Trajectory feature extraction and automatic sorting method for aviation sealing gluing

Abstract

The invention belongs to the technical field of automatic gluing and sealing of large structural parts in the field of aeronautical manufacturing, and particularly relates to a track feature extraction and automatic sorting method for aeronautical sealing gluing. The track feature extraction and automatic sorting method for aviation sealing gluing comprises the following steps: scanning aviation wing box test pieces through a line laser scanner, and collecting point cloud data; filtering the point cloud data by using an outlier filtering algorithm based on statistics; segmenting the filtered point cloud data by adopting an improved random sampling consistency algorithm to obtain an effective plane set; feature lines are extracted from the effective plane set; obtaining a multi-plane intersection point by using a matrix decomposition method; and orderly arranging the starting point, the middle point and the end point of the intersection line segment through a parameterized projection mechanism. According to the method, automatic sequencing and planning of the gluing paths are realized by utilizing an automatic trajectory optimization algorithm, so that the automation level and the operation efficiency of the sealing gluing process in the field of aeronautical manufacturing are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic gluing and sealing of large structural parts in the field of aviation manufacturing, and particularly relates to a trajectory feature extraction and automatic sorting method for aviation sealing gluing. Background Art

[0002] In the field of robotic automated gluing of high-precision aviation structural parts such as wing tank seals, the core challenge currently faced lies in the millimeter-level geometric deviation between the actual assembly and the design model. During the manufacturing process, the actual curved surface morphology of the wing tank seal structure often deviates from the theoretical CAD model due to deformation of the composite material, stress release of the skin riveting, and cumulative errors in the assembly of multiple components. This causes the glue gun posture to be inaccurate in the traditional robotic gluing system with a preset trajectory. Especially in areas such as the seam of the wing leading edge tank skin, this deviation will cause the glue gun's glue discharge angle to mismatch with the surface normal, resulting in process failures such as excessive glue layer thickness (more than ±0.5mm) and misalignment of the sealant and titanium alloy reinforcement, seriously affecting the aircraft's airtightness indicators.

[0003] Secondly, in the field of aviation manufacturing, with the frequent iterations and upgrades of aircraft structures and the increasing complexity of sealing trajectories (for example, the surge in the number of trajectories due to differences in fuel tank topology across aircraft models), traditional teaching trajectory correction methods that rely on manual experience have limitations. Although manual correction can compensate for machining deviations locally, the efficiency and consistency of manual correction cannot meet production requirements given the dynamically changing configuration requirements and the vast number of differentiated trajectories (such as the spatial distribution and characteristic differences of sealing paths between different aircraft models). Furthermore, the subjectivity of manual operation leads to fluctuations in trajectory quality, making it difficult to ensure process stability and beat control requirements in large-scale production. Summary of the Invention

[0004] The purpose of the present invention is to provide a trajectory feature extraction and automatic sorting method for aviation sealant coating. By integrating three-dimensional point cloud preprocessing technology, geometric feature extraction algorithms, and trajectory automatic sorting mechanisms, it effectively addresses the problem of loss of accuracy in robot offline programming caused by shape and position deviations in the coating of large aviation components (such as wing fuel tanks). This method also addresses the pain point of the low efficiency of traditional manual teaching methods in complex multi-track gluing operations, and utilizes an automated trajectory optimization algorithm to achieve automatic sorting and planning of gluing paths, thereby significantly improving the automation level and operating efficiency of the sealant coating process in the aviation manufacturing field.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0006] The trajectory feature extraction and automatic sorting method for aviation sealant coating includes the following steps:

[0007] 1) Scan the aviation wing box specimen using a line laser scanner to collect point cloud data;

[0008] 2) Use the statistical outlier filtering algorithm to filter the point cloud data;

[0009] 3) Using the improved random sampling consensus algorithm to segment the filtered point cloud data to obtain a valid plane set;

[0010] 4) Extract characteristic lines from the valid plane set;

[0011] 5) Use matrix decomposition method to obtain multi-plane intersection points;

[0012] 6) The starting point, middle point and end point of the intersection segment are arranged in order through the parametric projection mechanism.

