Automatic visual inspection system applied to charging pile production

Abstract

The invention relates to the technical field of industrial image recognition, in particular to an automatic visual inspection system applied to charging pile production. The method comprises the following steps: the image enhancement module is used for acquiring an omnibearing production monitoring image; carrying out image detail convolution enhancement on the omnibearing production monitoring image, and carrying out topological structure fitting to generate a component topological structure diagram; the twinborn model module is used for performing production process action recognition on the assembly topological structure diagram based on the omnibearing production monitoring image, performing three-dimensional simulation modeling and constructing a dynamic production twinborn model; and the deviation identification module is used for identifying the position deviation of the dynamic production twin model assembly according to a preset production planning log, then carrying out reference positioning deviation calculation, and generating assembly position deviation parameters. According to the invention, efficient and accurate charging pile production diagnosis detection is realized, so that the production process of the charging pile is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial image recognition, and in particular to an automated visual inspection system applied to the production of charging piles. Background Art

[0002] With the rapid development of the new energy vehicle industry, charging piles, as important power supply equipment, have become key constraints in production quality and reliability. As a key quality control method, the detection and analysis of the charging pile production process can effectively detect the appearance defects, assembly deviations and functional problems of charging piles, and improve the automation level of the production line and the quality standards of products. The traditional charging pile production detection system relies on manual inspections, which often have problems of low efficiency and poor diagnostic accuracy. It is difficult to meet the needs of modern charging pile manufacturing. In order to solve these challenges, an intelligent automated visual inspection system is needed. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes an automated visual inspection system applied to the production of charging piles to solve at least one of the above technical problems.

[0004] To achieve the above-mentioned object, the present invention provides an automated visual inspection system for charging pile production, wherein the automated visual inspection system for charging pile production comprises an image enhancement module, a twin model module, a deviation recognition module, a deviation compensation module, a defect detection module and a process optimization module; The image enhancement module is used to: obtain a full-scale production monitoring image; perform image detail convolution enhancement on the full-scale production monitoring image, and perform topological structure fitting to generate a component topological structure diagram; The twin model module is used to: identify production process actions on the component topology diagram based on the omnidirectional production monitoring image, and perform three-dimensional simulation modeling to build a dynamic production twin model; The deviation identification module is used to: identify the component position deviation of the dynamic production twin model according to the preset production planning log, and then calculate the reference positioning deviation to generate the component position deviation parameter; The deviation compensation module is used to: visually mark the dynamic production twin model based on the component position deviation parameter, and perform component positioning compensation calculation, thereby generating the component position deviation compensation parameter; The defect detection module is used to: perform surface texture defect detection on the detailed reconstructed omni-directional monitoring image to generate surface texture defect data; and perform geometric morphology analysis of the defect area to obtain defect type characteristics; The process optimization module is used to make charging pile abnormality correction decisions based on defect type characteristics and component position deviation compensation parameters, dynamically optimize the charging pile production process, and build an automated detection optimization model.

[0005] Preferably, the image enhancement module is used to: obtain an omnidirectional production monitoring image; perform image detail convolution enhancement on the omnidirectional production monitoring image, and perform topological structure fitting to generate a component topological structure diagram, specifically including: Real-time monitoring of the charging pile production process based on industrial cameras to obtain all-round production monitoring images; Perform image detail convolution enhancement on the omnidirectional production monitoring image to obtain a detailed reconstructed omnidirectional monitoring image; Perform multi-component visual recognition on the detailed reconstruction of the all-round monitoring image and mark the charging pile component nodes; Perform inter-component connection analysis on charging pile component nodes and extract component connection relationships; The topological structure of the detail-enhanced omnidirectional monitoring image is fitted according to the component connection relationship to generate a component topological structure diagram.

[0006] The present invention improves the quality of all-round production monitoring images through image enhancement, making the images clearer and easier to analyze. Image detail convolution enhancement and topological structure fitting help to extract and emphasize important features in the image, generate component topological structure diagrams, and provide a better data basis for subsequent analysis and identification. The twin model is based on all-round production monitoring images and can identify production process actions on component topological structure diagrams, achieving a seamless transition from monitoring images to action recognition. Through three-dimensional simulation modeling, a dynamic production twin model is constructed, providing an effective tool for simulation and optimization of production processes. The deviation identification module can accurately identify the deviation of component positions in the dynamic production twin model based on the production planning log, which is helpful to discover and locate problems in the production process. The component position deviation is generated through benchmark positioning deviation calculation. The difference parameters provide a basis for subsequent deviation compensation. The deviation compensation module visualizes the dynamic production twin model and performs compensation calculations based on the component position deviation parameters, thereby realizing the correction of component position deviation. The generated component position deviation compensation parameters improve the precision and accuracy of the production process and ensure production quality. The defect detection module performs surface texture defect detection by reconstructing the full range of monitoring images of details, which helps to discover defects in production. The geometric morphology analysis of the defect area provides defect type characteristics and provides guidance for subsequent processing and correction. The process optimization module is based on the defect type characteristics and component position deviation compensation parameters. It can make abnormal correction decisions for charging piles and dynamically optimize the production process. By building an automated detection optimization model, it can improve production efficiency, reduce costs, and ultimately improve product quality.

[0007] Preferably, the specific steps of performing image detail convolution enhancement on the omnidirectional production monitoring image to obtain a detail-reconstructed omnidirectional monitoring image are: Filter and reduce noise on all-round production monitoring images to generate noise-reduced and optimized monitoring images; Perform multi-scale decomposition on the noise reduction optimized monitoring image to obtain multiple frequency feature images; Perform convolution feature extraction on multiple frequency feature images to obtain the residual features of each image; Perform multiple rounds of iterative detail optimization on multiple frequency feature images according to the residual features of each image to generate multiple detail optimized feature images; Perform local pooling compression analysis on multiple detail optimized feature images to extract local implicit vectors of the images; Decoding and reconstructing the image according to the local latent vector of the image to obtain multiple super-pixel images; Multiple super-pixel images are fused and reconstructed to obtain a detailed reconstructed omnidirectional monitoring image.

[0008] The present invention reduces the noise in the image and improves the image quality and clarity through filtering and denoising processing. The generated denoised optimized monitoring image provides a cleaner and easier to process basis for subsequent processing. Multi-scale decomposition decomposes the image into frequency feature images of different scales, which is helpful to analyze and process the image features at different scales. The obtained multiple frequency feature images provide a richer data source for subsequent feature extraction and optimization. Through convolution feature extraction, residual features are extracted from each image to capture subtle changes and information in the image. These residual features provide a basis for subsequent detail optimization. Multiple rounds of iterative detail optimization can further optimize the frequency feature image according to the residual features of each image, improve the details and quality of the image, and the generated multiple detail optimized feature images can better retain the subtle features in the image. Image decoding and reconstruction reconstructs the image according to the local implicit vector to obtain multiple super-pixel images, which help to retain and reconstruct the detail information of the image. The super-pixel images obtained by reconstruction lay the foundation for the final detail reconstruction of the all-round monitoring image.

