Method and device for automatically detecting, hammering and correcting aircraft skin

Abstract

The invention discloses an aircraft skin automatic detection hammering shape correction method and device. According to the method, point cloud data of the to-be-corrected skin arc convex surface are obtained through a plurality of laser detectors, a to-be-corrected skin three-dimensional model is established, and curvature information of all data points of the to-be-corrected skin arc convex surface is calculated through a curvature analysis tool; according to the actual curvature deviation and the actual curvature deviation between the curvature of each data point of the arc convex surface of the to-be-corrected skin and the curvature of each data point of the arc convex surface of the standard skin, each correction block of the to-be-corrected skin is divided according to the actual curvature deviation, and each correction block forms a correction area; and then through combination of a target detection algorithm, a particle swarm optimization algorithm and finite element simulation, obtaining an optimal shape correction scheme of each shape correction area, using the optimal shape correction scheme to correct the shape of each shape correction point position, and after shape correction is completed, repeating the above steps until the convex curvature of the skin arc after shape correction meets the requirement. According to the invention, intelligent, automatic and precise shape correction of the skin to be corrected can be realized, and the shape correction efficiency is relatively high.

Description

Technical Field

[0001] The invention belongs to the field of aircraft skin manufacturing, and in particular relates to a method and a device for automatically detecting and hammering the shape of aircraft skin. Background Art

[0002] Aircraft skin, as a core component of an aircraft's exterior structure, significantly impacts the performance of key areas such as the fuselage, tail, and wings. Its design complexity and size are increasing, and the requirements for precision and surface quality are also increasing. Skin shaping technology is a key technology in skin processing. However, existing skin shaping devices and methods have low shaping efficiency and accuracy, making it impossible to achieve high-precision skin shaping. Therefore, developing a method that can achieve high-precision skin shaping is of great significance to the development of skin shaping technology. Summary of the Invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a method and device for automatically detecting and hammering the aircraft skin to correct its shape.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] The present invention provides an automatic detection and hammering correction method for aircraft skin, which is specifically as follows:

[0006] S1. Install the skin to be shaped in the arc-shaped groove of the calibration platform. The bottom surface of the arc-shaped groove is provided with multiple cylindrical pits arranged in a rectangular shape. A laser detector is fixed in each cylindrical pit. The calibration platform is fixed to a base plate. A robotic arm is provided on the base plate. The robotic arm drives the hammer device to move and rotate. The servo in the hammer device drives the turret base to rotate the cylinder group and the shaping hammer group. Each shaping hammer of the shaping hammer group is driven to move by a cylinder of the cylinder group.

[0007] S2. Each laser detector performs a three-dimensional laser scan on the corresponding scanning area of the arc convex surface of the skin to be shaped, obtaining point cloud data of each scanning area, with an overlapping area between each two adjacent scanning areas. The point cloud data of each scanning area is then pre-processed by registration, denoising, and streamlining to obtain pre-processed point cloud data of the arc convex surface of the skin to be shaped;

[0008] S3. Inputting the pre-processed point cloud data of the circular arc convex surface of the skin to be shaped and the pre-measured thickness parameters of the skin to be shaped into a three-dimensional modeling software, constructing a three-dimensional model of the skin to be shaped along the normal direction of the surface using a normal offset algorithm, and calculating the curvature information of the point cloud data of the circular arc convex surface of the skin to be shaped using a curvature analysis tool; then comparing the curvature information of the point cloud data of the circular arc convex surface of the skin to be shaped with the curvature information of the circular arc convex surface of a standard skin to obtain the actual curvature deviation between the curvature of each data point of the circular arc convex surface of the skin to be shaped and the curvature of the data point at the same position of the circular arc convex surface of the standard skin;

[0009] S4. Compare each actual curvature deviation with a preset curvature deviation threshold range. A continuous data point area or a single discrete data point where the actual curvature deviation exceeds the preset curvature deviation threshold range is defined as a correction block. All correction blocks constitute the correction area. Then, correction hammers are selected for each correction block in turn. Then, the positions of all correction points, correction force magnitudes, correction force directions, and correction paths in the correction area are obtained by combining a particle swarm optimization algorithm with finite element simulation.

[0010] S5, according to the corresponding shaping hammer shape, shaping point, shaping path, shaping force and shaping force direction of each shaping block obtained in step S4, each shaping block is calibrated in turn; wherein, when each shaping block is calibrated, the controller first controls the servo to drive the turret base to drive each cylinder and each shaping hammer to rotate, so that the shaping block corresponds to the shaping hammer of the corresponding hammer shape, and then drives the hammering device to move by the mechanical arm, so that the corresponding shaping hammer moves along the shaping path, and whenever the end center of the corresponding shaping hammer moves to the corresponding one When the shaping point is above a certain shaping point, the robotic arm stops driving the hammer device to move, drives the hammer device to deflect, and deflects the shaping hammer to the corresponding shaping force direction. The corresponding cylinder drives the shaping hammer to extend and retract, so that the shaping hammer strikes the shaping point with the corresponding shaping force and shaping force direction, wherein the shaping force of the shaping hammer is adjusted by controlling the air pump by the controller to change the air pressure in the corresponding cylinder; after completing the striking and shaping of the shaping point, the robotic arm drives the hammer device to continue moving, so that the shaping hammer moves to the next shaping point along the shaping path;

[0011] S6. Execute steps S2 and S3 to perform curvature detection on the actual calibrated skin arc convex surface, and obtain the curvature deviation after calibration between the curvature of each data point on the actual calibrated skin arc convex surface and the curvature of the same data point on the standard skin arc convex surface. If the curvature deviations after calibration are all within the preset curvature deviation threshold range, the calibration work of the skin to be calibrated is completed; otherwise, continue to execute step S4 until the curvature deviations after calibration are all within the preset curvature deviation threshold range.

[0012] Preferably, Geomagic Studio 2012 software is used to perform registration, denoising and streamlining pre-processing on the point cloud data of each scan area. The registration process is as follows:

[0013] S21. Importing point cloud data of two adjacent scan areas into Geomagic Studio 2012 software and converting them into point cloud views, dragging the point cloud views of the two scan areas into the same view window using a VBScript script, selecting multiple common points with identifiers from the two point cloud views, aligning the common points in the two point cloud views, and completing the initial splicing of the point cloud data in the two point cloud views, thereby completing the initial registration of the point cloud data of the two scan areas;

[0014] S22, repeat step S21 to complete the initial registration of point cloud data of all scan areas;

[0015] S23, detecting the average value and standard deviation of the distances between each group of data points corresponding to the positions of the overlapping point clouds in each of two adjacent point cloud views;

[0016] S24: Determine whether the average distance and standard deviation of each group of data points corresponding to the positions of the overlapping area point clouds in each of the two adjacent point cloud views exceed the corresponding preset threshold. If only the average distance exceeds the preset threshold, use the ICP algorithm to perform fine registration on the point cloud data of the corresponding two scanning areas after the initial registration. If the standard deviation exceeds the preset threshold, use the "local registration" tool to select the deviation area and re-perform the initial registration, and return to step S23 until the average distance and standard deviation of each group of data points corresponding to the positions of the overlapping area point clouds in all adjacent point cloud views do not exceed the corresponding preset threshold, thereby completing the fine registration of the point cloud data of all scanning areas.

[0017] The denoising process is as follows: using the denoising function of Geomagic Studio 2012 software to remove noise points from the point cloud data of all scanned areas after registration;

[0018] The streamlining process is as follows: using the "Uniform Sampling" function of Geomagic Studio 2012 software, the density of data points in the point cloud data of all scanned areas after denoising is adjusted by uniform thinning, and then a uniform grid is reconstructed. Finally, interpolation compensation is performed on the edges of the surface and areas with higher than the preset curvature.

[0019] Preferably, the shaping hammer group consists of shaping hammer one, shaping hammer two and shaping hammer three, and shaping hammer one and shaping hammer two are both cylindrical hammers, the ends of the cylindrical hammers are cylindrical, the length of the end of shaping hammer one is smaller than the length of the end of shaping hammer two, and shaping hammer three is a spherical hammer, the end of the spherical hammer is ball-shaped.

