Intelligent control method and system for full-automatic coating line of automobile parts

Abstract

The invention provides an intelligent control method and system for a full-automatic coating line of automobile parts, and relates to the field of control systems. The method comprises the following steps: acquiring three-dimensional point cloud data of an automobile part; based on the three-dimensional point cloud data, determining a high-complexity area on the surface of the automobile part as an area to be coated; calculating a covariance matrix of a normal vector in the to-be-coated area by adopting a principal component analysis algorithm, and extracting a principal direction feature vector through the covariance matrix; determining a rotation coordinate value of the rotation axis based on the main direction feature vector; and the sprayer is controlled to move to the autorotation coordinate value, and the automobile parts are coated. By the adoption of the technical scheme, when the surfaces of the automobile parts with different geometrical characteristics are sprayed, the geometrical characteristics of the automobile parts are combined, the rotation coordinate values are determined, when the sprayer is located at the rotation coordinate values, the uniform pattern layer can be completely sprayed to the surfaces of the automobile parts as much as possible, and the spraying effect is guaranteed.

Description

Technical Field

[0001] This application relates to the field of control systems, and in particular to an intelligent control method and system for a fully automated painting line for automotive parts. Background Technology

[0002] Automotive parts, also known as auto components, auto parts, or auto spare parts, refer to the components that make up the various units of a car and all consumable materials that serve the car. These include engine parts, transmission parts, braking parts, steering parts, running system parts, electrical and instrument system parts, body and accessories, and interior and exterior trim. The body itself includes bumpers, doors, fenders, windshields, pillars, seats, center console, hood, trunk lid, sunroof, roof, door locks, armrests, floor, and door sills.

[0003] However, automotive parts require coating during manufacturing, such as doors, hoods, trunk lids, sunroofs, roofs, door locks, and armrests. The diverse geometries of these parts present a fundamental challenge in coating inspection. Furthermore, the varying geometric characteristics of different parts, with their vastly different surface curvatures, unevenness, and spatial distribution, directly lead to deviations in the spraying viewing angle when the rotation axis is not properly determined. This results in certain critical surface areas not being adequately coated, affecting the overall coating effect.

[0004] Therefore, how to adapt the rotation axis to different shaped parts and realize the spraying of different shaped parts based on the rotation axis has become the key issue of fully automated coating. Summary of the Invention

[0005] In view of this, this application provides a load balancing control method and system for an intelligent controller, the main purpose of which is to achieve the spraying of parts of different shapes based on the rotation axis, so as to fully spray automotive parts.

[0006] To achieve the above objectives, the first aspect of this application discloses an intelligent control method for a fully automated painting line for automotive parts. The fully automated painting line includes a central controller and a sprayer. The method is applied to the central controller, which is connected to the sprayer. The method includes:

[0007] Acquire 3D point cloud data of automotive parts;

[0008] Based on the three-dimensional point cloud data, the high-complexity areas on the surface of automotive parts are identified as areas to be coated.

[0009] The principal component analysis algorithm is used to calculate the covariance matrix of the normal vector in the area to be coated, and the principal direction feature vector is extracted through the covariance matrix.

[0010] Based on the principal direction feature vector, the rotation coordinate values ​​of the rotation axis are determined;

[0011] The sprayer is controlled to move to the rotation coordinate value to perform painting on the automotive parts.

[0012] A second aspect of this application provides an intelligent control system for a fully automated painting line for automotive parts, the system comprising:

[0013] The acquisition module is used to acquire 3D point cloud data of automotive parts;

[0014] The first determining module is used to determine the high-complexity areas on the surface of automotive parts as areas to be coated based on the three-dimensional point cloud data.

[0015] The calculation module is used to calculate the covariance matrix of the normal vector in the area to be coated using the principal component analysis algorithm, and extract the principal direction feature vector through the covariance matrix;

[0016] The second determining module is used to determine the rotation coordinate value of the rotation axis based on the main direction feature vector;

[0017] The control module is used to control the sprayer to move to the rotation coordinate value and perform painting on the automotive parts.

[0018] A third aspect of this application provides an electronic device, comprising:

[0019] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform any of the methods disclosed in the first aspect.

[0020] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0021] In summary, the technical solution disclosed in this application includes: acquiring three-dimensional point cloud data of automotive parts; determining highly complex areas on the surface of automotive parts as areas to be coated based on the three-dimensional point cloud data; using principal component analysis (PCA) to calculate the covariance matrix of the normal vectors in the areas to be coated, and extracting principal direction eigenvectors through the covariance matrix; determining the rotation coordinates of the rotation axis based on the principal direction eigenvectors; and controlling the sprayer to move to the rotation coordinates to perform coating on the automotive parts. By adopting the technical solution of this application, when spraying automotive parts with different geometric characteristics, the rotation coordinates are determined in conjunction with the geometric characteristics of the automotive parts. When the sprayer is located at the rotation coordinates, a uniform coating layer can be sprayed onto the surface of the automotive parts as completely as possible, ensuring the coating effect.

[0022] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This paper presents a flowchart of an intelligent control method for a fully automated painting line for automotive parts, according to an embodiment of this application.

[0026] Figure 2 This paper presents a structural diagram of an intelligent control system for a fully automated painting line for automotive parts, as provided in an embodiment of this application. Detailed Implementation

[0027] To better understand the above-mentioned objectives, features, and advantages of this application, the solution of this application will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0028] To address the issue of how to adaptably select the rotation axis for parts of different shapes and how to achieve the spraying of parts of different shapes based on the rotation axis, this application provides the following embodiments to solve the above problems:

[0029] This embodiment provides an intelligent control method for a fully automated painting line for automotive parts, such as... Figure 1 The diagram shown is a flowchart of the method in this embodiment. The painting of automotive parts is performed by a paint sprayer. The technical solution of this application embodiment is used to clarify the control of the paint sprayer, so that the paint sprayer, according to the control, can achieve the painting of automotive parts. The method of this embodiment may specifically include the following steps:

[0030] Step 101: Obtain the 3D point cloud data of the automotive parts.

