Method and system for pose synthesis optimization of aircraft large component butt joint assembly based on measured point cloud model and geometric multi-constraint

Abstract

The application discloses a kind of based on measured point cloud model and geometric multi-constraint aircraft large component butt joint assembly pose synthesis optimization method and system, the pose synthesis optimization system includes three-dimensional scanner, laser tracker, pose adjustment positioner and computer integrated control system, wing and fuselage around frame butt joint surface point cloud are collected by three-dimensional scanner, global coordinate system is completed joint calibration using common target method and IR-SVD method;Key point DETR network is used to extract butt joint surface key point, and matching point pair is established by combining FPFH feature description and bidirectional matching strategy, and initial butt joint pose is solved by SVD;Multi-constraint optimization model with minimum butt joint surface gap square sum as target is constructed, gap tolerance, gap uniformity and pose adjustment control point deviation are fused, penalty function-quasi-newton method is used to iteratively solve optimal target pose, and wing is accurately butted.The application realizes uniform distribution of butt joint surface gap and pose precision collaborative optimization, significantly improves aircraft large component assembly quality and efficiency.

Description

Technical Field

[0001] This invention relates to the field of aircraft digital assembly technology, and in particular to a method and system for comprehensive optimization of the docking and assembly posture of large aircraft components based on measured point cloud models and geometric multi-constraints. Background Technology

[0002] Wing-body docking is a critical step in aircraft manufacturing, and its quality directly affects the aircraft's performance and service life. Traditional wing-body docking methods rely on manual tooling, requiring repeated manual adjustments and trial assembly, resulting in low efficiency, poor precision, and difficulty in guaranteeing assembly quality. With the development of digital assembly technology, automated docking technology for large components, relying on digital measuring equipment and attitude adjustment positioners, can efficiently and accurately complete component docking. Attitude adjustment accuracy directly affects the final assembly quality, and accurately solving the target pose of the component is the core of attitude adjustment. Currently, existing technologies mainly measure the attitude control points on the component and solve the pose accordingly. However, due to manufacturing errors and structural deformations, the attitude control points and docking surfaces of the component deviate significantly from the theoretical model. Using only attitude control points for pose solving will result in inaccurate docking features. Furthermore, the docking surfaces of large components are complex in shape, large in size, and difficult to allocate gaps, making it challenging to accurately solve the target pose to ensure uniform gaps on the docking surfaces.

[0003] Although existing research has yielded some results, it still has the following shortcomings: (1) The pose calculation relies on theoretical numerical models, which may lead to inaccurate optimal pose calculation due to deformation of actual products, resulting in assembly interference or gap deviation. (2) Using only pose control points as docking references makes it difficult for the docking surfaces to fit accurately during the assembly of large aircraft components, requiring analysis of the gap distribution of the docking surfaces. (3) Existing methods lack comprehensive modeling and collaborative optimization of various geometric constraints such as frame gap tolerance, gap uniformity, and attitude control point deviation, making it difficult to achieve reasonable distribution and uniformity of the gap of the docking surfaces while ensuring no interference.

[0004] Therefore, there is an urgent need for a pose optimization method that can be based on the actual shape of the docking surface and comprehensively consider various geometric constraints, so as to improve the uniformity of the docking surface gap, reduce the risk of interference, and improve the docking and assembly quality of large aircraft components. Summary of the Invention

[0005] The purpose of this invention is to solve the technical problems existing in the prior art and to provide a method and system for comprehensive optimization of the docking and assembly posture of large aircraft components based on measured point cloud models and geometric multi-constraints.

