Consistency detection, model training method and device, electronic equipment and storage medium

Abstract

The application discloses a consistency detection method and device, model training method and device, electronic equipment and storage medium. First, based on the initial coating measurement data, at most partially overlapping multi-frame target coating measurement data between adjacent two frames is obtained, the occurrence probability of coating defects in the target coating measurement data is increased, and the missed detection probability of the coating defects is reduced. Secondly, the target coating measurement data is classified and processed to determine the classification result corresponding to the target coating measurement data. Finally, the classification results corresponding to the target coating measurement data are summarized to generate a consistency detection result of a coating film roll. The influence of the possible error result of the classification processing on the final result of the coating consistency is reduced, so that the coating defects existing in the coating film roll can be more accurately detected, and the accuracy of the consistency detection result of the coating film roll is improved.

Description

Technical Field

[0001] This invention relates to the field of battery production equipment technology, and in particular to a consistency detection, model training method, apparatus, electronic device, and storage medium. Background Technology

[0002] The manufacturing process of lithium batteries generally includes steps such as stirring, coating, cold pressing, cutting, winding, formation, and capacity measurement. Among these, the coating process is the key step in lithium battery manufacturing. A uniformly stirred slurry is coated onto the positive and negative current collectors, and then dried, cold-pressed, slit, and die-cut to form the lithium battery electrodes. The consistency of the coating directly affects the battery's capacity, safety, and cost.

[0003] In related technologies, areal density meters are used to monitor the coated film rolls during the coating process, and the coating consistency of the coated film rolls is detected using the COV (Covariance) index. However, the accuracy of the coating consistency detection results in these related technologies needs to be improved. Summary of the Invention

[0004] In view of the above problems, this application provides a coating consistency detection, model training method, apparatus, electronic device and storage medium, which can improve the accuracy of coating roll consistency detection results in related technologies.

[0005] This application provides a coating consistency detection method, the method comprising: obtaining multiple frames of target coating measurement data based on initial coating measurement data of a coated film roll; wherein, at most, there is partial overlap between adjacent frames of target coating measurement data in the multiple frames of target coating measurement data; classifying the target coating measurement data to determine the classification result corresponding to the target coating measurement data; and summarizing the classification results corresponding to the target coating measurement data to generate a consistency detection result for the coated film roll.

[0006] In the technical solution of this application embodiment, firstly, multiple frames of target coating measurement data with at most partial overlap between adjacent frames are obtained based on initial coating measurement data, increasing the probability of coating defects appearing in the target coating measurement data and reducing the probability of missed detection of coating defects; secondly, the target coating measurement data is classified to determine the classification result corresponding to the target coating measurement data; finally, the classification results corresponding to the target coating measurement data are summarized to generate the consistency detection result of the coating film roll. This reduces the impact of possible erroneous results obtained from the classification process on the final coating consistency result, thereby enabling more accurate detection of coating defects in the coating film roll and improving the accuracy of the coating film roll consistency detection result.

[0007] In some embodiments, obtaining multiple frames of target coating measurement data based on the initial coating measurement data of the coated film roll includes: performing multiple sampling on the initial coating measurement data according to a preset step size and the size of the initial coating measurement data to obtain multiple frames of intermediate coating measurement data with at most partial overlap between adjacent frames; and generating the target coating measurement data based on the intermediate coating measurement data.

[0008] The embodiments of this application can improve the quality of target coating measurement data used to detect coating consistency, and provide an accurate data basis for detecting coating consistency.

[0009] In some embodiments, the classification result is obtained by classifying the target coating measurement data using a coating defect detection model; obtaining multiple frames of target coating measurement data based on the initial coating measurement data of the coating film roll includes: performing multiple sampling on the initial coating measurement data according to the size of the input data of the coating defect detection model to obtain multiple frames of the target coating measurement data.

[0010] In this embodiment, the coating defect detection model is able to better extract features from the target coating measurement data, providing an accurate data basis for detecting coating consistency.

[0011] In some embodiments, summarizing the classification results corresponding to the target coating measurement data to generate the consistency detection result of the coated film roll includes: determining the statistical quantity of each category of data for each category in the classification results; determining the target category data that meets the preset condition of the statistical quantity in the classification results corresponding to the target coating measurement data based on the statistical quantity of each category of data; and generating the consistency detection result of the coated film roll based on the target category data.

[0012] In this embodiment, by determining the target category data that meets the preset statistical quantity condition in the classification results corresponding to the target coating measurement data based on the statistical quantity of each category data, the category data that does not meet the preset statistical quantity condition is filtered out, thereby reducing the impact of this category data on the accuracy of the consistency detection results and improving the accuracy of the detection results of the coated film roll.

[0013] In some embodiments, the target category data includes an overall defect category corresponding to the entire coated film roll and / or a local defect category corresponding to a portion of the coated film roll; the generation of the consistency detection result of the coated film roll based on the target category data includes at least one of the following:

[0014] The overall defect category is used as the overall defect detection result of the coated film roll;

[0015] The local defect detection results of the coated film roll are determined according to the local defect category; wherein, the consistency detection results include the overall defect detection results and / or the local defect detection results.

[0016] In this embodiment, the consistency of the coated film roll is first tested from the perspectives of overall defects and local defects. Then, the failure mode of the coated film roll defects is comprehensively evaluated by the overall defect test results and the local defect test results, thereby improving the accuracy of defect detection and analysis.

[0017] In some embodiments, the classification result further includes the detection location corresponding to the local defect category; before determining the local defect detection result of the coated film roll according to the local defect category, the method further includes: determining the actual location of the local defect category in the coated film roll according to the detection location corresponding to the local defect category; wherein, the local defect detection result includes the actual location and the local defect category.

[0018] In this embodiment, by determining the actual position of the local defect category in the coated film roll based on the detection position corresponding to the local defect category, not only is the local defect category detected, but the position information of the local defect category is also accurately detected, thereby realizing the failure mode of automatic classification of consistent defects.

[0019] In some embodiments, the method further includes: searching a defect cause graph database based on the defect category in the consistency detection result to obtain the defect generation path between the defect category and the coating equipment operating parameters.

[0020] In this embodiment, the defect generation path can be automatically generated by the defect category in the consistency detection result, which shortens the coating weight defect resolution cycle and improves the overall consistency level of film roll products.

[0021] In some embodiments, the method further includes: acquiring unconfirmed operating parameters of the coating equipment related to influencing factors on the defect generation path; acquiring normal operating parameters of the coating equipment when producing qualified coated film rolls; and determining target abnormal factors causing consistency defects among the influencing factors on the defect generation path based on the comparison results between the unconfirmed operating parameters and the normal operating parameters.

[0022] In this embodiment, by identifying the target anomaly causing the consistency defect among the influencing factors along the defect generation path, the FTA model can be used to recommend equipment adjustment operations based on the detected category of coated film roll weight defect. On the one hand, suggestions are pushed to production operators for manual equipment; on the other hand, parameter adjustments are sent to the equipment controller for automated equipment.

[0023] This application provides a training method for a coating defect detection model. The method includes: obtaining multiple frames of coating measurement data samples based on historical coating measurement data of a coated film roll; wherein, there is at most partial overlap between coating measurement data samples of two adjacent frames in the multiple frames of coating measurement data samples; each frame of coating measurement data sample corresponds to a label; inputting each frame of coating measurement data sample into an initial detection model for classification processing to obtain a classification result corresponding to each frame of coating measurement data sample; updating the initial detection model according to the classification result and the label corresponding to each frame of coating measurement data sample to obtain the coating defect detection model.

[0024] The technical solution of this application embodiment realizes the training of a deep learning model for coating measurement data, and obtains a coating defect detection model for detecting defects. The coating defect detection model can automatically detect defect categories.

[0025] In some embodiments, obtaining multi-frame coating measurement data samples based on historical coating measurement data of the coated film roll includes: performing multiple sampling on the historical coating measurement data according to a preset step size and the size of the historical coating measurement data to obtain multiple frames of intermediate coating measurement data with at most partial overlap between adjacent frames; and generating the coating measurement data samples based on the intermediate coating measurement data.

[0026] In this embodiment, multiple frames of coating measurement data samples are obtained based on historical coating measurement data of the coating film roll, which expands the sample of defect categories with small data volume, reduces the impact of sample imbalance on model training effect, and improves sample balance.

[0027] In some embodiments, obtaining multiple frames of coating measurement data samples based on historical coating measurement data of the coated film roll includes: performing multiple sampling on the historical coating measurement data according to the size of the input data of the initial detection model to obtain multiple frames of the coating measurement data samples.

