indicating that the similarity between the two point sets is insufficient and accurate defect recognition results cannot be obtained. Proceed to the next step.

[0062] Traverse each point in the registered point cloud and calculate the Euclidean distance between each point and its corresponding point in the standard point cloud, the product of the modulus of the difference between the normal vector and the absolute value of the curvature difference, and the product g of the kth point k The calculation formula is:

[0063]

[0064] Among them, d k is the Euclidean distance between the kth point in the registered point cloud and its corresponding point in the standard point cloud, is the normal vector of the kth point in the registered point cloud, is the normal vector of the corresponding point of the kth point in the registered point cloud in the standard point cloud, k k is the curvature of the kth point in the registered point cloud, K k is the curvature of the corresponding point of the k-th point in the registered point cloud in the standard point cloud; if the g of the k-th point k If the value is greater than the set threshold, the point is determined to belong to the defective area of the rail tread 3, and all points belonging to the defective area are screened out to obtain point cloud 7 of defective area 1, point cloud 8 of defective area 2, and point cloud 9 of defective area 3, as shown in the figure. Figure 4 As shown, the defect identification process is completed.

[0065] The point cloud 7 of defect area 1, the point cloud 8 of defect area 2, and the point cloud 9 of defect area 3 are projected onto a horizontal plane tangent to the rail tread 3 and converted into a two-dimensional point cloud; the outer contour of the two-dimensional point cloud is extracted by the convex hull algorithm, that is, any two points in the two-dimensional point cloud are selected to connect into a straight line, and the remaining points in the two-dimensional point cloud are traversed. If all the remaining points are on one side of the straight line, the straight line is determined to be a boundary line, and all points in the two-dimensional point cloud are traversed, all boundary lines are screened out and combined into the outer contour of the two-dimensional point cloud, and the outer contour 10 of the two-dimensional point cloud 1, the outer contour 11 of the two-dimensional point cloud 2, and the outer contour 12 of the two-dimensional point cloud 3 are obtained, as shown in FIG. Figure 5 As shown; draw the minimum circumscribed rectangle of the outer contour of the two-dimensional point cloud, calculate the ratio of the length to the width of the minimum circumscribed rectangle, the calculated value of the outer contour 10 of the two-dimensional point cloud one is greater than the set threshold, and the defect corresponding to the outer contour is divided into crack 4, the calculated values of the outer contour 11 of the two-dimensional point cloud two and the outer contour 12 of the two-dimensional point cloud three are both less than the set threshold, and the defects corresponding to the two are divided into pits 13, completing the first-level classification of the partial binary tree structure, as shown Figure 6 shown.

[0066] Calculate the proportion of points whose curvature is less than the set threshold in the point cloud 8 of the defect area 2 and the point cloud 9 of the defect area 3. The calculated value of the point cloud 8 of the defect area 2 is greater than the set threshold, and it is divided into a peeling block 5. The calculated value of the point cloud 9 of the defect area 3 is less than the set threshold, and it is divided into a concave 6, completing the second-level classification of the partial binary tree structure, as shown in FIG. Figure 7 shown.

[0067] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying and classifying typical defects of rail components based on point cloud features. Surface point clouds of rail components such as rails and switches are acquired through 3D profiling and registered with standard point clouds. Surface defects of components are identified based on the characteristic differences between the two point clouds, resulting in a point cloud of the defective area. Based on the point cloud features of the defective area, the defects are classified into three types: cracks, spalling, and dents. This method achieves high-precision identification and classification of typical defects of rail components. The method is characterized by: The method comprises the following steps: S1, obtains surface point clouds of rails, switches and other track components through 3D profile measurement method; S2, registering the acquired point cloud with the standard point cloud of the rail component surface; S3, by analyzing the differences in spatial position, normal vector, curvature and other features of each point between the registered point cloud and the standard point cloud, the defects on the component surface are identified and the point cloud of the defective area is obtained; S4, based on the partial binary tree method and the point cloud shape of the defect area and the curvature of each point, the defects are classified into three types: cracks, spalling and depressions.

2. The method for identifying and classifying typical defects of rail components based on point cloud features according to claim 1, characterized in that: In S1, the surfaces of rail components such as rails and switches are measured using a high-precision 3D profile measuring instrument to obtain surface point clouds.

3. The method for identifying and classifying typical defects of rail components based on point cloud features according to claim 1, characterized in that: In S2, the acquired point cloud is registered with the standard point cloud of the rail component surface through the iterative closest point algorithm, so that the poses of the two point clouds remain consistent and the correspondence between the points in the two point clouds is obtained.

