A Visual Inspection Method for the Crimping Quality of Electric Power Hardware

Abstract

The present invention relates to a visual inspection method for the crimping quality of electric power fittings, comprising the following steps: S1. Acquisition of point cloud data: acquiring the spatial point cloud data of the electric power fittings to be measured and converting it into a general.txt file format as the original point cloud data; S2. Denoising of point cloud data: performing noise reduction processing on the point cloud model of the electric power fittings to be measured to remove the outlier points in the original point cloud data; S3. Registration of point cloud models: unifying the reference coordinate system of the target point cloud model with the reference coordinate system of the reference point cloud model; S4. Extraction of point cloud model parameters: segmenting the point cloud model of the electric power fittings to be measured and extracting the parameters of the point cloud model of the electric power fittings to be measured; S5. Visualization of point cloud difference values, including quantitative calculation of point cloud difference values and visualization of point cloud difference values. Compared with the prior art, the present invention has the advantages of high precision, high efficiency, high accuracy, etc.

Description

Technical Field

[0001] The present invention relates to the field of crimping quality detection of electric power fittings, and particularly to a visual detection method for the crimping quality of force fittings. Background Art

[0002] In China, for the connection of overhead transmission lines between conductors and ground wires, a large number of crimping-type electric power fittings such as strain clamps are used. They not only have to bear the charge passing through them, but also have to bear the tension of the conductor or ground wire, and are important stress-bearing and current-conducting equipment on power lines. However, the construction of strain clamps belongs to hidden works, and it is difficult to disassemble once installed. Therefore, ensuring the installation quality of crimping-type electric power fittings such as strain clamps is of great significance for ensuring the reliable operation of the line and ensuring safe power supply.

[0003] At present, the quality detection methods for crimping-type electric power fittings are mainly divided into: manual appearance dimension measurement, grip force test, and X-ray non-destructive testing.

[0004] 1) Manual appearance dimension measurement is to use a vernier caliper to measure and record the appearance dimensions of the strain clamp before and after crimping, mainly relying on manual measurement and visual identification. On the one hand, due to the influence of factors such as manual operation error and subjective judgment in using a vernier caliper for measurement, multiple measurements are often required for confirmation. However, even the results of multiple measurements by the same person will deviate, and it is difficult to ensure the accuracy of the measurement results. On the other hand, repeated measurement confirmation is time-consuming and laborious, and the measurement data needs to be manually copied, resulting in problems such as cumbersome recording, difficult format unification, and difficult digital preservation of results.

[0005] 2) The grip force test is a destructive sampling inspection method carried out in a laboratory. This method can intuitively detect the external and internal crimping states of electric power fittings. However, as a sampling test, the crimping state of the fittings obtained by the grip force test does not represent the crimping state of the fittings used in the line; at the same time, the sampling test will cause losses to the fittings and conductors, and is not suitable for on-site detection.

[0006] 3) X-ray non-destructive testing utilizes the principle that different metal structures absorb different rays, and transmits the signals received by the imager to the computer, which can conveniently, quickly, and accurately capture the images of the internal crimping state of electric power fittings without damaging the object to be detected. However, this method is mainly used to detect the internal crimping quality of electric power fittings, which is the second step of crimping quality detection, and cannot accurately measure the dimensions of external crimping.

[0007] The measurement of the dimensions of power fittings after crimping is particularly important. It is the first and most crucial step to inspect whether the crimping quality is qualified. If the detected dimensions do not meet the standards, re-crimping is required, and only after meeting the standards can the internal crimping quality be detected. Therefore, how to achieve efficient, high-precision, and digital detection of the dimensions of power fittings before and after crimping and visualize the detection results has become a technical problem to be solved. Summary of the Invention

[0008] The purpose of the present invention is to provide a visual inspection method for the crimping quality of power fittings to overcome the defects existing in the above-mentioned prior art.

[0009] The purpose of the present invention can be achieved through the following technical solutions:

[0010] A visual inspection method for the crimping quality of power fittings, the inspection method includes the following steps:

[0011] Step S1, point cloud data acquisition: Obtain the spatial point cloud data of the power fitting to be measured and convert it into a common.txt file format as the original point cloud data, and establish a point cloud model of the power fitting to be measured;

[0012] Step S2, point cloud data denoising: Use the radius filtering algorithm to perform denoising processing on the point cloud model of the power fitting to be measured, and remove the outlier points in the original point cloud data;

[0013] Step S3, point cloud model registration: Use the point cloud model of the power fitting with standard crimping as the reference point cloud model, and use the point cloud model of the power fitting to be measured as the target point cloud model to unify the reference coordinate system of the target point cloud model and the reference coordinate system of the reference point cloud model;

[0014] Step S4, point cloud model parameter extraction: Segment the point cloud model of the power fitting to be measured, and extract the parameters of the point cloud model of the power fitting to be measured through the model fitting algorithm;

[0015] Step S5, visualization of point cloud difference value: Quantitatively calculate the difference value between the registered reference point cloud model and the target point cloud model, convert the measured point cloud difference value into the corresponding color parameters, and assign the corresponding color map to the three-dimensional model for visual display of the detection results.

