Quality detection method, device and equipment for photovoltaic module

Abstract

The invention relates to the field of photovoltaic modules, in particular to a quality detection method, device and equipment for a photovoltaic module. The quality detection method for the photovoltaic module comprises the steps that a product picture of the photovoltaic module is obtained to serve as a first to-be-detected object, and the photovoltaic module is a product judged to be qualified after being detected by a production line; the first to-be-detected objects are initially detected through a multi-network detection model, a second to-be-detected object is screened out from the first to-be-detected objects, and the second to-be-detected object is a product possibly having abnormity in the first to-be-detected objects; extracting a third detection object from the second to-be-detected object, and sending the third detection object to a worker for reinspection; and processing the abnormal product according to a recheck result of a worker and a processing instruction issued according to the recheck result. According to the method, the defective components can be locked and intercepted in time before the products are put in storage, so that the sampling coverage rate is increased, and the sampling efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of photovoltaic modules, and in particular to a method, apparatus and equipment for quality testing of photovoltaic modules. Background Technology

[0002] During the quality judgment process of photovoltaic modules in the production line, due to abnormal reasons such as improper threshold settings of the testing equipment, unstable model algorithms, changes in testing standards, new defects, insufficient defect samples, unbalanced datasets, and inaccurate data labeling, testing failures and missed product inspections are prone to occur, resulting in defective finished products flowing into the market and causing customer complaints.

[0003] To intercept missed products, open box audit (OBA) sampling inspection is required. However, traditional manual OBA sampling inspection is time-consuming. Due to the high product yield, it is difficult to spot defective products, resulting in low efficiency, wasted packaging materials, and the risk of secondary damage to the products. Summary of the Invention

[0004] This invention provides a method, apparatus, and equipment for quality inspection of photovoltaic modules, which solves the problem of low efficiency in manual OBA sampling inspection in the prior art.

[0005] In a first aspect, embodiments of the present invention provide a method for quality inspection of photovoltaic modules, the method comprising: The photovoltaic module product image is obtained as the first object to be tested. The photovoltaic module is a product that has been judged to be qualified after being tested on the production line. The first object to be detected is initially detected by a multi-network detection model, and a second object to be detected is selected from the first object to be detected. The second object to be detected is a product that may be abnormal among the first object to be detected. A third test subject is randomly selected from the second test subjects and sent to staff for retesting; The abnormal products are processed according to the re-inspection results of the staff and the processing instructions issued based on the re-inspection results.

[0006] Optionally, the multi-network detection model includes a first sub-network model, a second sub-network model, a third sub-network model, and a fourth sub-network model; The initial detection of the first object to be detected using a multi-network detection model includes: The first sub-network model is used to perform back-end EL detection on each product image in the first object to be detected. The back-end EL detection is used to detect internal defects of the photovoltaic module. The second sub-network model is used to perform final FI inspection on the product images of the first detection object. The final FI inspection is used to detect external defects of the photovoltaic module. The nameplate detection is performed on the product images of the first detection object using the third sub-network model. The nameplate detection is used to detect nameplate defects of the photovoltaic module. The fourth sub-network model is used to perform graded detection on the product images in the first detection object. The graded detection is used to detect the appearance defects and performance defects of the photovoltaic module.

[0007] Optionally, the method of performing a preliminary detection on the first object to be detected using a multi-network detection model and selecting a second object to be detected from the first object to be detected further includes: By using each sub-network model in the multi-network detection model, the probability that each of the first objects to be detected may have a corresponding type of anomaly is determined; The product with a probability of any type of abnormality greater than zero among all the first objects to be tested is identified as the second object to be tested; The second detection objects are sorted according to the probability that each of them may be abnormal.

[0008] Optionally, the step of selecting a third test object from the second test object and sending it to staff for re-inspection includes: Based on the sorting order of the second detection objects, the third detection object is preferentially extracted from the second detection objects that have a higher probability of being abnormal.

[0009] Optionally, obtaining product images of photovoltaic modules as the first object to be detected includes: Simultaneously, the storage information of the first object to be tested is acquired, the storage information including the test line, test procedure, test time and component serial number of each first object to be tested; Based on the storage information, the first object to be detected is classified and stored in the MES file server.

[0010] Optionally, processing the abnormal product based on the staff's re-inspection results and the processing instructions issued based on the re-inspection results includes: When the processing instruction is an interception instruction, the abnormal product is retrieved through the MES file server; The abnormal products are transported to the rework workshop by an automated conveyor system controlled by the control commands, where staff will open the boxes and inspect them offline. When the re-inspection result of the third inspection object by the staff is abnormal, the interception command will be issued, and the MES file server is used to retrieve the abnormal product through the stored information.

