Automatic cutting method for aerial ground service lifting platform fused with image processing

Abstract

The invention discloses an automatic cutting method for an aviation ground service lifting platform fused with image processing, and particularly relates to the technical field of aviation maintenance auxiliary equipment. A multi-angle infrared image and a depth image of a target operation area of the aircraft are obtained, target contour features are recognized, and a three-dimensional contour feature set is generated; projecting the feature set to a three-dimensional model of the aircraft, constructing a point cloud picture of an operation area, estimating a height difference value and an edge curvature, and generating a basic matching profile curve of the lifting platform; the platform size parameters and the operation task boundary are combined, an optimal cutting track sequence is generated, and a numerical control cutting device is controlled to complete automatic cutting of the structural plate; after cutting is completed, contact pressure distribution of the edge of the platform is obtained, and error comparison and compensation cutting are carried out in combination with an aircraft model; when the fitting precision reaches a set threshold value, a completion signal is automatically output, the method has the advantages of being accurate in recognition, intelligent in cutting and self-adaptive in fitting, and automation and reliability of aviation ground service operation are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of aviation maintenance auxiliary equipment technology, and specifically to an automatic cutting method for aviation ground support lifting platforms that integrates image processing. Background Technology

[0002] Aviation ground lifting platforms, as crucial auxiliary equipment for supporting the maintenance, repair, assembly, and emergency rescue operations of large aircraft, are widely used in military and civil aviation bases. However, existing lifting platforms generally suffer from a mismatch between platform size and work area during operation preparation and scenario adaptation. This is particularly problematic in remote or irregularly shaped airframe maintenance tasks requiring rapid response, where the platform cannot achieve precise fit to complex contours, leading to low operational efficiency, high risk, and even potential damage to the aircraft's surface structure.

[0003] Currently, most mainstream lifting platform structures are fixed modular assemblies, lacking intelligent recognition and dynamic cutting capabilities. When encountering irregular spaces such as the complex air intake contours of military fighter jets, irregularly shaped pylon areas of UAVs, and rotor hub connection nodes of rotary-wing unmanned helicopters, the platform needs to be manually measured and cut, resulting in long deployment cycles, poor operational accuracy, and a high risk of collisions between the platform and the aircraft due to misjudgment, causing irreversible damage.

[0004] Especially in low-light environments at night, temporary deployment sites, and combat readiness states, the platform must have image guidance, adaptive recognition, and automatic cropping capabilities to meet the urgent needs of modern ground operations for speed, high security, and modular intelligence.

[0005] Therefore, there is an urgent need for an automatic cutting method for aviation ground support lifting platforms that integrates image processing, enabling adaptive cutting of platform structures in complex scenarios and improving the intelligence and reliability of aviation ground support operations. Summary of the Invention

[0006] The purpose of this invention is to provide an automatic cutting method for aviation ground support lifting platforms that integrates image processing, so as to overcome the shortcomings of the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an automatic cutting method for an aviation ground support lifting platform that integrates image processing, comprising:

[0008] Acquire multi-angle image data of the target operation area of ​​the aircraft, the image data including at least one set of infrared images and depth images, and perform feature recognition on the image data by fusing convolutional neural networks and edge extraction algorithms to generate a target contour feature set;

[0009] The target contour feature set is projected onto the three-dimensional spatial model of the aircraft surface to construct a three-dimensional point cloud map of the working area, and the height difference and edge curvature of different areas are estimated to generate the basic matching contour curve of the lifting platform.

[0010] Based on the basic matching contour curve of the lifting platform, and combined with the initial size parameters of the platform and the boundary of the task, an optimal cutting trajectory sequence for platform structure cutting is generated.

[0011] The optimal cutting trajectory sequence is input into the CNC cutting device to control the platform structure plate to automatically cut according to the optimized path;

[0012] The system provides feedback on the contact pressure distribution after the platform structure is cut, and compares the pressure distribution with the three-dimensional model of the aircraft surface in real time, adjusting the compensation cutting parameters for parts where the cutting error exceeds the threshold.

[0013] When the platform structure after cutting is detected to have a fitting accuracy with the target working area that meets the preset tolerance threshold, a cutting completion signal is output, and the platform automatically enters the locked deployment state.

[0014] Preferably, the step of constructing the three-dimensional point cloud map of the work area includes:

[0015] Based on the spatial coordinates of each three-dimensional edge point in the target contour feature set, a pre-established three-dimensional spatial model of the aircraft surface is called, and rigid coordinate transformation and scale matching processing are performed to make the edge points and the aircraft surface model in the same reference coordinate system.

[0016] After completing the coordinate alignment, the nearest neighbor search method is used to map each edge point to the corresponding mesh cell of the aircraft surface model, forming a set of surface constraint projection points.

[0017] Density interpolation and discrete point compensation are performed on the surface constraint projection point set to generate a continuously distributed three-dimensional point cloud map of the working area.