[0031] In 3D point cloud processing, 3D point cloud data can be generated by scanning the surface of an object using LiDAR. Taking the inspection of a car body surface as an example, LiDAR scanning can generate point cloud data containing millions of points, recording the 3D coordinates of the car body surface.

[0032] Voxel mesh filtering divides the point cloud into a fixed-size voxel mesh (e.g., with a side length of 0.01 meters) and downsamples by taking the average value of points within each voxel. For example, the original point cloud may contain 5 million points, which can be reduced to 500,000 points after downsampling, significantly reducing the computational load while preserving the overall geometry of the automotive parts, thus contributing to improved efficiency in subsequent processing.

[0033] The original point cloud may contain 5 million 3D point cloud data points. After downsampling, it is reduced to 500,000 points, which significantly reduces the amount of computation while preserving the overall geometry, which helps to improve the efficiency of subsequent processing.

[0034] In one possible implementation, statistical filtering is used to remove noise. Assuming there are outliers in the point cloud caused by ambient light interference, the density of each point's neighborhood can be calculated, and points with a density below a threshold (e.g., fewer than 5 points in the neighborhood) can be removed. For example, in a vehicle body point cloud, noisy points may appear at edges or in occluded areas; statistical filtering can produce a smooth point cloud that retains true surface features and improves data quality.

[0035] Furthermore, when collecting and computing point cloud data, GPUs can be used for concurrent computing, which greatly shortens the processing time of point cloud data and leaves sufficient time resources for subsequent calculations.

[0036] Step 102: Based on the 3D point cloud data, identify the high-complexity areas on the surface of the automotive parts as the areas to be coated.

[0037] In some embodiments, based on three-dimensional point cloud data, high-complexity areas on the surface of automotive parts are identified as areas to be coated, including:

[0038] The k-nearest neighbor algorithm is used to determine the neighborhood point set of each 3D point cloud data. Using the neighborhood point set, at least one surface normal vector is calculated on the surface of the automotive part, resulting in the surface normal vector distribution on the automotive part surface. Using the surface normal vector distribution, the least squares method is used to fit the local area to be coated, obtaining the curvature distribution data corresponding to the surface normal vector distribution. A mesh generation method is used to map the curvature distribution data into a two-dimensional matrix. The matrix element values ​​of the two-dimensional matrix are determined. If the matrix element values ​​exceed a preset curvature threshold, the local area to be coated corresponding to the two-dimensional matrix is ​​determined as a high-complexity area, and at least one high-complexity area is determined as the area to be coated.

[0039] The k-nearest neighbor algorithm is used to determine the neighborhood point set based on smooth point clouds. Setting k to 10, the 10 nearest neighbors of each point are calculated, forming the neighborhood point set. Based on this point set, the surface normal vector distribution is calculated through covariance analysis. For example, the normal vector distribution is uniform in smooth areas of a car body surface, while it varies significantly in edge areas. The normal vector distribution reflects the surface geometry and provides a basis for subsequent curvature calculations. For instance, based on the normal vector distribution, the least squares method is used to fit the local surface, extracting the principal curvatures K1 and K2, as well as the Gaussian curvature K. The principal curvature reflects the degree of bending of the surface in the principal direction, while the Gaussian curvature is the product of K1 and K2. For example, in flat areas of the car body, K1 and K2 are close to 0, and the Gaussian curvature K is also close to 0, while the curvature values ​​at the door handle are higher, with K1 possibly being 0.5, K2 0.3, and K 0.15, reflecting local complex geometry. This helps in identifying surface feature points.

[0040] Among them, when using the k-nearest neighbor algorithm to determine the neighborhood point set, CUDA or other GPU computing platforms can be used for parallel computing to accelerate the search of the k-nearest neighbor algorithm and quickly determine the neighborhood point set.

[0041] In one embodiment, curvature distribution data is mapped to a two-dimensional matrix. Assume the vehicle body surface is divided into a 100×100 grid, with each grid element recording the average Gaussian curvature of its corresponding region. If a preset curvature threshold of 0.1 is set, grids with curvature exceeding 0.1 are marked as high-complexity regions. For example, areas like door handles or wheel hubs, due to their high curvature, are marked as high-complexity regions, forming a surface complexity evaluation matrix. This matrix intuitively reflects geometric complexity, facilitating subsequent design optimization or defect detection. The preset curvature threshold can be calibrated through pre-experiments for different parts of different vehicle models, avoiding redundant calculations. For instance, by pre-calculating the Gaussian curvature distribution based on 100 samples of the same part from the same vehicle model, using the 95th percentile as the threshold, 95% of complex regions are covered, ultimately resulting in a preset threshold of 0.12.

[0042] In one possible implementation, a region growing algorithm is used to identify at least one connected high-complexity region based on a high-complexity region distribution matrix. The region growing algorithm uses a high-curvature grid as a seed point and checks the curvature values ​​of adjacent grids. If they meet a similarity condition (e.g., curvature difference less than 0.1), they are classified as the same connected region.

[0043] For example, a highly complex mesh at the edge of a car door can be extended to the entire edge region, forming a connected region. Boundary point sets are extracted from these connected regions, and boundary coordinates are calculated to obtain the set of boundary coordinates for the high-curvature region.

[0044] Step 103: Using principal component analysis algorithm, calculate the covariance matrix of the normal vector in the area to be coated, and extract the principal direction feature vector through the covariance matrix.

