Automatic generation method and system of PI film graph for reinforcing printing screen

Abstract

The invention relates to an automatic generation method and system of a PI film graph for reinforcing a printing screen, and belongs to the technical field of solar cell manufacturing, and the method comprises the steps that an original design image of the printing screen is acquired, and the original design image defines a printing area of the printing screen; obtaining at least one process parameter related to the PI film pattern; based on the original design image and the process parameters, generating a PI film filling pattern matched with the printing area through automatic image processing; wherein the automatic image processing is used for ensuring that the PI film filling pattern does not cover the printing area and a preset peripheral range, and keeping pattern interruption at a preset disconnection position of the printing area; and outputting the PI film filling pattern to be used by a subsequent cutting process. The method has the effect of realizing automatic, high-precision and rapid generation operation of the PI film reinforcing pattern.

Description

Technical Field

[0001] This application relates to the field of solar cell manufacturing technology, and in particular to an automatic method and system for generating PI film patterns for reinforcing printing screens. Background Technology

[0002] In the manufacturing process of solar cells, a printing screen is typically used to print grid electrode paste onto the cell surface to form conductive electrodes. Because the printing screen itself is relatively thin and lightweight, its mechanical strength is limited, especially in non-printed areas where structural support is insufficient. This makes it prone to deformation or damage during printing, affecting printing accuracy and the screen's lifespan. To improve the overall strength of the printing screen, current processes typically involve laminating a layer of PI film onto the non-printed areas for localized reinforcement.

[0003] Currently, the production of PI film patterns for reinforcement mainly relies on manual labor: operators need to manually identify the printing area and avoid the non-printing area according to the original design drawings of the printing screen (usually to check the film image), and draw the reinforcement pattern (i.e., the outline of the PI film) in the non-printing area based on experience, and ensure that the pattern maintains a safe distance from the electrode area, and perform corresponding pattern breaking processing at the break point of the electrode pattern, and finally generate a cutting pattern file that can be used by the laser cutting machine.

[0004] However, the above operations require each screen to be processed manually one by one, which is inefficient and cannot meet the needs of large-scale production. In addition, the consistency is poor, the accuracy of the pattern is affected by the experience of the personnel, and when processing complex patterns, it is easy to miss areas or make coverage errors, resulting in the PI film covering the electrode area, or the area that should be covered is not covered, which causes the reinforcement film to interfere with the electrode function or the reinforcement is insufficient. Therefore, it needs to be improved. Summary of the Invention

[0005] To achieve automated, high-precision, and rapid generation of PI film reinforcement patterns, this application provides an automatic generation method and system for PI film patterns used in printing screen reinforcement.

[0006] In a first aspect, this application provides a method for automatically generating PI film patterns for reinforcing printing screens, comprising: Obtain the original design image of the printing screen, wherein the original design image defines the printing area of ​​the printing screen; Obtain at least one process parameter related to the PI film pattern; Based on the original design image and the process parameters, an automated image processing method is used to generate a PI film filling pattern that is compatible with the printing area; wherein, the automated image processing is used to ensure that the PI film filling pattern does not cover the printing area and the preset surrounding area, and to maintain the pattern interruption at the preset break position of the printing area; Output the PI film filling pattern for use in subsequent cutting processes.

[0007] By adopting the above technical solution, a fully automated process for generating PI film cutting patterns, from screen printing design to reinforcement, was proposed and implemented, completely replacing the inefficient and error-prone traditional method that relies on human experience and manual drawing. Through the core step of "automated image processing," the generated PI film pattern is ensured to accurately fit the printing area, avoiding fatal errors such as PI film covering electrodes or missing areas that should be covered, fundamentally solving the core pain point of "high error rate" in manual operation. This reduces manual work that originally took hours to minutes or even seconds, and eliminates experience differences between different operators, ensuring high consistency and reliability of the generated pattern results.

[0008] Optionally, obtaining at least one process parameter related to the PI film pattern includes: Call the pre-trained process parameter recommendation model; Input the original design image into the process parameter recommendation model to obtain the recommended process parameter set output by the process parameter recommendation model; The process parameter recommendation model is a machine learning model trained on historical production data. The historical production data is associated with multiple sets of printing screen design features, the set of process parameters used, and the corresponding PI film cutting quality evaluation results.

[0009] By adopting the above technical solution and introducing a machine learning (ML)-based process parameter recommendation model, difficult-to-quantify expert experience is transformed into a repeatable and optimizable data model, thus achieving intelligent process parameter setting. The model is trained based on historical successful data and can recommend process parameters with a high probability of yielding excellent results for new network versions, overcoming the drawbacks of manual trial and error and conservative parameter settings, thereby improving process quality from the source. A data closed loop of "design features - process parameters - quality results" is established, enabling continuous accumulation and iterative optimization of production experience.

[0010] Optionally, the method further includes: After generating the PI film filling pattern, based on the original design image and the recommended process parameter set, virtual bonding and mechanical simulation are performed on the PI film filling pattern to predict the deformation and stress distribution of the printing screen after bonding the PI film filling pattern. Based on the results of mechanical simulation, the PI film filling pattern is adaptively adjusted; wherein, the adaptive adjustment includes: increasing the local support strength of the PI film filling pattern in areas where the predicted deformation or stress exceeds a preset threshold.

[0011] By adopting the above technical solutions, and introducing virtual simulation technology on the basis of automated generation, the mechanical reinforcement effect of the PI film can be predicted and optimized in advance before manufacturing, achieving a qualitative leap from "correct shape" to "reliable performance." It can automatically identify potential weak areas of the screen under stress and specifically increase local support strength, achieving "equal strength" or "lightweight" optimized design, significantly extending the screen's lifespan and printing stability. Through "stress testing" in a virtual environment, the physical trial-and-error costs caused by improper PI film design, such as premature screen scrapping and decreased printing yield, are significantly reduced.

[0012] Optionally, the subsequent cutting process is laser cutting, and the method further includes: The manufacturability of the PI film filling pattern is optimized for laser cutting process; wherein the manufacturability optimization includes at least one of the following optimization objectives: minimizing the total length of the laser cutting path of the PI film filling pattern; and ensuring that the estimated heat load distribution of the PI film filling pattern in the simulated laser cutting process meets a preset uniformity condition.

[0013] By adopting the above technical solutions, the optimization objective is extended from the functionality (reinforcement) of the PI film to its manufacturability, solving the subsequent production problem of "designed but difficult to cut". By "minimizing the cutting path", the processing time and idle movement of the laser cutting machine are directly reduced, improving equipment utilization and lowering energy consumption and production time costs. Through "uniform heat load distribution" optimization, carbonization, deformation, or edge quality deterioration of the PI film caused by localized overheating is effectively avoided, improving the dimensional accuracy and bonding yield of the finished PI film and reducing material waste.

[0014] Optionally, the method further includes: The online quality inspection data of the PI film obtained by laser cutting based on the PI film filling pattern, and / or the printing screen with the PI film attached, are obtained; wherein the online quality inspection data includes at least one of the following: the measured value of the key dimensions of the PI film, and the measured value of the deformation of the printing screen under a preset load. Based on the deviation between the measured values ​​of the key dimensions and the design values ​​of the corresponding positions in the PI film filling pattern, and / or based on the deviation between the measured values ​​of deformation and the corresponding deformation values ​​predicted by the mechanical simulation, the recommended process parameter model and / or the material mechanical parameters used in the mechanical simulation are adaptively adjusted.

[0015] By adopting the above technical solutions, a real-time physical closed loop of "design-manufacturing-inspection-feedback" is constructed, enabling the system to automatically calibrate the process model and simulation parameters according to actual production deviations. It can adaptively compensate for systematic errors caused by equipment status fluctuations, material batch differences, environmental changes, etc., significantly improving the long-term stability of production yield. This upgrades the static automated design system into an intelligent system with online learning and adaptive capabilities, which is a key step in practicing the "cyber-physical system" concept in Industry 4.0.

[0016] Optionally, the method further includes: Before outputting the PI film filling pattern, a pattern normalization operation is performed on the PI film filling pattern; the pattern normalization operation is used to eliminate non-manufacturable features in the pattern, the non-manufacturable features include at least isolated pattern areas or holes, and inner sharp corners.

[0017] By adopting the above technical solution, microscopic defects in the graphic that may lead to cutting failure or structural fragility (such as tiny islands, extremely fine connections, and sharp inner corners) are automatically repaired before output. This eliminates the tedious steps of manual inspection and repair of the graphic, ensuring that the graphics output by the automated process meet 100% of the laser cutting process requirements, truly achieving seamless integration. It eliminates problems such as cutter breakage, PI film fragment detachment and contamination, and localized overheating caused by graphic defects at the source, improving the reliability of the entire production process and product consistency.

[0018] Optionally, the step of generating a PI film filling pattern adapted to the printing area through automated image processing based on the original design image and the process parameters includes: The original design image is matched with a pre-stored template library; wherein the templates in the template library record a baseline screen design image, a baseline PI film filling image corresponding to the baseline screen design image, and a set of key driving dimension parameters; If a match is successful, the change in key driving dimension parameters of the original design image relative to the matched screen printing design image is calculated. Based on the change in the key driving dimension parameters, the reference PI film filling pattern is parametrically constrained and deformed to obtain the PI film filling pattern; wherein, the parametric constraint is used to maintain the avoidance relationship between the PI film filling pattern and the printing area during the pattern deformation process.

[0019] By adopting the above technical solutions, for serialized products or similar designs, template matching and parametric deformation reduce the graphic generation time from minutes of "recalculation" to seconds of "deformation adaptation," significantly improving order response speed. While rapidly transforming, the "parametric constraint deformation" mechanism ensures that the generated graphics inherit all design rules and performance advantages of the original optimized template (especially core avoidance relationships), preventing quality degradation caused by rapid output. While guaranteeing design quality, it provides an optimal efficiency solution for large-volume, multi-model orders, particularly suitable for product family and serialized production scenarios, achieving a balance between efficiency and customization.