[0006] To achieve the above objectives, the technical solution provided by this invention is: a comprehensive optimization system for the attitude of large aircraft component docking and assembly based on a measured point cloud model and geometric multi-constraints. The attitude optimization system includes physical hardware devices and software functional modules. The physical hardware devices include a 3D scanner, a laser tracker, an attitude adjuster, and a computer integrated control system, which provide data support and execution. The software functional modules are embedded in the computer integrated control system and are used to realize data processing, attitude calculation, and automated attitude adjustment control. The 3D scanner is used to collect point cloud data of the mating surface between the aircraft wing frame and the fuselage wing box frame, and send the point cloud data to the computer integrated control system. The laser tracker is used to construct a global coordinate system, measure the coordinates of the wing attitude control points, and calibrate the coordinate system of the 3D scanner with the global coordinate system. The attitude adjustment positioner is fixed to a rigid part of the aircraft wing and connected to the computer integrated control system to adjust the position and attitude of the aircraft wing according to drive commands. The computer-integrated control system is connected to the 3D scanner, laser tracker, and attitude adjuster respectively, and is used to perform attitude optimization and docking control. The software functional modules include: The global coordinate system joint calibration module is used to unify laser tracker data and 3D scanner data. It uses the common target method and the IR-SVD-based method to perform global coordinate system joint calibration and transform the data to the global coordinate system. The point cloud model construction module is used to construct a measured point cloud model of the docking surface between the wing and the fuselage frame using point cloud data collected by a 3D scanner, and to extract key points and solve the initial docking pose. The optimization model building module is used to establish a comprehensive optimization model with the goal of minimizing the sum of squares of the gaps between the docking surfaces, and includes gap tolerance, gap uniformity and attitude control point deviation constraints. The optimization solution module is used to solve the comprehensive optimization model using the initial docking pose as the initial value and the penalty function-quasi-Newton method to obtain the optimal target pose. The attitude adjustment driver module is used to generate drive instructions for the attitude adjustment positioner based on the optimal target pose.

[0007] This invention also provides a method for comprehensive optimization of the pose of large aircraft components during docking and assembly based on a measured point cloud model and geometric multi-constraints. The method is applied to the aforementioned pose comprehensive optimization system and specifically includes the following steps: Step 1: Global coordinate system joint calibration. A global coordinate system calibration model is established using the common target method. The global coordinate system transformation relationship is solved using the IR-SVD method, and the point cloud data collected by the 3D scanner is unified to the global coordinate system established by the laser tracker. Step 2: Construct a measured point cloud model of the wing-body docking frame, including: acquiring point cloud data of the docking surfaces of the aircraft wing frame and fuselage wing box frame using a 3D scanner, and performing point cloud denoising, stitching, and downsampling preprocessing; extracting key points of the point cloud on the docking surfaces of the wing frame and fuselage wing box frame using a KeypointDETR network; establishing matching point pairs between key points of the fuselage and wing frame docking surfaces based on the key points of the point cloud, using geometric feature description and bidirectional matching strategies; and using the matching point pairs, solving the initial docking pose of the wing relative to the fuselage using singular value decomposition. Step 3: Construct a geometrically multi-constraint-based wing-fuselage docking assembly attitude comprehensive optimization model, including: establishing an objective function with the goal of minimizing the sum of squares of the gaps between the matching points of the wing and fuselage frame docking surfaces; establishing and introducing frame gap tolerance constraints, frame gap uniformity constraints, and wing attitude control point deviation constraints; Step 4: Solve the pose integrated optimization model, including: using the initial docking pose as the initial optimization value, using the penalty function-quasi-Newton method to iteratively solve the objective function under the geometric constraints to obtain the optimal target pose of the wing relative to the fuselage; Step 5: Based on the optimal target pose, generate the drive command for the attitude adjustment positioner to drive the wing to the target pose and complete the docking assembly.

[0008] Preferably, in step 2, the keypoints of the point cloud are extracted using the KeypointDETR network, specifically including: A Transformer-based KeypointDETR network pre-trained model is adopted, which includes a point cloud encoder and a query decoder; The point cloud encoder is used to extract geometric features from the input point cloud. It aggregates each point based on a self-attention mechanism through a dynamic graph transformer module. Nearest neighbor feature; The query decoder is used to predict key points. It sets more queries than the actual number of key points. Each query contains a heatmap indicating the location of the key point and a probability value representing semantic consistency. By combining the heatmap and the probability value, the saliency estimation and correspondence estimation of the key points are completed, and the set of key points of the docking surface between the wing frame and the fuselage wing box frame is output.

[0009] Preferably, in step 2, a geometric feature description and two-way matching strategy is used to establish matching point pairs between key points on the fuselage and wing frame mating surfaces, specifically including: Step 21: For the extracted key points, calculate their fast point feature histogram as a geometric feature descriptor; Step 22: Perform bidirectional nearest neighbor coarse matching: Perform bidirectional nearest neighbor search on the wing key point set and the fuselage key point set in the feature space, set a distance threshold, and retain matching point pairs that satisfy bidirectional consistency; Step 23: Perform RANSAC geometric consistency screening: For each frame mating surface, based on the coarse matching point pairs, use the RANSAC algorithm to screen the interior point matching point pairs that satisfy the rigid body transformation model, and use them as the final matching point pairs.

