Assembly detection system and method for propellers of low-altitude aircraft and hovercar

Abstract

The invention relates to a low-altitude aircraft and hovercar propeller assembly detection system and method, and belongs to the field of low-altitude aircraft and hovercar detection. The system comprises an assembly module and a detection module, wherein the detection module comprises a paddle space shape and attitude detection algorithm module for automatically detecting the space shape and attitude of the paddle. According to the invention, the assembling efficiency is high, the assembling precision of each related part is effectively ensured, the problems of wrong assembly / neglected assembly and the like are effectively avoided, and the assembling error is controllable; the space shape of the propeller can be completely identified, rapid and effective detection can be realized, and the identification precision is high.

Description

Technical Field

[0001] This invention relates to the field of low-altitude aircraft and flying car testing technology, specifically an assembly testing system and method for low-altitude aircraft and flying car propellers. Background Technology

[0002] A crucial component of the low-altitude economy is low-altitude aircraft (or flying cars, the low-altitude manufacturing industry). Unlike traditional aerospace and general aviation industries, the projected fleet size of low-altitude aircraft is far greater than that of commercial aircraft, helicopters, and general aviation planes. Product assembly and testing require a shift from traditional manual methods to automated / semi-automated system design.

[0003] The propeller / propeller assembly (power system) of a low-altitude aircraft is its core power component, which is generally composed of electric drive and blades. Such products have complex structures, are difficult to manufacture, and have high quality requirements.

[0004] As the low-altitude economy is an emerging industry, most low-altitude aircraft manufacturers currently rely on manual methods for the assembly and testing of propellers / propeller components, lacking automated / semi-automated systems. This results in low overall efficiency, strong subjective factors, high error rates, and high labor costs.

[0005] There are numerous problems in the existing propeller / propeller assembly process for low-altitude aircraft. On the one hand, the assembly process lacks effective integration methods; each assembly step is independent, leading to low assembly efficiency and a tendency for accumulated assembly errors. For example, in traditional assembly, the installation of the blades and hub, and the connection of the motor and hub are performed separately. Differences in the techniques and experience of different operators make it difficult to guarantee assembly accuracy. On the other hand, the inspection process is also inadequate. Common inspection methods often only test some parameters of the propeller / propeller assembly, such as tip height, and cannot provide a comprehensive evaluation of the overall performance after assembly. Moreover, inspection equipment is usually separate from assembly equipment, making the inspection process cumbersome and unable to provide timely feedback on assembly quality issues. Once a problem is discovered, disassembly, adjustment, and re-inspection are required, significantly increasing production and time costs.

[0006] Based on this, the present invention is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide a semi-automated assembly and automated testing product for propellers of low-altitude aircraft and flying cars, solving problems such as low efficiency and low yield in mass production. Specifically, it aims to achieve efficient and precise assembly of each component of the propeller / propeller assembly, while simultaneously enabling rapid and comprehensive testing of various performance parameters of the propeller / propeller assembly during or after assembly, allowing for timely detection and adjustment of assembly problems, thereby improving product quality and production efficiency.

[0008] The technical solution of the present invention is as follows:

[0009] An assembly and testing system for a low-altitude aircraft and flying car propeller includes:

[0010] The assembly module includes a material distribution system and multiple workstations. The material distribution system distributes semi-finished propeller / propeller assembly products or their corresponding parts to different workstations via conveyor lines for assembly into propeller / propeller assembly.

[0011] The inspection module is used to receive the assembled propeller / propeller assembly and automatically inspect the component completion and overall appearance.

[0012] The detection module further includes a blade spatial shape and attitude detection algorithm module, which includes:

[0013] The vision system captures and analyzes images of the propeller blades using a vision camera.

[0014] The laser system uses a linear laser to scan and analyze the structural shape of the propeller blades.

[0015] The data fusion and analysis unit is used to achieve parameter fusion and modeling of the vision system and the laser system, and to complete the automatic detection of the spatial shape and attitude of the propeller blades.

[0016] As a further embodiment of the present invention, the blade spatial shape and attitude detection algorithm module further includes a vision and laser calibration unit.

[0017] The vision and laser calibration unit includes a vision calibration subunit and a laser calibration subunit;

[0018] The visual calibration subunit establishes a coordinate system by photographing the figure-eight mark on the visual calibration frame with a visual camera, and completes the calibration of the intrinsic and extrinsic parameters of the visual camera. The spatial coordinates and shape of the figure-eight mark are measured in advance by a laser tracker or a coordinate measuring machine.

[0019] The laser calibration subunit establishes a plane equation by measuring the calibration plane and positioning hole on the laser calibration frame through the laser system. It completes the position confirmation of the laser system, the correction of the movement step error, and the fitting of the actual movement direction vector of the movement axis by measuring three different positions. The shape and spatial coordinates of the calibration plane and positioning hole are pre-measured by a laser tracker or a coordinate measuring machine.

[0020] As a further aspect of the present invention, the intrinsic parameter calibration is achieved based on the mapping relationship between pixel coordinates (u,v) and visual camera coordinates (X,Y,Z), and the mapping relationship is as follows:

[0021] ;

[0022] , It is the equivalent focal length; , These are the coordinates of the principal point; , It is the distortion correction compensation amount for pixel coordinates;

[0023] Extrinsic parameter calibration via rotation matrix Translation vector , to convert three-dimensional points in the world coordinate system Converted to points in the visual camera coordinate system ;

[0024] .

[0025] As a further embodiment of the present invention, the data fusion and analysis unit includes an image data processing subunit, a laser point cloud extraction subunit, a vision-laser co-registration subunit, and a quality assessment subunit.