[0018] Preferably, the step of generating the basic matching contour curve of the lifting platform includes:

[0019] Based on the three-dimensional point cloud map of the work area, the point cloud is spatially partitioned according to the preset grid size, the maximum and minimum height values ​​of the edge points in each grid unit are extracted, and the height difference of the area is calculated.

[0020] For the edge point distribution of adjacent grid cells, fit curve segments and use the rate of change of the second derivative to estimate the edge curvature, and determine the continuity and abruptness of the edge changes;

[0021] Based on the joint weighted scoring rule of height difference and edge curvature, a representative set of edge path points is extracted as the control nodes of the lifting platform's cutting curve.

[0022] The control nodes are connected in spatial order, and the B-spline curve fitting method is used to generate the matching contour curve of the lifting platform foundation.

[0023] Preferably, the step of generating the optimal cutting trajectory sequence for platform structure trimming includes:

[0024] Based on the basic matching contour curve of the lifting platform, the control node sequence is extracted and the Euclidean distance between adjacent nodes is calculated to determine the local change amplitude of the curve.

[0025] Based on the initial size parameters of the lifting platform, a platform boundary constraint model is constructed. The matching contour curve and the platform boundary are geometrically trimmed to remove curve segments that exceed the processing range of the platform.

[0026] Based on the task boundary information, a multi-objective path optimization algorithm is adopted, setting path continuity, minimum tool travel distance and material utilization rate as optimization objectives, and reconstructing the cutting path topology;

[0027] The optimized path nodes are smoothed five times using Bezier curves to generate the optimal cutting trajectory sequence that the platform can process.

[0028] Preferably, the step of comparing the pressure distribution with the three-dimensional model of the aircraft surface in real time and adjusting the compensation cutting parameters for the portion of the cutting error exceeding the threshold includes:

[0029] After the platform structure is cut, real-time pressure data of the contact area between the platform edge and the aircraft surface is obtained, and a pressure distribution map is generated.

[0030] The peak contact pressure coordinates in the pressure distribution map are spatially registered with the three-dimensional model of the aircraft surface to calculate the deviation vector field between the cut contour and the theoretical fitting boundary.

[0031] If the magnitude of the local deviation vector exceeds the cutting error threshold, the initial cutting trajectory points of the corresponding edge segment are extracted.

[0032] Based on the deviation direction and tangential component, the coordinates of the trajectory control points are corrected, the cutting path is updated, and the machining control instructions for compensating cutting are regenerated.

[0033] Preferably, in the step of performing feature recognition on image data by fusing convolutional neural networks and edge extraction algorithms, a multi-scale convolutional neural network model based on the U-Net structure is used to segment the edge region of the infrared image. The neural network model includes at least five encoder layers and five decoder layers, and features are extracted through convolutional groups with a kernel size of 3×3.

[0034] Preferably, the edge extraction algorithm adopts the Canny edge detection algorithm, which performs image binarization processing based on the infrared image edge heatmap. The binarization threshold is automatically determined by Otsu's method, and the edge connection adopts an 8-neighborhood connection strategy.

[0035] Preferably, in the step of constructing a three-dimensional point cloud map of the work area, an inverse distance weighted interpolation method is used to fill in points in sparse areas, and the interpolation weight is the reciprocal of the square of the Euclidean distance between the interpolation point and the neighboring edge point.

[0036] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0037] 1. This invention combines infrared and depth images, and incorporates a fusion convolutional neural network and edge extraction algorithm to perform multi-angle, non-contact 3D contour recognition of the target operating area of ​​an aircraft, achieving high-precision modeling of complex aircraft shape boundaries. Compared to traditional methods that rely on manual measurement or fixed template matching, this invention can stably acquire target contour features under different aircraft types, different operating positions, and complex lighting conditions. Furthermore, it generates a basic contour curve that matches the aircraft surface height through height difference and edge curvature analysis, fundamentally improving the accuracy and adaptability of the elevator platform's cutting path and significantly reducing initial cutting errors.

[0038] 2. This invention constructs a closed-loop control process centered on optimal cutting trajectory generation, CNC automatic cutting, and pressure feedback compensation. By comparing the contact pressure distribution after cutting with a 3D model of the aircraft surface in real time, it can automatically identify and correct local areas exceeding the error threshold, and finally automatically complete the locking deployment when the bonding accuracy requirements are met. This technical solution not only realizes the automation and intelligence of the entire process of platform structure cutting and deployment, but also effectively improves cutting accuracy, work efficiency, and work safety, reduces manual intervention and repetitive processing costs, and is particularly suitable for aviation ground support scenarios with high precision and high reliability requirements. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0040] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] For examples, please refer to Figure 1 As shown in this embodiment, an automatic cutting method for an aviation ground support lifting platform that integrates image processing includes:

[0043] Acquire multi-angle image data of the target operation area of ​​the aircraft. The image data includes at least one set of infrared images and depth images. Perform feature recognition on the image data by fusing convolutional neural networks and edge extraction algorithms to generate a target contour feature set.