[0020] Secondly, this application provides an automatic generation system for PI film patterns used for reinforcing printing screens, comprising, A printing area recognition module is used to acquire the original design image of the printing screen, wherein the original design image defines the printing area of ​​the printing screen; The process parameter determination module is used to obtain at least one process parameter related to the PI film pattern. The PI film filling pattern generation module is used to generate a PI film filling pattern adapted to the printing area based on the original design image and the process parameters through automated image processing; wherein, the automated image processing is used to ensure that the PI film filling pattern does not cover the printing area and the preset surrounding range, and to maintain the pattern interruption at the preset break position of the printing area; The PI film filling pattern output module is used to output the PI film filling pattern for use in subsequent cutting processes.

[0021] Thirdly, this application provides an automatic generation device for PI film patterns for reinforcing printing screens, including a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any of the first aspects.

[0022] Fourthly, this application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described in any of the first aspects.

[0023] In summary, this application includes at least one of the following beneficial technical effects: This application is the first to propose and realize the fully automated generation of the entire process from screen design to reinforced PI film cutting pattern, completely replacing the inefficient and error-prone traditional method that relies on human experience and manual drawing. Furthermore, by automating the core image processing steps, it is ensured that the generated PI film pattern is accurately adapted to the printing area, avoiding fatal errors such as the PI film covering the electrode or not covering the area that should be covered, fundamentally solving the core pain point of high loss rate in manual operation. Furthermore, this application reduces the manual work that originally required several hours to minutes or even seconds, and eliminates the experience differences between different operators, ensuring a high degree of consistency and reliability in the graphic generation results. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a schematic flowchart of the automatic generation method for PI film patterns for reinforcing printing screens disclosed in the embodiments of this application.

[0026] Figure 2 This is a structural block diagram of the automatic generation system for PI film patterns for reinforcing printing screens disclosed in the embodiments of this application.

[0027] Explanation of reference numerals in the attached diagram: 501, Printing area identification module; 502, Process parameter determination module; 503, PI film filling pattern generation module; 504, PI film filling pattern output module. Detailed Implementation

[0028] The following is in conjunction with the appendix Figures 1-2 This application will be described in further detail.

[0029] This application discloses an automatic generation method for PI film patterns used for reinforcing printing screens (hereinafter referred to as the automatic generation method). The executing entity is an automatic generation system for PI film patterns used for reinforcing printing screens (hereinafter referred to as the automatic generation system). The following will describe the method in conjunction with the attached... Figure 1 This section elaborates on the specific process steps of the automatic generation system executing the automatic generation method.

[0030] S101, Obtain the original design image of the printing screen, wherein the original design image defines the printing area of ​​the printing screen.

[0031] S102, Obtain at least one process parameter related to the PI film pattern.

[0032] S103, based on the original design image and process parameters, generates a PI film filling pattern that matches the printing area through automated image processing; wherein, automated image processing is used to ensure that the PI film filling pattern does not cover the printing area and the preset surrounding range, and to keep the pattern interrupted at the preset break position in the printing area.

[0033] S104 outputs the PI film filling pattern for use in subsequent cutting processes.

[0034] In implementation, for step S101, the purpose of this step is to obtain the digital source file (i.e., the original design image) of the printing screen to be hardened. The original design image is the authoritative data source defining all graphic features on the printing screen, and is usually generated by computer-aided design software and used to manufacture the printing screen itself. The original design image can be, but is not limited to, the following formats: photoplot files, DXF files, Gerber files, GDSII stream files, or high-resolution bitmap images. These files typically come from the final design data provided by the screen designer.

[0035] The original design image precisely defines the printing area. In solar cell screen printing, the printing area typically refers to all the graphic portions that need to be printed with conductive paste to form the grid electrodes. These graphics are usually identified in the image by specific colors, grayscale values, or layers. For example, in a binarized image, black areas may represent the printed areas (grid lines), and white areas may represent the non-printed areas (blank substrate). The automated generation system reads this image to lay the data foundation for subsequent processing.

[0036] For S102, the purpose of this step is to introduce key process constraints that affect the generation of PI film patterns, making the generation process configurable and adaptable. Process parameters are key variables that determine the automated image processing to generate PI film patterns that meet engineering requirements. Specific process parameters include at least: Clearance width: This refers to the minimum safe distance that must be maintained between the edge of the PI film filling pattern and the edge of the printed area pattern. This parameter ensures that the PI film will never contact or cover the grid electrodes after lamination, preventing any impact on the conductivity of the print. Its typical value can be set based on process experience; for example, the default is 400 micrometers.

[0037] Outer frame range: Defines the maximum outer boundary of the PI film pattern. It is typically aligned with the effective area of ​​the printing screen or the standard size of the PI film raw material for cutting and positioning, avoiding material waste.

[0038] Through-the-line rule parameters: used to precisely define or identify preset break points. It can be the maximum allowed bridging width (the automatic generation system automatically finds all gaps in the printed area graphic that are smaller than this width and determines them as break points that need to be penetrated), or it can be a specific coordinate or layer (reads the marked break points directly from a specific layer or metadata in the design file).

[0039] The above process parameters can be obtained in several ways, such as: Method 1: Automatically generate a set of verified default parameter values ​​pre-stored in the system; Method 2: Operators can manually input or select the aforementioned default parameter values ​​from the drop-down menu through the human-machine interface; Method 3: Read the preset parameter set corresponding to a specific printing screen signal from the preset process configuration file (the set contains the corresponding process parameter values).

[0040] For S103, the purpose of this step is to automatically convert the original design image into a PI film-filled pattern with specific avoidance and interruption features, covering only non-printing areas, according to process rules. The following is a detailed explanation of the implementation principles of this automated image processing procedure: First, the automatic generation system analyzes the original design image to accurately identify the set of graphics representing the printing area. This is typically achieved through image processing techniques: if it's a vector graphic, the graphic object representing the printing area is directly read; if it's a bitmap, image binarization is performed, classifying pixels with color or grayscale values ​​higher / lower than a specific threshold as printing areas. Then, a contour-finding algorithm is used to extract the outer contour boundaries of all printing area graphics. At this point, the automatic generation system has established a digital geometric model of the printing area graphics in memory.

[0041] Next, to ensure that the final generated PI film-filled image does not cover the printed area and its preset perimeter, the automatic generation system automatically generates a clearance region based on the outer contour of the printed area graphic identified in the previous step and the clearance width parameter obtained in S102. This process is called "contour offset" or "graphic dilation" in computer graphics. The automatic generation system translates (offsets) each printed area graphic outwards along its normal direction by a precise distance equal to the set "clearance width," based on its outer contour. After this offset, the expanded area enclosed by the new boundaries of all printed area graphics is the clearance region. The clearance region completely encompasses the originally determined printed area and its surrounding safety range (i.e., the preset perimeter).

[0042] Then, in the original design image, the graphics (such as grid lines) in the printing area may have predetermined break points. These breaks are intentionally designed for electrical or process purposes. To ensure that the final generated PI film filling pattern also maintains these breaks, the automatic generation system needs to identify these break points. One implementation is that in the original design image, the predetermined break points may be represented by extremely short graphic gaps. The automatic generation system can automatically detect all positions that meet the condition of "gap width less than the maximum allowable bridging width" by analyzing the topology of the printing area graphics and mark them as "predetermined break points." Another approach is to embed the break point information as metadata in the original design image or a separate configuration file, which the automatic generation system can directly read.

[0043] Furthermore, when generating the final PI film filling pattern, the automatic generation system ensures that the pattern is completely cut off at these "preset break points," meaning that the PI film filling pattern also has a corresponding gap at these locations. This can be achieved by forcibly introducing a "cutting line" or "negative pattern" at these locations during the pattern processing.

[0044] Finally, after identifying the areas to be avoided (avoidance areas) and the locations where interruption is necessary (preset break points), the automatic generation system begins constructing the PI film filling pattern. This construction process essentially involves a series of Boolean operations: the automatic generation system first determines a valid working area containing all the graphics (usually the outer border of the printing screen), then uses this valid working area as an initial filling area, subtracts the avoidance area from it, and then, based on the resulting graphic, cuts at the determined preset break points to ensure the final graphic breaks at those points. After these operations, the final graphic is the PI film filling pattern. This PI film filling pattern perfectly fills all areas except the printing area and the safety avoidance zone, maintaining the correct break structure at the required locations, achieving compatibility with the printing area.

[0045] For S104, this step converts the digitized PI film filling pattern into a file usable by the production equipment. Specifically, the automated generation system converts the PI film filling pattern (usually vector data) generated in memory into an industry-standard file format that downstream laser cutting equipment can directly recognize and execute, such as DXF, AI, or NC code files specific to a particular cutting machine. The generated industry-standard file format can be saved to a designated directory or sent directly to the laser cutting machine's control system via a network. Operators or automated production lines can then access this file to drive the cutting machine to precisely cut the PI film raw material, obtaining PI reinforced films that correspond one-to-one with the printing screen.

[0046] Optionally, S102 specifically includes the following sub-steps: S1021, invoke the pre-trained process parameter recommendation model; S1022, Input the original design image into the process parameter recommendation model to obtain the recommended process parameter set output by the process parameter recommendation model; Among them, the process parameter recommendation model is a machine learning model trained on historical production data. The historical production data is associated with multiple sets of printing screen design features, the set of process parameters used, and the corresponding PI film cutting quality evaluation results.

[0047] In implementation, after acquiring the original design image (usually a binarized bitmap or vector graphics file), the automatic generation system executes a predetermined image analysis algorithm to calculate a set of fixed-dimensional numerical feature vectors F. The feature vector F includes at least the following components directly calculated through image processing: F1: Minimum linewidth (µm): First, a morphological skeletonization algorithm is used on the binary image identified as the printing area to obtain the central skeleton line of the raster graphic. Then, for each pixel on the central skeleton line, the shortest Euclidean distance from that point to the boundary of the original printing area graphic (non-printing area) is calculated. Twice this distance is the local linewidth at that point. By traversing all skeleton points, the minimum local linewidth value obtained is F1.

[0048] F2: Graphic Complexity Index: Calculated as (P²) / (4πA), where P is the total pixel perimeter of the outer contour of all printed areas, and A is the total pixel area of ​​all printed areas. This index is dimensionless and is used to quantify the irregularity of the graphic's shape.

[0049] F3: Percentage of non-printed area: The calculation formula is: (Total number of pixels in the image - Number of pixels in the printed area) / Total number of pixels in the image.