[0026] The image data processing subunit uses an adaptive filtering algorithm that fuses local texture gradients and dynamic orientation constraints to process the original image acquired by the vision system, and outputs the filtered target pixel values. and based on Extract visual feature data;

[0027] The laser point cloud extraction subunit uses a neighborhood-adaptive gradient weighting strategy to calculate the local curvature of each laser point cloud sampling point and outputs laser point cloud data with local curvature information.

[0028] The vision-laser collaborative registration subunit uses a visual feature-guided dynamic calibration strategy to transform the laser point cloud to the camera coordinate system, thereby unifying the coordinates of the visual feature data and the laser point cloud data and outputting a fused point cloud.

[0029] The quality assessment subunit calculates multi-dimensional assessment indicators and obtains a comprehensive quality score by comparing the fused point cloud with the standard model of the blade, thereby achieving quality determination.

[0030] As a further aspect of the present invention, the filtering output model of the adaptive filtering algorithm is as follows:

[0031] ;

[0032] In the formula:

[0033] The filtered target pixel value. Target pixel Neighborhood The original pixel values ​​within, It is a texture-aware adaptive neighborhood. Target pixel Corresponding texture-aware adaptive neighborhood The coordinates of any original pixel within the range, These are texture-aware dynamic weighting coefficients;

[0034] The texture-aware dynamic weighting coefficient The expression is:

[0035] ;

[0036] in, For target pixels The local texture principal direction angle, The cosine factor for the texture direction. The sine factor is the texture direction. It is the adaptive spatial variance in the x-direction. It is the adaptive spatial variance in the y-direction;

[0037] Gray-scale sensitivity self-adjustment factor: ; The local gradient magnitude of the target pixel. The maximum gradient magnitude of the entire image;

[0038] Target pixel The original pixel values ​​before adaptive filtering;

[0039] It is the adaptive variance baseline parameter of the grayscale dimension;

[0040] For local texture complexity factor:

[0041] ;

[0042] N is the number of neighboring pixels.

[0043] It is the neighborhood Inner pixel The local gradient magnitude.

[0044] As a further aspect of the present invention, the laser point cloud extraction subunit calculates the local curvature. The formula is:

[0045] ;

[0046] In the formula:

[0047] Let p be the adaptive neighborhood of the laser point cloud sampling point, and let the adaptive neighborhood radius be... ;

[0048] ;

[0049] Based on the radius,

[0050] It is the height gradient magnitude of the laser point cloud sampling point p.

[0051] The maximum global gradient of the point cloud;

[0052] where is the weight coefficient of the neighboring point q;

[0053] This is the weight normalization factor;

[0054] It is the height coordinate value of the laser point cloud sampling point p. It is the adaptive neighborhood of the laser point cloud sampling point p. The height coordinates of any neighboring point q within the interior.

[0055] As a further aspect of the present invention, the transformation formula for the vision-laser co-registration subunit to convert the laser point cloud to the camera coordinate system is as follows:

[0056] ;

[0057] In the formula:

[0058] It is a fused point cloud, which is the laser point cloud transformed into three-dimensional coordinates in the camera coordinate system;

[0059] It is the original point cloud in the laser coordinate system output by the laser system scanning the propeller blades;

[0060] The initial rotation matrix is ​​pre-calibrated. This is the pre-calibrated translation vector;

[0061] It is a rotation correction term.

[0062] ;

[0063] These are feature points on the blade edge. These are the feature points corresponding to the laser point cloud. The correction coefficient is m, where m is the feature point index and n is the number of feature points.

[0064] It is a translation correction term. .

[0065] As a further aspect of the present invention, the multi-dimensional evaluation indicators include spatial position deviation, curvature distribution matching degree, and blade normal vector deviation.

[0066] The spatial position deviation is the mean Euclidean distance between the fused point cloud and the corresponding points in the standard model. ;

[0067] ;

[0068] M represents the number of pairs of points, and t represents the index number of the pair. These are the 3D coordinates of the t-th point in the fused point cloud. These are the three-dimensional coordinates of the t-th point in the standard model.

[0069] It is the Euclidean distance between the corresponding points in the t-th group;

[0070] Curvature distribution matching degree The calculation formula is as follows:

[0071] ;

[0072] It is the covariance of the point cloud curvature and the standard model curvature.

[0073] It is the standard deviation of the curvature of the fused point cloud. It is the standard deviation of the curvature of the standard model;

[0074] The blade normal vector deviation is the mean angle between the blade edge normal vector and the standard model normal vector. ;

[0075] The formula for calculating the overall quality score S is:

[0076] ;

[0077] in, It is the threshold for the maximum deviation of the Euclidean distance. The maximum deviation threshold of the angle between the normal vectors; A, B, and C are all coefficients, and A+B+C=1;

[0078] If S is greater than or equal to the set value, the quality is deemed acceptable; otherwise, it is marked as an abnormal profile / curvature.

[0079] As a further aspect of the present invention, each workstation is equipped with a prompting unit for displaying the assembly sequence and requirements, and an instruction input unit for inputting transmission instructions.

[0080] Secondly, an assembly and inspection method for a low-altitude aircraft and flying car propeller, based on the aforementioned assembly and inspection system for a low-altitude aircraft and flying car propeller, includes the following steps:

[0081] Step 1: After the detection module starts, check whether the visual calibration subunit and the laser calibration subunit have completed pre-calibration; if not calibrated, issue a calibration prompt and recalibrate, measure, and perform fusion calculation; if calibrated, proceed to the next step.

