Workpiece vacuum coating surface quality identification method and system

Abstract

The invention provides a workpiece vacuum coating surface quality identification method and system, and relates to the technical field of vacuum coating, and the method comprises the steps: inputting a front workpiece image set into a three-dimensional reconstruction channel, outputting a front workpiece three-dimensional contour, carrying out the fitting of a front workpiece point cloud set according to the front workpiece three-dimensional contour, and generating a front workpiece three-dimensional model; inputting the rear workpiece image set into a three-dimensional reconstruction channel to generate a rear workpiece three-dimensional model; performing grid thickness deviation comparison on the front workpiece three-dimensional model and the rear workpiece three-dimensional model to generate a grid thickness deviation set; and carrying out coating quality analysis to generate a coating thickness detection report. According to the method, the technical problems that the coating thickness detection and analysis accuracy and efficiency are poor and the coating quality evaluation accuracy is low due to the fact that the workpiece structure is complex and diversified and an existing method cannot accurately detect the coating thickness of each part of the workpiece can be solved, the coating thickness detection intelligence and fineness can be improved, and the coating quality evaluation accuracy is improved. And the effect of accurately evaluating the coating quality is achieved.

Description

Technical Field

[0001] This application relates to the field of vacuum coating technology, and in particular to a method and system for identifying the surface quality of vacuum-coated workpieces. Background Technology

[0002] In modern manufacturing, vacuum coating technology is widely used to improve the performance and aesthetics of workpieces. Accurate assessment of coating quality is crucial for ensuring product performance and meeting customer needs. However, with the increasing complexity and diversity of workpiece structures, existing coating quality inspection methods face a series of technical challenges.

[0003] Currently, due to the complex and diverse structures of workpieces, existing methods for inspecting the coating quality of workpieces cannot accurately detect the coating thickness of different parts of the workpiece. This results in poor accuracy and efficiency in the detection and analysis of the coating thickness, leading to technical problems such as low accuracy and reliability in coating quality assessment. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for identifying the surface quality of vacuum-coated workpieces, in order to solve the technical problem that existing workpiece coating quality detection methods cannot accurately detect the coating thickness of various parts of the workpiece due to the complex and diverse structure of the workpiece, resulting in poor accuracy and efficiency of workpiece coating thickness detection and analysis, and low accuracy and reliability of coating quality assessment.

[0005] In view of the above problems, this application provides a method and system for identifying the surface quality of vacuum-coated workpieces.

[0006] In a first aspect, this application provides a method for identifying the surface quality of a workpiece under vacuum coating. This method is implemented using a workpiece vacuum coating surface quality identification system, comprising: placing the workpiece to be coated in a predetermined detection space for multi-angle data acquisition to obtain a front workpiece image set and a front workpiece point cloud set; inputting the front workpiece image set into a 3D reconstruction channel to output a 3D contour of the front workpiece; fitting the front workpiece point cloud set based on the 3D contour to generate a 3D model of the front workpiece, wherein the 3D reconstruction channel includes a ResNet module, a feature fusion module, a BSPNet module, and an edge extraction module. The process involves: placing the coated workpiece in a predetermined detection space for multi-angle data acquisition to obtain a post-workpiece image set and a post-workpiece point cloud set, where the coated workpiece is the workpiece to be coated after vacuum coating; inputting the post-workpiece image set into the 3D reconstruction channel to output the post-workpiece 3D contour; fitting the post-workpiece point cloud set based on the post-workpiece 3D contour to generate a post-workpiece 3D model; comparing the mesh thickness deviation between the pre-workpiece 3D model and the post-workpiece 3D model in a predetermined 3D comparison space to generate a mesh thickness deviation set; and performing workpiece coating quality analysis based on the mesh thickness deviation set to generate a coating thickness detection report.

[0007] Secondly, this application also provides a vacuum coating surface quality identification system for workpieces, used to execute a vacuum coating surface quality identification method for workpieces as described in the first aspect, comprising: a workpiece data acquisition unit for placing the workpiece to be coated in a predetermined detection space for multi-angle data acquisition, obtaining a front workpiece image set and a front workpiece point cloud set; a front workpiece 3D model generation unit for inputting the front workpiece image set into a 3D reconstruction channel, outputting a front workpiece 3D contour, fitting the front workpiece point cloud set according to the front workpiece 3D contour, and generating a front workpiece 3D model, wherein the 3D reconstruction channel includes a ResNet module, a feature fusion module, a BSPNet module, and an edge extraction module; and a coated workpiece data acquisition unit for... The coated workpiece is placed in a predetermined detection space for multi-angle data acquisition, resulting in a post-workpiece image set and a post-workpiece point cloud set. The coated workpiece is the workpiece to be coated after vacuum coating. A post-workpiece 3D model generation unit is used to input the post-workpiece image set into the 3D reconstruction channel, output the post-workpiece 3D contour, and fit the post-workpiece point cloud set according to the post-workpiece 3D contour to generate a post-workpiece 3D model. A workpiece deviation comparison unit is used to compare the mesh thickness deviation between the pre-workpiece 3D model and the post-workpiece 3D model in a predetermined 3D comparison space to generate a mesh thickness deviation set. A coating quality analysis unit is used to perform workpiece coating quality analysis based on the mesh thickness deviation set and generate a coating thickness detection report.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: 1. By acquiring multi-angle images and point clouds of the workpiece to be coated, a set of images and a set of point clouds of the workpiece are obtained. Then, the set of images is reconstructed and its contour is extracted using a 3D reconstruction channel to obtain the 3D contour of the workpiece. The point cloud of the workpiece is fitted with the 3D contour as a constraint, and a 3D model of the workpiece is constructed based on the point cloud fitting results. On the other hand, after the workpiece is vacuum coated, a 3D model of the workpiece is constructed based on the set of images and point clouds of the coated workpiece. Then, in a predetermined 3D comparison space, the 3D models of the workpiece and the workpiece are meshed according to a predetermined mesh size. The thickness deviation is compared based on the meshing results to obtain a set of mesh thickness deviations. Furthermore, a 3D model of the coating is constructed based on the set of mesh thickness deviations, and the watershed algorithm is used to divide the 3D model of the coating into regions to obtain a coating thickness distribution map. This can improve the intelligence and precision of the workpiece coating thickness detection, thereby improving the accuracy and efficiency of coating thickness detection, and thus achieving the technical effect of accurately evaluating the coating quality.