[0011] Optionally, the offline unpacking and inspection by staff includes: Based on preset sampling criteria, staff will handle the defective products according to the type, quantity, and severity of the abnormalities. The handling of defective products includes partial rework, batch rework, order rework, downgrading, and transfer to inventory.

[0012] Optionally, before performing the initial detection of the first object to be detected using a multi-network detection model, the method further includes: The training of the first sub-network model, the second sub-network model, the third sub-network model, and the fourth sub-network model is completed in advance; The model training methods include: Each sub-model has its corresponding image dataset acquired. The image dataset includes the EL image dataset, the FI image dataset, the nameplate image dataset, and the graded image dataset. Construct the first sub-network model, the second sub-network model, the third sub-network model, and the fourth sub-network model respectively; The first sub-network model is trained using the EL image dataset, the second sub-network model is trained using the FI image dataset, the third sub-network model is trained using the nameplate dataset, and the fourth sub-network model is trained using the graded dataset.

[0013] Optionally, the construction of the first sub-network model, the second sub-network model, the third sub-network model, and the fourth sub-network model respectively includes: The first sub-network model is constructed based on the Swin Transformer network; The second, third, and fourth sub-network models are constructed based on CNN convolutional neural networks.

[0014] Secondly, embodiments of the present invention provide a quality testing device for photovoltaic modules, the device comprising: The acquisition module acquires product images of photovoltaic modules as the first object to be inspected. The photovoltaic modules are products that have been determined to be qualified after being inspected on the production line. The initial inspection module performs an initial inspection on the first object to be inspected using a multi-network detection model, and selects a second object to be inspected from the first object to be inspected. The second object to be inspected is a product that may be abnormal among the first objects to be inspected. The extraction module extracts a third test object from the second test object and sends it to the staff for re-inspection; The processing module processes the abnormal products based on the re-inspection results from the staff and the processing instructions issued based on the re-inspection results.

[0015] Optionally, the initial inspection module further includes: a detection module, a determination module, and a sorting module; The detection module determines the probability that each of the first objects to be detected may have a corresponding type of anomaly through each sub-network model in the multi-network detection model; The determination module identifies products among all the first objects to be detected that have a probability of any type of abnormality greater than zero as the second objects to be detected. The sorting module sorts the second detection objects according to the probability that each of the second detection objects may be abnormal.

[0016] Optionally, the multi-network detection model in the detection module includes a first sub-network model, a second sub-network model, a third sub-network model, and a fourth sub-network model; The first sub-network model is used to perform back-end EL detection on each product image in the first object to be detected, and the back-end EL detection is used to detect internal defects of the photovoltaic module; The second sub-network model is used to perform final FI inspection on each product image in the first detection object, and the final FI inspection is used to detect external defects of the photovoltaic module; The third sub-network model is used to perform nameplate detection on the product images in the first detection object, and the nameplate detection is used to detect nameplate defects of the photovoltaic module. The fourth sub-network model is used to perform graded detection on the product images in the first detection object, and the graded detection is used to detect the appearance defects and performance defects of the photovoltaic module.

[0017] Optionally, the acquisition module further includes: a storage information acquisition module and a storage module; The storage information acquisition module simultaneously acquires the storage information of the first object to be tested, the storage information including the test line, test procedure, test time and component serial number of each of the first objects to be tested; The storage module categorizes and stores the first object to be detected in the MES file server according to the storage information.

[0018] Optionally, the processing module further includes: a retrieval module and a transport module; The retrieval module, when the processing instruction is an interception instruction, retrieves the abnormal products through the MES file server; The handling module controls the automated transport equipment to transport the detected abnormal products to the rework workshop, where staff will open the boxes and inspect them offline. When the re-inspection result of the third inspection object by the staff is abnormal, the interception command will be issued, and the MES file server is used to retrieve the abnormal product through the stored information.

[0019] Thirdly, embodiments of the present invention provide a quality testing device for photovoltaic modules, characterized in that it includes a quality testing apparatus for photovoltaic modules as described in any one of the second aspects, for performing a quality testing method for photovoltaic modules as described in any one of the first aspects.

[0020] In this embodiment of the invention, a multi-network detection model is used to perform non-physical sampling inspections of photovoltaic modules based on product images, replacing the original method of manual verification of physical samples. This allows for the timely identification and interception of defective modules before products enter the warehouse, improving sampling coverage and efficiency, and saving labor resources and packaging material waste caused by blindly opening boxes for sampling. Furthermore, when abnormal products are detected, timely manual processing also addresses the issues encountered in traditional offline manual sampling inspections. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 The diagram shown is a flowchart of a quality inspection method for photovoltaic modules provided in an embodiment of this application; Figure 2 The diagram shown is a structural schematic of a photovoltaic module quality inspection device provided in an embodiment of this application; Figure 3 The diagram shown is a structural schematic of another photovoltaic module quality inspection device provided in an embodiment of this application; Figure 4 The diagram shown is a structural schematic of a quality testing device for photovoltaic modules provided in an embodiment of this application. Detailed Implementation

[0023] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0024] It should be clear that the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0025] As Figure 1 shown, it is a flowchart of a method for detecting the quality of a photovoltaic module provided by an embodiment of the present invention. Refer to Figure 1 , the specific steps of this method include: S101, Obtain the product picture of the photovoltaic module as the first object to be detected. The photovoltaic module is a product determined to be qualified after being detected on the production line.