[0016] Further, the point cloud data acquisition in step S1 is obtained by using a high-precision handheld lidar scanner.

[0017] Further, the specific steps of step S2 are as follows:

[0018] S201. For any point p in the point cloud model of the power fitting to be measured i =(x i ,yi , z i ) T , determine a three-dimensional spherical space region with it as the center of the sphere and a radius of r, and calculate the total number C of point clouds within the r-neighborhood. r (p i )

[0019] S202. Statistically analyze the total number C of point clouds within the r-neighborhood of the point cloud model of the power fitting to be measured r (p i ) to determine the point cloud quantity judgment threshold C for removing noise point clouds T :

[0020]

[0021] Among them, represents the average total number of point clouds within the r-neighborhood of the point cloud model, N represents the number of points in the point cloud model, and σ r represents the standard deviation of the number of point clouds within the r-neighborhood of the point cloud model, and k represents the confidence interval coefficient;

[0022] S203. Remove the outliers in the original point cloud data according to the point cloud quantity judgment threshold C T .

[0023] Further, the specific step S203 is as follows: Traverse the point cloud data. If the total number C of point clouds of point p i is less than the point cloud quantity judgment threshold C r (p i ), then it is considered that this point p T is noise point cloud data and remove it; otherwise, retain the point cloud data. i

[0024] Further, the point cloud model registration in step S3 specifically includes the following steps:

[0025] S301. Point cloud feature extraction: Use the principal component analysis method and the coordinate transformation idea to extract the global feature descriptor GFD of the point cloud, use the fast point feature histogram algorithm to extract the local feature descriptor FPFH of the point cloud, and obtain the fused feature descriptor FFD of the point cloud by integrating GFD and FPFH;

[0026] S302. Coarse point cloud registration: Complete feature point matching by measuring the similarity of the fused feature descriptor FFD of the point cloud. Using the one-to-one correspondence of the matching point pairs, select the top m groups of matching point pairs with the highest similarity of local invariant features from high to low to calculate the corresponding rigid body transformation matrix to achieve the preliminary coincidence of the point cloud models;

[0027] S303, Point cloud fine registration: Based on the rough registration of the point cloud, the ICP algorithm is used for fine registration to obtain the optimal registration result of the point cloud model.

[0028] Furthermore, the parameters in step S4 include the crimping length and the opposite side moment of the crimping.

[0029] Furthermore, the extraction of the point cloud model parameters in step S4 specifically includes:

[0030] S401, Point cloud model segmentation: According to the registration result of the point cloud model, appropriate initial clustering points are selected in a specific area, and the improved k-means algorithm is used to realize the clustering segmentation of the point cloud model to obtain the final clustering segmentation result;

[0031] S402, Axial stratification of the point cloud model: Along the axis of the electrical fitting, several point clouds with a certain thickness are intercepted in layers, and each intercepted layer of the point cloud is projected onto a corresponding plane perpendicular to the central axis to obtain stratified point cloud slices;

[0032] S403, Extraction of point cloud model parameters: Six straight boundaries of the point cloud slice are extracted, and the data points located on the six straight boundaries are found. The found data points are used for model fitting to obtain the boundary straight line of the point cloud slice, and the parameters of the stratified point cloud are calculated according to the straight line equation.

[0033] Furthermore, the clustering segmentation in step S401 divides the point cloud model into a crimping area model and a non-crimping area model.

[0034] Furthermore, in step S403, the RANSAC algorithm is used to extract the six straight boundaries of the point cloud slice, and the total least squares method is used for model fitting of the found data points.