[0020] More preferably, the process of selecting the shape of each shape correction block is as follows:

[0021] ① Create the yolov11_data.yaml file in yolov11, adjust the value of the number of detection types nc to 2, and the value of the type name name to ['circle', 'rectangle'];

[0022] ② Modify the folder path to the pre-annotated dataset, which includes four types of annotations: circular dot areas, irregular areas, rectangular linear areas, and short rectangular linear areas. Then use the yolov11n.pt pre-trained model for training to obtain the best.pt weight file;

[0023] ③ Create the yolov11_predict.py file, modify the weights to the best.pt weight file, and perform each calibration block category test on the unlabeled validation set. If the consistency between each calibration block category and the actual one reaches more than 95%, the model training is completed. Otherwise, increase the number of training rounds and the amount of data in the dataset, return to step ②, and retrain the yolov11n.pt pre-trained model until the model training is completed;

[0024] ④ Import the point cloud data of the calibration block into Geomagic Studio 2012 software and convert it into a point cloud view. Use the trained model to perform shape target detection on the point cloud view of the calibration block. If the point cloud view of the calibration block is detected as a 'circle', it is classified as a circular point area. If the point cloud view of the calibration block is detected as a 'rectangle', calculate the diagonal length of the detection box. If the diagonal length is within the preset length range, it is classified as a short rectangular linear area. If the diagonal length is greater than any value within the preset length range, it is classified as a long rectangular linear area. If the point cloud view of the calibration block is not detected as a 'circle' or 'rectangle', it is classified as an irregular area.

[0025] ⑤ If the shape correction block is a circular point area or an irregular area, select the shape correction hammer three; if the shape correction block is a short rectangular linear area, select the shape correction hammer one; if the shape correction block is a rectangular linear area, select the shape correction hammer two.

[0026] Preferably, the process of planning the correction points, correction forces and correction paths of the correction block by combining the particle swarm optimization algorithm and finite element simulation is as follows:

[0027] (1) Determine the number of calibration points for each calibration block. 2 Take a calibration point, less than 10mm 2The calibration block has only one calibration point, and then the particle swarm optimization algorithm is used to plan the positions and calibration paths of all calibration points in the calibration area;

[0028] (2) According to the shape of the shaping hammer, the shaping point, the shaping path, the magnitude of the rough-selected shaping force and the shaping force direction of each shaping block, the finite element simulation shaping of each shaping block of the skin to be shaped is carried out in sequence by using Ansys software, and the curvature and normal direction of the skin after the simulation shaping and each data point on the skin after the simulation shaping are obtained; wherein, the initial shaping force direction is the normal direction at the shaping point;

[0029] (3) Calculate the simulated curvature deviation of the curvature of each data point on each calibration block of the skin arc convex surface after simulation calibration and the curvature of the same position on the standard skin arc convex surface, as well as the simulated direction deviation of the normal direction of each data point on each calibration block of the skin arc convex surface after simulation calibration and the normal direction of the same position on the standard skin arc convex surface. If each simulated curvature deviation does not exceed the preset curvature deviation threshold range, and each simulated direction deviation does not exceed the preset direction deviation threshold range, then the magnitude and direction of each calibration force of each calibration block are the optimal magnitude and direction of calibration force. Otherwise, adjust the magnitude and direction of calibration force according to the simulated curvature deviation and simulated direction deviation of all data points on each calibration block, and return Step (2); wherein, each time step (2) is returned to be executed, the adjusted rough selection correction force magnitude and correction force direction are used; the process of adjusting the correction force magnitude and correction force direction according to the simulation curvature deviation and simulation direction deviation of all data points on each correction block is as follows: if the simulation curvature deviation of any data point on the correction block is higher than the preset curvature deviation threshold range, the current correction force magnitude is reduced by 2%; if the simulation curvature deviation of any data point on the correction block is lower than the preset curvature deviation threshold range, the current correction force magnitude is increased by 2%; if the simulation direction deviation of any data point on the correction block exceeds the preset direction deviation threshold range, the current correction force direction is tilted 1° in the opposite direction of the deviation.

[0030] More preferably, the particle swarm optimization algorithm is used to plan the positions of all the calibration points and the calibration paths in the calibration area, and the specific process is as follows:

[0031] 1.1 Set the number of particles m, particle velocity, learning factors c1 and c2, and inertia weight ω;

[0032] 1.2 Randomly initialize all particles in the particle swarm; each particle represents a candidate correction scheme, which includes a correction point coordinate set, a rough correction force size set, and a correction path. The particle encoding structure is:

[0033]

[0034] points is the set of calibration point coordinates of all n calibration points in the calibration area. Except for the calibration block with only one discrete data point, the initial value of the calibration point coordinates and the subsequent iteration values are set to the coordinates of the corresponding discrete data point and remain unchanged. The initial value of the calibration point coordinates of each other calibration point is randomly generated within the calibration block. forces is the set of coarse-selected calibration force magnitudes corresponding to the n calibration points. The initial value of the coarse-selected calibration force magnitude corresponding to each calibration point is calculated according to the required deformation through the stress-strain formula. path_order is the calibration path. Each element in path_order represents the calibration order of each calibration point. The initial value of the calibration order of each calibration point is randomly generated.

[0035] 1.3 Calculate the fitness value of each particle according to the objective function. The objective function is

[0036] Fitness=α·Δ κ +β·E a +γ·L path +δ·D e

[0037] Among them, α, β, γ and δ are weight coefficients, Δ κ is the sum of the curvature deviations after correction at all correction points, E a is the sum of the energy consumption of all calibration points, and the path length L path is the sum of the Euclidean distances between every two adjacent calibration points on the calibration path, D e is the sum of deformation penalties of all correction points;

[0038] The sum of curvature deviations Δ after correction at all correction points κ The calculation process is

[0039]

[0040] Where, is the curvature radius of the i-th correction point after the corresponding rough correction force is applied, is the curvature radius of the standard skin at the same position as the i-th correction point, is the curvature radius of the i-th correction point before correction, F i is the rough selection correction force of the i-th correction point, r is the correction influence radius, which is three times the diameter of the corresponding correction hammer end, d j is the distance from the jth correction point to the ith correction point in the correction area, λ is the attenuation coefficient, n is the number of correction points in the correction area, E is the elastic modulus of the skin material to be corrected, v is the Poisson's ratio of the skin material to be corrected, D is the bending stiffness of the skin to be corrected, and h is the thickness of the skin to be corrected;

[0041] The total energy consumption of all correction points Ea The calculation process is

[0042] E=E hammer +E path

[0043]

[0044] Where, E hammer E is the total energy required for correction of all correction points in the correction area using the corresponding correction hammers, path is the total energy consumed by the manipulator in the correction area to drive the corresponding correction hammer to move along the correction path, |Δx i | is the displacement of the i-th correction point in the correction area during correction, and and The absolute value of the difference between the two is represented, η is the energy conversion efficiency, P arm is the rated power of the robot arm, t path is the working motion time of the robot arm, v arm is the moving speed of the robotic arm;

[0045] The sum of deformation penalties D for all correction points e The calculation formula is

[0046]

[0047] 1.4 Update the particle speed and position, and calculate the updated particle fitness value according to the objective function. If the updated particle fitness value is less than the particle fitness value before the update, the individual best position and the global best position are updated. Otherwise, the individual best position and the global best position remain unchanged. The update formulas for particle speed and particle position are:

[0048] v sq (t+1)=ωv sq (t)+c1r1(p sq (t)-x sq (t))+c2r2(p gq (t)-x sq (t))

[0049] x sq (t+1)=x sq (t)+v sq (t+1)

[0050] Where q is the iteration dimension of the particle. Each calibration point has 5 iteration dimensions, which are the three directional coordinates (x, y, z), the magnitude of the rough calibration force, and the calibration path. A particle with n calibration points has a total of 5n iteration dimensions; t represents the number of iterations; r1 and r2 are both uniform random numbers in the range [0, 1]; p sq(t) is the individual optimal solution position of the sth particle at the tth iteration, p gq (t) is the global optimal solution position at the tth iteration, x sq (t) is the position of the sth particle at the tth iteration, v sq (t) is the velocity of the sth particle at the tth iteration, s = 1, 2, 3, ..., m;

[0051] 1.5 Repeat step 1.4. When the rate of change of the updated particle fitness value is lower than the preset change threshold for multiple times in the iteration process, or the particle group speed is lower than the preset value, the iteration is terminated to obtain the optimal position of each calibration point and calibration path in the calibration area.

[0052] More preferably, the initial values of the rough-selected correction forces at the correction points in the same correction block are all equal, and are calculated by finite element analysis, as follows:

[0053] I Import the 3D model of the skin to be shaped into the Ansys software, set the boundary conditions, set the contact area between the skin to be shaped and the calibration platform to be rough, set the friction coefficient, set the skin material parameters, and divide the mesh.