4. The method for identifying and classifying typical defects of rail components based on point cloud features according to claim 1, characterized in that: The detailed steps of S3 are: S31, traverse each point in the registered point cloud and calculate the Euclidean distance between each point and its corresponding point in the standard point cloud. If the Euclidean distance between a point and its corresponding point is greater than a set threshold, it is determined that there is a defect on the surface of the rail component. Then continue to step S32. If there is no point with a Euclidean distance between its corresponding point and the point greater than the set threshold, it is determined that there is no defect on the surface of the rail component, and the defect identification process is completed. S32, traverse each point in the registered point cloud, first calculate the product of the modulus of the difference between the Euclidean distance between each point and its corresponding point in the standard point cloud and the normal vector, for the i-th point in the registered point cloud, the product f i The calculation formula is: Among them, d i is the Euclidean distance between the i-th point in the registered point cloud and its corresponding point in the standard point cloud, is the normal vector of point i in the registered point cloud, is the normal vector of the corresponding point of the i-th point in the registered point cloud in the standard point cloud; if the f of the i-th point i If it is greater than the set threshold, the point is put into the point set M; Then calculate the product of the Euclidean distance between each point and its corresponding point in the standard point cloud and the absolute value of the curvature difference. For the j-th point in the registered point cloud, the product h j The calculation formula is: h j =d j ·|k j -K j | Among them, d j is the Euclidean distance between the jth point in the registered point cloud and its corresponding point in the standard point cloud, k j is the curvature of the jth point in the registered point cloud, K j is the curvature of the corresponding point of the j-th point in the registered point cloud in the standard point cloud; if the h of the j-th point j If it is greater than the set threshold, the point is put into point set A; Calculate the Euclidean distance between each point in the point set M and each point in the point set A, and select a point A in the point set A i , if there is a point M in the point set M whose Euclidean distance is less than the set threshold i , then mark A i For similar points, traverse each point in point set A and get the number a1 of similar points in point set A; select a point M in point set M j , if there is a point A in point set A whose Euclidean distance is less than the set threshold j , then mark M j For similar points, traverse each point in the point set M and obtain the number m1 of similar points in the point set M; calculate the similarity value L between the point set M and the point set A based on the number of similar points in the point set M: Wherein, a is the number of points in point set A, and m is the number of points in point set M. If the similarity value is greater than the set threshold, it indicates that the similarity between the two point sets is sufficiently high, and the point set M with the larger number of points than the point set A is used as the defect recognition result, and the point cloud of the defect area is obtained, completing the defect recognition process. If the similarity value is less than or equal to the set threshold, it indicates that the similarity between the two point sets is insufficient, and an accurate defect recognition result cannot be obtained, and the process proceeds to step S33. S33, traverse each point in the registered point cloud, calculate the Euclidean distance between each point and its corresponding point in the standard point cloud, the product of the modulus of the difference between the normal vector and the absolute value of the difference in curvature, and for the kth point in the registered point cloud, the product g k The calculation formula is: Among them, d k is the Euclidean distance between the kth point in the registered point cloud and its corresponding point in the standard point cloud, is the normal vector of the kth point in the registered point cloud, is the normal vector of the corresponding point of the kth point in the registered point cloud in the standard point cloud, k k is the curvature of the kth point in the registered point cloud, K k is the curvature of the corresponding point of the k-th point in the registered point cloud in the standard point cloud; if the g of the k-th point k If the value is greater than the set threshold, the point is determined to belong to the defect area, and all points belonging to the defect area are screened out to obtain the point cloud of the defect area, completing the defect identification process.

5. The method for identifying and classifying typical defects of rail components based on point cloud features according to claim 1, characterized in that: The detailed steps of S4 are: S41, in the first layer of the partial binary tree structure, projecting the point cloud of the defect area onto a horizontal plane tangent to the surface of the rail component and converting it into a two-dimensional point cloud, extracting the outer contour of the two-dimensional point cloud using a convex hull algorithm and drawing the minimum circumscribed rectangle of the outer contour, calculating the ratio of the length to the width of the minimum circumscribed rectangle, and classifying the defect as a crack if the ratio is greater than a set threshold; and classifying the defect as a pit if the ratio is less than or equal to the set threshold; S42, in the second layer of the partial binary tree structure, calculate the proportion of points whose curvature is less than the set threshold in the point cloud of the defect area divided into pits. If the value is greater than the set threshold, the pits are further divided into peeling blocks. If the value is less than or equal to the set threshold, the pits are further divided into depressions.

Images