[0009] 2. A 3D reconstruction channel is constructed based on ResNet, feature fusion, BSPNet and edge extraction modules. Since the combination of these modules utilizes deep learning and computer vision technology, it can effectively process and analyze complex and diverse workpiece data, thereby achieving accurate 3D reconstruction and improving the accuracy and efficiency of workpiece 3D contour extraction.

[0010] 3. By using the watershed algorithm to divide the three-dimensional model of the coating into regions, a coating thickness distribution map is obtained. This allows for a clearer and more intuitive analysis of coating defects and coating uniformity, providing support for accurate evaluation of the coating quality of the workpiece in the future.

[0011] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in 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, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating a method for identifying the surface quality of a workpiece under vacuum coating, as described in this application. Figure 2 This is a schematic diagram of the process for constructing a three-dimensional reconstruction channel in a vacuum coating surface quality identification method for a workpiece according to this application; Figure 3 This is a schematic diagram of the structure of a vacuum coating surface quality identification system for a workpiece according to this application.

[0014] Explanation of reference numerals in the attached figures: The system includes: a workpiece data acquisition unit 11, a front workpiece 3D model generation unit 12, a coated workpiece data acquisition unit 13, a rear workpiece 3D model generation unit 14, a workpiece deviation comparison unit 15, and a coating quality analysis unit 16. Detailed Implementation

[0015] This application provides a method and system for identifying the surface quality of vacuum-coated workpieces. It solves the technical problem that existing workpiece coating quality inspection methods cannot accurately detect the coating thickness at different locations on the workpiece due to the complex and diverse structures of the workpiece. This results in poor accuracy and efficiency in coating thickness detection and analysis, leading to low accuracy and reliability in coating quality assessment. The method improves the intelligence and precision of workpiece coating thickness detection, thereby enhancing its accuracy and efficiency, and ultimately achieving accurate assessment of coating quality.

[0016] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0017] Example 1, please refer to the appendix. Figure 1 This application provides a method for identifying the surface quality of a workpiece after vacuum coating, which is applied to a system for identifying the surface quality of a workpiece after vacuum coating. The method specifically includes the following steps: Step 1: Place the workpiece to be coated in the predetermined detection space and collect data from multiple angles to obtain the front workpiece image set and the front workpiece point cloud set.

[0018] Specifically, firstly, a predetermined detection space is constructed, which includes multiple image sensors and multiple point cloud scanners. Each image sensor and each point cloud scanner has a different acquisition angle, which is used to acquire image data and point cloud data of the workpiece from all directions. Then, the workpiece to be coated is placed in the predetermined detection space for multi-angle data acquisition, thereby obtaining the front workpiece image set and front workpiece point cloud set of the workpiece to be coated.

[0019] Step 2: Input the image set of the previous workpiece into the 3D reconstruction channel, output the 3D contour of the previous workpiece, fit the point cloud set of the previous workpiece according to the 3D contour of the previous workpiece, and generate the 3D model of the previous workpiece. The 3D reconstruction channel includes a ResNet module, a feature fusion module, a BSPNet module, and an edge extraction module.

[0020] Specifically, a 3D reconstruction channel is constructed, which includes a ResNet module, a feature fusion module, a BSPNet module, and an edge extraction module connected in sequence. The ResNet module is built based on the ResNet18 framework and is used to extract features from the image, converting the image into feature vectors. The feature fusion module includes a feature fusion unit and a perceptual expansion unit. The feature fusion unit is used to fuse multiple feature vectors output by the ResNet module, that is, to fuse multiple feature vectors into one feature vector. The perceptual expansion unit is used to expand the dimension of the feature vector output by the feature fusion unit. The BSPNet module is built based on the binary space segmentation algorithm and is connected to the output layer of the feature fusion module. It is used to process the feature matrix output by the perceptual expansion unit based on the binary space segmentation algorithm. By recursively segmenting the feature space, the complex shape is decomposed into multiple simple convex shapes, and then the multiple convex shapes are merged to generate the 3D shape of the workpiece, thus completing the 3D reconstruction of the workpiece. The edge extraction module is built based on the edge detection algorithm and a convolutional neural network and is connected to the output layer of the BSPNet module. It is used to extract the contour features of the 3D shape of the workpiece and output the 3D contour of the workpiece. By constructing a 3D reconstruction channel based on ResNet, feature fusion, BSPNet, and edge extraction modules, the combination of these modules utilizes deep learning and computer vision technologies, thus effectively processing and analyzing complex and diverse workpiece data. This enables accurate 3D reconstruction and improves the accuracy and efficiency of workpiece 3D contour extraction.

[0021] Next, using the three-dimensional contour of the previous workpiece as a constraint, a point cloud fitting algorithm is used to fit the point cloud set of the previous workpiece to obtain the point cloud fitting result. Then, based on the point cloud fitting result, simulation modeling is performed within a 3D simulation platform to generate a 3D model of the previous workpiece. This involves importing the point cloud fitting result into the 3D simulation platform and using the tools and functions provided by the platform to construct the 3D model, such as polygon modeling and NURBS modeling. By using a point cloud fitting algorithm for point cloud fitting, the accuracy and reliability of the obtained point cloud fitting results can be improved, thereby enhancing the accuracy and reliability of the 3D model construction of the workpiece.

[0022] Step 3: Place the coated workpiece in the predetermined detection space for multi-angle data acquisition to obtain the workpiece image set and the workpiece point cloud set. The coated workpiece is the workpiece after vacuum coating.

[0023] Specifically, a coating operation is performed on the workpiece to be coated based on a predetermined vacuum coating scheme to obtain a coated workpiece; the coated workpiece is then placed in a predetermined detection space for multi-angle data acquisition to obtain a set of images of the coated workpiece and a set of point clouds of the coated workpiece.

[0024] Step 4: Input the image set of the post-workpiece into the 3D reconstruction channel, output the 3D contour of the post-workpiece, fit the point cloud set of the post-workpiece according to the 3D contour of the post-workpiece, and generate the 3D model of the post-workpiece.

[0025] Specifically, the image set of the subsequent workpiece is then input into the 3D reconstruction channel, and the contour features of the image set of the subsequent workpiece are extracted using the 3D reconstruction channel to obtain the 3D contour of the subsequent workpiece. Then, the 3D contour of the subsequent workpiece is used as a constraint, and the point cloud set of the subsequent workpiece is fitted using a point cloud fitting algorithm to obtain the point cloud fitting result of the subsequent workpiece. Further, simulation modeling is performed based on the point cloud fitting result of the subsequent workpiece to generate a 3D model of the subsequent workpiece.