[0013] The step 2) is specifically as follows:

[0014] Calculate each point p in the point cloud data i The average Euclidean distance d to its k nearest neighbors i , construct a global statistical distribution model and calculate the mean μ and standard deviation σ:

[0015]

[0016] Where N represents the total number of all points in the point cloud data;

[0017] Set the dynamic threshold T = μ + 3σ to remove outliers that do not meet the threshold.

[0018] The step 3) is specifically as follows:

[0019] The filtered point cloud is defined as the remaining point cloud set Q. Multi-planes are dynamically extracted through a random sampling mechanism in each iteration. In each iteration, three non-collinear points are randomly selected to construct a candidate plane equation ax+by+cz+d=0, and the Euclidean distance D from all points in the point cloud to the plane is calculated:

[0020]

[0021] Set the threshold distance D1 and count the number of inliers that meet the threshold condition. If the number of inliers exceeds 1000, it is determined to be a valid plane, extract the corresponding point cloud and update the remaining point cloud set Q. When the remaining point cloud scale |Q| < 1000, the iteration is automatically terminated to obtain the valid plane set.

[0022] The step 4) comprises the following steps:

[0023] 4.1) Based on the normal vectors n1 and n2 of the two planes, determine the intersection direction vector d through cross product operation:

[0024] d = n1 × n2;

[0025] 4.2) Solve the overdetermined system of equations consisting of the plane equations and directional constraints, and use the least squares method to optimize the coordinates of p0:

[0026]

[0027] Where p represents the coordinates of the point to be solved on the intersection line, d1 and d2 represent the constant terms of the two plane equations, reflecting the offset of the plane from the origin;

[0028] 4.3) Determine whether there is an actual intersection line segment between the two planes through the parameterized constraint mechanism of the actual distribution range of the point cloud.

[0029] The step 4.3) is specifically as follows:

[0030] For each plane point cloud, the vertex set representing the plane boundary is extracted by the three-dimensional convex hull algorithm {v i}, and project each vertex along the intersection direction d into the one-dimensional parameter space, vertex v i The projection parameter t i for:

[0031]

[0032] The projection parameters of all vertices are traversed by linear scanning {t i}, record the minimum projection value t of the two planes respectively min and the maximum projection value t max , forming the effective parameter interval t1∈[t 1,min ,t 1,max ] and t2∈[t 2,min ,t 2,max ];

[0033] Determine the effective parameter range of the intersection segment t by the intersection of the two plane parameter intervals intersext :

[0034] t intersect =[max(t 1,min ,t 2,min ),min(t 1,max ,t 2,max )]

[0035] If and only if max(t 1,min ,t 2,min )≤min(t 1,max ,t 2,max ), it is determined that there is an actual intersection line segment between the two planes, corresponding to the actual overlapping area of the point cloud; otherwise, it is considered that there is no valid intersection line between the two planes.

[0036] The step 5) is specifically as follows:

[0037] The plane equations of the three planes π1, π2, and π3 are combined to construct a linear equation system:

[0038]

[0039] The numerical stability of the coefficient matrix A is optimized by the column pivot QR decomposition method to solve x = A -1 b, where feature point x = [x, y, z] T ;

[0040] The matrix condition number cond(A) is introduced as the criterion for plane geometric relations: when cond(A)>10 6 When the three planes are determined to be coplanar or approximately parallel, the calculation is automatically terminated.

[0041] The step 6) is specifically as follows:

[0042] Based on the intersection direction vector d determined by the cross product of the plane normal vector and the endpoints of the parameter interval, the intersection line parameterization equation is defined as:

[0043] p(t)=p0+t·d

[0044] Where t represents the projection parameter value, t∈R;

[0045] Map the three-dimensional space points to the one-dimensional parameter space along the intersection line direction. The starting point and end point of the intersection line segment correspond to the endpoints of the parameter interval t intersect,min and t intersect,max , that is, the starting point is p0+t intersect,min d, the end point is p0+t intersect,max d. All intermediate points are sorted in ascending order according to their projection parameter values t. For points with the same projection parameter values, they are further sorted based on the direction of the local normal vector, or they are identified as duplicate points based on the point cloud density threshold and merged, ultimately achieving an unambiguous and continuous arrangement of trajectory points.