[0045] The surface normal vectors and boundary coordinate sets are obtained from the 3D point cloud data. Principal component analysis is used to reduce the dimensionality of the surface normal vector set to obtain the principal direction feature vectors.

[0046] For example, in the processing of point cloud data for automotive parts, obtaining surface normal vectors and boundary coordinates from the point cloud is a crucial step in analyzing geometric features. Point cloud data is typically generated through LiDAR scanning and contains three-dimensional coordinate information of the automotive part's surface. Surface normal vectors reflect the local surface orientation at each point and can be calculated through local plane fitting.

[0047] For example, for point cloud data of a car wheel hub, a point and its neighborhood are selected, and a local plane is fitted using the least squares method to obtain the normal vector. Assuming the normal vector of a point on the wheel hub surface is (0.7, 0.2, 0.7), reflecting the surface orientation at that point, the boundary coordinates are extracted using an edge detection algorithm.

[0048] For example, the three-dimensional coordinates of points on the edge of the wheel hub can form a boundary point set containing thousands of points, recording the precise location of the edge.

[0049] In one possible implementation, principal component analysis (PCA) is used to reduce the dimensionality of the surface normal vector set. PCA extracts the principal direction eigenvectors by calculating the covariance matrix of the normal vectors.

[0050] For example, the normal vector set of a wheel hub point cloud may contain tens of thousands of vectors. Through principal component analysis, principal direction eigenvectors, such as (0.9, 0.1, 0.4), can be obtained, reflecting the main geometric directions of the wheel hub surface. These vectors can serve as preliminary direction vectors for the rotation axis, which helps in subsequent axis determination and improves analysis efficiency.

[0051] Furthermore, when using principal component analysis for dimensionality reduction, the advantage of the principal component analysis algorithm supporting concurrent computation can be utilized to complete the processing of tens of thousands of vectors within 100ms, ensuring rapid dimensionality reduction.

[0052] Step 104: Determine the rotation coordinates of the rotation axis based on the principal direction feature vector.

[0053] In some embodiments, the initial direction vector of the sprayer is determined based on the principal direction feature vector.

[0054] Calculate the projected distance from at least one boundary point of the area to be coated to the preliminary direction vector; determine the boundary points whose projected distance is less than a preset threshold as axis neighborhood points, and obtain an axis neighborhood point set composed of at least one axis neighborhood point; fit the axis neighborhood point set into a spatial straight line using the least squares method, and determine the spatial coordinate points through which the spatial straight line passes; calculate the local feature mean and variance of each spatial coordinate point, and use a weighted average algorithm to combine and calculate the local feature mean and variance, and determine the rotation coordinate value of the rotation axis.

[0055] Determine the initial direction vector of the component's rotation axis. Based on the initial direction vector, calculate the projected distance from each point in the boundary coordinate set to the initial direction vector. If the projected distance is less than a preset threshold, it is marked as a neighborhood point of the axis, resulting in a neighborhood point set. For the neighborhood point set, fit a spatial line using the least squares method. Extract the direction vector of the line and the spatial coordinates of the points passing through it from the fitting result, obtaining the spatial coordinate set of the rotation axis. Based on the spatial coordinate set, calculate the mean and variance of the local features of each coordinate point. Use a weighted average algorithm to determine the spatial coordinates of the rotation center, obtaining the rotation center point of the rotation axis.

[0056] Based on the initial direction vector, the projected distance from the boundary coordinate points to this vector is calculated. Assuming the initial direction vector is (0.9, 0.1, 0.4), the projected distance threshold is set to 0.05. For the boundary point set along the wheel hub edge, the projected distance is calculated point by point. If the projected distance of a point is 0.03, which is less than the threshold, it is marked as a point in the axis neighborhood. This ultimately forms the axis neighborhood point set, which may contain hundreds of points concentrated in the rotationally symmetric region of the wheel hub. The projected distance threshold needs to be set according to the minimum step accuracy of the sprayer to ensure that the rotation axis positioning error is less than or equal to the sprayer's repeatability accuracy.

[0057] In one possible implementation, a spatial straight line is fitted using the least squares method for the set of points in the neighborhood of the axis.

[0058] For example, the neighborhood point set of the wheel hub's axis contains 500 points. Using the least squares method, the straight line direction vector is fitted as (0.95, 0.05, 0.3), and the point coordinates are (10, 20, 5). These parameters constitute the spatial coordinate set of the rotation axis, accurately describing the position of the wheel hub's rotation axis.

[0059] For example, based on a set of spatial coordinates, the mean and variance of local features for each coordinate point are calculated. For the hub axis point set, local features may include the spatial distribution density of the points, with a mean of 0.02 and a variance of 0.005, reflecting the distribution stability of the point set. A weighted average algorithm, combined with the point distribution density, is used to calculate the spatial coordinates of the rotation axis.

[0060] Step 105: Control the sprayer to move to the rotation coordinate value and perform painting on the automotive parts.

[0061] By adopting the technical solution of this embodiment, when spraying on the surface of automotive parts with different geometric characteristics, the rotation coordinate value is determined in combination with the geometric characteristics of the automotive parts. When the sprayer is located at the rotation coordinate value, a uniform layer can be sprayed onto the surface of the automotive parts as completely as possible, ensuring the spraying effect.

[0062] When spraying large automotive parts, the sprayer needs to perform the spraying process while constantly moving. Therefore, when controlling the sprayer, the central controller needs to consider not only the sprayer's rotation coordinates, but also the motion differences between the sprayer and the automotive parts, such as rotation angle or rotation speed.