[0082] Step 2: The vision system acquires the raw image of the propeller blade. The acquired image data processing subunit processes the image using an adaptive filtering algorithm and outputs the filtered target pixel value. ,based on Extracting blade edge feature points ;

[0083] Step 3: The laser system scans the original point cloud output by the propeller blades. The local curvature is calculated by the laser point cloud extraction sub-unit using a neighborhood-adaptive gradient weighting strategy. Extract the blade edge feature points corresponding to the laser point cloud. Outputs laser point cloud with local curvature information and corresponding feature points of the laser point cloud. ;

[0084] Step 4: Based on blade edge feature points Laser point cloud corresponding feature points And laser point clouds with local curvature information, and pre-calibrated initial rotation matrix Pre-calibrated translation vector The rotation correction term is calculated using the vision-laser co-registration subunit. Translation correction term The laser point cloud is converted into three-dimensional coordinates in the camera coordinate system, and the fused point cloud is output.

[0085] Step 5: By comparing the fused point cloud with the standard model, the spatial position deviation, curvature distribution matching degree, and blade normal vector deviation are calculated in the quality assessment sub-unit, and the comprehensive quality score S is output.

[0086] If S is greater than or equal to the set value, the quality is deemed acceptable; otherwise, it is marked as an abnormal profile / curvature.

[0087] Compared with the prior art, the beneficial effects of the present invention are:

[0088] 1. Improved Assembly Efficiency: This invention integrates the assembly and testing processes of propellers / propeller components, with each assembly and testing step working in close coordination, reducing waiting time and intermediate steps in traditional assembly and testing processes. It enables rapid and accurate delivery of materials to each installation station. Compared to manual assembly on a workbench, efficiency is significantly improved.

[0089] 2. Ensuring Assembly Quality: By employing a high-precision vision and laser system, combined with a parameter fusion algorithm, this system can accurately measure the dimensions of the rotor hub and blades of the propeller / propeller assembly after installation. This effectively ensures the assembly accuracy of all relevant components and effectively avoids problems such as incorrect assembly or omissions. It can control assembly errors within a very small range, thereby improving product quality and reliability.

[0090] 3. This invention employs a blade spatial shape and attitude detection algorithm module capable of automatically detecting the spatial shape and attitude of the blades. This overcomes the limitations of traditional visual technologies, which can only capture the appearance of a single blade from a specific angle and cannot fully identify the propeller's spatial shape, thus failing to achieve rapid and effective detection. This invention utilizes visual methods for shape assessment and laser technology for attitude assessment, fusing the parameters of both to establish a model. The innovative recognition algorithm is more suitable for propellers / propeller components, offering faster processing speed and higher accuracy. Attached Figure Description

[0091] Figure 1 This is a schematic diagram showing the distribution of the assembly and testing system for the propeller of the low-altitude aircraft and flying car described in this invention.

[0092] Figure 2 This is a schematic diagram of the structure of the vision calibration frame described in this invention;

[0093] Figure 3 This is a schematic diagram of the structure of the laser calibration frame described in this invention. Detailed Implementation

[0094] The present invention will be described in detail below with reference to specific embodiments. These embodiments are merely some, not all, implementations of the present invention. All other implementations obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0095] Example 1

[0096] like Figure 1 As shown, an assembly and testing system for a low-altitude aircraft and flying car propeller includes:

[0097] 1. Assembly module

[0098] 1.1 Material Distribution System: According to different needs, materials (semi-finished products of propellers / propeller assemblies or their corresponding parts) are distributed to different workstations through conveyor line 1, so as to carry out the assembly of each process.

[0099] 1.2 Multiple workstations 2: Each workstation is equipped with a prompting unit and an instruction input unit. The prompting unit is used to display prompts such as assembly sequence and assembly requirements to the operator. The instruction input unit is used by the operator to input transmission instructions after completing the assembly task of the current process, so as to transfer the semi-finished or finished propeller / propeller assembly to the next workstation.

[0100] Figure 1 The loading position 4 is used for loading materials (semi-finished propellers / propeller assemblies or their corresponding parts, etc.).

[0101] 2. Detection Module 3

[0102] After all assembly work of the propeller / propeller assembly is completed, the assembled propeller / propeller assembly is collected by the conveyor link on the conveyor line and automatically inspected. The inspection includes component completion inspection (omission / misassembly of parts and fasteners, etc.), overall appearance inspection, and key parameter inspection of the propeller / propeller assembly. Among them, the key parameters of the propeller / propeller assembly are the focus of the inspection. Taking the blade spatial shape and attitude detection as an example, it is detected by the blade spatial shape and attitude detection algorithm module.

[0103] 3. Blade spatial shape and attitude detection algorithm module

[0104] The spatial shape and attitude of the propeller blades in a propeller / propeller assembly directly affect the system's dynamic characteristics and are the most critical part of the entire system. Inspecting the spatial shape and attitude of the blades is directly related to the quality of the installation process; by measuring its main parameters, the assembly's qualification can be determined. Since the propeller blades are flexible bodies, their spatial attitude is not fixed, and there are many key parameters, with inconsistent evaluation standards across different products. Traditional manual methods require separate measurements of different parameters, which are difficult to operate and have poor accuracy and repeatability. Therefore, this invention introduces an automated solution. Traditional vision technology can only capture the appearance of a single blade from a certain angle and cannot completely identify the propeller's spatial shape, failing to achieve rapid and effective detection. This invention utilizes vision for shape assessment and laser for attitude assessment, fusing these two methods to build a model. The innovative recognition algorithm is more suitable for propellers / propeller assemblies, offering faster processing speed and higher accuracy.