[0035] Furthermore, the visualization of the point cloud difference value in step S5 specifically includes the following steps:

[0036] S501, Quantitative calculation of the point cloud difference value: The k-d tree algorithm and the nearest point correspondence idea are used to determine the point correspondence relationship between the target point cloud model and the reference point cloud model, and the corresponding point cloud difference value is calculated using the point correspondence relationship;

[0037] S502, Visualization method of the point cloud difference value: A color map matrix is established. According to the linear proportional relationship between the point cloud difference value and the maximum allowable error value, the corresponding color parameters are determined, and the corresponding color map is assigned to the three-dimensional model for visual display of the detection result.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. Based on the analysis of the crimping model of electrical fittings, the present invention transforms the parameter detection problem of electrical fittings into the model fitting of the outer contour of the axial section and the automatic extraction of the three-dimensional coordinates of the center point. By introducing artificial intelligence algorithms to replace humans, it effectively realizes the automatic, digital, and high-precision measurement of the appearance size parameters of electrical fittings, significantly improves the accuracy, efficiency, and digital level of the quality inspection of electrical fittings, and solves the challenges in engineering applications.

[0040] 2. The present invention adopts the point cloud fusion feature descriptor FFD that combines the advantages of GFD and FPFH to solve the problem of limited description performance of a single point cloud feature descriptor, effectively improves the accuracy of point cloud model registration, and enhances the accuracy of point cloud parameter extraction and deviation calculation.

[0041] 3. The present invention adopts the RANSAC algorithm and the total least squares method to realize the boundary line fitting of the point cloud slice, which has high robustness to noise points and fitting accuracy.

[0042] 4. The present invention realizes the three-dimensional visualization of the structural size deviation of electrical fittings, which can facilitate operators to intuitively locate the area and degree of structural damage of electrical fittings in the first time, can realize the rapid judgment of the crimping quality of electrical fittings, and effectively improves the detection efficiency and accuracy of the crimping quality of electrical fittings. Description of the Drawings

[0043] Figure 1 is the flowchart of the method of the present invention;

[0044] Figure 2 is the schematic diagram of the r-neighborhood of the point cloud of the present invention;

[0045] Figure 3 is the schematic diagram of the calculation of the r-neighborhood parameters of the point cloud of the present invention;

[0046] Figure 4 is the flowchart of the rough registration of the point cloud of the present invention;

[0047] Figure 5 is the flowchart of the fine registration of the point cloud of the present invention;

[0048] Figure 6 is the flowchart of the point cloud model segmentation algorithm of the electrical fitting of the present invention;

[0049] Figure 7 is the schematic diagram of the quantitative calculation of the point cloud difference value of the present invention;

[0050] Figure 8 is the initial position diagram of the point cloud models of the strain clamp #1 and the strain clamp #2 of the present invention;

[0051] Figure 9Point cloud model rough registration result diagram of strain clamp #1 and strain clamp #2 of the present invention;

[0052] Figure 10 Point cloud model fine registration result diagram of strain clamp #1 and strain clamp #2 of the present invention;

[0053] Figure 11 Initial position diagram of the point cloud model of strain clamp #1 and strain clamp #3 of the present invention;

[0054] Figure 12 Point cloud model rough registration result diagram of strain clamp #1 and strain clamp #3 of the present invention;

[0055] Figure 13 Point cloud model fine registration result diagram of strain clamp #1 and strain clamp #3 of the present invention;

[0056] Figure 14 Diagram of the opposite side moment distribution of strain clamp #1 of the present invention;

[0057] Figure 15 Diagram of the opposite side moment distribution of strain clamp #2 of the present invention;

[0058] Figure 16 Diagram of the opposite side moment distribution of strain clamp #3 of the present invention;

[0059] Figure 17 Visualization result diagram of the structural deviation of strain clamp #2 of the present invention;

[0060] Figure 18 Visualization result diagram of the structural deviation of strain clamp #3 of the present invention. Detailed implementation mode

[0061] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and the detailed implementation manner and specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0062] Embodiment 1

[0063] As Figure 1 shown, a visual inspection method for the crimping quality of electrical fittings includes the following steps:

[0064] Step S1, point cloud data acquisition: Use a high-precision handheld lidar scanner to obtain the spatial point cloud data of the electrical fitting to be measured, and convert it into a.txt file format as the original point cloud data, and establish a point cloud model of the electrical fitting to be measured.

[0065] Step S2, Point cloud data denoising: The radius filtering algorithm is used to denoise the point cloud model of the power fitting to be measured, and the outlier points in the original point cloud data are removed.