[0054] II Establish a coordinate system with the direction parallel to the central axis of the standard skin as the x-axis direction, the direction parallel to the radial direction of the standard skin as the z-axis direction, and the direction perpendicular to the x-axis and the z-axis as the y-axis direction; select the largest correction block in the skin to be corrected, and cut out a three-dimensional solid part with a fan-shaped cross section from the skin to be corrected within the correction block. The neutral surface of the three-dimensional solid part parallel to the correction block is t1, and the intersection line of the neutral surface t2 of the three-dimensional solid part parallel to the yoz plane and the correction block is a1. Calculate the mean curvature of each data point on the intersection line a2 of the neutral surface t2 and the neutral surface t1 The mean y1 of the distance between each data point on the intersection line a2 and each data point on the intersection line a1, and the central angle θ1 of the intersection line a1;

[0055] III Let the neutral surface parallel to the arc convex surface of the standard skin at the same position as the three-dimensional solid part of the skin to be calibrated be t3, let the intersection line of the neutral surface t4 parallel to the yoz plane at the same position as the three-dimensional solid part of the skin to be calibrated and the arc convex surface of the standard skin be a3, extract the mean curvature of each data point on the intersection line a4 of the neutral surface t4 and the neutral surface t3 The mean y0 of the distance between each data point on the intersection line a3 and each data point on the intersection line a4, and the central angle θ0 of the intersection line a3;

[0056] IV calculates the strain ε of the three-dimensional solid part parallel to the yoz plane on the skin to be shaped by the python script:

[0057]

[0058] in, is the arc length of the intersection line a1, is the arc length of the intersection line a3, l is the arc length of the intersection line a4;

[0059] V is calculated based on the elastic modulus E of the skin material to be shaped, and the stress of the three-dimensional solid part parallel to the yoz plane on the skin to be shaped is obtained by the formula σ=E·ε;

[0060] The VI uses a Python script to calculate the correction force F of each correction block based on the stress of the three-dimensional solid part parallel to the YOZ plane on the skin to be corrected. This is used as the initial value of the rough correction force of each correction point in the corresponding correction block. The correction force F of each correction block is calculated as follows:

[0061] F=∫ S σdA

[0062] Where S is the area of the corresponding calibration block, and dA is the infinitesimal area.

[0063] The invention discloses an automatic detection and hammering shape correction device for aircraft skin, comprising a hammering device, a mechanical arm, a calibration platform and a base plate.

[0064] The calibration platform is fixed to the base plate, and an arc-shaped groove is provided on the upper surface of the calibration platform. Positioning recesses are provided at the edges of both sides of the arc-shaped groove. A plurality of cylindrical pits arranged in a rectangular shape of X rows and Y columns are provided on the bottom surface of the arc-shaped groove, and a laser detector is fixed in each cylindrical pit.

[0065] The hammering device includes a turret base, a cylinder group and a correction hammer group. The turret base is driven to rotate by a servo, and the servo is driven to move and rotate by a mechanical arm arranged on the base plate. The cylinder group is arranged on the turret base and consists of three cylinders evenly distributed along the circumference. The correction hammer group consists of correction hammer one, correction hammer two and correction hammer three, and correction hammer one, correction hammer two and correction hammer three are respectively fixed to the piston rods of the three cylinders. Among them, correction hammer one and correction hammer two are both cylindrical hammers, the ends of the cylindrical hammers are cylindrical, and the length of the end of correction hammer one is smaller than the length of the end of correction hammer two. Correction hammer three is a spherical hammer, and the end of the spherical hammer is ball-headed.

[0066] Preferably, the robotic arm includes an upper arm, a lower arm, a wrist, a front end bracket, a connecting piece and a base, the base and the bottom plate constitute a rotating pair that rotates around a vertical axis and is driven by drive motor 1, the upper arm and the base constitute a rotating pair that rotates around a horizontal axis a and is driven by drive motor 2, the lower arm and the upper arm constitute a rotating pair that rotates around a horizontal axis b and is driven by drive motor 3, the wrist and the lower arm constitute a rotating pair and are driven by drive piece 1, and the rotation center axis of the wrist and the lower arm is perpendicular to the horizontal axis b, the front end bracket and the wrist constitute a rotating pair that rotates around a horizontal axis c and is driven by drive piece 2, and the connecting piece is fixed on the front end bracket; the horizontal axis a, the horizontal axis b and the horizontal axis c are parallel; the housing of the servo is fixed on the connecting piece, and the output shaft of the servo is fixed to the turret base.

[0067] More preferably, driving member 1 includes rotating shaft 1, stepping motor 1 and cylindrical gear pair, the small arm includes a connecting seat and a hollow shaft, the connecting seat is hinged to the large arm, the hollow shaft is fixed on the connecting seat, rotating shaft 1 is located in the hollow shaft, and forms a rotating pair with the hollow shaft, the shells of the two stepping motors 1 arranged at intervals are fixed to the connecting seat through the motor compartment, the output shafts of the two stepping motors 1 are connected to one end of rotating shaft 1 through two cylindrical gear pairs, and the other end of rotating shaft 1 is fixed to the wrist; driving member 2 includes stepping motor 2, a driving bevel gear and a driven bevel gear, the front end bracket includes rotating shaft 2 and a connecting plate, rotating shaft 2 and the wrist form a rotating pair, the shell of stepping motor 2 is fixed on the wrist, the output shaft of stepping motor 2 is fixed with the driving bevel gear, the driven bevel gear is fixed to the middle part of rotating shaft 2 and meshes with the driving bevel gear, the two ends of rotating shaft 2 are fixed to one end of the two connecting plates arranged at intervals, and the other ends of the two connecting plates are fixed to the two ends of the connecting member.

[0068] The present invention has the following beneficial effects:

[0069] The present invention can realize intelligent and precise shaping of the skin; specifically, the present invention performs three-dimensional laser scanning on the circular arc convex surface of the skin to be shaped by multiple laser detectors to obtain point cloud data of the circular arc convex surface of the skin to be shaped, and pre-processes the point cloud data, thereby improving the accuracy of subsequent calculations, thereby improving the shaping accuracy, and establishing a three-dimensional model of the skin to be shaped according to the pre-processed point cloud data, and using a curvature analysis tool to calculate the curvature information of each data point of the circular arc convex surface of the skin to be shaped, and compares it with the curvature information of the same position of the standard circular arc convex surface, and obtains the actual curvature deviation of the curvature of each data point of the circular arc convex surface of the skin to be shaped and the curvature of each data point of the circular arc convex surface of the standard skin, and according to the actual The curvature deviation divides the skin to be calibrated into various calibration blocks, and each calibration block constitutes a calibration area. Then, the target detection algorithm is used to divide the shape types of each calibration block, and the calibration hammer shape of each calibration block is selected. By combining the particle swarm optimization algorithm with finite element simulation, the calibration points, calibration force magnitude, calibration force direction and calibration path of the calibration area are obtained, and then the optimal calibration scheme for each calibration point is obtained, which further improves the calibration accuracy. Then, the optimal calibration scheme is used to calibrate the position of each calibration point, and the above steps are repeated after the calibration is completed until the curvature of the convex surface of the skin arc meets the requirements, thereby realizing intelligent automatic and precise calibration of the skin to be calibrated, and the calibration efficiency is high. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 Schematic diagram of the overall structure of the shape correction device of the present invention;

[0071] Figure 2 Schematic diagram of the structure of the robotic arm in the present invention;

[0072] Figure 3 This is a schematic structural diagram of the forearm, wrist, front end bracket and connector of the present invention;

[0073] Figure 4 Schematic diagram of the upper surface of the calibration platform in the present invention;

[0074] Figure 5 Schematic diagram of the structure of the cylindrical pit and the laser detector in the present invention;

[0075] Figure 6 It is a structural schematic diagram of the hammering device in the present invention;

[0076] Figure 7 Schematic diagram of the structure of the cylinder in the present invention;

[0077] Figure 8 is a cross-sectional view of the cylinder in the present invention;

[0078] Figure 9 is a flow chart of the present invention;

[0079] Figure 10 A flow chart for determining the correction points, correction force magnitude, and correction path in the present invention;

[0080] Figure 11 Schematic diagram of the working of the shape correction device of the present invention;

[0081] Figure 12 This is a schematic diagram of coordinate setting in the present invention;

[0082] Figure 13 It is a schematic diagram of the three-dimensional solid part cut out from the skin to be shaped in the present invention and the end face of the corresponding position on the standard skin. DETAILED DESCRIPTION

[0083] The present invention will be further described below with reference to the accompanying drawings.

[0084] like Figure 1 As shown, the present invention provides an automatic detection and hammering correction device for aircraft skin, comprising a hammering device A1, a mechanical arm A2, a calibration platform A3 and a base plate A4.

[0085] like Figure 2 and Figure 3 As shown, the robot arm A2 includes an upper arm B1, a lower arm B2, a wrist B4, a front bracket B5, a connecting piece B6, and a base B7. The base B7 and the bottom plate A4 form a revolving pair that rotates around a vertical axis and is driven by drive motor 1. The upper arm B1 and the base B7 form a revolving pair that rotates around a horizontal axis a and is driven by drive motor 2 B3. The lower arm B2 and the upper arm B1 form a revolving pair that rotates around a horizontal axis b and is driven by drive motor 3 F6. The wrist B4 and the lower arm B2 form a revolving pair that is driven by drive member 1, and the rotation center axis of the wrist B4 and the lower arm B2 is perpendicular to the horizontal axis b. The front bracket B5 and the wrist B4 form a revolving pair that rotates around a horizontal axis c and is driven by drive member 2. The horizontal axis a, the horizontal axis b, and the horizontal axis c are parallel. The connecting piece B6 is fixed to the front bracket B5 and is used to connect to the hammer device A1.