[0010] Preferably, in step 3, the expression for the frame gap tolerance constraint is:

[0011] in, For the first frame after posture adjustment The first on the first docking surface The gap value at each matching point This is the maximum permissible gap at this part of the frame; The expressions for the uniformity constraint of the frame gap include: (1) For the front and rear beams of the enclosure: ; (2) For the upper and lower wall panels of the enclosure: ; in, This represents the maximum allowable deviation in the gap between the front and rear beams. This represents the maximum allowable deviation in the gap between the upper and lower wall panels. For the first The average gap between the mating surfaces.

[0012] Preferably, in step 3, the expression for the wing attitude control point deviation constraint is: ; For the first on the wing The deviation between the coordinates of the pose control point after pose adjustment and its theoretical coordinates. This represents the maximum allowable deviation for this control point.

[0013] Preferably, in step 4, the objective function is:

[0014] in, The pose parameters to be optimized are: and For the first A pair of matching points on the mating surface and Let be the final rotation and translation matrix to be determined.

[0015] Preferably, in step 4, the penalty function-quasi-Newton method is used to iteratively solve the objective function under the geometric constraints, specifically including: Step 41: Construct the augmented objective function: ; in The original objective function is... For gap tolerance constraint penalty term The penalty factor Penalty term for gap uniformity constraint The penalty factor Penalty term for attitude control point deviation constraint The penalty factor Orthogonalization constraint penalty term The penalty factor These are the pose parameters to be optimized; Step 42: Initialize optimization variables For the initial docking pose, initialize the inverse of the Hessian matrix to an approximation of the identity matrix, and initialize each penalty factor; Step 43: In the In the next iteration, the gradient of the augmented objective function is calculated. ; Step 44: Calculate the search direction: ; The step size is determined by line search. ; Step 45: Update and optimize variables ; And based on the BFGS formula, the approximation of the Hessian matrix inverse is updated. ; Step 46: Stop iterating when the convergence condition is met, and output the optimal pose parameters. Otherwise, let Return to step 43.

[0016] Beneficial effects of this invention: This invention replaces the theoretical numerical model with a measured point cloud model, uses the KeypointDETR network to extract key points on the docking surface, and combines FPFH feature description, bidirectional nearest neighbor matching and RANSAC geometric consistency screening to effectively solve the problems of traditional assembly methods, such as reliance on pose control points for pose adjustment, assembly gap deviations due to component deformation, and high interference risk.

[0017] This invention establishes a geometric multi-constraint model that includes frame gap tolerance constraints, gap uniformity constraints, and attitude control point deviation constraints. The objective function is to minimize the sum of squares of gaps between matching point pairs. The solution is obtained using the penalty function-quasi-Newton method, which achieves comprehensive optimization of gap distribution and pose accuracy at the docking surface.

[0018] This invention effectively solves the problem of poor mating quality and easy interference caused by component manufacturing deformation in the assembly of aircraft frame-type wing-body docking. It improves the quality of docking assembly and has important theoretical significance and practical engineering value for improving aircraft assembly precision and reducing development costs. Attached Figure Description

[0019] The accompanying drawings, which are provided to further illustrate the invention and constitute a part of this invention, are illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention.

[0020] Figure 1 This is a schematic diagram of the aircraft frame-type wing-body docking assembly attitude optimization system in this invention; Figure 2 This is a schematic diagram of the aircraft frame-type wing-body docking assembly attitude optimization technology route in this invention; Figure 3 This is a schematic diagram of the process for constructing the measured point cloud model of the aircraft frame-type wing-body docking assembly in this invention; Figure 4 This is a schematic diagram of the KeypointDETR network architecture based on the extraction of key points from bounding point clouds in this invention; Figure 5 This is a schematic diagram of the tolerance distribution of the gap between the mating surfaces of the frame in this invention; Figure 6 This is a schematic diagram illustrating the uniformity of the gap between the mating surfaces of the frame in this invention; Figure 7 This is a schematic diagram of the deviation of the wing attitude control point in this invention; Figure 8 This is a schematic diagram of the laser tracker connection module in this invention; Figure 9 This is a schematic diagram of the global coordinate system establishment module in this invention; Figure 10 This is a schematic diagram of the global calibration module for the 3D scanner in this invention; Figure 11 This is a schematic diagram of the bounding point cloud denoising module in this invention; Figure 12 This is a schematic diagram of the bounding box point cloud downsampling module in this invention; Figure 13 This is a schematic diagram of the frame point cloud key point matching and filtering module in this invention; Figure 14This is a schematic diagram of the frame pose deviation calculation and correction module in this invention. Detailed Implementation

[0021] This section will describe in detail specific embodiments of the present invention. Preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the drawings is to supplement the textual description with graphics, so that people can intuitively and vividly understand each technical feature and overall technical solution of the present invention, but they should not be construed as limiting the scope of protection of the present invention.