[0026] Specifically, when the photovoltaic modules on several production lines are being produced, they need to be detected to determine whether they are qualified products. Since the production line aims to ensure production volume, the detection standards on the production line are usually relatively low, which may allow some abnormal products to pass the detection. Therefore, based on the product pictures obtained from the production line, the products determined to be qualified after being detected on the production line are used as the first objects to be detected for further detection.

[0027] Among them, all the original pictures of each production line during the subsequent electroluminescence testing (Electroluminescence Testing, EL testing), final inspection (Final Inspection, FI final inspection), nameplate inspection, and grading inspection are obtained as the first objects to be detected for detection.

[0028] In the embodiment of the present invention, when obtaining the product picture of the photovoltaic module as the first object to be detected, the storage information of the first object to be detected needs to be obtained simultaneously. The storage information of the first object to be detected includes the test line, test process, test time, and component serial number of each first object to be detected. According to the storage information, the first objects to be detected are classified and stored in the Manufacturing Execution System (MES) server for subsequent quick retrieval.

[0029] S102, Initially inspect the first object to be detected through a multi-network detection model, and screen out the second object to be detected from the first object to be detected. The second object to be detected is a product that may be abnormal among the first objects to be detected.

[0030] Specifically, the multi-network detection model includes four sub-network models: a first sub-network model, a second sub-network model, a third sub-network model, and a fourth sub-network model. In this embodiment of the invention, the multi-network detection model can be a deep neural network YOLO algorithm model.

[0031] When performing initial inspection of the first object to be inspected using a multi-network detection model, the first sub-network model is used to perform post-processing EL inspection on each product image in the first object to be inspected; the second sub-network model is used to perform final FI inspection on each product image in the first object to be inspected; the third sub-network model is used to perform nameplate inspection on each product image in the first object to be inspected; and the fourth sub-network model is used to perform graded inspection on each product image in the first object to be inspected.

[0032] Among them, the post-processing EL inspection is used to detect internal defects of photovoltaic modules, such as poor soldering, microcracks, and broken grids; the final FI inspection is used to detect external defects of photovoltaic modules, such as dirt, white spots on the front, chipped edges, missing corners, fragments, misalignment of front / back solder strips, busbar position misalignment, string spacing, misalignment, and foreign objects; the nameplate inspection is used to detect nameplate defects of photovoltaic modules, such as missing, damaged, bubbled, and missing double barcodes; and the graded inspection is used to detect appearance and performance defects of photovoltaic modules, such as scratches on the glass / cleanliness of the glass surface and poor junction box.

[0033] Each first object to be detected is input into a multi-network model, and the probability of each first object to be detected possibly having a corresponding type of anomaly is determined by each sub-network model in the multi-network model. For example, the probability of each first object to be detected having an internal defect is determined by the first sub-network model, and the probability of each first object to be detected having an external defect is determined by the second sub-network model.

[0034] Products with a probability greater than zero of any type of anomaly among all the first-stage test objects are designated as second-stage test objects. These second-stage test objects are then sorted according to the probability of anomalies they may exhibit. Generally, they are sorted from highest to lowest probability of anomaly.

[0035] Specifically, when ranking the second target object, the anomaly probabilities output by each sub-detection model are comprehensively ranked. That is, the anomaly type of the product is not distinguished, but only ranked according to the probability that the product may have an anomaly. If a product has the probability of having multiple types of anomalies, the product with the highest probability of having an anomaly is ranked.

[0036] In one specific embodiment, 10,000 first objects to be detected are input into a multi-network detection model. The first detection model identifies 100 products that may have various anomalies, and these are designated as second objects to be detected. Among the second objects, 50 products have a 40% probability of being abnormal, 20 products have a 60% probability of being abnormal, and 30 products have a 30% probability of being abnormal. The 100 second objects are then sorted according to their probability of being abnormal, from highest to lowest (60%, 40%, 30%).

[0037] S103, Select a third test object from the second test object and send it to the staff for re-inspection.

[0038] Specifically, a third sample is selected and sent to staff, who then manually review the image to determine whether to intercept the sample.