[0086] like Figure 4 and Figure 5 As shown, calibration platform A3 is fixed to base plate A4. Its arc-shaped top surface is defined by an arc-shaped groove, with positioning recesses defined along its edges. The bottom of the arc-shaped groove is defined by a plurality of cylindrical pits D2 arranged in a rectangular pattern of X rows by Y columns. A laser detector D1 is secured within each cylindrical pit D2. The bottom radius of the arc-shaped groove is consistent with the radius of the standard convex arc of the skin. In this embodiment, X = 20, Y = 40, and the diameter of the cylindrical pit D2 ranges from 29.95 to 30.05 mm, with a depth ranging from 24.95 to 25.05 mm.

[0087] like Figure 6 As shown, the hammering device A1 includes a turret base E5, a cylinder group and a shaping hammer group. The turret base E5 and the connecting piece B6 form a rotating pair and are driven to rotate by the servo E2. The cylinder group is arranged on the turret base E5 and consists of three cylinders evenly distributed along the circumference. The shaping hammer group consists of a shaping hammer 1 E3, a shaping hammer 2 E7 and a shaping hammer 3 E8, and the shaping hammer 1 E3, the shaping hammer 2 E7 and the shaping hammer 3 E8 are respectively fixed to the piston rods E10 of the three cylinders, wherein the shaping hammer 1 E3 and the shaping hammer 2 E7 are both cylindrical hammers, the ends of the cylindrical hammers are cylindrical, suitable for shaping of square linear areas, and the length of the end of the shaping hammer 1 E3 is smaller than the length of the end of the shaping hammer 2 E7. The shaping hammer 3 E8 is a spherical hammer, the end of the spherical hammer is ball-shaped, suitable for shaping of circular point areas and irregular areas.

[0088] As a preferred embodiment, the driving member 1 includes a rotating shaft 1, a stepper motor 1 F1 and a cylindrical gear pair F2, the small arm B2 includes a connecting base and a hollow shaft, the connecting base is hinged to the upper arm B1, the hollow shaft is fixed on the connecting base, the rotating shaft 1 is located in the hollow shaft, and forms a rotating pair with the hollow shaft, the housings of the two stepper motors 1 F1 arranged at intervals are fixed to the connecting base through the motor compartment F7, the output shafts of the two stepper motors 1 F1 are connected to one end of the rotating shaft 1 through two cylindrical gear pairs F2, and the other end of the rotating shaft 1 is fixed to the wrist B4.

[0089] As a preferred embodiment, the driving member 2 includes a stepper motor 2 F3, a driving bevel gear F4 and a driven bevel gear F5. The front end bracket B5 includes a rotating shaft 2 and a connecting plate. The rotating shaft 2 and the wrist B4 constitute a rotating pair. The housing of the stepper motor 2 F3 is fixed on the wrist B4. The output shaft of the stepper motor 2 F3 is fixed with the driving bevel gear F4. The driven bevel gear F5 is fixed to the middle of the rotating shaft 2 and meshes with the driving bevel gear F4. The two ends of the rotating shaft 2 are fixed to one end of two connecting plates arranged at intervals, and the other ends of the two connecting plates are fixed to the two ends of the connecting member B6.

[0090] As a preferred embodiment, Figure 7 As shown, the cylinder includes a cylinder head E9, a piston, a piston rod E10 and a cylinder body E11. The cylinder body E11 is fixed to the turret base E5 through the cylinder base E14. One end of the cylinder body E11 is provided with an air pipe that is integrally formed and connected to the inner cavity of the cylinder body E11. The other end is open and has an air hole. A sealing rubber tube E1 is connected to the air pipe. The cylinder head E9 is fixed to the opening (through a bolt E6). The piston is provided in the cylinder body E11 and forms a sliding pair with the cylinder body E11. The piston rod E10 passes through the circular hole formed in the cylinder head E9, is fixed to the piston, and forms a sliding pair with the circular hole. The piston will not cross the air hole when moving in the cylinder body E11.

[0091] More preferably, a sealing rubber ring E12 is embedded in the trachea to ensure airtightness between the trachea and the sealing rubber tube E1.

[0092] More preferably, if Figure 8 As shown, a plurality of wear-resistant rings E13 arranged at axial intervals are embedded on the piston.

[0093] Among them, each cylinder can be connected to an air pump (not shown in the figure) through an air pipe E4, and the servo, drive motor 1, drive motor 2 B3, drive motor 3 F6, stepper motor 1 F1, stepper motor 2 F3 and each air pump are all controlled by a controller.

[0094] like Figure 9 As shown, the present invention provides an automatic detection and hammering correction method for aircraft skin, which is specifically as follows:

[0095] S1, such as Figure 11 As shown, the skin to be shaped is embedded in the arc-shaped groove, and the edge convex part of the skin to be shaped is embedded in the positioning depression of the arc-shaped groove, so as to limit the position of the skin to be shaped and prevent the skin to be shaped from being offset during the shaping process.

[0096] S2, each laser detector D1 performs three-dimensional laser scanning on the corresponding scanning area of the arc convex surface of the skin to be calibrated, and obtains point cloud data of each scanning area, and there is an overlapping area between each two adjacent scanning areas. In this embodiment, the area of the overlapping area is 20% of the area of the scanning area; wherein, after the scanning is completed, if there is local data missing or abnormality, the corresponding scanning area is re-measured to ensure the integrity and availability of the point cloud data; then the point cloud data of each scanning area is stored in a file, and the point cloud data of each scanning area is pre-processed by registration, denoising and streamlining using Geomagic Studio 2012 software to obtain pre-processed point cloud data of the arc convex surface of the skin to be calibrated; wherein the registration process is as follows:

[0097] S21. Import the point cloud data of two adjacent scanning areas into Geomagic Studio 2012 software and convert them into point cloud views. Use VBScript to drag the point cloud views of the two scanning areas into the same view window. Select multiple common points with obvious marks (such as inflection points) from the two point cloud views, align the common points in the two point cloud views, and complete the initial splicing of the point cloud data in the two point cloud views, thereby completing the initial alignment of the point cloud data of the two scanning areas.

[0098] S22. Repeat step S21 to complete the initial registration of point cloud data in all scanned areas.

[0099] S23. Detect the average value and standard deviation of the distances between each group of data points corresponding to the positions between the point clouds in the overlapping area of each two adjacent point cloud views.

[0100] S24. Determine whether the distance average and standard deviation of each group of data points corresponding to the positions between the overlapping point clouds in each of the two adjacent point cloud views exceed the corresponding preset thresholds. If only the distance average exceeds the preset threshold, then use the ICP algorithm to fine-align the point cloud data of the two corresponding scan areas after initial registration. If the standard deviation exceeds the preset threshold, use the "Local Registration" tool to select the deviation area and re-initialize the registration, and return to step S23 until the distance average and standard deviation of each group of data points corresponding to the positions between the overlapping point clouds in all adjacent point cloud views do not exceed the corresponding preset thresholds, thereby completing the fine alignment of the point cloud data of all scan areas. In this embodiment, the preset threshold for the distance average is 0.008m, and the preset threshold for the standard deviation is 0.019m. When using the ICP algorithm to fine-align the point cloud data of the two corresponding scan areas after initial registration, the maximum number of iterations is set to 30, and the distance tolerance is 0.008m.

[0101] The denoising process is as follows: the denoising function of Geomagic Studio 2012 software is used to remove the noise points in the point cloud data of all scanned areas after registration, and the process is repeated three times.

[0102] The streamlining process is as follows: using the "uniform sampling" function of Geomagic Studio 2012 software, the density of data points in the point cloud data of all scanned areas after denoising is adjusted by uniform thinning, and some overly dense data points are randomly removed. The absolute spacing mode is used, and the spacing between adjacent data points is set to 0.010m. A uniform grid is reconstructed to ensure the uniform density of the point cloud and achieve the purpose of streamlining data. Then, interpolation compensation is performed on the edges of the surface and areas with higher than the preset curvature to avoid feature loss.

[0103] S3. Input the pre-processed point cloud data of the circular arc convex surface of the skin to be shaped and the pre-measured thickness parameters of the skin to be shaped into the 3D modeling software, construct a 3D model of the skin to be shaped along the normal direction of the surface through the normal offset algorithm, and use the curvature analysis tool (a tool that comes with the 3D modeling software) to calculate the curvature information of the point cloud data of the circular arc convex surface of the skin to be shaped; then compare the curvature information of the point cloud data of the circular arc convex surface of the skin to be shaped with the curvature information of the standard skin circular arc convex surface to obtain the actual curvature deviation between the curvature of each data point on the circular arc convex surface of the skin to be shaped and the curvature of the data point at the same position on the standard skin circular arc convex surface.