[0066] The point cloud data obtained by the high-precision handheld lidar scanner has relatively ideal data quality. The noise point clouds it contains often appear in the form of "outlier points", and the total amount of noise point clouds is relatively small. The method of the present invention uses the radius filtering algorithm for filtering and denoising, and the specific steps are as follows:

[0067] S201. For any point p i =(x i , y i , z i ) T in the point cloud model of the power fitting to be measured, determine a three-dimensional spherical space region with it as the center of the sphere and a radius of r, and calculate the total number C r (p i ) of the point clouds within the r neighborhood;

[0068] S202. Statistically analyze the total number C r (p i ) of the point clouds within the r neighborhood of the point cloud model of the power fitting to be measured, and determine the point cloud number judgment threshold C T :

[0069]

[0070] Wherein, represents the average total number of point clouds within the r neighborhood of the point cloud model, N represents the number of points in the point cloud model, and σ r represents the standard deviation of the number of point clouds within the r neighborhood of the point cloud model, and k represents the confidence interval coefficient (take k = 3);

[0071] S203. According to the point cloud number judgment threshold C T , remove the outlier points in the original point cloud data. Specifically: traverse the point cloud data. If the total number C i of the point cloud of point p r (p i ) is less than the point cloud number judgment threshold C T , then it is considered that this point p i is noise point cloud data and is removed; otherwise, the point cloud data is retained.

[0072] Step S3, Point cloud model registration: Use the point cloud model of the power fitting with standard crimping as the reference point cloud model, and use the point cloud model of the power fitting to be measured as the target point cloud model to unify the reference coordinate system of the target point cloud model and the reference coordinate system of the reference point cloud model.

[0073] S301. Point cloud feature extraction: The global feature descriptor GFD of the point cloud is extracted by using the principal component analysis method and the coordinate transformation idea. The local feature descriptor FPFH of the point cloud is extracted by using the fast point feature histogram algorithm. The fused feature descriptor FFD of the point cloud is obtained by integrating GFD and FPFH.

[0074] Among them, the global feature descriptor GFD constructs a reference coordinate system corresponding to the electric power fitting through the principal component analysis method PCA, and uses the idea of coordinate transformation to convert the three-dimensional coordinate information (x, y, z) of the point cloud into distance and angle parameters in the spherical coordinate system, so as to characterize the spatial position relationship between the point cloud and the electric power fitting. The specific steps are as follows:

[0075] 1) Solve the centroid coordinates of the electric power fitting according to the point cloud distribution and set them as the origin of the reference coordinate system;

[0076]

[0077] 2) Centralize the point cloud data, calculate its covariance matrix C, and perform eigenvalue decomposition on the covariance matrix C to solve the eigenvalues λ0≥λ1≥λ2 and their eigenvectors v0, v1, v2;

[0078] 3) Construct a reference coordinate system corresponding to the electric power fitting: o = p c , v o-x = v2, v o-y = v1, v o-z = v0;

[0079] 4) For any point p i in the point cloud data, calculate its global feature descriptor where:

[0080]

[0081] In the formula, the parameter C = 2ε(f xoy (p i )) - 1, ε(x) represents the step function and ε(0) = 1; f xoy (p) represents the plane equation of the xoy plane of the reference coordinate system, f xoy (p) = [(x, y, z) T - p c · v o-z .

[0082] The local feature descriptor FPFH of the point cloud parameterizes the difference degree between adjacent points by calculating the deviation angle between the surface normal and the normal at the adjacent points of a certain feature point, so as to obtain a complete description of the geometric attributes of the point cloud. The specific steps are as follows:

[0083] 1) AsFigure 2 As shown, determine the query point D q 's k-neighborhood; calculate the Euclidean distance d and the corresponding normal vector n between any two points D s and D t ; and the positional relationship between the corresponding normal vectors n s and n t ; define a local coordinate system UVW at one of the points, as Figure 3 shown; obtain the parametric feature representation:

[0084]

[0085] 2) Calculate the four groups of values between all point pairs within the k-neighborhood to represent the positional relationship between any two points; calculate the Simplified Point Feature Histogram (SPFH) and obtain the FPFH value by weighted summation.

[0086]

[0087] S302, Coarse Point Cloud Registration: As Figure 4 shown, complete feature point matching by measuring the similarity of the Feature Fusion Descriptor (FFD) of the point cloud. Using the one-to-one correspondence of the matching point pairs, select the top m groups of matching point pairs with the highest similarity of local invariant features from high to low and calculate the corresponding rigid body transformation matrix to achieve the preliminary coincidence of the point cloud model. The specific steps are as follows:

[0088] 1) Use the FFD as the similarity criterion to establish two corresponding point sets P i and Q i ; randomly select 3 feature point pairs (p i , q i ) from all feature point pairs, i = 1, 2, 3; calculate the corresponding initial rotation matrix R0 and translation matrix T0 using the singular value decomposition method.