[0039] Specifically, based on the sorting order of the second objects to be detected, the third object to be detected is selected first from the second objects that have a higher probability of being abnormal.

[0040] Based on the embodiments described above, when selecting the third detection object, it is first selected from 20 second detection objects with an abnormality probability of 60%, then from 50 second detection objects with an abnormality probability of 40%, and finally from 30 second detection objects with an abnormality probability of 30%.

[0041] S104. Based on the re-inspection results of the staff and the processing instructions issued based on the re-inspection results, the abnormal products are processed.

[0042] Specifically, if the staff re-examines the third sampled object and finds that it has no abnormalities, then the third sampled object will not be processed.

[0043] If staff determine that the re-inspection of the third sampled item indicates an anomaly, an interception command will be issued as a processing instruction. Upon receiving the interception command, the abnormal product will be retrieved through the MES file server.

[0044] The MES document service system will retrieve abnormal products identified by staff based on the storage information when storing the first object to be inspected.

[0045] Once an abnormal product is detected, the automated guided vehicle (AGV) is controlled via control commands to transport the abnormal product to the rework workshop, where staff will open the box and inspect it offline.

[0046] When staff open the boxes, they will take corresponding measures to handle defective products based on the type, quantity and severity of the abnormality of the defective products, according to the preset sampling standards. These measures include partial rework, batch rework, order rework, downgrading, and transfer to inventory.

[0047] Generally, the preset sampling standard is GB2828 standard.

[0048] Optionally, in some embodiments, when multiple products are identified as having the same type of anomaly, an automatic warning is issued, and closed-loop analysis and improvement are performed.

[0049] In this embodiment of the invention, a multi-network detection model is used to perform random inspections of photovoltaic modules based on product images without physical inspection, replacing the original method of manual verification of physical samples. This allows for the timely identification and interception of defective modules before products enter the warehouse, improving inspection coverage and efficiency, and saving labor resources and packaging material waste caused by blindly opening boxes for inspection. Furthermore, when abnormal products are detected, timely manual processing also addresses the issues encountered in traditional offline manual inspections.

[0050] Optionally, in some embodiments, it is necessary to pre-train the multi-network detection model. Specifically, it is necessary to train the first sub-network model, the second sub-network model, the third sub-network model, and the fourth sub-network model of the multi-network detection model separately.

[0051] Specifically, image datasets corresponding to each sub-model are obtained, namely, the EL image dataset corresponding to the first sub-network model, the FI image dataset corresponding to the second sub-network model, the nameplate image dataset corresponding to the third sub-network model, and the graded image dataset corresponding to the fourth sub-network model.

[0052] First, second, third, and fourth sub-network models were constructed respectively. The first sub-network model was trained using the EL image dataset, the second sub-network model using the FI image dataset, the third sub-network model using the nameplate dataset, and the fourth sub-network model using the graded dataset.

[0053] In the real-time example of this invention, the multi-network detection model is the YOLO model, the deep learning framework is PyTorch, and the virtual environment management tool is Anaconda.

[0054] Specifically, the first sub-network model needs to be built based on the Swing Transformer network, and the second, third, and fourth sub-network models need to be built based on the Convolutional Neural Network (CNN).

[0055] When building and training the first sub-network model, it is necessary to construct the first sub-network model through the Swing Transformer network. The Swing Transformer is a general computer vision backbone network based on the Transformer architecture. By introducing a sliding window mechanism and hierarchical downsampling, it solves the limitations of the Vision Transformer (ViT) in multi-task scenarios.

[0056] In traditional feature extraction based on convolutional principles, CNNs (Convolutional Neural Networks) focus more on local features while neglecting the connections between them. The Transformer architecture effectively addresses this issue. The Swin Transformer incorporates more prior knowledge from computer vision, implementing a hierarchical structure similar to CNNs. Through a flexible self-attention mechanism using paired Self-Window and Shifted-Window windows, the network achieves a globally comprehensive receptive field and global modeling capabilities even at shallower layers, while also improving computational performance.

[0057] Therefore, compared to the traditional CNN backbone network, the first sub-network model constructed by the Swing Transformer network for performing subsequent EL detection reduces computational complexity while maintaining global modeling capabilities. At the same time, it can grasp multi-scale features such as visual features, semantic features of the same target at different scales, and contextual features of the detected target and surrounding complex interference, to achieve hierarchical feature extraction and effectively resist interference.

[0058] In one specific embodiment, when training the first sub-network model, an EL image dataset is first collected, specifically including images of 13 types of defective products such as poor solder joints, hidden cracks, and broken grids in the downstream EL detection of battery cells, and the EL image dataset is labeled using the image annotation tool Labelimg.