[0104] S4. Compare each actual curvature deviation with the preset curvature deviation threshold range. The continuous data point area or individual discrete data points where the actual curvature deviation exceeds the preset curvature deviation threshold range is a calibration block. All calibration blocks constitute the calibration area. In this embodiment, the preset curvature deviation threshold range is -2×10 -3 μm -1~2×10 -3 μm -1 ; Then, the shape of the shape-correcting hammer is selected for each shape-correcting block in turn, and then the positions of all shape-correcting points, the size of the shape-correcting force, the direction of the shape-correcting force and the shape-correcting path of the shape-correcting area are obtained by combining the particle swarm optimization algorithm with the finite element simulation, such as Figure 10 shown.

[0105] The process of selecting the shape of each shape-correcting block is as follows:

[0106] ① Create the yolov11_data.yaml file in yolov11, adjust the value of the number of detection types nc to 2, and the value of the type name name to ['circle','rectangle'].

[0107] ② Modify the folder path to the pre-annotated dataset, which includes four types of annotations: circular dot areas, irregular areas, rectangular line areas (longer square line areas), and short square line areas. Then use the yolov11n.pt pre-trained model for training to obtain the best.pt weight file.

[0108] ③ Create the yolov11_predict.py file, modify the weights to the best.pt weight file, and perform various calibration block categories (circular dot areas, irregular areas, rectangular line areas) on the unlabeled validation set. If the degree of conformity between each calibration block category and the actual area reaches more than 95%, the model training is completed. Otherwise, increase the number of training rounds and the amount of data in the dataset, return to step ②, and retrain the yolov11n.pt pre-trained model until the model training is completed.

[0109] ④ Import the point cloud data of the calibration block into Geomagic Studio 2012 software and convert it into a point cloud view. Use the trained model to perform shape target detection on the point cloud view of the calibration block. If the point cloud view of the calibration block is detected as a 'circle', it is classified as a circular point area. If the point cloud view of the calibration block is detected as a 'rectangle', calculate the diagonal length of the detection box. When the diagonal length is 50mm to 90mm, it is classified as a short square linear area. When the diagonal length is greater than 90mm, it is classified as a long rectangular linear area. If the point cloud view of the calibration block is not detected as a 'circle' or 'rectangle', it is classified as an irregular area.

[0110] ⑤ If the calibration block is a circular dot area or an irregular area, select calibration hammer three E8; if the calibration block is a short rectangular linear area, select calibration hammer one E3; if the calibration block is a rectangular linear area, select calibration hammer two E7.

[0111] like Figure 10 As shown in the figure, by combining the particle swarm optimization algorithm with finite element simulation, the process of planning the correction points, correction forces and correction paths in the correction area is as follows:

[0112] (1) Determine the number of calibration points for each calibration block. 2 Take a calibration point, less than 10mm 2 The calibration block has only one calibration point, and then the particle swarm optimization algorithm is used to plan the positions and calibration paths of all calibration points in the calibration area:

[0113] 1.1 Set the number of particles m = 50, the particle velocity range is [-10m / s, 10m / s], the learning factor c1 = c2 = 2, and the inertia weight ω to an initial value of 0.9, which decreases linearly to 0.4 with the number of iterations;

[0114] 1.2 Randomly initialize all particles in the particle swarm; each particle represents a candidate correction scheme, which includes a correction point coordinate set, a rough correction force size set, and a correction path. The particle encoding structure is:

[0115]

[0116] points is the set of calibration point coordinates of all n calibration points in the calibration area. Except for the calibration block with only one discrete data point, the initial value of the calibration point coordinates and the subsequent iteration values are set to the coordinates of the corresponding discrete data point and remain unchanged. The initial value of the calibration point coordinates of each other calibration point is randomly generated within the calibration block. forces is the set of coarse-selected calibration force magnitudes corresponding to the n calibration points. The initial value of the coarse-selected calibration force magnitude corresponding to each calibration point is calculated according to the required deformation through the stress-strain formula. path_order is the calibration path. Each element in path_order represents the calibration order of each calibration point. The initial value of the calibration order of each calibration point is randomly generated.

[0117] 1.3 Calculate the fitness value of each particle according to the objective function. The objective function is

[0118] Fitness=α·Δ κ +β·E a +γ·L path +δ·D e

[0119] Among them, α, β, γ and δ are weight coefficients, Δ κ is the sum of the curvature deviations after correction at all correction points, E a is the sum of the energy consumption of all calibration points, and the path length L pathis the sum of the Euclidean distances between every two adjacent calibration points on the calibration path, D e is the sum of the deformation penalties of all correction points; the deformation penalty is a strong suppression of curvature deterioration. If the curvature of the correction area deteriorates due to correction, a deformation penalty is applied;

[0120] The sum of curvature deviations Δ after correction at all correction points κ The calculation process is

[0121]

[0122]

[0123] Where, is the curvature radius of the i-th correction point after the corresponding rough correction force is applied, is the curvature radius of the standard skin at the same position as the i-th correction point, is the curvature radius of the i-th correction point before correction, F i is the rough selection correction force of the i-th correction point, r is the correction influence radius, which is three times the diameter of the corresponding correction hammer end, d j is the distance from the jth correction point to the ith correction point in the correction area, λ is the attenuation coefficient, which is 10 mm, n is the number of correction points in the correction area, E is the elastic modulus of the skin material to be corrected, v is the Poisson's ratio of the skin material to be corrected, D is the bending stiffness of the skin to be corrected, and h is the thickness of the skin to be corrected.

[0124] The total energy consumption of all correction points E a The calculation process is

[0125] E a =E hammer +E path

[0126]

[0127] Where, E hammer E is the total energy required for correction of all correction points in the correction area using the corresponding correction hammers, path is the total energy consumed by the manipulator in the correction area to drive the corresponding correction hammer to move along the correction path, |Δx i | is the displacement of the i-th correction point in the correction area during correction, and and The absolute value of the difference between the two is used to represent the energy conversion efficiency, η is the energy conversion efficiency, and the empirical value is 0.7, P arm is the rated power of the robot arm, t path is the working motion time of the robot arm, v arm The moving speed of the robot arm is manually set.

[0128] The sum of deformation penalties D for all correction points e The calculation formula is

[0129]

[0130] 1.4 Update the particle speed and position, and calculate the updated particle fitness value according to the objective function. If the updated particle fitness value is less than the particle fitness value before the update, the individual optimal solution position and the global optimal solution position are updated. Otherwise, the individual optimal solution position and the global optimal solution position remain unchanged. The update formulas for particle speed and particle position are respectively

[0131] v sq (t+1)=ωv sq (t)+c1r1(p sq (t)-x sq (t))+c2r2(p qq (t)-x sq (t))

[0132] x sq (t+1)=x sq (t)+v sq (t+1)

[0133] Where q is the iteration dimension of the particle. Each calibration point has 5 iteration dimensions, which are the three directional coordinates (x, y, z), the magnitude of the rough calibration force, and the calibration path. Therefore, a particle with n calibration points has a total of 5n iteration dimensions; t represents the number of iterations, r1 and r2 are both uniform random numbers in the range of [0, 1]; p sq (t) is the individual optimal solution position of the sth particle at the tth iteration, p gq (t) is the global optimal solution position at the tth iteration, x sq (t) is the position of the sth particle at the tth iteration, v sq (t) is the velocity of the sth particle at the tth iteration, s = 1, 2, 3, …, m.

[0134] 1.5 Repeat step 1.4. When the rate of change of the updated particle fitness value is lower than the preset rate threshold for multiple times in the iteration process, or the particle group speed is lower than the preset value, the iteration is terminated to obtain the optimal position of each calibration point and calibration path in the calibration area.

[0135] The initial values of the rough-selected correction forces at each correction point in the same correction block are all equal and are obtained through finite element analysis and calculation, as follows:

[0136] I imported the 3D model of the skin to be calibrated into the Ansys software, set the boundary conditions, set the contact area between the skin to be calibrated and the calibration platform (the arc convex surface of the skin to be calibrated) to be rough, set the friction coefficient to 0.2, set the skin material parameters to aluminum-lithium alloy (2060), and autonomously divide the mesh through Python code.

[0137] II Figure 12 and Figure 13 As shown, a coordinate system is established with the direction parallel to the central axis of the standard skin as the x-axis direction, the direction parallel to the radial direction of the standard skin as the z-axis direction, and the direction perpendicular to the x-axis and the z-axis as the y-axis direction; the largest correction block in the skin to be corrected is selected, and a three-dimensional solid part with a fan-shaped cross section is cut out from the skin to be corrected within the correction block. The neutral plane parallel to the correction block in the three-dimensional solid part is t1, and the intersection line of the neutral plane t2 parallel to the yoz plane in the three-dimensional solid part and the correction block is a1. The mean curvature of each data point on the intersection line a2 of the neutral plane t2 and the neutral plane t1 is calculated. The mean y1 of the distance between each data point on the intersection line a2 and each data point on the intersection line a1, and the central angle θ1 included by the intersection line a1.