[0063] Therefore, when controlling the sprayer to perform painting on automotive parts, this embodiment further includes: the fully automated automotive parts painting line also includes a camera and a motion device, the motion device being connected to the automotive parts and used to control the rotation of the automotive parts, and a central controller being connected to both the motion device and the camera to control the sprayer to move to the rotation coordinate value and perform painting on the automotive parts, including:

[0064] After the control motion device moves the sprayer to the rotation coordinate value, the control camera captures the area to be coated, generating initial image data. A deep learning image recognition algorithm is used to identify the initial image data, determine the position to be coated and the coating requirement value at the position to be coated. Based on the position to be coated and the coating requirement value, an adaptive control algorithm is used to calculate the dynamic adjustment parameters of the motion device. The control motion device controls the rotation of the automotive parts according to the dynamic adjustment parameters, and controls the sprayer to perform coating on the automotive parts.

[0065] The image data captured by the camera first needs to undergo image preprocessing. Image preprocessing technology performs noise reduction and enhancement on the initial image data, and adjusts the acquisition resolution according to the boundary coordinates of complex areas to obtain a standardized image sequence of the coating surface.

[0066] The first image is obtained from the initial image data. A median filtering algorithm is used to calculate the median value of each pixel and its neighboring pixels, replacing the values ​​of noisy pixels to obtain the second image. For the second image, a histogram equalization method is used to adjust the pixel grayscale distribution and enhance the contrast of the coating area, resulting in the third image. Based on the third image, an edge detection algorithm is used to calculate the boundary coordinate point set of complex regions. If the curvature of the boundary coordinate point set is greater than a preset threshold, it is marked as a complex region, obtaining the location data of the complex region. For the location data of the complex region, an adaptive resolution adjustment method is used to dynamically adjust the camera acquisition resolution according to the density of the boundary coordinate point set, resulting in a standardized image sequence of the coating surface.

[0067] In one possible implementation, when acquiring the first image from the initial image data, a high-resolution industrial camera is typically used to capture raw image data of the surface of the automotive parts, including texture and grayscale information of the coated areas.

[0068] For example, an initial image of a car door surface with a pixel resolution of 1920×1080 and a grayscale value range of 0-255 records the brightness distribution of the surface coating. This image data provides the foundation for subsequent processing, ensuring the integrity of details. When applying a median filtering algorithm to the first image, the core principle is to smooth noise by calculating the median between each pixel and its neighboring pixels.

[0069] For example, for a pixel on the surface of a car door, the grayscale values ​​of its 3×3 neighborhood pixels are 150, 152, 255, 148, 149, 151, 147, 153, and 150. After sorting, the median value is 150. Replacing the original pixel value 255 with this value eliminates isolated noise points. This method effectively preserves edge details and avoids blurring key features.

[0070] In one possible implementation, when performing histogram equalization on the second image, the aim is to redistribute gray values ​​to enhance the contrast of the coated area.

[0071] For example, if the grayscale values ​​of a certain area on the car door surface are concentrated between 100-150, after equalization, they can be expanded to 50-200, making the details of the coating area more apparent. This processing can highlight the texture differences in the coating area, facilitating subsequent analysis. When using edge detection algorithms on the third image, the boundary coordinate point set of complex areas can be extracted by calculating the grayscale gradient.

[0072] For example, the curved surface at the edge of a car door may exhibit drastic grayscale changes. An edge detection algorithm identifies the set of boundary points and calculates their curvature. If the curvature value of a certain boundary segment reaches 0.8, exceeding a preset threshold of 0.5, it is marked as a complex region. This method can accurately locate the transition region of the curved surface, providing a basis for subsequent high-resolution data acquisition.

[0073] Initial image data is acquired from the camera. A convolutional neural network algorithm is used to calculate the pixel distribution features of each image, extracting texture and shape features of the area to be coated, thus obtaining the coating location and its coating requirement value. A clustering algorithm is used to group areas based on the coating location and coating requirement value. If the distribution density of the coating location and coating requirement value features in a group is greater than a preset threshold, the area to be coated is marked as a complex spraying area, obtaining the location data of the complex spraying area. An edge detection algorithm is used to calculate the boundary coordinate point set of the complex spraying area, and the continuity of the boundary coordinate point set is determined by combining pixel distribution features to identify key detection areas. For key detection areas, a grayscale analysis method is used to calculate the difference between the average pixel grayscale value within the area and a preset standard coating grayscale template. If the difference is greater than a preset threshold, it is judged as a low quality level, obtaining the coating quality level.

[0074] For example, when obtaining the first image sequence from a standardized image sequence of coating surfaces, multiple frames containing the coating surface can be directly extracted from the preprocessed image data. Assuming the image sequence of an automotive component surface is processed with a pixel resolution of 1920×1080, each frame records the grayscale information of the area to be coated. A convolutional neural network algorithm can be used to analyze pixel distribution characteristics and extract the coating location and coating requirements.

[0075] For example, the network model scans the image through convolution operations to generate feature maps, capturing the texture details and shape contours of defects such as scratches and bubbles.

[0076] For example, given an image of a car door surface, the model can identify linear scratches with a width of 2 mm and a length of 10 mm, or circular bubbles with a diameter of 1 mm, forming a defect feature dataset. This defect feature dataset can then be used to describe coating requirements.

[0077] For example, the configuration of a convolutional neural network algorithm is as follows:

[0078] First, the input data is a standardized image of the coating surface, with a fixed size of 224×224 pixels and 3 channels (RGB).

[0079] Feature Extraction Backbone Network: This network consists of 5 convolutional layers, with the kernel size decreasing layer by layer: 7×7→5×5→3×3→3×3→3×3. Shallow layers use larger kernels (7×7, 5×5) to broadly capture coarse-grained features in the image, such as large texture areas and contour information. Deeper layers use smaller kernels (3×3) to refine the extraction of local, subtle features, such as scratches, bubbles, or uneven coating areas. All convolutional layers have a stride of 1 and are padded with pixels to maintain the spatial resolution of the feature maps, ensuring no loss of detail. Each convolutional layer is followed by a batch normalization layer and a ReLU activation function to accelerate the training process and improve the model's expressive power.

