Process evaluation method and system for spatiotemporal traceability of magnetron sputtering web coating processes
By constructing a knowledge graph of coating defect causes and a Bayesian network, the problem of accurate identification and traceability of defects in the production of flexible thermal control films for spacecraft was solved, and efficient process evaluation and production optimization were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI YUDA INDUSTRIAL CO LTD
- Filing Date
- 2026-02-04
- Publication Date
- 2026-06-19
Smart Images
Figure CN122243257A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of thermal control thin film technology for spacecraft, and more specifically, to a spatiotemporally traceable process evaluation method and system for magnetron sputtering roll-to-roll coating. Background Technology
[0002] In the fabrication of flexible thermal control thin films for spacecraft, frequent defects such as incomplete plating, cracking, and electrical sparks can easily lead to batch scrapping. The traditional "manual visual inspection + offline" testing mode can no longer meet the requirements of high-quality development and low-cost manufacturing. Therefore, achieving accurate defect identification, in-depth analysis, and spatiotemporal tracing has become a core demand that the industry urgently needs to address.
[0003] In existing technologies, there have been attempts to detect the surface defects of metal sheets. For example, Chinese patent application number CN202211174874.X discloses a method for determining periodic defects on the surface of tin-plated steel sheets. It achieves rapid detection of periodic defects by combining an improved YOLOv5s convolutional neural network with semi-supervised learning and data augmentation. Although this technology utilizes the Backbone, Neck, and Head structure to improve feature extraction and classification capabilities, effectively solving the problem of "discovering" surface defects, for extremely complex production processes with highly coupled parameters, such as flexible thermal control films for spacecraft, simple image recognition often fails to reach the underlying mechanism of "why defects occur" and is difficult to handle the tracing of non-periodic, complex, and sudden defects.
[0004] Based on this, this solution, while leveraging data augmentation techniques to enhance model robustness, further incorporates data mining and knowledge graph technologies. Data mining extracts hidden patterns from massive datasets using statistical and machine learning methods, providing support for mitigating sample imbalance and reducing overfitting. The core technology of this solution lies in using a knowledge graph as the underlying architecture, which not only describes the characteristics of things but also excels at handling complex relationships. Through technological breakthroughs in representation learning and relational reasoning, a deep mapping mechanism is constructed based on the comprehensive process parameters of the magnetron sputtering roll-to-roll coating system and product defects (single / compound forms) such as incomplete coating, cracking, and electrical discharge machining.
[0005] Building upon this foundation, this method utilizes an expanded dataset to conduct quality tracing research based on a knowledge graph of coating defect causes. A Bayesian network-based coating defect tracing model is constructed, forming an evolvable knowledge base. By revealing the inherent spatiotemporal relationships between various parameters in the production process, a spatiotemporally traceable and efficient process evaluation method is ultimately proposed, achieving a technological leap from "defect detection" to "cause diagnosis and process evaluation." Summary of the Invention
[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a spatiotemporally traceable process evaluation method and system for magnetron sputtering roll-to-roll coating.
[0007] According to one aspect of the present invention, a spatiotemporally traceable process evaluation method for magnetron sputtering roll-to-roll coating includes:
[0008] Step S1: Collect data on the original defects; Step S2: Randomly transform the defect data using data augmentation methods to generate new defect data samples. Analyze the defect data using data mining techniques to enhance the scale and diversity of the defect dataset and generate a defect sample set. Step S3: Based on the defect sample set, summarize the causes of defects in the coating process and construct a quality traceability model based on the knowledge graph of coating defect causes.
[0009] Preferably, in step S1, the original defects include plating defects, electrical discharge defects, holes, perforations, scorching, vertical streaks, crystal points, and cracks.
