Aero-engine turbine rotor robot assembly point cloud feature recognition method and system

Abstract

The invention discloses an aero-engine turbine rotor robot assembly point cloud feature recognition method and system, and the method comprises the steps: calculating a rotation matrix of a point cloud according to the coordinate pose of the tail end of a robot and the hand-eye relation of the robot; extracting a first point cloud in a height range where the point cloud features are located, and solving normal vector information of the processed first point cloud; according to the solved normal vector information, establishing a boundary point judgment model and judging whether the to-be-monitored point is a boundary point to obtain an extracted second point cloud; performing three-dimensional circle model fitting on the second point cloud in combination with normal vector information, extracting fitted inner points, and gradually extracting a plurality of three-dimensional circle structures in the point cloud; and fitting the three-dimensional circle to obtain circle center, radius and axis information of a large circle with small circular holes, and obtaining an aero-engine turbine rotor robot assembly point cloud feature image. According to the method disclosed by the invention, accurate matching and attitude adjustment of the parts are realized by utilizing the high-resolution characteristic of the point cloud data, and the assembly precision is greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of aero-engine assembly manufacturing, and more specifically, relates to a method and system for identifying point cloud features of a power turbine rotor of an aero-engine by a robot during assembly. Background Art

[0002] With the continuous iteration of aero-engine technology, the manufacturing and assembly precision requirements for its core components are increasing day by day. As a key component of an aero-engine, the engine turbine rotor involves complex structural design and strict tolerance requirements, and its assembly quality directly affects the performance and service life of the engine. Traditional turbine rotor assembly methods mainly rely on manual operation or simple mechanical tooling. These methods have the following problems: low efficiency: Manual assembly is highly dependent on skilled workers, and the assembly speed is limited by the proficiency of the workers; insufficient accuracy: Due to human factors, assembly errors and consistency problems may occur, making it difficult to meet high-precision requirements; low automation level: Most of the existing tooling equipment is custom-designed, with poor adaptability, and it is difficult to flexibly handle the assembly requirements of different types of turbine rotors.

[0003] In recent years, point cloud data processing technology has been widely used in the industrial field. It can quickly obtain the spatial geometric information of components through a three-dimensional point cloud camera. Combining with feature recognition algorithms, it can achieve high-precision position and attitude matching. This provides a new solution for the automation of complex assembly tasks. For example, patent document CN 116341050 B discloses a robot intelligent construction method based on point cloud data. First, the precast components of the point cloud data of the prefabricated building are identified and segmented by using the geometric features of the precast components of the prefabricated building and the topological relationship features of the point cloud space to which they belong, and an API interface for calling point cloud data is developed based on Revit to read the point cloud data at the Revit end to achieve reverse modeling; then a precast component coding system is established based on the Omniclass information classification system, and data is transmitted between the BIM model established based on the IFC standard and the intelligent manufacturing end; then the components are divided and positioned, and the precast components based on Scan-vs-BIM are pre-assembled to obtain the optimal component assembly plan.

[0004] However, the application of point cloud technology in aircraft engine assembly is still in the exploratory stage. Especially in the high-precision and high-complexity assembly task of turbine rotor, its point cloud feature recognition faces the following difficulties: (1) In the process of aircraft engine automated assembly, there may be tiny assembly reference features (such as holes, notches, etc.) on the parts, which play an important role in determining the assembly posture. However, these point cloud features only account for a very small proportion of the overall structure, and their recognition difficulty is significantly increased compared with conventional point cloud recognition tasks. The core of this method is how to accurately extract target features from complex point cloud data while ensuring that the overall integrity and accuracy of the fitting point cloud are not affected; (2) Aircraft engine power turbine rotors usually have complex geometric structures and tiny assembly reference features. Their point cloud data presents high density and multi-noise characteristics, which significantly increases the difficulty of feature extraction and matching. Current point cloud algorithms usually rely on the saliency of the overall geometric structure in feature recognition, but are relatively less robust in extracting small features. When processing complex point cloud data of aircraft engines, the algorithm is easily affected by factors such as local noise, uneven sampling of point clouds, or sparse distribution, resulting in recognition accuracy that is difficult to meet requirements. In addition, since most aero-engine materials are high-strength metals with strong reflectivity, this further limits the integrity of feature information. In this context, the pre-processing steps of point cloud data, such as denoising, filtering, and segmentation, are particularly critical in the field of point cloud feature recognition for automated assembly of aero-engines. However, the pre-processing steps of existing point cloud feature recognition methods may cause feature loss or geometric deformation when processing complex point cloud data and tiny features, which will adversely affect the effects of subsequent feature extraction and recognition. Summary of the invention

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a point cloud feature recognition method and system for the robot assembly of aircraft engine power turbine rotors. Through the combination of regional filtering and statistical filtering, outliers in three-dimensional point cloud data can be effectively removed, the overall quality and reliability of point cloud data can be improved, and model fitting algorithms such as normal vector filtering and RANSAC three-dimensional circle fitting are used to finally realize the feature recognition of aircraft engine power turbine rotors. This method utilizes the high-resolution characteristics of point cloud data and combines feature recognition algorithms to achieve precise matching and posture adjustment of parts, greatly improve assembly accuracy, ensure that the assembly results meet strict tolerance requirements, and effectively solve many deficiencies of traditional manual assembly, such as efficiency and stability.