[0080] Next, the network's max pooling layer is responsible for performing the downsampling task. The pooling kernel size is 2×2, with a stride of 2, performing standard 2x downsampling to progressively compress the feature map size, increase the receptive field, and enhance the spatial invariance of the features.

[0081] The number of channels varies, increasing layer by layer: 32 → 64 → 128. As network depth increases and spatial size decreases, the number of channels multiplies, enabling the network to represent more complex patterns in a higher-dimensional feature space.

[0082] After the last convolutional layer, a Global Average Pooling (GAP) layer is used to reduce the dimensionality of each feature map to a scalar, replacing the traditional fully connected layer, which greatly reduces the number of parameters and effectively prevents overfitting. Finally, a Soft Max classification layer is added to output the probability distribution of coating requirements for each position to be coated. The coating requirements can be classified as: no processing required, minor touch-up coating, and focused spraying.

[0083] The model's loss function can be the cross-entropy loss function, suitable for multi-class classification tasks. The optimizer uses the Adam optimizer with an initial learning rate of 0.001, and a learning rate decay strategy can be adopted based on validation set performance. The training hyperparameter batch size is set to 32, and the number of training epochs is dynamically determined using an early stopping strategy.

[0084] In one possible implementation, a clustering algorithm is used to group the defect feature dataset according to the similarity of texture and shape features.

[0085] For example, K-means clustering can group scratch-like defects into one group and bubble-like defects into another. If the feature distribution density within a group is greater than a preset threshold, such as 10 defect points per square centimeter, it is marked as a complex defect region.

[0086] For example, dense scratches on the edge of a car door, with a distribution density of 12 points per square centimeter, exceed the threshold of 10 points and are thus marked as a complex defect area. Specifically, for the location data of complex defect areas, edge detection algorithms can be used to calculate the set of boundary coordinate points.

[0087] For example, Canny edge detection is used to identify the boundary contours of complex defect areas and generate a set of coordinate points.

[0088] For example, a scratch area on a car door surface might have a boundary point set containing 100 evenly spaced points, indicating good boundary continuity. By combining pixel distribution features and analyzing the grayscale variation trend of the point set, the integrity of the boundary can be determined. If the spacing between the points is less than 0.5 mm, it is considered continuous and can be identified as a key detection area. This method can accurately locate areas that require focused analysis.

[0089] For example, for key detection areas, grayscale analysis methods can calculate the difference between the average grayscale value of pixels within the area and the grayscale template of the standard coating. Assuming the average grayscale value of the standard template is 150 and the average grayscale value of the key detection area is 120, the difference is 30, which is greater than the preset threshold of 20, and is therefore judged as a low quality level.

[0090] Preferably, the stability of grayscale differences can be verified through a multi-frame image sequence to ensure reliable detection results.

[0091] In one possible implementation, the extension scheme could introduce multi-scale analysis, combining image sequences of different resolutions.

[0092] For example, for complex defect areas on car doors, in addition to the 1920×1080 resolution image, an additional 2560×1440 resolution image is acquired to analyze the texture changes of subtle defects. This multi-scale method can improve the accuracy of defect feature extraction and enhance the comprehensiveness of detection.

[0093] Image sequence data is obtained from a standardized coating surface image sequence. The pixel distribution features of the image sequence data are calculated using a grayscale histogram method, resulting in a pixel distribution feature set. For this pixel distribution feature set, a K-means clustering algorithm is used to group pixels based on their similarity. If the density of pixel distribution features in a group is greater than a preset threshold, it is marked as a key detection area, yielding key detection area location data. Based on the key detection area location data, an adaptive control algorithm is used to calculate the rotation speed parameter and detection angle parameter of the component. The rotation speed parameter S is calculated using the formula S=kD / T, where D represents the average distance of the boundary coordinate points of the key detection area location data, T represents the preset detection time, and k is a preset scaling factor. The detection angle parameter A is calculated using the formula A=arctan(H / W), where H represents the height of the key detection area, and W represents the width, resulting in a real-time adjustment parameter set. For the real-time adjustment parameter set, a motion control algorithm is used to generate motion control commands containing rotation speed and detection angle commands, yielding a motion control command set.

[0094] For example, for a sequence of images of a standardized coated surface, the grayscale histogram method can be used to analyze pixel distribution characteristics.

[0095] Specifically, a histogram is generated by calculating the grayscale value distribution of each frame in an image sequence, reflecting the brightness variation trend of the coating surface. Assuming an image sequence of automotive parts such as a hood is being processed, with a resolution of 1920×1080 and each frame containing 256 levels of grayscale values, the grayscale histogram shows that the grayscale values ​​of normal coating areas are concentrated in the range of 140 to 160, while defective areas such as scratches or dents have lower grayscale values, possibly in the range of 100 to 120. This method can quickly distinguish between normal areas and potential defective areas, providing a data foundation for subsequent analysis.

[0096] In one possible implementation, the K-means clustering algorithm can group pixels based on the similarity of their gray values, given the set of pixel distribution features generated from the gray-level histogram.

[0097] For example, the image data of the hood surface is divided into three groups: normal paint, minor defects, and severe defects. If the pixel distribution density of a certain group exceeds a preset threshold, such as 8 abnormal pixels per square centimeter, it is marked as a key detection area.

[0098] For example, the density of grayscale anomalies in the front area of ​​the hood is 10 per square centimeter, exceeding the threshold, and is therefore marked as a key detection area. This grouping method can effectively locate high-risk areas.