[0010] Preferably, in step S1, the methods for collecting data for different types of defects include: Plating defects: Apply localized lighting and use high-contrast settings to show the location and size of the plating defects; Electrical spark defects: Shoot under low light conditions to highlight the discharge trajectory of the electrical spark; Holes, perforations, and charring defects: The location and shape of the burn points are displayed using lighting of preset intensity; Vertical stripe defects: Use uniform lighting and ensure that the shot is taken perpendicularly to show the direction of the stripes; Crystal point defects: Select sufficient lighting conditions for shooting to make the small crystal point defects distributed on the edge more prominent; Breakage defects: Increasing the film roll tension makes the breakage defects more obvious.
[0011] Preferably, in step S1, the following conditions are met when collecting the original defects: uniform lighting equipment or soft light source is used, and the light is sufficient and evenly distributed; a high-resolution industrial camera is used, and appropriate focal length, exposure time and white balance are set; the defect sample is clean and the surface is smooth; the sample shooting angle includes front, side and oblique side.
[0012] Preferably, step S1 further includes: recording information for each defect sample and establishing a complete defect database, wherein the defect sample information includes defect type, collection date, and defect location.
[0013] Preferably, in step S2, the data augmentation method includes random rotation, mirroring, symmetry, blurring, brightness adjustment, contrast adjustment, saturation adjustment, and adding Gaussian noise, wherein: rotation is used to change the image orientation, adjust the background in the image, or remove unwanted elements; mirroring includes horizontal and vertical mirroring, used to change the image orientation while maintaining its shape and features; symmetry includes horizontal symmetry, vertical symmetry, and central symmetry, used to make the content on both sides of the image consistent relative to the center of symmetry; blurring is used to make the image more blurred, reduce noise and details in the image, and eliminate interference information in the image; changing saturation is used to change the color vibrancy of the image; changing brightness is used to change the brightness of the image; adjusting contrast is used to change the brightness difference of the image; increasing contrast is used to make the details of certain defects in the image clearer; adding Gaussian noise is used to simulate natural noise in the image by increasing random pixel value changes to change the visual effect, making the image more realistic or increasing its complexity.
[0014] Preferably, the quality traceability model is used for: Determine the type of defect: Based on the quality diagnostic report, determine the type of defect.
[0015] Determine the set of relational paths: Find all paths leading to the occurrence of the target defect in the knowledge graph of coating defect causes; Calculate the probability of a path leading to a defect: For each relational path, calculate the probability score of the target path leading to the target defect by referring to the spatiotemporal matching data. The spatiotemporal matching data includes various key parameters recorded during the coating process and the corresponding time and space information of the key parameters. Path ranking: Paths are ranked according to their scores to determine their importance and relevance. Paths with higher probability scores are considered to be more relevant to quality defects.
[0016] Application path sorting results: Based on the path sorting results, the top three paths with the highest probability are selected as the quality traceability results for display.
[0017] Preferably, the calculation of the probability score of the target path causing the target defect by comparing the spatiotemporal matching data specifically includes: Based on the spatiotemporal matching data of the target defect coating, obtain the set of events that led to the occurrence of the target defect. ; The probability score of the target path is calculated using the following formula:
[0018]
[0019]
[0020] in, This indicates the association weight between target events and target defects in the target path. Indicates events in the target path. The parentheses represent Iverson brackets, where the condition inside the brackets is 1 if satisfied and 0 if not satisfied.
[0021] Preferably, the step of determining the set of relational paths specifically involves finding paths on the knowledge graph using the Cypher language.
[0022] According to another aspect of the present invention, a spatiotemporally traceable process evaluation system for magnetron sputtering roll-to-roll coating is characterized by comprising: Module M1: Collects data on the original defects; Module M2: This module performs random transformations on defect data using data augmentation methods to generate new defect data samples. It then analyzes the defect data using data mining techniques to enhance the scale and diversity of the defect dataset and generate a defect sample set. Module M3: Based on the defect sample set, summarize the causes of defects in the coating process and construct a quality traceability model based on the knowledge graph of coating defect causes.
[0023] Compared with the prior art, the present invention has the following beneficial effects: First, it can very intuitively display the causal relationship between events (nodes).