[0006] In order to achieve the above object, according to a first aspect of the present invention, a method for identifying point cloud features of an aircraft engine turbine rotor assembly robot is provided, comprising:

[0007] S100: Calculate the rotation matrix of the point cloud according to the coordinate pose of the robot end and the robot's hand-eye relationship during shooting, so that the point cloud coordinates are transferred from the camera coordinate system to the robot base coordinate system;

[0008] S200: Screen the transformed point cloud data according to the Z-axis, extract the first point cloud within the height range where the point cloud features are located, sample and filter the extracted first point cloud, and solve the normal vector information of the processed first point cloud;

[0009] S300: Establish a boundary point judgment model according to the solved normal vector information and judge whether the point to be monitored is a boundary point. After judging all points for boundary point cloud, obtain the extracted second point cloud;

[0010] S400: Fit the three-dimensional circle model to the second point cloud in combination with the normal vector information, obtain the center, radius and axis information of the three-dimensional circle, extract the inliers fitted, and repeat the above model fitting and point cloud rejection steps for the remaining points to gradually extract multiple three-dimensional circle structures in the point cloud;

[0011] S500: First screen the three-dimensional circle structures fitted in the previous step according to the radius size, obtain the center, radius and axis information of the small circles accurately on the large circumference, extract the fitted small circle centers as the new three-dimensional circle circumferential point clouds, fit the three-dimensional circles to obtain the center, radius and axis information of the large circles with distributed small holes, and after visualizing the large circles, obtain the point cloud feature image of the aero-engine turbine rotor robot assembly.

[0012] Further, in step S200, the sampling is performed by dividing the space into voxels of a fixed size and selecting a representative point in each voxel to approximate the original point cloud. The specific steps are as follows:

[0013] S201: Calculate the bounding box of the point cloud and divide it into cubic grids with side lengths;

[0014] S202: Assign each point (x i , y i , z i ) to the corresponding voxel (i, j, k), where:

[0015]

[0016] where x min , y min , z min is the minimum coordinate of the bounding box;

[0017] S203: For each voxel, calculate the centroid of all points in it as the representative point:

[0018]

[0019] Where: N is the number of points in the voxel.

[0020] Further, in step S200, the filtering includes:

[0021] S204: For each point p in the point cloud i , determine its set of k nearest neighbor points N i ;

[0022] S205: Calculate the local density ρ i of point p i , defined as the ratio of the number of points in its neighborhood to the neighborhood volume:

[0023]

[0024] Where V i is the volume of the smallest sphere centered at p i and containing its k neighbors;

[0025] S206: Calculate the average distance i from point p

[0026]

[0027] to each point in its neighborhood i ,p j ), representing the Euclidean distance between point p i and p j ;

[0028] S207: Calculate the global mean μ and standard deviation σ of the average distances of all points:

[0029]

[0030] Where n is the total number of points in the point cloud;

[0031] S208: Set the adaptive threshold τ i according to the local density ρ i of point p i :

[0032]

[0033] Where α and β are adjustment parameters, and ρ max is the maximum local density in the point cloud;

[0034] S208: If the average distance d i of point p i exceeds its adaptive threshold τ i, it is marked as an outlier and removed.

[0035] Furthermore, the determination of the boundary points described in step S300 includes:

[0036] S301: Let p be the point to be detected, and n p be the normal vector of this point, and v be the observation direction vector;

[0037] S302: Calculate the angle θ between the normal vector n p of the point to be detected and the observation direction vector v:

[0038]

[0039] where: n p ·v is the dot product of the normal vector and the observation direction vector, and ||n p ||, ||v|| are their respective magnitudes;

[0040] S303: Judge the magnitude relationship between θ and the given threshold θ threshold :

[0041] When θ > θ threshold , this point is a boundary point. When θ < θ threshold , then this point is not a boundary point.

[0042] Furthermore, in step S400, the three-dimensional circle model fitting includes:

[0043] S401: Based on the normal vector information of the point cloud, select points with similar normal vector directions for sampling to increase the probability that the sampled points belong to the same circle;

[0044] S402: Dynamically adjust the distance threshold according to the local density and noise level of the point cloud to more effectively distinguish inliers and outliers and achieve model fitting;

[0045] S403: Introduce a multi-scale analysis method to perform multi-level verification on the fitting results to improve the robustness of the model.