[0009] Preferably, the twin model module is used to: identify production process actions on the component topology diagram based on the omnidirectional production monitoring image, and perform three-dimensional simulation modeling to build a dynamic production twin model, specifically including: Performing time-series multi-frame sampling on the detail-reconstructed omni-directional monitoring image to obtain a time-series frame sequence of the monitoring image; Perform frame-by-frame dynamic optical flow tracking on the monitoring image time sequence frame sequence to generate image frame dynamic optical flow tracking trajectory; Perform production process action recognition on the dynamic optical flow tracking trajectory of the image frame and extract production process action data; Conduct time-series process logic analysis on production process action data to obtain production process logic representation; The production process logic representation is used to perform three-dimensional simulation modeling of the component topology diagram and build a dynamic production twin model.

[0010] The present invention obtains a time-series frame sequence of a monitoring image through time-series multi-frame sampling, and provides information about the continuous changes of the monitored object in time. The time-series frame sequence provides a continuous data source for subsequent dynamic analysis. Frame-by-frame dynamic optical flow tracking can capture the motion information of objects in the monitoring image and generate a dynamic optical flow tracking trajectory for the image frame. The optical flow tracking trajectory provides the motion path of the object in the image, and provides a basis for subsequent action recognition and process analysis. By analyzing the optical flow tracking trajectory, the actions in the production process are identified, and related production process action data are extracted. The production process action data contains information about key actions in the production process, which helps to understand the production process and monitor the execution of the process. The time-series process logic analysis can perform time-series analysis on the production process action data and derive a logical representation of the production process. The production process logic representation provides an overall grasp of the production process, which helps to discover the rules and optimization space in the process. The production process logic representation is used to perform three-dimensional simulation modeling on the topological structure diagram of the component, and a dynamic production twin model is constructed. The dynamic production twin model helps to simulate and optimize the production process, and realize accurate modeling and simulation of the production environment and process.

[0011] Preferably, the specific steps of using the production process logic representation to perform three-dimensional simulation modeling on the component topology structure diagram and constructing a dynamic production twin model are: Extracting multiple key frame images based on the monitoring image time series frame sequence; Perform component feature point recognition on multiple key frame images and mark the feature points of multiple frame images; Performing multi-directional image posture analysis on a plurality of key frame images to generate multi-directional image posture distortion data; According to the multi-directional image posture distortion data, feature points of multiple frame images are matched in spatial position to obtain accurate spatial position data of feature points; Based on the precise spatial position data of the feature points, the component topology structure diagram is modeled with feature three-dimensional point cloud to construct a component topology three-dimensional point cloud model; Perform full-process process simulation based on the production process logic representation to obtain the full-process process simulation trajectory; According to the full-process process simulation trajectory, the component topology three-dimensional point cloud model is dynamically rendered to build a dynamic production twin model.

[0012] The present invention helps to capture the key moments in the production process by extracting key frame images, making subsequent analysis more efficient and accurate. These key frame images represent important stages and changes in the production process, provide targeted data sets, and help to accurately capture key information in the production process. By identifying and marking the feature points of components, associations are established between different frames, and the position and posture changes of components are tracked. The marking of these feature points provides a reliable basis for subsequent posture analysis and spatial position matching, allowing the system to more accurately understand the structure and movement of components. Multi-directional image posture analysis can integrate information from different angles and accurately capture the posture changes of components in three-dimensional space. This comprehensive analysis provides a more comprehensive perspective and provides more accurate posture distortion data for subsequent spatial position matching, which helps to build a more realistic three-dimensional model. Feature point spatial position matching is to determine the accurate spatial position of feature points between different frames, thereby providing feature points in three-dimensional space. The precise position data in the space is necessary for constructing a three-dimensional point cloud model. Through precise matching, the system can accurately restore the spatial layout of the component structure. The three-dimensional point cloud modeling based on the precise position data of the feature point space can restore the three-dimensional structure of the component to a digital model. This model provides an accurate description of the component topology, which provides a basis for subsequent process simulation and rendering, enabling the system to better simulate and analyze the production process. The full-process process simulation helps users understand the process logic and optimize the production process by simulating the dynamic changes of the entire production process. This simulation technology can provide decision makers with a comprehensive perspective and help them formulate more effective production strategies and process optimization plans. Dynamic process rendering applies simulation trajectories to the three-dimensional point cloud model to achieve dynamic visualization of the production process. This dynamic display method makes process changes more intuitive and visible, provides an intuitive reference for monitoring and optimization of the production process, and helps to improve production efficiency and quality.

[0013] Preferably, the deviation identification module is used to: identify the component position deviation of the dynamic production twin model according to a preset production planning log, and then calculate the reference positioning deviation to generate the component position deviation parameter, specifically including: Calculate the standard position of each component on the dynamic production twin model to generate the position coordinates of multiple components; Construct a component coordinate system network based on the position coordinates of multiple components; According to the preset production planning log, the component position deviation is identified in the component coordinate system network, and the position deviation components are extracted; The reference positioning deviation of the position deviation component is calculated to generate the component position deviation parameters.

[0014] By calculating the standard position of each component, the system can accurately determine the ideal position of each component in the production process, and real-time monitoring of the position of the component helps to detect and correct position deviations in production, thereby improving product quality and consistency. Understanding the standard position of the component helps to optimize the production process, improve production efficiency and reduce unnecessary waste. Establishing a component coordinate system network helps to understand the relative position relationship between the components on the production site as a whole, providing a basis for subsequent position deviation identification. By constructing a coordinate system network, the spatial association between the components can be better analyzed, helping to identify potential position deviation problems. The coordinate system network visualizes the position relationship between the components, helping users to more intuitively understand the structure and Layout, position deviation identification based on production planning logs can realize real-time monitoring and identification of position anomalies in the production process, and timely discover problems. Identifying position deviation components helps to quickly locate and solve potential production problems, improve production efficiency and product quality. By extracting position deviation components, management can make more accurate decisions based on data, optimize production processes and improve overall efficiency. By calculating position deviation parameters, quantifying the degree of position deviation, and improving positioning precision and accuracy, generating position deviation parameters helps to gain an in-depth understanding of the specific circumstances of position deviation and solve specific problems in production in a targeted manner. The generation of position deviation parameters provides data support for continuous improvement of production processes, helps optimize production links, and improves production capacity and quality.

[0015] Preferably, the deviation compensation module is used to: visually mark the dynamic production twin model based on the component position deviation parameter, and perform component positioning compensation calculation, so as to generate the component position deviation compensation parameter, specifically including: Calculate component deviation thresholds based on a preset production planning log; Compare the deviation degrees of component position deviation parameters according to component deviation thresholds, and extract components that exceed the deviation thresholds; Visually marking the dynamic production twin model based on the components exceeding the deviation threshold to generate a component deviation visualization model; The component positioning compensation calculation is performed on the component deviation visualization model to generate component position deviation compensation parameters.