[0046] The present invention has the following beneficial effects and advantages:

[0047] The present invention achieves efficient feature extraction of aviation sealant seams by constructing a horizontal-vertical dual-plane spatial analysis model, combining a three-dimensional point cloud feature analysis algorithm with deep integration of the PCL point cloud library. Based on the geometric principle of surface intersection lines, the reference trajectory is accurately generated, and relying on the automatic sorting algorithm of feature intersection lines, a continuous and breakpoint-free processing path topology network is automatically constructed. This method effectively suppresses the accumulation of trajectory fitting errors through algorithm optimization, while increasing the efficiency of traditional manual teaching by more than three times through fully automatic trajectory planning. Ultimately, it outputs a high-quality gluing path that is deeply coupled with the robot's kinematic parameters, significantly improving the robot's motion stability and consistency during continuous operation while ensuring the linear position accuracy of the sealant. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the process of the present invention;

[0049] Figure 2 The point cloud intent in the ASC format of the present invention;

[0050] Figure 3 This is a schematic diagram of the model file in STL format of the present invention;

[0051] Figure 4 This is a schematic diagram of converting an STL format file to a TXT format file according to the present invention;

[0052] Figure 5 This is a schematic diagram of the statistical filtering process performed on the point cloud model according to the present invention;

[0053] Figure 6 This is a schematic diagram of multi-plane segmentation of a point cloud model according to the present invention;

[0054] Figure 7 This is a schematic diagram of the intersection line obtained by intersecting the surface equations of the present invention;

[0055] Figure 8 The convex hull feature description and principle diagram of the present invention are as follows;

[0056] Figure 9 Schematic diagram of valid intersection segments obtained by the convex hull algorithm of the present invention;

[0057] Figure 10 This is a schematic diagram of the unique intersection obtained by the intersection of three surfaces in the present invention;

[0058] Figure 11 This is a schematic diagram of the automatic orderly arrangement of intersection segments of the present invention;

[0059] Figure 12 This is a schematic diagram of the trajectory path extracted in MATLAB by the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0061] like Figures 1 to 12 As shown, the present invention integrates the PCL point cloud library based on development platforms such as Visual Studio to complete 3D data preprocessing, and constructs a high-precision 3D model by performing noise reduction, sampling and other optimization processes on the original point cloud of the aviation wing box specimen. The system realizes the automatic extraction of sealant seam features and automatic trajectory sorting, significantly improving the efficiency of feature extraction and trajectory planning, and providing a complete implementation logic framework for the technical details shown in the accompanying drawings. Figure 1 The figure shows the overall flow chart of the present invention, which is described in detail in steps one to six.

[0062] The method of the present invention comprises the steps of:

[0063] Step 1: Point cloud acquisition

[0064] The point cloud data of the present invention is obtained from the scanning and acquisition of the aviation wing box test piece by a line laser scanner, and includes two data formats: ASC and STL. Figure 2 As shown in the figure, since the vertical surface of the ASC format point cloud data is parallel to the laser incident direction, the reflected signal intensity is significantly attenuated, resulting in a significant reduction in the point cloud density in the vertical area. The loss of local point clouds seriously affects the integrity of the geometric features. Figure 3 and Figure 4 As shown in the figure, by adaptively sampling the point cloud reconstruction of the STL mesh model, more than 90% of the original missing area data was successfully restored, and the curvature continuity error of the reconstructed point cloud was less than 0.1mm, which significantly improved the input quality of the feature intersection line extraction algorithm.

[0065] Step 2: Point cloud preprocessing

[0066] The present invention adopts a statistically based outlier filtering algorithm in the point cloud preprocessing stage. The specific process is as follows:

[0067] For the original point cloud data obtained by the line laser scanner, the noise interference is first eliminated based on the Statistical Outlier Removal (SOR) algorithm. The input original point cloud is subjected to noise filtering. The present invention adopts the statistical outlier filtering method to eliminate noise interference. The algorithm calculates the distance distribution characteristics of each point p by analyzing the local neighborhood of the point cloud. i The average Euclidean distance d to its k=50 nearest neighbors i , construct a global statistical distribution model and derive the mean μ and standard deviation σ:

[0068]

[0069] Set the dynamic threshold T = μ + 3σ to remove outliers. Automatically identify and remove outliers that exceed the threshold (such as sensor noise, floating flying points, etc.). Figure 5 As shown in the figure, the main structural integrity of the point cloud is preserved after preprocessing, and the surface smoothness is significantly improved, providing high-confidence data input for subsequent feature extraction.

