A plate dispensing automation control method and system based on visual detection

Abstract

This invention relates to the field of industrial automation control technology, and discloses a vision-based automated control method and system for sheet metal subcontracting. The method includes acquiring raw image data and production-related data and performing denoising processing; extracting multi-scale texture features to determine key regions; comparing and optimizing texture identification data with a texture comparison template library; generating a unique identification code and comparing it with production records; planning dynamic subcontracting path nodes; calibrating path deviations to obtain a stable execution scheme; and identifying dynamic interference to generate control commands to execute the subcontracting operation. This invention improves the accuracy of texture feature extraction through image preprocessing, ensures the reliability of identity verification based on template library comparison and identification codes, and dynamically plans paths and calibrates deviations using production data to adapt to dynamic production interference. This achieves fully automated and precise control of the sheet metal subcontracting process, effectively solving the problems of inaccurate and inefficient subcontracting in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control technology, and in particular to an automated control method and system for sheet metal sub-packaging based on vision inspection. Background Technology

[0002] In the process of modern manufacturing transforming towards intelligent manufacturing, the level of automation in the sheet metal processing and subcontracting stages directly determines production efficiency and product quality, becoming a core manifestation of industry competitiveness. With the popularization of large-scale production models, the demand for batch processing, precise classification, and efficient subcontracting of sheet metal is becoming increasingly urgent. In this process, rapid identification of sheet metal identity and accurate determination of surface characteristics have become key technological supports, and the maturity of these technologies directly affects the turnover efficiency of the entire production line.

[0003] Current mainstream sheet metal subcontracting solutions largely rely on traditional mechanical sensing or single-feature recognition technologies, which struggle to address the diverse challenges of complex production environments. When processing sheets from different batches with varying textures, these solutions often suffer from low accuracy due to a lack of deep analysis of subtle surface features. Furthermore, existing technologies are insufficiently adaptable to dynamic scenarios such as production cycle fluctuations and equipment load changes, easily leading to mismatches between subcontracting instructions and actual production conditions. Crucially, the application of industrial vision technology is not deep enough, failing to fully leverage its advantages in surface feature extraction and environmental interference resistance. This results in a lack of reliable logical connection between sheet metal identity verification and subcontracting path planning, potentially causing subcontracting errors and production stoppages.

[0004] Existing technologies struggle to achieve accurate identity verification through efficient matching of surface feature extraction with production records, resulting in inaccurate subcontracting and low production efficiency. Summary of the Invention

[0005] This invention provides an automated control method and system for sheet material subcontracting based on visual inspection, in order to solve the problems of inaccurate subcontracting and low production efficiency in the existing technology.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an automated control method for sheet metal subcontracting based on visual inspection, comprising: The original image data and production-related data of the board surface are acquired. The production-related data includes production record data, real-time production cycle data and production equipment load data. The original image data is then denoised to obtain clear image data. Based on the clear image data, multi-scale texture features are extracted and analyzed, and key areas of the board surface features are determined through texture clustering and integration. The key regions of the board surface features are matched and optimized with a pre-established texture comparison template library to detect the periodic repetition pattern of the texture and adjust the feature description to obtain the surface texture identification data of the board. Based on the surface texture identification data of the board, combined with the pre-established defect boundary information database, a unique identification code for the board is generated, which is compared with the production record data to obtain the board identification conclusion. Based on the board material authentication conclusion, obtain the corresponding subcontracting instruction sequence, combine it with the production association data, plan dynamic subcontracting path nodes, and obtain a subcontracting path node planning scheme. Based on the sub-package path node planning scheme, data frequency matching is performed, path planning conflict points are detected and deviations are calibrated, path nodes are updated, and a stable sub-package execution scheme is obtained. Based on the stable subcontracting execution scheme, dynamic interference in production is identified, automated control commands are generated, transmitted to production line equipment, and the sheet material subcontracting operation is executed.

[0007] Secondly, the present invention provides an automated control system for sheet metal subcontracting based on visual inspection, comprising: The data acquisition module acquires the original image data and production-related data of the board surface. The production-related data includes production record data, real-time production cycle data, and production equipment load data. The original image data is denoised to obtain clear image data. The feature extraction module extracts and analyzes multi-scale texture features based on the clear image data, and determines key areas of the board surface features through texture clustering and integration. The identifier generation module performs similarity matching and optimization between the key feature areas of the board surface and a pre-established texture comparison template library, detects the periodic repetition pattern of the texture and adjusts the feature description to obtain the texture identifier data of the board surface. The identity verification module generates a unique code for the board based on the surface texture identification data of the board and a pre-established defect boundary information database. The code is then compared with the production record data to obtain the board identity verification conclusion. The path planning module obtains the corresponding subcontracting instruction sequence based on the board material authentication conclusion, and plans dynamic subcontracting path nodes in conjunction with the production association data to obtain a subcontracting path node planning scheme. The path optimization module performs data frequency matching based on the sub-package path node planning scheme, detects path planning conflict points and calibrates deviations, updates path nodes, and obtains a stable sub-package execution scheme. The instruction execution module, based on the stable subcontracting execution scheme, identifies dynamic interference in production, generates automated control instructions, transmits them to the production line equipment, and executes the sheet metal subcontracting operation.

[0008] Compared with the prior art, the present invention has the following beneficial effects: This invention obtains clear image data by performing denoising processing on the original image data, including grayscale conversion, brightness suppression, adaptive filtering, and edge enhancement, providing a reliable foundation for multi-scale texture feature extraction and effectively improving the accuracy of identifying key areas of surface features on sheet materials.

[0009] This invention generates surface texture identification data for boards by integrating texture clustering and similarity matching optimization, and generates a unique identification code by combining it with a defect boundary information database and comparing it with production record data. This solves the problem of insufficient accuracy in board identification in existing technologies and ensures the reliability of identity verification conclusions.

[0010] (3) This invention combines production-related data to plan dynamic subcontracting path nodes, obtains a stable subcontracting execution scheme through data frequency matching, conflict point detection and deviation calibration, and generates automated control instructions by identifying dynamic interference in production, thereby realizing full-process automated control and improving the efficiency and stability of board subcontracting. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the automated control method for subcontracting of sheet metal based on visual inspection provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the automated control system for sheet material subcontracting based on vision detection provided in the second embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Reference Figure 1 The first embodiment of the present invention provides an automated control method for board subcontracting based on visual inspection, comprising the following steps: S11, acquire the original image data and production-related data of the board surface, the production-related data including production record data, real-time production cycle data and production equipment load data, and perform noise reduction processing on the original image data to obtain clear image data; S12, Based on the clear image data, multi-scale texture features are extracted and analyzed, and key areas of the board surface features are determined through texture clustering and integration. S13, perform similarity matching and optimization between the key areas of the board surface features and the pre-established texture comparison template library, detect the periodic repetition pattern of the texture and adjust the feature description to obtain the board surface texture identification data. S14. Based on the surface texture identification data of the board, combined with the pre-established defect boundary information database, a unique identification code for the board is generated, and compared with the production record data to obtain the board identification conclusion. S15, based on the board material authentication conclusion, obtain the corresponding subcontracting instruction sequence, and combine it with the production association data to plan dynamic subcontracting path nodes and obtain a subcontracting path node planning scheme. S16. Based on the sub-package path node planning scheme, perform data frequency matching, detect path planning conflict point data and calibrate deviations, update path nodes, and obtain a stable sub-package execution scheme. S17. Based on the stable subcontracting execution scheme, identify production dynamic interference, generate automated control commands, transmit them to the production line equipment, and execute the board subcontracting operation.

[0014] In step S11, it is necessary to acquire the original image data and production-related data of the board surface, and to perform noise reduction processing on the original image data to obtain clear image data, including: The original image data is converted into a first grayscale image, the first grayscale image is segmented according to a preset grayscale threshold range, and the brightness of overly bright areas is suppressed to obtain a second grayscale image. For the second grayscale image, adaptive filtering and denoising are performed on the high gradient intensity information region to obtain the third grayscale image; The edge features and texture details of the third grayscale image are extracted, and the image is sharpened and denoised by edge enhancement to obtain the clear image data.