[0046] Furthermore, step S401 includes:

[0047] Randomly select three points n a , n b , n c from the point cloud, and require the included angles of the normal vectors of these three points to be within a certain range:

[0048] ∠(n o , n b ) < θ max , ∠(n b , n c ) < θ max

[0049] where θ max is the preset maximum included angle threshold;

[0050] Calculate the center C, radius R and normal vector n of the fitted circle based on the three selected points;

[0051] Let the coordinates of the three points be p a =(x a , y a , z a ), p b =(x b , y b , z b ), p c =(x c , y c , z c ), then the parameters of the fitted circle are:

[0052]

[0053] The center C, radius R and normal vector n can be obtained by solving.

[0054] Furthermore, differentiating inliers and outliers in step S402 includes:

[0055] Calculate the distance from each point to the fitted circle, and determine whether it is an inlier according to the adaptive threshold. The distance d i from point p i to the fitted circle is:

[0056] d i =|||p i -C||-R

[0057] The adaptive distance threshold is:

[0058]

[0059] where ρ max is the maximum local density of the point cloud, and d 0 is the initial distance threshold;

[0060] If , then it is determined that p i is an inlier.

[0061] According to the second aspect of the present invention, there is provided an aeroengine turbine rotor robot assembly point cloud feature recognition system, including:

[0062] A coordinate transformation module, which is used to calculate the rotation matrix of the point cloud according to the coordinate pose of the end of the robot during shooting and the robot's hand-eye relationship, so as to transfer the point cloud coordinates from the camera coordinate system to the robot base coordinate system;

[0063] A point cloud processing module, which is used to screen the transformed point cloud data according to the Z-axis, extract the first point cloud within the height range where the point cloud features are located, sample and filter the extracted first point cloud, and solve the normal vector information of the processed first point cloud;

[0064] A boundary point judgment module, which is used to establish a boundary point judgment model according to the solved normal vector information and judge whether the point to be monitored is a boundary point, until the second point cloud is extracted after all points are judged as boundary point clouds;

[0065] A three-dimensional circle model fitting module, which is used to perform three-dimensional circle model fitting on the second point cloud combined with the normal vector information, obtain the center, radius and axis information of the three-dimensional circle, extract the inliers fitted, and repeat the above model fitting and point cloud rejection steps for the remaining points, and gradually extract multiple three-dimensional circle structures in the point cloud;

[0066] A point cloud feature image module, which is used to first screen the three-dimensional circle structures fitted in the previous step according to the radius size, obtain the center, radius and axis information of the small circles accurately on the large circumference, extract the centers of the fitted small circles as the new three-dimensional circle circumferential point clouds, fit the three-dimensional circles to obtain the center, radius and axis information of the large circles with distributed small holes, and obtain the point cloud feature image of the aero-engine turbine rotor robot assembly after visualizing the large circles.

[0067] According to the third aspect of the present invention, a terminal device is provided, including:

[0068] A memory, which is used to store a computer program;

[0069] A processor, which is used to implement the steps of the aero-engine turbine rotor robot assembly point cloud feature recognition method when executing the computer program.

[0070] According to the fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the aero-engine turbine rotor robot assembly point cloud feature recognition method are implemented.

[0071] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

[0072] 1. The method of the present invention can effectively remove outliers in three-dimensional point cloud data, improve the overall quality and reliability of the point cloud data, and utilize model fitting algorithms such as normal vector filtering and RANSAC three-dimensional circle fitting through the combination of regional filtering, statistical filtering, and neighborhood search algorithms. Finally, the feature recognition of the power turbine rotor of an aeroengine is realized. By utilizing the high-resolution characteristics of the point cloud data and combining with the feature recognition algorithm, this method realizes the precise matching and attitude adjustment of parts, greatly improves the assembly accuracy, ensures that the assembly result meets strict tolerance requirements, and can effectively solve many deficiencies such as the efficiency and stability of traditional manual assembly.

[0073] 2. The method of the present invention dynamically adjusts the threshold according to the local density to adapt to non-uniform density point cloud data, strictly controls in high-density areas, and relaxes in low-density areas, improves the accuracy of outlier detection, can more effectively process non-uniform density point cloud data, accurately removes outliers, reduces noise interference, and improves the overall quality of point cloud processing.