[0070] Step 3: Point cloud segmentation

[0071] The present invention adopts an improved RANSAC (Random Sample Consensus) algorithm in the multi-plane segmentation process to realize the automatic plane segmentation and extraction of complex aerospace structure point clouds. The specific process is as follows:

[0072] The algorithm initializes the preprocessed point cloud as the remaining point cloud set Q, and dynamically extracts multiple planes through a random sampling mechanism in each iteration. In each iteration, the algorithm randomly selects three non-collinear points to construct a candidate plane equation ax+by+cz+d=0 and calculates the Euclidean distance from all points in the point cloud to the plane:

[0073]

[0074] Set the threshold distance D = 0.3mm and count the number of inliers that meet the conditions. If the number of inliers exceeds 1000, it is determined to be a valid plane (such as Figure 6 The corresponding point cloud is extracted and the remaining point cloud set Q is updated. The iteration is automatically terminated when the remaining point cloud size |Q| is less than 1000, avoiding the over-segmentation problem caused by the fixed number of iterations in the traditional RANSAC algorithm and improving computational efficiency.

[0075] Step 4: Feature line extraction

[0076] like Figure 7 As shown in the figure, the present invention combines geometric constraints with parametric projection in the plane intersection calculation to ensure that the intersection segment is precisely confined to the actual overlapping area of the two plane point clouds. First, based on the normal vectors n1 and n2 of the two planes π1 and π2, the intersection direction vector is determined by cross product operation:

[0077] d=n1×n2 (4)

[0078] To determine the intersection base point p0, it is necessary to solve an overdetermined set of equations consisting of plane equations and direction constraints:

[0079]

[0080] The first two equations correspond to plane equations, and the third equation constrains the base point to lie along the intersection direction d. Because the actual point cloud data size is much larger than the number of equations (e.g., hundreds of points corresponding to three equations), the least squares method is used to optimize the coordinates of p0 to ensure the solution is robust to noise.

[0081] In order to avoid the error caused by the infinite extension of the intersection line, the present invention introduces a parameterized constraint mechanism for the actual distribution range of the point cloud. For each plane point cloud, the vertex set {v i}. And project each vertex along the intersection direction d into the one-dimensional parameter space. Specifically, the projection parameter of vertex vi is calculated by the formula:

[0082]

[0083] This parameter reflects the relative position of the vertex in the direction of the intersection line, such as Figure 8 shown.

[0084] The projection parameters of all vertices are traversed by linear scanning {t i}, record the minimum projection value t of the two planes respectively min and the maximum projection value t max , forming the effective parameter interval t1∈[t 1,min ,t 1,max ] and t2∈[t 2,min ,t 2,max ]. The time complexity of the extreme value search process is O(n), which avoids the computational redundancy of global sorting and improves efficiency.

[0085] The effective parameter range of the final intersection segment is determined by the intersection of the two plane parameter intervals:

[0086] t intersect =[max(t 1,min ,t 2,min ),min(t 1,max ,t 2,max )] (7)

[0087] If and only if max(t 1,min ,t 2,min )≤min(t 1,max ,t 2,max ), it is determined that there is an actual intersection line segment between the two planes, corresponding to the actual overlapping area of the point cloud; otherwise, it is considered that there is no valid intersection line between the two planes. Figure 9 As shown in the figure, this method transforms the three-dimensional geometric constraint problem into an efficient numerical operation through parametric projection and one-dimensional extreme value comparison. It not only avoids the defect of infinite extension of the intersection line in the traditional method, but also ensures that the intersection line is strictly limited to the coverage range of the point cloud, while taking into account both calculation accuracy and efficiency.