[0050] F4: Break Location Density (numbers / mm²): Performs horizontal and vertical projection analysis on the printed area graphic. Specifically, each row of pixels is scanned horizontally, and consecutive segments of printed area pixels are marked as a "horizontal segment". If the width of the non-printed area pixels between two adjacent "horizontal segments" is less than a preset threshold (e.g., 5 pixels), the location is recorded as a potential horizontal break location. The same method is used to perform a vertical scan to obtain vertical break locations. The two sets of locations are deduplicated (if a physical break is detected in both the horizontal and vertical scans, it is counted as 1), resulting in the total number of break locations N_gap.

[0051] Historical production data is stored in a relational database. Each record R_i contains: Key F_i: The feature vector extracted from historical task i. Key P_i: The set of process parameters actually used and verified to be effective in task i, which is a set of key-value pairs, such as {"avoidance width": value_a, "bevel radius": value_b, ...}. Key Q_i: The PI film cutting quality evaluation result corresponding to task i, i.e., the quality label. Q_i is a comprehensive score value, and its calculation method is predefined as follows: Q_i=w1* S_precision+w2 *Y_yield+w3 * L_life. Where: S_precision is the dimensional accuracy score. The deviations of the three key dimensions (d1, d2, d3) of the cut PI film from the design values ​​(D1, D2, D3) are measured using an optical measuring instrument and calculated as 1 / (1 + avg(|d_k - D_k|)), normalized to the 0-1 interval.

[0052] Y_yield represents the bonding yield. It is the percentage of screen printing plates that successfully bonded without rework out of the total number in this batch, with a value ranging from 0 to 1.

[0053] L_life is the normalized value of the lifespan improvement rate. It is calculated as (actual improvement cycle / target improvement cycle) and is limited to a maximum value of 1.

[0054] w1, w2, and w3 are preset weighting coefficients, and w1 + w2 + w3 = 1, for example (0.4, 0.4, 0.2). The value of Q_i ranges from 0 to 1, and a higher value indicates better overall quality.

[0055] The industrial parameter recommendation model uses a Gradient Boosting Decision Tree (GBDT) regression model, which is constructed sequentially from multiple weak learners (decision trees) to minimize prediction error. High-quality records with Q_i > Q_threshold (e.g., a threshold of 0.7) are selected from the historical database to form the training dataset D_train = {(F_i, P_i)}. Q_threshold is the quality screening threshold, and its value is determined by analyzing the distribution of Q_i in the historical data, for example, by taking the median or upper quartile of all historical Q_i values, to ensure that the training data comes from high-quality production batches.

[0056] The feature vector F_i is used as the model input, and each key parameter (such as the avoidance width) in the process parameter set P_i is trained as an independent regression objective. For multi-parameter outputs, multiple models or a single multi-output model can be trained. Mean squared error is used as the loss function L=Σ(y_true-y_pred)². A greedy algorithm is used to iteratively add new decision trees. The goal of constructing each new tree h_m(x) is to fit the residual (negative gradient) between the current model's predicted value F_{m-1}(x) and the true value y. The model update rule is: F_m(x)=F_{m-1}(x)+η*h_m(x), where η is the learning rate, used to control the influence of each tree. Cross-validation grid search is used to determine the optimal tree's maximum depth, learning rate, number of trees, and other hyperparameters. This is used to train the process parameter recommendation model. After training, the model parameters (including the node splitting features, splitting thresholds, and leaf node values ​​of each tree) are serialized into a file and integrated into the automatic generation system.

[0057] For a new printing screen, the automatic generation system first extracts a feature vector F (i.e., the printing screen design features) from the original design image. Then, it calls the deployed industrial parameter recommendation model, inputting the extracted feature vector F into it. Within the industrial parameter recommendation model, the feature vector F sequentially passes through each decision tree: starting from the root node, it determines whether to flow to the left or right child node based on the "split feature index" and "split threshold" stored in the node, until it reaches a leaf node, which stores a predicted value. The predicted values ​​from all trees are multiplied by the learning rate, summed, and then added to the initial predicted value to obtain the final recommended parameter value P_rec. The automatic generation system outputs and uses this P_rec as the process parameters for subsequent image generation.

[0058] Optionally, the automatic generation method may also include the following steps: S201 After generating the PI film filling pattern, based on the original design image and recommended process parameter set, virtual bonding and mechanical simulation are performed on the PI film filling pattern to predict the deformation and stress distribution of the printing screen after bonding the PI film filling pattern. S202, Based on the results of mechanical simulation, the PI film filling pattern is adaptively adjusted; wherein, the adaptive adjustment includes: increasing the local support strength of the PI film filling pattern in areas where the predicted deformation or stress exceeds a preset threshold.

[0059] In implementation, this scheme is executed after the initial generation of the PI film filling pattern, forming an analysis-optimization closed loop. Specifically, it includes the following steps: The automated generation system initiates a finite element analysis (FEA) kernel (e.g., integrating an open-source FEA solver, or calling the ANSYS / ABAQUS command line). Then, a 3D model is constructed: the graphic contour (i.e., the boundary between the printed and non-printed areas) in the original design image is used as a cross-section, and a 3D shell with a thickness of t_screen (e.g., 0.05 mm) is generated through an extrusion operation to represent the screen printing substrate. Similarly, the contour of the PI film filling pattern is used to generate a 3D shell with a thickness of t_PI (e.g., 0.1 mm) through an extrusion operation. In 3D space, the mid-surfaces of these two shells are adjusted to a distance of (t_screen + t_PI) / 2 to simulate the spatial relationship before bonding.

[0060] Then, material properties are assigned to the two 2D shells. For the screen shell, the material card sets YOUNG_MODULUS (elastic modulus) to 200000 (MPa) and POISSON_RATIO (Poisson's ratio) to 0.3. For the PI membrane shell, the material card sets YOUNG_MODULUS to 3000 and POISSON_RATIO to 0.34. In the solver, within the finite element solver, the lower surface of the PI membrane shell is defined as the "slave surface," and the upper surface of the screen shell as the "master surface." Using the solver's surface pairing function, the nearest "master surface" element is found for each "slave surface" node in its projection direction (usually the shell normal direction), and a node-face binding constraint is created, making the displacement of the slave surface node completely dependent on the interpolated displacement of its bound master surface element. This simulates adhesive bonding. ENCASTRE (fully constrained) boundary conditions are applied to all nodes on the four edges of the screen shell, i.e., UX=UY=UZ=0.

[0061] Next, based on the preset squeegee printing path parameters (including the printing direction and squeegee width W_blade, e.g., 50mm), a strip-shaped region with a width of W_blade is generated on the upper surface of the screen shell along the printing direction. A uniformly distributed pressure load p (e.g., 0.5 N / mm²) perpendicular to the surface is applied to the surface of all units within this region. The simulation calculation task is then submitted. The solver will output a result file (e.g., .odb, .rpt).

[0062] The system then automatically generates a file that parses the result and extracts two key data arrays: U_MAG: the resultant displacement of each node; and S_MISES: the Mises equivalent stress of each element.

[0063] Next, the automatic generation system, based on preset safety lines (deformation alarm line U_limit = 0.1 (mm), stress alarm line S_limit = 80 (MPa)), traverses all elements. If S_MISES[i] > S_limit, the ID of element i is added to the list stress_over_list. It then traverses all nodes; if U_MAG[j] > U_limit, the ID of the element to which node j belongs is added to the list disp_over_list. The two lists are merged and duplicates are removed to obtain the set of all problematic elements {E_problem}. To perform region division, the automatic generation system performs the following steps: 1. Coordinate mapping: For each cell in the set {E_problem}, obtain the three-dimensional coordinates of all its nodes, and project its center point onto a two-dimensional plane (XY plane) parallel to the screen plane to obtain the corresponding two-dimensional coordinate point P_i(x, y).

[0064] 2. Adjacency Determination and Clustering: Define a distance threshold D_thresh (e.g., 1.5 times the average side length of the cells). Initialize an empty list of regions Zones = []. Iterate through all 2D points P_i: 2.a If the current point P_i has not been visited, then use it as a seed to create a new region Zone_k.

[0065] 2.b For efficient searching, the automatic generation system can pre-store all unvisited points P_j using a spatial index structure (such as a KD-Tree). For the current region Zone_k, query the index for all unvisited points whose distance to any point within Zone_k is less than D_thresh, add them to Zone_k, and mark them as visited.

[0066] 2.c Repeat step 2.b until no new points can be added to Zone_k.

[0067] 2.d Add Zone_k to the Zones list and assign it a unique number k.

[0068] 3. Filter small regions: Traverse Zones and remove regions that contain fewer than the minimum number of points N_min (e.g., 3 points) to avoid unnecessary graphic adjustments to noisy points.

[0069] Ultimately, the list Zones stores the continuous problem regions Zone_k that require graphical adaptation adjustments.

[0070] Finally, the automatic generation system opens the original PI film vector graphics file (e.g., .dxf) and, for each Zone_k, determines the root cause of the problem: if the proportion of elements from the stress_over_list in Zone_k is high (e.g., >60%), it is marked as "stress-related". Otherwise, it is marked as "deformation-related".

[0071] Then perform the graphic modification operation: For "stress-type" regions: Perform a graphic expansion operation. In the original PI film vector graphic, select all graphic elements (such as line segments and polygons) that intersect with the circumscribed polygon of region Zone_k (i.e., the boundaries intersect or the graphic is located inside the polygon). Perform a graphic expansion operation on these selected graphic elements, using the OffsetPath function in a graphics algorithm library (such as ClipperLib) to offset the selected graphic elements outwards by delta_w (e.g., 0.02 mm), and then perform a Union operation on the old and new graphics.

[0072] For "deformed" regions: Perform local mesh refinement. If this is a mesh-filled area, update the mesh edge length parameter L within this Zone_k range to L / 1.5, and then regenerate the mesh within this region.

[0073] If Zone_k has a large area (e.g., >10 mm²): regardless of whether it's "stress type" or "deformation type", insert a predefined "cross" standard block at the centroid coordinates (Xc, Yc) of this area. After all modifications are complete, save it as a new version of the DXF file PI film graphic_v2.dxf.