[0059] The first sub-network model is built based on the transformer. The original backbone network CNN in the YOLO model is replaced with a Swin transformer network, and the neck network is improved using the attention mechanism module CBAM. The introduction of the attention mechanism enables more accurate defect identification.

[0060] In this method, the original CNN network in the YOLO network model is replaced with a Swin Transformer backbone network, which is connected before the Spatial Pyramid Pooling Fast (SPPF) module. The attention mechanism module CBAM is embedded after the first, second, and third C3k2 modules of the neck network to evaluate the model quality.

[0061] Furthermore, replacing the loss function with the SIoU loss function improves the model's localization accuracy, thereby achieving higher detection accuracy and robustness.

[0062] The network model training parameters are set as follows: 500 training epochs, batch size of 32, initial learning rate of 0.01, input image size of 640×640, SGD optimizer, and 4 threads.

[0063] The best.pt file was obtained by training with the EL image dataset. After validating and testing the first sub-network model according to the ratio of training set, validation set and test set of 6:2:2, the training of the first sub-network model was completed and it was deployed in the multi-network detection model.

[0064] When constructing and training the second sub-network model, wavelet convolutions (WTConv) can decompose the input image into different frequency components through wavelet transform, perform small-size convolutions on each frequency layer, and finally recombine the results through inverse wavelet transform, thereby enabling multi-scale analysis of the image.

[0065] Therefore, WTConv can expand the receptive field of convolution and effectively capture low-frequency information in images. It has excellent performance in multi-scale problems and small object problems for performing FI detection.

[0066] In one specific embodiment, when training the second sub-network model, the FI image dataset is first collected, which includes 35 types of defective product images such as photovoltaic module dirt, white spots on the front, chipped edges, missing corners, fragments, front / back solder strip offset, busbar position offset, string spacing, misalignment, and foreign objects. The dataset is labeled using the labeling tool Labelimg.

[0067] A second sub-network model is constructed based on a CNN network. First, the C3k2 module in the neck network is improved. The C3k2 module is improved using WTConv convolutions to construct a C3k2_WT module, which then replaces the original C3k2, thereby expanding the model's receptive field through WTConv. Next, the head layer of the second sub-network model is improved by constructing a DSConvWithWT depthwise separable convolution using WTConv convolutions. This DSConvWithWT convolution replaces the depthwise separable convolutions in the head layer, further expanding the receptive field of the head layer. This improves the AI ​​detection rate and robustness of the final FI (Firmware Detection) test, making the model more stable.

[0068] The network model training parameters are set as follows: 200 training epochs, batch size of 20, initial learning rate of 0.01, input image size of 640 × 640, SGD optimizer, and 4 threads.

[0069] The best.pt file was obtained by training on the FI image dataset. After validating and testing the model according to the ratio of training set, validation set and test set of 6:2:2, the training of the second sub-network model was completed and it was deployed in the multi-network detection model.

[0070] When constructing and training the third sub-network model, a CNN network is applied to construct the third sub-network model.

[0071] In one specific embodiment, a dataset of nameplate images is first collected, specifically including images of five types of defective products such as photovoltaic modules without missing labels, defects, bubbles, and double barcodes. The dataset is then labeled using the labeling tool Labelimg.

[0072] The network model training parameters are set as follows: 200 training epochs, batch size of 20, initial learning rate of 0.01, input image size of 640 × 640, SGD optimizer, and 4 threads.

[0073] The best.pt file was obtained by training on the nameplate image dataset. After validating and testing the model according to the ratio of training set, validation set and test set of 6:2:2, the training of the third sub-network model was completed and deployed in the multi-network detection model.

[0074] When building and training the fourth sub-network model, if the predicted box and the ground truth box have the same aspect ratio but different width and height values, the default bounding box regression loss function of the YOLO model may have the same value, which will reduce the convergence speed and accuracy of bounding box regression.

[0075] Therefore, we introduce a new IoU metric, MPDIoU (Intersection over Union with Minimum Points Distance), which is a loss function for efficient and accurate bounding box regression. MPDIoU directly minimizes the distance between the top-left and bottom-right corners of the predicted bounding box and the ground truth bounding box.

[0076] In one specific embodiment, a graded image dataset is first collected, specifically including images of 10 types of defective products such as glass scratches / glass surface cleanliness of photovoltaic modules and defective junction boxes (missing parts, missing junction box covers), and the dataset is labeled using the standard tool Labelimg.

[0077] When constructing the fourth network model, the MPDIoU loss function is introduced. Data augmentation increases the model's adaptability to objects of different sizes by scaling the images (such as random cropping, scaling, etc.). By rotating and flipping the images, the model's ability to recognize objects in different directions is enhanced. Optimizing the loss function can improve the AI ​​detection rate and robustness of graded detection, making the model more stable.