[0088] Step 5: Feature point extraction

[0089] This paper uses matrix decomposition to improve computational robustness in solving multi-plane intersections. To solve the intersection problem of three planes π1, π2, and π3, we first combine their plane equations to construct a linear system of equations:

[0090]

[0091] The numerical stability of the coefficient matrix A is optimized by the column pivot QR decomposition method to solve x = A -1 b. On this basis, the matrix condition number cond(A) is introduced as the criterion for plane geometric relations: when cond(A)>10 6 When , the three planes are judged to be coplanar or approximately parallel, and the calculation is automatically terminated to avoid ill-conditioned solutions. Figure 10As shown in the figure, this method can accurately determine the true intersection point (green circle) between the bottom plane and the two vertical planes, with a coordinate error within ±0.15 mm, effectively reflecting the spatial pose characteristics of the corners of aviation structures. By combining matrix decomposition with condition number threshold control, the method significantly reduces the misjudgment rate caused by sensor noise or missing point clouds while retaining valid intersection points.

[0092] Step 6: Automatic sorting

[0093] like Figure 11 As shown in Figure 2, the starting point, middle point, and end point of the intersection segment are arranged in an orderly manner through a parametric projection mechanism. Based on the intersection direction vector d (determined by the cross product of the plane normal vector) and the endpoints of the parameter interval, the intersection parameter equation is defined as:

[0094] p(t)=p0+t·d(t∈R) (9)

[0095] Map the three-dimensional space points to the one-dimensional parameter space along the intersection line. The starting point and end point of the intersection line segment correspond to the endpoints of the parameter interval t intersect,min and t intersect,max , that is, the starting point is p0+t intersect,min d, the end point is p0+t intersect,max d. All intermediate points are arranged in ascending order according to their projection parameter value t, ensuring that the point sequence strictly follows the natural extension direction of the intersection line. Figure 12 For points with the same projection parameter values (such as collinear points or dense point groups), secondary sorting is further performed based on the local normal vector direction, or duplicate points are determined by the point cloud density threshold and merged, ultimately achieving unambiguous continuous arrangement of trajectory points.

[0096] The technical solution of this invention uses the aforementioned point cloud processing flow to efficiently extract and automatically sort glue seam features. The resulting continuous processing path can be directly transmitted to the robot control system. Based on this path and incorporating glue coating process parameters (such as coating speed and glue output), the robot performs kinematic planning to accurately execute automated sealing and gluing operations on complex aerospace surfaces.

[0097] The algorithm characteristics and advantages of the present invention are further explained.

[0098] The core advantage of this algorithm lies in the efficiency and robustness of geometric calculations. By projecting three-dimensional geometric relationships into one-dimensional parameter space and combining it with an extreme value search algorithm (time complexity O(n)), the computational complexity of large-scale point cloud data is significantly reduced. At the same time, the range of the intersection segment is determined based on the projection parameters of the convex hull vertices, strictly ensuring that the intersection line only exists in the actual overlapping area of the two plane point clouds, completely eliminating the positioning deviation caused by the infinite extension of the intersection line in the traditional method, and the geometric calculation accuracy can reach ±0.05mm. This design not only improves the fault tolerance for noise points and locally missing points, but also achieves seamless adaptation to the robot kinematic model through a parameterized mechanism, ultimately achieving highly consistent and repeatable automated gluing operations on complex aviation structural parts.

Claims

1. Trajectory feature extraction and automatic sorting method for aviation sealing glue coating, characterized by: The following steps are involved: 1) Scan the aviation wing box specimen using a line laser scanner to collect point cloud data; 2) Use the statistical outlier filtering algorithm to filter the point cloud data; 3) Using the improved random sampling consensus algorithm to segment the filtered point cloud data to obtain a valid plane set; 4) Extract characteristic lines from the valid plane set; 5) Use matrix decomposition method to obtain multi-plane intersection points; 6) The starting point, middle point and end point of the intersection segment are arranged in order through the parametric projection mechanism.

2. The trajectory feature extraction and automatic sorting method for aviation sealant coating according to claim 1 is characterized in that: The step 2) is specifically as follows: Calculate each point p in the point cloud data i The average Euclidean distance d to its k nearest neighbors i , construct a global statistical distribution model and calculate the mean μ and standard deviation σ: Where N represents the total number of all points in the point cloud data; Set the dynamic threshold T = μ + 3σ to remove outliers that do not meet the threshold.