[0074] 3. The method of the present invention uses normal vector information to guide sampling to increase the probability of effective sampling, reduce the number of iterations, and perform adaptive threshold setting and multi-scale verification, which improves the algorithm's resistance to noise and outliers, can more effectively process the extraction of three-dimensional circle structures in complex point cloud data, and improves the accuracy and efficiency of three-dimensional circle model fitting. Description of the Drawings

[0075] Figure 1 It is a flow chart of the point cloud feature recognition method for the robotic assembly of the power turbine rotor of an aeroengine in an embodiment of the present invention;

[0076] Figure 2 It is the overall visualized point cloud data of the turbine rotor in an embodiment of the present invention;

[0077] Figure 3 It is a flow chart of the point cloud preprocessing steps in an embodiment of the present invention;

[0078] Figure 4 It is the visualized point cloud data of the turbine rotor after regional filtering in an embodiment of the present invention;

[0079] Figure 5 It is the visualized point cloud of the normal vector after uniform sampling and statistical filtering in an embodiment of the present invention;

[0080] Figure 6 It is a flow chart of the boundary estimation algorithm in an embodiment of the present invention;

[0081] Figure 7 It is the boundary point cloud data extracted in an embodiment of the present invention;

[0082] Figure 8The visualized point cloud data of the characteristic three-dimensional circle fitted for the embodiments of the present invention;

[0083] Figure 9 The fitting image of the large circle where the characteristic circles of the embodiments of the present invention are distributed;

[0084] Figure 10 The visualized image of the point cloud of the large and small characteristic points fitted in the overall turbine rotor point cloud for the embodiments of the present invention;

[0085] Figure 11 The schematic diagram of the organization of the point cloud feature recognition system for the robotic assembly of the power turbine rotor of an aeroengine in the embodiments of the present invention. Detailed implementation manners

[0086] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, 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 only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0087] Embodiment 1

[0088] As Figure 1 shown, the embodiments of the present invention propose a method for recognizing point cloud features in the robotic assembly of the power turbine rotor of an aeroengine, which is used to process and analyze complex three-dimensional point cloud data and extract the required point cloud features:

[0089] S100: Calculate the rotation matrix of the point cloud according to the coordinate pose of the robot end and the robot eye-in-hand relationship during shooting, so as to transfer the point cloud coordinates from the camera coordinate system to the robot base coordinate system;

[0090] S200: Screen the transformed point cloud data according to the Z axis, extract the first point cloud within the height range where the point cloud features are located, sample and filter the extracted first point cloud, and solve the normal vector information of the processed first point cloud;

[0091] S300: Establish a boundary point judgment model according to the solved normal vector information and judge whether the point to be monitored is a boundary point, until after all points are judged as boundary point clouds, the extracted second point cloud is obtained;

[0092] S400: Fit the three-dimensional circle model to the second point cloud in combination with the normal vector information, obtain the center, radius and axis information of the three-dimensional circle, extract the inliers fitted, and repeat the above model fitting and point cloud rejection steps for the remaining points to gradually extract multiple three-dimensional circle structures in the point cloud;

[0093] S500: First, filter the three-dimensional circular structure fitted in the previous step according to the radius size to obtain the center, radius, and axis information of the small circles accurately on the large circumference. Extract the center of the fitted small circle as the new three-dimensional circular circumference point cloud, fit a three-dimensional circle to obtain the center, radius, and axis information of the large circle where the small circular holes are distributed, and after visualizing the large circle, obtain the point cloud feature image of the aero-engine turbine rotor robot assembly.

[0094] Specifically, it includes

[0095] S101: Load the point cloud file. According to the coordinate pose of the robot end during shooting and the robot hand-eye relationship, calculate the rotation matrix of the point cloud, and obtain the 4x4 transformation matrix through matrix multiplication in OpenCV. Convert the transformation matrix to the Eigen::Matrix4f format, and use the coordinate transformation function to complete the coordinate system transformation of the point cloud, so that the point cloud coordinates are transferred from the camera coordinate system to the robot base coordinate system, which is convenient for subsequent filtering through the coordinate information of the point cloud. The visualized point cloud data of the turbine rotor is as Figure 2 shown.

[0096] S103: Preprocess the point cloud data, including uniform filtering, voxel sampling, Gaussian filtering, normal vector calculation, etc., as Figure 3 shown.

[0097] S1311: Since in the above steps, the point cloud data is transformed to the robot base coordinate system through the transformation matrix, filter the transformed point cloud data according to the z-axis, and extract the point cloud within the height range where the point cloud features are located. After this filtering, most of the miscellaneous points and noise point clouds are also filtered out at the same time. Sampling is to divide the space into voxels of a fixed size, and select a representative point in each voxel to approximate the original point cloud. The specific steps are as follows:

[0098] S201: Calculate the bounding box of the point cloud and divide it into cubic grids with side lengths;

[0099] S202: Assign each point (x i , y i , z i ) to the corresponding voxel (i, j, k), where:

[0100]

[0101] where x min , y min , z min is the minimum coordinate of the bounding box;

[0102] S203: For each voxel, calculate the centroid of all points in it as the representative point:

[0103]

[0104] Where: N is the number of points in the voxel.

