Lithium battery electrode surface defect identification and early warning method and system

Abstract

The invention discloses a lithium battery electrode surface defect identification early warning method and system, and relates to the technical field of defect detection.The method comprises the steps that historical defect identification records of a target production line are obtained and subjected to gray level enhancement processing, and multi-dimensional image features are extracted; performing clustering analysis on the historical data based on the features to obtain a sample data set; training the adaptive defect recognition model by using an integrated learning method, and obtaining a real-time image at a preset point location for defect recognition; and according to the multi-dimensional defect features, identification result judgment is carried out, and defect early warning is carried out. The technical problems of lack of detection pertinence, poor detection efficiency and poor detection accuracy for complex and changeable defects in the prior art are solved, and the technical effects of efficiently and accurately identifying the complex and changeable defects of the lithium battery electrode surface and the lithium battery diaphragm and remarkably improving the detection pertinence, the detection efficiency and the detection accuracy are achieved.

Description

Technical Field

[0001] This application relates to the technical field of defect detection, and particularly to a method and system for identifying and warning of defects on the surface of lithium battery electrodes. Background Art

[0002] With the wide application of lithium batteries in various electronic devices, electric vehicles and energy storage systems, the performance and safety of the batteries become particularly important. Defects on the surface of lithium battery electrodes directly affect the service life and safety of the batteries. Therefore, the early detection of electrode surface defects is crucial. However, traditional defect detection methods often rely on manual inspection or simple visual detection tools, which have problems such as slow detection speed, poor accuracy, and inability to handle complex and diverse defect types. With the complexity of production processes and the diversification of defect types, the existing technologies are difficult to meet the requirements of high efficiency and high accuracy. Therefore, there is an urgent need for a more intelligent and precise defect identification method that can automatically identify various defect types, provide efficient and highly accurate detection results, reduce manual intervention, and improve the overall production efficiency.

[0003] In the current related technologies, there are technical problems such as lack of detection pertinence, poor detection efficiency and detection accuracy in the face of complex and diverse defects. Summary of the Invention

[0004] This application solves the technical problems of lack of detection pertinence, poor detection efficiency and detection accuracy in the existing technologies in the face of complex and diverse defects by providing a method and system for identifying and warning of defects on the surface of lithium battery electrodes.

[0005] This application provides a method for identifying and warning of defects on the surface of lithium battery electrodes, including:

[0006] Obtain historical defect identification records based on the production characteristics set of the target production line, where the historical defect identification records correspond to a preset retrospective sampling window; based on the historical defect identification records, obtain historical image data, perform gray-scale enhancement processing on the historical image data, and based on the processing results, extract multi-dimensional gray-scale historical image features; according to the extracted multi-dimensional gray-scale historical image features, perform clustering analysis on historical monitoring data, divide the data into multiple clustering clusters, and output the multiple clustering clusters as multiple sample data groups; use the multiple sample data groups as training data, and construct and train an adaptive defect identification model based on the ensemble learning method, where the adaptive defect identification model includes multiple weak identifiers; at a preset defect identification point, obtain a real-time electrode surface image of the target production line through an image acquisition device, and transmit the real-time electrode surface image to the adaptive defect identification model for defect identification; obtain a defect identification result including multi-dimensional defect features, perform warning discrimination according to the preset production line electrode defect warning limit, and perform multi-path defect warning based on the discrimination result.

[0007] The present application provides a lithium battery electrode surface defect identification and early warning system, including:

[0008] An identification record acquisition module, which is used to acquire historical defect identification records based on the product feature set of the target production line, wherein the historical defect identification records correspond to a preset retrospective sampling window; a historical image data acquisition module, which is used to acquire historical image data based on the historical defect identification records, perform gray-scale enhancement processing on the historical image data, and extract multi-dimensional gray-scale historical image features based on the processing results; a data partitioning module, which is used to perform clustering analysis on historical monitoring data according to the extracted multi-dimensional gray-scale historical image features, partition the data into multiple clustering clusters, and output the multiple clustering clusters as multiple sample data groups; an identification model training module, which is used to use the multiple sample data groups as training data and construct and train an adaptive defect identification model based on the ensemble learning method, wherein the adaptive defect identification model includes multiple weak identifiers; a defect identification module, which is used to obtain the real-time electrode surface image of the target production line through an image acquisition device at a preset defect identification point, and transmit the real-time electrode surface image to the adaptive defect identification model for defect identification; an early warning discrimination module, which is used to obtain a defect identification result including multi-dimensional defect features, perform early warning discrimination according to the preset production line electrode defect early warning limit, and perform multi-path defect early warning based on the discrimination result.

[0009] It is intended to propose a lithium battery electrode surface defect identification and early warning method and system through the present application. First, the historical defect identification records of the target production line are acquired and gray-scale enhancement processing is performed to extract multi-dimensional image features; clustering analysis is performed on the historical data based on these features to obtain sample data groups; an adaptive defect identification model is trained using the ensemble learning method, and real-time images are acquired at preset points for defect identification; the identification result is judged according to multi-dimensional defect features, and defect early warning is performed, achieving the technical effects of efficiently and accurately identifying the complex and changeable lithium battery electrode surface and lithium battery separator defects, and significantly improving the pertinence, detection efficiency and detection accuracy of detection. Description of the Drawings

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments of the present invention will be briefly introduced below. Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations in the front or below do not necessarily need to be precisely executed in sequence. On the contrary, according to needs, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps of operations can be removed from these processes.

[0011] Figure 1 It is a schematic flow chart of a method for identifying and warning surface defects of a lithium battery electrode provided by an embodiment of the present application.

[0012] Figure 2 It is a schematic structural diagram of a system for identifying and warning surface defects of a lithium battery electrode provided by an embodiment of the present application.