[0089] 2) Traverse all feature point pairs and calculate the measurement error e j generated by any feature point pair (p j , q j ) under the action of the initial rigid body transformation matrix:

[0090] e j = ||p j - R0q j - T0||

[0091] Compare the measurement error e j with the error judgment threshold e t : If e j < e t , then consider the feature point pair (p j , q j) is the correct feature point pair; otherwise, the feature point pair (p j , q j ) is an incorrect feature point pair;

[0092] 3) Count the number n of correct feature point pairs k ; Select the rigid body transformation result corresponding to the most n k to obtain the initial rotation matrix R0 and translation matrix T0.

[0093] S303, Point cloud fine registration: On the basis of point cloud rough registration, further reduce the registration error between the reference point cloud model and the target point cloud model, so as to obtain a high-precision point cloud registration result. As Figure 5 shown, the method of the present invention uses the ICP algorithm for fine registration of the point cloud model, and the specific steps are as follows:.

[0094] 1) For any reference point cloud p i , traverse the data points in the target point cloud set, and select the corresponding matching point q k according to the principle of the nearest point, so as to obtain two corresponding point sets P i and Q i ;

[0095] 2) According to the one-to-one correspondence between the corresponding point sets P i and Q i , the objective function f(R k , T k ) can be constructed, and the optimal rotation matrix R k and translation matrix T k can be solved;

[0096]

[0097] 3) According to the rotation matrix R k and translation matrix T k , obtain the new target point cloud Q * ; Calculate the average distance error d between the target point cloud Q * with optimized position and the reference point cloud P:

[0098]

[0099] 4) Judge whether the calculated average distance error d meets the convergence condition: If the average distance error d is less than the judgment threshold τ set by the algorithm, the algorithm ends; otherwise, the algorithm continues to iterate until the number of algorithm iterations is greater than the preset maximum number of iterations, then stop the iterative calculation. Obtain the final rotation matrix R k and translation matrix T k .

[0100] Step S4, Point Cloud Model Parameter Extraction: Segment the point cloud model of the power fitting to be measured, and extract the parameters of the point cloud model of the power fitting to be measured through the model fitting algorithm.

[0101] S401, Point Cloud Model Segmentation: As Figure 6 shown, select appropriate initial clustering points in a specific area according to the point cloud model registration result, and use the improved k-means algorithm to achieve the clustering segmentation of the point cloud model, and divide the point cloud model into a crimping area model and a non-crimping area model. The specific steps are as follows:

[0102] 1) For any target point cloud p i , determine the set of point cloud data P i within its r neighborhood, and calculate the surface equation of the r neighborhood using the quadratic surface fitting method; according to the first fundamental form of the surface equation, calculate the first fundamental quantities E, F, and G; according to the second fundamental form of the surface equation, calculate the second fundamental quantities L, M, and N; using the first fundamental quantities and the second fundamental quantities, the Gaussian curvature K, mean curvature H, and principal curvatures k1 and k2 of the target point cloud p i can be calculated:

[0103]

[0104] 2) For any point cloud data p i , calculate its point cloud feature vector F(p i );

[0105] F(p i ) = [x i , y i , z i , K i , H i , k 1i , k 2i

[0106] 3) For the registered target point cloud model, its spatial range has been roughly determined; according to prior knowledge, randomly select a clustering center as the initial clustering center in different point cloud model regions respectively;

[0107] 4) For any point cloud data p i , calculate its point cloud feature distance D(p i , c i ) to each clustering center c j :

[0108] D(p i , c j ) = ||F(p i ) - F(c j )||​

[0109] 5) For any point cloud data p i , determine its belonging clustering region according to the point cloud feature distance from it to each clustering center c i :

[0110] S(p i ) = argmin j D(p i , c j )

[0111] where S i represents the i-th clustering region, and the clustering center of S i is equal to c i ; thus, all current clustering results {S1, S2, …, S N} are obtained;

[0112] 6) According to the current clustering results {S1, S2, …, S N}, re-update the clustering centers of each clustering region to obtain a new clustering center result {c1, c2, …, c N},

[0113]

[0114] 7) Repeat the above steps 4) to 6) until the updated clustering centers {c1, c2, …, c N} no longer change or reach the maximum number of iterations, then the clustering ends and the final clustering segmentation result is obtained.