[0028] In this embodiment, historical coating measurement data is multisampled based on the input data size of the initial detection model to obtain coating measurement data samples whose shape matches that of the initial detection model. This reduces the limitations imposed by historical coating measurement data on the initial detection model, improves the feature extraction capability of the initial detection model, and facilitates the training of the initial detection model.

[0029] In some embodiments, the method for determining the label includes: visualizing the historical coating measurement data to obtain a coating consistency distribution map; determining the category data corresponding to the historical coating measurement data when a category labeling operation occurs based on the coating consistency distribution map; and generating a label for each frame of coating measurement data sample based on the category data corresponding to the historical coating measurement data.

[0030] In this embodiment of the application, the historical coating measurement data is visualized to obtain a coating consistency distribution map, and in response to the coating consistency distribution map having a category labeling operation, the category data corresponding to the historical coating measurement data is determined to generate a label for each frame of coating measurement data sample.

[0031] In some embodiments, the category data includes an overall defect category corresponding to the entire coated film roll and / or a local defect category corresponding to a portion of the coated film roll; the coating consistency distribution map includes a heat map of coating measurement data; determining the category data corresponding to the historical coating measurement data when a category labeling operation occurs based on the coating consistency distribution map includes at least one of the following: determining the overall defect category corresponding to the historical coating measurement data when an overall labeling operation occurs on the coating measurement data heat map; or determining the local defect category corresponding to the historical coating measurement data when a local labeling operation occurs on the coating measurement data heat map.

[0032] In some embodiments, the local defect category corresponds to a labeled location; generating a label for each frame of coating measurement data sample based on the category data corresponding to the historical coating measurement data includes at least one of the following: using the overall defect category as the overall defect label for each frame of coating measurement data sample; or, when it is determined that the data range of the coating measurement data sample overlaps with the labeled location corresponding to the local defect category, generating local defect labels for coating measurement data samples with overlapping data ranges based on the local defect category and the labeled location.

[0033] In this embodiment, historical coating measurement data are first labeled from the perspectives of overall defects and local defects. Then, the initial detection model is trained comprehensively using overall defect labels and local defect labels to obtain a coating defect detection model that can comprehensively evaluate the failure modes of coating film roll defects, thereby improving the accuracy of defect detection and analysis.

[0034] This application provides a coating consistency testing device, the device comprising:

[0035] An initial data processing module is used to obtain multiple frames of target coating measurement data based on the initial coating measurement data of the coated film roll; wherein, at most, there is partial overlap between the target coating measurement data of two adjacent frames in the multiple frames of target coating measurement data;

[0036] The target data classification module is used to classify the target coating measurement data and determine the classification result corresponding to the target coating measurement data.

[0037] The classification result summary module is used to summarize the classification results corresponding to the target coating measurement data and generate the consistency detection results of the coated film roll.

[0038] This application provides a training device for a coating defect detection model, the device comprising:

[0039] The historical data processing module is used to obtain multiple frames of coating measurement data samples based on historical coating measurement data used to evaluate the coating consistency of the coated film roll; wherein, there is at most partial overlap between the coating measurement data samples of two adjacent frames in the multiple frames of coating measurement data samples; each frame of coating measurement data sample has a corresponding label;

[0040] The sample classification processing module is used to input the coating measurement data sample of each frame into the initial detection model for classification processing, and obtain the classification result corresponding to the coating measurement data sample of each frame;

[0041] The detection model update module is used to update the initial detection model based on the classification results and the labels corresponding to each frame of coating measurement data samples, so as to obtain the coating defect detection model.

[0042] This application provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the methods described in any of the above embodiments.

[0043] This application provides an electronic device, including: a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the method described in any of the above embodiments.

[0044] This application provides a computer program product that includes instructions that, when executed by a processor of an electronic device, implement the methods in any of the above embodiments.

[0045] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0046] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0047] Figure 1a This is a schematic diagram of the application environment for the coating consistency testing method provided according to the embodiments of this specification.

[0048] Figure 1b This is a schematic diagram of the coating consistency distribution provided according to the embodiments of this specification.

[0049] Figure 1c This is a schematic diagram of a multi-frame coating measurement data sample provided according to an embodiment of this specification.

[0050] Figure 2 This is a schematic flowchart of the coating consistency testing method provided according to the embodiments of this specification.

[0051] Figure 3a This is a flowchart illustrating the training method for the coating consistency detection method provided according to the embodiments of this specification.

[0052] Figure 3b This is a schematic diagram illustrating the test results of different models provided according to the embodiments of this specification.

[0053] Figure 4a This is a schematic flowchart of the coating consistency testing method provided according to the embodiments of this specification.

[0054] Figure 4b This is a schematic diagram of the ternary configuration provided according to the embodiments of this specification.

[0055] Figure 4c This is a schematic diagram of a tree structure for the periodic fluctuation defect categories provided according to the embodiments of this specification.

[0056] Figure 5 This is a flowchart illustrating the training method for the coating defect detection model provided according to the embodiments of this specification.

[0057] Figure 6 This is a flowchart illustrating the training method for the coating defect detection model provided according to the embodiments of this specification.

[0058] Figure 7 This is a flowchart illustrating the label determination method provided according to the embodiments of this specification.

[0059] Figure 8 This is a schematic diagram of the coating consistency testing device provided according to the embodiments of this specification.

[0060] Figure 9 This is a schematic diagram of the structure of the training device for the coating defect detection model provided according to the embodiments of this specification.

[0061] Figure 10 This is a block diagram of an electronic device provided according to an embodiment of this specification. Detailed Implementation

[0062] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0064] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0065] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0066] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0067] In some related technologies, beta-ray measurement equipment (also known as online areal density meter, online weight meter, or online thickness gauge) is used to measure the weight of the coated area on the substrate, and thresholds are set based on experience to identify defects that exceed tolerances. Specifically, coating weight control can be achieved using SPC (Statistical Process Control) methods, by setting upper and lower thresholds for coating weight to identify deviations during the coating process. Simultaneously, the consistency level of the coated film roll can be reflected through COV (Coefficient of Variation) indicators, such as calculating the mean, variance, and coefficient of variation of the coating weight, and then conducting detailed analysis and problem localization based on these indicators for different regions. However, on the one hand, due to the relatively large size of coated film rolls, the comprehensiveness and accuracy of the product defects and defect locations detected by the COV index assessment method need to be improved when evaluating the area and specific morphology of defects in the entire coated film roll product. On the other hand, during the coating process, automatic control principles can be used to control the production equipment. Some intelligent coating equipment can automatically adjust the coating head (die) based on the coating weight data. However, in the automatic control parameter tuning stage, the need to identify a more comprehensive range of coating weight defect morphologies increases the difficulty of setting compensation, resulting in a shorter debugging cycle.

[0068] In some related technologies, machine vision is used to detect defects on the coating surface. Machine vision can detect defects such as bubbles and cracks on the coating film surface, but it cannot evaluate the consistency of coating weight.

[0069] Therefore, this application provides a coating consistency detection method. First, initial coating measurement data for evaluating the coating consistency of a coated film roll is acquired. From the initial coating measurement data, multiple frames of target coating measurement data with at most partial overlap between adjacent frames are extracted. Second, the target coating measurement data are classified to determine the corresponding classification results. Finally, the classification results corresponding to the target coating measurement data are summarized to generate a consistency detection result for the coated film roll. In this application, the partial overlap between adjacent frames of target coating measurement data increases the probability of coating defects appearing in the target coating measurement data. This reduces the chance of missing coating defects when summarizing the classification results corresponding to the target coating measurement data, thus more accurately detecting coating defects present in the coated film roll.

[0070] Furthermore, to achieve intelligent detection of coating consistency defects, this application provides a training method for a coating defect detection model. First, multiple frames of coating measurement data samples are constructed based on historical coating measurement data of the coated film roll. At most, there is partial overlap between adjacent frames of coating measurement data samples, and each frame of coating measurement data samples is labeled. Second, an initial detection model is built, and the constructed coating measurement data samples and their corresponding labels are used to train the initial detection model to obtain a coating defect detection model. This not only enables intelligent detection of coating weight defects using the coating defect detection model but also achieves the goal of automatically classifying the failure modes of coating consistency defects, reducing manual workload and improving detection efficiency.

[0071] Please see Figure 1a , Figure 1a This is a schematic diagram illustrating the application environment of the coating consistency detection method provided in the scenario example of this specification. The application scenario includes a first electronic device 110 for constructing coating measurement data samples, a second electronic device 120 for training a coating defect detection model, and a third electronic device 130 for deploying the coating defect detection model.