[0013] Explanation of reference numerals: Identification record acquisition module 10, historical image data acquisition module 20, data division module 30, identification model training module 40, defect identification module 50, warning discrimination module 60. Detailed implementation manners

[0014] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specifically gives the detailed implementation manners of the present application.

[0015] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0016] In the following description, "some embodiments" are involved, which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first\second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0017] An embodiment of the present application provides a method for identifying and warning surface defects of a lithium battery electrode, as Figure 1 shown, the method includes:

[0018] Step S100: Obtain historical defect recognition records based on the product output feature set of the target production line, where the historical defect recognition records correspond to a preset retrospective sampling window. Specifically, first interact with the target production line to extract a product output feature set including time-series indexes of product quality and equipment failure. Then, perform time series analysis on it to obtain a product frequency-domain feature set. Adjust the sampling window size from small to large, and select the sampling window with a cumulative frequency greater than the preset frequency threshold as the retrospective sampling window. Taking the current moment as the sampling starting point, extract historical defect recognition records from the production management terminal based on the retrospective sampling window, and the records should cover various production scenarios and defect types to ensure data diversity and representativeness.

[0019] In a possible implementation, to obtain historical defect recognition records based on the product output feature set of the target production line, where the historical defect recognition records correspond to a preset retrospective sampling window, step S100 further includes step S110: Interact with the target production line to extract the product output feature set of the target production line, where the product output feature set of the target production line includes time-series indexes of product quality and equipment failure. Specifically, perform data interaction with the target production line, analyze the key information of the target production line, and thus extract the product output feature set of the target production line. The feature set covers time-series indexes of product quality and equipment failure. The time-series index of product quality can reflect the changes in the quality of the product over time during the production process, such as the fluctuation data of the product's dimensional accuracy in different production batches or different time periods, or the change sequence of the product's physical properties (such as strength, conductivity, etc.) in the production process. The time-series index of equipment failure records information such as the time nodes, failure types, and failure durations of various equipment on the production line during operation. These data are crucial for analyzing the stability and reliability of the production line.

[0020] Step S120: Perform time series analysis on the product output feature set of the target production line to obtain the product frequency-domain feature set of the target production line. Specifically, perform time series analysis on the extracted product output feature set of the target production line. The core role of time series analysis in this solution is to identify the periodic components. Through time series analysis methods such as Fourier transform, convert the time-series indexes of product quality and equipment failure originally in the time domain to the frequency domain, and then obtain the product frequency-domain feature set of the target production line. In the frequency domain, the periodic laws of the data can be observed more clearly. For example, whether certain quality problems occur frequently at specific production frequencies, or whether certain equipment failures show periodic occurrence laws.

[0021] Step S130: Based on the product frequency domain feature set, adjust the sampling window size from small to large, and select multiple sampling windows whose cumulative frequencies are all greater than a preset frequency threshold as the retrospective sampling windows. Herein, the multiple cumulative frequencies correspond to multiple index dimensions of the product quality time series index and the equipment failure time series index. Specifically, according to the obtained product frequency domain feature set, start the adjustment mechanism of the sampling window size. The adjustment process is carried out step by step from small to large. Calculate multiple cumulative frequencies, and the cumulative frequencies respectively correspond to multiple index dimensions of the product quality time series index and the equipment failure time series index. For the equipment failure index, by cumulatively calculating the product frequency domain features, if it is found that 95% (preset frequency threshold) of the failures show periodicity when the period is equal to T, then the sampling window size is determined as T at this time. Select multiple sampling windows whose cumulative frequencies are all greater than the preset frequency threshold. The selected sampling windows will be used as the retrospective sampling windows. The retrospective sampling window is essentially a sampling period within a specific time range determined according to the analysis requirements of the product stability or periodicity of the target production line. Its determination method can also be based on the distribution of historical defect identification records, and will be used later to accurately evaluate the defect occurrence law of the production line within a certain period of time.

[0022] Step S140: Taking the current moment as the sampling starting point, based on the retrospective sampling window, interact with the production management terminal of the target production line to extract the historical defect identification records. Specifically, taking the current moment as the sampling starting point, relying on the already determined retrospective sampling window, interact with the production management terminal of the target production line again. The purpose of this interaction is to extract historical defect identification records and the corresponding historical acquisition images. The historical defect identification records cover the detailed information of various defects that occurred on the production line during the time interval covered by the retrospective sampling window, such as the time when the defect was first detected, the specific location on the product, and the type it belongs to (such as surface scratches, dimensional deviations, material defects, etc.). The historical acquisition images are the product image data captured by the image acquisition equipment on the production line within the same retrospective sampling window time range. These images can intuitively show the appearance status of the product at different production stages, and play a key role in assisting the analysis of the specific situation where the defect occurs and more accurately locating the cause of the defect. Whether it is the morphological characteristics of the defect on the product surface or the comparison with the surrounding normal area can be clearly presented through the historical acquisition images. These historical defect identification records and historical acquisition images together provide a very important and comprehensive data basis for further exploring the root cause of the production line defects, predicting the occurrence of future defects, and formulating targeted improvement measures.

[0023] Step S200: Based on the historical defect recognition records, obtain historical image data, perform gray-scale enhancement processing on the historical image data, and extract multi-dimensional gray-scale historical image features based on the processing results. Specifically, based on the existing historical defect recognition records, extract the associated historical image data therefrom, which are derived from the images of products captured by image monitoring devices at different times and scenarios on the production line. Then, perform gray-scale enhancement processing on the historical image data. For example, use methods such as histogram equalization to redistribute the pixel gray-scale values, make the gray-scale histogram of the image evenly distributed, and highlight the details. Subsequently, based on the results of the gray-scale enhancement processing, use the gray-level co-occurrence matrix to extract multi-dimensional gray-scale historical image features. Among them, the information entropy can measure the complexity of the gray-scale distribution of the image. A large entropy value indicates that the gray-scale change of the image is complex, which is related to the variability of the texture structure on the electrode surface; the energy can reflect the uniformity of the gray-scale distribution and the thickness of the texture. A small energy value indicates fine texture, and a large energy value indicates uniform and regular texture; the inverse difference moment reflects the clarity and regularity of the texture. When the texture is regular, the inverse difference moment is large, and when there are defects, the inverse difference moment is small, providing a data basis for subsequent clustering analysis to explore the internal relationship between the image and product defects, and helping to identify and warn product defects.