[0099] Specifically, based on the location data of key inspection areas, the adaptive control algorithm can calculate the inspection parameters of the components. Assuming the average distance between the boundary points of the key inspection area at the front of the hood is 50 mm, the preset inspection time is 2 seconds, and the scaling factor k is 0.5, the calculated rotation speed parameter S is 12.5 revolutions per minute. The inspection angle parameter is calculated using the ratio of the area's height to its width; assuming a height of 20 mm and a width of 40 mm, the angle A is approximately 26.6 degrees. These parameters ensure that the inspection equipment can accurately focus on the defect area.

[0100] For example, based on a real-time adjustment set of parameters, the motion control algorithm generates motion control commands. For the key inspection area at the front of the hood, the generated rotation speed command is 12.5 revolutions per minute, and the inspection angle command is 26.6 degrees. These control commands drive the inspection equipment to adjust the lens angle and rotation speed, focusing on the defect area. This method optimizes equipment operating efficiency and ensures that the inspection process covers critical areas.

[0101] In one possible implementation, the scalable approach improves accuracy by dynamically adjusting the detection parameters.

[0102] For example, for complex areas like the hood, tilt angle detection is introduced, and image sequences under different lighting conditions are used to verify the stability of defect features. This approach can capture defect details more comprehensively and improve the reliability of detection.

[0103] When considering the pixel distribution characteristics in the image, the shooting quality of the initial image data is further considered. Since shooting quality is affected by lighting, in some embodiments, the fully automated painting line for automotive parts also includes a light source device for illuminating the automotive parts and controlling the camera to capture the area to be painted, generating initial image data, including:

[0104] The system controls the camera to capture images of the area to be coated, acquiring image data to be adjusted; based on the acquired image data, the system determines the initial light source angle of the light source device; the system adjusts the initial light source angle until the angle between the irradiated light and the surface normal vector distribution is less than a preset angle value when the adjusted light source device irradiates the automotive parts; and the system controls the camera to capture images of the light source device irradiating the area to be coated, generating initial image data.

[0105] In one possible implementation, calculating the light reflection intensity I requires considering both the reflection coefficient R and the angle θ between the light source and the normal vector. Assuming the reflection coefficient at a point on the hood is 0.8 and the light source direction vector is (0.5, 0.5, 0.7), calculating cosθ yields an I value of approximately 0.65. This intensity value reflects the brightness performance of that point under a specific light source. By employing a light source angle optimization algorithm, the distribution of reflection intensity can be made more uniform by adjusting the light source position.

[0106] For example, the initial light source angle was 45 degrees, which was then adjusted to 30 degrees after optimization to reduce shadow areas and improve image clarity.

[0107] Specifically, when adjusting the camera pose, the angle α between the camera's line of sight and the surface normal vector needs to be calculated. Assuming the normal vector at a point is (0.6, 0.3, 0.7) and the camera's line of sight is (0.4, 0.4, 0.8), α is calculated to be approximately 20 degrees. If the preset threshold is 30 degrees, no camera position adjustment is needed; if α is 35 degrees, the camera angle needs to be adjusted to within 25 degrees to ensure the image captures surface details. Optimized camera pose parameters can improve image quality.

[0108] In some embodiments, after the painting is completed, a step of verifying the integrity of the painting can be determined. Specifically, after controlling the sprayer to move to the rotation coordinate value and performing the painting on the automotive parts, the process further includes:

[0109] Multiple coating images are captured when the automotive parts are rotated at various angles; multiple coating images are combined using image registration and feature matching algorithms to obtain a layer image set; the coating integrity of the automotive parts surface is determined based on the layer image set.

[0110] A sequence of coating images of automotive parts surfaces is acquired using a multi-angle inspection device. The pixel distribution of the image sequence is calculated using a grayscale histogram method to obtain a first pixel feature set. For this first pixel feature set, a boundary extraction algorithm is used to calculate the set of boundary coordinate points of the coating region, resulting in a boundary feature set. Based on the boundary feature set, an image registration algorithm is used to align the image sequence, and a feature matching algorithm is used to calculate corresponding points between the images, resulting in an aligned second image sequence. If there are detection blind spots in the second image sequence, interpolation methods are used to supplement the missing data, and a data fusion algorithm is used to integrate the data from the second image sequence to determine the complete quality assessment result of the surface coating.

[0111] For example, for image sequences of automotive component surface coatings acquired by multi-angle inspection equipment, the grayscale histogram method generates a first pixel feature set by analyzing the distribution of pixel grayscale values. Taking the surface of an automotive component such as a hood as an example, assuming the image sequence resolution is 1920×1080 and the grayscale value range is 0 to 255, the grayscale histogram shows that the pixel grayscale values ​​in normal coating areas are concentrated in the range of 160 to 180, while the grayscale values ​​in areas where scratches or coating peeling may exist are lower, distributed in the range of 80 to 100. By statistically analyzing the pixel distribution, a first pixel feature set is generated, which can effectively reflect the uniformity of the coating surface and potential abnormal areas, providing basic data for subsequent boundary extraction.

[0112] In one possible implementation, for the first pixel feature set, the Canny edge detection algorithm is used to generate a set of boundary coordinate points to form a boundary feature set.

[0113] For example, the boundary point set of a scratch area on the hood surface contains approximately 150 coordinate points, and the outline presents an irregular line shape. By analyzing the density of the point set distribution, it is determined whether the area is a complex surface. Assuming the boundary point set density of a certain area is 10 points per square centimeter, which is below the complex surface threshold of 20 points per square centimeter, it is classified as a simple area. If the density is higher than the threshold, it is marked as a complex surface, facilitating subsequent parameter adjustments. This method can accurately distinguish the characteristics of different areas.