[0105] 3.1 Vision System: Images of the propeller blades are captured and analyzed using a vision camera.

[0106] 3.2 Laser System: The structure and shape of the propeller blade are scanned and analyzed by a linear laser.

[0107] 3.3 Data Fusion and Analysis Unit (Parameter Fusion Algorithm): Used to realize parameter fusion and modeling of the vision system and the laser system, and to complete the automatic detection of the spatial shape and attitude of the propeller blade.

[0108] 3.4 Vision and Laser Calibration Unit:

[0109] 3.4.1 Visual Calibration Subunit:

[0110] The vision system calibrates the intrinsic and extrinsic parameters of the vision camera through a vision calibration subunit. The vision camera captures an image of a figure-eight marker to establish a coordinate system, thus completing the calibration of the intrinsic and extrinsic parameters.

[0111] Since the figure-eight shaped markers are precisely measured in spatial coordinates and shape using laser trackers or coordinate measuring machines, these parameters are known and accurate. For example... Figure 2 As shown, when the vision system takes a picture of the figure-eight mark 6 on the vision calibration frame 5, it can determine the theoretical value of the photograph by accurately judging its spatial coordinates and shape, and compare it with the photographed value to calibrate the intrinsic parameters (focal length, principal point coordinates, distortion coefficient, etc.) and extrinsic parameters (spatial rotation matrix R and translation vector t) of the vision camera.

[0112] Intrinsic parameter calibration is achieved based on the mapping relationship between pixel coordinates (u,v) and visual camera coordinates (X,Y,Z), as follows:

[0113] ;

[0114] , It is the equivalent focal length; , These are the coordinates of the principal point; , It is the distortion correction compensation amount for pixel coordinates;

[0115] Extrinsic parameter calibration via rotation matrix Translation vector , to convert three-dimensional points in the world coordinate system Converted to points in the visual camera coordinate system ;

[0116] .

[0117] 3.4.2 Laser Calibration Subunit:

[0118] like Figure 3 As shown, the laser system is calibrated through the laser calibration subunit. The laser system measures the calibration plane 8 and positioning hole 9 on the laser calibration frame 7 to establish the plane equation. By measuring three different positions, the position of the laser system at different positions is confirmed, the motion step error is corrected, the actual motion direction vector of the motion axis is fitted, and the deviation between the actual value and the theoretical value of the motion step is calculated.

[0119] Similarly, the calibration plane 8 and positioning hole 9 on the laser calibration frame 7 are measured with accurate shape and spatial coordinates by a laser tracker or a coordinate measuring machine, and can be used as calibration parameters.

[0120] 3.5 The data fusion and analysis unit includes an image data processing subunit:

[0121] To address the characteristic that while the blade edges are sharp, they are susceptible to high-frequency noise interference, the image data processing subunit employs an adaptive filtering algorithm that integrates local texture gradients and dynamic directional constraints. This involves deeply coupling the "directional features of the blade surface texture" with the "spatial / grayscale constraints of neighboring pixels" to accurately preserve the blade's edge and texture details while suppressing noise. The filtering output model of this adaptive filtering algorithm is as follows:

[0122] ;

[0123] In the formula:

[0124] The filtered target pixel value is the pixel value of the original image of the propeller captured by the vision system after being filtered by the image data processing subunit. Its core function is to suppress noise, preserve the edge / texture details of the propeller, and provide high-quality image data for subsequent visual feature extraction.

[0125] It is the core intermediate output of visual measurement, not the "visual measurement result" directly used for fusion. Specific logic:

[0126] The "visual measurement results" used later for fusion with laser point cloud data are based on Further extracted visual feature data (such as blade edge feature points) (The three-dimensional coordinate association information corresponding to the texture contour); the final fusion is of "visual feature data" and "laser point cloud data". It is the basic preprocessed data for visual measurement results, which indirectly supports the fusion process.

[0127] Target pixel Neighborhood The original pixel values ​​within, It is a texture-aware adaptive neighborhood, and the neighborhood range is dynamically adjusted by the local texture gradient of the target pixel. The neighborhood shrinks in areas with large texture gradients (such as the edge of a paddle), and expands in areas with gentle textures.

[0128] Target pixel Corresponding texture-aware adaptive neighborhood The coordinates of any original pixel within the range, and the coordinates of the target pixel. The pixel coordinate system (using discrete integer coordinates) of the same image is used to traverse all original pixels in the neighborhood to calculate the filtered output. .

[0129] For texture-aware dynamic weighting coefficients, the core innovation is the introduction of a "local texture direction constraint term" and a "parameter self-adjustment factor," the specific expressions of which are as follows:

[0130] ;

[0131] Innovation Point 1: Spatial distance term constrained by texture direction

[0132] For target pixels The local texture principal direction angle (obtained by statistical analysis of the grayscale gradient direction of the pixel's neighborhood, such as the extension direction of the blade edge);

[0133] The cosine factor for the texture direction. The sine factor is the texture direction. The spatial distance weight decay rate along the main texture direction is reduced (preserving texture continuity), while the decay rate perpendicular to the texture direction is increased (enhancing edge constraints).

[0134] It is the adaptive spatial variance in the x-direction. It is the adaptive spatial variance in the y-direction.

[0135] Its value is positively correlated with the gradient magnitude of the main direction of the texture (the larger the gradient, the smaller the variance, and the stronger the spatial constraint).