[0024] In a possible implementation manner, based on the historical defect recognition records, obtain historical image data, perform gray-scale enhancement processing on the historical image data, and extract multi-dimensional gray-scale historical image features. Step S200 further includes step S210. The multi-dimensional gray-scale historical image features are extracted through the gray-level co-occurrence matrix, and the multi-dimensional gray-scale historical image features at least include information entropy, energy, and inverse difference. Specifically, after obtaining the historical image data processed by gray-scale enhancement, first construct a gray-level co-occurrence matrix, determine parameters such as the gray-level quantization series, calculation direction, and pixel pair distance, and then calculate the information entropy based on this matrix. According to the formula ENT = -∑ i ∑ j P(i, j)log(P(i, j)), which can reflect the complexity of the gray-scale distribution of the image. A large entropy value indicates that the gray-scale change of the image is complex and can be used to identify various situations on the electrode surface; then calculate the energy. According to the formula ASM = ∑∑P(i, j) 2 , the energy can reflect the uniformity of the gray-scale distribution and the thickness of the texture. The energy of the uniform and smooth electrode surface is large, and the energy is small when there are defects; finally, calculate the inverse difference. Through the formula It reflects the clarity and regularity of the texture. The inverse difference moment value of the electrode surface without defects is large, and it decreases when there are defects. Extracting these multi-dimensional gray-scale historical image features can comprehensively describe the texture characteristics, providing key data basis for subsequent clustering analysis and constructing a defect recognition model to accurately identify the defects on the surface of lithium battery electrodes.

[0025] Step S300: According to the extracted multi-dimensional gray-scale historical image features, perform clustering analysis on the historical monitoring data, divide the data into multiple clustering clusters, and output the multiple clustering clusters as multiple sample data groups. Specifically, first integrate the multi-dimensional gray-scale historical image features such as information entropy, energy, and inverse difference with the historical monitoring data, clean and preprocess the data, and process missing values and outliers to ensure accuracy and reliability. For density-based clustering algorithms such as DBSCAN, set parameters such as the neighborhood radius and the minimum number of points according to the data characteristics, determine the optimal values through multiple experiments and evaluation of clustering quality indicators, use this algorithm to cluster according to the density relationship of data points and visualize the results. According to the visualized evaluation effect, if it is not good, adjust the parameters or replace the algorithm. Then calculate the within-cluster consistency measurement indicators such as the average distance from the data points within the clustering cluster to the center, and discard the last 5% of the data in each cluster according to this to clean the data and reduce the influence of noise and outliers. Finally, output the cleaned clustering clusters as sample data groups, and their corresponding image feature categories provide a comprehensive and representative data basis for constructing a training adaptive defect recognition model, and improve the recognition and generalization ability of the model for the surface defects of lithium battery electrodes.

[0026] In a possible implementation manner, according to the extracted multi-dimensional gray-scale historical image features, perform clustering analysis on the historical monitoring data, divide the data into multiple clustering clusters, and output the multiple clustering clusters as multiple sample data groups. Step S300 further includes step S310: Adopt a density-based clustering method to process the multi-dimensional gray-scale historical image features and obtain multiple initial clustering clusters. Specifically, adopt a density-based clustering method to process the multi-dimensional gray-scale historical image features. Density-based clustering algorithms (such as DBSCAN) are based on the concept that clusters are formed in regions where data points are densely distributed in space, and data points in sparse regions are regarded as noise or do not belong to any cluster. When processing multi-dimensional gray-scale historical image features, the algorithm determines the clustering clusters based on the distance between data points and the number of data points within a certain distance range. For example, for the multi-dimensional data space composed of features such as the information entropy, energy, and inverse difference of an image, the algorithm calculates the number of data points within a specific radius (neighborhood radius eps) around each data point. If this number exceeds the set minimum number of points minPts, then this data point and the data points within its neighborhood may form the core part of a clustering cluster. As the algorithm progresses, data points that are density-connected to the core part will also be included in the corresponding clustering cluster, thereby obtaining multiple initial clustering clusters.

[0027] Step S320: Traverse the multiple initial clustering clusters. According to the within-cluster consistency metric, discard 5% of the data corresponding to the within-cluster consistency metric in each of the initial clustering clusters for data cleaning, and obtain multiple clustering clusters. Specifically, traverse the multiple initial clustering clusters and calculate the within-cluster consistency metric for each cluster. Multiple methods are used to measure the within-cluster consistency, such as calculating the sum or average distance of each data point to the center of its belonging clustering cluster. Taking the calculation of the average distance as an example, for an initial clustering cluster, first determine its clustering center (which can be the point formed by the average values of all data points in the cluster in each dimension), then calculate the distance of each data point in the cluster to this center (such as the Euclidean distance), and then calculate the average of these distances as the consistency metric for the cluster. According to this metric, perform data screening on each initial clustering cluster and discard 5% of the data corresponding to the within-cluster consistency metric. This is because these data points are relatively far from the clustering center and may be outliers or data that interfere with the representativeness of the clustering cluster. Discarding them can improve the data quality, make the data within each clustering cluster more compact and consistent, and thus obtain multiple cleaned clustering clusters.