[0016] The present invention standardizes the concept of deviation by calculating the deviation threshold, which helps to unify the judgment standard. The setting of the deviation threshold helps to determine the benchmark of the component position, and further performs deviation comparison and analysis. Identifying components that exceed the deviation threshold helps to promptly discover position anomalies and improve the stability and quality of the production process. Prioritizing components that exceed the deviation threshold improves the efficiency and accuracy of problem solving. Visually marking deviation components can intuitively display position deviations, which helps to analyze and understand problems. Generating a visual model can facilitate exchanges and communication between different teams and speed up problem solving. By calculating compensation parameters, the position of components can be accurately adjusted to reduce deviations and improve production accuracy. Generating compensation parameters can help to continuously improve production processes, reduce deviations, and improve product quality and consistency.

[0017] Preferably, the defect detection module is used to: perform surface texture defect detection on the detail reconstructed omni-directional monitoring image to generate surface texture defect data; and perform geometric morphology analysis of the defect area to obtain defect type characteristics, specifically including: The texture details of the detailed reconstruction omni-directional monitoring image are deeply mined to generate the surface texture detail features of the charging pile; Perform surface texture defect detection on the surface texture detail features of the charging pile to generate surface texture defect data; Performing defect area geometric morphology analysis on the surface texture defect data to generate texture defect area geometric morphology data; The defect type is inferred based on the geometric data of the texture defect area to obtain the defect type characteristics.

[0018] The present invention deeply mines texture details and accurately extracts texture features of the charging pile surface, which helps to identify different texture patterns and structures. Generating texture detail features helps to enrich the data set and improve the accuracy and robustness of model training. Deep mining of texture details better characterizes the visual information of the charging pile surface and helps subsequent defect detection and analysis. Through surface texture defect detection, an automated defect detection process is realized, and production efficiency and quality consistency are improved. Detecting surface texture defects helps to monitor problems in the production process in real time, discover and deal with defects in time, and reduce the defective rate. Defect data is generated for accumulating historical data and further improving the performance of models and algorithms. Through geometric morphology analysis, morphological features of defective areas are extracted to help further understand and classify different types of defects. Analyzing geometric morphology data helps to accurately locate defective areas and provide guidance for subsequent processing and repair. Understanding the geometric morphology of defective areas helps to improve the production process and improve product quality and consistency. By inferring the defect type, automatic classification and labeling are realized, and the efficiency and accuracy of defect identification are improved. Defect type inference based on geometric morphology data helps data-driven decision-making and improvement, and provides a basis for quality management and production optimization.

[0019] Preferably, the process optimization module is used to make abnormal correction decisions for charging piles based on defect type characteristics and component position deviation compensation parameters, and dynamically optimize the production process of charging piles to build an automated detection optimization model, specifically including: Perform sequential process dependency analysis based on production process logic representation to generate production process sequence dependency data; Perform compensation process constraint analysis on component position deviation compensation parameters according to production process sequence dependency data to obtain deviation compensation constraint data; Make charging pile abnormality correction decisions based on defect type characteristics and deviation compensation constraint data to generate abnormality correction strategies; Based on the abnormal correction strategy, the charging pile production process is dynamically optimized, and an automated detection optimization model is built to perform automated visual inspection operations.

[0020] The present invention optimizes the production process and improves production efficiency and product quality by analyzing process sequence dependency data. Understanding the temporal dependency between processes helps to accurately grasp the production progress and process execution sequence. The generated dependency data provides data support for subsequent decision-making and helps to formulate more reasonable production plans and scheduling strategies. Through constraint analysis of compensation parameters, the deviation compensation strategy is optimized to improve the accuracy and efficiency of component position adjustment. The generated deviation compensation constraint data helps to increase the stability of the production process and reduce the impact of deviations. The constraint data can be used to adjust process parameters in real time, adapt to changes in the production environment, and maintain production stability and consistency. According to defect characteristics and compensation constraint data, abnormal correction decisions are made to achieve real-time response and processing of production abnormalities. The formulation of abnormal correction strategies helps to quickly resolve abnormal situations that occur in the production process and reduce defective product rates and production delays. Data-based decisions help to improve the accuracy and efficiency of production decisions and improve production management levels. Dynamic optimization of production processes adjusts production processes according to real-time conditions to improve production efficiency and product quality. The construction of an automated detection optimization model helps to realize automated visual detection operations, reduce human intervention, and improve detection efficiency and consistency. Through dynamic optimization, the production process is continuously improved to gradually improve production efficiency and product quality levels. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a schematic diagram of the steps of an automated visual inspection system for charging pile production according to the present invention; DETAILED DESCRIPTION

[0022] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0023] The present application example provides an automated visual inspection system for charging pile production. The execution subject of the automated visual inspection system for charging pile production includes but is not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as the general computing node of the present application, and the data processing platform includes but is not limited to: at least one of an audio image management system, an information management system, and a cloud data management system.

[0024] See also Figure 1 , is an automated visual inspection system for charging pile production provided by the present invention, and the automated visual inspection system for charging pile production includes an image enhancement module, a twin model module, a deviation recognition module, a deviation compensation module, a defect detection module and a process optimization module; The image enhancement module is used to: obtain a full-scale production monitoring image; perform image detail convolution enhancement on the full-scale production monitoring image, and perform topological structure fitting to generate a component topological structure diagram; The twin model module is used to: identify production process actions on the component topology diagram based on the omnidirectional production monitoring image, and perform three-dimensional simulation modeling to build a dynamic production twin model; The deviation identification module is used to: identify the component position deviation of the dynamic production twin model according to the preset production planning log, and then calculate the reference positioning deviation to generate the component position deviation parameter; The deviation compensation module is used to: visually mark the dynamic production twin model based on the component position deviation parameter, and perform component positioning compensation calculation, thereby generating the component position deviation compensation parameter; The defect detection module is used to: perform surface texture defect detection on the detailed reconstructed omni-directional monitoring image to generate surface texture defect data; and perform geometric morphology analysis of the defect area to obtain defect type characteristics; The process optimization module is used to make charging pile abnormality correction decisions based on defect type characteristics and component position deviation compensation parameters, dynamically optimize the charging pile production process, and build an automated detection optimization model.

[0025] In the embodiment of the present invention, refer to Figure 1 , is a schematic diagram of the steps of an automated visual inspection system for charging pile production according to the present invention. In this example, the steps of the automated visual inspection system for charging pile production include: The image enhancement module is used to: obtain a full-scale production monitoring image; perform image detail convolution enhancement on the full-scale production monitoring image, and perform topological structure fitting to generate a component topological structure diagram; In this embodiment, multiple cameras are arranged at key positions of the production line to ensure that the entire production process can be covered and images can be captured from multiple angles. The parameters of the camera (such as resolution, frame rate, exposure time) are configured to optimize the image quality to ensure that clear images can be obtained under different lighting conditions. The camera is started for real-time image acquisition to obtain all-round production monitoring images, which are stored as image sequences or video files for subsequent processing. A suitable convolution kernel is selected for image detail enhancement. Commonly used convolution kernels include sharpening kernels (such as Laplace operators) or edge detection kernels (such as Sobel operators). A convolution operation is applied to the obtained all-round production monitoring images. The specific steps are as follows: sliding the convolution kernel to the image, calculating the convolution value, updating each pixel value of the image, and sampling. Use boundary processing methods (such as zero padding or mirroring) to process edge pixels during the convolution process, extract key features such as edges, corners, and contours from the enhanced image, and use algorithms such as Canny edge detection or Harris corner detection to achieve this. Select a suitable mathematical model to represent the topological structure of the component, such as the node and edge model in graph theory, or use a more complex geometric model to map the extracted feature points and edge information to the topological structure model for fitting. Use the RANSAC algorithm to handle noise and outliers to ensure the robustness of the fitting. Construct a component topological structure diagram based on the geometric relationship and connection method between components, verify whether the fitted topological structure conforms to the geometric characteristics of the actual production components, and perform manual adjustments or re-fitting when necessary.