[0074] The automatic generation system takes the PI membrane graphic_v2.dxf as new input, automatically jumps back to the first step, and remodels, loads, and calculates. It then checks again to see if any elements exceed the safety limit, implements a loop, and ensures the number of loops does not exceed the preset number N_max = 5. If both the stress_over_list and disp_over_list extracted in any iteration are empty, it is considered that the optimization has converged, and the loop immediately stops. The DXF file of the last successful loop is output as the final PI membrane cutting diagram.

[0075] Optionally, the subsequent cutting process is laser cutting, and the automatic generation method also includes the following steps: The manufacturability of PI film filling patterns is optimized for laser cutting processes. The manufacturability optimization includes at least one of the following optimization objectives: minimizing the total length of the laser cutting path of the PI film filling pattern; and ensuring that the estimated heat load distribution of the PI film filling pattern in the simulated laser cutting process meets the preset uniformity condition.

[0076] In implementation, this optional approach aims to ensure that the generated PI film-filled pattern not only meets functional requirements but is also easily cut and manufactured efficiently and with high quality. This process can be intervened at any stage before the final output pattern for further optimization. It includes the following steps: The automatic generation system reads the PI film filling pattern to be optimized (this pattern may have already been optimized using S1021-S1022 or S201-S202) and determines the optimization objectives to be performed. The automatic generation system is configured with a database of basic process parameters for the laser cutting machine, including: laser power P_laser, cutting speed V_cut, spot diameter D_spot, etc.

[0077] I. Optimization Objective (A): The implementation process for minimizing the total length of the laser cutting path of the PI film filling pattern is as follows: Step A1: The PI film filling pattern is stored in an industry-standard vector format, such as DXF or Gerber files. The automatic generation system reads the DXF file and extracts all geometric primitives representing the pattern boundaries, including line segments, arcs, and Bézier curves. For uniform processing, all non-linear primitives are converted into a sequence of short line segments: For the circular arc, set the maximum chord height error δ_max (e.g., 0.001 mm). Calculate the bisector angle θ = 2 * arccos(1 - δ_max / r) based on the radius r and δ_max, dividing the circular arc into several chord segments.

[0078] For the Bézier curve, the de Castelli algorithm is used to recursively subdivide it until the maximum distance between all subdivided segments and the original curve is less than δ_max.

[0079] Generating an ordered set S of cut line segments includes the following sub-steps: Step A.1 Define the vertex coordinate list: For each closed contour, the automatic generation system extracts a series of vertex coordinates sequentially from its discretized polygon contour. If a contour has M vertices, then the vertex coordinate list is obtained as follows: Coordinates = [ (x_1, y_1), (x_2, y_2), ..., (x_M, y_M) ], where (x_M, y_M) and (x_1, y_1) are the same point, indicating that the contour is closed.

[0080] Step A.2: Generate line segment coordinate pairs: Based on the vertex order, pair consecutive vertices of the contour to generate a set of line segments representing each side of the contour. The specific steps are as follows: Initialize an empty list segment_list_for_this_contour = [].

[0081] For index i looping from 0 to M-2: Let the starting point coordinate of the line segment be start_point = (x_i, y_i). Let the ending point coordinate of the line segment be end_point = (x_{i+1}, y_{i+1}). Define the line segment as a coordinate pair (start_point, end_point) and add it to segment_list_for_this_contour.

[0082] Handling closed edges: Finally, generate the line segment connecting the last vertex and the first vertex: start_point = (x_{M-1}, y_{M-1}), end_point = (x_0, y_0), and also add this coordinate pair to the list.

[0083] Step A.3: Summarize and construct set S: Iterate through all contours and append all line segment coordinate pairs in segment_list_for_this_contour for each contour to a general list in order.

[0084] Suppose there are a total of N line segments, then this list is the ordered set of line segments S. We can represent it as: S = [seg_1, seg_2, ..., seg_N]; where, for the k-th element seg_k in set S, it is explicitly a coordinate pair: seg_k = (start_k, end_k). Here, start_k and end_k are two-dimensional coordinate variables, whose values ​​are assigned in the loop of step A.2 above. For example, start_k = (x_i, y_i), end_k = (x_{i+1}, y_{i+1}). For continuous line segments seg_k and seg_{k+1} belonging to the same contour, they satisfy a geometric relationship: the coordinates of end_k are equal to the coordinates of start_{k+1}. This is precisely the meaning of "ordered".

[0085] Step A2 (Constructing the cutting path planning graph model G=(V, E)): First, for each line segment seg_k in set S, create two graph vertices: v_k_s represents "cutting from the starting point of seg_k" and v_k_e represents "cutting from the ending point of seg_k" (or "cutting to the ending point of seg_k").

[0086] Then, create a new vertex in the graph and name it v_start. The spatial coordinates of v_start (x_start, y_start) are determined based on the initial safe position of the laser cutter or the preset alignment point of the PI film material, and are used as input for the automatic generation system configuration parameters.

[0087] Then define the edge set E and the weight W: Cut edges: Add an undirected edge between each pair (v_k_s, v_k_e), with weight w = ||end_k - start_k|| (segment length).

[0088] Initial empty edge shift: For each v_k_s, add a directed edge (v_start → v_k_s), with weight w = β * ||start_k - (x_s, y_s)||. β is the lift penalty coefficient (e.g., 1.05).

[0089] Ordinary empty edge shift: For any two vertices v_a and v_b (not the start and end points of the same line segment), if their spatial points are directly reachable, then add a bidirectional edge. The weight w = α * ||point(v_a) - point(v_b)||. If the line segments seg_a and seg_b to which v_a and v_b belong are not on the same contour, then α = α_lift (lift penalty, as in 1.1), otherwise α = 1.

[0090] Step A3 (Calculate the current path length baseline): Run the nearest neighbor greedy algorithm on graph G: Starting from v_start, always select the edge with the smallest weight and move it to the endpoint of the untraversed "cut edge" until all cut edges have been traversed. Calculate the total weight of this path as the current optimal path length L_current.

[0091] Step A4 (Graphical Iterative Optimization Based on Path Analysis): The automatic generation system enters an optimized main loop, each loop containing the following sub-steps: A4.1 Analyze the path obtained in step A3 and identify: High-frequency empty-shifted connection pairs: Count the number of empty-shifted edges used and select the top K pairs (e.g., K=3) of vertices (v_i, v_j) that are frequently connected.

[0092] Ultra-short cut segments: Mark all segments with a length (seg_k) < L_short (e.g., 0.15 mm).

[0093] Isolated Contours: The automatic generation system maintains a contour set C, initially with each original closed contour as one element. During path analysis, whenever a lift edge (i.e., a free-movement edge) satisfying α = α_lift is encountered, and this edge connects vertices belonging to two different contours C_i and C_j, these two contours are considered to be visited separately in this cutting path. The number of times each contour C_i is connected to other contours via a non-lift lift edge (α = 1) is counted. Contours that are only connected to other contours via lift lift edges are marked as "isolated contours".

[0094] A4.2 Perform geometric modifications to the graphic: Rule R1 (Merging Regions): For each high-frequency spatial shift connection pair (v_i, v_j), find its corresponding original graphic regions Region_i and Region_j. (Note that when generating set S in step A1, each seg_k is pre-labeled with the ID tag contour_id of its original contour. Thus, vertex v_i is naturally associated with the unique contour_id through its source line segment seg_k, thereby finding the "corresponding original graphic region".) By calling the convex hull calculation function in the computational geometry library, calculate the convex hull polygon containing all vertices of Region_i and Region_j, and replace the original two regions with this new convex hull polygon.

[0095] Rule R2 (Delete Short Lines): For each ultra-short cutting line segment seg_k, delete vertices start_k and end_k from its associated polygon, and directly connect the predecessor vertex of seg_k to its successor vertex.

[0096] Rule R3 (Bridging Isolated Contours): For each isolated contour, calculate its nearest point pair (p_a, p_b) to the main graphic region. Add a rectangle centered on p_a and p_b with a width of W_bridge (e.g., 0.2 mm) to connect the two. Immediately after each modification, perform coverage verification: check if the new graphic completely covers all sample point sets in the original graphic that must be covered by the PI film (d_sample (e.g., 0.05 mm) within the outer rectangular region of the original graphic during process initialization). For each point located at a grid intersection and falling within the area of ​​the original graphic that must be covered by the PI film (i.e., the area after avoidance and mechanical optimization), add it to the required sample point set P_must. During verification, check if the new graphic (polygon set) contains every point in P_must (i.e., whether the point is inside the polygon or on its boundary).

[0097] A4.3 Update Model and Recalculate: With the modified new graph, re-execute steps A1 to A3 to generate a new graph model G' and calculate the new shortest path length L_new.

[0098] A4.4 Judgment and Convergence: If L_new < L_current and the coverage verification passes, accept the current modification, set L_current = L_new, and enter the next loop with the new graph. If the current modification is rejected, revert to the previous graph and try the next priority modification rule (modification rules are tried in the order of R1 (merge regions), R2 (delete short lines), and R3 (bridge isolated contours). When a high-priority rule cannot produce a valid modification, try a low-priority rule). If L_current cannot be reduced by more than ε (e.g., 0.1%) after M consecutive loops (e.g., M=5), or the total number of loops reaches the upper limit, the optimization terminates.

[0099] A4.5 Output: Outputs the last accepted graph with the shortest path, as the optimized PI film filling graph _optimized.

[0100] II. Optimization Objective (B): The implementation process for ensuring that the estimated heat load distribution of the PI film filling pattern meets the preset uniformity conditions during the simulated laser cutting process is as follows: Step B1 (Thermodynamic Model Construction and Initialization): The automatic generation system reads the preset laser cutting process parameter database to obtain key parameters: laser power P_laser (unit: W), cutting speed V_cut (unit: mm / s), and laser spot radius r_spot (unit: mm). The automatic generation system uses a simplified formula based on the classic Rosenthal moving point heat source model to quickly estimate the temperature rise caused by laser irradiation. For a point on the material surface, when the laser passes along a straight path at a speed V_cut, the peak temperature rise ΔT_peak at that point can be approximately expressed as a function positively correlated with P_laser / V_cut, and decreases exponentially with the increase of the vertical distance d from the point to the laser path. A safe temperature rise threshold ΔT_safe (e.g., 200°C) and a critical temperature rise threshold ΔT_critical (e.g., 300°C, which may lead to carbonization) are set for the PI film material. These thresholds serve as the basis for subsequent judgments.