[0028] Step S330: Output the multiple clustering clusters as multiple sample data groups, where each sample data group corresponds to one type of image feature category. Specifically, output the cleaned clustering clusters as multiple sample data groups. Each sample data group corresponds to one type of image feature category, and here the image feature category can be understood as a classification manifestation of data quality or data texture. For example, a sample data group may represent a type of data with relatively regular image texture and high data quality (such as the information entropy being within a certain range, the energy being relatively stable, and the inverse difference being large indicating clear texture), while another sample data group may correspond to the situation where the image texture is relatively complex and the data quality has certain fluctuations (such as a large information entropy, low energy, and small inverse difference indicating disordered texture). These sample data groups will serve as the basic data for subsequent model construction and training, providing strong support for accurately identifying different types of image features and product defects, and helping to improve the accuracy and reliability of the entire system for defect recognition on the surface of lithium battery electrodes.

[0029] Step S400: Using the multiple sample data groups as training data, construct and train an adaptive defect recognition model based on the ensemble learning method. Among them, the adaptive defect recognition model includes multiple weak recognizers. Specifically, first deeply analyze the multiple sample data groups to obtain distribution probability information and define it as sample weights. Then, adjust the sample weights according to the preset weight constraint limits. If the weight is greater than the upper limit, compress the data; if it is less than the lower limit, expand the samples. After that, generate a standard sample data set. Subsequently, normalize the sample weights and calculate the model construction ratio, and generate multiple weak recognizer clusters accordingly. Then, randomly sample from the standard sample data groups as the sample space for each weak recognizer cluster, and use the sampling results to train the weak recognizers. Algorithms such as decision trees and support vector machines can be selected and optimized. Finally, based on the ensemble learning method, integrate the output results of each weak recognizer cluster through a distributor connected to the multiple weak recognizer clusters to generate an adaptive defect recognition model. This model can allocate the input data features to the appropriate weak recognizer clusters for processing, thereby accurately identifying and adaptively processing the defects on the surface of lithium battery electrodes.

[0030] In a possible implementation, using the multiple sample data groups as training data, an adaptive defect recognition model is constructed and trained based on the ensemble learning method. The adaptive defect recognition model includes multiple weak recognizers. Step S400 further includes step S410, which analyzes and obtains the distribution probability information of the multiple sample data groups, defines it as the sample weights of the multiple sample data groups, and performs sample adjustment based on a preset weight constraint limit to obtain a standard sample data set. Specifically, analyze the multiple sample data groups, and determine the distribution probability information of each sample data group in the overall data set through statistical methods. For example, calculate the proportion of the number of data in each sample data group to the total amount of all sample data, and use this as its distribution probability. Define the distribution probability information as the sample weight of the corresponding sample data group. The sample weight can intuitively reflect the relative importance degree of each sample data group in the process of building the model. Perform sample adjustment according to the preset weight constraint limit. The weight constraint limit includes a constraint upper limit and a constraint lower limit. The two limits are determined by comprehensively considering various factors such as data balance, representativeness, and the effectiveness of model training. If the sample weight of a certain sample data group is greater than the constraint upper limit, it indicates that the proportion of this sample data group in the data set is too large, which may cause the model training to be overly biased towards this part of the data, resulting in overfitting. At this time, use data compression methods, such as randomly selecting some data or performing feature fusion on the data to reduce its dimension and reduce the data volume of this sample data group. On the contrary, if the sample weight is less than the constraint lower limit, it means that the representativeness of this sample data group in the data set is insufficient and may be ignored by the model. In response to this situation, use sample expansion means, such as copying existing data, performing data interpolation, or introducing data similar to this sample data group, to enhance its influence in the data set. After such adjustment, a standard sample data set is obtained, ensuring that each sample data group can play an appropriate role in model training and improving the overall quality and representativeness of the data.

[0031] Step S420: Normalize the sample weights, obtain the model construction ratios of multiple sample data groups, and generate multiple weak recognizer clusters based on the model construction ratios. Specifically, perform a normalization operation on the adjusted sample weights. The purpose of normalization is to make the sum of all sample weights equal to 1, so as to more accurately allocate resources in the subsequent model construction process. Through the normalized sample weights, calculate the model construction ratios of multiple sample data groups. The model construction ratio clarifies the resource ratio that each sample data group should be allocated when constructing a weak recognizer cluster. For example, if the model construction ratio of a sample data group is 0.25, it means that 25% of the resources (including data volume, computing resources, etc.) should be allocated to the part related to this sample data group when constructing a weak recognizer cluster. According to these model construction ratios, generate multiple weak recognizer clusters accordingly. Each weak recognizer cluster will focus on processing sample data groups with specific features and weights, thereby improving the model's processing efficiency and adaptability to different types of data.

[0032] Step S430: Respectively use multiple standard sample data groups in the standard sample dataset as the sample spaces of multiple weak recognizer clusters, conduct random sampling, and use the random sampling results as training data to train multiple weak recognizers in each weak recognizer cluster respectively. Specifically, respectively use multiple standard sample data groups in the standard sample dataset as the sample spaces of multiple weak recognizer clusters, and perform random sampling operations within each sample space. Random sampling can effectively increase the diversity of training data and reduce the risk of model overfitting. Use the results obtained from random sampling as training data to prepare for subsequent training of multiple weak recognizers in each weak recognizer cluster. Take the random sampling results as the training basis and train multiple weak recognizers in each weak recognizer cluster. A weak recognizer can be a simple classifier or a base model, such as a decision tree, a support vector machine, etc. During the training process, each weak recognizer will learn the specific relationship pattern between image texture features and product defects according to the characteristics of the sample data group to which it belongs. For example, a certain weak recognizer may be good at identifying defect situations in image data with specific texture complexity and energy features. By continuously optimizing the model parameters, it can more accurately identify defects in corresponding types of data.