[0044] This invention provides an image processing method for high-precision identification of the three-dimensional contour of an aircraft target operating area in a field environment, which is a preliminary step in the automatic cropping path calculation of a lifting platform. This method relies on multiple sub-processes, including image acquisition, depth estimation, neural network recognition, and edge coordinate reconstruction, forming a complete image processing chain, and ultimately outputting a set of three-dimensional contour features suitable for cropping path generation.

[0045] The image acquisition device consists of an infrared imager (e.g., FLIR A65) and a depth camera (e.g., Intel RealSense D435). The infrared imager captures surface temperature distribution images, with an infrared imaging band ranging from 7.5 to 13 micrometers. The depth camera uses active infrared structured light, supporting a depth ranging accuracy of ±2 mm and a maximum field of view of 85° × 58°. Each image acquisition unit is installed on the top, front, and left sides of the lifting platform and can be rotated horizontally by ±90° via a mechanical turntable to ensure that the acquisition angle covers the target area.

[0046] Image acquisition process: Before the operation begins, the elevator platform is moved to a position below the target area of ​​the aircraft; after the image acquisition program is started, each acquisition unit acquires infrared and depth images from three angles (forward, oblique, and lateral), named IRi and DEPi (i=1~3); each image is time-stamped and spatial registration correction is performed to ensure that the infrared image and the corresponding depth map correspond one-to-one at the pixel level; all image data are stored in the image buffer after acquisition and enter the next stage of processing.

[0047] To improve image recognition accuracy, the acquired images are processed as follows:

[0048] Infrared image preprocessing: Thermal image noise reduction is performed using a two-dimensional Gaussian filter with a kernel size of 5×5 and a standard deviation σ of 2.0; Max-Min normalization is performed on the image to map pixel values ​​to between 0 and 1 for subsequent input into the neural network model; Contrast-Limited Adaptive Histogram Equalization (CLAHE) is performed to improve the thermal gradient contrast of the target edge region.

[0049] Depth image processing: Confidence assessment is performed on all depth pixels in each depth image, and pixels with confidence scores below 0.85 are removed; bilateral filtering is performed on the remaining depth images with parameters set as follows: spatial standard deviation of 5 pixels and intensity standard deviation of 25 pixel units; a depth map-infrared map matching mapping table is constructed to achieve accurate matching of depth values ​​corresponding to each infrared image pixel.

[0050] The process of building a convolutional neural network model and identifying edges includes:

[0051] Model architecture design: A multi-scale convolutional neural network model based on the U-Net structure is adopted to perform segmentation and recognition of aircraft edge regions in images; the network input is the registered infrared image (size 1280×720), and the output is an edge probability heatmap of the same size as the input; the encoder consists of 5 convolutional layers, each layer including 2 convolutional operations, 1 batch normalization layer and 1 ReLU activation function; the kernel size of the 1st to 5th convolutional layers is (3,3) respectively, and the number of output channels is 32, 64, 128, 256 and 512; the decoder uses transposed convolution to restore the spatial dimension and skips connections with the encoder to achieve fine-grained restoration of edge information.

[0052] Training method: A public image dataset containing complex aircraft component structures (such as AeroStruct-5K) was used, with a total of 5000 training samples. The labels were manually drawn edge segmentation masks. The loss function adopted a hybrid form of Dice coefficient loss and cross-entropy loss, and the learning rate was initially set to 0.0005. The network was trained for 100 rounds on an NVIDIA graphics card platform with a batch size of 16. During training, the images were randomly rotated and mirrored.

[0053] Inference output processing: Threshold the edge heatmap generated by inference, use Otsu's method to automatically select the binarization threshold, and extract the initial edge mask; perform morphological closing operation (structuring element size is 5×5) on the edge mask to fill the edge break area; use the contour tracking algorithm to extract all closed boundary curves and calculate the length of each boundary, retaining only the main contour with a length greater than 50 pixels.

[0054] 3D edge coordinate reconstruction and feature set generation include:

[0055] Pixel-coordinate mapping method: Each infrared image pixel coordinate (u, v) corresponds to the depth value d of the pixel in its matching depth image; according to the camera intrinsic parameter matrix, the pixel coordinates are converted into three-dimensional coordinates (X, Y, Z) in the camera coordinate system, and the calculation formula is: X=(u-cx)×d / fx; Y=(v-cy)×d / fy; Z=d; where (cx, cy) are the optical center coordinates, (fx, fy) are the focal length parameters, and d is the depth value, in millimeters.

[0056] Unified coordinate transformation: The above camera coordinate points are transformed to the lifting platform coordinate system through the extrinsic parameter matrix; the lifting platform coordinate system is defined as follows: the origin is located at the center of the platform, the X-axis points to the front of the platform, the Y-axis points to the left, and the Z-axis points upward; the edge 3D points from different viewpoints are uniformly registered, duplicate points are removed, and the error tolerance is set to 2 mm.

[0057] Contour feature set generation: The final three-dimensional point set composed of all edge points is denoted as: Fset = {Pi(xi,yi, zi) | i ∈ [1, N]}; where Pi is the position coordinate of the i-th edge point in the platform coordinate system, and N is the total number of edge points.