[0072] In this scenario example, during the production process of the coated film roll, several measuring points are set on the coated film roll. The coating weight at these measuring points is detected using a β-ray probe, obtaining historical coating measurement data for the coated film roll. Further, the historical coating measurement data is visualized using a first electronic device 110 to obtain a coating consistency distribution map. This coating consistency distribution map can be displayed on the interface of the first electronic device 110. Please refer to [link to relevant documentation]. Figure 1b The coating consistency distribution map can be a coating measurement data heatmap (such as a coating weight heatmap). The horizontal and vertical axes of the coating measurement data heatmap correspond to the measurement point positions on the coating film roll, and the color changes in the coating measurement data heatmap are used to represent the coating weight at the measurement point positions.

[0073] In this scenario example, the coating consistency distribution map comprehensively displays the coating weight data of the coating roll, clearly showing the weight situation of each region of the coating roll. Based on the weight situation of each region of the coating roll, data related to coating weight defects are labeled. Specifically, the labeling methods include overall labeling for overall defects and local labeling for local defects. Overall defects can be defects relative to the entire coating roll and can include at least one of the following: U-weight, inverted U-weight, single-column heavier, single-column lighter, and periodic fluctuations. Local defects can be defects relative to a specific part of the coating roll and can include at least one of the following: single-coat area U-weight, weight tilt, edge lighter, process variation, and initial coating fluctuations. By clearly defining the defect categories related to coating consistency from both overall and local defect labeling perspectives, and further, using training samples labeled using both overall and local methods, the model can learn the defect characteristics of the coating roll, thereby enabling a comprehensive analysis of the coating roll.

[0074] In this scenario example, because the shape of the historical coating measurement data is difficult to adapt to the size of the input data of the convolutional neural network, preprocessing (such as multisample processing) is required based on the historical coating measurement data to construct coating measurement data samples. Preprocessing can also reduce the impact of imbalanced samples on model training. During preprocessing, data augmentation can be performed on samples of some categories to achieve a sample balancing strategy.

[0075] The preprocessing process of historical coating measurement data is illustrated exemplarily. The size of the historical coating measurement data is m*n; where m is the data length and n is the data width. Considering the product characteristics of the coated film roll, m is several times greater than n. Therefore, the m*n historical coating measurement data is multisampled according to a preset sampling step size s, data width n, and preset data frame length j to obtain several preliminary coating measurement data samples. In some embodiments, the size of the preliminary coating measurement data samples is n*j. Further, the preset data frame length j can be determined by the size x*x of the input data of the coating defect detection model, for example, j equals x. It should be noted that the sampling step size s can be understood as the data interval between two adjacent coating measurement data samples. The number of data frames obtained by multisampling and the overlap ratio between two adjacent coating measurement data samples can be changed by adjusting the sampling step size s. Since the input data of the coating defect detection model has a size of x*x, matrix filling is performed on the preliminary coating measurement data sample with a size of n*j to fill the dimensionality difference between the preliminary coating measurement data sample and the model's input data.

[0076] After the above data preprocessing, the model's input data is standardized to a square matrix x*x. A coated film roll is divided into multiple frames of coating measurement data samples. Please refer to... Figure 1c , Figure 1c The image shows a multi-frame sample of coating measurement data. Each frame of the coating measurement data sample has a corresponding label.

[0077] In this scenario example, the second electronic device 120 trains an initial detection model based on each frame of coating measurement data samples and their corresponding labels to obtain a coating defect detection model. This model is then deployed in the third electronic device 130. The third electronic device 130 acquires the initial coating measurement data of the coating roll to be inspected, performs multiple sampling on the initial coating measurement data, and obtains multiple frames of target coating measurement data. At most, there is partial overlap between adjacent frames of target coating measurement data. Each frame of target coating measurement data is input into the coating defect detection model for classification processing, and the model outputs the category data corresponding to each frame of target coating measurement data. The third electronic device 130 determines the statistical quantity of each category of data; and based on the statistical quantity of each category of data, it determines the target category data that meets the preset statistical quantity conditions from the classification results corresponding to the target coating measurement data. The target category data includes the overall defect category corresponding to the entire coating roll and / or the local defect category corresponding to a part of the coating roll; the overall defect category is used as the overall defect detection result of the coating roll. The local defect detection results of the coated film roll are determined based on the local defect category. The consistency detection results of the coated film roll to be inspected are determined based on the overall defect detection results and / or local defect detection results. Defect categories are predicted using a coating defect detection model, reducing reliance on the experience of technicians when analyzing coating measurement data and improving analysis efficiency.

[0078] In this scenario example, various influencing factors and corresponding solutions can be enumerated based on historical experience and mechanistic knowledge, constructing a knowledge graph stored in the defect cause graph database. The knowledge graph model in this scenario example can be a directed acyclic graph. The defect categories from the consistency detection results are searched in the defect cause graph database to obtain the defect generation path between the defect category and the coating equipment operating parameters. The defect generation path has influencing factors; the operating parameters to be confirmed related to these influencing factors are identified. The normal operating parameters of the coating equipment when producing qualified coated film rolls are obtained. The operating parameters to be confirmed are compared with the normal operating parameters. Operating parameters that do not match the normal operating parameters are identified as abnormal operating parameters, and the influencing factors corresponding to these abnormal operating parameters are identified as the target abnormal factors causing consistency defects.

[0079] In this scenario example, by identifying the target anomaly factors, an optimization plan can be provided based on the corresponding SOP (Standard Operating Procedure) and the target monitoring parameter range, guiding equipment maintenance personnel to handle the anomaly. This scenario example achieves the intelligent generation of coating consistency defect handling suggestions, realizing the goal of outputting equipment adjustment and optimization plans according to the defect category. This can shorten the resolution cycle of coating consistency defects and improve the consistency and quality of coated film roll products.

[0080] This specification provides a method for detecting coating consistency. Please refer to [link / reference]. Figure 2 The coating consistency testing method includes the following steps:

[0081] S210. Based on the initial coating measurement data of the coated film roll, obtain multiple frames of target coating measurement data.

[0082] In this multi-frame target coating measurement data set, there is at most partial overlap between adjacent frames. The initial coating measurement data can be obtained by monitoring the weight of the coating area during the coating roll production process using measuring equipment, and is used to assess coating consistency. The initial coating measurement data can be the areal density data of the coating roll, the weight data of the coating roll, or the thickness data of the coating roll.

[0083] In some cases, considering the large size of the coated film roll and the fact that some defect categories are localized while others are global, in order to comprehensively analyze the consistency performance of the coated film roll, the initial coating measurement data of the coated film roll can be broken down into smaller parts to comprehensively test the consistency of the coated film roll.

[0084] Specifically, for different location ranges, different target coating measurement data are extracted from the initial coating measurement data of the coating film roll. To reduce the probability of missing defect categories, the target coating measurement data of two adjacent frames may or may not overlap with the corresponding location ranges in the initial coating measurement data, thus obtaining target coating measurement data of at most partial overlap between adjacent frames. In other embodiments, the initial coating measurement data of the coating film roll can also be copied into multiple copies, each copy of the initial coating measurement data corresponding to a different segmentation range. Target coating measurement data is then obtained by cropping each copy of the initial coating measurement data according to the different segmentation ranges. Similarly, the target coating measurement data of two adjacent frames may or may not overlap with the corresponding segmentation ranges in the initial coating measurement data.

[0085] S220. Classify the target coating measurement data and determine the classification result corresponding to the target coating measurement data.

[0086] Specifically, consistency-related features are extracted from any frame of target coating measurement data to obtain data features used for classifying consistency defects. The classification result corresponding to that frame of target coating measurement data is determined based on the extracted data features. It should be noted that when classifying the target coating measurement data to obtain a specific defect category, the classification result corresponding to the target coating measurement data can be either the defect category itself. If the target coating measurement data is classified to obtain a normal category (i.e., no defect category is detected), the classification result corresponding to the target coating measurement data is the normal category. Therefore, the classification result corresponding to the target coating measurement data can be either the defect category or the normal category. Furthermore, the classification result can also include the specific location of the defect category within the target coating measurement data.

[0087] S230. Summarize the classification results corresponding to the target coating measurement data and generate the consistency test results of the coated film roll.