[0033] Step S440: Based on the ensemble learning method, generate the adaptive defect recognition model according to multiple trained weak recognizer clusters, wherein the adaptive defect recognition model is embedded with a distributor connected to the multiple weak recognizer clusters. Specifically, based on the ensemble learning method, construct an adaptive defect recognition model according to multiple trained weak recognizer clusters. Ensemble learning can significantly improve the accuracy and generalization ability of the model by integrating the output results of multiple weak recognizers. Inside the adaptive defect recognition model, there is a distributor connected to the multiple weak recognizer clusters. The core function of the distributor is to accurately allocate the input data to the corresponding weak recognizer cluster according to the image feature category of the input data. For example, when an image data of the surface of a lithium battery electrode is input, the distributor will quickly determine which sample data group the texture features, energy features, etc. of the image data match most, and then allocate it to the corresponding weak recognizer cluster. Each weak recognizer cluster respectively performs recognition processing on the input data and outputs its own results. Finally, the model integrates and analyzes these results to determine whether there are defects on the electrode surface, as well as the type, degree, etc. of the defects, realizing the efficient and accurate recognition and adaptive processing of the defects on the surface of the lithium battery electrode.

[0034] In a possible implementation manner, analyze and obtain the distribution probability information of multiple sample data groups, define it as the sample weights of the multiple sample data groups, and perform sample adjustment based on a preset weight constraint limit to obtain a standard sample data set. Step S410 further includes step S411: respectively compare multiple sample weights with the preset weight constraint limit, wherein the weight constraint limit includes a constraint upper limit and a constraint lower limit. Specifically, take out multiple sample weights one by one and carefully compare them with the preset weight constraint limit. The constraint upper limit and the constraint lower limit in the weight constraint limit are determined by comprehensively considering various factors such as the structural analysis of the entire data set, the diversity requirements of the data, and the stability requirements of subsequent model training. For example, if through a large number of experiments and data analysis, it is found that when the weight of a certain sample data group exceeds 0.3 (this is an example of the constraint upper limit), the model will overly rely on this group of data during training, resulting in a decrease in the recognition ability for other types of data; while when the weight is lower than 0.1 (this is an example of the constraint lower limit), the role of this sample data group in model training is negligible and it cannot fully reflect the image feature category information it represents.

[0035] Step S412: If the sample weight is greater than the constraint upper limit, data compression is performed on the corresponding sample data group; if the sample weight is less than the constraint upper limit, sample expansion is performed on the corresponding sample data group. Specifically, when it is found that the weight of a certain sample is greater than the constraint upper limit, data compression needs to be performed on the corresponding sample data group. The purpose of data compression is to reduce the data volume and its influence in the overall data set while retaining the core feature information of the sample data group. Various data compression methods can be used, such as random sampling, that is, randomly selecting a certain proportion (such as determining the sampling proportion according to the degree of weight exceeding the upper limit) of data points from the sample data group for retention and discarding other data points; or using a feature selection algorithm to screen out some features that contribute more to the characteristics of the sample data group, thereby reducing the data dimension and achieving data volume compression. For example, for a sample data group with a sample weight of 0.35 (greater than the constraint upper limit of 0.3), if the random sampling method is used, perhaps 70% of the data points will be selected and only 70% of the original data volume will be retained, making the weight of the sample data group close to or lower than the constraint upper limit after adjustment. If the sample weight is less than the constraint upper limit, sample expansion is performed on the corresponding sample data group. The purpose of sample expansion is to enhance the representativeness of the sample data group in the data set so that it can better provide effective information for model training. Sample expansion includes methods such as mutation expansion, resampling, and collecting data from homologous production lines. Mutation expansion can generate new data points by making small random changes to the data points in the sample data group (such as adding a small range of random noise in a certain dimension of the image feature data), increasing data diversity; resampling is to repeatedly draw data points from the sample data group with replacement and merge the drawn data points with the original data group to expand the scale of the data group; collecting data from homologous production lines is to obtain relevant data from other production lines with similar production conditions and processes as the current production line and add it to the sample data group to further enrich the information of the data group. For example, for a sample data group with a sample weight of 0.08 (less than the constraint upper limit of 0.1), if the resampling method is used, assuming the original data group has 100 data points, it may increase to 150 data points after resampling, enhancing the weight of the sample data group in the data set.

[0036] Step S413: Output the data compression result and the sample expansion result as the standard sample data set. Specifically, integrate the result after data compression processing and the result after sample expansion to form the standard sample data set. This standard sample data set has a more reasonable sample weight distribution, and each sample data group can provide balanced and effective data support for subsequent model construction and training within an appropriate weight range, helping to improve the adaptability and accuracy of the model to different image feature category data, thereby enhancing the performance of the entire defect recognition system.

[0037] Step S500: At the preset defect recognition points, use an image acquisition device to obtain the real-time electrode surface image of the target production line, and transmit the real-time electrode surface image to the adaptive defect recognition model for defect recognition. Specifically, determine the preset defect recognition points based on in-depth analysis of the target production line and accurate prediction of the areas where product defects are likely to occur. Then, install image acquisition devices such as industrial cameras with high resolution, good imaging quality, stable performance, and the ability to adapt to the production environment at key positions. Continuously capture the electrode surface of the target production line at a specific acquisition frequency (such as 10 frames per second) to obtain the real-time electrode surface image. During this process, perform preliminary image processing such as adjusting brightness and contrast. Transmit the image to the computing system where the adaptive defect recognition model is located through high-speed data transmission lines (such as industrial Ethernet and dedicated data lines), and use data verification and retransmission mechanisms to ensure the integrity and timeliness of the data. Once the image data is transmitted successfully, immediately trigger the adaptive defect recognition model to perform defect recognition processing, so as to achieve real-time monitoring and rapid identification of defects on the electrode surface of the target production line, timely detect quality problems in production, and implement measures such as alarming operators and automatically adjusting production process parameters, thereby ensuring product quality and production efficiency.