[0114] For example, based on a set of boundary features, an image registration algorithm is used to align image sequences. Suppose a sequence of images of a hood surface comes from three cameras at different angles, and the difference in viewpoints causes image misalignment. The SIFT feature matching algorithm is used to extract key points from each image; for example, if 500 key points are detected, 400 of them are successfully paired across multiple images. By calculating the transformation matrix between the paired points, image alignment is achieved, generating a second image sequence. This alignment method ensures consistency of content across multiple image angles, improving the accuracy of subsequent analysis.

[0115] In one possible implementation, if the second image sequence has detection blind spots, such as data loss due to viewpoint obstruction at the edge of a hood, interpolation methods can be used to fill in the gaps. Assuming the blind spot area is 20 square centimeters, bilinear interpolation is used to estimate the grayscale value of the missing area based on the grayscale values ​​of surrounding pixels, generating the completed image data. This method effectively fills in blind spots and improves data integrity.

[0116] For example, for the second image sequence, a data fusion algorithm is used to integrate multi-angle image data to generate a complete quality assessment result for the surface coating. The fusion process combines the grayscale values ​​and boundary features of the three image sequences, calculating the overall grayscale distribution using a weighted average method. If a region shows low grayscale values ​​(e.g., in the 90-100 range) in multiple image angles, and the boundary point set displays an irregular shape, it is identified as a coating defect, such as scratches or peeling. This fusion method integrates multi-source data, improving the reliability of the assessment.

[0117] The second aspect of this application discloses an intelligent control system for a fully automated painting line for automotive parts, such as... Figure 2 As shown, the system includes:

[0118] The acquisition module 21 is used to acquire three-dimensional point cloud data of automotive parts;

[0119] The first determining module 22 is used to determine the high-complexity area on the surface of the automotive parts as the area to be coated based on the three-dimensional point cloud data.

[0120] Calculation module 23 is used to calculate the covariance matrix of the normal vector in the area to be coated using the principal component analysis algorithm, and extract the principal direction feature vector through the covariance matrix;

[0121] The second determining module 24 is used to determine the rotation coordinate value of the rotation axis based on the main direction feature vector;

[0122] The control module 25 is used to control the sprayer to move to the rotation coordinate value and perform painting on the automotive parts.

[0123] In one possible implementation, the first determining module 22 is specifically used for:

[0124] The k-nearest neighbor algorithm is used to determine the neighborhood point set of each of the three-dimensional point cloud data.

[0125] Using the neighborhood point set, at least one surface normal vector of the automotive component surface is calculated to obtain the surface normal vector distribution on the automotive component surface.

[0126] Using the surface normal vector distribution, the least squares method is used to fit the local area to be coated to obtain curvature distribution data corresponding to the surface normal vector distribution;

[0127] The curvature distribution data is mapped onto a two-dimensional matrix using a grid partitioning method;

[0128] The matrix element values ​​of the two-dimensional matrix are determined. If the matrix element values ​​exceed a preset curvature threshold, the local area to be coated corresponding to the two-dimensional matrix is ​​determined to be a high-complexity area, and at least one of the high-complexity areas is determined to be a coating area.

[0129] In one possible implementation, the first determining module 22 is specifically used for:

[0130] A region growing algorithm is used to connect at least one of the high-complexity regions, and the connected at least one of the high-complexity regions is the region to be painted.

[0131] The second determining module 24 is specifically used for:

[0132] Based on the main direction feature vector, the preliminary direction vector of the sprayer is determined;

[0133] Calculate the projected distance from at least one boundary point of the area to be coated to the preliminary direction vector;

[0134] Boundary points whose projection distance is less than a preset threshold are identified as axis neighborhood points, thus obtaining an axis neighborhood point set composed of at least one of the axis neighborhood points;

[0135] The least squares method is used to fit the set of neighborhood points of the axis into a spatial straight line, and the spatial coordinate points through which the spatial straight line passes are determined.

[0136] Calculate the local feature mean and variance of each spatial coordinate point, and use a weighted average algorithm to combine the local feature mean and variance to determine the rotation coordinate value of the rotation axis.

[0137] In one possible implementation, the control module 25 is specifically used for:

[0138] After controlling the motion device to move the sprayer to the rotation coordinate value, control the camera to capture the area to be coated and generate initial image data;

[0139] The initial image data is identified using a deep learning image recognition algorithm to determine the coating location of the area to be coated and the coating requirement value of the coating location.

[0140] Based on the position to be coated and the coating requirement value, an adaptive control algorithm is used to calculate the dynamic adjustment parameters of the motion device;

[0141] The motion control device controls the rotation of the automotive parts according to the dynamic adjustment parameters, and controls the sprayer to perform coating on the automotive parts.

[0142] In one possible implementation, the control module 25 is specifically used for:

[0143] Control the camera to capture images of the area to be coated, and obtain image data to be adjusted;

[0144] Based on the acquired image data to be adjusted, the initial light source angle of the light source device is determined;

[0145] Adjust the initial light source angle until, when the adjusted light source illuminates the automotive component, the angle between the irradiated light and the surface normal vector distribution is less than a preset angle value;

[0146] The camera is controlled to capture initial image data when the light source device illuminates the area to be coated by the illumination light.

[0147] In one possible implementation, the intelligent control system for a fully automated automotive parts painting line also includes:

[0148] Verification module 26 is used to acquire multiple coating images taken when the automotive parts are rotated at multiple angle values;

[0149] By combining the multiple coating images using image registration and feature matching algorithms, a layer image set is obtained.

[0150] The coating integrity of the automotive component surface is determined based on the layer image set.