[0026] The twin model module is used to: identify production process actions on the component topology diagram based on the omnidirectional production monitoring image, and perform three-dimensional simulation modeling to build a dynamic production twin model; In this embodiment, key frames are extracted from the omnidirectional production monitoring images. These frames should contain important production process actions. Computer vision algorithms (such as HOG, SIFT or CNN) are used to extract key feature points in the image to facilitate subsequent action recognition. Suitable machine learning or deep learning models are selected for action recognition. For example, convolutional neural networks (CNN) or long short-term memory networks (LSTM) are used to analyze time series image data. Model training is performed using labeled process action data, and model parameters are adjusted through cross-validation to improve recognition accuracy. The identified process actions are organized into structured data, and the data of each action is recorded. Timestamp, duration and corresponding component information. According to the timestamp, the corresponding process actions are constructed into an action sequence in sequence for subsequent analysis. Suitable 3D modeling software, such as Blender, Maya or CAD tools, are selected to build the 3D model. Based on the component topology diagram, a 3D model of each component is constructed to ensure that the geometry of the model is consistent with the actual component. The identified production process actions are mapped to the 3D model, and the parameters of the action (such as motion trajectory, speed and time) are set to achieve dynamic simulation. The 3D model is connected to real-time data to generate a dynamic production twin model to ensure that the model can reflect changes in the production process in real time.

[0027] The deviation identification module is used to: identify the component position deviation of the dynamic production twin model according to the preset production planning log, and then calculate the reference positioning deviation to generate the component position deviation parameter; In this embodiment, a preset production planning log is extracted from the production management system to ensure that the log contains information such as the expected position, motion trajectory and timestamp of the component. The position data of the current component is extracted from the dynamic production twin model. This data should include the coordinates (x, y, z) of each component in three-dimensional space and the relevant timestamp. The extracted component position data is formatted to ensure that it is consistent with the data structure of the production planning log for comparison. An appropriate calculation method is selected to identify the deviation of the component position, usually using Euclidean distance or other suitable distance metrics. The actual position of the component in the dynamic production twin model is compared with the preset position in the production planning log, and the deviation value of each component is calculated and recorded. The benchmark component or position is determined according to the production planning log. Usually, the key component or center point is selected as the benchmark. The position deviation of all components relative to the benchmark component is calculated, and the deviation value of each component relative to the benchmark is recorded.

[0028] The deviation compensation module is used to: visually mark the dynamic production twin model based on the component position deviation parameter, and perform component positioning compensation calculation, thereby generating the component position deviation compensation parameter; In this embodiment, component position deviation parameters are collected, including the actual position and standard position of each component, and a suitable visualization tool (such as Matplotlib, Mayavi or Unity) is selected to generate a three-dimensional visualization model. In the dynamic production twin model, the actual position of each component is compared with the standard position, and the components whose position deviation exceeds the preset threshold are visually marked (such as changing color or shape) for quick identification, and a dynamic production twin model with deviation marks is generated to ensure that the visualization effect is clear and intuitive. According to the visualization results, the components that need positioning compensation and the corresponding compensation targets (such as adjusting the actual position to the standard position) are determined, and a suitable compensation calculation method (such as geometric compensation, linear interpolation, or a method based on an optimization algorithm) is selected for each component that needs positioning compensation. For the components to be compensated, compensation calculation is performed based on their position deviation parameters to determine the new position coordinates. The calculated component position deviation compensation parameters, including the compensation value and the final position coordinates, are recorded. An appropriate verification method (such as repeated measurement, simulation test) is selected to confirm the compensation effect. The position of the compensated component is compared with the standard position again to verify the effectiveness of the compensation. If there is a new deviation, record it and feed it back to the compensation calculation process for further adjustment. An appropriate database or data storage method (such as SQL database, NoSQL database or CSV file) is selected to record the compensation parameters and results. The compensation parameters of each component, the key data in the compensation calculation process and the verification results are recorded for subsequent analysis and optimization. The compensation data is analyzed regularly to identify potential systematic deviations and improvement opportunities.

[0029] The defect detection module is used to: perform surface texture defect detection on the detailed reconstructed omni-directional monitoring image to generate surface texture defect data; and perform geometric morphology analysis of the defect area to obtain defect type characteristics; In this embodiment, a suitable defect detection algorithm is selected, such as a convolutional neural network (CNN), a support vector machine (SVM) or a traditional image processing method (such as edge detection, threshold segmentation), the selected defect detection algorithm is applied to the monitoring image, the surface texture defects are identified and marked, the detected defect location, type and severity are recorded, and a surface texture defect data set is generated, which includes the coordinates, size and feature description of each defect. A suitable geometric morphology analysis method is selected, such as morphological processing, region growing algorithm or contour extraction (such as Canny edge detection), and the defect area is extracted from the monitoring image according to the detected defect location to ensure that the extracted area is complete and accurate. The geometric morphology analysis algorithm is applied to the extracted defect area to calculate the geometric features of the defect, such as area, perimeter, shape factor, etc., to generate defect type feature data, and the geometric morphology parameters of each defect are recorded for subsequent analysis. A suitable feature extraction model (such as a decision tree, random forest or deep learning model) is selected to classify the defect type, and the geometric morphology data and texture features are combined and input into the defect type classification model to infer the defect type (such as scratches, dents, bubbles, etc.), and the type features of each defect are recorded to ensure that each defect has a clear type identification and description.

[0030] The process optimization module is used to make charging pile abnormality correction decisions based on defect type characteristics and component position deviation compensation parameters, dynamically optimize the charging pile production process, and build an automated detection optimization model.