[0015] It should be noted that the raw image data is the initial image data of the board surface obtained by high-resolution industrial cameras and other image acquisition equipment. It may contain RGB three-channel color information. Its source is real-time acquisition at the designated inspection station on the production line. The acquisition frequency is synchronized with the production cycle and is usually set to 1-2 times per minute to ensure that the inspection needs of each board are covered.

[0016] The production record data in the production-related data comes from a pre-established production record database. This database contains basic information such as production batches, specifications, material types, and production times of the boards. It adopts a distributed architecture design, consisting of a data storage layer, a data processing layer, and a data interface layer. The data storage layer uses a combination of MySQL relational database and MongoDB non-relational database. MySQL is used to store structured data (such as production batch number, board specifications, material type, production time, operator number, etc.), and MongoDB is used to store unstructured data (such as equipment operation logs and original quality inspection reports during the production process). The two are interconnected through data association keys (such as production batch number). Data sources include real-time data uploaded by production line equipment controllers, basic information entered manually, and inspection results fed back by quality inspection equipment. All data must be verified before being stored to ensure data accuracy. The update mechanism combines real-time updates and scheduled updates. Data with high real-time requirements, such as equipment operation data and production progress data, is updated in seconds through message queues, while data with lower real-time requirements, such as statistical analysis data and historical archive data, is updated hourly.

[0017] Real-time production cycle data is collected by devices such as speed sensors and counters deployed on the production line, reflecting the processing efficiency of the sheet metal per unit time; production equipment load data is fed back in real time by the equipment controller, including parameters such as equipment operating power, occupancy rate, and operating temperature, which are used to determine the operating status of the equipment.

[0018] The preset grayscale threshold range is determined based on statistical analysis of a large number of board image samples with different materials and lighting conditions. It is determined according to the grayscale value distribution range of the normal area on the board surface. By statistically analyzing the concentrated range of grayscale values ​​of the samples, the range that covers 95% of normal pixels is taken as the initial threshold range. The specific value can be adjusted according to the actual lighting changes and board material in the production environment. The adjustment step is 5 grayscale levels. For example, the upper limit of the threshold can be appropriately increased in a strong light environment, and the lower limit of the threshold can be appropriately decreased in a weak light environment.

[0019] In this step, the first step is to acquire the raw image data. An industrial camera (with a resolution of at least 1920x1080 pixels) is used to photograph the surface of the board. During acquisition, the vertical distance between the camera and the board surface must be kept constant at 50-80cm, and the lens focal length must be adjusted to a clear focus state to avoid image distortion caused by shooting angle deviations or blurry focus. After acquisition, the raw color image data is converted into a first grayscale image. A weighted average method is used for grayscale conversion. The formula Gray=0.299×R+0.587×G+0.114×B is applied to calculate the grayscale value of each pixel, where R, G, and B are the red, green, and blue channel intensity values ​​of the corresponding pixel in the original color image (range 0-255). The weighting coefficients 0.299, 0.587, and 0.114 are standard coefficients determined based on the characteristics of human visual perception and can be directly applied to grayscale conversion of various color images without additional adjustment. This method assigns different weights to the human eye’s sensitivity to red, green and blue colors to more realistically preserve image brightness information, while reducing data dimensionality and reducing the computational load of subsequent processing. The calculated Gray value is the grayscale value of the pixel after conversion, and its range is also between 0 and 255.

[0020] Furthermore, a brightness distribution analysis is performed on the first grayscale image. By statistically analyzing the image's grayscale histogram, the distribution range of grayscale values ​​is determined. Then, the image is segmented according to a preset grayscale threshold range to determine whether there are overly bright areas, i.e., areas where the grayscale value exceeds the upper limit of the threshold. If overly bright areas exist, linear brightness suppression processing is performed. By reducing the grayscale value of the overly bright areas, the overall brightness distribution of the image tends to be more balanced, avoiding interference from overly bright areas with subsequent edge detection and feature extraction. The suppression intensity can be adaptively adjusted according to the extent to which the grayscale value of the overly bright areas exceeds the threshold; the greater the exceedance, the higher the suppression intensity.

[0021] Furthermore, for the second grayscale image, it is necessary to identify regions with high gradient intensity information. These regions typically correspond to key areas such as edges and defects on the surface of the board. Gradient detection is used to calculate the grayscale change between each pixel and its neighboring pixels. When the change exceeds a preset gradient threshold, it is identified as a region with high gradient intensity information. An adaptive median filtering algorithm is then applied to this region for denoising. The filtering window size can be dynamically adjusted between 3×3 and 7×7. When the proportion of noisy pixels within the window is high, the window size is automatically expanded until the proportion of noisy pixels falls below a set standard or the window reaches its maximum size. This effectively preserves edge details while smoothing noise, avoiding image blurring problems caused by traditional filtering algorithms.

[0022] Finally, edge feature extraction is performed on the third grayscale image. Edge detection utilizes the Canny edge detection algorithm, which accurately locates image edges through multi-step processing. First, the image is smoothed to reduce noise. Then, the gradient direction and intensity of each pixel are calculated. Next, non-maximum suppression is used to eliminate spurious edge responses. Finally, double threshold detection and edge connection are applied to obtain clear texture details. Subsequently, Laplacian sharpening is used for edge enhancement. By highlighting the grayscale contrast of edge regions, texture details and defect features become more apparent, ultimately resulting in clear image data that meets the requirements of subsequent processing.

[0023] For example, suppose a high-resolution industrial camera with a resolution of 1920x1080 pixels is used to acquire raw image data of the surface of a steel plate. This image is an RGB color image containing three channels of information. First, it is converted into a first grayscale image using a weighted average method. This conversion preserves the brightness distribution characteristics of the steel plate surface while removing data redundancy caused by color information. Brightness distribution analysis is performed on the first grayscale image, setting a preset grayscale threshold range of 100-255. It is found that some areas have grayscale values ​​reaching 240 due to uneven illumination, exceeding the upper limit of the threshold, and are identified as overly bright areas. After applying linear brightness suppression processing, the grayscale value of this area is adjusted to a more reasonable range, forming a second grayscale image with more balanced overall brightness. Gradient detection is used to calculate the gradient intensity of the second grayscale image. A certain area has a scratch, and its grayscale value suddenly changes from 120 to 180, a significantly larger change than other areas. This area is identified as a high gradient intensity information area. After applying adaptive median filtering with a 3×3 window, the noise in this area is effectively smoothed, while the scratch edges are preserved, resulting in a third grayscale image. The Canny edge detection algorithm was used to extract edge features from the third grayscale image, detecting fine scratches and texture edges on the steel plate surface. Then, the Laplacian sharpening algorithm was used to enhance the edges, significantly improving the recognition of scratches and textures, and finally obtaining clear image data, which provides a reliable image foundation for subsequent multi-scale texture feature extraction.

[0024] In step S12, based on the clear image data, multi-scale texture features need to be extracted and analyzed. Through texture clustering and integration, key areas of the board surface features are determined, including: The multi-scale texture features include multi-scale texture roughness information and local texture directionality features; The clear image data is subjected to multi-scale filtering decomposition to obtain multi-scale texture roughness information and determine the roughness distribution data. Based on the roughness distribution data, local texture directional features are extracted by directional gradient calculation to determine local directional distribution data; For the local directional distribution data, key points of texture change are filtered out by a preset change threshold range to identify potential key areas; For the potential key areas, multi-scale texture features are integrated through texture clustering to determine the key areas of the board surface features.

[0025] It should be noted that multi-scale texture features reflect the inherent properties of the board surface texture at different scales. Among them, multi-scale texture roughness information is used to describe the roughness of the board surface, and local texture directionality features are used to characterize the direction and variation of the texture. Together, they constitute the core feature dimensions of the board surface texture. The scale division of multi-scale filtering decomposition is based on the common size range of board surface texture, usually divided into three scales: low frequency (1-2 mm), mid frequency (0.5-1 mm), and high frequency (0.1-0.5 mm), corresponding to macro-roughness, meso-roughness, and micro-roughness, respectively. This division is determined based on statistical analysis of a large number of texture samples from different types of boards, and the scale range can be adjusted according to the material characteristics of the board (such as steel plate, wood plate, and plastic plate).