[0022] In a preferred embodiment of the present invention, the objective of this invention is to provide a method and system for comprehensive optimization of the attitude of large aircraft component docking and assembly based on a measured point cloud model and multiple geometric constraints. By constructing a measured point cloud model of the docking surface, establishing a multi-constraint comprehensive optimization model, and efficiently solving it, the aim is to improve the uniformity of the wing-body docking clearance and the assembly accuracy. The invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Example 1

[0023] This embodiment provides an aircraft frame-type wing-body docking assembly attitude optimization system for implementing the method of the present invention. The attitude optimization system includes physical hardware devices and software functional modules; the physical hardware devices include a 3D scanner, a laser tracker, an attitude adjuster, and a computer integrated control system, which provide data support and execution carrier.

[0024] Figure 1 This is a schematic diagram illustrating the composition of an aircraft frame-type wing-body docking assembly attitude optimization system provided in an embodiment of the present invention. (See also...) Figure 1 The physical hardware includes: a 3D scanner, a laser tracker, an attitude adjuster, and a computer-integrated control system.

[0025] The 3D scanner (MarvelScan) is used to collect point cloud data of the enclosure area and construct a measured point cloud model of the enclosure. The laser tracker (Leica-AT402 model) is used to construct a global coordinate system, measure the coordinates of the attitude control points, and calibrate the scanner coordinate system with the global coordinate system. The attitude adjustment and positioning fixture consists of multiple CNC positioners connected in parallel, responsible for wing support, spatial attitude adjustment, and precise docking.

[0026] The software functional modules include: The global coordinate system joint calibration module is used to unify laser tracker data and 3D scanner data. It uses the common target method and the IR-SVD-based method to perform global coordinate system joint calibration and transform the data to the global coordinate system. The point cloud model construction module is used to construct a measured point cloud model of the docking surface between the wing and the fuselage frame using point cloud data collected by a 3D scanner, and to extract key points and solve the initial docking pose. The optimization model building module is used to establish a comprehensive optimization model with the goal of minimizing the sum of squares of the gaps between the docking surfaces, and includes gap tolerance, gap uniformity and attitude control point deviation constraints. The optimization solution module is used to solve the comprehensive optimization model using the initial docking pose as the initial value and the penalty function-quasi-Newton method to obtain the optimal target pose. The attitude adjustment driver module is used to generate drive instructions for the attitude adjustment positioner based on the optimal target pose.

[0027] The software functional modules are embedded in the computer integrated control system and are responsible for the processing and analysis of all measurement data, the calculation and optimization of component poses, the motion control of the positioner, and the real-time feedback of status information. Example 2

[0028] This embodiment provides a method for optimizing the attitude of aircraft frame-type wing-body docking assembly based on measured point cloud and geometric multiple constraints. This method can be applied to the system described in Embodiment 1. Figure 2 A flowchart illustrating the method provided in an embodiment of the present invention. See also... Figure 2 The method includes the following steps: Step 1: Collection of bounding point cloud data.

[0029] A 3D scanner was used for calibration. After calibration, the parameters were adjusted to scan the mating surfaces of the wing frame and fuselage frame and collect their point cloud data.

[0030] Step 2: Global coordinate system joint calibration.

[0031] A joint calibration of the global coordinate system is performed using the common target method and the IR-SVD-based method. Let... , The target point coordinates in the scanner coordinate system. , For coordinates in the laser tracker's coordinate system, the transformation relationship between the two coordinate systems should satisfy: ; In the formula, , This represents the transformation relationship between the laser tracker and the global coordinate system. , The transformation relationship between the 3D scanner and the global coordinate system is to be determined. This allows for the establishment of a coordinate system.

[0032] Step 3: Construct a measured point cloud model of the frame-type wing-body docking and solve for the initial pose.

[0033] Figure 3 The process of constructing the measured point cloud model is demonstrated. Specifically, it includes: (1) Point cloud data preprocessing. Point cloud data of the wing frame and fuselage wing box frame were collected using a 3D scanner and preprocessed by denoising, stitching and downsampling to obtain a clean and concise measured point cloud model.