[0136] Value selection rule: relative to the target pixel The gradient magnitudes in the main direction of the texture are positively correlated—the larger the gradient magnitude (such as the edge of a paddle), the smaller the variance, and the stronger the spatial constraint on neighboring pixels in the x / y direction; the smaller the gradient magnitude (such as a flat area), the larger the variance, and the looser the spatial constraint, which is suitable for differentiated filtering requirements.

[0137] Innovation Point 2: Dynamically Adjusted Gray-Scale Similarity Item

[0138] Gray-scale sensitivity self-adjustment factor: ; The local gradient magnitude of the target pixel. The maximum gradient magnitude of the entire image; edge regions ( big) Reduce the impact of grayscale differences on weights (to prevent edges from being smoothed).

[0139] For local texture complexity factor:

[0140] ;

[0141] N is the number of neighboring pixels, representing regions with complex textures. Increase the size, and the grayscale constraint is more relaxed (preserving texture details).

[0142] Target pixel The original pixel values ​​(i.e., the original grayscale data of the image) before adaptive filtering, and the original pixel values ​​of neighboring pixels. They belong to the same data type and are the core input of the filtering algorithm;

[0143] It is an adaptive variance baseline parameter for the grayscale dimension; its core function is to adjust the weight decay rate of the grayscale similarity term, serving as a dynamic weighting coefficient for grayscale differences in texture-aware applications. The baseline threshold for the degree of influence, and the local texture complexity factor. Collaboratively achieve differentiated grayscale constraints for different texture regions;

[0144] It is the neighborhood Inner pixel The local gradient magnitude is calculated from the grayscale gradient of that pixel (reflecting the pixel's...). The degree of texture change at the location, and the local gradient magnitude of the target pixel. These are parameters of the same dimension, used to calculate the local texture complexity factor. .

[0145] Innovation Point 3: Texture-Aware Adaptive Neighborhood

[0146] Neighborhood range The calculation method is as follows:

[0147] ;

[0148] In the formula Based on the radius of the neighborhood, the edge region of the blade ( The neighborhood of (large) is reduced to (Avoid blurry edges), flat areas ( The neighborhood of (small) expands to (Enhanced noise suppression).

[0149] The essential difference from existing technologies

[0150] Difference 1: From "Indifferential Spatial Constraint" to "Texture Direction Constraint": Existing bilateral filtering constrains spatial distance is isotropic, while this algorithm... , Achieving differentiated spatial constraints along the blade texture direction better suits the geometric characteristics of the propeller blades, a special industrial part of this invention.

[0151] Difference 2: Introduction , Dynamic factors that are strongly correlated with local textures are no longer preset by humans, but are calculated in real time from the texture features of the image itself.

[0152] Difference 3: The neighborhood range changes dynamically with the local texture gradient, which solves the contradiction of "smoothing noise in a large neighborhood but blurring the edges, and preserving the edges in a small neighborhood but with insufficient noise suppression".

[0153] 3.6 The data fusion and analysis unit also includes a laser point cloud extraction subunit.

[0154] Laser point cloud extraction subunit:

[0155] Scope of application: Curvature extraction can be performed on all surfaces of a propeller (including complex curved surfaces of the blades, hub planes, etc.).

[0156] Extraction performance: It performs better on complex surfaces—its “adaptive neighborhood + gradient weighting” strategy is specifically designed to solve the problem of accurate curvature calculation for complex surfaces, while only basic adaptive adjustment is needed to meet the requirements for simple planes (such as wheel hub planes).

[0157] Output: Not simply curvature values, but "laser point cloud data with local curvature information"—that is, each point cloud sampling point is associated with a corresponding local curvature. It also includes spatial coordinate information, providing complete data for subsequent fusion and quality assessment.

[0158] To address the laser scanning requirements of the complex curved surface of the propeller blades, a neighborhood-adaptive gradient weighting strategy is employed to calculate the local curvature of each point cloud sampling point, thereby enhancing the curvature sensitivity of edge regions. This is known as texture-aware local curvature extraction, and the specific process is as follows:

[0159] After the propeller blades are scanned by the laser system, the neighborhood range of each laser point cloud sampling point p is dynamically adjusted according to its height gradient amplitude, and differentiated weights are assigned to the height deviations of points within the neighborhood, as well as the local curvature. The calculation formula is optimized as follows:

[0160] ;

[0161] In the formula:

[0162] Let p be the adaptive neighborhood of the laser point cloud sampling point, and let the adaptive neighborhood radius be... ; ;

[0163] Based on the radius, It is the height gradient magnitude of the laser point cloud sampling point p. The maximum global gradient of the point cloud;

[0164] The neighborhood of the blade edge (large gradient) shrinks, while the neighborhood of the hub plane (small gradient) expands.

[0165] is the weight coefficient of the neighborhood point q; the larger the gradient (edge ​​region), the higher the weight, which enhances the accuracy of the edge curvature.

[0166] This is a weight normalization factor; it eliminates the influence of neighborhood range differences on curvature.

[0167] This refers to the height coordinates of the laser point cloud sampling point p, indicating the spatial height data along the laser scanning direction, and is used to calculate local curvature. The core foundation;

[0168] It is the adaptive neighborhood of the laser point cloud sampling point p. The height coordinates of any neighborhood point q within the inner region, and Height data, belonging to the same laser scanning coordinate system, is used to calculate the height difference between p and q. This leads to the derivation of the local curvature.

[0169] 3.7 Fusion of visual measurement results and laser point cloud data:

[0170] The core of the fusion is two types of preprocessed feature data:

[0171] Visual measurement results: based on Further extracted visual feature data (such as blade edge feature points) (The three-dimensional coordinate association information of the texture contour), rather than the original image.