[0058] The target contour feature set Fset provides a direct input basis for the subsequent construction of the lifting platform cutting path. Its edge continuity and three-dimensional accuracy are controlled within ±2 mm, which meets the requirements of high-precision operation.

[0059] The target contour feature set is projected onto the three-dimensional spatial model of the aircraft surface to construct a three-dimensional point cloud map of the working area, and the height difference and edge curvature of different areas are estimated to generate the basic matching contour curve of the lifting platform.

[0060] In this invention, after image feature extraction is completed, the three-dimensional edge points in the target contour feature set need to be accurately mapped onto the aircraft surface model, and a point cloud map and contour curve of the working area for cropping path calculation are generated accordingly. Specifically, the steps include:

[0061] The three-dimensional spatial model of the aircraft surface is a pre-acquired CAD format model or a laser scanning point cloud reconstruction model, containing high-precision surface mesh data, with coordinate units in millimeters, and the reference coordinate system is defined as the body coordinate system;

[0062] Each 3D edge point Pi in the target contour feature set is represented as (xi, yi, zi), with the initial coordinate system being the lifting platform coordinate system. A rigid coordinate transformation is applied to the edge point set, with the transformation matrix including a rotation matrix R and a translation vector T, and the transformation formula is as follows: X′=R×X+T; where X is the original coordinate vector, [xi, yi, zi]. TX′ is the transformed coordinate, R is a 3×3 direction cosine matrix, and T is a 3×1 translation vector.

[0063] Meanwhile, a scale matching factor S is applied to the edge point set to ensure that the coordinate units of the transformed edge points are consistent with the aircraft model. S is obtained by linear fitting of the edge point cloud bounding box and the corresponding region bounding box of the model.

[0064] The set of edge points after coordinate transformation is denoted as P′. For each mesh cell in the 3D model mesh of the aircraft surface, its center point coordinates are defined as Cj. For each edge point P′i, the Euclidean distance between it and each Cj is calculated, and the mesh cell corresponding to the minimum distance is taken as its mapping target to complete the spatial projection of the edge point onto the model surface. The set of all edge points that have been projected is defined as the surface constraint projection point set Q={Qk(xk,yk,zk)}, where Qk is the k-th edge point after projection. To improve the spatial integrity of the edge region, the point set Q is resampled with a regular mesh, and the mesh resolution is 10 mm. In the sparse edge point region, the inverse distance weighted interpolation method is used to fill in the missing points. The interpolation function is: f(P)=Σ(wi×Qi) / Σwi, wi=1 / d(P,Qi). 2 Where P is the interpolation point position, Qi is the neighboring edge point, d is the spatial distance function, and wi is the inverse square distance weighting factor; after interpolation, all points are combined to obtain a continuously distributed 3D point cloud map of the work area Dset={Dm(xm,ym,zm)}.

[0065] The three-dimensional point cloud map Dset is spatially divided into equal-sized cubic grid blocks with a unit side length of 30 mm. The Z-axis coordinates of all points Dm in each cubic grid unit are extracted, and the maximum height Hmax and minimum height Hmin of the unit are calculated. The height difference ΔH is defined as: ΔH = Hmax - Hmin. ΔH is used to describe the degree of edge undulation of the region and is recorded as a grid height attribute.

[0066] Between adjacent spatial grids, the coordinate sequence of edge points Dm is extracted and subjected to local curve fitting; the curve is fitted using the three-point median method, and the curvature K is calculated, defined as: K=|x'·y"-y'·x"| / (x' 2 +y' 2 ) (3 / 2) Where x′, y′ are the first derivatives and x″, y″ are the second derivatives, obtained by the finite difference method; the maximum curvature of each curve segment is used as the edge abruptness index of the region for subsequent control point extraction.

[0067] For all grid cells, the height difference threshold Th is set to 8 mm and the curvature threshold Tk is set to 0.02. If a grid cell ΔH>Th and K>Tk, its center point is marked as a potential control node. All potential nodes are clustered and merged, and redundant nodes with a distance of less than 15 mm are removed to obtain the edge path point set Cset={Cn(xn,yn,zn)}. Each Cn is an important control point of the lifting platform cutting path.

[0068] Sort the control points in Cset according to their XY plane projection order to form an ordered node sequence;

[0069] Perform B-spline curve fitting on the ordered node sequence, using the fourth-order non-uniform rational B-spline (NURBS) form;

[0070] Set the fitting smoothness error to no more than ±3 mm, and output the basic matching contour curve Lcurve of the lifting platform.

[0071] The Lcurve will serve as the geometric basis for generating subsequent CNC cutting paths, possessing characteristics such as high adaptability to aircraft surface edges, continuous manufacturing capability, and controllable error.

[0072] Based on the basic matching contour curve of the lifting platform, and combined with the initial size parameters of the platform and the boundary of the task, an optimal cutting trajectory sequence for platform structure cutting is generated.