[0031] In this embodiment, a suitable decision support model (such as a decision tree, random forest, fuzzy logic or a rule-based system) is selected to generate anomaly correction decisions, defect type characteristics and component position deviation compensation parameters are input into the decision model, the impact is analyzed, and a charging pile abnormality correction decision is generated. The specific content of the correction decision, including the required compensation measures, adjustment plans and expected results, is recorded. According to the correction decision, optimization goals and evaluation indicators (such as production efficiency, defect rate, resource utilization, etc.) are set, and a suitable dynamic optimization algorithm (such as genetic algorithm, particle swarm optimization or simulated annealing) is selected for process optimization. Based on the optimization goal and the existing production process, the optimization algorithm is run to dynamically adjust the production process and process parameters, the results of each optimization are recorded, and it is analyzed whether the adjusted process has achieved the expected goal. The framework of the automated detection optimization model is designed, including data input, processing modules and output feedback mechanisms, and suitable image processing and machine learning algorithms (such as convolutional neural networks, support vector machines, etc.) are selected for automated detection. Based on the production data and optimization strategies of the charging piles, an automated detection optimization model is constructed to ensure that the model can monitor abnormal conditions in the production process in real time, and the training process and parameter settings of the model are recorded to ensure the repeatability and scalability of the model.

[0032] In this embodiment, the image enhancement module is used to: obtain a full-scale production monitoring image; perform image detail convolution enhancement on the full-scale production monitoring image, and perform topological structure fitting to generate a component topological structure diagram, specifically including: Real-time monitoring of the charging pile production process based on industrial cameras to obtain all-round production monitoring images; Perform image detail convolution enhancement on the omnidirectional production monitoring image to obtain a detailed reconstructed omnidirectional monitoring image; Perform multi-component visual recognition on the detailed reconstruction of the all-round monitoring image and mark the charging pile component nodes; Perform inter-component connection analysis on charging pile component nodes and extract component connection relationships; The topological structure of the detail-enhanced omnidirectional monitoring image is fitted according to the component connection relationship to generate a component topological structure diagram.

[0033] In this embodiment, multiple industrial cameras are installed at key positions of the charging pile production line to ensure coverage of all aspects of the production process. The cameras are ensured to have high resolution and appropriate shooting angles to obtain clear images. The cameras and data acquisition system are configured, and real-time image capture parameters (such as frame rate, exposure time, etc.) are set. The image acquisition program is started to regularly capture all-round monitoring images of the charging pile production process. The collected all-round monitoring images are pre-processed, including denoising, grayscale, and normalization, to improve the effect of subsequent processing. An image processing library (such as OpenCV or TensorFlow) is used to implement convolution operations. Image details are enhanced by applying detail enhancement convolution kernels (such as Laplace filters or high-pass filters). The enhanced images are recorded to ensure that the details are clearly reconstructed to facilitate subsequent visual recognition. A suitable computer vision algorithm (such as CNN, YOLO, Faster The R-CNN is used to identify components in the image, and the recognition model is trained using the labeled training set to improve the recognition accuracy. The trained model is applied to the detailed reconstructed omnidirectional monitoring image to identify and mark the component nodes of the charging pile (such as battery, charging interface, controller, etc.), set the connection rules and relationships between components (such as electrical connection, physical connection, etc.), and prepare the corresponding analysis algorithm. Based on the marked component node data, the connection relationship between the nodes is analyzed to determine which components are connected to each other, and a connection relationship diagram is generated. The connection information between each component, including the connection type and strength, is recorded, and a suitable graphics processing algorithm (such as the minimum spanning tree algorithm) is selected to generate the topological structure of the component. According to the extracted component connection relationship, the topological structure is fitted for the detailed enhanced omnidirectional monitoring image to generate a topological structure diagram of the charging pile component, ensuring that the topological structure diagram can clearly show the connection relationship and overall structure between the components, and the generated topological structure diagram is saved as a graphic file (such as SVG, PNG) or a data file (such as JSON) for subsequent analysis and visualization.

[0034] In this embodiment, the specific steps of performing image detail convolution enhancement on the omnidirectional production monitoring image to obtain a detail-reconstructed omnidirectional monitoring image are: Filter and reduce noise on all-round production monitoring images to generate noise-reduced and optimized monitoring images; Perform multi-scale decomposition on the noise reduction optimized monitoring image to obtain multiple frequency feature images; Perform convolution feature extraction on multiple frequency feature images to obtain the residual features of each image; Perform multiple rounds of iterative detail optimization on multiple frequency feature images according to the residual features of each image to generate multiple detail optimized feature images; Perform local pooling compression analysis on multiple detail optimized feature images to extract local implicit vectors of the images; Decoding and reconstructing the image according to the local latent vector of the image to obtain multiple super-pixel images; Multiple super-pixel images are fused and reconstructed to obtain a detailed reconstructed omnidirectional monitoring image.

[0035] In this embodiment, a suitable denoising filter (such as a Gaussian filter, a median filter or a bilateral filter) is selected, and according to the characteristics of the image noise, an image processing library (such as OpenCV) is used to filter the omnidirectional monitoring image, process each pixel to reduce noise interference, record the processed image to ensure that the noise is significantly reduced, generate a denoised optimized monitoring image, select a suitable multi-scale decomposition method (such as wavelet transform, Laplace pyramid or Gaussian pyramid) to extract features of different frequencies, perform multi-scale decomposition on the denoised optimized monitoring image to obtain multiple frequency feature images, ensure that each frequency feature image can represent different image details and structural information, record the frequency features of each image, design a suitable convolution kernel (such as Sobel operator, Laplacian operator or custom convolution kernel) for feature extraction, apply convolution operation to each frequency feature image, extract the edge and detail information of the image, obtain the residual feature of each image, record each residual feature image to ensure that it contains important structural information, select a suitable iterative optimization algorithm (such as Adam optimization, SGD or adaptive learning rate algorithm) to improve image detail. Section, according to the residual features of each image, multiple rounds of iterative detail optimization are performed on multiple frequency feature images to gradually enhance the detail performance of the image, and the images after each round of optimization are recorded to ensure that the details are gradually clear, and multiple detail optimization feature images are generated. A local pooling method (such as maximum pooling, average pooling or adaptive pooling) is selected to extract local latent vectors, and local pooling analysis is performed on multiple detail optimization feature images to compress image information and extract local feature vectors. The local latent vector of each image is recorded to provide a basis for subsequent decoding and reconstruction, and a suitable image decoding and reconstruction algorithm (such as inverse convolution, back propagation or GAN reconstruction) is selected to generate super-pixel images. According to the extracted local latent vector of the image, each image is decoded and reconstructed to obtain multiple super-pixel images, ensuring that the reconstructed super-pixel image can reflect the detail information of the original image, and a suitable image fusion algorithm (such as weighted average, maximum selection or multi-resolution fusion) is selected to merge the super-pixel images, and multiple super-pixel images are fused and reconstructed to generate a full-range monitoring image with detail reconstruction, ensuring that the fused image can fully display each detail and improve the overall visual quality.

[0036] In this embodiment, the twin model module is used to: identify the production process actions of the component topology diagram based on the omnidirectional production monitoring image, and perform three-dimensional simulation modeling to build a dynamic production twin model, specifically including: Performing time-series multi-frame sampling on the detail-reconstructed omni-directional monitoring image to obtain a time-series frame sequence of the monitoring image; Perform frame-by-frame dynamic optical flow tracking on the monitoring image time sequence frame sequence to generate image frame dynamic optical flow tracking trajectory; Perform production process action recognition on the dynamic optical flow tracking trajectory of the image frame and extract production process action data; Conduct time-series process logic analysis on production process action data to obtain production process logic representation; The production process logic representation is used to perform three-dimensional simulation modeling of the component topology diagram and build a dynamic production twin model.