[0101] Step B2 (Cut Path Generation and Thermal Load Pre-analysis): Automatic Generation System Calls and Optimization Objective (A) Using the same method as steps A1-A3, the current PI film filling pattern is converted into an ordered set of line segments S, and an initial cutting path Path is generated (e.g., a path obtained using a nearest neighbor greedy algorithm). This path determines the laser head's movement sequence. The automatic generation system then analyzes the pattern geometry and pre-marks high-risk areas prone to heat accumulation. Interior sharp angle: Identifies all vertices whose interior angle is less than θ_risk (e.g., 120 degrees).

[0102] Short segments: Mark all segments whose length L_seg is equivalent to the spot diameter 2*r_spot (e.g., L_seg < 3 * r_spot).

[0103] Dense Intersection Region: Extract the endpoints of all line segments in the graph. Draw a circle with each endpoint p_i as the center and r_spot as the radius. Count the number N_i of other line segment endpoints that fall within this circle. If N_i ≥ N_threshold (e.g., 3), then mark p_i as a "dense intersection node". Cluster all dense intersection nodes whose distance to each other is less than r_spot to form a "dense intersection node".

[0104] Step B3 (Based on thermal model simulation and hot zone marking): The automatic generation system simplifies the laser as a constant heat source moving along the path. For a point p(t) on the path at time t, the instantaneous heat flux density I(q, t) caused by it to any point q(x, y) on the material surface is approximated by the following Gaussian distribution model: I(q, t) = (η * P_laser) / (π * r_spot²) * exp( - ||qp(t)||² / r_spot²); where η is the absorptivity of the material to the laser (as a preset parameter, for example, 0.8), and ||q-p(t)|| is the Euclidean distance from point q to the laser center p(t).

[0105] For a fixed point q, the cumulative heat load H(q) received during the entire cutting process is obtained by integrating the instantaneous heat flux density along the laser path over time. To simplify the calculation, a discrete superposition method is used: H(q)≈Σ_t [I(q,t)*Δt]; where the summation iterates through all discrete time steps Δt (Δt = d_grid / V_cut). This H(q) is a dimensionless relative value used to compare the degree of heat accumulation at different locations.

[0106] Create a 2D grid with a resolution of d_grid (e.g., 0.02 mm) on the bounding rectangular region of the graphic. For each grid point g, calculate H(g) according to its coordinates using the formula above. Calculate the average H_avg and standard deviation σ of all grid points' H values. Mark hot zones: group all grid points satisfying H(g) > H_avg + 2σ into the same connected region; each connected region is a HotZone_q. Mark overheating hazard zones: mark all grid point regions satisfying H(g) > H_avg + 4σ as overheating hazard zones.

[0107] Step B4 (Targeted Modification of Graphic Geometry): The system executes corresponding geometric modification rules based on the different risk characteristics identified in steps B2 and B3: Rule R_B1: Dynamic Rounding of Inner Sharp Corners (for Geometric Features and Hot Zones): Calculate a required rounding radius R_needed based on the estimated heat load value H_corner at the sharp corner (taken from the heat load map H). The calculation formula is: R_needed = R_base + k * (H_corner - H_avg), where R_base is the base safety radius (e.g., 0.05 mm) and k is a scaling factor. After calculation, the automatic generation system performs a verification-adjustment loop: based on the current R_needed value, it simulates rounding replacement on the graph and recalculates the heat load value H_corner_new for the corrected area locally (only this local area is recalculated, not the entire graph). If H_corner_new > H_risk, R_needed is increased by a step size ΔR (e.g., 0.01 mm), and this verification is repeated until H_corner_new ≤ H_risk or R_needed reaches the maximum allowable value R_max. Finally, the rounded corner is replaced using R_needed that meets the conditions (the sharp corner is replaced with an arc with a radius of R_needed that is tangent to both sides).

[0108] Rule R_B2: Merging or eliminating short segments and minute structures: For an identified short segment seg_short (whose length L_short satisfies L_short < 3 * r_spot): Find the preceding segment seg_prev and the following segment seg_next that are geometrically connected to it. If the lengths of seg_prev and seg_next are both greater than μ times L_short (e.g., μ = 2), i.e., satisfying: length(seg_prev) > μ * L_short and length(seg_next) > μ * L_short, then it is determined that the short segment is connected to longer segments at both ends, meeting the merging condition. In this case, delete vertices start_k and end_k (i.e., the two endpoints of the short segment). Connect the non-common endpoints of seg_prev and seg_next directly with a new straight line segment to form a simpler outline.

[0109] Rule R_B3: Introduce "thermal fracture" for overheating hazard zones: For the "overheating hazard zone" HotZone_q, calculate the minimum bounding rectangle of its pixel set. If the aspect ratio (long side / short side) of this rectangle is greater than the threshold Aspect_ratio_thresh (e.g., 3), it is determined to be a "strip structure." The automatic generation system inserts a tiny gap of width W_break at the midpoint of this rectangular area, perpendicular to its long side, breaking the originally continuous graphic at this point. This gap acts as a "thermal fracture," preventing heat from accumulating along long distances in the graphic. The introduction of this gap must not affect the overall support strength of the graphic, and its location must avoid high-stress load-bearing areas determined in mechanical simulations (such as the schemes described in S201-S202).

[0110] Step B5: Apply rules R_B1, R_B2, and R_B3 sequentially to modify the graph. After each modification, immediately perform graph coverage verification (using the same method as step A4.2 of optimization objective (A), checking whether the required sampling point set P_must is still covered). For the modified graph, repeat steps B2 and B3 to generate a new cutting path and heat load distribution map H_new. Calculate the heat load evaluation indicators of the new graph, such as H_max_new and the area of ​​the H_risk region Area_risk_new. If H_max_new decreases and Area_risk_new decreases, and the coverage verification passes, the modification is accepted. If the indicators deteriorate or the modification is ineffective, roll back (e.g., try other modification objects under the same rule or directly proceed to the next priority rule). When several consecutive iterations fail to significantly improve the heat load indicators (e.g., H_max decrease < 1%, Area_risk decrease < 5%), or there are no more suitable inner sharp corners or short line segments to modify, the optimization terminates.

[0111] Finally, after process optimization for optimization objective (A) and / or optimization objective (B), the automatic generation system generates the final PI film filling pattern _final.

[0112] Optionally, the automatic generation method may also include the following steps: S301, acquire online quality inspection data of the PI film obtained by laser cutting based on the PI film filling pattern, and / or the printing screen with the PI film attached; wherein, the online quality inspection data includes at least one of the following: the measured value of the key dimensions of the PI film, and the measured value of the deformation of the printing screen under a preset load; S302, based on the deviation between the measured value of the key dimension and the design value of the corresponding position in the PI film filling pattern, and / or based on the deviation between the measured value of deformation and the corresponding deformation value predicted by the mechanical simulation, adaptively adjust the material mechanical parameters used in the process parameter recommendation model and / or mechanical simulation.

[0113] In implementation, after the PI film filling pattern is generated and has been used in production (i.e., laser cutting and bonding to the printing screen have been completed), the automatic generation system starts and executes the following closed-loop adaptive optimization process to achieve model self-correction. The specific steps of this process are as follows: Step C1: The automatic generation system controls a preset industrial camera to photograph the cut PI film, obtaining a color image Img_color. Then, Img_color is converted to a grayscale image Img_gray. To suppress noise, Gaussian filtering is applied to Img_gray, with the convolution kernel size set to 5×5 pixels. Next, the automatic generation system uses the Sobel operator to calculate the gradient of Img_gray; specifically, the image is convolved with a horizontal Sobel kernel Sobel_x and a vertical Sobel kernel Sobel_y to obtain gradient components Gx and Gy. The gradient magnitude is calculated as G = sqrt(Gx^2 + Gy^2). A high threshold high_thresh and a low threshold low_thresh are set, and G undergoes double-threshold hysteresis processing: pixels in G with values ​​greater than high_thresh are marked as strong edges; pixels with values ​​between [low_thresh, high_thresh] and adjacent to any strong edge pixel are marked as weak edges; the remaining pixels are set as background. The final result is a binarized edge image called Edges.

[0114] Then, the automatic generation system projects the edge pixels in the Edges image vertically along the direction perpendicular to the length of the support strip under test, generating a one-dimensional projection curve profile, where the value at each position represents the number of edge pixels at that location. This profile curve is then fitted using a Gaussian function model g(x) = A * exp(-(x-μ)² / (2σ²)). The fitting result yields the center positions μ1 and μ2 of two Gaussian peaks, corresponding to the two edges of the support strip, respectively. The pixel distance Δpixel = |μ2 - μ1| between the two center points is calculated.

[0115] Based on the camera's pre-calibration results (pixel equivalent to pixel_to_mm), the physical distance d_pixel = Δpixel * pixel_to_mm is calculated. Simultaneously, the average value σ_avg of the σ parameters of the two Gaussian peaks is taken to compensate for edge diffusion effects. Finally, the support strip width m1 = d_pixel - 2 * σ_avg * pixel_to_mm.

[0116] Similarly, for the corresponding corner radius M_2, the coordinates of the edge pixels in the corresponding arc region in Edges are extracted, and the center and radius are solved using the least squares circle fitting algorithm. The center lines of the two parallel edges are extracted, and the perpendicular distance M_3 between the two lines is calculated after straight line fitting. After completing the above measurements, the system automatically generates the actual size vector M_actual = (m1, m2, m3).

[0117] Step C2 (Online Deformation Data Acquisition): At the screen lamination station, the system controls a calibrated force-controlled actuator to apply a vertical standard test load of magnitude F_test (e.g., 5 Newtons) to a specific measurement point on the surface of the screen with the PI film already laminated. A micro-displacement sensor (e.g., a sensor using the laser triangulation principle) mounted coaxially with the actuator monitors the vertical displacement of this point in real time and outputs an analog voltage signal raw_voltage (unit: volts V) proportional to the displacement.