[0026] Step 5: Within the predetermined three-dimensional comparison space, compare the mesh thickness deviation between the three-dimensional model of the previous workpiece and the three-dimensional model of the subsequent workpiece to generate a mesh thickness deviation set.

[0027] Specifically, within a predetermined 3D comparison space, the 3D models of the preceding and following workpieces are first meshed according to a predetermined mesh size. Mesh division divides the 3D model into multiple small units, each representing a mesh. Next, a thickness deviation comparison is performed based on the mesh division results. At each mesh, the coating thickness deviation between the preceding and following workpieces is calculated, which is the difference between the thickness of the following workpiece and the thickness of the preceding workpiece. This coating thickness deviation represents the coating thickness at the mesh location, resulting in a set of mesh thickness deviations. By meshing the 3D model and comparing coating thickness deviations based on the mesh division results, the precision of coating thickness analysis can be improved, thereby enhancing the accuracy of coating thickness detection.

[0028] Step 6: Analyze the workpiece coating quality based on the mesh thickness deviation set and generate a coating thickness inspection report.

[0029] Specifically, simulation modeling is performed based on the mesh thickness deviation set to construct a three-dimensional model of the coating. The watershed algorithm is then used to divide the three-dimensional model of the coating into regions to obtain a coating thickness distribution map. This coating thickness distribution map is then used as a coating thickness detection report. The watershed algorithm is an image segmentation technique. Its basic idea is to treat objects in an image as terrain and determine the state of each pixel in the image by calculating the gray-level differences between pixels, thereby finding the segmentation boundary. Here, the existing watershed algorithm is modified by using mesh thickness deviation instead of image gray-level values ​​for region segmentation.

[0030] By using the watershed algorithm to divide the three-dimensional model of the coating into regions, a coating thickness distribution map is obtained. This allows for a clearer and more intuitive analysis of coating defects and coating uniformity, providing support for accurate evaluation of the coating quality of the workpiece in the future.

[0031] The aforementioned method for identifying the surface quality of vacuum-coated workpieces is applied to a vacuum-coated surface quality identification system for workpieces. It can solve the technical problem that existing workpiece coating quality detection methods cannot accurately detect the coating thickness of various parts of the workpiece due to the complex and diverse structure of the workpiece, resulting in poor accuracy and efficiency of workpiece coating thickness detection and analysis, and thus low accuracy and reliability of coating quality assessment. First, the workpiece to be coated is placed in a predetermined detection space for multi-angle data acquisition, resulting in a front workpiece image set and a front workpiece point cloud set. Then, the front workpiece image set is input into a 3D reconstruction channel, outputting a front workpiece 3D contour. The front workpiece point cloud set is fitted based on the front workpiece 3D contour to generate a front workpiece 3D model. The 3D reconstruction channel includes a ResNet module, a feature fusion module, a BSPNet module, and an edge extraction module. Next, the coated workpiece is placed in the predetermined detection space for multi-angle data acquisition, resulting in a rear workpiece image set and a rear workpiece point cloud set. The coated workpiece is the workpiece to be coated after vacuum coating. Then, the rear workpiece image set is input into the 3D reconstruction channel, outputting a rear workpiece 3D contour. The rear workpiece point cloud set is fitted based on the rear workpiece 3D contour to generate a rear workpiece 3D model. Further, within a predetermined 3D comparison space, the front workpiece 3D model and the rear workpiece 3D model are compared for mesh thickness deviation, generating a mesh thickness deviation set. Finally, based on the mesh thickness deviation set, workpiece coating quality analysis is performed, generating a coating thickness detection report. It can improve the intelligence and precision of workpiece coating thickness detection, thereby improving the accuracy and efficiency of coating thickness detection, and thus achieving the technical effect of accurately evaluating coating quality.

[0032] Furthermore, the predetermined testing space is constructed. Step one of this application includes: Read the reliable coverage area of ​​the image sensor and the 3D scanner; analyze the number and angle of deployment based on the reliable coverage area to determine the equipment deployment information; construct the predetermined detection space according to the equipment deployment information.

[0033] Specifically, first, the technical specifications of the image sensor and 3D scanner are obtained to understand their parameters such as field of view, resolution, and accuracy. Next, based on the equipment specifications, the reliable coverage range of each sensor and scanner in the predetermined detection space is calculated. For example, for image sensors, their field of view and resolution are considered to determine the workpiece portion that can be covered; for 3D scanners, their scanning range and accuracy are considered to determine the workpiece details that can be scanned. Further analysis of the workpiece area to be covered determines the number of sensors and scanners needed, ensuring that each sensor and scanner covers every corner of the workpiece while avoiding redundant coverage. Then, based on the analysis results, the placement position and angle of each sensor and scanner are determined, ensuring that the placement information meets the data acquisition requirements while considering operational convenience and safety. Finally, the predetermined detection space is constructed according to the equipment placement information, i.e., the sensors and scanners are installed, ensuring they are correctly installed at the predetermined positions and angles. Constructing the predetermined detection space provides support for image acquisition and point cloud acquisition of the workpiece.

[0034] Furthermore, a three-dimensional reconstruction channel is constructed, as shown in the attached diagram. Figure 2 As shown, step two of this application includes: A ResNet module is constructed based on the ResNet18 framework. This ResNet module converts images into feature vectors, each containing 256 feature elements. The feature fusion module includes a feature fusion unit and a perceptual expansion unit. The feature fusion unit includes a feature fusion function that calculates the mean of feature elements with mapping relationships. The perceptual expansion unit includes four linear layers to expand the dimension of the feature vectors. The output of the perceptual expansion unit is the feature matrix output by the fourth linear layer. A BSPNet module is constructed based on a binary space segmentation algorithm. This BSPNet module includes a signed distance function, an activation function, and a pooling layer. An edge extraction module is constructed based on an edge detection algorithm and a convolutional neural network. This edge extraction module extracts contour features from the output of the BSPNet module. The ResNet module, the feature fusion module, the BSPNet module, and the edge extraction module are connected sequentially to generate the 3D reconstruction channel.

[0035] Specifically, the ResNet18 framework is a relatively small network containing 18 layers, including 17 residual blocks. Each residual block consists of two 3x3 convolutional layers and a shortcut connection. The shortcut connection allows input features to be directly passed to the output, reducing information loss between network layers, ensuring information integrity, and improving model training efficiency. Based on the ResNet18 framework, a ResNet module is built. This module converts images into feature vectors containing pixel information for each view, with each feature vector containing 256 feature elements, each representing a key feature in the input image.