[0037] In this embodiment, the frequency and interval of time series sampling (such as sampling 10 frames per second) are determined to ensure that dynamic changes in the production process are captured, continuous image frames are extracted from the omnidirectional monitoring image reconstructed with details to form a monitoring image time series frame sequence, the timestamp of each image frame is recorded to ensure the time order of the frame sequence, a suitable optical flow calculation method (such as the Lucas-Kanade method or the Horn-Schunck method) is selected for frame-by-frame dynamic optical flow tracking, the monitoring image time series frame sequence is processed frame by frame, the optical flow between each frame is calculated, the image frame dynamic optical flow tracking trajectory is generated, the speed and direction of each optical flow is recorded to form an optical flow tracking trajectory data set, a suitable action recognition algorithm (such as gesture recognition, convolutional neural network (CNN) or recurrent neural network (RNN)) is selected for process action recognition, the generated optical flow tracking trajectory is input into the action recognition model, and the key actions in the production process (such as assembly) are identified. , disassembly, testing, etc.), extract and record the time, duration and corresponding characteristics of each action, generate a production process action data set, select an appropriate logic analysis method (such as state machine, Petri net or logistic regression) to analyze the process action, perform sequential process logic analysis based on the extracted production process action data, identify the logical relationship between each action (such as cause and effect, sequential relationship), generate a production process logic representation, record the before and after relationship and dependency of each action, select a suitable 3D modeling software (such as Blender, Unity or MATLAB) to build a dynamic production twin model, use the production process logic representation and component topology diagram to build a dynamic 3D simulation model, ensure that the model can simulate various process actions and component interactions in the production process, record key parameters and states, run the model in a simulation environment, verify its consistency with the actual production process, and adjust the model parameters to improve accuracy.

[0038] In this embodiment, the specific steps of using the production process logic representation to perform three-dimensional simulation modeling on the component topology diagram and constructing a dynamic production twin model are as follows: Extracting multiple key frame images based on the monitoring image time series frame sequence; Perform component feature point recognition on multiple key frame images and mark the feature points of multiple frame images; Performing multi-directional image posture analysis on a plurality of key frame images to generate multi-directional image posture distortion data; According to the multi-directional image posture distortion data, feature points of multiple frame images are matched in spatial position to obtain accurate spatial position data of feature points; Based on the precise spatial position data of the feature points, the component topology structure diagram is modeled with feature three-dimensional point cloud, and a component topology three-dimensional point cloud model is constructed; Perform full-process process simulation based on the production process logic representation to obtain the full-process process simulation trajectory; According to the full-process process simulation trajectory, the component topology three-dimensional point cloud model is dynamically rendered to build a dynamic production twin model.

[0039] In this embodiment, a standard for key frame extraction (such as motion change, scene change or time interval) is determined, multiple key frame images that meet the standard are extracted from the monitoring image time series frame sequence to ensure that they can represent important stages in the production process, the timestamp of each key frame and its position in the time series are recorded, a suitable feature point detection algorithm (such as SIFT, SURF, ORB, etc.) is selected to identify component feature points, the feature point detection algorithm is applied to the extracted multiple key frame images, the component feature points in each image are identified and marked, the coordinates and descriptors of each feature point are recorded, a feature point data set is generated, a suitable posture estimation method (such as PnP algorithm, ICP algorithm) is selected for image posture analysis, a multi-directional posture analysis is performed on multiple key frame images, the posture distortion data (such as rotation matrix and translation vector) of each image is calculated, the posture distortion data is recorded for subsequent feature point matching, and a suitable feature point matching algorithm (such as FLANN, BFMat) is selected. cher) to match the spatial position of feature points. According to the multi-directional posture distortion data, the matching algorithm is used to match the spatial position of feature points of multiple frame images to obtain the spatial precise position data of the feature points. The matching results are recorded to ensure that the position of the feature points in the three-dimensional space is accurate. The point cloud modeling software (such as PCL, Meshlab) is used to construct the feature three-dimensional point cloud. Based on the spatial precise position data of the feature points, the feature three-dimensional point cloud modeling of the component topology structure diagram is performed to generate the three-dimensional point cloud model of the component to ensure that the point cloud model can truly reflect the shape and characteristics of the component. The point cloud data is recorded, and the appropriate process simulation tool (such as AnyLogic, Simul8) is selected for full-process process simulation. Based on the previously extracted production process logic representation, the full-process process simulation is performed to generate the process simulation trajectory, and the key parameters and states in the simulation process are recorded to ensure the validity of the simulation results. The appropriate rendering engine (such as Unity, Unreal Engine) is selected for dynamic process rendering. According to the full-process process simulation trajectory, the component topology three-dimensional point cloud model is dynamically rendered to build a dynamic production twin model to ensure that the rendering effect is realistic and can dynamically display the component interaction and behavior in the production process.

[0040] In this embodiment, the deviation identification module is used to: identify the component position deviation of the dynamic production twin model according to the preset production planning log, and then calculate the reference positioning deviation to generate the component position deviation parameter, which specifically includes: Calculate the standard position of each component on the dynamic production twin model to generate the position coordinates of multiple components; Construct a component coordinate system network based on the position coordinates of multiple components; According to the preset production planning log, the component position deviation is identified in the component coordinate system network, and the position deviation components are extracted; The reference positioning deviation is calculated for the position deviation component to generate the component position deviation parameters.

[0041] In this embodiment, the standard position of each component is determined (such as based on a design drawing or production specification), the ideal coordinates of each component are recorded, the position of each component in the dynamic production twin model is calculated, the actual position coordinates in the three-dimensional space are extracted, the position coordinates of the generated multiple components are recorded, and it is ensured that each coordinate corresponds to its corresponding component one by one, the structure of the component coordinate system network is designed, including the relationship and connection method between the components, to ensure that the relative positions of the components can be reflected, a coordinate system network model is constructed according to the position coordinates of the generated multiple components, the relative position relationship between the components is recorded, and the component coordinate system network is visualized using a graphical modeling tool (such as Graphviz or MATLAB). The network is convenient for analysis. The preset production planning log is obtained, the deviation requirements of the target position and the actual position of the component are recorded, the component coordinate system network is analyzed, the deviation between the actual position and the standard position is identified, and the position deviation component is extracted. The threshold judgment method (such as setting the allowable deviation range) is used to mark the components with deviations exceeding the range, and the appropriate deviation calculation method (such as Euclidean distance calculation and Manhattan distance calculation) is selected to quantify the displacement deviation of the component. The reference positioning deviation is calculated for each identified position deviation component, and the specific position deviation parameters of the component (such as deviation value and deviation direction) are generated. The deviation parameters of each component are recorded to ensure that the deviation of each component can be tracked and managed.