[0138] III Let the neutral surface parallel to the arc convex surface of the standard skin at the same position as the three-dimensional solid part of the skin to be calibrated be t3, let the intersection line of the neutral surface t4 parallel to the yoz plane at the same position as the three-dimensional solid part of the skin to be calibrated and the arc convex surface of the standard skin be a3, extract the mean curvature of each data point on the intersection line a4 of the neutral surface t4 and the neutral surface t3 The mean y0 of the distance between each data point on the intersection line a3 and each data point on the intersection line a4, and the central angle θ0 included by the intersection line a3.

[0139] IV calculates the strain ε of the three-dimensional solid part parallel to the yoz plane on the skin to be shaped by the python script:

[0140]

[0141] in, is the arc length of the intersection line a1, is the arc length of the intersection line a3, and l is the arc length of the intersection line a4.

[0142] V is calculated based on the elastic modulus E of the skin material to be shaped, and the formula σ=E·ε (i.e., the stress-strain formula) is used to obtain the stress of the three-dimensional solid part of the skin to be shaped parallel to the yoz plane.

[0143] The VI uses a Python script to calculate the correction force F of each correction block based on the stress of the three-dimensional solid part parallel to the YOZ plane on the skin to be corrected. This is used as the initial value of the rough correction force of each correction point in the corresponding correction block. The correction force F of each correction block is calculated as follows:

[0144] F=∫ S σdA

[0145] Where S is the area of the corresponding calibration block, and dA is the infinitesimal area.

[0146] (2) According to the shape of the correction hammer, correction point, correction path, rough correction force and correction force direction of each correction block, the correction blocks of the skin to be corrected are sequentially subjected to finite element simulation correction by Ansys software, and the curvature and normal direction of the skin after simulation correction and each data point on the skin after simulation correction are obtained; among which, the initial correction force direction is the normal direction at the correction point.

[0147] (3) Calculate the simulated curvature deviation of the curvature of each data point on each calibration block of the skin arc convex surface after simulation calibration and the curvature of the same position on the standard skin arc convex surface, as well as the simulated direction deviation of the normal direction of each data point on each calibration block of the skin arc convex surface after simulation calibration and the normal direction of the same position on the standard skin arc convex surface. If each simulated curvature deviation does not exceed the preset curvature deviation threshold range, and each simulated direction deviation does not exceed the preset direction deviation threshold range, then the magnitude and direction of each calibration force of each calibration block are the optimal magnitude and direction of calibration force. Otherwise, adjust the magnitude and direction of calibration force according to the simulated curvature deviation and simulated direction deviation of all data points on each calibration block, and return Step (2); wherein, each time step (2) is returned to be executed, the adjusted rough selection correction force magnitude and correction force direction are used; the process of adjusting the correction force magnitude and correction force direction according to the simulation curvature deviation and simulation direction deviation of all data points on each correction block is as follows: if the simulation curvature deviation of any data point on the correction block is higher than the preset curvature deviation threshold range, the current correction force magnitude is reduced by 2%, if the simulation curvature deviation of any data point on the correction block is lower than the preset curvature deviation threshold range, the current correction force magnitude is increased by 2%, if the simulation direction deviation of any data point on the correction block exceeds the preset direction deviation threshold range, the current correction force direction is tilted 1° in the opposite direction of the deviation. wherein, after optimization in step (1), there will not be a situation on the correction block where the simulation curvature deviation of some data points is higher than the preset curvature deviation threshold range and the simulation curvature deviation of some data points is lower than the preset curvature deviation threshold range, nor will there be a situation where the simulation direction deviations of different data points are opposite.

[0148] S5, according to the corresponding shaping hammer shape, shaping point, shaping path, shaping force and shaping force direction of each shaping block obtained in step S4, each shaping block is calibrated in turn; wherein, when each shaping block is calibrated, the controller first controls the servo E2 to drive the turret base E5 to drive each cylinder and each shaping hammer to rotate, so that the shaping block corresponds to the shaping hammer of the corresponding hammer shape, and then drives the hammering device A1 to move through the mechanical arm A2, so that the corresponding shaping hammer moves along the shaping path, and whenever the end center of the corresponding shaping hammer moves to When it is above a corresponding correction point, the robot arm A2 stops driving the hammer device A1 to move, and drives the hammer device A1 to deflect, so that the direction of the correction hammer is deflected to the corresponding correction force direction, and the corresponding cylinder drives the correction hammer to extend and retract, so that the correction hammer strikes the correction point with the corresponding correction force size and correction force direction, and the correction force size of the correction hammer is adjusted by controlling the air pressure in the corresponding cylinder. After completing the knocking correction of the correction point, the robot arm A2 drives the hammer device A1 to continue moving, so that the correction hammer moves to the next correction point along the correction path.

[0149] S6. Execute steps S2 and S3 to perform curvature detection on the actual calibrated skin arc convex surface, and obtain the curvature deviation after calibration between the curvature of each data point on the actual calibrated skin arc convex surface and the curvature of the same data point on the standard skin arc convex surface. If the curvature deviations after calibration are all within the preset curvature deviation threshold range, the calibration work of the skin to be calibrated is completed. Otherwise, continue to execute step S4 until the curvature deviations after calibration are all within the preset curvature deviation threshold range, and the calibration work of the skin to be calibrated is completed.

Claims

1. A method for automatically detecting and correcting the shape of aircraft skin by hammering, characterized in that: The details are as follows: S1. Install the skin to be shaped in the arc-shaped groove of the calibration platform. The bottom surface of the arc-shaped groove is provided with multiple cylindrical pits arranged in a rectangular shape. A laser detector is fixed in each cylindrical pit. The calibration platform is fixed to a base plate. A robotic arm is provided on the base plate. The robotic arm drives the hammer device to move and rotate. The servo in the hammer device drives the turret base to rotate the cylinder group and the shaping hammer group. Each shaping hammer of the shaping hammer group is driven to move by a cylinder of the cylinder group. S2. Each laser detector performs a three-dimensional laser scan on the corresponding scanning area of the arc convex surface of the skin to be shaped, obtaining point cloud data of each scanning area, with an overlapping area between each two adjacent scanning areas. The point cloud data of each scanning area is then pre-processed by registration, denoising, and streamlining to obtain pre-processed point cloud data of the arc convex surface of the skin to be shaped; S3. Inputting the pre-processed point cloud data of the circular arc convex surface of the skin to be shaped and the pre-measured thickness parameters of the skin to be shaped into a three-dimensional modeling software, constructing a three-dimensional model of the skin to be shaped along the normal direction of the surface using a normal offset algorithm, and calculating the curvature information of the point cloud data of the circular arc convex surface of the skin to be shaped using a curvature analysis tool; then comparing the curvature information of the point cloud data of the circular arc convex surface of the skin to be shaped with the curvature information of the circular arc convex surface of a standard skin to obtain the actual curvature deviation between the curvature of each data point of the circular arc convex surface of the skin to be shaped and the curvature of the data point at the same position of the circular arc convex surface of the standard skin; S4. Compare each actual curvature deviation with a preset curvature deviation threshold range. A continuous data point area or a single discrete data point where the actual curvature deviation exceeds the preset curvature deviation threshold range is defined as a correction block. All correction blocks constitute the correction area. Then, correction hammers are selected for each correction block in turn. Then, the positions of all correction points, correction force magnitudes, correction force directions, and correction paths in the correction area are obtained by combining a particle swarm optimization algorithm with finite element simulation. S5, according to the corresponding shaping hammer shape, shaping point, shaping path, shaping force and shaping force direction of each shaping block obtained in step S4, each shaping block is calibrated in turn; wherein, when each shaping block is calibrated, the controller first controls the servo to drive the turret base to drive each cylinder and each shaping hammer to rotate, so that the shaping block corresponds to the shaping hammer of the corresponding hammer shape, and then drives the hammering device to move by the mechanical arm, so that the corresponding shaping hammer moves along the shaping path, and whenever the end center of the corresponding shaping hammer moves to the corresponding one When the shaping point is above a certain shaping point, the robotic arm stops driving the hammer device to move, drives the hammer device to deflect, and deflects the shaping hammer to the corresponding shaping force direction. The corresponding cylinder drives the shaping hammer to extend and retract, so that the shaping hammer strikes the shaping point with the corresponding shaping force and shaping force direction, wherein the shaping force of the shaping hammer is adjusted by controlling the air pump by the controller to change the air pressure in the corresponding cylinder; after completing the striking and shaping of the shaping point, the robotic arm drives the hammer device to continue moving, so that the shaping hammer moves to the next shaping point along the shaping path; S6. Execute steps S2 and S3 to perform curvature detection on the actual calibrated skin arc convex surface, and obtain the curvature deviation after calibration between the curvature of each data point on the actual calibrated skin arc convex surface and the curvature of the same data point on the standard skin arc convex surface. If the curvature deviations after calibration are all within the preset curvature deviation threshold range, the calibration work of the skin to be calibrated is completed; otherwise, continue to execute step S4 until the curvature deviations after calibration are all within the preset curvature deviation threshold range.