[0115] S402. Axial stratification of the point cloud model: Axially stratify the point cloud along the power hardware to intercept several point clouds with a certain thickness, and project each intercepted layer of point cloud onto a corresponding plane perpendicular to the central axis to obtain stratified point cloud slices.

[0116] The lidar point cloud data is a discrete data form. If simply choosing the way of the intersection of the point cloud and the plane to obtain the contour of the cross-section of the power hardware point cloud model, it may lead to a small number of contour point clouds, which is not enough to restore the real situation of the power hardware model. Therefore, the method of the present invention introduces the idea of slicing, and uses point cloud slices with a certain thickness to replace the simple plane. Set an appropriate number n of axial stratifications, and divide the point cloud model into n point cloud stratifications {S1, S2, …, S n} along the central axis direction of the power hardware.

[0117] S403. Point cloud model parameter extraction: The RANSAC algorithm is used to extract the six straight-line boundaries of the point cloud slice, and the data points located on the six straight-line boundaries are found. The total least squares method is used to perform model fitting on the found data points to obtain the boundary straight line of the point cloud slice, and the parameters of the stratified point cloud are calculated according to the straight-line equation.

[0118] For electrical hardware, the main object of quality inspection in its crimping area; before crimping, the electrical hardware is a regular cylindrical structure, and after crimping, it becomes a relatively regular hexagonal prism. Therefore, the main parameters reflecting the crimping situation of the electrical hardware include the crimping length and the opposite side distance. The method of the present invention uses the idea of model fitting to fit the stratified cross-section model of the electrical hardware and extract parameters such as the opposite side distance. The specific steps are as follows:

[0119] 1) For any stratified data set S of a sub-model i ∈{S1, S2, …, S n}, determine its projection data in the axial direction, and use the RANSAC algorithm to perform straight-line detection and extraction on it, so as to obtain the six straight-line boundaries of the stratified data:

[0120] a. Randomly select two data points from the current stratified data set S i , and determine the corresponding straight-line equation y = kx + b according to the two data points;

[0121] b. Traverse any point p i in the stratified data set S i , calculate the distance from point p i to the fitted straight line,

[0122]

[0123] Compare d i with the judgment threshold d t : If d i < d t , then it is considered that point p i is an "inlier" of the fitted straight line; otherwise, it is considered that point p i is an "outlier" of the fitted straight line;

[0124] c. Count the number N k of "inliers" of the fitted straight line; select the straight-line model with the largest N k ;

[0125] 2) Divide the six straight-line equations extracted in step 1) into 3 groups of parallel line groups, and calculate the distances d1, d2, d3 between each group of parallel lines; by solving the maximum value of the distances of the three groups of parallel lines, the corresponding opposite side distance can be determined:

[0126] SL = max(d1, d2, d3)

[0127] 3) According to the six straight-line equations extracted in step 1), the corresponding six straight-line intersection points (x j , y j , z j ) can also be determined, where j = 1, 2, …, 6. The center point of the layered cross-section can be determined according to the coordinate values of the intersection points:

[0128]

[0129] 4) Determine the set {c1, c2, …, c r} of the center points of all sub-model layers. Take c1 as the starting point and c r as the ending point, and thus determine the fitting equation l c of the central axis of the electrical fitting; calculate the distances {d2, d3, …, d k-1} from the remaining layer centers to the fitting straight-line equation; thereby, the maximum chord height h of the electrical fitting can be determined;

[0130] h = max(d2, d3, …, d k-1 )

[0131] 5) Statistically record all appearance dimension parameters of the electrical fitting: crimping length L; crimping opposite side moment S L ; maximum chord height h at the bending part.

[0132] Step S5, Visualization of point cloud difference value: Quantitatively calculate the difference value between the registered reference point cloud model and the target point cloud model, convert the measured point cloud difference value into corresponding color parameters, and assign the corresponding color map to the 3D model for visual display of the detection result.

[0133] S501, Quantitatively calculate the point cloud difference value: As Figure 7 shown, quantitatively calculating the point cloud difference value is to calculate the difference between the target point cloud model and the reference point cloud model by comparing the registered reference point cloud model and the target point cloud model; if the electrical fitting has undergone a certain deformation, then Figure 7 there is a corresponding deformation deviation on the surface in

[0134] 1) For any point q i in the target point cloud of the electrical fitting, use the k-d tree nearest neighbor search algorithm to find the data point p i closest to it in the reference point cloud, and calculate the distance d i between point q i and point p M ;

[0135]

[0136] In the formula, (x pi ,y pi ,z pi ) represents the point p i The three-dimensional coordinate value, (x qi ,y qi ,z qi ) represents the point q i The three-dimensional coordinate value of .