[0038] Step S600: Obtain the defect recognition result including multi-dimensional defect features, conduct early warning discrimination according to the preset warning limits for production line electrode defects, and perform multi-path defect early warning based on the discrimination result. Specifically, after obtaining the defect recognition result including multi-dimensional defect features such as defect category, area, and quantity from the adaptive defect recognition model, conduct discrimination based on the warning limits for production line electrode defects set by considering various factors. Compare each dimension of defect features with the corresponding warning limits one by one. If the result shows that there are defects, that is, the warning limits are not met, trigger the multi-path defect early warning mechanism, including sending visual and auditory alarms to operators, transmitting early warning information to the production management system to record defect details for subsequent analysis and optimization, and also sending it to the mobile terminal of the relevant person in charge through network communication for their timely decision-making, thereby effectively preventing a large number of defective products from appearing and ensuring product quality and production efficiency.

[0039] In a possible implementation, a defect recognition result including multi-dimensional defect features is obtained, early warning discrimination is performed according to a preset production line electrode defect early warning limit, and multi-path defect early warning is performed based on the discrimination result. Step S600 further includes step S610, where a defect recognition result including multi-dimensional defect features is obtained, early warning discrimination is performed according to a preset production line electrode defect early warning limit, and multi-path defect early warning is performed based on the discrimination result. Among them, the multi-dimensional defect features at least include defect category features, defect area features, and defect quantity features. When any one of the multi-dimensional defect features does not meet the production line electrode defect early warning limit, the discrimination result is that there is a defect, and a corresponding early warning instruction is generated for multi-path defect early warning. Specifically, an adaptive defect recognition model is used to process the real-time electrode surface image of the target production line to obtain a defect recognition result including defect category, area, and quantity features. According to the production line electrode defect early warning limit determined based on comprehensive product quality standards, production experience, and customer requirements, the defect category, area, and quantity features are compared with it. If any one of the features exceeds the corresponding early warning limit, the discrimination result is that there is a defect and an early warning instruction is generated. Immediately, a warning pop-up window will appear on the production site monitoring system and an alarm sound will sound, and a defect data report will be sent to the production management system for traceability and process optimization. It will also be sent to the mobile devices of management personnel through text messages or push notifications, etc., so that they can organize technical personnel to troubleshoot, adjust process parameters, or arrange equipment maintenance, reduce the output of defective products, and ensure production and product quality.

[0040] In the embodiment of the present application, the historical defect recognition records of the target production line are obtained and gray-scale enhancement processing is performed to extract multi-dimensional image features; clustering analysis is performed on the historical data based on these features to obtain a sample data group; an adaptive defect recognition model is trained using an ensemble learning method, and real-time images are obtained at preset points for defect recognition; the recognition result is judged according to the multi-dimensional defect features, and defect early warning is performed, achieving the technical effects of efficiently and accurately identifying complex and changeable surface defects of lithium battery electrodes, and significantly improving the pertinence, detection efficiency, and detection accuracy of detection.

[0041] In the above text, with reference to Figure 1 A method for identifying and warning surface defects of lithium battery electrodes according to an embodiment of the present invention is described in detail. Next, with reference to Figure 2 A system for identifying and warning surface defects of lithium battery electrodes according to an embodiment of the present invention will be described.

[0042] A lithium battery electrode surface defect identification and early warning system according to an embodiment of the present invention solves the technical problems in the prior art that in the face of complex and changeable defects, there is a lack of detection pertinence, and the detection efficiency and detection accuracy are not good, and achieves the technical effects of efficiently and accurately identifying the complex and changeable defects on the surface of lithium battery electrodes and lithium battery diaphragms, and significantly improving the detection pertinence, detection efficiency and detection accuracy. A lithium battery electrode surface defect identification and early warning system includes: an identification record acquisition module 10, a historical image data acquisition module 20, a data division module 30, an identification model training module 40, a defect identification module 50, and an early warning discrimination module 60.

[0043] The identification record acquisition module 10 is used to acquire historical defect identification records based on the production characteristics set of the target production line, wherein the historical defect identification records correspond to a preset retrospective sampling window.

[0044] The historical image data acquisition module 20 is used to acquire historical image data based on the historical defect identification records, perform gray-scale enhancement processing on the historical image data, and extract multi-dimensional gray-scale historical image features based on the processing results.

[0045] The data division module 30 is used to perform cluster analysis on the historical monitoring data according to the extracted multi-dimensional gray-scale historical image features, divide the data into multiple cluster clusters, and output the multiple cluster clusters as multiple sample data groups.

[0046] The identification model training module 40 is used to use the multiple sample data groups as training data, construct and train an adaptive defect identification model based on the ensemble learning method, wherein the adaptive defect identification model includes multiple weak identifiers.

[0047] The defect identification module 50 is used to obtain the real-time electrode surface image of the target production line through an image acquisition device at a preset defect identification point, and transmit the real-time electrode surface image to the adaptive defect identification model for defect identification.

[0048] The early warning discrimination module 60 is used to obtain a defect identification result including multi-dimensional defect features, perform early warning discrimination according to a preset production line electrode defect early warning limit, and perform multi-path defect early warning based on the discrimination result.

[0049] Next, the specific configuration of the recognition record acquisition module 10 will be described in detail. As described above, historical defect recognition records based on the product output feature set of the target production line are obtained, where the historical defect recognition records correspond to a preset retrospective sampling window. The recognition record acquisition module 10 further includes: a feature set extraction unit for interacting with the target production line to extract the product output feature set of the target production line, where the product output feature set of the target production line includes production quality time series indicators and equipment failure time series indicators; a time series analysis unit for performing time series analysis on the product output feature set of the target production line to obtain the product frequency domain feature set of the target production line; a retrospective sampling window determination unit for, based on the product frequency domain feature set, adjusting the sampling window size from small to large, and selecting multiple sampling windows with cumulative frequencies all greater than a preset frequency threshold as the retrospective sampling window, where the multiple cumulative frequencies correspond to multiple index dimensions of the production quality time series indicators and the equipment failure time series indicators; and a production management terminal interaction unit for, with the current moment as the sampling starting point, based on the retrospective sampling window, interacting with the production management terminal of the target production line to extract the historical defect recognition records.