[0078] The network model training parameters are set as follows: 500 training epochs, batch size of 20, initial learning rate of 0.01, input image size of 640 × 640, SGD optimizer, and 4 threads.

[0079] The best.pt file was obtained by training on the graded image dataset. After validating and testing the model according to the ratio of training set, validation set and test set of 6:2:2, the training of the fourth sub-network model was completed and deployed in the multi-network detection model.

[0080] Corresponding to the above-described quality inspection method for photovoltaic modules, this application also provides a quality inspection device for photovoltaic modules. See [link to relevant documentation]. Figure 2 This is a schematic diagram of the structure of a quality inspection device for photovoltaic modules provided in an embodiment of this application. The quality inspection device for photovoltaic modules may include: an acquisition module 210, a preliminary inspection module 220, an extraction module 230, and a processing module 240.

[0081] The acquisition module 210 acquires product images of photovoltaic modules as the first object to be inspected. The photovoltaic modules are products that have been determined to be qualified after being inspected on the production line.

[0082] The initial inspection module 220 performs an initial inspection on the first object to be inspected through a multi-network detection model, and selects a second object to be inspected from the first object to be inspected. The second object to be inspected is a product that may be abnormal among the first objects to be inspected.

[0083] The extraction module 230 extracts a third test object from the second test object and sends it to the staff for re-inspection.

[0084] The processing module 240 processes abnormal products based on the re-inspection results of the staff and the processing instructions issued based on the re-inspection results.

[0085] like Figure 3 The diagram shown is a structural schematic of another photovoltaic module quality inspection device provided in an embodiment of this application. See also... Figure 3 The initial inspection module 220 also includes a detection module 221, a determination module 222, and a sorting module 223. The acquisition module 210 also includes an information acquisition module 211 and a storage module 212. The processing module 240 also includes a retrieval module 241 and a transport module 242.

[0086] The detection module 221 determines the probability that each first object to be detected may have a corresponding type of anomaly through the sub-network models in the multi-network detection model.

[0087] The determination module 222 identifies products among all the first objects to be tested that have a probability of any type of abnormality greater than zero as the second objects to be tested.

[0088] The sorting module 223 sorts the second detection objects according to the probability that each second detection object may be abnormal.

[0089] The multi-network detection model in the detection module includes a first sub-network model, a second sub-network model, a third sub-network model, and a fourth sub-network model. The first sub-network model is used to perform back-end EL inspection on the product images of the first object to be inspected. Back-end EL inspection is used to detect internal defects in photovoltaic modules.

[0090] The second sub-network model is used to perform final FI inspection on the product images of each product in the first detection object. The final FI inspection is used to detect external defects of photovoltaic modules. The third sub-network model is used to perform nameplate detection on the product images in the first detection object. The nameplate detection is used to detect nameplate defects in photovoltaic modules.

[0091] The fourth sub-network model is used to perform graded detection on the product images in the first detection object. Graded detection is used to detect appearance defects and performance defects of photovoltaic modules.

[0092] The storage information acquisition module 211 simultaneously acquires the storage information of the first object to be tested, including the test line, test procedure, test time and component serial number of each first object to be tested.

[0093] Storage module 212, based on the storage information, classifies and stores the first object to be detected in the MES file server.

[0094] The retrieval module 241, when the processing instruction is an interception instruction, retrieves abnormal products through the MES file server; The handling module 242 controls the automatic transport equipment to transport the detected abnormal products to the rework workshop through control commands, where staff will open the boxes and inspect them offline.

[0095] When the re-inspection result of the third inspection object is abnormal, an interception command will be issued, and the MES file server will be used to retrieve the abnormal product by storing information.

[0096] Figure 4 This is a structural diagram of the photovoltaic module quality testing equipment described in this specification. The photovoltaic module quality testing equipment can specifically include, for example: Figure 1 The photovoltaic module quality inspection device shown is an image forming device used to perform the photovoltaic module quality inspection method provided in this embodiment. Figure 4 As shown, the photovoltaic module quality inspection device may include at least one processor; and at least one memory communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor can execute the photovoltaic module quality inspection method provided in this embodiment by calling the program instructions.

[0097] The aforementioned photovoltaic module quality testing equipment can be a device capable of intelligent dialogue with the user, such as a cloud server. This specification does not limit the specific form of the photovoltaic module quality testing equipment in this embodiment. It is understood that the photovoltaic module quality testing equipment here refers to the machine mentioned in the method embodiment.

[0098] Figure 4 A block diagram of a quality inspection device suitable for implementing the embodiments of this specification for an exemplary photovoltaic module is shown. Figure 4 The photovoltaic module quality testing equipment shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments described in this specification.