[0036] The feature fusion module includes a feature fusion unit and a perceptual expansion unit. The feature fusion unit includes a feature fusion function, which is used to calculate the mean of feature elements with mapping relationships, thereby aggregating multiple feature vectors. That is, the feature fusion unit fuses multiple feature vectors into a vector containing 256 feature elements. The perceptual expansion unit includes a first linear layer, a second linear layer, a third linear layer, and a fourth linear layer, which are used to expand the dimension of the feature vector. The first linear layer, the second linear layer, the third linear layer, and the fourth linear layer are connected in sequence to expand the dimension of the output result of the previous layer. The output result of the perceptual expansion unit is the feature matrix output by the fourth linear layer. Through the combination of the four linear layers, the perceptual expansion unit can expand the input feature vector to a higher-dimensional space, thereby providing more useful information for subsequent analysis.

[0037] A BSPNet module is constructed based on a binary space segmentation algorithm. The BSPNet module is a neural network framework for implementing convex shape decomposition. It processes the feature matrix output by the perceptual extension unit based on the binary space segmentation algorithm. By recursively segmenting the feature space, it decomposes complex shapes into multiple simple convex shapes. Then, it merges multiple convex shapes to generate the three-dimensional shape of the workpiece and completes the three-dimensional reconstruction of the workpiece. The BSPNet module includes a signed distance function, an activation function, and a pooling layer.

[0038] An edge extraction module is constructed based on an edge detection algorithm and a convolutional neural network. This module extracts contour features from the output of the BSPNet module through supervised training using sample data. Finally, the ResNet module, feature fusion module, BSPNet module, and edge extraction module are connected sequentially to generate the 3D reconstruction channel. By constructing a 3D reconstruction channel based on the ResNet, feature fusion, BSPNet, and edge extraction modules, and leveraging deep learning and computer vision technologies, this combination effectively processes and analyzes complex and diverse workpiece data, thereby achieving accurate 3D reconstruction and improving the accuracy and efficiency of workpiece 3D contour extraction.

[0039] Furthermore, in constructing the BSPNet module, this application also includes the following steps: The symbolic distance function is used to calculate the symbolic distance from a point in space to each plane. The expression for the symbolic distance function is: ;in, The sign distance between the characterization point X and the dividing plane, if If the value is greater than 0, the characterization point X is outside the dividing plane. A value less than or equal to 0 indicates that point X lies inside the dividing plane, where X is a point in space and n is the normal vector of the dividing plane. The distance from the dividing plane to the origin. Let n be the dot product of vector n and vector X. If it is a sign function, then If greater than 0, then If equals 1, If it equals 0, then If equals 0, If less than 0, then equal The activation function is used to retain inputs greater than 0 and filter inputs less than or equal to 0; the pooling layer is used to merge the convex shape primitives obtained after filtering by the activation function to generate the target three-dimensional shape.

[0040] Specifically, the symbolic distance function is used to calculate the symbolic distance from a point in space to each plane, and the expression of the symbolic distance function is: In the symbolic distance function, The sign distance between the characterization point X and the dividing plane, if If the value is greater than 0, the characterization point X is outside the dividing plane. A value less than or equal to 0 indicates that point X lies inside the dividing plane, where X is a point in space and n is the normal vector of the dividing plane. The distance from the dividing plane to the origin. Let n be the dot product of vector n and vector X. If it is a sign function, then If greater than 0, then If equals 1, If it equals 0, then If equals 0, If less than 0, then equal Next, based on the signed distance function, the feature matrix output by the fourth linear layer is spatially partitioned to obtain multiple segmentation planes. The activation function is used to retain inputs greater than 0 and filter inputs less than or equal to 0. Then, the multiple segmentation planes processed by the activation function are combined into a convex shape to obtain multiple convex shape primitives. Finally, the multiple convex shape primitives are merged through a pooling layer to output the target 3D shape.

[0041] Furthermore, in constructing the edge extraction module, this application also includes the following steps: Using the attribute information of the workpiece to be coated as a constraint, information retrieval of source workpieces is performed based on the Industrial Internet of Things to obtain a set of sample workpiece 3D shapes. An edge detection algorithm is used to extract contours from the sample workpiece 3D shapes to generate a set of sample 3D contours. The contour extraction effect of the sample 3D contours is evaluated, and sample 3D contours that meet the expected indicators are extracted to construct a standard 3D contour set. The standard workpiece 3D shape set corresponding to the standard 3D contour set is obtained. A sample training set is generated based on the standard workpiece 3D shape set and the standard 3D contour set, and the sample training set is divided into Q equal parts to obtain Q training sets. Using the standard workpiece 3D shape as input and the standard 3D contour as output, the convolutional neural network is supervised and cross-validated using the Q training sets to obtain Q convergent edge extraction units. An edge extraction module is built in parallel based on the Q convergent edge extraction units, where the output of the edge extraction module is the fitting result of the contours output by the Q convergent edge extraction units.

[0042] Specifically, the process involves acquiring the attribute information of the workpiece to be coated, including parameters such as its size, material, and shape. Then, using this attribute information as a constraint, a search for similar workpieces is performed based on the Industrial Internet of Things (IIoT). A matching algorithm retrieves historical data of coated workpieces with similar attributes, obtaining a set of sample workpiece 3D shapes. Next, an edge detection algorithm is used to extract the contours of this set, generating a sample 3D contour set. Common edge detection algorithms include Sobel, Canny, and Prewitt algorithms. The appropriate algorithm can be selected based on the specific circumstances. The edge detection algorithm calculates the Gaussian weighted gradient of each point in the input data and determines the magnitude and direction of the gradient to extract edges from the image. The extracted edges are then refined and processed to improve the clarity and continuity of the contours, using methods such as edge joining, edge smoothing, and edge refinement to enhance the extraction effect.

[0043] Next, the contour extraction effect is evaluated on multiple sample 3D contours in the sample 3D contour set, resulting in multiple effect evaluation results. For example, quality evaluation indicators are used to assess the contour extraction effect, including contour clarity, continuity, and completeness. Further, sample 3D contours that meet the expected indicators are extracted to construct a standard 3D contour set. The expected indicators are contour extraction quality standards, which can be set according to actual needs. The standard workpiece 3D shape set corresponding to the standard 3D contour set is then obtained. Then, a sample training set is generated based on the standard workpiece 3D shape set and the standard 3D contour set, and the sample training set is divided into Q equal parts, resulting in Q training sets.

