Intelligent defect detection and quality control method, equipment and medium in photovoltaic production

By integrating sensors and data acquisition systems and combining deep learning algorithms to detect defects of photovoltaic cells, the data complexity and defect identification problems in photovoltaic manufacturing are solved, and fast and accurate defect identification and quality control are achieved, which significantly reduces the rate of defective products. Improve production efficiency.

CN118961748AInactive Publication Date: 2024-11-15HEFEI ZHE TOWER TECH CO LTD +1
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411015513.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The data volume in the photovoltaic cell manufacturing process is large and complex, and traditional quality control methods are difficult to identify the root causes of defects in a timely and accurate manner, resulting in high defective product rates and low production efficiency.

Method used

Integrated sensors and data acquisition systems are used to obtain EL data in real time, and deep learning algorithms are used to combine machine learning and artificial intelligence technology for defect detection, providing visual tools and report generation functions to quickly and accurately identify defect types and causes of poor products.

Benefits of technology

It significantly improves the efficiency and accuracy of defect detection, reduces the rate of defective products, improves production efficiency and quality control capabilities, helps photovoltaic manufacturing companies trace and solve defect sources, and promotes industry development.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118961748A_ABST
    Figure CN118961748A_ABST
Patent Text Reader

Abstract

According to the intelligent defect detection and quality control method and equipment in photovoltaic production, defect detection and classification are carried out on the data based on electroluminescence detection data in the photovoltaic manufacturing industry. According to the method, the ADC technology and an advanced data analysis algorithm are combined, so that the defective products are accurately positioned. Specifically, by acquiring and analyzing EL data in real time, the defect type of the photovoltaic cell can be quickly and accurately identified and classified, and the efficiency and accuracy of defect detection are remarkably improved. For a large amount of complex data generated in the photovoltaic cell manufacturing process, an efficient data management and processing method is adopted. Through the integrated sensor and the data acquisition system, key data in the production process are acquired and processed in real time, and comprehensive data support is provided for defect detection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic manufacturing technology, and in particular to an intelligent defect detection and quality control method, equipment and storage medium in photovoltaic production. Background Art

[0002] In the field of photovoltaic manufacturing, ensuring product quality is of vital importance. However, the manufacturing process of photovoltaic cells is also faced with variations in process parameters, equipment status and other factors, which may lead to defects in cells and affect the performance and quality of photovoltaic products. Traditional quality control methods mainly rely on process monitoring and product testing, but these methods may not be able to identify the root cause of defects in a timely and accurate manner, resulting in an increase in the rate of defective products and a decrease in production efficiency.

[0003] To solve this problem, defect detection and analysis technology based on ADC (Automated Defect Classification) has been introduced in recent years. ADC technology uses advanced image processing and machine learning algorithms to monitor and analyze key parameters and image data in the photovoltaic cell manufacturing process in real time. By analyzing these data, defects such as hidden cracks, concentric circles, dirt, black spots, black dots, and edge dirt on the surface of the cell can be accurately identified and located, which is conducive to the later tracing of the cause of defective products.

[0004] However, existing ADC technology still faces some challenges in the field of photovoltaic manufacturing. First, the amount of data in the photovoltaic cell manufacturing process is huge and complex, containing multiple parameters and image features, so efficient data management and processing methods are required. Secondly, defect classification requires accurate models and algorithms to identify the root type of defects and perform real-time analysis in complex production environments. In addition, the traceability of fault diagnosis and corrective measures is also one of the challenges, especially in large-scale production, where the source of defects needs to be found quickly and accurately and corresponding measures need to be taken to minimize the number of defective products.

[0005] Therefore, further research and development of semi-automated ADC defect detection and analysis technology is needed to improve the quality control capabilities of photovoltaic industry cell manufacturing, reduce the defective product rate, and improve production efficiency. This technology will bring a higher level of quality assurance and production optimization to the photovoltaic industry. Summary of the invention

[0006] The present invention proposes an intelligent defect detection and quality control method, device and storage medium in photovoltaic production, which can solve at least one of the technical problems in the background technology.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] An intelligent defect detection and quality control method in photovoltaic production, first of all, the present invention acquires EL data in real time by integrating sensors and data acquisition systems in the photovoltaic manufacturing process of solar cells. These data include process parameters of cell production, sensor readings, equipment status, etc., providing comprehensive monitoring of the entire production process.