[0099] like Figure 4 As shown, the photovoltaic module quality testing equipment is presented in the form of a general-purpose computing device. The components of the photovoltaic module quality testing equipment may include, but are not limited to: one or more processors 410, communication interface 420, memory 430, and communication bus 440 connecting different system components (including memory 430, communication interface 420 and processor 410).

[0100] Communication bus 440 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, Industry Standard Architecture (ISA) buses, Micro Channel Architecture (MAC) buses, Enhanced ISA buses, Video Electronics Standards Association (VESA) local buses, and Peripheral Component Interconnect (PCI) buses.

[0101] Quality inspection equipment for photovoltaic modules typically includes a variety of computer-readable media. These media can be any available media that can be accessed by the quality inspection equipment for photovoltaic modules, including volatile and non-volatile media, and movable and non-movable media.

[0102] Memory 430 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The photovoltaic module quality inspection equipment may further include other removable / non-removable, volatile / non-volatile computer system storage media. Memory 430 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments described herein.

[0103] A program / utility having a set (at least one) of program modules may be stored in memory 430. Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. The program modules typically perform the functions and / or methods described in the embodiments of this specification.

[0104] The processor 410 executes various functional applications and data processing by running programs stored in the memory 430, such as implementing the quality inspection method for photovoltaic modules provided in the embodiments shown in this specification.

[0105] This specification provides a non-transitory computer-readable storage medium that stores computer instructions that cause the computer to execute the photovoltaic module quality inspection method provided in the embodiments shown in this specification.

[0106] The aforementioned non-transitory computer-readable storage medium may be any combination of one or more computer-readable media. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0107] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0108] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0109] Computer program code for performing the operations described herein can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0110] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0111] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this specification, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0112] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this specification includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which the embodiments of this specification pertain.

[0113] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0114] It should be noted that the terminals involved in the embodiments of this specification may include, but are not limited to, personal computers (PCs), personal digital assistants (PDAs), wireless handheld devices, tablet computers, mobile phones, MP3 players, MP4 players, etc.

[0115] In the embodiments provided in this specification, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0116] Furthermore, the functional units in the various embodiments of this specification can be integrated into a single processor, or each unit can exist physically separately, or two or more units can be integrated into a single unit. The integrated units described above can be implemented in hardware or in a combination of hardware and software functional units.

[0117] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this specification.

[0118] The above description is merely a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of protection of this specification.

Claims

1. A method for quality inspection of photovoltaic modules, characterized in that, The method includes: The photovoltaic module product image is obtained as the first object to be tested. The photovoltaic module is a product that has been judged to be qualified after being tested on the production line. The first object to be detected is initially detected by a multi-network detection model, and a second object to be detected is selected from the first object to be detected. The second object to be detected is a product that may be abnormal among the first object to be detected. A third test subject is randomly selected from the second test subjects and sent to staff for retesting; The abnormal products are processed according to the re-inspection results of the staff and the processing instructions issued based on the re-inspection results.

2. The method according to claim 1, characterized in that, The multi-network detection model includes a first sub-network model, a second sub-network model, a third sub-network model, and a fourth sub-network model; The initial detection of the first object to be detected using a multi-network detection model includes: The first sub-network model is used to perform back-end EL detection on each product image in the first object to be detected. The back-end EL detection is used to detect internal defects of the photovoltaic module. The second sub-network model is used to perform final FI inspection on the product images of the first detection object. The final FI inspection is used to detect external defects of the photovoltaic module. The nameplate detection is performed on the product images of the first detection object using the third sub-network model. The nameplate detection is used to detect nameplate defects of the photovoltaic module. The fourth sub-network model is used to perform graded detection on the product images in the first detection object. The graded detection is used to detect the appearance defects and performance defects of the photovoltaic module.

3. The method according to claim 1, characterized in that, The method further includes: performing a preliminary detection on the first object to be detected using a multi-network detection model, and selecting a second object to be detected from the first object to be detected; By using each sub-network model in the multi-network detection model, the probability that each of the first objects to be detected may have a corresponding type of anomaly is determined; The product with a probability of any type of abnormality greater than zero among all the first objects to be tested is identified as the second object to be tested; The second detection objects are sorted according to the probability that each of them may be abnormal.

4. The method according to claim 3, characterized in that, The step of selecting a third test subject from the second test subject and sending it to staff for re-testing includes: Based on the sorting order of the second detection objects, the third detection object is preferentially extracted from the second detection objects that have a higher probability of being abnormal.

5. The method according to claim 1, characterized in that, The acquisition of product images of photovoltaic modules as the first object to be inspected includes: Simultaneously, the storage information of the first object to be tested is acquired, the storage information including the test line, test procedure, test time and component serial number of each first object to be tested; Based on the storage information, the first object to be detected is classified and stored in the MES file server.