[0044] Next, an initial edge extraction unit is constructed based on a convolutional neural network. This initial edge extraction unit is a convolutional neural network model that can be iteratively optimized in machine learning. The input data of the input layer of the initial edge extraction unit is the 3D shape of the workpiece, and the output data of the output layer is the 3D contour of the workpiece. Then, using the standard 3D shape of the workpiece as input and the standard 3D contour as output, the initial edge extraction unit is subjected to supervised training and cross-validation using Q training sets. For example, gradient descent algorithm and loss function can be used for supervised training. After meeting the preset number of training iterations, the initial edge extraction unit is validated by interactive training sets, resulting in Q convergent edge extraction units that meet the expected training constraints. Finally, an edge extraction module is built in parallel based on the Q convergent edge extraction units. The output of the edge extraction module is the fitting result of the output contour of the Q convergent edge extraction units.

[0045] By building an edge extraction module based on Q convergent edge extraction units, the contour extraction error of a single edge extraction unit can be reduced, thereby improving the accuracy and reliability of workpiece contour extraction.

[0046] Furthermore, generating a three-dimensional model of the pre-workpiece, step two of this application includes: Using the three-dimensional contour of the preceding workpiece as a constraint, a point cloud random fitting is performed on the point cloud set of the preceding workpiece to obtain a first point cloud fitting result, and a first point cloud fitting degree is calculated, wherein the first point cloud fitting degree is the ratio of the number of point clouds falling within the three-dimensional contour of the preceding workpiece to the total number of point clouds in the point cloud set of the preceding workpiece; again using the three-dimensional contour of the preceding workpiece as a constraint, a point cloud random fitting is performed on the point cloud set of the preceding workpiece to obtain a second point cloud fitting result and a second point cloud fitting degree; iterative fitting is performed until a predetermined number of fitting times is met, and the point cloud fitting result with the highest current point cloud fitting degree is output as the optimal point cloud fitting result, and a three-dimensional model of the preceding workpiece is constructed based on the optimal point cloud fitting result.

[0047] Specifically, using the three-dimensional contour of the preceding workpiece as a constraint, a point cloud random fitting is performed on the point cloud set of the preceding workpiece, that is, the point cloud data is randomly placed into the three-dimensional contour of the preceding workpiece to obtain a first point cloud fitting result, and a first point cloud fitting degree is calculated. The first point cloud fitting degree is the ratio of the number of points falling into the three-dimensional contour of the preceding workpiece to the total number of points in the point cloud set of the preceding workpiece. The larger the point cloud fitting degree, the better the point cloud fitting effect. Then, using the three-dimensional contour of the preceding workpiece as a constraint, a point cloud random fitting is performed on the point cloud set of the preceding workpiece again to obtain a second point cloud fitting result and a second point cloud fitting degree. The same method is used for iterative fitting until a predetermined number of fittings is met. The predetermined number of fittings can be set according to the point cloud fitting accuracy requirements. The higher the point cloud fitting accuracy requirements, the larger the predetermined number of fittings. The point cloud fitting result with the highest current point cloud fitting degree is output as the optimal point cloud fitting result. Then, simulation modeling is performed based on the optimal point cloud fitting result to generate a three-dimensional model of the preceding workpiece.

[0048] By using the workpiece's 3D contour as a constraint and performing point cloud fitting based on a point cloud fitting algorithm, the accuracy and reliability of the point cloud fitting results can be improved, thereby enhancing the precision and reliability of the workpiece's 3D model construction.

[0049] Furthermore, a set of mesh thickness deviations is generated. Step five of this application includes: Within a predetermined three-dimensional comparison space, the three-dimensional models of the preceding and following workpieces are divided into meshes based on a predetermined mesh size, generating a mesh thickness set for the preceding workpiece and a mesh thickness set for the following workpiece. Each workpiece mesh has a positioning mark, and the workpiece mesh thickness is the average thickness of the workpiece coating within the mesh. The mesh thickness set for the preceding and following workpieces is then compared by mapping deviation to obtain the mesh thickness deviation set.

[0050] Specifically, within a predetermined 3D comparison space, firstly, the 3D models of the preceding and following workpieces are meshed based on a predetermined mesh size, generating a mesh thickness set for the preceding and following workpieces. The predetermined mesh size can be set according to the expected comparison precision; the higher the expected comparison precision, the smaller the predetermined mesh size. Each workpiece mesh has a positioning mark, meaning each mesh unit is assigned a unique positioning mark to accurately locate and compare each mesh unit in subsequent comparisons. The workpiece mesh thickness is the average coating thickness of the workpiece within the mesh. Then, based on the positioning marks, the mesh thickness sets of the preceding and following workpieces are compared by mapping deviation, i.e., the mesh thickness of the following workpiece is subtracted from the mesh thickness of the preceding workpiece, resulting in a mesh thickness deviation set, where the mesh thickness deviation is the coating thickness at the mesh position.

[0051] Furthermore, a coating thickness inspection report is generated. Step six of this application includes: The mesh thickness deviation set is fused according to the positioning marks to generate a three-dimensional model of the coating. Taking the minimum thickness deviation of the three-dimensional model of the coating as the starting point, the three-dimensional model of the coating is simulated to be filled with water. The first thickness deviation that the water surface submerges is set as the first stage thickness deviation. The first deviation value between the first stage thickness deviation and the minimum thickness deviation is calculated. If the first deviation value meets the predetermined deviation threshold, the first watershed is set at the first thickness deviation. If the first deviation value does not meet the predetermined deviation threshold, the simulated water filling continues until the water surface submerges the maximum thickness deviation, then the water filling stops, resulting in multiple watersheds. The three-dimensional model of the coating is divided into regions according to the multiple watersheds to generate a coating thickness distribution map, which is used as a coating thickness detection report.

[0052] Specifically, the mesh thickness deviation set is fused according to the positioning marks, and a three-dimensional model of the coating is obtained based on the fusion result. Then, starting from the minimum thickness deviation of the coating three-dimensional model, water is simulated to fill the coating three-dimensional model, and the first thickness deviation submerged by the water surface is set as the first stage thickness deviation. Then, the first deviation value between the first stage thickness deviation and the minimum thickness deviation is calculated. The first deviation value is the thickness difference between the first stage thickness deviation and the minimum thickness deviation. Then, the first deviation value is judged according to a predetermined deviation threshold, which can be set by those skilled in the art based on experience. If the first deviation value is greater than or equal to the predetermined deviation threshold, a first watershed is set at the first thickness deviation. If the first deviation value is less than the predetermined deviation threshold, the simulated water filling continues until the water surface submerges the maximum thickness deviation, then the water filling stops, resulting in multiple watersheds. Finally, the coating three-dimensional model is divided into regions according to the multiple watersheds, and a coating thickness distribution map is generated based on the region division result. The coating thickness distribution map is used as a coating thickness detection report.