2. The method for automatically detecting and correcting the shape of aircraft skin by hammering according to claim 1, characterized in that: Geomagic Studio 2012 software was used to perform registration, denoising, and streamlined preprocessing of the point cloud data of each scan area. The registration process was as follows: S21. Importing point cloud data of two adjacent scan areas into Geomagic Studio 2012 software and converting them into point cloud views, dragging the point cloud views of the two scan areas into the same view window using a VBScript script, selecting multiple common points with identifiers from the two point cloud views, aligning the common points in the two point cloud views, and completing the initial splicing of the point cloud data in the two point cloud views, thereby completing the initial registration of the point cloud data of the two scan areas; S22, repeat step S21 to complete the initial registration of point cloud data of all scan areas; S23, detecting the average value and standard deviation of the distances between each group of data points corresponding to the positions of the overlapping point clouds in each of two adjacent point cloud views; S24: Determine whether the average distance and standard deviation of each group of data points corresponding to the positions of the overlapping point clouds in each of the two adjacent point cloud views exceed the corresponding preset thresholds. If only the average distance exceeds the preset threshold, use the ICP algorithm to perform fine registration on the point cloud data of the corresponding two scan areas after initial registration. If the standard deviation exceeds the preset threshold, use the "Local Registration" tool to select the deviation area and re-perform the initial registration. Return to step S23 until the average distance and standard deviation of each group of data points corresponding to the positions of the overlapping point clouds in all adjacent point cloud views do not exceed the corresponding preset thresholds, thereby completing the fine registration of the point cloud data of all scan areas. The denoising process is as follows: using the denoising function of Geomagic Studio 2012 software to remove noise points from the point cloud data of all scanned areas after registration; The streamlining process involves using the "Uniform Sampling" function in Geomagic Studio 2012 software to uniformly thin out the density of data points in the denoised point cloud data of all scanned areas. A uniform mesh is then reconstructed, and interpolation compensation is performed on surface edges and areas with curvature exceeding a preset value.

3. The method for automatically detecting and correcting the shape of aircraft skin by hammering according to claim 1, characterized in that: The shaping hammer group consists of shaping hammer one, shaping hammer two and shaping hammer three, and shaping hammer one and shaping hammer two are both cylindrical hammers, the ends of the cylindrical hammers are cylindrical, the length of the end of shaping hammer one is smaller than the length of the end of shaping hammer two, and shaping hammer three is a spherical hammer, the end of the spherical hammer is ball-shaped.

4. The method for automatically detecting and correcting the shape of aircraft skin by hammering according to claim 3, characterized in that: The process of selecting the shape of each shape block is as follows: ① Create the yolov11_data.yaml file in yolov11, adjust the value of the number of detection types nc to 2, and the value of the type name name to ['circle', 'rectangle']; ② Modify the folder path to the pre-annotated dataset, which includes four types of annotations: circular dot areas, irregular areas, rectangular linear areas, and short rectangular linear areas. Then use the yolov11n.pt pre-trained model for training to obtain the best.pt weight file; ③ Create the yolov11_predict.py file, modify the weights to the best.pt weight file, and perform each calibration block category test on the unlabeled validation set. If the degree of conformity between each calibration block category and the actual one reaches more than 95%, the model training is completed. Otherwise, increase the number of training rounds and the amount of data in the dataset, return to step ②, and retrain the yolov11n.pt pre-trained model until the model training is completed; ④ Import the point cloud data of the calibration block into Geomagic Studio 2012 software and convert it into a point cloud view. Use the trained model to perform shape target detection on the point cloud view of the calibration block. If the point cloud view of the calibration block is detected as a 'circle', it is classified as a circular point area. If the point cloud view of the calibration block is detected as a 'rectangle', calculate the diagonal length of the detection box. If the diagonal length is within the preset length range, it is classified as a short rectangular linear area. If the diagonal length is greater than any value within the preset length range, it is classified as a long rectangular linear area. If the point cloud view of the calibration block is not detected as a 'circle' or 'rectangle', it is classified as an irregular area. ⑤ If the shape correction block is a circular point area or an irregular area, select the shape correction hammer three; if the shape correction block is a short rectangular linear area, select the shape correction hammer one; if the shape correction block is a rectangular linear area, select the shape correction hammer two.

5. The method for automatically detecting and correcting the shape of aircraft skin by hammering according to claim 1, characterized in that: By combining the particle swarm optimization algorithm with finite element simulation, the process of planning the correction points, correction forces, and correction paths in the correction area is as follows: (1) Determine the number of calibration points for each calibration block. 2 Take a calibration point, less than 10mm 2 The calibration block has only one calibration point, and then the particle swarm optimization algorithm is used to plan the positions and calibration paths of all calibration points in the calibration area; (2) According to the shape of the shaping hammer, the shaping point, the shaping path, the magnitude of the rough-selected shaping force and the shaping force direction of each shaping block, the finite element simulation shaping of each shaping block of the skin to be shaped is carried out in sequence by using Ansys software, and the curvature and normal direction of the skin after the simulation shaping and each data point on the skin after the simulation shaping are obtained; wherein, the initial shaping force direction is the normal direction at the shaping point; (3) Calculate the simulated curvature deviation of the curvature of each data point on each calibration block of the skin arc convex surface after simulation calibration and the curvature of the same position on the standard skin arc convex surface, as well as the simulated direction deviation of the normal direction of each data point on each calibration block of the skin arc convex surface after simulation calibration and the normal direction of the same position on the standard skin arc convex surface. If each simulated curvature deviation does not exceed the preset curvature deviation threshold range, and each simulated direction deviation does not exceed the preset direction deviation threshold range, then the magnitude and direction of each calibration force of each calibration block are the optimal magnitude and direction of calibration force. Otherwise, adjust the magnitude and direction of calibration force according to the simulated curvature deviation and simulated direction deviation of all data points on each calibration block, and return Step (2); wherein, each time step (2) is returned to be executed, the adjusted rough selection correction force magnitude and correction force direction are used; the process of adjusting the correction force magnitude and correction force direction according to the simulation curvature deviation and simulation direction deviation of all data points on each correction block is as follows: if the simulation curvature deviation of any data point on the correction block is higher than the preset curvature deviation threshold range, the current correction force magnitude is reduced by 2%; if the simulation curvature deviation of any data point on the correction block is lower than the preset curvature deviation threshold range, the current correction force magnitude is increased by 2%; if the simulation direction deviation of any data point on the correction block exceeds the preset direction deviation threshold range, the current correction force direction is tilted 1° in the opposite direction of the deviation.