[0042] In this embodiment, the dynamic production twin model is visually marked based on the component position deviation parameter, and the component positioning compensation calculation is performed to generate the component position deviation compensation parameter, which is specifically used for: Calculate component deviation thresholds based on a preset production planning log; Compare the deviation degrees of component position deviation parameters according to component deviation thresholds, and extract components that exceed the deviation thresholds; Visually marking the dynamic production twin model based on the components exceeding the deviation threshold to generate a component deviation visualization model; The component positioning compensation calculation is performed on the component deviation visualization model to generate component position deviation compensation parameters.

[0043] In this embodiment, according to the production requirements and component characteristics, the deviation threshold of each component (such as the maximum allowable deviation range) is set, relevant data is extracted from the preset production planning log, the deviation threshold of each component is calculated and recorded, and a deviation threshold data set is generated to ensure that the threshold of each component corresponds to its position parameter one by one. A suitable comparison method (such as absolute difference calculation, relative difference calculation) is selected to measure the relationship between the actual position deviation and the threshold, the component position deviation parameters are compared with the corresponding deviation threshold one by one, and the components exceeding the deviation threshold are extracted. These components exceeding the threshold and their deviation degrees are recorded to form a data set of components exceeding the deviation, and a suitable visualization tool (such as MATLAB, Python's Matplotlib or Plotly) is used to generate a visualization model. Based on the components that exceed the deviation threshold, the dynamic production twin model is visually marked, and different colors or graphics are used to identify the deviation components. A component deviation visualization model is generated to ensure that the model can clearly show which components have deviations and their degree. An appropriate compensation calculation method (such as geometric compensation, linear interpolation, or compensation based on an optimization algorithm) is selected to adjust the component position. For each component that exceeds the deviation threshold, a positioning compensation calculation is performed based on its deviation parameters, and the component position deviation compensation parameters are generated. The calculated compensation parameters are recorded to ensure that the component can be effectively adjusted to its standard position.

[0044] In this embodiment, the defect detection module is used to: perform surface texture defect detection on the detailed reconstructed omni-directional monitoring image to generate surface texture defect data; and perform geometric morphology analysis of the defect area to obtain defect type characteristics, specifically including: The texture details of the detailed reconstruction omni-directional monitoring image are deeply mined to generate the surface texture detail features of the charging pile; Perform surface texture defect detection on the surface texture detail features of the charging pile to generate surface texture defect data; Performing defect area geometric morphology analysis on the surface texture defect data to generate texture defect area geometric morphology data; The defect type is inferred based on the geometric data of the texture defect area to obtain the defect type characteristics.

[0045] In this embodiment, a suitable texture feature extraction algorithm (such as local binary pattern (LBP), gray level co-occurrence matrix (GLCM) or Gabor filter) is selected to perform texture detail analysis, and the selected texture analysis method is applied to the omnidirectional monitoring image with detail reconstruction to extract the texture detail features of the charging pile surface, generate a texture feature data set, including texture contrast, energy, entropy and other indicators to ensure that the details of the surface texture can be fully described, and a suitable defect detection algorithm (such as a convolutional neural network (CNN) based on deep learning, an image processing method (such as edge detection or threshold segmentation)) is selected to perform texture defect detection, and the extracted texture features are input into the defect detection model to identify the texture defects on the charging pile surface and generate defect data, and record the detected defect location, type and severity. Gravity, form a surface texture defect data set, select appropriate geometric morphology analysis tools (such as morphological processing, region growing algorithm or contour analysis) to perform geometric morphology analysis of the defect area, perform geometric morphology analysis on the detected texture defect area, extract the geometric morphology features of the defect area (such as area, perimeter, shape factor, etc.), record the generated geometric morphology data of the texture defect area to ensure that it can reflect the geometric characteristics of the defects, select appropriate defect type inference model (such as support vector machine (SVM), decision tree or deep learning model) for defect classification, combine the geometric morphology data with texture features, input them into the inference model to infer the defect type, obtain the defect type features (such as scratches, dents, bubbles, etc.), record the inference results, and ensure that each defect has a clear type identification.

[0046] In this embodiment, the process optimization module is used to make abnormal correction decisions for charging piles based on defect type characteristics and component position deviation compensation parameters, and dynamically optimize the production process of charging piles to build an automated detection optimization model, specifically including: Perform sequential process dependency analysis based on production process logic representation to generate production process sequence dependency data; Perform compensation process constraint analysis on component position deviation compensation parameters according to production process sequence dependency data to obtain deviation compensation constraint data; Make charging pile abnormality correction decisions based on defect type characteristics and deviation compensation constraint data to generate abnormality correction strategies; Based on the abnormal correction strategy, the charging pile production process is dynamically optimized, and an automated detection optimization model is built to perform automated visual inspection operations.

[0047] In this embodiment, a process dependency graph is constructed based on the logical representation of the production process, and the dependency relationship between each process step (such as sequence and parallelism) is defined. By analyzing the time series of each process step, the dependency relationship in the process flow is identified, and the production process sequence dependency data is generated. A graph theory algorithm (such as topological sorting) is used to ensure the rationality and executability of the process sequence, and the dependency relationship of each process step is recorded. According to the component position deviation compensation parameters, the constraint conditions of the compensation process (such as the maximum allowable deviation) are defined, and the production process sequence dependency data is analyzed to identify the constraint relationship for deviation compensation under different process steps, and the deviation compensation constraint data is generated. A constraint optimization algorithm (such as linear programming or nonlinear programming) is used to analyze and record the constraints to ensure that the compensation measures can be effectively implemented in the process flow, and a suitable decision support model (such as a decision tree, random forest or fuzzy logic system) is selected to generate an abnormality correction strategy based on The defect type characteristics and deviation compensation constraint data are input into the decision model for analysis, and the charging pile abnormality correction decision is generated. The applicable conditions and expected effects of each correction strategy are recorded to ensure that the strategy can effectively respond to different types of abnormalities. Based on the abnormality correction strategy, an automated detection optimization model is constructed, and dynamic optimization is performed in combination with the real-time data of the production process. The production process is dynamically adjusted using optimization algorithms (such as genetic algorithms, ant colony algorithms, or particle swarm optimization) to ensure that the process flow can adapt to changes in real time. The optimization results are recorded to ensure that each adjustment can improve production efficiency and product quality. An automated visual inspection system is established, and appropriate image processing and machine learning algorithms (such as convolutional neural networks) are selected for defect detection. According to the dynamically optimized process flow and abnormality correction strategy, automated visual inspection operations are performed to monitor abnormal conditions in the production process in real time, record the inspection results and feedback information, and ensure that the system can continuously learn and optimize.