[0053] By using the watershed algorithm to divide the three-dimensional model of the coating into regions, a coating thickness distribution map is obtained. This allows for a clearer and more intuitive analysis of coating defects and coating uniformity, providing support for accurate evaluation of the coating quality of the workpiece in the future.

[0054] In summary, the method for identifying the surface quality of a workpiece under vacuum coating provided in this application has the following technical advantages: 1. By acquiring multi-angle images and point clouds of the workpiece to be coated, a set of images and a set of point clouds of the workpiece are obtained. Then, the set of images is reconstructed and its contour is extracted using a 3D reconstruction channel to obtain the 3D contour of the workpiece. The point cloud of the workpiece is fitted with the 3D contour as a constraint, and a 3D model of the workpiece is constructed based on the point cloud fitting results. On the other hand, after the workpiece is vacuum coated, a 3D model of the workpiece is constructed based on the set of images and point clouds of the coated workpiece. Then, in a predetermined 3D comparison space, the 3D models of the workpiece and the workpiece are meshed according to a predetermined mesh size. The thickness deviation is compared based on the meshing results to obtain a set of mesh thickness deviations. Furthermore, a 3D model of the coating is constructed based on the set of mesh thickness deviations, and the watershed algorithm is used to divide the 3D model of the coating into regions to obtain a coating thickness distribution map. This can improve the intelligence and precision of the workpiece coating thickness detection, thereby improving the accuracy and efficiency of coating thickness detection, and thus achieving the technical effect of accurately evaluating the coating quality.

[0055] 2. A 3D reconstruction channel is constructed based on ResNet, feature fusion, BSPNet and edge extraction modules. Since the combination of these modules utilizes deep learning and computer vision technology, it can effectively process and analyze complex and diverse workpiece data, thereby achieving accurate 3D reconstruction and improving the accuracy and efficiency of workpiece 3D contour extraction.

[0056] 3. By using the watershed algorithm to divide the three-dimensional model of the coating into regions, a coating thickness distribution map is obtained. This allows for a clearer and more intuitive analysis of coating defects and coating uniformity, providing support for accurate evaluation of the coating quality of the workpiece in the future.

[0057] Example 2: Based on the same inventive concept as the vacuum coating surface quality identification method for a workpiece in the foregoing examples, this application also provides a vacuum coating surface quality identification system for a workpiece. Please refer to the appendix. Figure 3 ,include: The workpiece data acquisition unit 11 is used to place the workpiece to be coated in a predetermined detection space for multi-angle data acquisition, and obtain the front workpiece image set and the front workpiece point cloud set.

[0058] The pre-workpiece 3D model generation unit 12 is used to input the pre-workpiece image set into the 3D reconstruction channel, output the pre-workpiece 3D contour, fit the pre-workpiece point cloud set according to the pre-workpiece 3D contour, and generate the pre-workpiece 3D model. The 3D reconstruction channel includes a ResNet module, a feature fusion module, a BSPNet module, and an edge extraction module.

[0059] The coated workpiece data acquisition unit 13 is used to place the coated workpiece in a predetermined detection space for multi-angle data acquisition, and obtain the post-workpiece image set and post-workpiece point cloud set. The coated workpiece is the workpiece after vacuum coating of the workpiece to be coated.

[0060] The post-workpiece 3D model generation unit 14 is used to input the post-workpiece image set into the 3D reconstruction channel, output the post-workpiece 3D contour, and fit the post-workpiece point cloud set according to the post-workpiece 3D contour to generate the post-workpiece 3D model.

[0061] The workpiece deviation comparison unit 15 is used to compare the mesh thickness deviation of the front workpiece 3D model and the rear workpiece 3D model in a predetermined 3D comparison space to generate a mesh thickness deviation set.

[0062] The coating quality analysis unit 16 is used to perform workpiece coating quality analysis based on the grid thickness deviation set and generate a coating thickness detection report.

[0063] Furthermore, the workpiece data acquisition unit 11 in the vacuum coating surface quality identification system for a workpiece is also used for: Read the reliable coverage area of ​​the image sensor and the 3D scanner; analyze the number and angle of deployment based on the reliable coverage area to determine the equipment deployment information; construct the predetermined detection space according to the equipment deployment information.

[0064] Furthermore, the pre-workpiece 3D model generation unit 12 in the vacuum coating surface quality identification system for a workpiece is also used for: A ResNet module is constructed based on the ResNet18 framework. This ResNet module converts images into feature vectors, each containing 256 feature elements. The feature fusion module includes a feature fusion unit and a perceptual expansion unit. The feature fusion unit includes a feature fusion function that calculates the mean of feature elements with mapping relationships. The perceptual expansion unit includes four linear layers to expand the dimension of the feature vectors. The output of the perceptual expansion unit is the feature matrix output by the fourth linear layer. A BSPNet module is constructed based on a binary space segmentation algorithm. This BSPNet module includes a signed distance function, an activation function, and a pooling layer. An edge extraction module is constructed based on an edge detection algorithm and a convolutional neural network. This edge extraction module extracts contour features from the output of the BSPNet module. The ResNet module, the feature fusion module, the BSPNet module, and the edge extraction module are connected sequentially to generate the 3D reconstruction channel.

[0065] Furthermore, the pre-workpiece 3D model generation unit 12 in the vacuum coating surface quality identification system for a workpiece is also used for: The symbolic distance function is used to calculate the symbolic distance from a point in space to each plane. The expression for the symbolic distance function is: ;in, The sign distance between the characterization point X and the dividing plane, if If the value is greater than 0, the characterization point X is outside the dividing plane. A value less than or equal to 0 indicates that point X lies inside the dividing plane, where X is a point in space and n is the normal vector of the dividing plane. The distance from the dividing plane to the origin. Let n be the dot product of vector n and vector X. If it is a sign function, then If greater than 0, then If equals 1, If it equals 0, then If equals 0, If less than 0, then equal The activation function is used to retain inputs greater than 0 and filter inputs less than or equal to 0; the pooling layer is used to merge the convex shape primitives obtained after filtering by the activation function to generate the target three-dimensional shape.