[0026] The preset threshold range is used to filter key points where the texture direction changes significantly. It is set based on the statistical analysis of the direction change amplitude in normal texture areas, and by calculating the standard deviation of the sample texture direction. The change threshold range is set to ±2. Specifically, the statistical analysis of the directional variation amplitude of the normal texture area refers to selecting areas with uniform texture and no defects from a large number of qualified board material samples as normal areas, and calculating the standard deviation of the texture direction angle of all pixels in these areas. For each sample's normal region, the mean and standard deviation of its orientation angle are first calculated. Then, the standard deviations of all samples are averaged to obtain the final standard deviation used to set the threshold. The preset variation threshold range is the mean plus or minus twice the standard deviation, which can cover approximately 95% of normal texture orientation variations. Roughness distribution data is the spatial distribution matrix of roughness values ​​on the image at various scales. Local orientation distribution data is the angular distribution data of texture orientation of each pixel (angle range 0-180°). Potential key regions are areas where texture variations are relatively concentrated. Key feature regions of the board surface are regions determined after clustering and integration that can reflect the core features of the board surface. Their area is usually not less than 10% of the total image area and must include the main texture features and potential defect areas.

[0027] In this step, the sharp image data is first subjected to multi-scale filtering decomposition. Then, Gaussian blurring is applied to the sharp image data, with the standard deviation of the Gaussian kernel... Using values ​​of 1.0, 2.0, and 3.0 respectively, corresponding to low-frequency, mid-frequency, and high-frequency scales, the formula for calculating Gaussian blur is as follows: , where (x,y) are pixel coordinates; then the blurred image is downsampled, retaining pixels in even rows and even columns to obtain decomposed images at each scale; The standard deviation is the Gaussian kernel. For images decomposed at various scales, the Gray-Level Co-occurrence Matrix (GLCM) is used to extract multi-scale texture roughness information. The calculation parameters of the GLCM are set to distance d (e.g., 1, 3, 5 pixels) and direction. Including 0°, 45°, 90°, and 135°, roughness values ​​at each scale are obtained by calculating the texture contrast, correlation, energy, entropy, and other feature parameters of the gray-level co-occurrence matrix, and then using a weighted summation method. The weights are determined based on the contribution of each scale to the roughness: low-frequency scales have a weight of 0.3, mid-frequency scales have a weight of 0.5, and high-frequency scales have a weight of 0.2, with a total weight of 1. This weight allocation is determined based on fitting experimental data and can be adjusted according to actual texture features. The roughness values ​​at each scale are then mapped to the corresponding positions in the original image to form roughness distribution data.

[0028] Furthermore, based on the roughness distribution data, local texture directionality features are extracted using Histogram of Oriented Gradients (HOG). First, the image is divided into 16×16 pixel cells, and the gradient direction and gradient intensity of each pixel in each cell are calculated. The gradient direction is represented by an unsigned angle of 0-180°, which is divided into 9 bins (angle range, each bin corresponding to 20°). The sum of the gradient intensities in each bin is counted to obtain the orientation histogram of the cell. Then, 2×2 cells are grouped into a block, and the orientation histogram within the block is normalized using the L2-Hys normalization method. First, L2 normalization is performed, then the normalized values ​​are limited to within 0.2, and finally L2 normalization is performed again to obtain the local texture directionality features, thereby determining the local orientation distribution data.

[0029] Furthermore, for the local directional distribution data, the directional difference value between each pixel and its neighboring pixels (3×3 window) is calculated. If the directional difference value exceeds a preset change threshold, the pixel is marked as a texture change key point. Connectivity analysis is performed on all texture change key points. Using the eight-neighborhood connectivity algorithm, key points with a distance of less than 5 pixels are grouped into the same connected region, and each connected region is a potential key region.

[0030] Finally, for potential key regions, the K-means clustering algorithm is used to integrate multi-scale texture features. The K value for clustering is determined based on the number of potential key regions, typically K=3-5. Euclidean distance is used as the similarity metric, and the calculation formula is as follows: ,in The weights for each feature dimension are: roughness feature weight 0.6, directionality feature weight 0.4. and These are the feature values ​​of the two samples respectively; the cluster centers are converged through iterative calculation, and each cluster after clustering is taken as a key feature region of the board surface.

[0031] It is worth noting that the number of scales in the multi-scale filtering decomposition can be adjusted according to the complexity of the surface texture of the board. The more complex the texture, the more scales can be added (e.g., 4-5 scales) to capture texture features more comprehensively. The initial cluster centers of the K-means clustering algorithm are selected using the K-means++ algorithm to avoid the influence of initial values ​​on the clustering results. The clustering convergence condition is set to 100 iterations or a change in the cluster centers of less than 0.001. In addition, during feature extraction, the feature data needs to be normalized to map all feature values ​​to the [0,1] interval. The value is determined based on the contribution of features to texture recognition: through a large number of experiments on board samples of different materials and texture types, the discriminative power and stability of each feature (such as roughness and orientation) in distinguishing different board textures are statistically analyzed. Features with higher discriminative power and stronger stability are assigned higher weights. The larger the value, the better. For example, if roughness features are better at distinguishing texture differences than directional features, then set the weight of roughness features to 0.6 and the weight of directional features to 0.4, with a total weight of 1. This weight can be finely adjusted based on the accuracy of texture recognition in practical applications.

[0032] It should be noted that since the feature parameters such as contrast, correlation, energy, and entropy calculated by the gray-level co-occurrence matrix have different dimensions and numerical ranges, they must be normalized before weighted summation to eliminate the influence of dimensions and make each feature comparable in magnitude. For example, min-max normalization can be used to map the minimum value of each feature parameter to 0 and the maximum value to 1 across all samples or image regions, and the other values ​​of the parameter are mapped to the [0,1] interval in a linear proportion. After normalization, the roughness value will be within the [0,1] interval and has a clear relative meaning. The larger the value, the coarser the texture at that scale. Assuming that the contrast is 0.8, the correlation is 0.7, the energy is 0.6, and the entropy is 0.5 after normalization, the roughness value is 0.3×0.8+0.5×0.7+0.2×0.6+0.1×0.5=0.76.

[0033] Furthermore, local texture directionality features are extracted, and the image is divided into 16×16 pixel cells. The gradient directions of pixels within a certain cell are mainly concentrated in the 0°-20° and 160°-180° ranges, with corresponding gradient intensity sums of 80 and 75 within their respective bins. After normalization, the orientation histogram of that cell is obtained. The neighborhood orientation difference value of each pixel is calculated, with a preset change threshold range of ±30°. =15°), a pixel with its own orientation of 90° and neighboring pixels mostly with orientations of 60°, has an orientation difference of 30°, which just reaches the upper limit of the threshold and is marked as a key point of texture change. Using the eight-neighbor connectivity algorithm, these key points are clustered into 3 connected regions, i.e., 3 potential key regions. The K-means clustering algorithm (K=3) is used to integrate multi-scale texture features and calculate the feature similarity of each potential key region. Regions with roughness values ​​of 0.8-1.2 and orientations concentrated between 45°-90° are clustered into one class as key regions of plate surface features. This region contains the main texture changes on the steel plate surface and a minor scratch defect.

[0034] In step S13, the key feature regions of the board surface need to be matched and optimized with a pre-established texture comparison template library for similarity, the periodic repetition pattern of the texture is detected and the feature description is adjusted to obtain the board surface texture identification data, including: The feature distribution data of key regions on the surface of the board are extracted and matched with a pre-established texture comparison template library to determine the preliminary similarity distribution data. If the preliminary similarity distribution data is lower than the preset similarity threshold, then the texture contrast difference of the key feature areas on the surface of the board is weighted and calculated to obtain the optimized similarity distribution data. For the optimized similarity distribution data, detect the periodic repetition pattern of the texture and determine the distribution of patterns that can be optimized. Based on the optimizable pattern distribution, the feature description is adjusted to obtain the surface texture identification data of the board.