3. The trajectory feature extraction and automatic sorting method for aviation sealant coating according to claim 1 is characterized in that: The step 3) is specifically as follows: The filtered point cloud is defined as the remaining point cloud set Q. Multi-planes are dynamically extracted through a random sampling mechanism in each iteration. In each iteration, three non-collinear points are randomly selected to construct a candidate plane equation ax+by+cz+d=0, and the Euclidean distance D from all points in the point cloud to the plane is calculated: Set the threshold distance D1 and count the number of inliers that meet the threshold condition. If the number of inliers exceeds 1000, it is determined to be a valid plane, extract the corresponding point cloud and update the remaining point cloud set Q. When the remaining point cloud scale |Q| < 1000, the iteration is automatically terminated to obtain the valid plane set.

4. The trajectory feature extraction and automatic sorting method for aviation sealant coating according to claim 1 is characterized in that: The step 4) comprises the following steps: 4.1) Based on the normal vectors n1 and n2 of the two planes, determine the intersection direction vector d through cross product operation: d = n1 × n2; 4.2) Solve the overdetermined system of equations consisting of the plane equations and directional constraints, and use the least squares method to optimize the coordinates of p0: Where p represents the coordinates of the point to be solved on the intersection line, d1 and d2 represent the constant terms of the two plane equations, reflecting the offset of the plane from the origin; 4.3) Determine whether there is an actual intersection line segment between the two planes through the parameterized constraint mechanism of the actual distribution range of the point cloud.

5. The trajectory feature extraction and automatic sorting method for aviation sealant coating according to claim 4 is characterized in that: The step 4.3) is specifically as follows: For each plane point cloud, the vertex set representing the plane boundary is extracted by the three-dimensional convex hull algorithm {v i }, and project each vertex along the intersection direction d into the one-dimensional parameter space, vertex v i The projection parameter t i for: The projection parameters of all vertices are traversed by linear scanning {t i }, record the minimum projection value t of the two planes respectively min and the maximum projection value t max , forming the effective parameter interval t1∈[t 1,min ,t 1,max ] and t2∈[t 2,min ,t 2,max ]; Determine the effective parameter range of the intersection segment t by the intersection of the two plane parameter intervals intersect : t intersect =[max(t 1,min ,t 2,min ),min(t 1,max ,t 2,max )] If and only if max(t 1,min ,t 2,min )≤min(t 1,max ,t 2,max ), it is determined that there is an actual intersection line segment between the two planes, corresponding to the actual overlapping area of the point cloud; otherwise, it is considered that there is no valid intersection line between the two planes.

6. The trajectory feature extraction and automatic sorting method for aviation sealant coating according to claim 1 is characterized in that: The step 5) is specifically as follows: The plane equations of the three planes π1, π2, and π3 are combined to construct a linear equation system: The numerical stability of the coefficient matrix A is optimized by the column pivot QR decomposition method to solve x = A -1 b, where feature point x = [x, y, z] T ; The matrix condition number cond(A) is introduced as the criterion for plane geometric relations: when cond(A)>10 6 When the three planes are determined to be coplanar or approximately parallel, the calculation is automatically terminated.

7. The trajectory feature extraction and automatic sorting method for aviation sealant coating according to claim 1 is characterized in that: The step 6) is specifically as follows: Based on the intersection direction vector d determined by the cross product of the plane normal vector and the endpoints of the parameter interval, the intersection line parameterization equation is defined as: p(t)=p0+t·d Where t represents the projection parameter value, t∈R; Map the three-dimensional space points to the one-dimensional parameter space along the intersection line direction. The starting point and end point of the intersection line segment correspond to the endpoints of the parameter interval t intersect,min and t intersect,max , that is, the starting point is p0+t intersect,min d, the end point is p0+t intersect,max d. All intermediate points are sorted in ascending order according to their projection parameter values t. For points with the same projection parameter values, they are further sorted based on the direction of the local normal vector, or they are identified as duplicate points based on the point cloud density threshold and merged, ultimately achieving an unambiguous and continuous arrangement of trajectory points.

Images