[0009] Secondly, based on the acquired EL data, the present invention uses an advanced deep learning algorithm for defect detection. The algorithm combines machine learning, statistical analysis, and artificial intelligence technology to perform high-performance processing and analysis of large-scale EL data. By performing feature extraction and defect detection on the data, the present invention can quickly and accurately identify the type of defect and the cause of defective products.

[0010] Furthermore, the present invention provides visualization tools and report generation functions for intuitively presenting defect detection results. Through charts, trend analysis, and fault trees, engineers and operators can clearly understand the root causes of defective products and take corresponding corrective measures.

[0011] This paper describes an analysis process for defect detection in the photovoltaic cell manufacturing process under limited hardware resources (only 4 NVIDIA A10 GPUs). The process mainly includes the following steps:

[0012] S1: Information acquisition, real-time testing and diagnosis of produced cells, such as optical testing, to obtain EL data information of cells;

[0013] Optical testing includes SEM, TEM, etc.;

[0014] S2: Get EL data and configuration files;

[0015] S3: Manually collect defect type image data, perform model training using a specified algorithm (here the EfficientNetV2 algorithm is used), and manually import the trained model into the inference platform;

[0016] S4: Parse and process the EL data requested in real time to obtain the data of the algorithm structure;

[0017] S5: Further perform defect detection on the preprocessed data and give the predicted defect type based on the trained model;

[0018] S6: Further manual review is performed to verify and determine the root cause of the failure based on the defect type;

[0019] S7: Summarize the results of the re-judgment to obtain the pass / miss scores. For models with scores lower than a given threshold (e.g., 99.7%), update the training set data and retrain, and repeat steps S3 to S6.

[0020] S8: Take measures, take corresponding measures according to the root cause of the defect.

[0021] Further, in step S1: the request method of information condition, the following step S11:

[0022] S11: Filter the corresponding product name, production line, equipment name, and machine name within the specified time, and perform EL testing and diagnostic operations on the battery cell, such as optical testing, to obtain the history information of the battery cell device;

[0023] Further, in step S2: according to the data requested in step S1, EL data and configuration files in the photovoltaic cell manufacturing process are obtained, which specifically includes the following steps S21 to S23:

[0024] S21: After the back-end ADC parsing data interface obtains the EL data within the specified time, it parses the information of the corresponding table structure;

[0025] S22: Through step S21, redundant data is filtered out, and data is stored according to the structural information in the database table;

[0026] S23: through step S22, extract the image path and the corresponding unique identification ID information, and produce the information in the form of Kafka communication for subsequent inference service consumption;

[0027] Further, in step S3: the historical EL image data is manually reviewed and classified and collected, model training is performed and the algorithm model is exported to the reasoning service, which specifically includes the following steps S31 to S33;

[0028] S31: Manually collect typical defect category image data,

[0029] S32: Based on step S31, the collected image data is uploaded to the specified training set path, and the training script is started to perform algorithm model training;

[0030] S33: Based on step S32, manually import the trained model into the inference platform;

[0031] Further, in step S4: the EL data requested in real time is parsed and processed to obtain data of the algorithm structure, which specifically includes the following steps S41 to S43:

[0032] S41: pre-processing the data of the request service monitored in real time, and parsing the image path;

[0033] S42: Based on step S41, when the image path file cannot be found, returning content with an identifier "image not found" and a classification result of "None";

[0034] S43: Based on step S42, when the image path file is found, the image path in the request is processed into an inferable image array;

[0035] Furthermore, in step S5: a predicted defect type is given according to the trained model, which specifically includes the following steps S51 to S53:

[0036] S51: Based on the classification algorithm model of step S33, further performing defect detection prediction on the data pre-processed based on step S41;

[0037] S52: Based on step S51, the classification result predicted by the image, together with the unique ID information of the image and the correlation coefficient of the classification result, are packaged and returned;

[0038] S53: Based on step S51, if the correlation coefficient of the classification prediction does not reach the given minimum threshold (set as 0.1 here), the category is thrown out as "other" for manual re-judgment;

[0039] Furthermore, in step S6: according to the defect type, further manual review is performed to verify and determine the root cause of the failure, which specifically includes the following steps S61 to S62:

[0040] S61: Manual re-judgment is performed by manually selecting a specified product name, production line, equipment name, machine name, and a specified number of images of different defect types in a specified time period;

[0041] S62: Based on the image of the defect type in step S61, an experienced engineer determines the root cause of the failure;

[0042] Furthermore, in step S7: the results of the re-judgment are summarized, and the training data of the models that do not meet the standards are collected and re-trained. The specific contents include the following steps S71 to S73:

[0043] S71: Summarize the results of the re-judgment to obtain the pass / miss detection score;

[0044] S72: For the model whose score in step S71 is higher than a given threshold, continue to monitor new request services;

[0045] S73: For the model whose score in step S71 is lower than a given threshold, the training set data is updated and the model is retrained, and steps S3 to S6 are repeated;

[0046] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0047] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0048] It can be seen from the above technical solutions that the intelligent defect detection and quality control method in photovoltaic production of the present invention specifically provides a method based on electroluminescence (EL) detection data in photovoltaic manufacturing industry, and performs defect detection and classification on the data. The method combines ADC technology and advanced data analysis algorithms to achieve accurate positioning of defective products.

[0049] Specifically, by acquiring and analyzing EL data in real time, the present invention can quickly and accurately identify and classify the defect types of photovoltaic cells, significantly improving the efficiency and accuracy of defect detection.

[0050] In the face of the large amount of complex data generated during the manufacturing process of photovoltaic cells, the present invention adopts an efficient data management and processing method. By integrating sensors and data acquisition systems, key data in the production process can be acquired and processed in real time, providing comprehensive data support for defect detection.

[0051] The present invention uses advanced deep learning algorithms for defect detection, which can perform high-performance processing and analysis of large-scale EL data. The application of deep learning algorithms makes defect identification more accurate, helping engineers and operators to quickly find the cause of defective products and take corresponding corrective measures.

[0052] The present invention provides visualization tools and report generation functions for intuitively presenting defect detection results. Through charts, trend analysis, and fault trees, users can clearly understand the root causes of defective products and take corresponding measures to improve quality control capabilities during production.

[0053] By applying the method of the present invention, photovoltaic manufacturing companies can significantly reduce the defective product rate, improve production efficiency and product quality. Advanced defect detection technology helps manufacturers trace and solve the problems caused by defective products, and promote the development of the photovoltaic manufacturing industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flowchart of the present invention;

[0055] Figure 2The figure is a schematic diagram of the process of an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0057] Defect detection and product quality control of photovoltaic cells are crucial in the manufacturing process. Traditional detection methods are difficult to accurately identify and locate the root cause of defects, so an efficient defect detection technology is needed to help manufacturers trace and solve problems caused by defective products. This technology can effectively improve the performance and quality of photovoltaic products and promote the development of the photovoltaic manufacturing industry.

[0058] refer to Figure 1 to Figure 2 The present invention proposes a semi-automatic defect detection method for EL data of photovoltaic cell products based on deep learning, including the following steps S1 to S8:

[0059] Information acquisition, real-time testing and diagnosis operations are performed on the produced cells, such as optical testing, to obtain EL data information of the cells, as follows step S11:

[0060] S11: Filter the corresponding product name, production line, equipment name, and machine name within the specified time, and perform EL testing and diagnostic operations on the battery cell, such as optical testing, to obtain the history information of the battery cell device;

[0061] According to step S1, requesting data, obtaining EL data and configuration files in the photovoltaic cell manufacturing process, specifically includes the following steps S21 to S23:

[0062] S21: After the back-end ADC parsing data interface obtains the EL data within the specified time, it parses the information of the corresponding table structure;

[0063] S22: Through step S21, redundant data is filtered out, and data is stored according to the structural information in the database table;

[0064] S23: through step S22, extract the image path and the corresponding unique identification ID information, and produce the information in the form of Kafka communication for subsequent inference service consumption;

[0065] Manually collect defect type image data, and then perform model training for the specified algorithm (here the improved version of EfficientNet algorithm is used). The improved version of EfficientNet algorithm adopts a hierarchical design, and balances the depth, width and resolution of the network through the compound scaling method. It combines standard convolution and Mobile Inverted Bottleneck Conv (MBConv), reduces the amount of calculation by introducing point-by-point convolution and depth convolution, improves the network's expression ability, and forms a Fused-MBConv layer that adapts to feature extraction at different levels. The SiLU (Sigmoid-Weighted Linear Unit) activation function is used to enhance nonlinear expression ability. The asymmetric residual block design is introduced to improve the efficiency of feature transfer. Through dynamic progressive learning rate adjustment, the oscillation in the initial training stage is avoided, and the convergence speed of the model is improved. In the process of data enhancement, the Mixup and CutMix technologies are combined to improve the generalization ability and robustness of the model. By adjusting the memory usage, the use of video memory is reduced, and large-scale training and reasoning are optimized.