[0035] It should be noted that the feature distribution data of key areas on the surface of the board material refers to the spatial distribution and statistical characteristics of multi-scale texture features (roughness, directionality) within that area, including parameters such as feature mean, variance, extreme values, and distribution density. The pre-established texture comparison template library is a set of standard texture templates used for comparison. Its architecture adopts a "classification storage + index retrieval" design, divided into a template classification layer, a template storage layer, and an index layer. The template classification layer classifies the materials (e.g., steel, aluminum, wood), specifications, and processing technology into three levels. The template storage layer uses a distributed file system to store the feature data and image data of the templates, supporting fast retrieval. The index layer uses a B+ tree index structure, using the core feature parameters of the template (e.g., average roughness, main texture direction) as the index key to improve retrieval efficiency. The data sources for the texture comparison template library include two parts: first, texture data from standard board samples, acquired through high-precision acquisition equipment in a laboratory environment to ensure the standardization of the templates; second, texture data from qualified boards verified in actual production, which are manually screened and quality checked before being added to the library, enriching the diversity of the template library. The update mechanism employs a combination of periodic and incremental updates. Periodic updates occur quarterly, validating templates in the template library, deleting outdated or invalid templates, and adding standard templates for new types of boards. Incremental updates are real-time; when qualified boards with new textures appear in production, their texture features are automatically extracted and added to the library, while the index structure is updated simultaneously. The template library includes a unique template identifier, basic board information (material, specifications, process), texture feature data (multi-scale roughness, directional distribution), template image data, creation time, and update time.

[0036] The preset similarity threshold is a critical value for judging the degree of feature matching. It is set based on a large amount of experimental data comparing qualified boards with template libraries. The similarity is calculated by taking the average similarity of successfully matched cases. and standard deviation Set the similarity threshold to The value typically ranges from 0.7 to 0.8, and can be adjusted according to the matching accuracy requirements of production. For high accuracy requirements, the threshold can be increased to 0.85, while for lower accuracy requirements, it can be decreased to 0.65. Texture contrast difference refers to the difference in texture contrast between the key feature areas of the board surface and the most similar template in the template library. Texture contrast data is calculated from the gray-level co-occurrence matrix of the key feature areas of the board surface. Periodic repetition patterns refer to the regular repetition characteristics of textures in space, including parameters such as repetition period, repetition direction, and repetition intensity. Optimizable pattern distribution refers to periodic patterns that can improve matching accuracy by adjusting feature descriptions. Board surface texture identification data is a standardized description of the board surface texture features, including core feature parameters, periodic pattern parameters, and optimization adjustment records, used to uniquely characterize the surface texture characteristics of the board.

[0037] In this step, feature distribution data of key areas on the board surface is first extracted. A combination of feature statistics and spatial sampling is used to uniformly sample the key areas (sampling interval of 5 pixels). Multi-scale roughness values ​​and texture orientation angles are extracted for each sampling point. Statistical parameters such as the feature mean, variance, main texture direction, and texture contrast of the region are calculated to form a feature distribution data vector. Subsequently, similarity matching is performed with a pre-established texture comparison template library. An index layer quickly retrieves the template most similar to the feature distribution data to be matched. The cosine similarity between the data to be matched and each template is calculated, with the similarity value ranging from [0,1]. A value closer to 1 indicates a higher matching degree. The calculated similarity values ​​are mapped to the corresponding positions of the key areas to form preliminary similarity distribution data.

[0038] Furthermore, it is determined whether the preliminary similarity distribution data is lower than a preset similarity threshold. If it is lower than the threshold, a weighted calculation is performed on the texture contrast difference. The purpose is to fine-tune the similarity based on the texture contrast difference when the initial matching is poor. The adjustment range should be strictly controlled to avoid distortion of the similarity value. For example, the weighted calculation formula is as follows: ,in For the initial similarity, For contrast difference, This is a scaling factor for the contrast difference (e.g., taking the standard deviation of contrast in the template library). The hyperbolic tangent function maps the input to the interval (-1, 1). An adjustment factor less than 0.2 (e.g., 0.1) is used to ensure the optimized similarity. exist The similarity distribution data is optimized by taking small fluctuations in the vicinity, which do not exceed the reasonable range of [0,1].

[0039] Furthermore, for the optimized similarity distribution data, the autocorrelation function method is used to detect the periodic repetition pattern of the texture. The formula for calculating the autocorrelation function is as follows: Where I(x,y) is the pixel grayscale value. =( , Let E be the displacement vector and E be the expectation operator. By calculating the autocorrelation coefficient under different displacement vectors, when the autocorrelation coefficient reaches a peak, the corresponding displacement vector is the repetition period of the texture. The peak intensity reflects the repetition intensity. Based on this, the periodic repetition pattern (repetition period, direction, intensity) of the texture is determined, and the distribution of optimizable patterns (i.e., regions with clear repetition patterns and high intensity) is judged.

[0040] Finally, for the optimizable pattern distribution, the feature description is adjusted, periodic pattern parameters (repetition period value, repetition direction angle, repetition intensity) are added, and the parts in the original feature description that do not match the periodic pattern are corrected. A standardized feature description format is adopted, and the key feature parameters are organized in the order of "material-roughness range-main texture direction-periodic parameter" to form the surface texture identification data of the board. This data is stored in JSON format for easy subsequent processing and transmission.

[0041] It is worth noting that when detecting periodic repetition patterns in textures, if the autocorrelation coefficient does not have a significant peak, it indicates that the texture does not have significant periodicity. Therefore, there is no need to adjust the feature description; the feature description corresponding to the optimized similarity distribution data can be directly used as the texture identification data. Furthermore, the weights in the weighted calculation can be dynamically adjusted based on feedback from the quality of the sheet metal in actual production. If it is found that the weight allocation for a certain type of region is unreasonable, leading to a large matching error, the weight values ​​can be adjusted through the backend management system.

[0042] In step S14, based on the surface texture identification data of the board material and combined with a pre-established defect boundary information database, a unique identification code for the board material needs to be generated. This code is then compared with the production record data to obtain the board material identification conclusion, including: Based on the surface texture identification data of the board material, combined with the pre-established defect boundary information database, the boundary information features of the board material are extracted to form a surface feature dataset; Based on the surface feature dataset, a unique code for the board is generated by encoding and integrating the surface texture identification data of the board with the boundary information features of the board. The unique identification code of the board is matched with the production record data, and the matching degree is calculated. If the matching degree meets the preset matching threshold, the identity is confirmed to be consistent, and the board identification conclusion is obtained.

[0043] It should be noted that the pre-established defect boundary information database is a database storing boundary feature data of common defects in various types of sheet materials (such as scratches, dents, and cracks). Its architecture adopts a two-level storage structure of "defect type - boundary feature." Defect types are divided into three main categories: surface defects, edge defects, and internal defects, each further subdivided into specific defect types. Boundary feature data stores the boundary shape parameters and boundary grayscale characteristics of the defects. Data sources include boundary feature data from laboratory-simulated defect samples and defect boundary data detected in actual production. All data is labeled and verified to ensure accuracy. The update mechanism involves a comprehensive update every six months, supplementing boundary feature data for new defects while deleting invalid data. When a new defect is discovered in production, a temporary update process is initiated to promptly add its boundary feature data to the database. The defect boundary information database includes the unique defect identifier, defect type, boundary shape parameters, boundary grayscale characteristics, corresponding sheet material, and creation time.

[0044] The board boundary information features are the boundary feature parameters of surface defects, including the length, width, area, grayscale difference, and gradient distribution of the defect boundary. The surface feature dataset is a collection of integrated board surface texture identification data and boundary information features, including core texture parameters, periodic pattern parameters, and defect boundary parameters. The board unique identification code is used to uniquely identify a single board piece, adopting a "fixed bit + variable bit" structure. The fixed bits are 10 bits (including 4 bits for enterprise identification, 2 bits for production year, 2 bits for production line number, and 2 bits for material code), and the variable bits are 22 bits (including 8 bits for texture feature hash value, 8 bits for defect boundary feature hash value, and 6 bits for random check code), for a total length of 32 bits, ensuring the uniqueness and security of the code.

[0045] The preset matching threshold is a critical value for judging the degree of matching between the identity code and the production record data. It is established based on matching experimental data of a large number of qualified boards and is calculated by taking the average matching degree of successful matching cases. and standard deviation Set the preset matching threshold as The value is typically set to 0.9, but can be adjusted according to production management requirements. It can be increased to 0.95 for strict management and decreased to 0.85 for lenient management. The board material identity verification conclusion includes three situations: identity consistent, identity inconsistent, and identity questionable. These correspond to matching degrees meeting the threshold, not meeting the threshold, and close to the threshold (within ±0.05), respectively.