[0151] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0152] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0153] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0154] Based on the above, Figure 1 The method illustrated in this application also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the method corresponding to any embodiment. The storage medium may further include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device and supports the operation of the information processing program and other software and / or programs. The network communication module is used to realize communication between the components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0155] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or it can be implemented by hardware. Compared with the prior art, this embodiment obtains three-dimensional point cloud data of automotive parts; based on the three-dimensional point cloud data, it determines the high-complexity areas on the surface of the automotive parts as the areas to be coated; it uses principal component analysis to calculate the covariance matrix of the normal vectors in the areas to be coated, and extracts the principal direction eigenvectors through the covariance matrix; based on the principal direction eigenvectors, it determines the rotation coordinates of the rotation axis; and it controls the sprayer to move to the rotation coordinates to perform the coating of the automotive parts. By adopting the technical solution of this application, when spraying automotive parts with different geometric characteristics, the rotation coordinates are determined in combination with the geometric characteristics of the automotive parts. When the sprayer is located at the rotation coordinates, a uniform layer can be sprayed onto the surface of the automotive parts as completely as possible, ensuring the coating effect.

[0156] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An intelligent control method for a fully automated painting line for automotive parts, characterized in that, The fully automated painting line for automotive parts includes a central controller and a sprayer. The method is applied to the central controller, which is connected to the sprayer. The method includes: Acquire 3D point cloud data of automotive parts; Based on the three-dimensional point cloud data, the high-complexity areas on the surface of the automotive parts are identified as areas to be coated. The principal component analysis algorithm is used to calculate the covariance matrix of the normal vector in the area to be coated, and the principal direction feature vector is extracted through the covariance matrix. Based on the principal direction feature vector, the rotation coordinate values ​​of the rotation axis are determined; The sprayer is controlled to move to the rotation coordinate value to perform painting on the automotive parts.

2. The method according to claim 1, characterized in that, Based on the three-dimensional point cloud data, determining the highly complex areas on the surface of automotive parts as areas to be painted includes: The k-nearest neighbor algorithm is used to determine the neighborhood point set of each of the three-dimensional point cloud data. Using the neighborhood point set, at least one surface normal vector of the automotive component surface is calculated to obtain the surface normal vector distribution on the automotive component surface. Using the surface normal vector distribution, the least squares method is used to fit the local area to be coated to obtain curvature distribution data corresponding to the surface normal vector distribution; The curvature distribution data is mapped onto a two-dimensional matrix using a grid partitioning method; The matrix element values ​​of the two-dimensional matrix are determined. If the matrix element values ​​exceed a preset curvature threshold, the local area to be coated corresponding to the two-dimensional matrix is ​​determined to be a high-complexity area, and at least one of the high-complexity areas is determined to be a coating area.

3. The method according to claim 2, characterized in that, The step of determining at least one of the highly complex regions as areas to be painted includes: A region growing algorithm is used to connect at least one of the high-complexity regions, and the connected at least one of the high-complexity regions is the region to be painted.

4. The method according to claim 2, characterized in that, Determining the rotation coordinates of the rotation axis based on the principal direction feature vector includes: Based on the main direction feature vector, the preliminary direction vector of the sprayer is determined; Calculate the projected distance from at least one boundary point of the area to be coated to the preliminary direction vector; Boundary points whose projection distance is less than a preset threshold are identified as axis neighborhood points, thus obtaining an axis neighborhood point set composed of at least one of the axis neighborhood points; The least squares method is used to fit the set of neighborhood points of the axis into a spatial straight line, and the spatial coordinate points through which the spatial straight line passes are determined. Calculate the local feature mean and variance of each spatial coordinate point, and use a weighted average algorithm to combine the local feature mean and variance to determine the rotation coordinate value of the rotation axis.

5. The method according to claim 2, characterized in that, The fully automated painting line for automotive parts also includes a camera and a motion device. The motion device is connected to the automotive parts and is used to control the rotation of the automotive parts. The central controller is connected to both the motion device and the camera. Controlling the sprayer to move to the rotation coordinate value to perform painting on the automotive parts includes: After controlling the motion device to move the sprayer to the rotation coordinate value, control the camera to capture the area to be coated and generate initial image data; The initial image data is identified using a deep learning image recognition algorithm to determine the coating location of the area to be coated and the coating requirement value of the coating location. Based on the position to be coated and the coating requirement value, an adaptive control algorithm is used to calculate the dynamic adjustment parameters of the motion device; The motion control device controls the rotation of the automotive parts according to the dynamic adjustment parameters, and controls the sprayer to perform coating on the automotive parts.

6. The method according to claim 5, characterized in that, The fully automated painting line for automotive parts also includes a light source device for illuminating the automotive parts. The control of the camera to capture images of the area to be painted and generate initial image data includes: Control the camera to capture images of the area to be coated, and obtain image data to be adjusted; Based on the acquired image data to be adjusted, the initial light source angle of the light source device is determined; Adjust the initial light source angle until, when the adjusted light source illuminates the automotive component, the angle between the irradiated light and the surface normal vector distribution is less than a preset angle value; The camera is controlled to capture initial image data when the light source device illuminates the area to be coated by the illumination light.

7. The method according to claim 1, characterized in that, After controlling the sprayer to move to the rotation coordinate value and performing the painting on the automotive parts, the method further includes: Multiple coating images were captured when the automotive component rotated at multiple angle values; By combining the multiple coating images using image registration and feature matching algorithms, a layer image set is obtained. The coating integrity of the automotive component surface is determined based on the layer image set.

8. An intelligent control system for a fully automated painting line for automotive parts, characterized in that, The system includes: The acquisition module is used to acquire 3D point cloud data of automotive parts; The first determining module is used to determine the high-complexity areas on the surface of automotive parts as areas to be coated based on the three-dimensional point cloud data. The calculation module is used to calculate the covariance matrix of the normal vector in the area to be coated using the principal component analysis algorithm, and extract the principal direction feature vector through the covariance matrix; The second determining module is used to determine the rotation coordinate value of the rotation axis based on the main direction feature vector; The control module is used to control the sprayer to move to the rotation coordinate value and perform painting on the automotive parts.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-7.

Images