[0048] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0049] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be 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 the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An automated visual inspection system for charging pile production, characterized in that: The automated visual inspection system applied to charging pile production includes an image enhancement module, a twin model module, a deviation recognition module, a deviation compensation module, a defect detection module and a process optimization module; The image enhancement module is used to: obtain a full range of production monitoring images; perform image detail convolution enhancement on the full range of production monitoring images, and perform topological structure fitting to generate a component topological structure diagram; The twin model module is used to: identify production process actions on the component topology diagram based on the omnidirectional production monitoring image, and perform three-dimensional simulation modeling to build a dynamic production twin model; The deviation identification module is used to: identify the component position deviation of the dynamic production twin model according to the preset production planning log, and then calculate the reference positioning deviation to generate the component position deviation parameter; The deviation compensation module is used to: visually mark the dynamic production twin model based on the component position deviation parameter, and perform component positioning compensation calculation, thereby generating the component position deviation compensation parameter; The defect detection module is used to: perform surface texture defect detection on the detailed reconstructed omni-directional monitoring image to generate surface texture defect data; and perform geometric morphology analysis of the defect area to obtain defect type characteristics; The process optimization module is used to make charging pile abnormality correction decisions based on defect type characteristics and component position deviation compensation parameters, dynamically optimize the charging pile production process, and build an automated detection optimization model.

2. The automated visual inspection system for charging pile production according to claim 1 is characterized in that: The image enhancement module is used to: obtain a full range of production monitoring images; perform image detail convolution enhancement on the full range of production monitoring images, and perform topological structure fitting to generate a component topological structure diagram, specifically including: Real-time monitoring of the charging pile production process based on industrial cameras to obtain all-round production monitoring images; Perform image detail convolution enhancement on the omnidirectional production monitoring image to obtain a detailed reconstructed omnidirectional monitoring image; Perform multi-component visual recognition on the detailed reconstruction of the all-round monitoring image and mark the charging pile component nodes; Perform inter-component connection analysis on charging pile component nodes and extract component connection relationships; The topological structure of the detail-enhanced omnidirectional monitoring image is fitted according to the component connection relationship to generate a component topological structure diagram.

3. The automated visual inspection system for charging pile production according to claim 2 is characterized in that: The specific steps of performing image detail convolution enhancement on the omnidirectional production monitoring image to obtain a detail-reconstructed omnidirectional monitoring image are as follows: Filter and reduce noise on all-round production monitoring images to generate noise-reduced and optimized monitoring images; Perform multi-scale decomposition on the noise reduction optimized monitoring image to obtain multiple frequency feature images; Perform convolution feature extraction on multiple frequency feature images to obtain the residual features of each image; Perform multiple rounds of iterative detail optimization on multiple frequency feature images according to the residual features of each image to generate multiple detail optimized feature images; Perform local pooling compression analysis on multiple detail optimized feature images to extract local implicit vectors of the images; Decoding and reconstructing the image according to the local latent vector of the image to obtain multiple super-pixel images; Multiple super-pixel images are fused and reconstructed to obtain a detailed reconstructed omnidirectional monitoring image.

4. The automated visual inspection system for charging pile production according to claim 1 is characterized in that: The twin model module is used to: identify the production process actions of the component topology diagram based on the omnidirectional production monitoring image, and perform three-dimensional simulation modeling to build a dynamic production twin model, specifically including: Performing time-series multi-frame sampling on the detail-reconstructed omni-directional monitoring image to obtain a time-series frame sequence of the monitoring image; Perform frame-by-frame dynamic optical flow tracking on the monitoring image time sequence frame sequence to generate image frame dynamic optical flow tracking trajectory; Perform production process action recognition on the dynamic optical flow tracking trajectory of the image frame and extract production process action data; Conduct time-series process logic analysis on production process action data to obtain production process logic representation; The production process logic representation is used to perform three-dimensional simulation modeling of the component topology diagram and build a dynamic production twin model.

5. The automated visual inspection system for charging pile production according to claim 4 is characterized in that: The specific steps of using the production process logic representation to perform three-dimensional simulation modeling on the component topology diagram and constructing a dynamic production twin model are as follows: Extracting multiple key frame images based on the monitoring image time series frame sequence; Perform component feature point recognition on multiple key frame images and mark the feature points of multiple frame images; Performing multi-directional image posture analysis on a plurality of key frame images to generate multi-directional image posture distortion data; According to the multi-directional image posture distortion data, feature points of multiple frame images are matched in spatial position to obtain accurate spatial position data of feature points; Based on the precise spatial position data of the feature points, the component topology structure diagram is modeled with feature three-dimensional point cloud to construct a component topology three-dimensional point cloud model; Perform full-process process simulation based on the production process logic representation to obtain the full-process process simulation trajectory; According to the full-process process simulation trajectory, the component topology three-dimensional point cloud model is dynamically rendered to build a dynamic production twin model.

6. The automated visual inspection system for charging pile production according to claim 1 is characterized in that: The deviation identification module is used to: identify the component position deviation of the dynamic production twin model according to the preset production planning log, and then calculate the reference positioning deviation to generate the component position deviation parameter, which specifically includes: Calculate the standard position of each component on the dynamic production twin model to generate the position coordinates of multiple components; Construct a component coordinate system network based on the position coordinates of multiple components; According to the preset production planning log, the component position deviation is identified in the component coordinate system network, and the position deviation components are extracted; The reference positioning deviation of the position deviation component is calculated to generate the component position deviation parameters.

7. The automated visual inspection system for charging pile production according to claim 1 is characterized in that: The deviation compensation module is used to: visually mark the dynamic production twin model based on the component position deviation parameter, and perform component positioning compensation calculation, so as to generate the component position deviation compensation parameter, specifically including: Calculate component deviation thresholds based on a preset production planning log; Compare the deviation degrees of component position deviation parameters according to component deviation thresholds, and extract components that exceed the deviation thresholds; Visually marking the dynamic production twin model based on the components exceeding the deviation threshold to generate a component deviation visualization model; The component positioning compensation calculation is performed on the component deviation visualization model to generate component position deviation compensation parameters.

8. The automated visual inspection system for charging pile production according to claim 1 is characterized in that: The defect detection module is used to: perform surface texture defect detection on the detailed reconstructed omni-directional monitoring image to generate surface texture defect data; and perform geometric morphology analysis of the defect area to obtain defect type characteristics, specifically including: The texture details of the detailed reconstruction omni-directional monitoring image are deeply mined to generate the surface texture detail features of the charging pile; Perform surface texture defect detection on the surface texture detail features of the charging pile to generate surface texture defect data; Performing defect area geometric morphology analysis on the surface texture defect data to generate texture defect area geometric morphology data; The defect type is inferred based on the geometric data of the texture defect area to obtain the defect type characteristics.

9. The automated visual inspection system for charging pile production according to claim 1, characterized in that: The process optimization module is used to make abnormal correction decisions for charging piles based on defect type characteristics and component position deviation compensation parameters, and dynamically optimize the production process of charging piles to build an automated detection optimization model, specifically including: Perform sequential process dependency analysis based on production process logic representation to generate production process sequence dependency data; Perform compensation process constraint analysis on component position deviation compensation parameters according to production process sequence dependency data to obtain deviation compensation constraint data; Make charging pile abnormality correction decisions based on defect type characteristics and deviation compensation constraint data to generate abnormality correction strategies; Based on the abnormal correction strategy, the charging pile production process is dynamically optimized, and an automated detection optimization model is built to perform automated visual inspection operations.

Images