[0046] In this step, firstly, based on the surface texture identification data of the board material and combined with the defect boundary information database, the boundary information features of the board material are extracted. The Canny edge detection algorithm is used to extract the defect boundaries of key feature areas on the board material surface, obtaining the pixel coordinate sequence of the boundary; then, the shape parameters of the boundary are calculated, and the length (sum of distances between adjacent pixels), width (average thickness of the boundary), area (number of pixels enclosed by the boundary), and curvature (degree of curvature of the boundary curve, calculated after fitting the curve using the least squares method) of the boundary are calculated through the pixel coordinate sequence; at the same time, the gray-level features of the boundary are calculated, and the mean gray-level difference between boundary pixels and background pixels and the gradient intensity distribution of boundary pixels are statistically analyzed to form boundary information features; the boundary information features are integrated with the surface texture identification data of the board material to form a surface feature dataset.

[0047] Furthermore, a unique identification code for the board material is generated using a step-by-step encoding method. First, the core texture parameters (average roughness, main texture direction, repetition period) in the surface feature dataset are hashed using the SHA-256 hash algorithm, and the first 8 bits of the hash value are used as the texture feature hash value. Second, the defect boundary features (length, width, area, average grayscale difference) are hashed using the same SHA-256 algorithm, and the first 8 bits of the hash value are used as the defect boundary feature hash value. Third, a 6-bit random check code is generated using a pseudo-random number generation algorithm, with the seed being the hash value of the current timestamp and the board surface feature data to ensure randomness. Fourth, the fixed bits (company identifier, production year, production line number, material code) and the variable bits (texture feature hash value, defect boundary feature hash value, random check code) are concatenated sequentially to form a 32-bit unique identification code for the board material.

[0048] Furthermore, the unique identification code of the board material is matched with the production record data. The matching process consists of two steps: First, an exact match is performed to search the production record database for a completely identical identification code. If a match is found, the matching degree is directly determined to be 1.0. If no exact match is found, a fuzzy match is performed to extract key information such as material, specifications, and production batch from the production record data. This information is then correlated with the fixed-bit information and the feature hash value in the variable bits of the identification code. A weighted sum is used to calculate the matching degree, with the following weights: material code matching weight 0.3, specification information matching weight 0.2, production batch matching weight 0.2, texture feature hash value matching weight 0.2, and defect boundary feature hash value matching weight 0.1, with a total weight of 1. The weight values ​​are determined through experimental statistical analysis. The fitting process involves collecting thousands of matching samples from different board materials, calculating the contribution of a single dimension's correct match to overall identity verification (i.e., the probability of overall verification being correct when that dimension is correctly matched), and then normalizing the contribution of each dimension (dividing the contribution of each dimension by the sum of the contributions of all dimensions). The final fitting yields the aforementioned weight allocation. The weights of each dimension can be fine-tuned based on feedback from misjudgments in actual production. The matching degree calculation formula is M=w_match1×m1+w_match2×m2+w_match3×m3+w_match4×m4+w_match5×m5, where mi is the matching score for each dimension (1 point for a perfect match, 0.5 points for a partial match, and 0 points for no match), and w_matchi is the weight value for each dimension.

[0049] Finally, it is determined whether the matching degree meets the preset matching threshold. If the matching degree is ≥ the preset matching threshold, the identity is confirmed to be consistent; if the matching degree is < the preset matching threshold, the identity is confirmed to be inconsistent; if the matching degree is within ±0.05 of the preset matching threshold, the identity is determined to be questionable and manual review is required to obtain the final board material identity verification conclusion.

[0050] For example, based on the surface texture identification data of the sheet metal (material Q235 steel plate, main texture direction 60°, repetition period 5mm), combined with the defect boundary information database, a scratch defect was extracted from the sheet metal, with a boundary length of 10mm, a width of 0.5mm, and an area of ​​8mm². 2 The average grayscale difference between the boundary and the background is 30, forming a surface feature dataset. SHA-256 hash calculation is performed on the core texture parameters (average roughness 0.9, main texture direction 60°, repetition period 5mm), yielding the first 8 bits of the hash value as "3F7A2D9C"; the defect boundary features (length 10mm, width 0.5mm, area 8mm²) are also analyzed. 2 The average grayscale difference (30) is used to perform hash calculation, and the first 8 bits of the hash value are "8B4E6C1A". The current timestamp "1715368800000" and the hash value of the surface feature data are used as seeds to generate a random verification code "729513". The fixed bits are the enterprise identifier "ABC1", the production year "24", the production line number "03" and the material code "ST01". After splicing, the unique identification code of the board is obtained as "ABC12403ST013F7A2D9C8B4E6C1A729513". The code is matched against the production record database. An exact match was not found, so a fuzzy match was performed. The material code "ST01" matches the Q235 steel plate material code in the production record (m1=1), the specification information (1500mm×1000mm×5mm) matches the record (m2=1), the production batch "B2024051001" matches the record (m3=1), the texture feature hash value matches (m4=1), and the defect boundary feature hash value matches (m5=1). The matching degree M is calculated as M=0.3×1+0.2×1+0.2×1+0.2×1+0.1×1=1.0. The preset matching threshold is 0.9. The matching degree meets the threshold, confirming the identity is consistent, and the plate identity verification conclusion is obtained.

[0051] In step S15, it is necessary to obtain the corresponding subcontracting instruction sequence based on the board material authentication conclusion, and combine it with the production association data to plan dynamic subcontracting path nodes, thereby obtaining a subcontracting path node planning scheme, including: Based on the board material authentication conclusion, the corresponding subcontracting instruction sequence is obtained from the pre-established instruction database, and the detailed content of the target subcontracting instruction is parsed to obtain the subcontracting instruction. Based on the detailed content of the target subcontracting instruction, the range of instruction adjustment is determined by data integration and matching with the real-time production cycle data; Based on the instruction adjustment range, analyze the production equipment load data to determine whether it meets the preset load threshold range; If the conditions are met, the node allocation is optimized by adjusting the range based on the instructions, and dynamic sub-package path nodes are planned to obtain a sub-package path node planning scheme.

[0052] It should be noted that the pre-established instruction database stores subcontracting instruction sequences. Its architecture adopts a three-level mapping structure: "authentication result - instruction type - instruction content." The first-level node represents the authentication conclusion (identity matches, identity does not match, identity is questionable); the second-level node represents the instruction type (normal subcontracting instruction, exception handling instruction, review instruction); and the third-level node represents the specific instruction sequence. The data source is the subcontracting rules formulated by the enterprise's production management department, including subcontracting instructions for normal production processes, contingency plans for exceptions, and operational instructions for review processes. All instructions are entered into the database after process verification. The update mechanism is a quarterly update, revising instructions based on production process optimization and subcontracting rule adjustments; when new authentication scenarios emerge, immediate updates are performed to supplement the corresponding instruction sequences. The instruction database contains the unique identifier of the instruction sequence, the corresponding authentication conclusion, the instruction type, detailed instruction content (subcontracting target area, priority, operation steps, execution time limit), creation time, and update time.

[0053] The detailed content of the target subcontracting instruction refers to the specific subcontracting operation instruction corresponding to the current board material authentication conclusion, including information such as the target warehouse, storage area, stacking method, priority level, and executing equipment. Real-time production cycle data is the number of boards processed and transferred on the production line per unit time, usually measured in "pieces / minute," reflecting the busyness of the production process. The instruction adjustment range is the interval within which the target subcontracting instruction is adjusted based on the real-time production cycle data, including the execution time adjustment range and priority adjustment range. Production equipment load data is the operating status data of subcontracting-related equipment (such as transfer robots, conveyor belts, and stacker cranes), including parameters such as equipment operating power, occupancy rate, and remaining capacity.

[0054] The preset load threshold range is the critical range for determining whether the equipment can undertake the current subcontracted task. It is set based on the equipment's rated load parameters and actual operating experience data. The preset load threshold range is calculated to be 70%-90% of the equipment's rated load. It can be adjusted according to the aging degree and maintenance status of the equipment. For aging equipment, the lower limit of the threshold can be appropriately reduced to 60%.

[0055] In this step, the corresponding subcontracting instruction sequence is first retrieved from the instruction database based on the board material authentication conclusion. Using an index retrieval method, with the authentication conclusion as the search key, the corresponding instruction type is quickly located, and the instruction sequence is obtained. The instruction sequence is then parsed using an XML parser to extract the structured data of the instructions, revealing detailed information about the target subcontracting instruction, including the subcontracting target area, priority level, execution device, and execution time limit.