[0088] In some cases, on the one hand, considering that the classification results obtained from classifying the target coating measurement data may be correct or incorrect—for example, target coating measurement data that originally had no defects may be falsely detected as defective, or target coating measurement data that originally had defects may be falsely detected as normal, or target coating measurement data that originally had defect category A may be falsely detected as defect category B; on the other hand, the classification results corresponding to the target coating measurement data are for a portion of the coated film roll, not the entire coated film roll. Therefore, based on the classification results corresponding to the target coating measurement data, it is necessary to further summarize the classification results corresponding to the target coating measurement data to obtain the consistency detection results of the coated film roll.

[0089] Specifically, the consistency defects that may exist in the coated film roll are several known defect categories. Classifying the target coating measurement data yields classification results that also appear within these known defect categories. The classification results for each frame of target coating measurement data may be the same or different. For any category among the known defect categories and normal categories, a voting method can be used to statistically analyze the classification results for each frame of target coating measurement data. Initially, the number of votes for any category is a preset value. Each time a category appears in the classification results corresponding to the target coating measurement data, the corresponding number of votes for that category is increased, and so on, until the number of votes for each category is obtained. The consistency detection result of the coated film roll is determined based on the number of votes for each category. For example, an unweighted voting method or a weighted voting method can be used, and the corresponding weights can be set according to the actual situation.

[0090] The aforementioned coating consistency detection method first obtains multiple frames of target coating measurement data with at most partial overlap between adjacent frames based on initial coating measurement data. This increases the probability of coating defects appearing in the target coating measurement data and reduces the probability of missed detection. Second, the target coating measurement data is classified to determine the corresponding classification results. Finally, the classification results corresponding to the target coating measurement data are summarized to generate the consistency detection results of the coated film roll. This reduces the impact of possible erroneous results from the classification process on the final coating consistency result, thereby enabling more accurate detection of coating defects in the coated film roll and improving the accuracy of the coating roll consistency detection results.

[0091] In some implementations, please refer to Figure 3a Obtaining multiple frames of target coating measurement data based on the initial coating measurement data of the coated film roll may include the following steps:

[0092] S310. Based on the preset step size and the size of the initial coating measurement data, perform multiple sampling on the initial coating measurement data to obtain multiple frames of intermediate coating measurement data with at most partial overlap between adjacent frames.

[0093] S320. Generate target coating measurement data based on intermediate coating measurement data.

[0094] The preset step size can be set according to the actual situation. For example, if the accuracy of previous consistency test results exceeds the preset requirements, the preset step size can be increased accordingly; if the accuracy of previous consistency test results does not meet the preset requirements, the preset step size can be decreased accordingly.

[0095] In some cases, on the one hand, because the length of the initial coating measurement data is several times its width, the aspect ratio of the initial coating measurement data affects the accuracy of the coating consistency detection results to some extent. Therefore, multiple sampling processing of the initial coating measurement data is required. On the other hand, different defects may correspond to different locations or different data ranges in the initial coating measurement data. To reduce the missed detection of defect categories, there is overlap between adjacent frames of measurement data obtained through multiple sampling. Furthermore, as mentioned above, as the accuracy of the consistency detection results improves, the overlap between adjacent frames of measurement data can be gradually reduced until there is no overlap, in order to reduce the amount of data processing.

[0096] Specifically, the initial coating measurement data can be denoted as m*n, where m is the data length and n is the data width. The preset step size can be denoted as s. Since the data length is several times the data width, the data width n is used as the data extraction size to extract intermediate coating measurement data of size n*n from the initial coating measurement data. After obtaining the first frame of intermediate coating measurement data, the starting point of data extraction is moved along the data length direction by a preset step size s. Using the moved starting point as the new starting point, intermediate coating measurement data of size n*n is extracted again from the initial coating measurement data. This process is repeated to perform multiple sampling on the initial coating measurement data, resulting in multiple frames of intermediate coating measurement data with at most partial overlap between adjacent frames. After obtaining the intermediate coating measurement data from the multiple sampling of the initial coating measurement data, the intermediate coating measurement data needs to be processed by filling, padding, or any other method to generate the target coating measurement data.

[0097] It should be noted that the preset step size s can be smaller than the data width n. As the accuracy of consistency detection results improves, in order to reduce the amount of data processing, the preset step size s can be larger than the data width n.

[0098] In the above embodiments, multiple frames of intermediate coating measurement data are obtained by multi-sampling the initial coating measurement data according to the preset step size and the size of the initial coating measurement data; and target coating measurement data is generated based on the intermediate coating measurement data, thereby improving the quality of the target coating measurement data used to detect coating consistency and providing an accurate data basis for detecting coating consistency.

[0099] In some implementations, the classification result is obtained by classifying the target coating measurement data using a coating defect detection model. Obtaining multiple frames of target coating measurement data based on the initial coating measurement data of the coating film roll may include: performing multiple sampling of the initial coating measurement data according to the size of the input data of the coating defect detection model to obtain multiple frames of target coating measurement data.

[0100] The coating defect detection model can be any one of SqueezeNet, MobileNet, ResNet18, ResNet34, or Inception v3. In this embodiment, SqueezeNet, MobileNet, ResNet18, ResNet34, and Inception v3 have different feature extraction capabilities. Please refer to [link / reference]. Figure 3b , Figure 3bThe test results for different models are shown. Due to the small dimensionality of the coating measurement data for some film rolls, feature vanishing occurs in large-scale models, affecting the accuracy of the results. However, for small-scale models, the test results show that the training and testing accuracy are higher, making the ResNet18 model a suitable choice. It should be noted that, to better accommodate different film roll products, the actual application model can be designed with multiple input data sizes to reduce incompatibility caused by model scale differences. This ensures consistent defect detection accuracy for most products and improves the model's robustness.

[0101] In some cases, the aspect ratio of the initial coating measurement data can affect the feature extraction capability of the neural network model to a certain extent. Therefore, it is necessary to process the initial coating measurement data according to the size of the input data of the coating defect detection model in order to obtain target coating measurement data that matches the size of the input data.

[0102] Specifically, the initial coating measurement data size can be denoted as m*n, where m is the data length and n is the data width. The preset step size can be denoted as s. In some embodiments, target coating measurement data of size x*x can be extracted from the initial coating measurement data size of m*n according to the input data size x*x of the coating defect detection model. In other embodiments, intermediate coating measurement data of size n*n can be extracted from the initial coating measurement data using the data width n as the data extraction size, and then the intermediate coating measurement data of size n*n can be filled according to the input data size x*x of the coating defect detection model to obtain the target coating measurement data.

[0103] In the above embodiments, by performing multiple sampling of the initial coating measurement data according to the input data size of the coating defect detection model, multiple frames of target coating measurement data are obtained. This helps the coating defect detection model to better extract features from the target coating measurement data, providing an accurate data basis for detecting coating consistency.

[0104] In some implementations, please refer to Figure 4a The process of summarizing the classification results corresponding to the target coating measurement data and generating consistency detection results for the coated film rolls may include the following steps:

[0105] S410. For each category of data in the classification results, determine the statistical quantity of each category of data.

[0106] S420. Based on the statistical quantity of each category of data, determine the target category data that meets the preset statistical quantity conditions in the classification results corresponding to the target coating measurement data.

[0107] S430. Generate consistency test results for coated film rolls based on target category data.

[0108] In some cases, since the object of classification processing is the target coating measurement data obtained by multiple sampling, rather than the entire coating roll, it is necessary to assemble the classification results of each frame of target coating measurement data to obtain the consistency detection results of the coating roll.

[0109] Specifically, each frame of target coating measurement data is classified to obtain the corresponding classification results. These results are then aggregated, and each frame's classification result includes at least one category of data. Different frames of coating measurement data may contain the same category or different categories. The categories that appear in each frame's classification results are determined, and for any given category, the number of times that category appears is counted to obtain the statistical count for that category. It should be noted that equal-weighted voting or unequal-weighted weighted voting can be used to assemble the classification results to obtain the consistency detection results for the coated film roll.

[0110] The data is sorted according to the statistical quantity of each category, and the category data that meets the preset statistical quantity condition is determined as the target category data. For example, the category data with a statistical quantity exceeding a preset threshold can be determined as the target category data, or the category data with the top N statistical quantities can be used as the target category data, where N can be 3 or 5. After obtaining the target category data from the classification results, the target classification data can be assembled to obtain the consistency detection results of the coated film roll.

[0111] In the above embodiments, by determining the target category data that meets the preset statistical quantity condition in the classification results corresponding to the target coating measurement data based on the statistical quantity of each category data, filtering out the category data that does not meet the preset statistical quantity condition, reducing the impact of this category data on the accuracy of the consistency detection results, and improving the accuracy of the detection results of the coated film roll.