[0066] Furthermore, the pre-workpiece 3D model generation unit 12 in the vacuum coating surface quality identification system for a workpiece is also used for: Using the attribute information of the workpiece to be coated as a constraint, information retrieval of source workpieces is performed based on the Industrial Internet of Things to obtain a set of sample workpiece 3D shapes. An edge detection algorithm is used to extract contours from the sample workpiece 3D shapes to generate a set of sample 3D contours. The contour extraction effect of the sample 3D contours is evaluated, and sample 3D contours that meet the expected indicators are extracted to construct a standard 3D contour set. The standard workpiece 3D shape set corresponding to the standard 3D contour set is obtained. A sample training set is generated based on the standard workpiece 3D shape set and the standard 3D contour set, and the sample training set is divided into Q equal parts to obtain Q training sets. Using the standard workpiece 3D shape as input and the standard 3D contour as output, the convolutional neural network is supervised and cross-validated using the Q training sets to obtain Q convergent edge extraction units. An edge extraction module is built in parallel based on the Q convergent edge extraction units, where the output of the edge extraction module is the fitting result of the contours output by the Q convergent edge extraction units.

[0067] Furthermore, the pre-workpiece 3D model generation unit 12 in the vacuum coating surface quality identification system for a workpiece is also used for: Using the three-dimensional contour of the preceding workpiece as a constraint, a point cloud random fitting is performed on the point cloud set of the preceding workpiece to obtain a first point cloud fitting result, and a first point cloud fitting degree is calculated, wherein the first point cloud fitting degree is the ratio of the number of point clouds falling within the three-dimensional contour of the preceding workpiece to the total number of point clouds in the point cloud set of the preceding workpiece; again using the three-dimensional contour of the preceding workpiece as a constraint, a point cloud random fitting is performed on the point cloud set of the preceding workpiece to obtain a second point cloud fitting result and a second point cloud fitting degree; iterative fitting is performed until a predetermined number of fitting times is met, and the point cloud fitting result with the highest current point cloud fitting degree is output as the optimal point cloud fitting result, and a three-dimensional model of the preceding workpiece is constructed based on the optimal point cloud fitting result.

[0068] Furthermore, the workpiece deviation comparison unit 15 in the vacuum coating surface quality identification system for a workpiece is also used for: Within a predetermined three-dimensional comparison space, the three-dimensional models of the preceding and following workpieces are divided into meshes based on a predetermined mesh size, generating a mesh thickness set for the preceding workpiece and a mesh thickness set for the following workpiece. Each workpiece mesh has a positioning mark, and the workpiece mesh thickness is the average thickness of the workpiece coating within the mesh. The mesh thickness set for the preceding and following workpieces is then compared by mapping deviation to obtain the mesh thickness deviation set.

[0069] Furthermore, the coating quality analysis unit 16 in the vacuum coating surface quality identification system for a workpiece is also used for: The mesh thickness deviation set is fused according to the positioning marks to generate a three-dimensional model of the coating. Taking the minimum thickness deviation of the three-dimensional model of the coating as the starting point, the three-dimensional model of the coating is simulated to be filled with water. The first thickness deviation that the water surface submerges is set as the first stage thickness deviation. The first deviation value between the first stage thickness deviation and the minimum thickness deviation is calculated. If the first deviation value meets the predetermined deviation threshold, the first watershed is set at the first thickness deviation. If the first deviation value does not meet the predetermined deviation threshold, the simulated water filling continues until the water surface submerges the maximum thickness deviation, then the water filling stops, resulting in multiple watersheds. The three-dimensional model of the coating is divided into regions according to the multiple watersheds to generate a coating thickness distribution map, which is used as a coating thickness detection report.

[0070] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. The method and specific examples for identifying the surface quality of a workpiece under vacuum coating in Embodiment 1 described above are also applicable to the system for identifying the surface quality of a workpiece under vacuum coating in this embodiment. Through the foregoing detailed description of the method for identifying the surface quality of a workpiece under vacuum coating, those skilled in the art can clearly understand the system for identifying the surface quality of a workpiece under vacuum coating in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here. As for the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and relevant parts can be referred to in the method section.

[0071] The above description of the disclosed embodiments enables those skilled in the art to make or use 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 shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0072] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for identifying the surface quality of a vacuum-coated workpiece, characterized in that include: The workpiece to be coated is placed in a predetermined detection space for multi-angle data acquisition, resulting in a set of images of the workpiece and a set of point clouds of the workpiece. The image set of the previous workpiece is input into the 3D reconstruction channel, and the 3D contour of the previous workpiece is output. The point cloud set of the previous workpiece is fitted according to the 3D contour of the previous workpiece to generate a 3D model of the previous workpiece. The 3D reconstruction channel includes a ResNet module, a feature fusion module, a BSPNet module, and an edge extraction module. The coated workpiece is placed in a predetermined detection space for multi-angle data acquisition, resulting in a set of images and a set of point clouds of the workpiece. The coated workpiece is the workpiece after vacuum coating. The image set of the post-workpiece is input into the three-dimensional reconstruction channel, and the three-dimensional contour of the post-workpiece is output. The point cloud set of the post-workpiece is fitted according to the three-dimensional contour of the post-workpiece to generate a three-dimensional model of the post-workpiece. Within a predetermined three-dimensional comparison space, the mesh thickness deviation of the previous workpiece three-dimensional model and the subsequent workpiece three-dimensional model is compared to generate a mesh thickness deviation set; The workpiece coating quality is analyzed based on the mesh thickness deviation set, and a coating thickness inspection report is generated.

2. The method for identifying the surface quality of a workpiece under vacuum coating according to claim 1, characterized in that, Constructing a predetermined detection space includes: Read the reliable coverage area of ​​image sensors and 3D scanners; Based on the trusted coverage area, the number and angle of deployment are analyzed to determine the equipment deployment information; The predetermined testing space is constructed based on the equipment deployment information.

3. The method for identifying the surface quality of a workpiece under vacuum coating according to claim 1, characterized in that, Constructing a 3D reconstruction channel includes: A ResNet module is built based on the ResNet18 framework. The ResNet module is used to convert images into feature vectors, where each feature vector contains 256 feature elements. The feature fusion module includes a feature fusion unit and a perceptual expansion unit. The feature fusion unit includes a feature fusion function, which is used to calculate the mean of feature elements that have a mapping relationship. The perceptual expansion unit includes four linear layers, which are used to expand the dimension of the feature vector. The output of the perceptual expansion unit is the feature matrix output by the fourth linear layer. A BSPNet module is constructed based on the binary space partitioning algorithm, wherein the BSPNet module includes a symbolic distance function, an activation function, and a pooling layer; An edge extraction module is constructed based on an edge detection algorithm and a convolutional neural network. The edge extraction module is used to extract contour features from the output of the BSPNet module. The ResNet module, the feature fusion module, the BSPNet module, and the edge extraction module are connected in sequence to generate the 3D reconstruction channel.