[0118] The automatic generation system reads the instantaneous value of raw_voltage via an analog-to-digital converter. To convert the voltage value into a physical displacement value, the system accesses the calibration data file uniquely bound to this micro-displacement sensor. This file records the sensor's calibration parameters: sensitivity coefficient S (unit: mm / V) and zero-point offset voltage V_offset (unit: V). The sensitivity coefficient S represents the actual displacement change corresponding to each volt change in voltage; the zero-point offset voltage V_offset represents the theoretical output voltage of the sensor at zero displacement.

[0119] The automatic generation system calculates the actual displacement value D_actual (unit: mm) based on the following linear transformation formula: D_actual = S * (raw_voltage - V_offset). Where: (raw_voltage - V_offset) calculates the effective voltage change relative to zero. S * (raw_voltage - V_offset) is the physical displacement represented by this voltage change. The calculated D_actual is the measured vertical displacement value of the screen measurement point under load F_test, which will be used for subsequent deviation calculations.

[0120] Step C3: The automatic generation system reads the DXF file of the PI film filling pattern generated by S101-S104 corresponding to this production task. In this file, it locates the graphic features (such as specific line segments or circles) corresponding to the online measurement positions M_1, M_2, and M_3, and directly obtains their design dimension values ​​d1, d2, and d3 by analyzing their geometric parameters, forming the design value vector M_design = (d1, d2, d3). The automatic generation system accesses the mechanical simulation result file (usually in .odb or .dat format) associated with this task. In this file, it locates the finite element node N_sim that corresponds to the actual measurement point position by node number or node coordinates, and reads the displacement output value of this node in the Z direction (vertical direction) as the simulation predicted displacement D_sim. Then, it calculates the deviation according to the following formula: Relative deviation of dimensions: ΔM_i = (m_i - d_i) / d_i, where i = 1, 2, 3.

[0121] Absolute deviation of deformation: ΔD = D_actual - D_sim.

[0122] Step C4 (Adaptive adjustment of tool compensation values ​​in the recommended process parameter model): The automatic generation system executes the following steps: C4.1 Triggering Adjustment: Calculate the arithmetic mean of the dimensional deviations ΔM_avg = (ΔM_1 + ΔM_2 + ΔM_3) / 3. The system automatically generates a preset dimensional deviation threshold δ_m (e.g., 0.02, representing 2%). If |ΔM_avg| > δ_m, a systematic dimensional deviation is determined, triggering the tool compensation value adjustment process.

[0123] C4.2 Querying the Current Tool Offset Value: The process parameter recommendation model internally maintains a tool offset value lookup table. This table uses the combination of the feature vector's hash value Hash(F) and the material thickness t as the primary key. The automatic generation system calculates key = (Hash(F_current), t_current) based on the current task's feature vector F_current and material thickness t_current, and retrieves the currently used tool offset value Offset_old from this lookup table.

[0124] C4.3 Calculate the new tool compensation value: The automatic generation system calculates the new tool compensation value based on the following linear feedback control formula: Offset_new = Offset_old - K_p * ΔM_avg. Where K_p is a proportional control coefficient (e.g., 0.1 mm), which physically means the number of millimeters the tool offset needs to be adjusted when the dimensional deviation is 100%. A positive ΔM_avg indicates that the actual object size is too large; therefore, the new tool offset value Offset_new should be less than Offset_old, causing the subsequent cutting path to shift inward, thus cutting a smaller shape. This formula is based on the classic control principle that deviation is proportional to the correction amount.

[0125] C4.4 Update Lookup Table: The automatic generation system updates the offset value stored in the entry with key (Hash(F_current), t_current) in the lookup table to Offset_new.

[0126] Step C5 (Elastic modulus inversion calibration of material parameters in mechanical simulation): The automatic generation system executes the following sub-steps: C5.1 Constructing a Local Inversion Model: The automatic generation system extracts all elements, nodes, and constraint information within a radius of R_local (e.g., 20 mm) centered on the measurement point N_sim from the complete finite element model completed in steps S201-S202, forming an independent, smaller-scale local finite element model. This local finite element model inherits the boundary conditions of the global model and applies the same concentrated force load F_test at node N_sim. The initial value of the material elastic modulus of this local finite element model is set to E_old.

[0127] C5.2 defines and solves a one-dimensional optimization problem: the objective is to find an elastic modulus E such that the predicted displacement U(E) of the local finite element model is infinitely close to the measured displacement D_actual. The error function is defined as e(E) = U(E) - D_actual, and the optimization objective is to minimize e(E)^2.

[0128] One-dimensional optimization is performed using the golden section search method (because it is stable and does not require gradient calculation): a. Determine the search interval [E_low, E_high], for example [0.5*E_old, 2.0*E_old].

[0129] b. Take two points within the interval: E_a = E_low + 0.382*(E_high - E_low) and E_b = E_low + 0.618*(E_high - E_low).

[0130] c. Calculate e(E_a)^2 and e(E_b)^2 respectively (the local model needs to be solved once for each calculation).

[0131] d. If e(E_a)^2 < e(E_b)^2, then let E_high = E_b; otherwise let E_low = E_a.

[0132] e. Repeat step bd until the interval length (E_high - E_low) < tolerance_E (e.g., 1 MPa).

[0133] f. Take the midpoint of the final interval as the optimal solution, i.e., the elastic modulus correction ΔE.

[0134] C5.3 Calculate and update material parameters: Calculate the new elastic modulus E_new = E_old + ΔE.

[0135] C5.4 Update Material Database: The system automatically generates and maintains a material batch parameter database. This database uses the "production batch number" of the PI film material as the primary key. The system uses the Batch_ID of the PI film material used in the current task as the query key and the calculated E_new as the value to update or insert into this database. Subsequently, any simulation task specifying the use of the Batch_ID batch material will read E_new from this database as the input parameter when running the mechanical simulations in S201-S202.

[0136] Step C6 (Closed-loop Effectiveness): After completing the above adjustments and calibrations, the internal model of the automatic generation system has been updated. For the update to take effect, the automatic generation system needs to apply the new knowledge to subsequent tasks. When a new printing screen design task arrives, the automatic generation system first extracts the feature vector F_new from its original design image according to steps S1021-S1022. The automatic generation system calculates the similarity between F_new and the feature vector F_current of the previously adjusted task. Specifically, cosine similarity is used for calculation: similarity = (F_new · F_current) / (||F_new|| * ||F_current||); where · represents the vector dot product, and ||·|| represents the Euclidean norm of the vector. The automatic generation system presets a feature similarity threshold S_feat (e.g., 0.85). If similarity ≥ S_feat, the new task is determined to have "similar features" to the historical task.

[0137] If a new task is determined to be similar to a historical task F_current, the automatic generation system, when performing inference using the model from steps S1021-S1022, will prioritize querying the updated OffsetMap for its internal offset value. Specifically, the automatic generation system queries the OffsetMap using (F_new, t_material) as the key. If a record exists (i.e., a direct match), that value is used directly; otherwise, it searches for F' (based on cosine similarity) that is most similar to F_new, and uses the value of OffsetMap[(F', t_material)] as the basis, then outputs it after model fine-tuning.

[0138] If the PI film material batch number specified in the new task is the same as that of the historical task that has completed calibration, the system will directly call the calibrated elastic modulus E_new corresponding to that batch number as the input parameter when running the mechanical simulation of S201-S202.

[0139] Optionally, the automatic generation method may also include the following steps: S401, before outputting the PI film filling pattern, a pattern normalization operation is performed on the PI film filling pattern; the pattern normalization operation is used to eliminate non-manufacturable features in the pattern, which at least include isolated pattern areas or holes, and inner sharp corners.

[0140] In practice, after generating the PI film filling pattern but before outputting it, the automatic generation system automatically executes a pattern normalization and repair process, including the following steps: Step D1 (Graphic Analysis and Contour Classification): The automatic generation system reads the DXF format file of the PI film-filled graphic. The system iterates through each LWPOLYLINE or POLYLINE entity in the file, extracting the (X, Y) coordinates of all vertices sequentially from their vertex data segments, forming an ordered vertex coordinate sequence V = [(x1, y1), (x2, y2), ..., (xn, yn)]. For closed entities, the last vertex of the sequence has the same coordinates as the first vertex.

[0141] For each polygon defined by the vertex sequence V, the automatic generation system uses the shoelace formula to calculate its directed area S = 0.5 * Σ_{i=1}^{n-1} (x_i*y_{i+1} - x_{i+1}*y_i), where (x_{n+1}, y_{n+1}) is equivalent to (x_1, y_1).

[0142] If the calculated directed area S > 0, then the sequence of polygon vertices is determined to be a counterclockwise loop and identified as the "outer contour of the graphic".

[0143] If S < 0, it is determined to be a clockwise wrap and identified as "the outline of a hole inside the graphic".

[0144] The system organizes all contours in a tree-like data structure: each polygon identified as an outer contour is the root node, and all polygons identified as holes whose centroids fall inside the outer contour polygon are added as child nodes of the root node, forming a "contour-hole" hierarchy.

[0145] Step D2 (Minor Structure Removal): The automatic generation system iterates through all polygonal contour nodes obtained in Step D1. For each node, its actual area Area = |S| (i.e., the absolute value of the directed area mentioned above). If Area < A_min, then a deletion operation is performed. If the node is an outer contour node (root node), then remove it along with all its child nodes (holes) from the data structure.

[0146] If the node is a hole node (child node), then delete the hole node record from the child node list of its parent node (the outer contour to which it belongs) to fill the hole.

[0147] Step D3 (Narrow Connectivity Merging): D3.1 (Graphics Rasterization): The automatic generation system calculates the bounding rectangle of the remaining graphics and sets the rasterized pixel size (pixel_size, e.g., 0.0254 mm, corresponding to 1000 dots per inch). A two-dimensional binary array I with a width of W_pixel (equal to the width of the bounding rectangle divided by pixel_size and rounded up) and a height of H_pixel is created, initially with all elements set to 0, representing the background color (black). For each outer contour and its internal holes in the data structure (hole areas are considered background and not filled), the automatic generation system uses a scanline seed filling algorithm: for each row (scanline) of the image, the intersection points of that row with all edges of the polygon are calculated; the X coordinates of these intersection points are sorted and paired; in the pixel segment between each pair of intersection points, the element at the corresponding position in array I is set to 1, representing the foreground color (white).