[0056] Furthermore, based on the detailed content of the target subcontracting instruction, data integration and matching are performed with real-time production cycle data to analyze the matching relationship between real-time production cycle and instruction execution time limit. If the real-time production cycle is 50 pieces / minute, the current number of boards to be subcontracted is 10, and the estimated processing time is 12 minutes, which is less than the instruction execution time limit of 30 minutes, so no adjustment to the execution time is needed; if the real-time production cycle is 30 pieces / minute, the number of boards to be subcontracted is 10, and the estimated processing time is 20 minutes, which is still within the execution time limit, the instruction adjustment range is determined to be "priority maintained at level one, execution time can be adjusted within 15-30 minutes".

[0057] Furthermore, based on the instruction adjustment range, the load data of the production equipment is analyzed. The current load rate of the transfer robot R05 is 75%, and the current load rate of the stacker crane S03 is 70%. The preset load threshold range is 70%-90%. It is determined that the load data of both devices conforms to the preset load threshold range and is capable of executing subcontracting tasks. If the equipment load data exceeds the preset threshold range, the instruction adjustment range needs to be adjusted, such as lowering the subcontracting priority, extending the execution time, or replacing the device with one that has a lower load.

[0058] Finally, by combining instruction adjustment range optimization with node allocation, in one possible implementation, Dijkstra's shortest path algorithm can be used to plan dynamic subcontracting path nodes. First, a path node map of the production workshop is constructed, including production workstations (starting nodes), transfer nodes (such as robot docking points and conveyor belt connection points), and the target warehouse area (ending nodes). The weight of each node is its equipment processing time and congestion coefficient. Based on Dijkstra's algorithm, the shortest path from the starting node to the ending node is calculated. According to the equipment load of each node, priority tasks are assigned to devices with lower loads to avoid equipment overload. The execution order and operation instructions of each node are determined, forming a complete subcontracting path node planning scheme.

[0059] It's worth noting that an XML parser is used during instruction parsing to ensure structured reading of instruction data and avoid parsing errors. During path planning, if multiple shortest paths exist, the path with the lowest total load on node devices is selected to ensure efficient execution. When real-time production cycle time fluctuates significantly (e.g., a sudden 50% drop), the instruction adjustment range needs to be recalculated, potentially requiring extended execution time or lower priority. If production equipment load data suddenly exceeds a threshold, backup equipment must be activated or the subcontracting order adjusted to ensure smooth execution of subcontracted tasks. Furthermore, the subcontracting path node planning scheme must include contingency plans, such as automatic switching to a backup node in case of a node failure, ensuring uninterrupted subcontracting processes.

[0060] For example, the board authentication conclusion is "identity consistent". The corresponding normal subcontracting instruction sequence is retrieved from the instruction database, and the detailed content of the target subcontracting instruction is parsed: the subcontracting target area is shelf 5 in warehouse B area, priority level 2, the executing equipment is transfer robot R08 and stacker crane S07, and the execution time limit is 25 minutes. The real-time production cycle data is 40 boards / minute, the current number of boards to be subcontracted is 8, and the estimated processing time is 12 minutes. The instruction adjustment range is determined to be "execution time 12-25 minutes, priority level 2". Production equipment load data is obtained: the current load rate of transfer robot R08 is 72%, and the current load rate of stacker crane S07 is 78%. The preset load threshold range is 70%-90%, which meets the threshold requirements. The Dijkstra algorithm is used to plan dynamic subcontracting path nodes, constructing a path node map. The starting node is production station P05, intermediate nodes include M07 (R08 stopping point), C05 (conveyor belt connection point), and S07 (stacker crane working point), and the ending node is shelf T15 in warehouse B area. Calculate the weights of each node: P05→M07 weight is 2 (processing time 2 minutes), M07→C05 weight is 3 (processing time 3 minutes), C05→S07 weight is 4 (processing time 4 minutes), S07→T15 weight is 5 (processing time 5 minutes), and the total weight of the shortest path is 14. The corresponding path node sequence is P05→M07→C05→S07→T15. Arrange the transfer task of R08 in minutes 1-3, the conveyor belt transport in minutes 3-7, and the stacking operation of stacker crane S07 in minutes 7-12. The estimated total execution time is 12 minutes. Within the instruction adjustment range, a subcontracting path node planning scheme is formed.

[0061] In step S16, based on the sub-package path node planning scheme, data frequency matching is performed, path planning conflict point data is detected and deviations are calibrated, path nodes are updated, and a stable sub-package execution scheme is obtained, including: Based on the sub-package path node planning scheme, path planning data and data acquisition frequency data are obtained, and the frequency matching deviation range is determined after integration. Based on the frequency matching deviation range, conflict point data of path planning is obtained through conflict detection and deviation analysis and comparison to obtain deviation calibration parameters. Calculate the fusion deviation value of the deviation calibration parameter. If the fusion deviation value exceeds the preset deviation threshold, adjust the sub-package path and update the path node to obtain a stable sub-package execution scheme.

[0062] It should be noted that path planning data is the core data in the subcontracted path node planning scheme, including the path node sequence, the execution time of each node, operation instructions, equipment allocation information, etc. Data collection frequency refers to the frequency at which path planning-related data (such as node equipment status and path congestion) is collected, usually in units of "times / minute". This frequency is set according to the real-time requirements of the data; the node equipment status data collection frequency is 2 times / minute, and the path congestion data collection frequency is 1 time / minute. The frequency matching deviation range is the range of difference between the path planning data update frequency and the data collection frequency, used to determine whether the two are synchronized. Conflict detection refers to detecting whether there are node conflicts (such as the same node being occupied by multiple subcontracted tasks simultaneously), equipment conflicts (such as the same device being assigned by multiple tasks), or time conflicts (such as overlapping task execution times) in the path planning. Deviation analysis and comparison refers to analyzing the deviation between conflict point data and normal path planning data to determine the type (node ​​deviation, equipment deviation, time deviation) and degree of deviation. Deviation calibration parameters are adjustment parameters used to calibrate these deviations, including node adjustment parameters (such as replacing intermediate nodes), equipment adjustment parameters (such as replacing execution equipment), and time adjustment parameters (such as delaying execution time).

[0063] The fusion deviation value is a weighted fusion of the deviation levels corresponding to various deviation calibration parameters. The weight allocation is based on the degree of influence of the deviation on the subcontracting process. The weight values ​​are derived from the influence of each deviation type on the stability of subcontracting. Through numerous subcontracting scenario experiments, the probability of subcontracting failure or efficiency reduction caused by deviations such as node conflicts, equipment conflicts, and time conflicts is statistically analyzed. The higher the probability of influence, the greater the weight. For example, node conflicts have the greatest impact on path smoothness, with a weight of 0.4; equipment conflicts and time conflicts have the next greatest impact, with a weight of 0.3, and the total weight is 1. The preset deviation threshold is the critical value for determining whether the fusion deviation value needs to be adjusted for the path. It is set based on the deviation data statistics of a large number of path planning cases. By calculating the mean μ and standard deviation σ of the fusion deviation value of normal cases, the preset deviation threshold is set to μ+2σ, usually 0.3. It can be adjusted according to the stability requirements of the subcontracting process. When the stability requirements are high, it can be reduced to 0.2, and when the requirements are low, it can be increased to 0.4.

[0064] In this step, firstly, based on the subcontracted path node planning scheme, path planning data is extracted, including the path node sequence, execution time of each node, equipment allocation information, etc.; simultaneously, data collection frequency data is obtained to clarify the collection frequency of various types of data. The update frequency of path planning data is integrated and compared with the data collection frequency. The update frequency of path planning data is 1 time / minute (synchronized with the production cycle), the collection frequency of equipment status data is 2 times / minute, and the collection frequency of path congestion data is 1 time / minute. The frequency matching deviation is calculated. The frequency deviation of equipment status data is |2-1| / 1=100%, and the frequency deviation of path congestion data is |1-1| / 1=0%. The overall frequency matching deviation range is determined to be 0%-100%.