[0172] Laser point cloud data: After scanning by the laser system, the "3D point cloud data with local curvature information" (including spatial coordinates and curvature features) is processed by the laser point cloud extraction sub-unit, rather than simply curvature values.

[0173] The essence of fusion: Through the vision-laser collaborative registration subunit, the laser point cloud is transformed into the camera coordinate system, realizing the coordinate unification and information complementarity of "visual feature data" and "laser point cloud data".

[0174] The data fusion and analysis unit also includes a vision-laser co-registration subunit and a quality assessment subunit.

[0175] 3.7.1 To eliminate the influence of propeller installation deviation on registration accuracy, the vision-laser collaborative registration subunit adopts a vision feature-guided dynamic calibration strategy to realize the transformation from laser point cloud to camera coordinate system (different from the "fixed calibration parameters" of existing conventional technologies), that is, vision-laser "dynamic collaborative registration" coordinate fusion, the transformation formula is:

[0176] ;

[0177] In the formula:

[0178] After dynamic collaborative registration, the laser point cloud is transformed into three-dimensional coordinates in the camera coordinate system. (i.e., fused point cloud, which is the three-dimensional coordinates of laser point cloud transformed into camera coordinate system); representing the spatial position information of each sampling point on the blade surface under a unified reference (camera coordinate system), providing three-dimensional data of a unified dimension for subsequent comparison with standard models.

[0179] It is the original point cloud in the laser coordinate system output by the laser system scanning blades (the input data of the laser point cloud extraction sub-unit, which includes spatial position information);

[0180] The above formula means: using a pre-calibrated initial rotation matrix Pre-calibrated translation vector and rotation correction item Translation correction term The original point cloud in the laser coordinate system The point cloud is converted into a fused point cloud in the camera coordinate system, realizing the unification of the coordinate system of visual and laser data.

[0181] The initial rotation matrix is ​​pre-calibrated. This is the pre-calibrated translation vector;

[0182] It is a rotation correction term. ; through blade edge feature points Feature points corresponding to laser point clouds The deviation is compensated in real time for the rotation matrix error. The correction coefficient is used to adjust the compensation intensity for rotational errors; m is the feature point index, and n is the number of feature points.

[0183] It is a translation correction term. Eliminate translational biases in initial calibration to achieve dynamic co-registration between vision and laser.

[0184] and This refers to the dynamic compensation parameter data acquired in real time during the measurement phase.

[0185] These two corrections ( and The data is generated in real time during the measurement phase, not during the calibration phase. The core basis for this is as follows:

[0186] From the perspective of parameter dependence: the calculation of these two correction terms depends on the blade edge feature points acquired in real time during the current measurement process. Laser point cloud corresponding feature points (These are dynamic data from each measurement, not fixed reference data from the calibration phase.)

[0187] From the functional description: It is "real-time compensation for rotation matrix error". It is "to achieve dynamic collaborative registration between vision and laser"—"real-time" and "dynamic" both refer to the instantaneous adjustment during the measurement process, rather than the generation of fixed parameters in the calibration stage.

[0188] The calibration phase only generates the initial values. (Initial rotation matrix) (Translation vectors) (These two are fixed reference parameters); and It is a real-time fine-tuning item for the initial calibration parameters based on the current data during measurement.

[0189] 3.7.2 The quality assessment subunit compares the fused point cloud with the standard model to calculate multi-dimensional assessment indicators and obtain a comprehensive quality score, thereby realizing the quality judgment of the propeller / propeller assembly; the multi-dimensional assessment indicators include spatial position deviation, curvature distribution matching degree, and blade normal vector deviation.

[0190] The fused point cloud contains the propeller's spatial coordinates and local curvature information, and is compared with the standard model of the propeller blades using three-dimensional feature indices:

[0191] 1) Spatial position deviation: This is the mean Euclidean distance between the fused point cloud and the corresponding points in the standard model. ;

[0192] ;

[0193] M represents the number of pairs of points, and t represents the index number of the pair. These are the 3D coordinates of the t-th point in the fused point cloud. These are the three-dimensional coordinates of the t-th point in the standard model. It is the Euclidean distance between the corresponding points in the t-th group;

[0194] 2) Curvature distribution matching degree The calculation formula is as follows:

[0195] ;

[0196] It is the covariance of the point cloud curvature and the standard model curvature.

[0197] It is the standard deviation of the curvature of the fused point cloud. It is the standard deviation of the curvature of the standard model.

[0198] 3) Blade normal vector deviation: the mean angle between the blade edge normal vector and the standard model normal vector. .

[0199] 4) Finally, the overall quality score S:

[0200]

[0201] ( It is the threshold for the maximum deviation of the Euclidean distance. (The maximum deviation threshold of the angle between the normal vectors). When S≥0.85, it is judged as qualified; otherwise, it is marked as an abnormal contour / curvature.

[0202] Example 2

[0203] The detection method of the blade spatial shape and attitude detection algorithm module described in Example 1 includes the following steps:

[0204] Step 1, Initialization and Calibration Check:

[0205] After the detection module is started, it first checks whether the vision system calibration subunit and the laser calibration subunit have completed pre-calibration;

[0206] If not calibrated: A prompt will be issued indicating that calibration is required, and then recalibrate, measure, and perform fusion calculation;

[0207] If already calibrated: Proceed directly to the next step.