[0024] Second, it allows for bidirectional reasoning, meaning you can reason from cause to effect or from effect to cause.
[0025] Third, new evidence can be used to overturn previous reasoning.
[0026] Fourth, by monitoring changes in key indicators, potential defects and risks can be identified in a timely manner, and corresponding measures can be taken for adjustment and optimization.
[0027] Fifth, all nodes are visible, which facilitates the analysis of data related to coating defects, reveals the relationships between different factors, and identifies the key factors and patterns that lead to defects.
[0028] In summary, the spatiotemporally traceable process evaluation method for magnetron sputtering roll-to-roll coating based on digital twins described in this invention reveals the intrinsic connections between various production processes and process parameters, analyzes and summarizes the causes and potential patterns of coating defects, solves the problem of timely control of coating defects in traditional production lines, improves the accuracy of coating defect quality traceability, and provides strong support for high-quality, high-efficiency, and high-stability production. It is of great significance for the mass production, diversification, and high-quality manufacturing of flexible thermal control films for spacecraft. Attached Figure Description
[0029] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 The diagram illustrates the architecture of the quality traceability model. Figure 2 A portion of the data augmentation set of an embodiment is shown schematically; Figure 3 A schematic diagram illustrating the entire path leading to the defect in the embodiment is shown. Figure 4 The quality traceability results of the embodiment are illustrated schematically.
[0030] Figure 5 The image is located at X1 in Table 1.
[0031] Figure 6 The image shown is located at X2 in Table 2.
[0032] Figure 7 The image shown is located at X3 in Table 2.
[0033] Figure 8 The image shown is located at X4 in Table 2.
[0034] Figure 9 The image shown is located at X5 in Table 2.
[0035] Figure 10 The image shown is located at X6 in Table 2.
[0036] Figure 11 The image shown is located at X7 in Table 2. Detailed Implementation
[0037] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.
[0038] Example 1: This embodiment discloses a spatiotemporally traceable process evaluation method for magnetron sputtering roll-to-roll coating, including: Step S1: Collect data on the original defects; Step S2: Randomly transform the defect data using data augmentation methods to generate new defect data samples. Analyze the defect data using data mining techniques to enhance the scale and diversity of the defect dataset and generate a defect sample set. Step S3: Based on the defect sample set, summarize the causes of defects in the coating process and construct a quality traceability model based on the knowledge graph of coating defect causes.
[0039] In this embodiment, in step S1, the original defects include plating defects, electrical discharge defects, holes, perforations, scorching, vertical streaks, crystal points, and cracks.
[0040] In this embodiment, step S1 includes the following methods for collecting data on different types of defects: Plating defects: Apply localized lighting and use high-contrast settings to show the location and size of the plating defects; Electrical spark defects: Shoot under low light conditions to highlight the discharge trajectory of the electrical spark; Holes, perforations, and charring defects: The location and shape of the burn points are displayed using lighting of preset intensity; Vertical stripe defects: Use uniform lighting and ensure that the shot is taken perpendicularly to show the direction of the stripes; Crystal point defects: Select sufficient lighting conditions for shooting to make the small crystal point defects distributed on the edge more prominent; Breakage defects: Increasing the film roll tension makes the breakage defects more obvious.
[0041] In this embodiment, in step S1, the following conditions are met when collecting the original defect: Use uniform lighting equipment or soft light sources to ensure sufficient and evenly distributed illumination; Use a high-resolution industrial camera and set the appropriate focal length, exposure time, and white balance. The defective sample is clean and has a smooth surface; The sample was photographed from the front, side, and oblique angles.
[0042] In this embodiment, step S1 further includes: recording information for each defect sample and establishing a complete defect database, wherein the defect sample information includes defect type, collection date, and defect location.