[0050] Next, the specific configuration of the historical image data acquisition module 20 will be described in detail. As described above, based on the historical defect recognition records, historical image data is obtained, gray-scale enhancement processing is performed on the historical image data, and based on the processing result, multi-dimensional gray-scale historical image features are extracted. The historical image data acquisition module 20 further includes: a gray-level co-occurrence matrix extraction unit for extracting the multi-dimensional gray-scale historical image features through a gray-level co-occurrence matrix, and the multi-dimensional gray-scale historical image features at least include information entropy, energy, and inverse difference.

[0051] Next, the specific configuration of the data partitioning module 30 will be described in detail. As described above, according to the extracted multi-dimensional gray-scale historical image features, clustering analysis is performed on the historical monitoring data, the data is partitioned into multiple clustering clusters, and the multiple clustering clusters are output as multiple sample data groups. The data partitioning module 30 further includes: an initial clustering cluster acquisition unit for processing the multi-dimensional gray-scale historical image features using a density-based clustering method to obtain multiple initial clustering clusters; a data cleaning unit for traversing the multiple initial clustering clusters and discarding 5% of the data corresponding to the intra-cluster consistency metric in each initial clustering cluster for data cleaning to obtain multiple clustering clusters; and a sample data group output unit for outputting the multiple clustering clusters as multiple sample data groups, where each sample data group corresponds to a type of image feature category.

[0052] Next, the specific configuration of the recognition model training module 40 will be described in detail. As described above, using multiple sample data groups as training data, an adaptive defect recognition model is constructed and trained based on the ensemble learning method. Among them, the adaptive defect recognition model includes multiple weak recognizers. The recognition model training module 40 further includes: a distribution probability information analysis unit, which is used to analyze and obtain the distribution probability information of multiple sample data groups, defined as the sample weights of multiple sample data groups, and perform sample adjustment based on a preset weight constraint limit to obtain a standard sample data set; a model construction ratio acquisition unit, which is used to normalize the sample weights, obtain the model construction ratios of multiple sample data groups, and generate multiple weak recognizer clusters correspondingly based on the model construction ratios; a random sampling unit, which is used to respectively use multiple standard sample data groups in the standard sample data set as the sample spaces of multiple weak recognizer clusters for random sampling, and use the random sampling results as training data to respectively train multiple weak recognizers in each weak recognizer cluster; an adaptive defect recognition model generation unit, which is used to generate the adaptive defect recognition model based on the ensemble learning method according to multiple trained weak recognizer clusters. Among them, the adaptive defect recognition model is embedded with a distributor connected to multiple weak recognizer clusters.

[0053] Among them, analyzing and obtaining the distribution probability information of multiple sample data groups, defined as the sample weights of multiple sample data groups, and performing sample adjustment based on a preset weight constraint limit to obtain a standard sample data set. The distribution probability information analysis unit further includes: a weight constraint limit comparison subunit, which is used to respectively compare the sample weights of multiple sample data groups with the preset weight constraint limit. Among them, the weight constraint limit includes a constraint upper limit and a constraint lower limit; a data compression subunit, which is used to perform data compression on the corresponding sample data group if the sample weight is greater than the constraint upper limit, and perform sample expansion on the corresponding sample data group if the sample weight is less than the constraint upper limit; a standard sample data set determination subunit, which is used to output the data compression result and the sample expansion result as the standard sample data set.

[0054] Next, the specific configuration of the early warning discrimination module 60 will be described in detail. As described above, a defect recognition result including multi-dimensional defect features is obtained, early warning discrimination is performed according to a preset production line electrode defect early warning limit, and multi-path defect early warning is performed based on the discrimination result. The early warning discrimination module 60 further includes: a multi-path defect early warning generation unit, which is used to obtain a defect recognition result including multi-dimensional defect features, perform early warning discrimination according to a preset production line electrode defect early warning limit, and perform multi-path defect early warning based on the discrimination result. Among them, the multi-dimensional defect features at least include defect category features, defect area features, and defect quantity features. When any one of the multi-dimensional defect features does not meet the production line electrode defect early warning limit, the discrimination result is that there is a defect, and a warning instruction is correspondingly generated for multi-path defect early warning.

[0055] The lithium battery electrode surface defect recognition and early warning system provided by the embodiment of the present invention can execute the lithium battery electrode surface defect recognition and early warning method provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution of the method.

[0056] Although this application makes various references to certain modules in the system according to the embodiments of this application, however, any number of different modules can be used and run on the user terminal and / or the server. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0057] The above specific implementation manners do not constitute a limitation to the protection scope of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of this application shall be included within the protection scope of this application. In some cases, the actions or steps recorded in this application can be executed in a different order from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In certain embodiments, multi-tasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for identifying and warning surface defects of a lithium battery electrode, characterized in that, The method includes: Obtaining historical defect recognition records based on the product output feature set of the target production line, where the historical defect recognition records correspond to a preset retrospective sampling window; Based on the historical defect recognition records, obtaining historical image data, performing grayscale enhancement processing on the historical image data, and extracting multi-dimensional grayscale historical image features based on the processing results; According to the extracted multi-dimensional grayscale historical image features, performing clustering analysis on historical monitoring data, dividing the data into multiple clustering clusters, and outputting the multiple clustering clusters as multiple sample data groups; Using the multiple sample data groups as training data, constructing and training an adaptive defect recognition model based on an ensemble learning method, where the adaptive defect recognition model includes multiple weak recognizers; At a preset defect recognition point, obtaining a real-time electrode surface image of the target production line through an image acquisition device, and transmitting the real-time electrode surface image to the adaptive defect recognition model for defect recognition; Obtaining a defect recognition result including multi-dimensional defect features, performing early warning discrimination according to a preset production line electrode defect early warning limit, and performing multi-path defect early warning based on the discrimination result.