[0112] In some implementations, the target category data includes an overall defect category corresponding to the entire coated film roll and / or a local defect category corresponding to a portion of the coated film roll. Generating a consistency inspection result for the coated film roll based on the target category data includes at least one of the following: using the overall defect category as the overall defect inspection result for the coated film roll; or determining the local defect inspection result for the coated film roll based on the local defect category; wherein the consistency inspection result includes the overall defect inspection result and / or the local defect inspection result.

[0113] Among these, the consistency defects of the coated film roll can be reflected in the coating weight. Partial coating weight defects manifest as overall defects in the coated film roll, while partial defects manifest as localized defects. Overall defects can be categorized as U-weight, inverted U-weight, single-row heavier, single-row lighter, or periodic fluctuations. Local defects can be categorized as single-film area U-weight, weight tilt, edge lighter, process variation, or initial coating fluctuations.

[0114] Specifically, if the target category data retained from the classification results corresponding to the target coating measurement data includes an overall defect category, since the overall defect category corresponds to the entire coated film roll, the overall defect category included in the target category data can be directly used as the overall defect detection result of the coated film roll. If the target category data retained from the classification results corresponding to the target coating measurement data includes a local defect category, since the local defect category corresponds to a part of the coated film roll, and the local defect category corresponds to the target coating measurement data, it is necessary to assemble the data according to the local defect category and map it onto the coated film roll, thereby determining the local defect detection result of the coated film roll based on the local defect category.

[0115] If only local defect categories are detected from the coated film roll, the consistency test result includes the local defect detection result; if only overall defect categories may be detected from the coated film roll, the consistency test result includes the overall defect detection result; if both local and overall defect categories are detected from the coated film roll, the consistency test result includes both the overall defect detection result and the local defect detection result.

[0116] In the above implementation, the consistency of the coated film roll is first tested from the perspectives of overall defects and local defects. Then, the failure mode of the coated film roll defects is comprehensively evaluated by using the overall defect test results and the local defect test results, thereby improving the accuracy of defect detection and analysis.

[0117] In some implementations, the classification result also includes the detection location corresponding to the local defect category. Before determining the local defect detection result of the coated film roll based on the local defect category, the method may further include: determining the actual location of the local defect category in the coated film roll based on the detection location corresponding to the local defect category; wherein, the local defect detection result includes the actual location and the local defect category.

[0118] In some cases, during the classification and processing of target coating measurement data to determine the defect categories within the data, the detection location of each defect category within the target coating measurement data can also be determined. As mentioned earlier, the overall defect category refers to the entire coated film roll, and its location within the target coating measurement data can be disregarded; the overall defect category can be directly used as the overall defect detection result for the coated film roll. However, for local defect categories, which refer to a specific part of the coated film roll, it is necessary to detect the actual location of the local defect category within the coated film roll.

[0119] Specifically, the target coating measurement data is classified to determine the corresponding local defect categories and their detection positions within the target coating measurement data. Since the target coating measurement data is obtained through multiple sampling of the initial coating measurement data, its extraction position within the initial coating measurement data is known. Furthermore, to determine the actual position of the local defect category within the coating roll, it can be calculated based on the detection position and extraction position of the local defect category in the target coating measurement data and the initial coating measurement data. The local defect category and its corresponding actual position are then combined to form the local defect detection result.

[0120] In the above embodiments, by determining the actual position of the local defect category in the coated film roll according to the detection position corresponding to the local defect category, not only is the local defect category detected, but the position information of the local defect category is also accurately detected, thereby realizing the failure mode of automatic classification of consistent defects.

[0121] In some implementations, the method may further include: searching a defect cause graph database based on the defect category in the consistency detection results to obtain the defect generation path between the defect category and the coating equipment operating parameters.

[0122] The defect cause graph database can be obtained by constructing a knowledge graph by enumerating various influencing factors and corresponding solutions based on historical experience and mechanistic knowledge. For example, it can be analyzed using FTA (Fault Tree Analysis) tools. Based on the experience of technical personnel, possible causes of coating consistency defects are enumerated, and the equipment problems causing the failures are analyzed one by one for each possible cause, down to individual parameter leaf nodes, corresponding to field inspection items or equipment parameters. The knowledge base for a single failure mode expands into a tree structure. The defect cause graph database includes knowledge nodes and reasoning logic. For example, the knowledge graph model can use a directed acyclic graph. Here, the knowledge graph is a network structure without a central point. Please refer to [link to relevant documentation]. Figure 4b The basic triplet configurations are as follows:

[0123] 1) Failure Mode -> Influencing Factors: The direct cause of failure.

[0124] 2) Influencing factors -> Influencing factors: Direct causes or observable factors.

[0125] 3) Monitoring parameters -> Influencing factors: Parameters that can be monitored by means of sampling, product inspection, visual inspection, offline measurement, etc. of the equipment control system can be used to determine whether the influencing factors meet the conditions that cause the influence.

[0126] 4) Influencing Factors -> Handling SOP: SOP (Standard Operation Process) is a standardized operating instruction written based on equipment design, process design and lean production process design. It is used to regulate and guide the elimination of abnormalities of influencing factors and bring the monitored parameters back to the normal range.

[0127] Specifically, the consistency inspection results include the defect categories corresponding to defects present in the coated film roll. These defect categories are then searched in the defect cause graph database to obtain the defect generation path between the defect category and the coating equipment operating parameters. For a specific defect category, a tree-like subgraph can be retrieved based on the triplet path, starting from that defect category; this is the FTA tree for that defect category. Please refer to [link to relevant documentation]. Figure 4c Taking the periodic fluctuation defect category as an example, Figure 4c This is a tree structure representing the categories of periodic fluctuation defects. The first-level influencing factors are: periodic fluctuations in pump inlet pressure, periodic fluctuations in oven temperature, and back roller runout. Oven temperature can be monitored using sensor values ​​output by the equipment control system; back roller runout requires measurement during machine shutdown; periodic fluctuations in pump inlet pressure are caused by secondary influencing factors such as gap fluctuations, liquid level difference, and pump speed fluctuations. These secondary influencing factors correspond to control parameters in the equipment control system. For example, gap fluctuations correspond to real-time left and right gap changes, liquid level difference corresponds to pump inlet pressure, pump speed fluctuations correspond to pump inlet pressure, and pump speed fluctuations correspond to real-time pump speed.

[0128] In some implementations, coating equipment operating parameters can be obtained before determining the defect generation path. Based on the defect category in the consistency test results and the coating equipment operating parameters, a failure path (defect generation path) between the defect pattern and the abnormal parameters is searched in a defect cause graph database. Each defect generation path can be used to describe the possible causes of the corresponding defect.

[0129] In this embodiment, the defect generation path can be automatically generated by the defect category in the consistency detection results, which shortens the coating weight defect resolution cycle and improves the overall consistency level of film roll products.

[0130] In some implementations, please refer to Figure 5 The method may also include the following steps:

[0131] S510. Obtain the unconfirmed operating parameters of the coating equipment related to the influencing factors along the defect generation path.

[0132] S520. Obtain the normal operating parameters when the coating equipment produces qualified coated film rolls.

[0133] S530. Based on the comparison results between the operating parameters to be confirmed and the normal operating parameters, identify the target abnormal factors that cause the consistency defect from the influencing factors on the defect generation path.

[0134] Specifically, after determining the defect generation path, the operational parameters to be confirmed related to the influencing factors along the defect generation path are obtained, thus generating a monitoring parameter checklist. Normal operating parameters for the coating equipment when producing qualified coated film rolls are retrieved from the production monitoring database. An operational parameter to be confirmed is obtained from the monitoring parameter checklist, and this parameter is compared with the normal operating parameters to obtain a comparison result. Based on this comparison result, the target anomaly causing the consistency defect is identified among the influencing factors along the defect generation path.

[0135] For example, if the operating parameter to be confirmed differs from the normal operating parameter, it can be determined that the operating parameter to be confirmed is an abnormal monitoring parameter, and further, it can be added to the list of abnormal factors. If the operating parameter to be confirmed is the same as the normal operating parameter, it can be determined that the operating parameter to be confirmed is a normal monitoring parameter. Similarly, another operating parameter to be confirmed is obtained from the monitoring parameter check table, and the above comparison process with the normal operating parameter is repeated until all operating parameters to be confirmed in the monitoring parameter check table have been compared.