[0105] The visualized point cloud after region filtering is as Figure 4 shown.

[0106] S1313: Use the uniform sampling algorithm to sample based on a uniform probability distribution to achieve the purpose of reducing the data volume. After uniform sampling, the data volume contained in the point cloud will be greatly reduced, reducing unnecessary repeated calculations and improving the calculation efficiency.

[0107] S1315: Perform statistical filtering on the point cloud after uniform sampling. The principle is: perform statistical analysis on the points in the neighborhood of each point and calculate the average distance from it to all neighboring points. If the result obtained is a Gaussian distribution, whose shape is determined by the mean and standard deviation, then the points whose average distance is outside the standard range (defined by the global average distance and variance) can be defined as outliers and removed from the data. Statistical filtering can help eliminate outliers and reduce noise interference. In this problem, it can greatly reduce the noise impact caused by environmental factors and camera accuracy factors.

[0108] Specifically, the filtering includes:

[0109] S204: For each point p in the point cloud i , determine its set N of k nearest neighbor points i ;

[0110] S205: Calculate the local density ρ i of point p i , defined as the ratio of the number of points in its neighborhood to the neighborhood volume:

[0111]

[0112] Where V i is the volume of the smallest sphere centered at p i and containing its k neighbors;

[0113] S206: Calculate the average distance d i from point p i to each point in its neighborhood:

[0114]

[0115] Where, (p i , p j ) represents the Euclidean distance between point p i and p j ;

[0116] S207: Calculate the global mean μ and standard deviation σ of the average distances of all points:

[0117]

[0118] where n is the total number of points in the point cloud;

[0119] S208: According to the local density ρ i of point p i , set the adaptive threshold τ i :

[0120]

[0121] where α and β are adjustment parameters, and ρ max is the maximum local density in the point cloud;

[0122] S208: If the average distance i of point p exceeds its adaptive threshold τ i , then mark it as an outlier and remove it.

[0123] S1317: Use the NormalEstimation function to estimate the normal vectors for the filtered point cloud. The normal vector information will correspond one by one to the information of the points in the point cloud. The visualized point cloud of the normal vectors solved after uniform sampling and statistical filtering is as Figure 5 shown.

[0124] S105: According to the extracted normal vector information, use the BoundaryEstimation function to estimate and extract the boundary function. The specific process is as Figure 6 shown:

[0125] S1511: Let p be the point to be detected, and n p be the normal vector of this point, and v be the viewing direction vector (generally, it can be the vector from the viewpoint to point p).

[0126] S1513: The judgment formula is:

[0127]

[0128] where: n p ·v is the dot product of the normal vector and the viewing direction vector, and ||n p ||, ||v|| are their respective magnitudes.

[0129] S1515: Judge the size relationship between the included angle θ and the given threshold θ threshold .

[0130] S1517: When θ > θ threshold , this point is a boundary point. When θ < θthreshold , then this point is not a boundary point.

[0131] After performing boundary point cloud estimation calculations for all points, the extracted boundary point cloud can be obtained. As Figure 7 shown, next, feature fitting will be performed on this boundary point cloud.

[0132] S107: Use the RANSAC algorithm combined with the normal vector to fit a three-dimensional circle model to the boundary point cloud, obtain the center, radius, and axis information of the three-dimensional circle, extract the inliers fitted, and repeat the above model fitting and point cloud removal process for the remaining points to gradually extract multiple three-dimensional circle structures in the point cloud. That is the required feature point cloud. The RANSAC algorithm process is as follows:

[0133] (1) Initialize parameters: Set the parameters of the model, such as the maximum number of iterations N, the distance threshold d, and the consistency score threshold T.

[0134] (2) Random sampling: Randomly select the minimum number of points from the data to fit the model. For example, fitting a three-dimensional circle requires 3 points.

[0135] (3) Model fitting: Construct a model based on the sampled points in the random sampling and calculate the parameters of the model.

[0136] (4) Consistency check: Calculate the distance from each data point to the model. If the distance is less than the threshold d, then this point is considered consistent with the model.

[0137] (5) Verification and update: Calculate the number of consistent data points. If the number exceeds the current best number of consistent points, then update the best model parameters.

[0138] (6) Iteration termination: Repeat the above steps until the maximum number of iterations is reached or the score of the model meets the requirements.

[0139] In this problem, for the extracted boundary points, each point has three-dimensional coordinates (x i , y i , z i ). The goal is to find a three-dimensional circle model that contains the most inliers. Randomly sampling three points can determine the center (x c , y c , z c ) and the radius r. The center coordinates and radius can be calculated using the least squares method or geometric relationships. The distance calculation method is:

[0140]

[0141] Then calculate the distance from each data point to the model. If the distance is less than the threshold d, the point is considered to be consistent with the model. After calculating the consistency of all points, determine whether to update the optimal model parameters according to the number of consistent data points.