[0073] In this invention, after generating the basic matching contour curve of the lifting platform, in order to achieve precise structural cutting and efficient manufacturing, it is necessary to further generate the optimal cutting trajectory based on the contour curve to meet the objectives of dimensional boundaries, processing accuracy, and path optimization. The specific steps are as follows:

[0074] Control node extraction method: In the matching contour curve of the lifting platform foundation, each curve segment is defined by B-spline control points; extract all control points to form a control node sequence, denoted as Cset={Ci(xi,yi,zi)}, where Ci is the three-dimensional coordinate of the i-th node; the nodes are arranged in the order of the curve to ensure the continuity of the spatial topology.

[0075] Euclidean distance calculation and local change judgment: For any two adjacent control nodes Ci and C(i+1), calculate their Euclidean distance Di. The calculation method is as follows: If the change in Di for three consecutive times exceeds the preset amplitude threshold Td (Td is 1.5 times the previous average), the region is marked as a high-frequency region of curve change; this judgment is used for node encryption processing in subsequent path reconstruction to ensure path accuracy.

[0076] Initial dimension parameter input: The initial dimension parameters of the lifting platform include the length L, width W, and cutting reachable boundary S of the processing area; the processing boundary of the platform is defined as a two-dimensional space rectangle, and the boundary constraint model is denoted as Bbox={(x,y)|0≤x≤L,0≤y≤W}.

[0077] Geometric clipping implementation: Project the matching contour curve onto the XY plane; use the boundary constraint model Bbox to perform Boolean operations (using line segment-rectangle intersection judgment) on curve segments that exceed the boundary to cut the curve; add control nodes at the endpoints of the clipped curves to ensure geometric continuity, and record the clipping points as processing boundary control points.

[0078] The task boundary is provided by the input task model, including the minimum processing error requirement Emin (in millimeters) and the path tolerance range Tp (in millimeters); at the same time, the material layout diagram is input to obtain the material usage boundary, which constitutes the overall processing constraint set.

[0079] An improved ant colony algorithm is used for path search and reconstruction. The objective function Ftotal consists of three parts:

[0080] F1: Path continuity score, which calculates the rate of change of direction between control points;

[0081] F2: Minimize the total path length Ltotal;

[0082] F3: Maximize material utilization rate (effective path coverage area / total material area).

[0083] The overall objective function is in the form: Ftotal = α × F1 + β × F2 + γ × (1 - Urate), where α = 0.4, β = 0.3, and γ = 0.3 are empirical weighting coefficients. The algorithm iteration count is set to 100 times, and the convergence accuracy is set to an error variation of less than 0.5 mm.

[0084] The output path control node sequence Popt={Pj(xj,yj,zj)} has been optimized to meet the requirements of processing continuity and path economy.

[0085] For the path control node sequence Popt, perform five Bezier curve fittings in groups of five.

[0086] The control points for each curve are from Pj to P(j+4), and the fitted curve Bj(t) is expressed as: Bj(t)=Σ(C(5,k)×(1-t)) (5-k) ×t k ×P(j+k)), where t∈[0,1], k=0~5.

[0087] Curve continuity and smoothness adjustment: Adjacent Bezier curve segments are connected by first derivatives to ensure tangent continuity; the distance between control points is not less than 5 mm to ensure stable movement path of the machining tool; if any fitted segment deviates from the original path by more than 1.5 mm, the segment is reverted and reconstructed to improve fitting accuracy.

[0088] All Bezier curve segments are sequentially spliced ​​together to form the optimal cutting trajectory sequence Tseq; Tseq serves as the trajectory input for the motion control of the subsequent CNC cutting device, meeting the target requirements of high adaptability, high continuity, and high material utilization.

[0089] The optimal cutting trajectory sequence is input into the CNC cutting device, which controls the platform structure plate to automatically cut according to the optimized path.

[0090] In this invention, after generating the optimal cutting trajectory sequence, the trajectory needs to be accurately input into the CNC cutting device to control the lifting platform structural plate to automatically cut along the path, ensuring that the edge of the formed structure can be precisely fitted to the target working area of ​​the aircraft. This process includes steps such as trajectory data conversion, CNC command generation, plate fixing preprocessing, tool path execution, and dynamic adjustment of the cutting state, specifically implemented as follows:

[0091] The optimal cutting trajectory sequence Tseq consists of multiple fifth-order Bezier curve segments, each of which is expressed as a set of control points.

[0092] To adapt to the input format accepted by standard CNC cutting devices, the Bezier curve is discretized into equally spaced path points, with the path point spacing set to 1 mm.

[0093] All path points are arranged in curve order to form a cutting path point sequence Pcut={Pi(xi,yi)}, where i is the discrete point number.

[0094] CNC instruction generation: Based on the path point sequence Pcut, control instructions are generated according to the standard G-code format; for each path point Pi, the corresponding linear interpolation instruction G01 Xxi Yyi Ff is generated, where F is the feed rate (in millimeters / minute), which is set to 800 by default; a tool down movement instruction G00 Z-5 is added at the starting point of the trajectory, and a tool up movement instruction G00Z+5 is added at the ending point to ensure the safe entry and exit of the tool from the sheet metal.