[0034] (2) Keypoint extraction based on the KeypointDETR network. Keypoints were extracted from the point clouds of the wing and fuselage bounding boxes using a pre-trained KeypointDETR network model based on Transformer. This network architecture includes a point cloud encoder and a query decoder. The point cloud encoder aggregates each point using a self-attention mechanism through a dynamic graph transformer module. k Geometric features are extracted using nearest neighbor features. The feature aggregation formula of the point cloud encoder is as follows: ; in, It is implemented using a multilayer perceptron (MLP). , and This is achieved through one-dimensional convolution. This represents the softmax operation applied to the self-attention weights. Represents element-wise multiplication; terminology This indicates the position code.

[0035] The query decoder uses multi-head self-attention and cross-attention mechanisms to decode features into multiple heatmaps and probabilities, ultimately determining the 3D coordinates and semantic labels of key points.

[0036] (3) Key point matching. To establish the correspondence between key points of the wing and fuselage, a three-stage strategy of "geometric feature description - bidirectional nearest neighbor - RANSAC screening" is adopted.

[0037] Phase 1: Geometric feature description.

[0038] The fast point feature histogram descriptor construction method is as follows: establish a local coordinate system based on the normal vector for each keypoint. ; In the formula, These are the three directions of the local coordinate system.

[0039] Calculate three feature angles within the neighborhood Create a simplified point feature histogram: ; The simplified histogram of neighboring points is then weighted and fused to obtain a fast point feature histogram.

[0040] .

[0041] Phase 2: Initial matching of bidirectional nearest neighbors.

[0042] Bidirectional nearest neighbor search satisfies: for wing critical points To the key points of the fuselage positive mapping Simultaneously, the reverse mapping from key points on the fuselage to key points on the wing. If the two match, they are retained as coarse matching point pairs.

[0043] Phase 3: RANSAC geometric consistency screening.

[0044] Based on coarse-matched point pairs, an interior point selection strategy using partial-matching interfaces is employed. Let the first... Initial matching set of each mating surface , which includes interior point sets and external point set The interior points satisfy the rigid body transformation model, whose expression is as follows: ; (4) Solving for the initial pose.

[0045] Based on the selected matching point pairs, the initial pose of each docking surface is solved using the SVD decomposition method. The local transformation results of the four docking surfaces are then fused. The fusion strategy is as follows: ; Obtain the initial docking pose of the wing relative to the fuselage. .

[0046] Step 4: Construct a comprehensive optimization model for the wing-body docking assembly posture based on multiple geometric constraints.

[0047] To address the uneven gap distribution caused by relying solely on the initial pose, this step establishes an optimization model that includes multiple geometric constraints and an objective function.

[0048] Geometric constraint one: Establish tolerance constraints for the frame gaps. For example... Figure 5 As shown, to ensure no interference occurs, the gap value of each point after pose adjustment must be within the allowable range. Let the extracted bounding box matching point pair set be... , After adjusting the posture, the frame is the first The gap at each matching point is: ; The constraints are:

[0049] in, It is the normal vector. This represents the maximum permissible clearance.

[0050] Geometric constraint two: Establish a constraint on the uniformity of the frame gap. For example... Figure 6 As shown, to ensure uniform assembly stress, the gap distribution of the front and rear beams and the upper and lower wall panels should be as consistent as possible. Let the average gap between the mating surfaces of the frame be denoted as... , The expression is: ; Define the formula for calculating the uniformity of the gap between the front and rear beams of the enclosure. : ; Define the formula for calculating the uniformity of the gap between the upper and lower wall panels of the enclosure. : ; in, This represents the maximum allowable deviation in the gap between the front and rear beams. This represents the maximum allowable deviation in the gap between the upper and lower wall panels. For the first The average gap between the mating surfaces.

[0051] Geometric constraint three: Establish deviation constraints for wing attitude control points. For example... Figure 7 As shown, the final position of the attitude control point on the wing needs to be close to the theoretical position. In the aircraft assembly coordinate system, the theoretical coordinates of the attitude control point are marked as follows: The measured coordinates are as follows The coordinates after pose adjustment are: ; The coordinate deviation is: ; The constraints are: ; Define the objective function: Minimize the sum of squares of the gaps between all matching point pairs, thus maximizing the fit of the mating surfaces. The objective function is: ; in, Represented as target pose parameters, including rotation matrix Translation matrix .

[0052] Step 5: Solve the optimization model based on the penalty function-quasi-Newton method.