[0142] Specifically, randomly select three points n a , n b , n c from the point cloud, and require the included angles of the normal vectors of these three points to be within a certain range:

[0143] ∠(n a , n b ) < θ max ∠(n b , n c ) < θ max

[0144] where θ max is the preset maximum included angle threshold;

[0145] According to the three selected points, calculate the center C, radius R, and normal vector n of the fitted circle;

[0146] Let the coordinates of the three points be p a = (x a , y a , z a ), p b = (x b , y b , z b ), p c = (x c , y c , z c ). Then the parameters of the fitted circle are:

[0147]

[0148] The center C, radius R, and normal vector n can be obtained by solving.

[0149] Calculate the distance from each point to the fitted circle, and determine whether it is an inlier according to the adaptive threshold. The distance d i from the point p i to the fitted circle is:

[0150] d i = ||p i - C|| - R|

[0151] The adaptive distance threshold is:

[0152]

[0153] where ρmax is the maximum local density of the point cloud, d 0 is the initial distance threshold;

[0154] If at this time, then it is determined that p i is an inlier.

[0155] In this example, the results obtained by fitting multiple three-dimensional circles are as Figure 8 shown.

[0156] S109: First, filter the circles fitted in the previous step according to the radius size to obtain the accurate center, radius, and axis information of the small circles on the large circle circumference. Extract the center of the fitted small circle as the new three-dimensional circle circumference point cloud. Fit the three-dimensional circle to obtain the center, radius, and axis information of the large circle with distributed small circular holes. After visualizing the large circle, the point cloud image as Figure 9 shown is obtained.

[0157] The obtained center and radius data of the large circle can play an important role in the positioning of the aero-engine power turbine rotor. Because coordinate transformation was performed previously, the coordinate information obtained now is in the robot base coordinate system. In the control algorithm, the three-dimensional information of the center can be input into the end position of the robot to control the robot to pick up the turbine rotor for assembly.

[0158] The position of the finally identified three-dimensional circle visualization information in the original point cloud is shown as Figure 10 shown.

[0159] So far, this embodiment has solved the problem of aero-engine power turbine rotor feature recognition using point cloud processing technology. By loading a point cloud camera at the end fixture of an industrial robot to collect point cloud data of the power turbine rotor, and using point cloud related preprocessing algorithms such as uniform sampling, boundary estimation algorithms, and model fitting algorithms such as RANSAC three-dimensional circle fitting, the task of surface hole feature recognition and positioning of the power turbine rotor has been finally achieved. This example can be used for the aero-engine robot automatic assembly task. By accurately identifying and positioning the complex geometric features of the aero-engine power turbine rotor, the assembly accuracy and efficiency are improved. Compared with the prior art, the present invention reduces human errors, lowers production costs, and improves production safety. Through accurate feature recognition and robot control technology, efficient and accurate automatic assembly is achieved.

[0160] Embodiment 2

[0161] In another embodiment of the present invention, an aero-engine turbine rotor robot assembly point cloud feature recognition system is provided, including:

[0162] A coordinate transformation module, which is used to calculate the rotation matrix of the point cloud according to the coordinate pose of the robot end during shooting and the robot hand-eye relationship, so as to transfer the point cloud coordinates from the camera coordinate system to the robot base coordinate system;

[0163] A point cloud processing module, which is used to screen the transformed point cloud data according to the Z-axis, extract the first point cloud within the height range where the point cloud features are located, sample and filter the extracted first point cloud, and solve the normal vector information of the processed first point cloud;

[0164] A boundary point judgment module, which is used to establish a boundary point judgment model according to the solved normal vector information and judge whether the point to be monitored is a boundary point, until the second point cloud is extracted after all points are judged for boundary point cloud;

[0165] A three-dimensional circle model fitting module, which is used to fit the second point cloud combined with the normal vector information to obtain the center, radius and axis information of the three-dimensional circle, extract the inliers fitted, and repeat the above model fitting and point cloud rejection steps for the remaining points to gradually extract multiple three-dimensional circle structures in the point cloud;

[0166] A point cloud feature image module, which is used to screen the three-dimensional circle structures fitted in the previous step according to the radius size to obtain the center, radius and axis information of the small circles accurately on the large circumference, extract the center of the fitted small circles as the new three-dimensional circle circumferential point cloud, fit the three-dimensional circle to obtain the center, radius and axis information of the large circle with distributed small holes, and obtain the point cloud feature image of the aero-engine turbine rotor robot assembly after visualizing the large circle.