[0095] The lifting platform structure is made of aluminum alloy modular flat plate with a thickness of 8 mm and an anodized surface. The plate is placed above the cutting platform and fixed by a four-point vacuum adsorption device to ensure stability and zero-position reference during the cutting process.

[0096] After starting the CNC cutting device, perform the automatic zero-point return operation to confirm the mechanical origin position; use the laser pointer to mark the starting point coordinates at the lower left corner of the plate and align them with the starting point of the cutting trajectory. The coordinate error shall not exceed ±0.5 mm.

[0097] The cutting process uses an alloy steel straight cutter with high-frequency vibration and a cutter radius of 2 mm, which is suitable for precision machining of aluminum alloy sheets. The cutter is mounted on a CNC cutting head and has XY plane dual-axis linkage capability, with a path following accuracy of no less than ±0.2 mm.

[0098] Cutting execution process: The path points are moved sequentially according to the generated G-code sequence; the tool moves continuously between each path segment, maintaining a stable speed and uniform feed; if the tool vibration amplitude exceeds the set threshold (e.g., ±0.3 mm) during the cutting process, the control program automatically decelerates to 60% and issues an alarm.

[0099] During the cutting process, the cutting resistance changes are monitored in real time by a tool load sensor;

[0100] When the load exceeds the normal range (e.g., the current value exceeds 120% of the rated value), a path pause command is triggered, and the abnormal location point is recorded.

[0101] For the abnormal path segment, resample the path points and extend the toolpath outward by 0.5 mm;

[0102] Automatically generate compensation G-code segments and append them for execution to ensure the integrity of the cropped outline.

[0103] Finally, after the above steps, the platform structural plates are precisely cut, and the resulting structural edges are spatially precisely adapted to the target area of ​​the aircraft. This ensures good structural continuity, boundary integrity, and strength retention, providing a reliable processing foundation for subsequent platform assembly and deployment.

[0104] The system provides feedback on the contact pressure distribution after the platform structure is cut, and compares the pressure distribution with the three-dimensional model of the aircraft surface in real time, adjusting the compensation cutting parameters for parts where the cutting error exceeds the threshold.

[0105] In this invention, after the initial cutting of the platform structure, to ensure precise alignment between the cut edges and the target operating area of ​​the aircraft, error correction is achieved through pressure sensing, 3D model comparison, and trajectory compensation. The specific steps are as follows:

[0106] Flexible pressure sensors (model TekscanFlexiForce A401) are evenly distributed every 20 mm along the cut edges of the platform structure, with the sensor sensing surface integrated on the back of the plate.

[0107] Each sensor pre-positions itself in space and records its three-dimensional coordinates (xi,yi,zi) in the platform coordinate system;

[0108] After the platform comes into contact with the aircraft surface, it collects the instantaneous contact pressure value Pi of each sensor, in kilopascals, with the sampling frequency set to 100 Hz and the sampling duration to 2 seconds.

[0109] The collected pressure values ​​are visualized as point clouds according to their corresponding coordinates to form an initial pressure point set Pset={Pi,xi,yi,zi}.

[0110] Spatial expansion is performed on the point set using inverse distance weighted interpolation, with the interpolation function weights defined as wi = 1 / di. 2 , where di is the Euclidean distance between the current interpolation point and Pi;

[0111] The interpolation results are mapped onto the XY plane to form a two-dimensional pressure distribution map Pmap(x,y), where the pixel value corresponding to each position (x,y) represents the unit contact pressure at that position.

[0112] Calculate the average pressure value Pavg for all locations in the Pmap;

[0113] Set the pressure peak threshold Tp=1.3×Pavg, extract all pixel positions that satisfy P(x,y)≥Tp, and map them back to the original three-dimensional coordinates to form the pressure peak point set Ppeak={Qj(xj,yj,zj)}.

[0114] Ppeak is taken as the source point cloud, and the point set of the corresponding region of the 3D model of the aircraft surface is taken as the target point cloud, denoted as Smodel;

[0115] Point cloud registration is performed using the Iterative Closest Point (ICP) algorithm, and the steps include:

[0116] Initialize the rotation matrix R as the identity matrix and the translation vector T as the zero vector;

[0117] For each point Qj in Ppeak, search for its nearest point Sk in Smodel;

[0118] Minimize the error function E=Σ∥RQj+T-Sk∥ 2 The singular value decomposition method is used to solve for R and T;

[0119] Update the coordinates Q'j=RQj+T to achieve rigid registration from Ppeak to Smodel.

[0120] For each registered Q′j, calculate its normal projection point Sj* on Smodel;

[0121] Construct a deviation vector Ej=Q'j-Sj* to represent the spatial error between the actual contact point and the theoretical bonding point;

[0122] Calculate the magnitude |Ej| of Ej to determine the error amplitude, while retaining its directional information.