[0043] In this embodiment, step S2, the data augmentation method includes random rotation, mirroring, symmetry, blurring, changing brightness, adjusting contrast, changing saturation, and adding Gaussian noise. Specifically: rotation is used to change the image orientation, adjust the background in the image, or remove unwanted elements; mirroring includes horizontal and vertical mirroring, used to change the image orientation while maintaining its shape and features; symmetry includes horizontal, vertical, and central symmetry, used to make the content on both sides of the image consistent relative to the center of symmetry; blurring is used to make the image more blurred, reduce noise and details in the image, and eliminate interfering information in the image; changing saturation is used to change the vividness of the image colors; changing brightness is used to change the brightness of the image; adjusting contrast is used to change the brightness difference in the image; increasing contrast is used to make the details of certain defects in the image clearer; adding Gaussian noise is used to simulate natural noise in the image by increasing random pixel value changes to change the visual effect, making the image more realistic or increasing its complexity.
[0044] In this embodiment, the quality traceability model is used for: Determine the type of defect: Based on the quality diagnostic report, determine the type of defect.
[0045] Determine the set of relational paths: Find all paths leading to the occurrence of the target defect in the knowledge graph of coating defect causes; Calculate the probability of a path leading to a defect: For each relational path, calculate the probability score of the target path leading to the target defect by referring to the spatiotemporal matching data. The spatiotemporal matching data includes various key parameters recorded during the coating process and the corresponding time and space information of the key parameters. Path ranking: Paths are ranked according to their scores to determine their importance and relevance. Paths with higher probability scores are considered to be more relevant to quality defects.
[0046] Application path sorting results: Based on the path sorting results, the top three paths with the highest probability are selected as the quality traceability results for display.
[0047] In this embodiment, calculating the probability score of the target path causing the target defect by comparing the spatiotemporal matching data specifically includes: Based on the spatiotemporal matching data of the target defect coating, obtain the set of events that led to the occurrence of the target defect. ; The probability score of the target path is calculated using the following formula:
[0048]
[0049]
[0050] in, This indicates the association weight between target events and target defects in the target path. Indicates events in the target path. The parentheses represent Iverson brackets, where the condition inside the brackets is 1 if satisfied and 0 if not satisfied.
[0051] In this embodiment, the step of determining the set of relational paths specifically involves finding paths on the knowledge graph using the Cypher language.
[0052] Example 2: This explanation uses a common coating defect, incomplete coating, as an example. Other defects are also applicable to this method. For details on the quality traceability model architecture, please refer to [link / reference needed]. Figure 1 The specific steps are as follows: Step S100: Using a high-resolution industrial camera under sufficient and uniform lighting, select an appropriate angle, focal length and exposure time to collect data on the original defects.
[0053] Specifically, the operational requirements for step S100 include: For the acquisition of original defects, uniform lighting equipment or soft light sources should be used, ensuring sufficient and evenly distributed illumination. A high-resolution industrial camera should be used, with appropriate focal length, exposure time, and white balance settings to ensure color accuracy and consistency of the image. Defect samples should be clean and have smooth surfaces. Sample shooting angles should include front, side, and oblique angles to help eliminate interference factors. Detailed records should be kept for each defect sample, including defect type, acquisition date, and defect location, establishing a complete defect database. Furthermore, different methods should be used to highlight defect characteristics for different types of defects. For example, incomplete plating defects require localized lighting and high-contrast settings to display the location and size of the incomplete plating. The final original acquisition data is shown in the table below.
[0054] Table 1 Defects in Magnetron Sputtering Wrap-in Coating
[0055] Step S200: By using data augmentation methods, the defect data is randomly transformed to generate new data samples. Based on data mining techniques, the defect data is analyzed to further enhance the scale and diversity of the defect dataset.
[0056] Specifically, through verification, changing the brightness has little effect on improving the accuracy of the model during training for plating defects. Therefore, data augmentation for plating defects is performed using rotation, saturation adjustment, and blurring. Rotation can change the image orientation, adjust the background, or remove unwanted elements; changing the saturation alters the color vibrancy, helping to better highlight defect information; blurring makes the image more indistinct, reducing noise and details, and eliminating interfering information. The data augmentation methods and effects are shown in the table, and the data augmentation set is as follows: Figure 2 As shown.