6. The method for automatically detecting and correcting the shape of aircraft skin by hammering according to claim 5, characterized in that: The particle swarm optimization algorithm is used to plan the positions of all the correction points and the correction paths in the correction area. The specific process is as follows: 1).1Set the number of particles m, particle velocity, learning factors c1 and c2, and inertia weight ω; 1).2 Randomly initialize all particles in the particle swarm; each particle represents a candidate correction scheme, which includes a correction point coordinate set, a rough correction force size set, and a correction path. The particle encoding structure is: points is the set of calibration point coordinates of all n calibration points in the calibration area. Except for the calibration block with only one discrete data point, the initial value of the calibration point coordinates and the subsequent iteration values are set to the coordinates of the corresponding discrete data point and remain unchanged. The initial value of the calibration point coordinates of each other calibration point is randomly generated within the calibration block. forces is the set of coarse-selected calibration force magnitudes corresponding to the n calibration points. The initial value of the coarse-selected calibration force magnitude corresponding to each calibration point is calculated according to the required deformation through the stress-strain formula. path_order is the calibration path. Each element in path_order represents the calibration order of each calibration point. The initial value of the calibration order of each calibration point is randomly generated. 1).3 Calculate the fitness value of each particle according to the objective function. The objective function is Fitness=a·D κ +β·E a +γ·L path +δ·D e Among them, α, β, γ and δ are weight coefficients, Δ κ is the sum of the curvature deviations after correction at all correction points, E a is the sum of the energy consumption of all calibration points, and the path length L path is the sum of the Euclidean distances between every two adjacent calibration points on the calibration path, D e is the sum of deformation penalties of all correction points; The sum of curvature deviations Δ after correction at all correction points κ The calculation process is Where, is the curvature radius of the i-th correction point after the corresponding rough correction force is applied, is the curvature radius of the standard skin at the same position as the i-th correction point, is the curvature radius of the i-th correction point before correction, F i is the rough selection correction force of the i-th correction point, r is the correction influence radius, which is three times the diameter of the corresponding correction hammer end, d j is the distance from the jth correction point to the ith correction point in the correction area, λ is the attenuation coefficient, n is the number of correction points in the correction area, E is the elastic modulus of the skin material to be corrected, v is the Poisson's ratio of the skin material to be corrected, D is the bending stiffness of the skin to be corrected, and h is the thickness of the skin to be corrected; The total energy consumption of all correction points E a The calculation process is E=E hammer +E path Where, E hammer E is the total energy required for correction of all correction points in the correction area using the corresponding correction hammers, path is the total energy consumed by the manipulator in the correction area to drive the corresponding correction hammer to move along the correction path, |Δx i | is the displacement of the i-th correction point in the correction area during correction, and and The absolute value of the difference between the two is represented, η is the energy conversion efficiency, P arm is the rated power of the robot arm, t path is the working motion time of the robot arm, v arm is the moving speed of the robotic arm; The sum of deformation penalties D for all correction points e The calculation formula is 1).4 Update the particle speed and position, and calculate the updated particle fitness value according to the objective function. If the updated particle fitness value is less than the particle fitness value before the update, the individual best position and the global best position are updated. Otherwise, the individual best position and the global best position remain unchanged. The update formulas for particle speed and particle position are v sq (t+1)=ωv sq (t)+c1r1(p sq (t)-x sq (t))+c2r2(p gq (t)-x sq (t)) x sq (t+1)=x sq (t)+v sq (t+1) Where q is the iteration dimension of the particle. Each calibration point has 5 iteration dimensions, which are the three directional coordinates (x, y, z), the magnitude of the rough calibration force, and the calibration path. A particle with n calibration points has a total of 5n iteration dimensions; t represents the number of iterations; r1 and r2 are both uniform random numbers in the range [0, 1]; p sq (t) is the individual optimal solution position of the sth particle at the tth iteration, p gq (t) is the global optimal solution position at the tth iteration, x sq (t) is the position of the sth particle at the tth iteration, v sq (t) is the velocity of the sth particle at the tth iteration, s = 1, 2, 3, …, m; 1).5 Repeat step 1.

4. When the rate of change of the updated particle fitness value is lower than the preset change threshold for multiple times in the iteration process, or the particle group speed is lower than the preset value, the iteration is terminated to obtain the optimal position of each calibration point and calibration path in the calibration area.

7. The method for automatically detecting and correcting the shape of aircraft skin by hammering according to claim 6, characterized in that: The initial values of the rough-selected correction forces at each correction point in the same correction block are all equal, and are obtained through finite element analysis and calculation, as follows: I. Import the 3D model of the skin to be shaped into the Ansys software, set the boundary conditions, set the contact area between the skin to be shaped and the calibration platform to be rough, set the friction coefficient, set the skin material parameters, and divide the mesh; II Establish a coordinate system with the direction parallel to the central axis of the standard skin as the x-axis direction, the direction parallel to the radial direction of the standard skin as the z-axis direction, and the direction perpendicular to the x-axis and the z-axis as the y-axis direction; select the largest correction block in the skin to be corrected, and cut out a three-dimensional solid part with a fan-shaped cross section from the skin to be corrected within the correction block. The neutral surface of the three-dimensional solid part parallel to the correction block is t1, and the intersection line of the neutral surface t2 of the three-dimensional solid part parallel to the yoz plane and the correction block is a1. Calculate the mean curvature of each data point on the intersection line a2 of the neutral surface t2 and the neutral surface t1 The mean y1 of the distance between each data point on the intersection line a2 and each data point on the intersection line a1, and the central angle θ1 of the intersection line a1; III Let the neutral surface parallel to the arc convex surface of the standard skin at the same position as the three-dimensional solid part of the skin to be calibrated be t3, let the intersection line of the neutral surface t4 parallel to the yoz plane at the same position as the three-dimensional solid part of the skin to be calibrated and the arc convex surface of the standard skin be a3, extract the mean curvature of each data point on the intersection line a4 of the neutral surface t4 and the neutral surface t3 The mean y0 of the distance between each data point on the intersection line a3 and each data point on the intersection line a4, and the central angle θ0 of the intersection line a3; IV calculates the strain ε of the three-dimensional solid part parallel to the yoz plane on the skin to be shaped by the python script: in, is the arc length of the intersection line a1, is the arc length of the intersection line a3, l is the arc length of the intersection line a4; V is calculated based on the elastic modulus E of the skin material to be shaped, and the stress of the three-dimensional solid part parallel to the yoz plane on the skin to be shaped is obtained by the formula σ=E·ε; The VI uses a Python script to calculate the correction force F of each correction block based on the stress of the three-dimensional solid part parallel to the YOZ plane on the skin to be corrected. This is used as the initial value of the rough correction force of each correction point in the corresponding correction block. The correction force F of each correction block is calculated as follows: S=∫ s σdA Where S is the area of the corresponding calibration block, and dA is the infinitesimal area.

8. A hammering and shaping device for use in an automatic detection and hammering shaping method for aircraft skin according to any one of claims 1 to 7, characterized in that: The device comprises a hammering device, a robotic arm, a calibration platform and a base plate. The calibration platform is fixed to the base plate. The upper surface of the calibration platform is provided with an arc-shaped groove. Positioning recesses are provided at both sides of the arc-shaped groove. The bottom surface of the arc-shaped groove is provided with a plurality of cylindrical pits arranged in an X-row × Y-column rectangular pattern. A laser detector is fixed in each cylindrical pit. The hammering device includes a turret base, a cylinder group and a correction hammer group. The turret base is driven to rotate by a servo, and the servo is driven to move and rotate by a mechanical arm arranged on the base plate. The cylinder group is arranged on the turret base and consists of three cylinders evenly distributed along the circumference. The correction hammer group consists of correction hammer one, correction hammer two and correction hammer three, and correction hammer one, correction hammer two and correction hammer three are respectively fixed to the piston rods of the three cylinders. Among them, correction hammer one and correction hammer two are both cylindrical hammers, the ends of the cylindrical hammers are cylindrical, and the length of the end of correction hammer one is smaller than the length of the end of correction hammer two. Correction hammer three is a spherical hammer, and the end of the spherical hammer is ball-headed.

9. The device used in the method for automatically detecting and correcting the shape of aircraft skin by hammering according to claim 8, characterized in that: The robotic arm includes an upper arm, a lower arm, a wrist, a front end bracket, a connecting piece and a base. The base and the bottom plate constitute a rotating pair that rotates around a vertical axis and are driven by a driving motor 1. The upper arm and the base constitute a rotating pair that rotates around a horizontal axis a and are driven by a driving motor 2. The lower arm and the upper arm constitute a rotating pair that rotates around a horizontal axis b and are driven by a driving motor 3. The wrist and the lower arm constitute a rotating pair and are driven by a driving piece 1, and the rotation center axis of the wrist and the lower arm is perpendicular to the horizontal axis b. The front end bracket and the wrist constitute a rotating pair that rotates around a horizontal axis c and are driven by a driving piece 2. The connecting piece is fixed on the front end bracket, and the horizontal axis a, the horizontal axis b and the horizontal axis c are parallel; the housing of the servo is fixed on the connecting piece, and the output shaft of the servo is fixed to the turret base.

10. The device used in the method for automatically detecting and correcting the shape of aircraft skin by hammering according to claim 9, characterized in that: The driving member 1 includes a rotating shaft 1, a stepper motor 1 and a cylindrical gear pair. The small arm includes a connecting seat and a hollow shaft. The connecting seat is hinged to the large arm. The hollow shaft is fixed on the connecting seat. The rotating shaft 1 is located in the hollow shaft and forms a rotating pair with the hollow shaft. The shells of the two stepper motors 1 arranged at intervals are fixed to the connecting seat through the motor compartment. The output shafts of the two stepper motors 1 are connected to one end of the rotating shaft 1 through two cylindrical gear pairs, and the other end of the rotating shaft 1 is fixed to the wrist; the driving member 2 includes a stepper motor 2, a driving bevel gear and a driven bevel gear. The front end bracket includes a rotating shaft 2 and a connecting plate. The rotating shaft 2 and the wrist form a rotating pair. The shell of the stepper motor 2 is fixed on the wrist. The output shaft of the stepper motor 2 is fixed with the driving bevel gear. The driven bevel gear is fixed to the middle of the rotating shaft 2 and meshes with the driving bevel gear. The two ends of the rotating shaft 2 are fixed to one end of the two connecting plates arranged at intervals, and the other ends of the two connecting plates are fixed to the two ends of the connecting member.

Images