[0123] Set the error tolerance threshold Te = 2 mm, and the judgment condition is |Ej|>Te; mark all edge segments corresponding to Ej that meet this condition, and find their corresponding initial cutting trajectory point set Cerr = {Cn(xn,yn)}; Cerr is the path node set of the error exceeding the limit area, which serves as the basis for subsequent compensation cutting reconstruction.

[0124] For each point Cn in Cerr, its corresponding deviation vector En is decomposed into the following directions:

[0125] Normal component En⊥: points in the direction of the normal to the aircraft surface;

[0126] Tangential component En∥: Located in the tangential direction of the contour boundary;

[0127] The correction strategy is to offset the coordinates of Cn in the opposite direction along the normal direction. The offset is equal to the magnitude of En⊥. The updated coordinates are: Cn'=Cn-α×En⊥, where α is the compensation coefficient, which is set to 1.0 by default.

[0128] Combine all Cn′ into a new set of control points Ccorr;

[0129] Perform Bezier curve fitting on Corr four times in sequence to reconstruct the local compensation clipping path Pcorr;

[0130] Numerical control instructions are regenerated based on Pcorr, using G01 linear interpolation, with each path segment having a length of no more than 1 millimeter;

[0131] The instruction is appended to the end of the original G code to perform local correction processing operations.

[0132] The above method enables automated compensation cutting based on measured pressure feedback, combined with aircraft surface model error mapping, and precise correction through trajectory control points. This effectively solves structural error problems caused by inconsistent sheet stress, processing offset, or bonding curvature, and improves the bonding accuracy and operational reliability of the platform structure.

[0133] When the platform structure after cutting is detected to have a fitting accuracy with the target working area that meets the preset tolerance threshold, a cutting completion signal is output, and the platform automatically enters the locked deployment state.

[0134] In this invention, to ensure that the platform structure can safely and stably fit the target operating area of ​​the aircraft after automatic cutting, it is necessary to determine the fitting accuracy between the platform edge and the aircraft surface. If the preset tolerance threshold is met, a cutting completion signal is output, and the system automatically enters a locked deployment state to ensure the reliability and structural stability of subsequent operations. The specific implementation of this process is as follows:

[0135] After performing the compensation trimming operation, the contact pressure data between the platform structure edge and the aircraft surface is repeatedly collected to generate an updated pressure distribution map Pmapfinal.

[0136] Extract the actual contact point set Qfinal={Qj(xj,yj,zj)} corresponding to the pressure peak point and perform point cloud registration with the surface point set Smodel of the three-dimensional model of the aircraft surface; calculate the Euclidean distance from all Qj points to their nearest theoretical contact point Sj*, and define it as the contact error ej.

[0137] The fit tolerance threshold Tfit is set to 1.5 mm to measure edge fit performance;

[0138] The average value (eavg) and maximum value (emax) of all fitting errors (ej) are calculated, and the judgment criteria are as follows:

[0139] If eavg≤1.0 mm and emax≤1.5 mm, then the fitting accuracy is considered to meet the deployment requirements.

[0140] Once the bonding accuracy meets the set threshold, the controller triggers a "cutting complete" status signal.

[0141] The signal is a digital high level, the flag bit is set to 1, and it is synchronously written into the task status register for subsequent operations to read.

[0142] The controller sends this signal to the main operation control unit via the bus interface;

[0143] At the same time, update the status of the current cropping task number to "completed", and record the completion timestamp and trajectory number for data traceability.

[0144] Upon receiving the cutting completion signal, the control program automatically enters the deployment and locking phase; it activates the electric locking hook actuators deployed along the edge of the platform, deploying a set of locking hooks every 50 millimeters, each set including an electrically controlled clamping block and an elastic limit arm;

[0145] The electronically controlled clamping block driver is activated, pushing the locking hook toward the aircraft surface until the clamping force sensor feedback reaches more than 5 Newtons; the limit arm locks to prevent structural movement due to vibration or airflow, achieving a stable fit between the platform structure and the aircraft; after all locking point status feedback is confirmed to be correct, the controller outputs a "deployment complete" flag, and the task transitions to the operation preparation state.

[0146] Through the above process, the quality of platform structure cutting can be automatically determined and the structure can be automatically locked without relying on manual intervention. It has the advantages of high safety, high efficiency and closed-loop control of the whole process, and is especially suitable for high-risk and high-requirement aviation ground service environments.