[0057] Table 2. Methods and Effects of Expanding Data on Plating Defects
[0058] Step S300: Summarize the causes of defects in the coating process and construct a quality tracing model based on a knowledge graph of coating defect causes. This quality tracing model is based on a path ranking algorithm. Through this model, the importance of different quality event paths in the knowledge graph can be assessed, thus helping to identify critical paths and relevant nodes in the quality tracing process. The input data for the quality tracing model includes spatiotemporal matching data from the coating process and corresponding quality diagnostic reports. The spatiotemporal matching data includes various key parameters recorded during the coating process, such as average particle bombardment velocity, coating roller tension, condenser inlet water temperature, vacuum level, substrate defects, actual operating conditions, and related temporal and spatial information.
[0059] Specifically, step S300 includes the following steps: Step S310: When a defect with the defect category label of incomplete plating occurs, it is passed as evidence to the inference engine. The probability distribution of each attribute feature variable in the case of incomplete plating defect is queried, and the results are shown in the table.
[0060] Table 3. Probability distribution of characteristic variables under plating defects.
[0061] Step S320: Use Cypher language to find all paths leading to the defect on the coating defect cause knowledge graph. When the input spatiotemporal matching data and quality diagnosis report of the coating production process indicate a defect type of incomplete coating, this method can obtain all paths leading to the defect. There are a total of 23 defect cause paths, see details. Figure 3 .
[0062] Step S330: For each relationship path, compare the input data with the events in the path and calculate the probability that the path causes a defect. The calculation method is as follows: Based on the spatiotemporal matching data of the target defect coating, obtain the set of events that led to the occurrence of the target defect. ; The probability score of the target path is calculated using the following formula:
[0063]
[0064]
[0065] in, This indicates the association weight between target events and target defects in the target path. Indicates events in the target path. The parentheses represent Iverson brackets, where the condition inside the brackets is 1 if satisfied and 0 if not satisfied.
[0066] Table 4 Probability of Path for Missing Plating Defect
[0067] Step S340: Path Ranking. Paths are ranked according to their scores to determine their importance and relevance. Paths with higher probability scores are considered more relevant to quality defects.
[0068] Step S350: Apply path ranking results. Based on the path ranking results, select the top three paths with the highest probabilities as the quality traceability results for visualization. See details. Figure 4 .
[0069] The present invention also provides a spatiotemporally traceable process evaluation system for magnetron sputtering roll-to-roll coating. The spatiotemporally traceable process evaluation system for magnetron sputtering roll-to-roll coating can be implemented by executing the process steps of the spatiotemporally traceable process evaluation method for magnetron sputtering roll-to-roll coating. That is, those skilled in the art can understand the spatiotemporally traceable process evaluation method for magnetron sputtering roll-to-roll coating as a preferred embodiment of the spatiotemporally traceable process evaluation system for magnetron sputtering roll-to-roll coating.
[0070] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.
[0071] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.
Claims
1. A process evaluation method for spatiotemporal traceability of a magnetron sputter web-coating process, characterized in that include: Step S1: Collect data on the original defects; Step S2: Randomly transform the defect data using data augmentation methods to generate new defect data samples. Analyze the defect data using data mining techniques to enhance the scale and diversity of the defect dataset and generate a defect sample set. Step S3: Based on the defect sample set, summarize the causes of defects in the coating process and construct a quality traceability model based on the knowledge graph of coating defect causes.
2. The method of claim 1, wherein, In step S1, the original defects include plating defects, electrical discharge defects, holes, perforations, scorching, vertical streaks, crystal points, and cracks.