[0208] Step 2, Visual Data Acquisition and Processing:

[0209] The vision system acquires raw images of the propeller blades, which are then processed by the image data processing subunit. An adaptive filtering algorithm that fuses local texture gradients and dynamic orientation constraints is used to process and output the filtered target pixel values. ;based on Extracting blade edge feature points (Visual feature data, including 3D coordinates in the camera coordinate system);

[0210] Step 3, Laser Point Cloud Acquisition and Processing:

[0211] The laser system scans the original point cloud in the laser coordinate system output from the propeller blades. The laser point cloud extraction subunit employs a neighborhood-adaptive gradient weighting strategy to calculate the local curvature of each point cloud sampling point p. And extract the blade edge feature points corresponding to the laser point cloud. Output laser point cloud with local curvature information Laser point cloud corresponding feature points .

[0212] Step 4: Visual-laser co-registration subunit processing:

[0213] Blade edge feature points Laser point cloud corresponding feature points and laser point clouds with local curvature information Pre-calibrated initial rotation matrix Pre-calibrated translation vector The rotation correction term is calculated using the vision-laser co-registration subunit. Translation correction term Convert laser point clouds into three-dimensional coordinates in the camera coordinate system. Output the fused point cloud (including 3D coordinates in the camera coordinate system).

[0214] Step 5, Quality Assessment and Judgment:

[0215] Using the fused point cloud from step 4 and the standard model of the propeller blades, the spatial position deviation, curvature distribution matching degree, and blade normal vector deviation are calculated in the quality assessment sub-unit, and the comprehensive quality score S is output.

[0216] If the overall quality score S is greater than or equal to the set value (e.g., 0.85), it is considered to be of acceptable quality; otherwise, it is marked as an abnormal profile / curvature.

[0217] The differences between this invention and the prior art are as follows:

[0218] Curvature calculation: Existing technologies mostly use "average deviation of fixed neighborhood". This invention introduces "adaptive neighborhood + gradient weighting" to adapt to the irregular curved surface of the propeller.

[0219] Coordinate fusion: Existing technologies mostly employ "pre-calibrated fixed coordinates" , This invention incorporates "visual feature-guided dynamic correction" to eliminate installation deviations.

[0220] Quality assessment: Existing technologies mostly use "single-space comparison", while this invention uses multi-dimensional indicators of "space + curvature + normal vector", which results in higher accuracy.

[0221] Furthermore, it should be understood that those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. An assembly and testing system for a propeller of a low-altitude aircraft and flying car, characterized in that, include: The assembly module includes a material distribution system and multiple workstations. The material distribution system distributes semi-finished propeller / propeller assembly products or their corresponding parts to different workstations via conveyor lines for assembly into propeller / propeller assembly. The inspection module is used to receive the assembled propeller / propeller assembly and automatically inspect the component completion and overall appearance. The detection module further includes a blade spatial shape and attitude detection algorithm module, which includes: The vision system captures and analyzes images of the propeller blades using a vision camera. The laser system uses a linear laser to scan and analyze the structural shape of the propeller blades. The data fusion and analysis unit is used to achieve parameter fusion and modeling of the vision system and the laser system, and to complete the automatic detection of the spatial shape and attitude of the propeller blades.

2. The assembly and testing system for a low-altitude aircraft and flying car propeller according to claim 1, characterized in that: The blade spatial shape and attitude detection algorithm module also includes a vision and laser calibration unit. The vision and laser calibration unit includes a vision calibration subunit and a laser calibration subunit; The visual calibration subunit establishes a coordinate system by photographing the figure-eight mark on the visual calibration frame with a visual camera, and completes the calibration of the intrinsic and extrinsic parameters of the visual camera. The spatial coordinates and shape of the figure-eight mark are measured in advance by a laser tracker or a coordinate measuring machine. The laser calibration subunit establishes a plane equation by measuring the calibration plane and positioning hole on the laser calibration frame through the laser system. It completes the position confirmation of the laser system, the correction of the movement step error, and the fitting of the actual movement direction vector of the movement axis by measuring three different positions. The shape and spatial coordinates of the calibration plane and positioning hole are pre-measured by a laser tracker or a coordinate measuring machine.

3. The assembly and testing system for a low-altitude aircraft and flying car propeller according to claim 2, characterized in that: Intrinsic parameter calibration is achieved based on the mapping relationship between pixel coordinates (u,v) and visual camera coordinates (X,Y,Z), as follows: ； , It is the equivalent focal length; , These are the coordinates of the principal point; , It is the distortion correction compensation amount for pixel coordinates; Extrinsic parameter calibration via rotation matrix Translation vector , three-dimensional points in the world coordinate system Points converted to the visual camera coordinate system ; 。 4. The assembly and testing system for a low-altitude aircraft and flying car propeller according to claim 1, characterized in that: The data fusion and analysis unit includes an image data processing subunit, a laser point cloud extraction subunit, a vision-laser co-regulation subunit, and a quality assessment subunit. The image data processing subunit uses an adaptive filtering algorithm that fuses local texture gradients and dynamic orientation constraints to process the original image acquired by the vision system, and outputs the filtered target pixel values. and based on Extract visual feature data; The laser point cloud extraction subunit uses a neighborhood-adaptive gradient weighting strategy to calculate the local curvature of each laser point cloud sampling point and outputs laser point cloud data with local curvature information. The vision-laser collaborative registration subunit uses a visual feature-guided dynamic calibration strategy to transform the laser point cloud to the camera coordinate system, thereby unifying the coordinates of the visual feature data and the laser point cloud data and outputting a fused point cloud. The quality assessment subunit calculates multi-dimensional assessment indicators and obtains a comprehensive quality score by comparing the fused point cloud with the standard model of the blade, thereby achieving quality determination.