[0137] 2) Set point q i The physical corresponding point is point If the electrical fittings have no structural deviation, the surface will not have deformation deviation, and point q i and Point Completely overlap, then point p i and Point The distance between R It is equivalent to the point cloud registration error, which can be expressed by the root mean square error of the distance between the matching point pairs;

[0138]

[0139] 4) According to point p i , click q i and Point The spatial geometric relationship of the physical corresponding point q can be calculated i and The distance between T , this distance is also equivalent to the quantitative calculation result of the difference value of the point cloud.

[0140]

[0141] S502, point cloud difference value visualization method: establish a color map matrix, clarify the corresponding color parameters according to the linear proportional relationship between the point cloud difference value and the maximum allowable error value, and assign the corresponding color map to the three-dimensional model, so as to display the structural deviation in a three-dimensional visualization form on the power fitting point cloud model. Set the maximum allowable error value E max And the color map matrix Color, if point p i The point cloud difference value at is calculated as d T , then with point p i The corresponding color matrix index can be determined as ind i ,

[0142]

[0143] In the formula, ind maxRepresents the maximum index value of the color matrix.

[0144] In this embodiment, a tension clamp with qualified crimping and two tension clamps with unqualified crimping are selected for visual inspection of the crimping quality of electrical fittings. Among them, the tension clamp with qualified crimping is numbered #1, and the two tension clamps with unqualified crimping are numbered #2 and #3 respectively.

[0145] The point cloud data of the three tension clamps are collected by a handheld lidar scanner, and the point cloud data of the three tension clamps are denoised. As Figures 8 - 13 shown, taking the point cloud model of the tension clamp #1 as the reference point cloud model and the point cloud models of the tension clamps #2 and #3 as the target point cloud models, the operations of point cloud model registration are carried out respectively.

[0146] After the registration is completed, the parameters of the tension clamp #1, the tension clamp #2, and the tension clamp #3 are extracted respectively. The extracted parameter results are shown in Table 1, and the distribution of the side moments is as Figures 14 - 16 shown.

[0147] Table 1 Parameter extraction results

[0148] Measuring parameter Strain clamp #1 Strain clamp #2 Strain clamp #3 <![CDATA[Length L0 of non-pressed area of aluminum tube / mm]]> 230.45204 220.55126 217.89625 <![CDATA[Crimping length L1 of aluminum pipe on the steel anchor side / mm]]> 118.38518 121.16169 126.90010 <![CDATA[Pressing length L2 of aluminum pipe on wire side / mm]]> 278.87448 295.14273 291.11778 Chord height h / mm at the maximum bending position 2.16661 4.66855 13.18316 Total length L / mm of the aluminum tube of the strain clamp 627.71171 636.85568 544.91798

[0149] As Figure 17 and 18 shown, after the parameters of the tension clamps are extracted, the visual processing of the structural deviation of the tension clamps #2 and #3 is carried out respectively to obtain the final detection visual result.

[0150] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations according to the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention based on the concept of the present invention through logical analysis, reasoning or limited experiments should be within the protection scope determined by the claims.

Claims

1. A visual inspection method for the crimping quality of electrical fittings, characterized in that, The described detection method includes the following steps: Step S1, Point cloud data acquisition: Obtain the spatial point cloud data of the power fitting to be measured, convert it into the.txt file format as the original point cloud data, and establish the point cloud model of the power fitting to be measured. Step S2, Point cloud data denoising: Use the radius filtering algorithm to perform noise reduction processing on the point cloud model of the power fitting to be measured, and remove the outlier points in the original point cloud data. Step S3, Point cloud model registration: Use the point cloud model of the power fitting with standard crimping as the reference point cloud model, and the point cloud model of the power fitting to be measured as the target point cloud model, and unify the reference coordinate system of the target point cloud model with the reference coordinate system of the reference point cloud model. Step S4, Point cloud model parameter extraction: Segment the point cloud model of the power fitting to be measured, and extract the parameters of the point cloud model of the power fitting to be measured through the model fitting algorithm. Step S5, Visualization of point cloud difference value: Quantitatively calculate the difference value between the registered reference point cloud model and the target point cloud model, convert the measured point cloud difference value into the corresponding color parameter, and assign the corresponding color map to the 3D model for visual display of the detection result.