[0065] Furthermore, based on the frequency matching deviation range, a conflict detection algorithm is employed to detect path planning conflict points. Conflict detection is performed from three dimensions: time, space, and device. The time dimension detects whether the execution times of each task overlap; the space dimension detects whether each path node is occupied simultaneously; and the device dimension detects whether each device is repeatedly assigned. By traversing all currently executing and pending sub-tasks, the node, device, and time information of the current path planning scheme are compared to determine path planning conflict points, including conflicting nodes, conflicting devices, conflicting time intervals, and overlapping times. Deviation analysis and comparison are performed on the conflict point data to calculate node conflict deviation, device conflict deviation, and time conflict deviation, thereby obtaining deviation calibration parameters.

[0066] Furthermore, the fusion deviation value of the deviation calibration parameters is calculated using a weighted summation. It is then determined whether the fusion deviation value exceeds a preset deviation threshold of 0.3. If it does, the sub-package path needs to be adjusted and the path nodes updated to form a stable sub-package execution plan, including the adjusted path node sequence, device allocation, execution time, and operation instructions.

[0067] For example, based on the subcontracted path node planning scheme, the obtained path planning data is the path node sequence P05→M07→C05→S07→T15, with an execution time of 1-12 minutes, and the devices assigned are R08 and S07; the data collection frequency is 2 times / minute for device status and 1 time / minute for path congestion. After integration, the frequency matching deviation range is determined to be 0%-100%. A conflict detection algorithm identifies conflict points: node M07 is occupied by another task within 3-5 minutes, and device R08 is repeatedly assigned within 1-3 minutes, with a 2-minute time overlap. Deviation analysis and comparison yields a node conflict deviation of 0.5, a device conflict deviation of 0.4, and a time conflict deviation of 0.3. The deviation calibration parameters are replacing node M07 with M08, device R08 with R09, and delaying execution by 2 minutes. The calculated fusion deviation value is 0.4×0.5+0.3×0.4+0.3×0.3=0.41, exceeding the preset deviation threshold of 0.3. The path node sequence was adjusted to P05→M08→C05→S07→T15, with an execution time of 3-15 minutes. The devices were assigned as R09 and S07. After another check, no conflict points were found, and the fusion deviation value was reduced to 0.15, which met the requirements, resulting in a stable sub-package execution scheme.

[0068] In step S17, based on the stable subcontracting execution scheme, it is necessary to identify production dynamic interference, generate automated control commands, transmit them to the production line equipment, and execute the sheet metal subcontracting operation, including: Based on the stable subcontracting execution scheme, real-time monitoring data of the production environment is obtained to identify the type and intensity of dynamic interference in production. Based on the type and intensity of the production dynamic interference, corresponding dynamic signal data is generated and converted into automated control commands that conform to the preset command threshold range; The automated control commands are transmitted to the production line equipment to drive the subcontracting operation process and execute the sheet metal subcontracting operation.

[0069] It should be noted that real-time monitoring data of the production environment is collected through various sensors deployed in the production workshop, including environmental data, equipment operation data, and logistics data, at a collection frequency of once per second to ensure real-time capture of production dynamics. Production dynamic interference refers to various factors affecting the normal execution of subcontracting operations. These are categorized into three main types: environmental interference, equipment interference, and logistics interference, each further subdivided into specific interference types. Interference intensity is classified into three levels: low, medium, and high, based on the degree of impact on subcontracting operations: low-level interference has no significant impact on subcontracting operations and requires no adjustment of control commands; medium-level interference may lead to a decrease in subcontracting efficiency and requires fine-tuning of control commands; high-level interference may lead to subcontracting failure or equipment malfunction, requiring significant adjustments to control commands or suspension of operations. Dynamic signal data consists of signal parameters generated based on the type and intensity of interference, used for adjusting control equipment, including voltage signals, current signals, and frequency signals.

[0070] The preset command threshold range is the effective value range of the automation control command. It is established based on the rated parameters and safe operating range of the production line equipment. For example, the threshold range for equipment operating speed is 50%-110% of the rated speed, and the threshold range for voltage signals is 220V±10%. These can be adjusted according to the equipment's performance parameters. Automation control commands convert dynamic signal data into operation instructions that the equipment can recognize, including speed adjustment commands, power adjustment commands, direction adjustment commands, and pause commands. Production line equipment, including transfer robots, conveyor belts, stacker cranes, and warehouse management systems, can receive and execute automation control commands.

[0071] In this step, firstly, based on a stable sub-package execution scheme, real-time monitoring data of the production environment is acquired through a sensor network. This data is then subjected to interference identification. In one possible implementation, a support vector machine (SVM) can be used for interference identification. This SVM is trained on monitoring data samples with different interference types and intensities, enabling automatic identification of interference types and intensities. The real-time monitoring data is then input into the SVM, which outputs the interference type and intensity.

[0072] Furthermore, based on the type and intensity of interference, corresponding dynamic signal data is generated, establishing a mapping relationship between interference type, intensity, and signal parameters. For abnormally high temperatures, a voltage signal is generated to reduce the cooling system power; for abnormal robot vibration, a frequency signal is generated to reduce the operating speed; and for low conveyor belt speed, a current signal is generated to increase the conveyor belt speed. The dynamic signal data is converted into automated control commands, encapsulated into command formats according to equipment communication protocols (such as Modbus), and these commands are checked to ensure they conform to preset command threshold ranges. The automated control commands are then transmitted to the corresponding production line equipment via industrial Ethernet, using the TCP / IP protocol for data transmission. After receiving the commands, the equipment returns a reception confirmation signal. Finally, the production line equipment executes the automated control commands to complete the sheet metal repackaging operation.

[0073] For example, based on a stable sub-package execution scheme, real-time monitoring data of the production environment is obtained: temperature 35℃, humidity 65%, robot R09 vibration frequency 5Hz, and conveyor belt C05 speed 0.8m / s. The interference type and intensity are identified using a support vector machine as abnormal high temperature (medium), excessive humidity (low), abnormal robot vibration (medium), and low conveyor belt speed (medium). A fuzzy control algorithm is used to generate dynamic signal data: cooling system voltage signal 0.8V, robot frequency signal 50Hz, and conveyor belt current signal 5A. This is converted into automation control commands: "COOL_POWER=80%", "ROBOT_SPEED=85%", and "CONVEYOR_SPEED=1.1m / s", all of which conform to the preset command threshold range. The commands are transmitted to the equipment via industrial Ethernet. After receiving and confirming the commands, the equipment executes them: the cooling system adjusts its power, and the robot and conveyor belt adjust their speeds, restoring all parameters to normal. Following the path node sequence P05→M08→C05→S07→T15, the transfer and stacking of the sheet metal were completed in 3-15 minutes, which meets the requirements of the stable subcontracting execution plan, and the sheet metal subcontracting operation was successfully completed.

[0074] In summary, this invention discloses a vision-based automated control method for sheet metal subcontracting. The method includes acquiring raw image data and production-related data and performing denoising processing; extracting multi-scale texture features to determine key regions; comparing and optimizing texture identification data against a texture comparison template library; generating a unique identification code and comparing it with production records; planning dynamic subcontracting path nodes; calibrating path deviations to obtain a stable execution scheme; and identifying dynamic interference to generate control commands and execute subcontracting operations. This invention improves the accuracy of texture feature extraction through image preprocessing, ensures the reliability of identity verification based on template library comparison and identification codes, and dynamically plans paths and calibrates deviations using production data to adapt to dynamic production interference. This achieves fully automated and precise control of the sheet metal subcontracting process, effectively solving the problems of inaccurate and inefficient subcontracting in existing technologies.

[0075] Reference Figure 2 The second embodiment of the present invention provides an automated control system for board subcontracting based on vision detection, comprising: The data acquisition module acquires the original image data and production-related data of the board surface. The production-related data includes production record data, real-time production cycle data, and production equipment load data. The original image data is denoised to obtain clear image data. The feature extraction module extracts and analyzes multi-scale texture features based on the clear image data, and determines key areas of the board surface features through texture clustering and integration. The identifier generation module performs similarity matching and optimization between the key feature areas of the board surface and a pre-established texture comparison template library, detects the periodic repetition pattern of the texture and adjusts the feature description to obtain the texture identifier data of the board surface. The identity verification module generates a unique code for the board based on the surface texture identification data of the board and a pre-established defect boundary information database. The code is then compared with the production record data to obtain the board identity verification conclusion. The path planning module obtains the corresponding subcontracting instruction sequence based on the board material authentication conclusion, and plans dynamic subcontracting path nodes in conjunction with the production association data to obtain a subcontracting path node planning scheme. The path optimization module performs data frequency matching based on the sub-package path node planning scheme, detects path planning conflict points and calibrates deviations, updates path nodes, and obtains a stable sub-package execution scheme. The instruction execution module, based on the stable subcontracting execution scheme, identifies dynamic interference in production, generates automated control instructions, transmits them to the production line equipment, and executes the sheet metal subcontracting operation.