[0066] Dynamic progressive learning rate adjustment is a key strategy in the improved version of EfficientNet. It improves the convergence speed and stability of the model by gradually adjusting the learning rate during training. In the initial stage of training, a smaller learning rate is used for learning to avoid instability caused by large gradient updates. This process is called learning rate warm-up. The learning rate gradually increases from a very small value to the preset initial learning rate, which can usually be completed within the first few epochs.

[0067] Learning rate warm-up formula:

[0068]

[0069] Among them, lr t is the learning rate at time step t, lr0 is the preset initial learning rate, t is the current number of time steps, and t w is the number of time steps in the warm-up phase.

[0070] Learning rate formula with cosine annealing:

[0071]

[0072] Among them, lr min is the minimum learning rate, lr max is the initial learning rate, t is the current time step, and T is the total time step.

[0073] At different stages of training, the learning rate adjustment strategy may be different to adapt to the training needs of different stages. For example, a faster learning rate decay strategy can be used in the first half of training, while a slower learning rate decay strategy can be used in the second half of training to carefully optimize the model parameters.

[0074] The cross-entropy loss function is mainly used in the training process to measure the difference between the predicted results and the true labels. This is a loss function widely used in classification tasks, especially for multi-class classification problems.

[0075] The formula for the cross entropy loss function is:

[0076]

[0077] Where: N is the number of samples. C is the number of categories. i,c is the true label of sample i. If the sample belongs to category c, then y i,c =1; otherwise y i,c =0. i,c is the probability that sample i is predicted to be of category c.

[0078] The goal of the cross entropy loss function is to minimize the difference between the true label and the predicted probability, thereby improving the accuracy of the classification model.

[0079] Here, the trained model is manually imported into the inference platform, which specifically includes the following steps S31 to S33;

[0080] S31: Manually collect typical defect category image data,

[0081] S32: Based on step S31, the collected image data is uploaded to the specified training set path, and the training script is started to perform algorithm model training on a GPU card with 1 NVIDIA A10;

[0082] S33: Based on step S32, manually import the trained model into a concurrent inference platform with three NVIDIA A10 GPU cards;

[0083] The data of step S2 is parsed and processed to obtain the data of the algorithm structure, which specifically includes the following steps S41 to S43:

[0084] S41: Based on step S23, data preprocessing is performed on the request service monitored in real time to parse out the image path;

[0085] S42: Based on step S41, when the image path file cannot be found, returning content with an identifier "image not found" and a classification result of "None";

[0086] S43: Based on step S42, when the image path file is found, the image path in the request is processed into an inferable image array;

[0087] S5: Further perform defect detection on the preprocessed data and give the predicted defect type based on the trained model;

[0088] Based on the request data of step S43, the predicted defect type is given according to the model trained in step S33, and the specific contents include the following steps S51 to S53:

[0089] S51: Based on the classification algorithm model of step S33, perform trap detection prediction on the data pre-processed based on step S41;

[0090] S52: Based on step S51, the classification result predicted by the image, together with the unique ID information of the image and the correlation coefficient of the classification result, are packaged and returned;

[0091] S53: Based on step S51, if the correlation coefficient of the classification prediction does not reach the given minimum threshold (set as 0.1 here), the category is thrown out as "other" for manual re-judgment;

[0092] Based on the summary data obtained in step S52, further manual review is performed according to the defect type to verify and determine the root cause of the failure, which specifically includes the following steps S61 to S62:

[0093] S61: Based on the summary data of step S52, manually re-judgment is performed by manually selecting a specified product name, production line, equipment name, machine name, and a specified number of images of different defect types in a specified time period;

[0094] S62: Based on the image of the defect type in step S61, an experienced engineer determines the root cause of the failure;

[0095] Based on step S61, the results of the re-judgment are summarized, and the training data of the models that do not meet the standards are collected and re-trained. The specific contents include the following steps S71 to S73:

[0096] S71: Based on step S61, the results of the re-judgment are summarized to obtain the pass / miss detection score;

[0097] S72: For the model whose score in step S71 is higher than a given threshold (here the threshold is 99.7%), continue to monitor new request services;

[0098] S73: For the model whose score in step S71 is lower than a given threshold (here the threshold is 99.7%), the training set data is updated and the model is retrained, and steps S3 to S6 are repeated;

[0099] Based on step S62, measures are taken, and corresponding measures are taken according to the root cause of the defect.