[0136] In some implementations, anomaly handling suggestions can be generated based on target anomalies in the anomaly factor list. Continuing with the example of periodic fluctuation defects, the monitoring parameter list is compared item by item to automatically identify abnormal parameters and pinpoint the abnormal influencing factors causing periodic fluctuation failure modes. Corresponding SOPs and handling solutions for the target monitoring parameter range can also be provided to guide equipment maintenance personnel in handling anomalies. This includes a work order and feedback system, allowing optimization solutions to be issued via email, internal messaging, and the device's IoT channel, and feedback on results to be received. In other implementations, it can be determined whether the defect category in the consistency detection results is intermittent. If it is intermittent, no anomaly handling suggestions need to be generated; if it is not intermittent, anomaly handling suggestions can be generated based on target anomalies in the anomaly factor list to help improve the quality of coating process products.

[0137] In the above embodiments, by identifying the target anomaly factors causing consistency defects among the influencing factors along the defect generation path, the FTA model can recommend equipment adjustment operations based on the category of detected coated film roll weight defects. On the one hand, suggestions are pushed to production operators for manual equipment; on the other hand, parameter adjustments are sent to the equipment controller for automated equipment.

[0138] This specification provides a method for detecting coating consistency, which includes the following steps:

[0139] S602. Based on the preset step size and the size of the initial coating measurement data, perform multiple sampling on the initial coating measurement data to obtain multiple frames of intermediate coating measurement data with at most partial overlap between adjacent frames.

[0140] S604. Generate target coating measurement data based on intermediate coating measurement data.

[0141] Among them, there is at most partial overlap between the target coating measurement data of two adjacent frames in the multi-frame target coating measurement data.

[0142] S606. Classify the target coating measurement data and determine the classification result corresponding to the target coating measurement data.

[0143] S608. For each category of data in the classification results, determine the statistical quantity of each category of data.

[0144] S610. Based on the statistical quantity of each category of data, determine the target category data that meets the preset statistical quantity conditions in the classification results corresponding to the target coating measurement data.

[0145] The target category data includes the overall defect category of the corresponding coated film roll and / or the local defect category of the corresponding coated film roll.

[0146] S612. The overall defect category is used as the overall defect detection result of the coated film roll.

[0147] S614. Based on the detection location corresponding to the local defect category, determine the actual location of the local defect category in the coated film roll.

[0148] S616. Determine the local defect detection results of the coated film roll based on the actual location and local defect type.

[0149] The consistency test results include overall defect test results and / or local defect test results.

[0150] S618. Based on the defect category in the consistency test results, search the defect cause graph database to obtain the defect generation path between the defect category and the coating equipment operating parameters.

[0151] S620. Obtain the unconfirmed operating parameters of the coating equipment related to the influencing factors along the defect generation path.

[0152] S622. Obtain the normal operating parameters of the coating equipment when producing qualified coated film rolls.

[0153] S624. Based on the comparison results between the operating parameters to be confirmed and the normal operating parameters, identify the target abnormal factors that cause the consistency defect from the influencing factors on the defect generation path.

[0154] In the above embodiments, the closed-loop operation from online defect detection to problem handling verification not only realizes intelligent control of coating weight consistency, but also reduces production costs and improves overall product quality.

[0155] This specification provides a training method for a coating defect detection model. Please refer to [link to relevant documentation]. Figure 6 The training method for this coating defect detection model may include the following steps:

[0156] S710. Based on historical coating measurement data of the coated film roll, obtain multi-frame coating measurement data samples.

[0157] S720. Input each frame of coating measurement data sample into the initial detection model for classification processing to obtain the classification result corresponding to each frame of coating measurement data sample.

[0158] S730. Update the initial detection model based on the classification results and the labels corresponding to each frame of coating measurement data samples to obtain the coating defect detection model.

[0159] Among them, there is at most partial overlap between coating measurement data samples of two adjacent frames in the multi-frame coating measurement data sample; each frame of coating measurement data sample corresponds to a label.

[0160] In the past, during the production of coated film rolls, the weight of the coated area was monitored using measuring equipment to obtain historical coating measurement data for the coated film rolls. In this embodiment, a weight data acquisition module can periodically collect coating measurement data, perform streaming calculations of indicators, and write the data to a database.

[0161] Specifically, firstly, multiple sampling or cropping can be performed on historical coating measurement data of the coated film roll to obtain multiple frames of coating measurement data samples. Secondly, each frame of coating measurement data sample is pre-labeled with a corresponding label. Each frame of coating measurement data sample is input into an initial detection model, which classifies each frame to obtain a classification result. Finally, the model loss value is determined based on the classification result and label of each frame of coating measurement data sample. The parameters of the initial detection model are updated based on the model loss value, and the updated detection model is trained using the coating measurement data samples until the model stops training, thus obtaining the coating defect detection model.

[0162] In this embodiment, a self-learning module can be provided to monitor model performance and control model retraining. The initial detection model can be built using a deep learning model and output results in one-hot encoding. Furthermore, the initial detection model can be built using one deep learning model, multiple deep learning models, or different deep learning models. Additionally, a model management module can be provided to manage the model's weight files, label valid models, and implement model rollback.

[0163] The training method for the aforementioned coating defect detection model involves several steps. First, by using historical coating measurement data from the coated film roll, multiple frames of coating measurement data samples are obtained. This increases the probability of coating defects appearing in the samples, allowing for the construction of more coating measurement data samples and, to some extent, amplifying the sample data, which is beneficial for training the initial detection model. Second, each frame of coating measurement data samples is classified to determine the classification result. Finally, the initial detection model is updated based on the classification results and the labels corresponding to each frame of coating measurement data samples to obtain the coating defect detection model. This method achieves deep learning model training on coating measurement data, resulting in a coating defect detection model that can automatically detect defect categories.

[0164] In some implementations, obtaining multi-frame coating measurement data samples based on historical coating measurement data of the coated film roll may include: performing multiple sampling on the historical coating measurement data according to a preset step size and the size of the historical coating measurement data to obtain multiple frames of intermediate coating measurement data with at most partial overlap between adjacent frames; and generating coating measurement data samples based on the intermediate coating measurement data.

[0165] Specifically, the size of the historical coating measurement data can be denoted as m*n, where m is the data length and n is the data width. The preset step size can be denoted as s. Since the data length is several times the data width, the data width n is used as the data extraction size to extract intermediate coating measurement data of size n*n from the historical coating measurement data. After obtaining the first frame of intermediate coating measurement data, the starting point of data extraction is moved along the data length direction by a preset step size s. Using the moved starting point as the new starting point, intermediate coating measurement data of size n*n is extracted again from the historical coating measurement data. This process is repeated to perform multiple sampling on the historical coating measurement data, resulting in multiple frames of intermediate coating measurement data with at most partial overlap between adjacent frames. After obtaining the intermediate coating measurement data from the historical coating measurement data through multiple sampling, the intermediate coating measurement data needs to be processed by filling, padding, or other methods to generate coating measurement data samples.

[0166] In the above embodiments, by using historical coating measurement data of the coating film roll, multiple frames of coating measurement data samples are obtained, which expands the sample of defect categories with less data, reduces the impact of sample imbalance on the model training effect, and improves sample balance.

[0167] In some implementations, obtaining multi-frame coating measurement data samples based on historical coating measurement data of the coated film roll includes: performing multiple sampling of historical coating measurement data according to the size of the input data of the initial detection model to obtain multi-frame coating measurement data samples.

[0168] The initial detection model can be any one of SqueezeNet, MobileNet, ResNet18, ResNet34, or Inception v3. In this implementation, SqueezeNet, MobileNet, ResNet18, ResNet34, and Inception v3 have different feature extraction capabilities. Because the coating measurement data for some membrane rolls has a small dimensionality, feature vanishing occurs in large-scale models, affecting the accuracy of the results. However, for small-scale models, test results show that the training and testing accuracy are higher; therefore, the ResNet18 model can be selected.

[0169] Specifically, coating measurement data samples of size x*x can be extracted from the initial coating measurement data of size m*n, based on the input data size x*x of the coating defect detection model. Alternatively, intermediate coating measurement data of size n*n can be extracted from the initial coating measurement data using the data width n as the data extraction size, and then the n*n intermediate coating measurement data can be filled with the input data size x*x of the coating defect detection model to obtain coating measurement data samples.

[0170] In the above implementation, historical coating measurement data is multisampled based on the input data size of the initial detection model to obtain coating measurement data samples whose shape matches that of the initial detection model. This reduces the limitations imposed by historical coating measurement data on the initial detection model, improves the feature extraction capability of the initial detection model, and facilitates the training of the initial detection model.