[0053] The solution is obtained using the penalty function-quasi-Newton method.

[0054] (1) Construct the augmented objective function. Transform each constraint into a penalty term and add it to the original objective function:

[0055] In the formula, For gap tolerance constraint penalty term The penalty factor Penalty term for gap uniformity constraint The penalty factor Penalty term for attitude control point deviation constraint The penalty factor Orthogonalization constraint penalty term The penalty factor, the penalty function consists of four parts: Clearance tolerance constraint penalty : ; Gap uniformity constraint penalty term : ; Attitude control point deviation constraint penalty : ; Orthogonalization constraint penalty term : ; (2) Iterative optimization. Using the obtained initial pose... As optimization variables The initial value of the Hessian matrix is ​​approximately equal to the identity matrix. Initialize each penalty factor. Then iterate: Calculate the gradient of the current augmented objective function: ; Calculate the search direction: ; The step size is determined by a line search that satisfies the Armijo condition. .

[0056] Update and optimize variables: ; The approximation of the Hessian matrix inverse is updated using the BFGS formula: ; in, To optimize the target position difference, This represents the gradient difference.

[0057] (3) Convergence judgment. Iteration stops when any of the following conditions are met: ; Output optimal pose parameters That is, the final .

[0058] Step 6: Perform docking and assembly.

[0059] The optimal target pose obtained by the solution is converted into the driving quantity of each attitude adjustment positioner, and the control system drives the wing to move to the target pose to complete the precise docking with the fuselage.

[0060] Example 3 This embodiment presents a method and system for comprehensive optimization of the attitude of large aircraft component docking and assembly based on measured point cloud models and geometric multi-constraints. The developed software functional modules are used for aircraft frame-type wing-body docking and assembly. The software functional modules include a laser tracker connection module, a global coordinate system establishment module, a global coordinate system calibration module, a frame point cloud data processing module, a frame attitude optimization calculation module, and a frame attitude optimization execution module.

[0061] like Figure 8 As shown, in the laser tracker connection module, an IP address of the same category as the software is set to facilitate the connection and initialization of the laser tracker with the software, preparing for subsequent measurement work.

[0062] like Figure 9 As shown, in the global coordinate system establishment module, after the laser tracker is connected to the software and initialized, the laser tracker sequentially measures the reference points set on the ground or tooling, fits the laser tracker coordinate system with the global coordinate system according to the singular value decomposition method, and solves the coordinate transformation parameters.

[0063] like Figure 10 As shown, in the global coordinate system calibration module, a common target is used as a "bridge" to obtain the coordinates of its marker points in the scanner's own coordinate system using a 3D scanner. Similarly, a laser tracker is used to obtain the coordinates of its target points in the global coordinate system. The IR-SVD method is used to accurately solve the transformation parameters between the scanner coordinate system and the assembly global coordinate system, thereby achieving the unification of the two coordinate systems.

[0064] like Figures 11 to 13As shown, the bounding box point cloud data processing module includes a bounding box point cloud denoising module, a bounding box point cloud downsampling module, and a bounding box point cloud key point matching and filtering module. The bounding box point cloud files obtained from the scanner are imported into the software in PCD or PLY format, and the data is read. First, the bounding box point cloud data is denoised to remove noise points, and multi-view point cloud stitching is performed. Second, the bounding box point cloud is simplified according to the downsampling algorithm, preserving bounding box features while reducing subsequent point cloud processing time and improving point cloud processing efficiency. Finally, effective matching point pairs are extracted from the bounding box point cloud using the KeypointDETR network, FPFH descriptor, bidirectional nearest neighbor method, and RANSAC filtering strategy, and the relative pose deviation of the bounding box is calculated.

[0065] like Figure 14 As shown, in the frame pose deviation calculation and correction module, firstly, the coordinates of the fuselage wing box matching points and the wing frame matching points are imported. Secondly, based on the frame gap deviation constraint, the frame gap uniformity deviation constraint, and the pose control point deviation constraint, a geometrically multi-constraint frame-wing-body docking pose optimization model is constructed. The final wing pose is calculated using the penalty function-quasi-Newton method, yielding the final pose deviation of the wing frame during the frame-wing-body docking assembly process. Then, the driving quantities of each positioner are calculated based on inverse kinematics, and the trajectory planning of the attitude adjustment tooling is performed. Finally, the positioners are driven to correct the pose deviation and the results are fed back.