3. The method of claim 2, wherein, In step S1, the methods for collecting data for different types of defects include: Plating defects: Apply localized lighting and use high-contrast settings to show the location and size of the plating defects; Electrical spark defects: Shoot under low light conditions to highlight the discharge trajectory of the electrical spark; Holes, perforations, and charring defects: The location and shape of the burn points are displayed using lighting of preset intensity; Vertical stripe defects: Use uniform lighting and ensure that the shot is taken perpendicularly to show the direction of the stripes; Crystal point defects: Select sufficient lighting conditions for shooting to make the small crystal point defects distributed on the edge more prominent; Breakage defects: Increasing the film roll tension makes the breakage defects more obvious.
4. The method of claim 1, wherein, In step S1, the following conditions must be met when collecting the original defects: Use uniform lighting equipment or soft light sources to ensure sufficient and evenly distributed illumination; Use a high-resolution industrial camera and set the appropriate focal length, exposure time, and white balance. The defective sample is clean and has a smooth surface; The sample was photographed from the front, side, and oblique angles.
5. The method according to claim 1, characterized in that, Step S1 further includes: recording information for each defect sample and establishing a complete defect database, wherein the defect sample information includes defect type, collection date, and defect location.
6. The method according to claim 1, characterized in that, In step S2, the data augmentation method includes random rotation, mirroring, symmetry, blurring, brightness adjustment, contrast adjustment, saturation adjustment, and adding Gaussian noise. Specifically: rotation is used to change the image orientation, adjust the background, or remove unwanted elements; mirroring includes horizontal and vertical mirroring, used to change the image orientation while maintaining its shape and features; symmetry includes horizontal, vertical, and central symmetry, used to make the content on both sides of the image consistent relative to the center of symmetry; blurring makes the image more indistinct, reduces noise and details, and eliminates interfering information; changing saturation alters the color vibrancy; changing brightness alters the image's brightness; adjusting contrast alters the image's brightness differences; increasing contrast makes details of certain defects in the image clearer; adding Gaussian noise simulates natural noise in the image by increasing random pixel value variations to change the visual effect, making the image more realistic or increasing its complexity.
7. The method according to claim 1, characterized in that, The quality traceability model is used for: Determine the type of defect: Based on the quality diagnostic report, determine the type of defect. Determine the set of relational paths: Find all paths leading to the occurrence of the target defect in the knowledge graph of coating defect causes; Calculate the probability of a path leading to a defect: For each relational path, calculate the probability score of the target path leading to the target defect by referring to the spatiotemporal matching data. The spatiotemporal matching data includes various key parameters recorded during the coating process and the corresponding time and space information of the key parameters. Path ranking: Paths are ranked according to their scores to determine their importance and relevance. Paths with higher probability scores are considered to be more relevant to quality defects. Application path sorting results: Based on the path sorting results, the top three paths with the highest probability are selected as the quality traceability results for display.
8. The method according to claim 7, characterized in that, The calculation of the probability score of the target path causing the target defect based on the spatiotemporal matching data specifically includes: Based on the spatiotemporal matching data of the target defect coating, obtain the set of events that led to the occurrence of the target defect. ; The probability score of the target path is calculated using the following formula: in, This indicates the association weight between target events and target defects in the target path. Indicates events in the target path. The parentheses represent Iverson brackets, where the condition inside the brackets is 1 if satisfied and 0 if not satisfied.
9. The method according to claim 7, characterized in that, In the step of determining the set of relational paths, paths are specifically found on the knowledge graph using the Cypher language.
10. A spatiotemporally traceable process evaluation system for magnetron sputtering roll-to-roll coating, characterized in that, include: Module M1: Collects data on the original defects; Module M2: This module performs random transformations on defect data using data augmentation methods to generate new defect data samples. It then analyzes the defect data using data mining techniques to enhance the scale and diversity of the defect dataset and generate a defect sample set. Module M3: Based on the defect sample set, summarize the causes of defects in the coating process and construct a quality traceability model based on the knowledge graph of coating defect causes.