6. The method according to claim 5, characterized in that, The process of handling the abnormal products based on the re-inspection results of the staff and the processing instructions issued based on the re-inspection results includes: When the processing instruction is an interception instruction, the abnormal product is retrieved through the MES file server; The abnormal products are transported to the rework workshop by an automated conveyor system controlled by the control commands, where staff will open the boxes and inspect them offline. When the re-inspection result of the third inspection object by the staff is abnormal, the interception command will be issued, and the MES file server is used to retrieve the abnormal product through the stored information.

7. The method according to claim 6, characterized in that, The offline unpacking and inspection by staff includes: Based on preset sampling criteria, staff will handle the defective products according to the type, quantity, and severity of the abnormalities. The handling of defective products includes partial rework, batch rework, order rework, downgrading, and transfer to inventory.

8. The method according to claim 2, characterized in that, Before performing the initial detection of the first object to be detected using a multi-network detection model, the method further includes: The training of the first sub-network model, the second sub-network model, the third sub-network model, and the fourth sub-network model is completed in advance; The model training methods include: Each sub-model has its corresponding image dataset acquired. The image dataset includes the EL image dataset, the FI image dataset, the nameplate image dataset, and the graded image dataset. Construct the first sub-network model, the second sub-network model, the third sub-network model, and the fourth sub-network model respectively; The first sub-network model is trained using the EL image dataset, the second sub-network model is trained using the FI image dataset, the third sub-network model is trained using the nameplate dataset, and the fourth sub-network model is trained using the graded dataset.

9. The method according to claim 8, characterized in that, The construction of the first sub-network model, the second sub-network model, the third sub-network model, and the fourth sub-network model includes: The first sub-network model is constructed based on the Swin Transformer network; The second, third, and fourth sub-network models are constructed based on CNN convolutional neural networks.

10. A quality inspection device for photovoltaic modules, characterized in that, The device includes: The acquisition module acquires product images of photovoltaic modules as the first object to be inspected. The photovoltaic modules are products that have been determined to be qualified after being inspected on the production line. The initial inspection module performs an initial inspection on the first object to be inspected using a multi-network detection model, and selects a second object to be inspected from the first object to be inspected. The second object to be inspected is a product that may be abnormal among the first objects to be inspected. The extraction module extracts a third test object from the second test object and sends it to the staff for re-inspection; The processing module processes the abnormal products based on the re-inspection results from the staff and the processing instructions issued based on the re-inspection results.

11. The apparatus according to claim 10, characterized in that, The initial inspection module also includes: a detection module, a determination module, and a sorting module; The detection module determines the probability that each of the first objects to be detected may have a corresponding type of anomaly through each sub-network model in the multi-network detection model; The determination module identifies products among all the first objects to be detected that have a probability of any type of abnormality greater than zero as the second objects to be detected. The sorting module sorts the second detection objects according to the probability that each of the second detection objects may be abnormal.

12. The apparatus according to claim 11, characterized in that, The multi-network detection model in the detection module includes a first sub-network model, a second sub-network model, a third sub-network model, and a fourth sub-network model; The first sub-network model is used to perform back-end EL detection on each product image in the first object to be detected, and the back-end EL detection is used to detect internal defects of the photovoltaic module; The second sub-network model is used to perform final FI inspection on each product image in the first detection object, and the final FI inspection is used to detect external defects of the photovoltaic module; The third sub-network model is used to perform nameplate detection on the product images in the first detection object, and the nameplate detection is used to detect nameplate defects of the photovoltaic module. The fourth sub-network model is used to perform graded detection on the product images in the first detection object, and the graded detection is used to detect the appearance defects and performance defects of the photovoltaic module.

13. The apparatus according to claim 10, characterized in that, The acquisition module further includes: an information acquisition module and a storage module; The storage information acquisition module simultaneously acquires the storage information of the first object to be tested, the storage information including the test line, test procedure, test time and component serial number of each of the first objects to be tested; The storage module categorizes and stores the first object to be detected in the MES file server according to the storage information.

14. The apparatus according to claim 10, characterized in that, The processing module further includes: a retrieval module and a transport module; The retrieval module, when the processing instruction is an interception instruction, retrieves the abnormal products through the MES file server; The handling module controls the automated transport equipment to transport the detected abnormal products to the rework workshop, where staff will open the boxes and inspect them offline. When the re-inspection result of the third inspection object by the staff is abnormal, the interception command will be issued, and the MES file server is used to retrieve the abnormal product through the stored information.

15. A quality inspection device for photovoltaic modules, characterized in that, The photovoltaic module quality testing device as described in any one of claims 10 to 14 is used to perform the photovoltaic module quality testing method as described in any one of claims 1 to 9.

Images