[0100] Introduce advanced optical testing equipment and technology to improve the precision and accuracy of EL data.

[0101] Strengthen environmental management of production sites, ensure a clean and tidy production environment, and reduce pollution sources.

[0102] Optimize material selection and processing technology to improve the manufacturing quality and stability of battery cells.

[0103] Implement a strict quality management system, including testing, evaluation and feedback mechanisms, to identify and resolve problems in a timely manner.

[0104] Train and improve employee skills to ensure that operating specifications and equipment maintenance are effectively implemented.

[0105] Through the above steps S11 to S71, the pass rate and missed rate based on product, line, equipment, machine and other groups can be output. The smaller the pass rate and missed rate scores are, the higher the accuracy of the algorithm model is and the higher the use value is. Under the premise of the parameter settings in this example, the final accuracy is as high as 99.7%. At the same time, 33 images per second can be achieved on a GPU card of model NVIDIA A10. Here, 3 NVIDIA A10 GPU cards are used for concurrent consumption of messages and real-time reasoning, which is much faster than the original manual analysis and more comprehensive in the range of analyzed data, indicating that the results and effects of this method are very good. The types that have the greatest impact on battery cell defects (similarly, the worst impact type can be obtained) can greatly improve the work efficiency of photovoltaic-related engineers, provide useful guidance for production yield, and point out the direction for subsequent searches for potential process problems.

[0106] Based on the fact that manufacturers need to manually conduct quality inspections on the products they produce, with 100,000 solar cells inspected daily, each quality inspector can re-judge 1,000 cells in one hour, and manufacturers need nearly 12 people every day. Since the launch of the present invention, one NVIDIA A10 GPU card can infer 33 images per second, and three NVIDIA A10 GPU cards are fully capable of handling the work of these 12 people. Due to long-term fatigue work, manual re-judgment will result in more misjudgments, but the reasoning of the algorithm model will not result in such a situation. The present invention can greatly improve the work efficiency of photovoltaic-related engineers and provide useful guidance for the yield rate in production.

[0107] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0108] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0109] In another embodiment provided in the present application, a computer program product comprising instructions is also provided, which, when executed on a computer, enables the computer to execute any of the intelligent defect detection and quality control methods in photovoltaic production in the above embodiments.

[0110] It is understandable that the system, device and storage medium provided in the embodiments of the present invention correspond to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above methods.

[0111] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0112] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0113] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0114] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent defect detection and quality control method in photovoltaic production, characterized in that: The following steps are included: S1: Test and diagnose the produced cells in real time to obtain EL data information of the cells; S2: Get EL data and corresponding configuration files; S3: Manually collect defect type image data, set up algorithm model training, and manually import the trained model into the inference platform; S4: parsing and processing the EL data requested in real time in step S2 to obtain data of the algorithm structure; S5: further perform defect detection on the data preprocessed by S4, and give the predicted defect type based on the model trained by S3; S6: Further manual review is performed to verify and determine the root cause of the failure based on the defect type; S7: Summarize the results of the re-judgment to obtain the pass / miss scores. For the models whose scores are lower than the given threshold, update the training set data and retrain, and repeat steps S3 to S6. S8: Determine the root cause of the defects detected by the trained model and take corresponding measures.

2. The intelligent defect detection and quality control method in photovoltaic production according to claim 1, characterized in that: Step S1 specifically includes screening the corresponding product name, production line, equipment name, and machine name within a specified time, and performing EL testing and diagnosis operations on the battery cell to obtain the history information of the battery cell device.