2. The visual inspection method for the crimping quality of electric power fittings according to claim 1, wherein The point cloud data acquisition in step S1 is obtained by using a high-precision handheld lidar scanner.

3. A visual inspection method for the crimping quality of electrical fittings according to claim 1, characterized in that, The specific steps of step S2 are as follows: S201. For any point p in the point cloud model of the power hardware to be measured i =(x i , y i , z i ), T determine a three-dimensional spherical space region with it as the center of the sphere and a radius of r, and calculate the total number C of points in the r-neighborhood r (p i ); S202. Statistically analyze the total number C of point clouds within the r-neighborhood of the point cloud model of the power hardware to be measured r (p i ) to determine the point cloud quantity judgment threshold C for noise point cloud removal T : Among them, represents the total average number of point clouds within the r-neighborhood of the point cloud model, N represents the number of points in the point cloud model, and σ r represents the standard deviation of the number of point clouds within the r-neighborhood of the point cloud model, and k represents the confidence interval coefficient; S203. Determine the threshold C based on the number of point clouds T , and remove the outliers in the original point cloud data.

4. A visual inspection method for the crimping quality of electrical hardware according to claim 3, characterized in that, The specific step S203 is as follows: traverse the point cloud data, such as point p i with the total number of point clouds C r (p i ) is less than the point cloud quantity judgment threshold C T , then it is considered that the point p i is noise point cloud data and is removed; otherwise, the point cloud data is retained.

5. A visual inspection method for the crimping quality of electrical fittings according to claim 1, characterized in that, The point cloud model registration in step S3 specifically includes the following steps: S301, Point cloud feature extraction: Use the principal component analysis method and the coordinate transformation idea to extract the global feature descriptor GFD of the point cloud, use the fast point feature histogram algorithm to extract the local feature descriptor FPFH of the point cloud, and obtain the fused feature descriptor FFD of the point cloud by integrating GFD and FPFH. S302, Coarse point cloud registration: Complete the feature point matching by measuring the similarity of the fused feature descriptor FFD of the point cloud, use the one-to-one correspondence relationship of the matching point pairs, select the top m groups of matching point pairs with the highest similarity of local invariant features from high to low to calculate the corresponding rigid body transformation matrix, and realize the preliminary coincidence of the point cloud models. S303, Fine point cloud registration: On the basis of the coarse point cloud registration, use the ICP algorithm for refined registration to obtain the optimal point cloud model registration result.

6. The visual inspection method for the crimping quality of electric power fittings according to claim 1, characterized in that, The parameters in step S4 include the crimping length and the crimping opposite side moment.

7. A visual inspection method for the crimping quality of electrical fittings according to claim 1, characterized in that, The point cloud model parameter extraction in step S4 specifically includes: S401, Point cloud model segmentation: Select appropriate initial clustering points in a specific area according to the point cloud model registration result, and use the improved k-means algorithm to realize the clustering segmentation of the point cloud model to obtain the final clustering segmentation result. S402, Axial layer segmentation of the point cloud model: Axially segment the point cloud along the power fitting to intercept several point clouds with a certain thickness, and project each intercepted layer of point cloud onto the corresponding plane perpendicular to the central axis to obtain the layered point cloud slices. S403, Point cloud model parameter extraction: Extract the six straight line boundaries of the point cloud slice, find the data points located on the six straight line boundaries, perform model fitting on the found data points to obtain the boundary straight line of the point cloud slice, and calculate the parameters of the layered point cloud according to the straight line equation.

8. A visual inspection method for the crimping quality of electric power fittings according to claim 7, characterized in that The clustering segmentation in step S401 divides the point cloud model into a crimping area model and a non-crimping area model.

9. A visual inspection method for the crimping quality of electric power fittings according to claim 7, characterized in that, In step S403, the RANSAC algorithm is used to extract the six straight-line boundaries of the point cloud slice, and the total least squares method is used to fit the found data points to a model.

10. A visual inspection method for the crimping quality of electrical fittings according to claim 1, characterized in that, The visualization of the point cloud difference value in step S5 specifically includes the following steps: S501. Quantitative calculation of the point cloud difference value: The k-d tree algorithm and the nearest point correspondence idea are used to determine the point correspondence relationship between the target point cloud model and the reference point cloud model, and the corresponding point cloud difference value is calculated using the point correspondence relationship; S502. Visualization method of the point cloud difference value: Establish a color map matrix, determine the corresponding color parameters according to the linear proportional relationship between the point cloud difference value and the maximum allowable error value, and assign the corresponding color map to the three-dimensional model for visual display of the detection result.

Images