[0147] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An automatic cutting method for an aviation ground support lifting platform that integrates image processing, characterized in that: include: Acquire multi-angle image data of the target operation area of ​​the aircraft, the image data including at least one set of infrared images and depth images, and perform feature recognition on the image data by fusing convolutional neural networks and edge extraction algorithms to generate a target contour feature set; The target contour feature set is projected onto the three-dimensional spatial model of the aircraft surface to construct a three-dimensional point cloud map of the working area, and the height difference and edge curvature of different areas are estimated to generate the basic matching contour curve of the lifting platform. Based on the basic matching contour curve of the lifting platform, and combined with the initial size parameters of the platform and the boundary of the task, an optimal cutting trajectory sequence for platform structure cutting is generated. The optimal cutting trajectory sequence is input into the CNC cutting device to control the platform structure plate to automatically cut according to the optimized path; The system provides feedback on the contact pressure distribution after the platform structure is cut, and compares the pressure distribution with the three-dimensional model of the aircraft surface in real time, adjusting the compensation cutting parameters for parts where the cutting error exceeds the threshold. When the platform structure after cutting is detected to have a fitting accuracy with the target working area that meets the preset tolerance threshold, a cutting completion signal is output, and the platform automatically enters the locked deployment state.

2. The automatic cutting method for an aviation ground support lifting platform according to claim 1, characterized in that: The steps for constructing the 3D point cloud map of the work area include: Based on the spatial coordinates of each three-dimensional edge point in the target contour feature set, a pre-established three-dimensional spatial model of the aircraft surface is called, and rigid coordinate transformation and scale matching processing are performed to make the edge points and the aircraft surface model in the same reference coordinate system. After completing the coordinate alignment, the nearest neighbor search method is used to map each edge point to the corresponding mesh cell of the aircraft surface model, forming a set of surface constraint projection points. Density interpolation and discrete point compensation are performed on the surface constraint projection point set to generate a continuously distributed three-dimensional point cloud map of the working area.

3. The automatic cutting method for an aviation ground support lifting platform according to claim 2, characterized in that: The steps for generating the basic matching contour curve of the lifting platform include: Based on the three-dimensional point cloud map of the work area, the point cloud is spatially partitioned according to the preset grid size, the maximum and minimum height values ​​of the edge points in each grid unit are extracted, and the height difference of the area is calculated. For the edge point distribution of adjacent grid cells, fit curve segments and use the rate of change of the second derivative to estimate the edge curvature, and determine the continuity and abruptness of the edge changes; Based on the joint weighted scoring rule of height difference and edge curvature, a representative set of edge path points is extracted as the control nodes of the lifting platform's cutting curve. The control nodes are connected in spatial order, and the B-spline curve fitting method is used to generate the matching contour curve of the lifting platform foundation.

4. The automatic cutting method for an aviation ground support lifting platform according to claim 1, characterized in that: The step of generating the optimal cutting trajectory sequence for platform structure trimming includes: Based on the basic matching contour curve of the lifting platform, the control node sequence is extracted and the Euclidean distance between adjacent nodes is calculated to determine the local change amplitude of the curve. Based on the initial size parameters of the lifting platform, a platform boundary constraint model is constructed. The matching contour curve and the platform boundary are geometrically trimmed to remove curve segments that exceed the processing range of the platform. Based on the task boundary information, a multi-objective path optimization algorithm is adopted, setting path continuity, minimum tool travel distance and material utilization rate as optimization objectives, and reconstructing the cutting path topology; The optimized path nodes are smoothed five times using Bezier curves to generate the optimal cutting trajectory sequence that the platform can process.

5. The automatic cutting method for an aviation ground support lifting platform according to claim 1, characterized in that: The step of comparing the pressure distribution with the three-dimensional model of the aircraft surface in real time and adjusting the compensation cutting parameters for the portion of the cutting error exceeding the threshold includes: After the platform structure is cut, real-time pressure data of the contact area between the platform edge and the aircraft surface is obtained, and a pressure distribution map is generated. The peak contact pressure coordinates in the pressure distribution map are spatially registered with the three-dimensional model of the aircraft surface to calculate the deviation vector field between the cut contour and the theoretical fitting boundary. If the magnitude of the local deviation vector exceeds the cutting error threshold, the initial cutting trajectory points of the corresponding edge segment are extracted. Based on the deviation direction and tangential component, the coordinates of the trajectory control points are corrected, the cutting path is updated, and the machining control instructions for compensating cutting are regenerated.

6. The automatic cutting method for an aviation ground support lifting platform according to claim 1, characterized in that: In the step of performing feature recognition on image data by fusing convolutional neural networks and edge extraction algorithms, a multi-scale convolutional neural network model based on the U-Net structure is used to segment the edge region of the infrared image. The neural network model includes at least five encoder layers and five decoder layers, and features are extracted through convolutional groups with a kernel size of 3×3.

7. The automatic cutting method for an aviation ground support lifting platform according to claim 6, characterized in that: The edge extraction algorithm adopts the Canny edge detection algorithm, which performs image binarization processing based on the infrared image edge heatmap. The binarization threshold is automatically determined by Otsu's method, and the edge connection adopts the 8-neighborhood connection strategy.

8. The automatic cutting method for an aviation ground support lifting platform according to claim 2, characterized in that: In the step of constructing a three-dimensional point cloud map of the work area, an inverse distance weighted interpolation method is used to fill in points in sparse areas. The interpolation weight is the reciprocal of the square of the Euclidean distance between the interpolation point and the neighboring edge point.

Images