4. The method for identifying the surface quality of a workpiece under vacuum coating according to claim 3, characterized in that, Building the BSPNet module includes: The symbolic distance function is used to calculate the symbolic distance from a point in space to each plane. The expression for the symbolic distance function is: ; Wherein, the sign distance between the characterization point X and the dividing plane, if If the value is greater than 0, the characterization point X is outside the dividing plane. A value less than or equal to 0 indicates that point X lies inside the dividing plane, where X is a point in space and n is the normal vector of the dividing plane. The distance from the dividing plane to the origin. Let n be the dot product of vector n and vector X. If it is a sign function, then If it is greater than 0, then If equals 1, If it equals 0, then If equals 0, If less than 0, then equal ; The activation function is used to retain inputs greater than 0 and filter inputs less than or equal to 0; The pooling layer is used to merge the convex primitives obtained after filtering by the activation function to generate the target three-dimensional shape.

5. The method for identifying the surface quality of a workpiece under vacuum coating according to claim 4, characterized in that, The edge extraction module is constructed, including: Using the attribute information of the workpiece to be coated as a constraint, the information retrieval of the same workpiece is carried out based on the Industrial Internet of Things to obtain the three-dimensional shape set of the sample workpiece; Based on the edge detection algorithm, the contour of the three-dimensional shape set of the sample workpiece is extracted to generate a sample three-dimensional contour set; The contour extraction effect of the sample 3D contour set is evaluated, and the sample 3D contours that meet the expected indicators are extracted to construct a standard 3D contour set. The standard workpiece 3D shape set corresponding to the standard 3D contour set is then obtained. A sample training set is generated based on the standard workpiece three-dimensional shape set and the standard three-dimensional contour set, and the sample training set is divided into Q equal parts to obtain Q training sets; Using the standard workpiece's 3D shape as input and the standard 3D contour as output, a convolutional neural network is trained and cross-validated using Q training sets to obtain Q convergent edge extraction units. An edge extraction module is then built in parallel based on these Q convergent edge extraction units, where the output of the edge extraction module is the fitting result of the contour output by the Q convergent edge extraction units.

6. The method for identifying the surface quality of a workpiece under vacuum coating according to claim 1, characterized in that, Generate a 3D model of the pre-workpiece, including: Using the three-dimensional contour of the preceding workpiece as a constraint, the point cloud set of the preceding workpiece is randomly fitted to obtain the first point cloud fitting result, and the first point cloud fitting degree is calculated. The first point cloud fitting degree is the ratio of the number of point clouds falling within the three-dimensional contour of the preceding workpiece to the total number of point clouds in the point cloud set of the preceding workpiece. Again, using the three-dimensional contour of the previous workpiece as a constraint, the point cloud set of the previous workpiece is randomly fitted to obtain the second point cloud fitting result and the second point cloud fitting degree. Perform iterative fitting until the predetermined number of fittings is met, and output the point cloud fitting result with the highest current point cloud fitting degree as the optimal point cloud fitting result. Based on the optimal point cloud fitting result, construct the three-dimensional model of the workpiece.

7. The method for identifying the surface quality of a workpiece under vacuum coating according to claim 1, characterized in that, Generate a set of mesh thickness deviations, including: Within a predetermined three-dimensional comparison space, the three-dimensional models of the preceding and following workpieces are divided into meshes based on a predetermined mesh size, generating a mesh thickness set for the preceding workpiece and a mesh thickness set for the following workpiece. Each workpiece mesh has a positioning mark, and the workpiece mesh thickness is the average thickness of the workpiece coating within the mesh. The mesh thickness set of the preceding workpiece and the mesh thickness set of the following workpiece are compared by mapping deviation to obtain the mesh thickness deviation set.

8. The method for identifying the surface quality of a vacuum-coated workpiece according to claim 7, characterized in that, Generate a coating thickness inspection report, including: The mesh thickness deviation set is fused according to the positioning marks to generate a three-dimensional model of the coating; Starting from the minimum thickness deviation of the coating three-dimensional model, the coating three-dimensional model is simulated to be filled with water, and the first thickness deviation after the water surface is submerged is set as the first stage thickness deviation. Calculate the first deviation value between the first stage thickness deviation and the minimum thickness deviation. If the first deviation value meets the predetermined deviation threshold, then set the first watershed at the first thickness deviation. If the first deviation value does not meet the predetermined deviation threshold, the simulated watering continues until the water surface exceeds the maximum thickness deviation, at which point the watering stops, resulting in multiple watersheds. The three-dimensional model of the coating is divided into regions based on multiple watersheds to generate a coating thickness distribution map, which is then used as a coating thickness detection report.

9. A vacuum coating surface quality identification system for workpieces, characterized in that, The steps for implementing the vacuum coating surface quality identification method for a workpiece according to any one of claims 1 to 8 include: The workpiece data acquisition unit is used to place the workpiece to be coated in a predetermined detection space to acquire data from multiple angles, thereby obtaining a set of images of the workpiece and a set of point clouds of the workpiece. The pre-workpiece 3D model generation unit is used to input the pre-workpiece image set into the 3D reconstruction channel, output the pre-workpiece 3D contour, fit the pre-workpiece point cloud set according to the pre-workpiece 3D contour, and generate the pre-workpiece 3D model. The 3D reconstruction channel includes a ResNet module, a feature fusion module, a BSPNet module, and an edge extraction module. The coated workpiece data acquisition unit is used to place the coated workpiece in a predetermined detection space for multi-angle data acquisition, and obtain the workpiece image set and the workpiece point cloud set. The coated workpiece is the workpiece after vacuum coating. The post-workpiece 3D model generation unit is used to input the post-workpiece image set into the 3D reconstruction channel, output the post-workpiece 3D contour, and fit the post-workpiece point cloud set according to the post-workpiece 3D contour to generate the post-workpiece 3D model. The workpiece deviation comparison unit is used to compare the mesh thickness deviation of the previous workpiece 3D model and the subsequent workpiece 3D model in a predetermined 3D comparison space, and generate a mesh thickness deviation set. The coating quality analysis unit is used to perform coating quality analysis on the workpiece based on the grid thickness deviation set and generate a coating thickness detection report.

Images