5. The assembly and testing system for a low-altitude aircraft and flying car propeller according to claim 4, characterized in that: The filtering output model of the adaptive filtering algorithm is as follows: ； In the formula: The filtered target pixel value. Target pixel Neighborhood The original pixel values ​​within, It is a texture-aware adaptive neighborhood. Target pixel Corresponding texture-aware adaptive neighborhood The coordinates of any original pixel within the range, These are texture-aware dynamic weighting coefficients; The texture-aware dynamic weighting coefficient The expression is: ； in, For target pixels The local texture principal direction angle, The cosine factor for the texture direction. The sine factor is the texture direction. It is the adaptive spatial variance in the x-direction. It is the adaptive spatial variance in the y-direction; Gray-scale sensitivity self-adjustment factor: ; The local gradient magnitude of the target pixel. The maximum gradient magnitude of the entire image; Target pixel The original pixel values ​​before adaptive filtering; It is the adaptive variance baseline parameter of the grayscale dimension; For local texture complexity factor: ； N is the number of neighboring pixels. It is the neighborhood Inner pixel The local gradient magnitude.

6. The assembly and testing system for a low-altitude aircraft and flying car propeller according to claim 4, characterized in that: The laser point cloud extraction subunit calculates local curvature. The formula is: ； In the formula: Let p be the adaptive neighborhood of the laser point cloud sampling point, and let the adaptive neighborhood radius be... ; ； Based on the radius, It is the height gradient magnitude of the laser point cloud sampling point p. The maximum global gradient of the point cloud; where is the weight coefficient of the neighboring point q; This is the weight normalization factor; It is the height coordinate value of the laser point cloud sampling point p. It is the adaptive neighborhood of the laser point cloud sampling point p. The height coordinates of any neighboring point q within the interior.

7. The assembly and testing system for a low-altitude aircraft and flying car propeller according to claim 4, characterized in that: The transformation formula for converting the laser point cloud to the camera coordinate system by the vision-laser co-registration subunit is as follows: ； In the formula: It is a fused point cloud, which is the laser point cloud transformed into three-dimensional coordinates in the camera coordinate system; It is the original point cloud in the laser coordinate system output by the laser system scanning the propeller blades; The initial rotation matrix is ​​pre-calibrated. This is the pre-calibrated translation vector; It is a rotation correction term. ； These are feature points on the blade edge. These are the feature points corresponding to the laser point cloud. The correction coefficient is m, where m is the feature point index and n is the number of feature points. It is a translation correction term. .

8. The assembly and testing system for a low-altitude aircraft and flying car propeller according to claim 4, characterized in that: The multi-dimensional evaluation indicators include spatial position deviation, curvature distribution matching degree, and blade normal vector deviation; The spatial position deviation is the mean Euclidean distance between the fused point cloud and the corresponding points in the standard model. ; ； M represents the number of pairs of points, and t represents the index number of the pair. These are the 3D coordinates of the t-th point in the fused point cloud. These are the three-dimensional coordinates of the t-th point in the standard model. It is the Euclidean distance between the corresponding points in the t-th group; Curvature distribution matching degree The calculation formula is as follows: ； It is the covariance of the point cloud curvature and the standard model curvature. It is the standard deviation of the curvature of the fused point cloud. It is the standard deviation of the curvature of the standard model; The blade normal vector deviation is the mean angle between the blade edge normal vector and the standard model normal vector. ; The formula for calculating the overall quality score S is: ； in, It is the threshold for the maximum deviation of the Euclidean distance. The maximum deviation threshold of the angle between the normal vectors; A, B, and C are all coefficients, and A+B+C=1; If S is greater than or equal to the set value, the quality is deemed acceptable; otherwise, it is marked as an abnormal profile / curvature.

9. The assembly and testing system for a low-altitude aircraft and flying car propeller according to claim 1, characterized in that: Each workstation is equipped with a prompting unit for displaying the assembly sequence and requirements, and an instruction input unit for inputting transmission commands.

10. An assembly and inspection method for a propeller of a low-altitude aircraft and a flying car, characterized in that: The assembly and testing system for a low-altitude aircraft and flying car propeller according to any one of claims 1-9 includes the following steps: Step 1: After the detection module starts, check whether the visual calibration subunit and the laser calibration subunit have completed pre-calibration; if not calibrated, issue a calibration prompt and recalibrate, measure, and perform fusion calculation; if calibrated, proceed to the next step. Step 2: The vision system acquires the raw image of the propeller blade. The acquired image data processing subunit processes the image using an adaptive filtering algorithm and outputs the filtered target pixel value. ,based on Extracting blade edge feature points ; Step 3: The laser system scans the original point cloud output by the propeller blades. The local curvature is calculated by the laser point cloud extraction sub-unit using a neighborhood-adaptive gradient weighting strategy. Extract the blade edge feature points corresponding to the laser point cloud. Outputs laser point cloud with local curvature information and corresponding feature points of the laser point cloud. ; Step 4: Based on blade edge feature points Laser point cloud corresponding feature points And laser point clouds with local curvature information, and pre-calibrated initial rotation matrix Pre-calibrated translation vector The rotation correction term is calculated using the vision-laser co-registration subunit. Translation correction term The laser point cloud is converted into three-dimensional coordinates in the camera coordinate system, and the fused point cloud is output. Step 5: By comparing the fused point cloud with the standard model, the spatial position deviation, curvature distribution matching degree, and blade normal vector deviation are calculated in the quality assessment sub-unit, and the comprehensive quality score S is output. If S is greater than or equal to the set value, the quality is deemed acceptable; otherwise, it is marked as an abnormal profile / curvature.

Images