[0066] Verification of the technical effects of this invention As can be seen from the above specific implementation methods: (1) This invention replaces the theoretical numerical model with the measured point cloud model, uses the KeypointDETR network to extract key points of the docking surface, and combines FPFH feature description, bidirectional nearest neighbor matching and RANSAC geometric consistency screening. This invention effectively solves the problems of traditional assembly methods that rely on pose control points for pose adjustment, and the high risk of interference caused by component deformation in the docking assembly method between actual products and theoretical numerical models.

[0067] (2) This invention establishes a geometric multi-constraint model that includes frame gap tolerance constraints, gap uniformity constraints and attitude control point deviation constraints. The objective function is to minimize the sum of squares of gaps between matching points. The penalty function-quasi-Newton method is used to solve the problem, thereby achieving comprehensive optimization of the gap distribution and pose accuracy of the docking surface.

[0068] Therefore, this invention effectively solves the problem of poor mating quality and easy interference caused by component manufacturing deformation in the assembly of aircraft frame-type wing-body docking, improves the quality of docking assembly, and has important theoretical significance and practical engineering value for improving aircraft assembly precision and reducing development costs.

[0069] Without causing conflict, those skilled in the art can freely combine and use the above-mentioned additional technical features.

[0070] The above description is only a preferred embodiment of the present invention. Any technical solution that achieves the purpose of the present invention by essentially the same means is within the protection scope of the present invention.

Claims

1. A comprehensive optimization system for the pose of large aircraft component docking and assembly based on a measured point cloud model and multiple geometric constraints, characterized in that: The pose optimization system includes physical hardware devices and software functional modules; the physical hardware devices include a 3D scanner, a laser tracker, a pose adjuster and positioner, and a computer-integrated control system, which are used to provide data support and execution carrier. The software functional modules are embedded in the computer integrated control system to realize data processing, pose calculation and automated attitude adjustment control; The 3D scanner is used to collect point cloud data of the mating surface between the aircraft wing frame and the fuselage wing box frame, and send the point cloud data to the computer integrated control system. The laser tracker is used to construct a global coordinate system, measure the coordinates of the wing attitude control points, and calibrate the coordinate system of the 3D scanner with the global coordinate system. The attitude adjustment positioner is fixed to a rigid part of the aircraft wing and connected to the computer integrated control system to adjust the position and attitude of the aircraft wing according to drive commands. The computer-integrated control system is connected to the 3D scanner, laser tracker, and attitude adjuster respectively, and is used to perform attitude optimization and docking control. The software functional modules include: The global coordinate system joint calibration module is used to unify laser tracker data and 3D scanner data. It uses the common target method and the IR-SVD-based method to perform global coordinate system joint calibration and transform the data to the global coordinate system. The point cloud model construction module is used to construct a measured point cloud model of the docking surface between the wing and the fuselage frame using point cloud data collected by a 3D scanner, and to extract key points and solve the initial docking pose. The optimization model building module is used to establish a comprehensive optimization model with the goal of minimizing the sum of squares of the gaps between the docking surfaces, and includes gap tolerance, gap uniformity and attitude control point deviation constraints. The optimization solution module is used to solve the comprehensive optimization model using the initial docking pose as the initial value and the penalty function-quasi-Newton method to obtain the optimal target pose. The attitude adjustment driver module is used to generate drive instructions for the attitude adjustment positioner based on the optimal target pose.

2. A method for comprehensive optimization of the pose of large aircraft components during assembly based on a measured point cloud model and geometric multi-constraints, wherein the method is applied to the pose comprehensive optimization system described in claim 1, characterized in that: The method specifically includes the following steps: Step 1: Global coordinate system joint calibration. A global coordinate system calibration model is established using the common target method. The global coordinate system transformation relationship is solved using the IR-SVD method, and the point cloud data collected by the 3D scanner is unified to the global coordinate system established by the laser tracker. Step 2: Construct a measured point cloud model of the wing-body docking frame, including: acquiring point cloud data of the docking surfaces of the aircraft wing frame and fuselage wing box frame using a 3D scanner, and performing point cloud denoising, stitching, and downsampling preprocessing; extracting key points of the point cloud on the docking surfaces of the wing frame and fuselage wing box frame using a KeypointDETR network; establishing matching point pairs between key points of the fuselage and wing frame docking surfaces based on the key points of the point cloud, using geometric feature description and bidirectional matching strategies; and using the matching point pairs, solving the initial docking pose of the wing relative to the fuselage using singular value decomposition. Step 3: Construct a geometrically multi-constraint-based wing-fuselage docking assembly attitude comprehensive optimization model, including: establishing an objective function with the goal of minimizing the sum of squares of the gaps between the matching points of the wing and fuselage frame docking surfaces; establishing and introducing frame gap tolerance constraints, frame gap uniformity constraints, and wing attitude control point deviation constraints; Step 4: Solve the pose integrated optimization model, including: using the initial docking pose as the initial optimization value, using the penalty function-quasi-Newton method to iteratively solve the objective function under the geometric constraints to obtain the optimal target pose of the wing relative to the fuselage; Step 5: Based on the optimal target pose, generate the drive command for the attitude adjustment positioner to drive the wing to the target pose and complete the docking assembly.