[0148] D3.2 (Distance Transform Calculation): On the binary image I, the system calculates the Euclidean distance from each foreground pixel (pixel with a value of 1) to the nearest background pixel (pixel with a value of 0), generating a distance map DT. This is implemented using a two-pass scanning algorithm. First scan (from left to right, top to bottom): For each pixel (i, j) in the image, if I[i][j] = 1, then set DT[i][j] = 0. Otherwise, DT[i][j] takes the square of min(DT[i-1][j] + 1, DT[i][j-1] + 1, DT[i-1][j-1] + √2, DT[i-1][j+1] + √2). This step uses an integer approximation distance.

[0149] The second scan (from right to left, from bottom to top): Update DT[i][j] again, taking the value as the square of min( DT[i][j],DT[i+1][j]+ 1, DT[i][j+1]+ 1, DT[i+1][j+1]+ √2, DT[i+1][j-1]+ √2 ).

[0150] Finally, take the square root of each element value in DT to obtain the precise Euclidean distance d for each pixel.

[0151] D3.3 (Skeleton Extraction and Narrow Point Identification): The automatic generation system iteratively thins the binary image I to extract its central skeleton line. In each iteration, the system examines all foreground pixels in parallel, determining whether a pixel is a deletable boundary pixel (i.e., deletion without disrupting connectivity) based on the specific black-and-white distribution pattern of its eight neighboring pixels (8-neighborhood). All deletable pixels are marked and deleted together. This iterative process is repeated until no pixels can be deleted, ultimately resulting in a skeleton image Skel with a width of only one pixel. A standard implementation of this thinning process is called the Zhang-Suen parallel thinning algorithm.

[0152] Subsequently, the automatic generation system iterates through each foreground pixel p in the Skel, with coordinates (i, j). It reads the distance value d corresponding to that location from the distance map DT. The local physical width w of the graphic at point p is then calculated as: w = 2 * d * pixel_size. If w is less than the preset minimum width threshold W_min (e.g., 0.03 mm), the coordinates of pixel p are recorded as a "narrow point".

[0153] D3.4 (Perform Morphological Closing Operation): The automatic generation system constructs a circular structuring element B with radius r, where r is calculated by rounding up according to r = ceil(W_min / (2 * pixel_size)) to ensure that the structuring element can cover narrow regions with a physical width less than W_min. Then, a morphological closing operation is performed on image I, consisting of successive dilation and erosion operations: Dilation operation: Generates an image I_dilate. For each pixel in I_dilate, if there is at least one foreground pixel (value 1) in the original image I, centered on that pixel and covered by structuring element B, then that pixel is set to 1 in I_dilate; otherwise, it is set to 0.

[0154] Erosion operation: Generate an image I_closed from the I_dilate obtained in the previous step. For each pixel in I_closed, if all pixels in the area covered by structuring element B centered on that pixel in I_dilate are foreground pixels (value 1), then that pixel in I_closed is set to 1; otherwise, it is set to 0.

[0155] This closing operation can merge narrow connections while keeping the dimensions of other areas essentially unchanged.

[0156] D3.5 (Vectorization): The automatic generation system performs contour tracing on the binary image I_closed. Specifically, it scans the image, and when it finds an unvisited foreground pixel, it uses that pixel as a starting point, following the rule of "always searching for the next unvisited foreground pixel within the 4-connected or 8-connected neighborhood of the current pixel," tracing and recording the coordinates of all pixels on the entire closed contour boundary, forming a pixel chain. For each pixel chain, the automatic generation system applies a polygon approximation algorithm: setting a distance tolerance ε (e.g., ε = pixel_size). This algorithm works recursively, retaining the endpoint pixels that have the greatest impact on the overall shape of the chain, while deleting intermediate pixels whose vertical distance to the chord defined by the key endpoints is less than ε. Ultimately, each pixel chain is simplified to an ordered set of vertices, forming a vector polygon. The system updates its internal data structure with these new polygons.

[0157] Step D4 (Perform Inner Sharp Corner Smoothing): The automatic generation system iterates through the vertex list of each contour in the current vector graphics. For vertex v_i, calculate vector e_prev = v_i - v_{i-1}, e_next = v_{i+1} - v_i. Calculate the angle θ between these two edge vectors using the formula: θ = arccos((dot product of e_prev and e_next) / (length of vector e_prev * length of vector e_next)). If the calculated angle θ is less than the preset minimum angle threshold θ_min (e.g., 15°), then v_i is determined to be an inner sharp corner that needs smoothing, and the following sub-steps are executed: D4.1: Calculate the normalized angle bisector direction vector b of the apex. First, normalize e_prev and e_next respectively, obtaining u_prev = e_prev / ||e_prev|| and u_next = e_next / ||e_next||. Then calculate b = (u_prev + u_next) / ||u_prev + u_next||.

[0158] D4.2: Calculate the center O of the arc used for smoothing. According to geometry, the center O lies in the direction of the angle bisector b, and its distance to the vertex v_i is R_fillet / sin(θ / 2), where R_fillet is the preset fillet radius. Therefore, the center coordinates O = v_i + (R_fillet / sin(θ / 2)) * b.

[0159] D4.3: Calculate the feet of the perpendiculars from the center O of the circle to the edges e_prev and e_next. These two feet are the points of tangency T_prev and T_next between the arc and the two edges. Taking the calculation of T_prev as an example: Calculate the dot product of the projection length proj = (O - v_i) of vector O - v_i onto the direction of unit vector u_prev and u_prev. Then the point of tangency T_prev = v_i + proj * u_prev. The automatic generation system needs to verify whether T_prev is indeed located on the line segment v_{i-1}v_i (i.e., the value of proj is between 0 and ||e_prev||).

[0160] D4.4: Perform graphic replacement. Replace the original polyline path v_{i-1}→v_i→v_{i+1} with a new path v_{i-1} → T_prev → (an arc from T_prev to T_next) → T_next → v_{i+1}. The arc has center O and radius R_fillet, starting at T_prev and ending at T_next.

[0161] Step D5 (Perform contour self-intersection detection and repair): The automatic generation system collects line segments of all contours in the current drawing. A scanline algorithm is used to detect intersections between all non-adjacent line segments. This algorithm uses a vertical scanline moving from left to right, maintaining a state table of line segments intersecting with the scanline, and checking whether adjacent line segments in the state table intersect near the current position of the scanline. Whenever a pair of intersecting line segments is detected, the system splits these two line segments into two shorter line segments at the intersection point.

[0162] All line segments (including original segments and newly generated segments) are treated as undirected edges. The system executes the following loop to reconstruct the closed contour: a. Start from any line segment endpoint P that has not yet been used by any cycle, and select an edge E with P as the current edge.

[0163] b. Let Q be the other endpoint of the current edge E. At point Q, find all edges with Q as their endpoint (excluding the edge E that has just been traversed). From these edges, select the one that minimizes the rotation angle when turning from the current edge E to the selected edge, and use it as the next edge. This selection rule is called the "leftmost turn rule".

[0164] c. Add the selected edge to the cycle being built and mark it as used. Update the current edge E to this newly selected edge.

[0165] d. Repeat steps b and c until you return to the starting endpoint P, thus forming a closed polygonal loop.

[0166] The loop ends when all edges have been used, resulting in a new set of closed loops.

[0167] For each newly generated closed loop, the system uses the method described in step D1 (i.e., calculate the oriented area and determine the surrounding direction based on its positive or negative sign) to determine whether the loop is an outer contour (counterclockwise) or a hole (clockwise), and reconstructs the tree-like hierarchical structure of the graphic data accordingly.

[0168] Finally, the automatic generation system will encode the repaired and reconstructed graphic data into a standard DXF format file for output.

[0169] Optionally, S103, "based on the original design image and process parameters, generating a PI film filling pattern adapted to the printing area through automated image processing," includes the following steps: S1031, Match the original design image with the pre-stored template library; wherein, the template library records the baseline screen design image, the baseline PI film filling image corresponding to the baseline screen design image, and a set of key driving dimension parameters. S1032, If the match is successful, calculate the change in key driving dimension parameters of the original design image relative to the matched screen design image; S1033, based on the change of key driving dimension parameters, the reference PI film filling pattern is parametrically constrained and deformed to obtain the PI film filling pattern; wherein, the parametric constraint is used to maintain the avoidance relationship between the PI film filling pattern and the printing area during the pattern deformation process.

[0170] In practice, when the automatic generation system receives a new original design image of the printing screen, if a rapid transformation mode is desired, the following process is executed sequentially: Step E1 (Image Preprocessing and Feature Extraction): Step E1.1: The automatic generation system reads in the original design image I_new (e.g., a binarized image, with white representing the printing area and black representing the background). First, the image size is normalized to a standard size (e.g., 1000×1000 pixels).

[0171] Step E1.2 (Calculate the eigenvector F_new of the Histogram of Oriented Gradients (HOG)): Calculate the gradient: For each pixel in the normalized image, calculate its gradient values ​​g_x and g_y in the horizontal (x) and vertical (y) directions, respectively (e.g., using a convolution kernel of [-1, 0, 1]). Calculate the gradient magnitude m = sqrt(g_x^2 + g_y^2) and the gradient direction θ = arctan(g_y / g_x) (range 0-180 degrees) for that pixel.

[0172] Cellular division and orientation histogram generation: The entire image is divided into several non-overlapping cellular units (e.g., each unit is 8×8 pixels). For each cellular unit, an orientation histogram is generated based on the gradient direction θ and magnitude m of all pixels within it. This histogram divides the 0-180 degree range into 9 directional intervals (bins). For each pixel, its magnitude m is linearly interpolated and accumulated into the two adjacent directional intervals based on its direction θ.

[0173] Intra-block normalization and feature concatenation: Adjacent 2×2 cell units are grouped into a block. The histogram values ​​of all cell units within a block are concatenated into a vector, and this vector is normalized using the L2 norm (i.e., each element in the vector is divided by the L2 norm of the vector). A sliding window is used to traverse all blocks of the image, and the normalized vectors of all blocks are concatenated sequentially to form a high-dimensional feature vector F_new.

[0174] Step E2 (Template library matching): E2.1: The template library is a database where each record (a template_k) contains three parts of data: T_design_k: A design image of a baseline screen.

[0175] T_pi_k: Vector data of the PI film filling pattern that is fully adapted to the reference screen and optimized according to claims 1-4.