2. The method for identifying and warning surface defects of a lithium battery electrode according to claim 1, wherein Obtaining historical defect recognition records based on the product output feature set of the target production line, where the historical defect recognition records correspond to a preset retrospective sampling window, including: Interacting with the target production line to extract the product output feature set of the target production line, where the product output feature set of the target production line includes product quality time series indicators and equipment failure time series indicators; Performing time series analysis on the product output feature set of the target production line to obtain the product frequency domain feature set of the target production line; Based on the product frequency domain feature set, adjusting the sampling window size from small to large, and selecting multiple sampling windows with cumulative frequencies all greater than a preset frequency threshold as the retrospective sampling window, where the multiple cumulative frequencies correspond to multiple index dimensions of the product quality time series indicator and the equipment failure time series indicator; Taking the current moment as the sampling starting point, based on the retrospective sampling window, interacting with the production management terminal of the target production line to extract the historical defect recognition records.

3. The method for identifying and warning surface defects of a lithium battery electrode according to claim 2, characterized in that The multi-dimensional grayscale historical image features are extracted through a gray-level co-occurrence matrix, and the multi-dimensional grayscale historical image features at least include information entropy, energy, and inverse difference.

4. The method for identifying and warning surface defects of a lithium battery electrode according to claim 3, wherein, According to the extracted multi-dimensional grayscale historical image features, performing clustering analysis on historical monitoring data, dividing the data into multiple clustering clusters, and outputting the multiple clustering clusters as multiple sample data groups, including: Using a density-based clustering method to process the multi-dimensional grayscale historical image features to obtain multiple initial clustering clusters; Traversing the multiple initial clustering clusters, according to the within-cluster consistency metric, discarding 5% of the data corresponding to the within-cluster consistency metric in each initial clustering cluster for data cleaning to obtain multiple clustering clusters; Outputting the multiple clustering clusters as multiple sample data groups, where each sample data group corresponds to a type of image feature category.

5. The method for identifying and warning surface defects of a lithium battery electrode according to claim 4, characterized in that, Using the multiple sample data groups as training data, constructing and training an adaptive defect recognition model based on an ensemble learning method, including: Analyze and obtain the distribution probability information of multiple said sample data groups, which is defined as the sample weights of multiple said sample data groups, and perform sample adjustment based on a preset weight constraint limit to obtain a standard sample data set; Normalize the sample weights, obtain the model construction ratios of multiple said sample data groups, and generate multiple weak recognizer clusters based on the model construction ratios; Use multiple standard sample data groups in the standard sample data set as the sample spaces of multiple said weak recognizer clusters respectively, conduct random sampling, and use the random sampling results as training data to train multiple weak recognizers in each said weak recognizer cluster respectively; Based on the ensemble learning method, generate the adaptive defect recognition model according to multiple said weak recognizer clusters that have been trained, wherein the adaptive defect recognition model is embedded with a distributor connected to multiple said weak recognizer clusters.

6. The method for identifying and warning surface defects of a lithium battery electrode according to claim 5, wherein, Performing sample adjustment based on a preset weight constraint limit includes: Compare the sample weights of multiple said sample data groups with the preset weight constraint limit respectively, wherein the weight constraint limit includes a constraint upper limit and a constraint lower limit; If the sample weight is greater than the constraint upper limit, perform data compression on the corresponding sample data group. If the sample weight is less than the constraint upper limit, perform sample expansion on the corresponding sample data group; Output the data compression result and the sample expansion result as the standard sample data set.

7. The method for identifying and warning surface defects of a lithium battery electrode according to claim 1, wherein Obtain a defect recognition result including multi-dimensional defect features, conduct early warning discrimination according to a preset production line electrode defect early warning limit, and perform multi-path defect early warning based on the discrimination result. Among them, the multi-dimensional defect features at least include defect category features, defect area features, and defect quantity features. When any one of the multi-dimensional defect features does not meet the production line electrode defect early warning limit, the discrimination result is that there is a defect, and a corresponding early warning instruction is generated for multi-path defect early warning.

8. A lithium battery electrode surface defect identification and early warning system, characterized in that, The system is used to implement the method for identifying and warning surface defects of lithium battery electrodes according to any one of claims 1-7. The system includes: An identification record acquisition module, which is used to acquire historical defect identification records based on the product feature set of the target production line, wherein the historical defect identification records correspond to a preset retrospective sampling window; A historical image data acquisition module, which is used to acquire historical image data based on the historical defect identification records, perform gray-scale enhancement processing on the historical image data, and extract multi-dimensional gray-scale historical image features based on the processing results; A data partitioning module, which is used to perform clustering analysis on historical monitoring data according to the extracted multi-dimensional gray-scale historical image features, partition the data into multiple clustering clusters, and output multiple said clustering clusters as multiple sample data groups; An identification model training module, which is used to use multiple said sample data groups as training data and construct and train an adaptive defect recognition model based on the ensemble learning method, wherein the adaptive defect recognition model includes multiple weak recognizers; A defect recognition module, which is used to obtain the real-time electrode surface image of the target production line at the preset defect recognition points through an image acquisition device, and transmit the real-time electrode surface image to the adaptive defect recognition model for defect recognition; A warning discrimination module, which is used to obtain the defect recognition result including multi-dimensional defect features, perform warning discrimination according to the preset production line electrode defect warning limit, and perform multi-path defect warning based on the discrimination result.

Images