3. The intelligent defect detection and quality control method in photovoltaic production according to claim 1, characterized in that: Step S2 specifically includes: S21: After the back-end ADC parsing data interface obtains the EL data within a given time, it parses the information of the corresponding table structure; S22: Through step S21, redundant data is filtered out, and data is stored according to the structural information in the database table; S23: Through step S22, the image path and the corresponding unique identification ID information are extracted, and the information is produced in the form of Kafka communication for subsequent inference service consumption.

4. The intelligent defect detection and quality control method in photovoltaic production according to claim 1, characterized in that: In step S3, the improved version of Effici entNet algorithm is used to train the model; The improved EfficientNet algorithm adopts a hierarchical design, using the Compound Scaling Method to balance the depth, width, and resolution of the network; it combines standard convolution with Mobile Inverted Bottleneck Conv (MBConv), reduces the amount of calculation by introducing point-by-point convolution and depth-wise convolution, improves the network's expressiveness, and forms a Fused-MBConv layer that adapts to feature extraction at different levels; The SiLU (Sigmoid-Weighted Linear Unit) activation function is used to enhance nonlinear expression capabilities. An asymmetric residual block design is introduced to improve the efficiency of feature transfer. Dynamic progressive learning rate adjustment is used to avoid oscillations in the initial training phase and improve the convergence speed of the model. Mixup and CutMix techniques are combined in the data enhancement process to improve the generalization and robustness of the model. Memory usage is reduced by adjusting memory usage to optimize large-scale training and inference.

5. The intelligent defect detection and quality control method in photovoltaic production according to claim 4, characterized in that: Dynamic progressive learning rate adjustment is a key strategy in the improved version of EfficientNet. It improves the convergence speed and stability of the model by gradually adjusting the learning rate during the training process. In the initial stage of training, a smaller learning rate is used to avoid instability caused by large gradient updates. This process is called learning rate warm-up. Learning rate warm-up formula: Among them, lr t is the learning rate at time step t, lr0 is the preset initial learning rate, t is the current number of time steps, and t w is the number of time steps in the warm-up phase; Learning rate formula with cosine annealing: Among them, l rmin is the minimum learning rate, lr max is the initial learning rate, t is the current time step, and T is the total time step.

6. The intelligent defect detection and quality control method in photovoltaic production according to claim 5, characterized in that: The cross entropy loss function is used during training to measure the difference between the predicted results and the true labels; The formula for the cross entropy loss function is: Where: N is the number of samples; C is the number of categories; y i,c is the true label of sample i. If the sample belongs to category c, then y i,c =1; otherwise y i,c =0; p i,c is the probability that sample i is predicted to be category c; The goal of the cross entropy loss function is to minimize the difference between the true label and the predicted probability, thereby improving the accuracy of the classification model.

7. The intelligent defect detection and quality control method in photovoltaic production according to claim 2, characterized in that: S4 specifically includes: S41: Based on step S23, data preprocessing is performed on the request service monitored in real time to parse out the image path; S42: Based on step S41, when the image path file cannot be found, the content with the mark "image not found" and the classification result as "None" is returned; S43: Based on step S42, when the image path file is found, the image path in the request is processed into an inferable image array.

8. The intelligent defect detection and quality control method in photovoltaic production according to claim 7, characterized in that: Step S5 specifically includes, based on the request data of step S43, providing a predicted defect type according to the model trained in step S33, and specifically includes the following steps S51 to S53: S51: Based on the classification algorithm model of step S33, perform trap detection prediction on the data pre-processed based on step S41; S52: Based on step S51, the classification result predicted by the image, together with the unique ID information of the image and the correlation coefficient of the classification result, are packaged and returned; S53: Based on step S51, if the correlation coefficient of the classification prediction does not reach the given minimum threshold, the category "other" is discarded for manual re-judgment.

9. The intelligent defect detection and quality control method in photovoltaic production according to claim 8, characterized in that: Step S6 specifically includes the following: S61: Based on the summary data of step S52, manually re-judgment is performed by manually selecting a specified product name, production line, equipment name, machine name, and a specified number of images of different defect types in a specified time period; S62: Based on the image of the defect type in step S61, the root cause of the failure is determined by experienced engineers.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 6.

Citation Information

Patent Citations

  • Image classification method based on improved OfficientNet-B0

    CN117333705A

  • Defect detection method and device of defect generation mode based on Bessel function

    CN118247212A

  • Photovoltaic module evaluation method and system based on AI visual inspection algorithm

    CN118261878A