[0176] T_params_k: A set of key driving dimension parameters, such as [outer frame length L0, outer frame width W0, number of main grid lines N0, spacing of fine grid lines P0]. These parameters are either manually measured from T_design_k or automatically identified and recorded when the template is created.

[0177] E2.2 (Feature Comparison and Similarity Calculation): For each template_k in the template library, the system has pre-calculated and stored the HOG feature vector F_temp_k of its baseline image T_design_k. The automatic generation system calculates the cosine similarity sim_k between F_new and each F_temp_k. The formula for calculating the cosine similarity is: sim_k = (dot product of F_new and F_temp_k) / (L2 norm of F_new * L2 norm of F_temp_k).

[0178] E2.3 (Match Determination): The automatic generation system presets a similarity threshold S_match (e.g., 0.85). If the maximum value sim_max among all sim_k is greater than S_match, a match is determined to be successful, and the template index k of sim_max is recorded. Otherwise, no matching template is determined, and the process proceeds to the general generation process S101-S104.

[0179] Step E3 (Critical Dimension Remeasurement and Variation Calculation): The automatic generation system automatically remeasures the same set of parameters defined in T_params_k of template_k on image I_new. For example: Measure the outer frame length L_new: Use the Hough transform to detect all long straight line segments in I_new. Among all detected near-horizontal straight line segments, select a pair of line segments with similar Y coordinates located on opposite sides of the image, and calculate the horizontal distance between the midpoints of these two line segments as L_new.

[0180] Measuring the fine grid line spacing P_new: Select a column of vertical fine grid lines in the center region of the image, and measure the average vertical distance between the center lines of two adjacent grid lines, which is taken as P_new.

[0181] Then, the system automatically generates a parameter list measured from I_new as Params_new = [L_new, W_new, N_new, P_new, ...]. It calculates the difference between this list and the template parameter T_params_k, obtaining the change vector ΔParams = Params_new - T_params_k.

[0182] Step E4 (parametric graphic deformation): When the template_k is imported into the library, a set of control points C_k = {c_1, c_2, ..., c_m} are defined on its PI membrane graph T_pi_k, and linear hybrid skinning weights w_{j1}, w_{j2}, ..., w_{jm} relative to these control points are calculated for each vertex v_j on the graph. The weights satisfy Σ_i w_{ji} = 1 and w_{ji}≥0. The weight matrix W is obtained by solving the following least squares problem: with all vertex coordinates V_k as observations and control point coordinates C_k as unknowns, minimize ||V_k - W * C_k||^2, while satisfying the constraint of the weight sum.

[0183] The automatic generation system calculates the new position C_new of the control points based on ΔParams and preset driving rules. For example, the rule stipulates that if ΔL ≠ 0, the X-coordinate of all control points located on the left and right boundaries of the graphic should be scaled by a factor of L_new / L0. Similar rules are applied to other parameter changes.

[0184] According to the linear blending skin formula, the new coordinates of all vertices are calculated: V_pre = W * C_new. V_pre constitutes the PI film pattern after initial deformation.

[0185] Step E5 (Constraint Verification and Iterative Correction): Binarize the printed area in I_new. Calculate the minimum Euclidean distance d_min between the boundary of the initial deformed pattern V_pre and the boundary of the printed area in I_new. If d_min is less than the avoidance width D_avoid defined in claim 1, then initiate the correction process. The system constructs an optimization problem: Optimization variable: Position correction ΔC of control points.

[0186] Objective function: Minimize ||ΔC||^2, that is, adjust the control points as little as possible.

[0187] Constraint: The minimum distance d_min_final between the modified vertex V_final = W * (C_new + ΔC) and the printed area is greater than or equal to D_avoid.

[0188] The automatic generation system uses a quadratic programming solver (e.g., the effective set method) to solve the constrained optimization problem described above, obtaining the optimal control point correction ΔC_opt. The final control point position is C_final = C_new + ΔC_opt, and the corresponding final vertex coordinates are V_final = W * C_final.

[0189] Step E6: The automatic generation system outputs the vector graphic defined by V_final as the PI film filling graphic generated for the new I_new network version. This graphic can then be sent to the S401 normalized repair process for final processing.

[0190] This application also discloses an automatic generation system for PI film patterns used to reinforce printing screens. (See also...) Figure 2 ,include: The printing area recognition module 501 is used to acquire the original design image of the printing screen, wherein the original design image defines the printing area of ​​the printing screen; The process parameter determination module 502 is used to obtain at least one process parameter related to the PI film pattern. The PI film filling pattern generation module 503 is used to generate a PI film filling pattern adapted to the printing area based on the original design image and the process parameters through automated image processing; wherein, the automated image processing is used to ensure that the PI film filling pattern does not cover the printing area and the preset surrounding range, and to keep the pattern interrupted at the preset break position of the printing area; The PI film filling pattern output module 504 is used to output the PI film filling pattern for use in subsequent cutting processes.

[0191] This application also discloses an automatic generation device for PI film patterns for reinforcing printing screens. The automatic generation device for PI film patterns for reinforcing printing screens includes a memory and a processor. The memory stores a computer program that can be loaded by the processor and executed as described above for the automatic generation method of PI film patterns for reinforcing printing screens.

[0192] This application also discloses a computer-readable storage medium that stores a computer program that can be loaded by a processor and executed as described above in the automatic generation method for reinforcing PI film patterns for printing screens. The computer-readable storage medium includes, for example, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0193] It should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0194] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit the scope of protection of the application. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

Claims

1. A method for automatically generating PI film patterns for reinforcing printing screens, characterized in that, include: Obtain the original design image of the printing screen, wherein the original design image defines the printing area of ​​the printing screen; Obtain at least one process parameter related to the PI film pattern; Based on the original design image and the process parameters, an automated image processing method is used to generate a PI film filling pattern that is compatible with the printing area; wherein, the automated image processing is used to ensure that the PI film filling pattern does not cover the printing area and the preset surrounding area, and to maintain the pattern interruption at the preset break position of the printing area; Output the PI film filling pattern for use in subsequent cutting processes.

2. The method for automatically generating PI film patterns for reinforcing printing screens according to claim 1, characterized in that, The acquisition of at least one process parameter related to the PI film pattern includes: Call the pre-trained process parameter recommendation model; Input the original design image into the process parameter recommendation model to obtain the recommended process parameter set output by the process parameter recommendation model; The process parameter recommendation model is a machine learning model trained on historical production data. The historical production data is associated with multiple sets of printing screen design features, the set of process parameters used, and the corresponding PI film cutting quality evaluation results.

3. The method for automatically generating PI film patterns for reinforcing printing screens according to claim 2, characterized in that, The method further includes: After generating the PI film filling pattern, based on the original design image and the recommended process parameter set, virtual bonding and mechanical simulation are performed on the PI film filling pattern to predict the deformation and stress distribution of the printing screen after bonding the PI film filling pattern. Based on the results of mechanical simulation, the PI film filling pattern is adaptively adjusted; wherein, the adaptive adjustment includes: increasing the local support strength of the PI film filling pattern in areas where the predicted deformation or stress exceeds a preset threshold.

4. The method for automatically generating PI film patterns for reinforcing printing screens according to any one of claims 1-3, characterized in that, The subsequent cutting process is laser cutting, and the method further includes: The manufacturability of the PI film filling pattern is optimized for laser cutting process; wherein the manufacturability optimization includes at least one of the following optimization objectives: minimizing the total length of the laser cutting path of the PI film filling pattern; and ensuring that the estimated heat load distribution of the PI film filling pattern in the simulated laser cutting process meets a preset uniformity condition.

5. The method for automatically generating PI film patterns for reinforcing printing screens according to claim 3, characterized in that, The method further includes: The online quality inspection data of the PI film obtained by laser cutting based on the PI film filling pattern, and / or the printing screen with the PI film attached, are obtained; wherein the online quality inspection data includes at least one of the following: the measured value of the key dimensions of the PI film, and the measured value of the deformation of the printing screen under a preset load. Based on the deviation between the measured values ​​of the key dimensions and the design values ​​of the corresponding positions in the PI film filling pattern, and / or based on the deviation between the measured values ​​of deformation and the corresponding deformation values ​​predicted by the mechanical simulation, the recommended process parameter model and / or the material mechanical parameters used in the mechanical simulation are adaptively adjusted.

6. The method for automatically generating PI film patterns for reinforcing printing screens according to claim 1, characterized in that, The method further includes: Before outputting the PI film filling pattern, a pattern normalization operation is performed on the PI film filling pattern; the pattern normalization operation is used to eliminate non-manufacturable features in the pattern, the non-manufacturable features include at least isolated pattern areas or holes, and inner sharp corners.

7. The method for automatically generating PI film patterns for reinforcing printing screens according to claim 1, characterized in that, The step of generating a PI film filling pattern adapted to the printing area through automated image processing, based on the original design image and the process parameters, includes: The original design image is matched with a pre-stored template library; wherein the templates in the template library record a baseline screen design image, a baseline PI film filling image corresponding to the baseline screen design image, and a set of key driving dimension parameters; If a match is successful, the change in key driving dimension parameters of the original design image relative to the matched screen printing design image is calculated. Based on the change in the key driving dimension parameters, the reference PI film filling pattern is parametrically constrained and deformed to obtain the PI film filling pattern; wherein, the parametric constraint is used to maintain the avoidance relationship between the PI film filling pattern and the printing area during the pattern deformation process.

8. An automatic generation system for PI film patterns used for reinforcing printing screens, characterized in that, include, The printing area recognition module (501) is used to acquire the original design image of the printing screen, wherein the original design image defines the printing area of ​​the printing screen; The process parameter determination module (502) is used to obtain at least one process parameter related to the PI film pattern; The PI film filling pattern generation module (503) is used to generate a PI film filling pattern that matches the printing area based on the original design image and the process parameters through automated image processing; wherein, the automated image processing is used to ensure that the PI film filling pattern does not cover the printing area and the preset surrounding range, and to keep the pattern interrupted at the preset break position of the printing area; The PI film filling pattern output module (504) is used to output the PI film filling pattern for use in subsequent cutting processes.

9. An automatic generation device for PI film patterns used to reinforce printing screens, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 7.

Images