[0167] This application also provides a terminal device. Please refer to Figure 11 and the terminal device may include:

[0168] A memory, which is used to store computer programs;

[0169] A processor, which can implement the steps of any one of the above aero-engine power turbine rotor robot assembly point cloud feature recognition methods when executing the computer program.

[0170] As Figure 11 shown, it is a schematic structural diagram of the terminal device. The terminal device may include: a processor 10, a memory 11, a communication interface 12 and a communication bus 13. The processor 10, the memory 11 and the communication interface 12 all complete communication with each other through the communication bus 13.

[0171] In the embodiments of this application, the processor 10 may be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic devices, etc.

[0172] The processor 10 can call the program stored in the memory 11. Specifically, the processor 10 can execute the operations in the embodiments of the abnormal IP recognition method.

[0173] The memory 11 is used to store one or more programs. The program can include program codes, and the program codes include computer operation instructions. In the embodiments of the present application, the memory 11 stores at least a program for implementing the following functions:

[0174] When the drawing software is started, the drawing terminal obtains the login usage information input by the user;

[0175] Send the login usage information to the cloud computing platform so that the cloud computing platform can perform permission verification on the login usage information and send a verification passed message when the permission verification is passed;

[0176] When receiving the verification passed message, send a normal enable message;

[0177] Send project query information to the project center of the cloud computing platform so that the project center can send the corresponding project list to the drawing software;

[0178] Receive the project list and perform drawing data exchange with the data exchange center of the cloud computing platform based on the project list;

[0179] When the drawing is completed, drawing result data is obtained.

[0180] In a possible implementation, the memory 11 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function, etc.; the data storage area can store the data created during use.

[0181] In addition, the memory 11 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device or other volatile solid-state storage devices.

[0182] The communication interface 12 can be the interface of a communication module, used to connect to other devices or systems.

[0183] Of course, it should be noted that Figure 3 the shown structure does not limit the terminal device in the embodiments of the present application. In actual applications, the terminal device may include more or fewer components than Figure 3 the shown ones, or combine some components.

[0184] The present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above data processing methods can be implemented.

[0185] The computer-readable storage medium may include: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0186] For the introduction of the computer-readable storage medium provided by the present application, please refer to the above method embodiments, and the present application will not elaborate herein.

[0187] The embodiments in the specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description in the method part.

[0188] Those skilled in the art can further realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0189] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be directly implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0190] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying point cloud features of an aircraft engine turbine rotor robot assembly, characterized in that: include: S100: Calculate the rotation matrix of the point cloud according to the coordinate position of the robot end during shooting and the hand-eye relationship of the robot, so that the point cloud coordinates are transferred from the camera coordinate system to the robot base coordinate system; S200: filtering the transformed point cloud data according to the Z axis, extracting the first point cloud within the height range where the point cloud feature is located, sampling and filtering the extracted first point cloud, and solving the normal vector information of the processed first point cloud; S300: establishing a boundary point judgment model according to the solved normal vector information and judging whether the point to be monitored is a boundary point, until all points are subjected to boundary point cloud judgment to obtain a second point cloud extracted; S400: fitting a three-dimensional circle model on the second point cloud in combination with the normal vector information to obtain the center, radius and axis information of the three-dimensional circle, extracting the fitted inner points, repeating the above model fitting and point cloud elimination steps for the remaining points, and gradually extracting multiple three-dimensional circle structures in the point cloud; S5 00: First, filter the 3D circle structure fitted in the previous step according to the radius size to obtain the center, radius and axis information of the small circle accurately on the large circumference, extract the center of the fitted small circle as the new 3D circle circumference point cloud, fit the 3D circle to obtain the center, radius and axis information of the large circle with distributed small circular holes, and visualize the large circle to obtain the point cloud feature image of the aircraft engine turbine rotor robot assembly.

2. The method for identifying point cloud features of an aircraft engine turbine rotor robot assembly according to claim 1, characterized in that: In step S200, the sampling is performed by dividing the space into voxels of fixed size and selecting a representative point in each voxel to approximate the original point cloud. The specific steps are as follows: S201: Calculate the bounding box of the point cloud and divide it into cubic grids with side lengths of 1. S202: For each point (x i ,y i ,z i ) is assigned to the corresponding voxel (i,j,k), where: Among them, x min ,y min ,z min is the minimum coordinate of the bounding box; S203: For each voxel, calculate the centroid of all points therein as the representative point: Where: N is the number of points in the voxel.