3. The pose synthesis optimization method according to claim 2, characterized in that: In step 2, the keypoints of the point cloud are extracted using the KeypointDETR network for the wing frame mating surface and the fuselage wing box frame mating surface, specifically including: A Transformer-based KeypointDETR network pre-trained model is adopted, which includes a point cloud encoder and a query decoder; The point cloud encoder is used to extract geometric features from the input point cloud. It aggregates each point based on a self-attention mechanism through a dynamic graph transformer module. k Nearest neighbor feature; The query decoder is used to predict key points. It sets more queries than the actual number of key points. Each query contains a heatmap indicating the location of the key point and a probability value representing semantic consistency. By combining the heatmap and the probability value, the saliency estimation and correspondence estimation of the key points are completed, and the set of key points of the docking surface between the wing frame and the fuselage wing box frame is output.

4. The pose synthesis optimization method according to claim 2, characterized in that: In step 2, a geometric feature description and two-way matching strategy are used to establish matching point pairs between key points on the fuselage and wing frame mating surfaces. Specifically, this includes: Step 21: For the extracted key points, calculate their fast point feature histogram as a geometric feature descriptor; Step 22: Perform bidirectional nearest neighbor coarse matching: Perform bidirectional nearest neighbor search on the wing key point set and the fuselage key point set in the feature space, set a distance threshold, and retain matching point pairs that satisfy bidirectional consistency; Step 23: Perform RANSAC geometric consistency screening: For each frame mating surface, based on the coarse matching point pairs, use the RANSAC algorithm to screen the interior point matching point pairs that satisfy the rigid body transformation model, and use them as the final matching point pairs.

5. The pose synthesis optimization method according to claim 2, characterized in that: In step 3, the expression for the frame gap tolerance constraint is: ； in, For the first frame after posture adjustment The first on the first docking surface The gap value at each matching point This is the maximum permissible gap at this part of the frame; The expressions for the uniformity constraint of the frame gap include: (1) For the front and rear beams of the enclosure: ； (2) For the upper and lower wall panels of the enclosure: ； in, This represents the maximum allowable deviation in the gap between the front and rear beams. This represents the maximum allowable deviation in the gap between the upper and lower wall panels. For the first The average gap between the mating surfaces.

6. The pose synthesis optimization method according to claim 2, characterized in that: In step 3, the expression for the wing attitude control point deviation constraint is: ; For the first on the wing The deviation between the coordinates of the pose control point after pose adjustment and its theoretical coordinates. This represents the maximum allowable deviation for this control point.

7. The pose synthesis optimization method according to claim 2, characterized in that: In step 4, the objective function is: ； in, The pose parameters to be optimized are: and For the first A pair of matching points on the mating surface and Let be the final rotation and translation matrix to be determined.

8. The pose synthesis optimization method according to claim 7, characterized in that: In step 4, the penalty function-quasi-Newton method is used to iteratively solve the objective function under the geometric constraints, specifically including: Step 41: Construct the augmented objective function: ； in The original objective function is... For gap tolerance constraint penalty term The penalty factor Penalty term for gap uniformity constraint The penalty factor Penalty term for attitude control point deviation constraint The penalty factor Orthogonalization constraint penalty term The penalty factor These are the pose parameters to be optimized; Step 42: Initialize optimization variables For the initial docking pose, initialize the inverse of the Hessian matrix to an approximation of the identity matrix, and initialize each penalty factor; Step 43: In the In the next iteration, the gradient of the augmented objective function is calculated. ; Step 44: Calculate the search direction: ； The step size is determined by line search. ; Step 45: Update and optimize variables ； And based on the BFGS formula, the approximation of the Hessian matrix inverse is updated. ; Step 46: Stop iterating when the convergence condition is met, and output the optimal pose parameters. Otherwise, let Return to step 43.

Images