[0171] In some implementations, please refer to Figure 7 The method for determining the label may include the following steps:

[0172] S810. Visualize the historical coating measurement data to obtain a coating consistency distribution map.

[0173] S820. In the case of category labeling operation based on the coating consistency distribution map, determine the category data corresponding to the historical coating measurement data.

[0174] S830: Generate labels for each frame of coating measurement data samples based on category data corresponding to historical coating measurement data.

[0175] The coating consistency distribution map can be in the form of a heatmap, such as a coating weight heatmap generated based on historical coating weight data. The horizontal and vertical axes of the coating weight heatmap correspond to the measurement points on the coated film roll, and the color changes in the heatmap represent the coating weight at each measurement point. Alternatively, the coating consistency distribution map can be in the form of a scatter plot.

[0176] Specifically, historical coating measurement data is visualized to obtain a coating consistency distribution map. This map comprehensively displays the consistency of the coated film roll, allowing labelers to fully observe the coating weight data and clearly understand the consistency of various unknown areas. Labelers can then use the coating consistency distribution map to issue category labeling operations based on their observations. In response to these operations, the category data corresponding to the historical coating measurement data is determined, and corresponding labels are generated.

[0177] For example, a data visualization and annotation module is provided. Historical coating weight data is displayed using scatter plots or heatmaps via a web client. For instance, the annotation interface uses a heatmap to display historical coating weight data. Annotators can select at least one category from the following—U-weight, inverted U-weight, single-column heavier, single-column lighter, periodic fluctuation, single-film area U-weight, weight tilt, edge lighter, process variation, initial coating fluctuation, and normal—to annotate the coating weight heatmap based on the actual situation of the historical coating weight data. If the annotator observes three defect categories—periodic fluctuation, single-film area U-weight, and weight tilt—and can issue annotation operations corresponding to these three defect categories, the defect category corresponding to the historical coating measurement data is determined to be periodic fluctuation, single-film area U-weight, or weight tilt. If the annotator does not observe any defect categories and can issue a normal annotation operation, the category of data corresponding to the historical coating measurement data is determined to be normal.

[0178] In the above embodiments, by visualizing historical coating measurement data, a coating consistency distribution map is obtained. In response to the coating consistency distribution map having a category labeling operation, the category data corresponding to the historical coating measurement data is determined to generate a label for each frame of coating measurement data sample.

[0179] In some implementations, the category data includes overall defect categories corresponding to the entire coated film roll and / or local defect categories corresponding to a portion of the coated film roll. The coating consistency distribution map includes a heatmap of coating measurement data. When a category labeling operation occurs based on the coating consistency distribution map, determining the category data corresponding to historical coating measurement data may include at least one of the following: determining the overall defect category corresponding to historical coating measurement data when an overall labeling operation occurs on the coating measurement data heatmap; or determining the local defect category corresponding to historical coating measurement data when a local labeling operation occurs on the coating measurement data heatmap.

[0180] Specifically, a heatmap of coating measurement data is displayed. In response to an overall annotation operation on the heatmap, the corresponding overall defect category is determined. Similarly, in response to a local annotation operation on the heatmap, the corresponding local defect category is determined. In the above embodiment, performing an overall or local annotation operation on the coating measurement data heatmap generates corresponding labels. These labels can be used for defect detection model training and model iterative optimization in intelligent software systems.

[0181] In some embodiments, local defect categories correspond to labeled positions. Generating a label for each frame of coating measurement data sample based on category data corresponding to historical coating measurement data includes at least one of the following: taking the overall defect category as the overall defect label of each frame of coating measurement data sample; or, in a case where it is determined that the data range of the coating measurement data sample overlaps with the labeled position corresponding to the local defect category, generating a local defect label for the coating measurement data sample with overlapping data ranges based on the local defect category and the labeled position.

[0182] Specifically, when the annotator determines that the historical coating measurement data has an overall defect category, since the overall defect category corresponds to the whole of the coating film roll, the overall defect category can be directly used as the overall defect label of each frame of coating measurement data sample generated based on the historical coating measurement data.

[0183] When the annotator determines that the historical coating measurement data has a local defect category, since the local defect category corresponds to a local part of the coating film roll, it is necessary to determine the labeled position corresponding to the local defect category. When generating each frame of coating measurement data sample based on the historical coating measurement data, it is determined whether there is an overlap between the data range of the coating measurement data sample and the labeled position corresponding to the local defect category. If there is an overlap between the data range of the coating measurement data sample and the labeled position corresponding to the local defect category, a local defect label for the coating measurement data sample with overlapping data ranges is generated based on the local defect category and the labeled position. If there is no overlap between the data range of the coating measurement data sample and the labeled position corresponding to the local defect category, the coating measurement data sample does not have a corresponding local defect label.

[0184] In the above embodiments, firstly, the historical coating measurement data is labeled from two perspectives of overall defects and local defects, and then the initial detection model is comprehensively trained by the overall defect labels and local defect labels, so as to obtain a coating defect detection model capable of comprehensively evaluating the failure modes of coating film roll defects, which improves the accuracy of defect detection and analysis.

[0185] In some embodiments, after the model completes training, the calculation accuracy of the coating defect detection model can be counted. When the calculation accuracy exceeds the limit, the model can continue to be automatically trained and stored in the model library, and whether the model is qualified is determined through regression testing. If the model is unqualified, the number of unqualified times is recorded. If the number of unqualified times exceeds a preset number threshold (such as three times), manual parameter adjustment can be performed, and after manual parameter adjustment, whether the model is qualified is determined again through regression testing. It can be seen that a closed performance loop for the coating defect detection model is realized, thereby improving the intelligent capability of the system to adapt to the continuous upgrading of different film roll products.

[0186] This specification provides an embodiment for training a coating defect detection model, which includes the following steps:

[0187] S902. Visualize the historical coating measurement data to obtain a coating consistency distribution map.

[0188] S904. In the case of category labeling operations based on the coating consistency distribution map, determine the category data corresponding to the historical coating measurement data.

[0189] The category data includes the overall defect category of the corresponding coated film roll and / or the local defect category of the corresponding coated film roll; the coating consistency distribution map includes a heat map of coating measurement data. Specifically, when an overall annotation operation occurs on the coating measurement data heat map, the overall defect category corresponding to the historical coating measurement data is determined; or, when a local annotation operation occurs on the coating measurement data heat map, the local defect category corresponding to the historical coating measurement data is determined.

[0190] S906. Based on the size of the input data of the initial detection model, perform multiple sampling on the historical coating measurement data to obtain multiple frames of coating measurement data samples.

[0191] Among them, there is at most partial overlap between coating measurement data samples of two adjacent frames in the multi-frame coating measurement data sample; each frame of coating measurement data sample corresponds to a label.

[0192] S908. Generate labels for each frame of coating measurement data samples based on the category data corresponding to the historical coating measurement data.

[0193] Among them, the local defect category corresponds to the labeling position; specifically, the overall defect category is used as the overall defect label for each frame of coating measurement data sample; when it is determined that the data range of the coating measurement data sample overlaps with the labeling position corresponding to the local defect category, the local defect label of the coating measurement data sample with overlapping data range is generated based on the local defect category and the labeling position.

[0194] S910. Input each frame of coating measurement data sample into the initial detection model for classification processing to obtain the classification result corresponding to each frame of coating measurement data sample;

[0195] S912. Update the initial detection model based on the classification results and the labels corresponding to each frame of coating measurement data samples to obtain the coating defect detection model.

[0196] This specification provides a coating consistency testing device 800. Please refer to [link / reference]. Figure 8The coating consistency detection device 800 includes: an initial data processing module 810, a target data classification module 820, and a classification result summary module 830.

[0197] The initial data processing module 810 is used to obtain multiple frames of target coating measurement data based on the initial coating measurement data of the coating film roll; wherein, there is at most partial overlap between the target coating measurement data of two adjacent frames in the multiple frames of target coating measurement data.

[0198] The target data classification module 820 is used to classify the target coating measurement data and determine the classification result corresponding to the target coating measurement data.

[0199] The classification result summary module 830 is used to summarize the classification results corresponding to the target coating measurement data and generate the consistency detection results of the coating film roll.

[0200] It should be noted that for the description of the coating consistency testing device in the embodiments of this specification, please refer to the description of the coating consistency testing method in this specification, which will not be repeated here.