[0076] It should be noted that the vision-based automated control system for board subcontracting provided in this embodiment of the invention is used to execute all the process steps of the vision-based automated control method for board subcontracting described in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0077] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0078] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for automated control of board material subcontracting based on vision inspection, characterized in that, include: The original image data and production-related data of the board surface are acquired. The production-related data includes production record data, real-time production cycle data and production equipment load data. The original image data is then denoised to obtain clear image data. Based on the clear image data, multi-scale texture features are extracted and analyzed, and key areas of the board surface features are determined through texture clustering and integration. The key regions of the board surface features are matched and optimized with a pre-established texture comparison template library to detect the periodic repetition pattern of the texture and adjust the feature description to obtain the surface texture identification data of the board. Based on the surface texture identification data of the board, combined with the pre-established defect boundary information database, a unique identification code for the board is generated, which is compared with the production record data to obtain the board identification conclusion. Based on the board material authentication conclusion, obtain the corresponding subcontracting instruction sequence, combine it with the production association data, plan dynamic subcontracting path nodes, and obtain a subcontracting path node planning scheme. Based on the sub-package path node planning scheme, data frequency matching is performed, path planning conflict points are detected and deviations are calibrated, path nodes are updated, and a stable sub-package execution scheme is obtained. Based on the stable subcontracting execution scheme, dynamic interference in production is identified, automated control commands are generated, transmitted to production line equipment, and the sheet material subcontracting operation is executed.

2. The automated control method for board subcontracting based on vision inspection according to claim 1, characterized in that, The process of acquiring raw image data and production-related data of the board surface, and performing noise reduction processing on the raw image data to obtain clear image data includes: The original image data is converted into a first grayscale image, the first grayscale image is segmented according to a preset grayscale threshold range, and the brightness of overly bright areas is suppressed to obtain a second grayscale image. For the second grayscale image, adaptive filtering and denoising are performed on the high gradient intensity information region to obtain the third grayscale image; The edge features and texture details of the third grayscale image are extracted, and the image is sharpened and denoised by edge enhancement to obtain the clear image data.

3. The automated control method for board subcontracting based on vision inspection according to claim 1, characterized in that, Based on the clear image data, multi-scale texture feature extraction and analysis are performed. Through texture clustering and integration, key regions of the board surface feature are determined, including: The multi-scale texture features include multi-scale texture roughness information and local texture directionality features; The clear image data is subjected to multi-scale filtering decomposition to obtain multi-scale texture roughness information and determine the roughness distribution data. Based on the roughness distribution data, local texture directional features are extracted by directional gradient calculation to determine local directional distribution data; For the local directional distribution data, key points of texture change are filtered out by a preset change threshold range to identify potential key areas; For the potential key areas, multi-scale texture features are integrated through texture clustering to determine the key areas of the board surface features.

4. The automated control method for sheet metal subcontracting based on vision inspection according to claim 1, characterized in that, The process of matching and optimizing the key feature regions of the board surface with a pre-established texture comparison template library, detecting periodic repetition patterns of the texture and adjusting the feature description to obtain the board surface texture identification data includes: The feature distribution data of key regions on the surface of the board are extracted and matched with a pre-established texture comparison template library to determine the preliminary similarity distribution data. If the preliminary similarity distribution data is lower than the preset similarity threshold, then the texture contrast difference of the key feature areas on the surface of the board is weighted and calculated to obtain the optimized similarity distribution data. For the optimized similarity distribution data, detect the periodic repetition pattern of the texture and determine the distribution of patterns that can be optimized. Based on the optimizable pattern distribution, the feature description is adjusted to obtain the surface texture identification data of the board.

5. The automated control method for board subcontracting based on vision inspection according to claim 1, characterized in that, The process involves generating a unique identification code for the board based on the surface texture identification data of the board material, combined with a pre-established defect boundary information database. This code is then compared with the production record data to obtain a board material identity verification conclusion, including: Based on the surface texture identification data of the board material, combined with the pre-established defect boundary information database, the boundary information features of the board material are extracted to form a surface feature dataset; Based on the surface feature dataset, a unique code for the board is generated by encoding and integrating the surface texture identification data of the board with the boundary information features of the board. The unique identification code of the board is matched with the production record data, and the matching degree is calculated. If the matching degree meets the preset matching threshold, the identity is confirmed to be consistent, and the board identification conclusion is obtained.

6. The automated control method for sheet metal subcontracting based on vision inspection according to claim 1, characterized in that, The step of obtaining the corresponding subcontracting instruction sequence based on the board material authentication conclusion, and combining it with the production-related data to plan dynamic subcontracting path nodes, results in a subcontracting path node planning scheme, including: Based on the board material authentication conclusion, the corresponding subcontracting instruction sequence is obtained from the pre-established instruction database, and the detailed content of the target subcontracting instruction is parsed to obtain the subcontracting instruction. Based on the detailed content of the target subcontracting instruction, the range of instruction adjustment is determined by data integration and matching with the real-time production cycle data; Based on the instruction adjustment range, analyze the production equipment load data to determine whether it meets the preset load threshold range; If the conditions are met, the node allocation is optimized by adjusting the range based on the instructions, and dynamic sub-package path nodes are planned to obtain a sub-package path node planning scheme.

7. The automated control method for board subcontracting based on vision inspection according to claim 1, characterized in that, The process of performing data frequency matching, detecting conflict points in path planning data and calibrating deviations, updating path nodes, and obtaining a stable sub-package execution scheme based on the sub-package path node planning scheme includes: Based on the subcontracted path node planning scheme, path planning data and data acquisition frequency data are obtained, and the frequency matching deviation range is determined after integration. Based on the frequency matching deviation range, conflict point data of path planning is obtained through conflict detection and deviation analysis and comparison to obtain deviation calibration parameters. Calculate the fusion deviation value of the deviation calibration parameter. If the fusion deviation value exceeds the preset deviation threshold, adjust the sub-package path and update the path node to obtain a stable sub-package execution scheme.

8. The automated control method for sheet metal subcontracting based on vision inspection according to claim 1, characterized in that, The process of identifying dynamic interference in production based on the stable subcontracting execution scheme, generating automated control commands, transmitting them to the production line equipment, and executing the sheet metal subcontracting operation includes: Based on the stable subcontracting execution scheme, real-time monitoring data of the production environment is obtained to identify the type and intensity of dynamic interference in production. Based on the type and intensity of the production dynamic interference, corresponding dynamic signal data is generated and converted into automated control commands that conform to the preset command threshold range; The automated control commands are transmitted to the production line equipment to drive the subcontracting operation process and execute the sheet metal subcontracting operation.

9. An automated control system for sheet metal subcontracting based on vision inspection, characterized in that, include: The data acquisition module acquires the original image data and production-related data of the board surface. The production-related data includes production record data, real-time production cycle data, and production equipment load data. The original image data is denoised to obtain clear image data. The feature extraction module extracts and analyzes multi-scale texture features based on the clear image data, and determines key areas of the board surface features through texture clustering and integration. The identifier generation module performs similarity matching and optimization between the key feature areas of the board surface and a pre-established texture comparison template library, detects the periodic repetition pattern of the texture and adjusts the feature description to obtain the texture identifier data of the board surface. The identity verification module generates a unique code for the board based on the surface texture identification data of the board and a pre-established defect boundary information database. The code is then compared with the production record data to obtain the board identity verification conclusion. The path planning module obtains the corresponding subcontracting instruction sequence based on the board material authentication conclusion, and plans dynamic subcontracting path nodes in conjunction with the production association data to obtain a subcontracting path node planning scheme. The path optimization module performs data frequency matching based on the sub-package path node planning scheme, detects path planning conflict points and calibrates deviations, updates path nodes, and obtains a stable sub-package execution scheme. The instruction execution module, based on the stable subcontracting execution scheme, identifies dynamic interference in production, generates automated control instructions, transmits them to the production line equipment, and executes the sheet metal subcontracting operation.

Images