3. The method for identifying point cloud features of an aircraft engine turbine rotor robot assembly according to claim 2, characterized in that: In step S200, the filtering includes: S204: For each point p in the point cloud i , determine its k nearest neighbor point set N i ; S205: Calculate point p i The local density ρ i , defined as the ratio of the number of points in its neighborhood to the volume of the neighborhood: Among them, V i For p i The minimum spherical volume with centered on , including its k neighbors; S206: Calculate point p i The average distance d to each point in its neighborhood i : Among them, (p i ,p j ) represents point p i and p j The Euclidean distance between S207: Calculate the global mean μ and standard deviation σ of the average distance of all points: Where n is the total number of points in the point cloud; S208: According to point p i The local density ρ i , set the adaptive threshold τ i : Among them, α and β are adjustment parameters, ρ max is the maximum local density in the point cloud; S208: If point p i The average distance Exceeding its adaptive threshold τ i , it is marked as an outlier and removed.

4. A method for identifying point cloud features of an aircraft engine turbine rotor robot assembly according to any one of claims 1 to 3, characterized in that: The determination of the boundary points in step S300 includes: S301: Let p be the point to be detected, n p is the normal vector of the point, and v is the viewing direction vector; S302: Calculate the normal vector n of the point to be detected p The angle θ with the viewing direction vector v: Where: n p v is the dot product of the normal vector and the viewing direction vector, ||n p ||,||v|| are their respective modulus lengths; S303: Determine whether θ is equal to a given threshold value θ threshold Size relationship: When θ>θ threshold , this point is a boundary point, when θ<θ threshold , then the point is not a boundary point.

5. The method for identifying point cloud features of an aircraft engine turbine rotor assembly robot according to claim 3, characterized in that: In step S400, the three-dimensional circle model fitting includes: S401: based on the normal vector information of the point cloud, select points with similar normal vector directions for sampling to increase the probability that the sampling points belong to the same circle; S402: dynamically adjusting the distance threshold according to the local density and noise level of the point cloud to more effectively distinguish the inner points from the outer points and achieve model fitting; S403: Introduce multi-scale analysis methods to verify the fitting results at multiple levels to improve the robustness of the model.

6. The method for identifying point cloud features of an aircraft engine turbine rotor robot assembly according to claim 5, characterized in that: Step S401 includes: Randomly select three points n from the point cloud a 、n b 、n c , requiring the normal vector angles of these three points to be within a certain range: ∠(n,n b )<θ max ,∠(n b ,n c )<θ max Among them, θ max is the preset maximum angle threshold; According to the three selected points, calculate the center C, radius R and normal vector n of the fitting circle; Assume the coordinates of the three points are p a =(x a ,y a 、z a ), p b =(x b ,y b 、z b ), p c =(x c ,y c 、z c ), then the parameters of the fitted circle are: By solving, we can obtain the center C, radius R and normal vector n.

7. The method for identifying point cloud features of an aircraft engine turbine rotor robot assembly according to claim 6, characterized in that: The step S402 of distinguishing the inner points from the outer points includes: Calculate the distance from each point to the fitted circle and determine whether it is an inner point based on the adaptive threshold. i The distance d to the fitted circle i for: d i =|p i -C||-R| Adaptive distance threshold for: Among them, ρ max is the maximum local density of the point cloud, d0 is the initial distance threshold; like When p i For the inner point.

8. An aero-engine turbine rotor robot assembly point cloud feature recognition system, characterized in that: include: The coordinate conversion module is used to calculate the rotation matrix of the point cloud according to the coordinate position of the robot end and the robot hand-eye relationship during shooting, so that the point cloud coordinates are transferred from the camera coordinate system to the robot base coordinate system; The point cloud preprocessing module is used to filter the transformed point cloud data according to the Z axis, extract the first point cloud within the height range where the point cloud feature is located, sample and filter the extracted first point cloud, and solve the normal vector information of the processed first point cloud; The boundary point judgment module is used to establish a boundary point judgment model according to the solved normal vector information, and judge whether the point to be monitored is a boundary point, until all points are subjected to boundary point cloud judgment to obtain the extracted second point cloud; A three-dimensional circle model fitting module is used to fit the three-dimensional circle model to the second point cloud in combination with the normal vector information, obtain the center, radius and axis information of the three-dimensional circle, extract the fitted inner points, repeat the above model fitting and point cloud elimination steps for the remaining points, and gradually extract multiple three-dimensional circle structures in the point cloud; The point cloud feature image module is used to first screen the three-dimensional circle structure fitted in the previous step according to the radius size, obtain the center, radius and axis information of the small circle accurately on the large circle, extract the fitted small circle center as the new three-dimensional circle circumference point cloud, fit the three-dimensional circle to obtain the center, radius and axis information of the large circle with distributed small circular holes, and visualize the large circle to obtain the point cloud feature image of the aircraft engine turbine rotor robot assembly.

9. A terminal device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the method for robotic assembly of aero-engine turbine rotors as claimed in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for robotic assembly of aero-engine turbine rotors as claimed in any one of claims 1 to 7.

Images