[0201] This specification provides a training device for a coating defect detection model. Please refer to [link / reference]. Figure 9 The model training device 900 includes: a historical data processing module 910, a sample classification processing module 920, and a detection model update module 930.

[0202] The historical data processing module 910 is used to obtain multiple frames of coating measurement data samples based on historical coating measurement data used to evaluate the coating consistency of the coated film roll; wherein, there is at most partial overlap between two adjacent frames of coating measurement data samples in the multiple frames of coating measurement data samples; each frame of coating measurement data sample corresponds to a label.

[0203] The sample classification processing module 920 is used to input the coating measurement data sample of each frame into the initial detection model for classification processing, and obtain the classification result corresponding to the coating measurement data sample of each frame.

[0204] The detection model update module 930 is used to update the initial detection model based on the classification results and the labels corresponding to the coating measurement data samples of each frame, so as to obtain the coating defect detection model.

[0205] It should be noted that for the description of the training device for the coating defect detection model in the embodiments of this specification, please refer to the description of the training method for the coating defect detection model in this specification, which will not be repeated here.

[0206] This specification provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the methods described in any of the above embodiments.

[0207] In some embodiments, an electronic device is provided, which may be a terminal or a server, and its internal structure diagram may be as follows: Figure 10 As shown, the electronic device includes a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the methods described in any of the above embodiments. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0208] Those skilled in the art will understand that Figure 10 The structures shown are merely block diagrams of a portion of the structures related to the solutions disclosed in this specification, and do not constitute a limitation on the electronic devices to which the solutions disclosed in this specification are applied. Specifically, the electronic devices may include more or fewer components than those shown in the figures, or combine certain components, or have different component arrangements.

[0209] This specification provides an electronic device, including: a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the method described in any of the above embodiments.

[0210] The embodiments of this specification also provide a computer program product, which includes instructions that can be executed by a processor of an electronic device to implement the method steps described above.

[0211] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0212] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable aggregated logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0213] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0214] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0215] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for detecting coating consistency, characterized in that, The method includes: Based on the initial coating measurement data of the coated film roll, multiple frames of target coating measurement data are obtained; wherein, at most, there is partial overlap between the target coating measurement data of two adjacent frames in the multiple frames of target coating measurement data; The target coating measurement data is classified and processed to determine the classification result corresponding to the target coating measurement data; The classification results corresponding to the target coating measurement data are summarized to generate the consistency detection results of the coated film roll; The method further includes: Based on the defect category in the consistency detection result, a search is performed in the defect cause graph database to obtain the defect generation path between the defect category and the coating equipment operating parameters. Obtain the unconfirmed operating parameters of the coating equipment related to the influencing factors along the defect generation path; Obtain the normal operating parameters of the coating equipment when it produces qualified coated film rolls; Based on the comparison results between the operating parameters to be confirmed and the normal operating parameters, the target abnormal factors causing the consistency defect are identified among the influencing factors on the defect generation path.

2. The method according to claim 1, characterized in that, The initial coating measurement data based on the coated film roll is used to obtain multiple frames of target coating measurement data, including: Based on the preset step size and the size of the initial coating measurement data, the initial coating measurement data is sampled multiple times to obtain multiple frames of intermediate coating measurement data with at most partial overlap between adjacent frames; The target coating measurement data is generated based on the intermediate coating measurement data.

3. The method according to claim 1, characterized in that, The classification result is obtained by classifying the target coating measurement data using a coating defect detection model; the initial coating measurement data based on the coating film roll is used to obtain multiple frames of target coating measurement data, including: Based on the size of the input data of the coating defect detection model, the initial coating measurement data is sampled multiple times to obtain multiple frames of the target coating measurement data.

4. The method according to claim 1, characterized in that, The process of summarizing the classification results corresponding to the target coating measurement data to generate the consistency detection results of the coated film roll includes: For each category of data in the classification results, determine the statistical quantity of each category of data; Based on the statistical quantity of each category of data, target category data that meets the preset statistical quantity condition is determined from the classification results corresponding to the target coating measurement data. The consistency test results of the coated film roll are generated based on the target category data.

5. The method according to claim 4, characterized in that, The target category data includes the overall defect category of the corresponding coated film roll and / or the local defect category of the corresponding part of the coated film roll; the generation of the consistency detection result of the coated film roll based on the target category data includes at least one of the following: The overall defect category is used as the overall defect detection result of the coated film roll; The local defect detection results of the coated film roll are determined according to the local defect category; wherein, the consistency detection results include the overall defect detection results and / or the local defect detection results.

6. The method according to claim 5, characterized in that, The classification result also includes the detection location corresponding to the local defect category; before determining the local defect detection result of the coated film roll according to the local defect category, the method further includes: Based on the detection location corresponding to the local defect category, the actual location of the local defect category in the coated film roll is determined; wherein, the local defect detection result includes the actual location and the local defect category.

7. The method according to claim 3, characterized in that, The method further includes: Based on historical coating measurement data of the coated film roll, multiple frames of coating measurement data samples are obtained; wherein, there is at most partial overlap between the coating measurement data samples of two adjacent frames in the multiple frames of coating measurement data samples; each frame of coating measurement data sample corresponds to a label; Each frame of coating measurement data sample is input into the initial detection model for classification processing to obtain the classification result corresponding to each frame of coating measurement data sample; The initial detection model is updated based on the classification results and the labels corresponding to each frame of coating measurement data samples to obtain the coating defect detection model.

8. The method according to claim 7, characterized in that, The historical coating measurement data based on the coated film rolls yields multi-frame coating measurement data samples, including: Based on the preset step size and the size of the historical coating measurement data, the historical coating measurement data is sampled multiple times to obtain multiple frames of intermediate coating measurement data with at most partial overlap between adjacent frames. The coating measurement data sample is generated based on the intermediate coating measurement data.

9. The method according to claim 7, characterized in that, The historical coating measurement data based on the coated film rolls yielded multi-frame coating measurement data samples, including: Based on the size of the input data of the initial detection model, the historical coating measurement data is multisampled to obtain multiple frames of coating measurement data samples.

10. The method according to claim 7, characterized in that, The method for determining the label includes: The historical coating measurement data are visualized to obtain a coating consistency distribution map; In the event that a category labeling operation has occurred based on the coating consistency distribution map, the category data corresponding to the historical coating measurement data is determined; A label is generated for each frame of coating measurement data sample based on the category data corresponding to the historical coating measurement data.

11. The method according to claim 10, characterized in that, The category data includes the overall defect category of the corresponding coated film roll and / or the local defect category of the corresponding part of the coated film roll; the coating consistency distribution map includes a heat map of coating measurement data; when a category labeling operation occurs based on the coating consistency distribution map, the category data corresponding to the historical coating measurement data is determined, including at least one of the following: In the event of an overall annotation operation on the coating measurement data heatmap, the overall defect category corresponding to the historical coating measurement data is determined; In the event of a local annotation operation occurring on the heatmap of the coating measurement data, the local defect category corresponding to the historical coating measurement data is determined.

12. The method according to claim 11, characterized in that, The local defect category corresponds to a labeled location; the label generated for each frame of coating measurement data sample based on the category data corresponding to the historical coating measurement data includes at least one of the following: The overall defect category is used as the overall defect label for each frame of coating measurement data sample; If it is determined that the data range of the coating measurement data sample overlaps with the labeling position corresponding to the local defect category, a local defect label is generated for the coating measurement data sample with overlapping data ranges based on the local defect category and the labeling position.

13. A coating consistency testing device, characterized in that, The device includes: An initial data processing module is used to obtain multiple frames of target coating measurement data based on the initial coating measurement data of the coated film roll; wherein, at most, there is partial overlap between the target coating measurement data of two adjacent frames in the multiple frames of target coating measurement data; The target data classification module is used to classify the target coating measurement data and determine the classification result corresponding to the target coating measurement data. The classification result summary module is used to summarize the classification results corresponding to the target coating measurement data and generate the consistency detection results of the coated film roll. The device is further configured to: search a defect cause graph database based on the defect category in the consistency detection result to obtain the defect generation path between the defect category and the operating parameters of the coating equipment; obtain the unconfirmed operating parameters of the coating equipment related to the influencing factors on the defect generation path; obtain the normal operating parameters of the coating equipment when producing qualified coated film rolls; and determine the target abnormal factor causing the consistency defect among the influencing factors on the defect generation path based on the comparison result between the unconfirmed operating parameters and the normal operating parameters.

14. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements the method described in any one of claims 1-12.

15. An electronic device, characterized in that, include: A memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method according to any one of claims 1-12.

Images