Solar cell defect detection method based on multi-mode sensing technology

Abstract

The invention discloses a solar cell defect detection method based on a multi-mode sensing technology. The system comprises the following steps: 1, multi-modal data acquisition: acquiring multi-modal data in real time by using four sensing technologies of near-infrared laser, area array visible light, PL and EL; 2, multi-modal data processing: carrying out denoising, edge detection and feature extraction on the acquired near-infrared, visible light, PL and EL images by adopting a mixed image processing algorithm to obtain key defect features; 3, multi-modal data fusion: combining a multi-modal data fusion method with a decision-making level strategy, synthesizing different technical data, improving the detection precision, and classifying, positioning and labeling the defects of cracks, broken gates, scratches, depressions, poor welding and hot spots by using a convolutional neural network to ensure accurate distinguishing of surface and internal defects; and 4, defect marking and production line adjustment: marking defects and adjusting the production line through a feedback and automatic control module, and feeding back a real-time visual result to a management layer.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection of solar cells, and in particular to a method for detecting defects of solar cells based on multimodal sensing technology. Background Art

[0002] With the popularization and development of solar power generation technology, the production quality of solar cells directly affects their photoelectric conversion efficiency and service life. However, during the production process of solar cells, various defects such as cracks, disconnections, scratches, dents, cold solder joints and hot spots often occur, which can significantly reduce the performance of the cell. Therefore, efficient and accurate defect detection methods are particularly important.

[0003] Existing defect detection methods usually rely on a single technical means, such as visible light technology for surface detection, thermal imaging technology, or internal defect detection technology through laser scanning. However, these technical methods often have the limitations of a single technology, such as the inability to effectively detect internal defects in battery cells or insufficient recognition accuracy of surface defects. Therefore, how to combine multiple sensing technologies to achieve comprehensive detection of surface and internal defects of battery cells has become a technical challenge that needs to be solved urgently. Summary of the invention

[0004] In order to solve the shortcomings of the prior art, the present invention provides a solar cell defect detection method based on multimodal sensing technology. By combining near-infrared laser, array visible light, PL photoluminescence and EL electroluminescence technology, multi-peak data of solar cells are comprehensively collected. Through multimodal data fusion algorithm and deep learning technology, the surface and internal defects of solar cells are accurately located and identified, greatly improving the accuracy and efficiency of defect detection.

[0005] The solar cell defect detection method of the present invention comprises the following steps:

[0006] Step 1: Use four sensing technologies, including near-infrared laser, area array visible light, PL-photoluminescence, and EL-electroluminescence, to collect multimodal data of solar cells in real time; these technologies are used to identify surface and internal defects of solar cells;

[0007] Step 2: Apply traditional image processing algorithms to hybrid multimodal data analysis to process the collected NIR images, visible light images, PL images, and EL images; including denoising, edge detection, and feature extraction to obtain key defect features of surface and internal defects from different sensing modes;

[0008] Step 3: Adopt a multi-modal data fusion method and combine a decision-level fusion strategy to comprehensively analyze data from different technologies, improving the detection accuracy; by integrating multiple data layers, the system gives full play to the advantages of each technology and minimizes errors caused by the limitations of a single technology; on this basis, apply a convolutional neural network (CNN) for deep learning analysis, classification, and localization to identify defect types such as cracks, broken grids, scratches, dents, poor soldering, and hotspots, and assign corresponding labels to different defect types to ensure accurate distinction between surface defects and internal defects;

[0009] Step 4: Mark the defects according to the defect detection results of multi-modal data fusion and adjust the production line through the feedback and automation control module.

[0010] Furthermore, the specific process of Step 1 is as follows:

[0011] By implementing four sensing technologies: near-infrared laser, area array visible light, PL - photoluminescence, and EL - electroluminescence, the system collects multi-modal data of solar cells in real time on the production line and analyzes it through subsequent intelligent algorithms; by precisely adjusting the sensor acquisition parameters and applying these technologies, the system can identify and locate surface and internal defects of solar cells, thus achieving efficient and accurate detection of solar cells;

[0012] The four types of sensing technologies are as follows:

[0013] Near-infrared laser technology:

[0014] The near-infrared laser technology irradiates the surface of the solar cell with a laser and measures the intensities of the reflected and transmitted lasers to detect internal defects of the solar cell; the near-infrared laser can penetrate the surface layer of the cell, thus detecting internal defects such as cracks, broken grids, scratches, dents, poor soldering, and hotspots;

[0015] A group of near-infrared laser sensors is installed on the production line of the cell. The laser beam penetrates the cell and is reflected, and the sensor measures the intensity of the reflected light; the presence or absence of defects in the cell is inferred based on the change in the reflected light;

[0016] The intensity of the reflected light is described by the following formula:

[0017] I ref =I0·e -α·d

[0018] In the formula, I ref is the intensity of the reflected light, I0 is the initial laser intensity, α is the absorption coefficient of the near-infrared laser in the cell material, d is the thickness of the cell, and e -α·d represents the attenuation ratio of the laser after passing through the cell thickness d, and is used to calculate the intensity of the reflected light I refKey factors; among them, the absorption coefficient α affects the attenuation degree of the laser and varies with different materials, and the thickness d of the cell directly affects the laser transmission ability;

[0019] Area array visible light technology:

[0020] The area array visible light imaging technology illuminates the surface of the solar cell and uses an imaging device to capture the reflected light image, thereby identifying surface defects; subsequently, applying image processing technology to analyze the light intensity change of each pixel can detect the smallest flaw on the surface of the solar cell;

[0021] Install a high-resolution area array visible light camera to perform real-time imaging on the surface of the cell; the resolution of this camera is 3000×2000 pixels, and each pixel corresponds to a tiny area on the surface of the cell; through the collected image, use traditional image processing algorithms to analyze the image, thereby identifying cracks, broken grids, scratches, depressions, false soldering, and hot spot defects on the surface;

[0022] The image processing of area array visible light imaging identifies edges through image gradient calculation; the image f(x, y) represents the collected image of the cell surface, and edge detection is performed through the gradient of the image;

[0023] Gradient calculation formula:

[0024]

[0025] In the formula, are the gradients of the image in the x and y directions, representing the degree of image brightness change; edge detection is performed by calculating the magnitude of the gradient to determine whether it is a defect area; through the image gradient, the defect characteristics on the surface of the cell can be accurately extracted;

[0026] Among them, f(x, y) is the image gray value, representing the light intensity at the (x, y) coordinate position on the surface of the cell; the gradient reflects the brightness change in different regions of the image and helps to identify edges and defects;

[0027] PL - Photoluminescence technology:

[0028] The PL - Photoluminescence technology detects internal defects of the solar cell by exciting the solar cell and measuring the emitted optical signal; changes in the PL signal intensity and wavelength reveal electrical defects inside the solar cell: lattice defects, structural defects, and material inhomogeneities;

[0029] The intensity I of the PL signal PL is related to the concentration of defects, the intensity of the excitation source, and the material properties of the cell; the PL signal intensity is expressed as:

[0030]

[0031] In the formula, I PL (T) is the photoluminescence intensity at temperature T; A is a constant that depends on the type and structure of the material; E g is the bandgap of the material; E def is the defect energy level; k is the Boltzmann constant; by comparing the PL signal intensities at different wavelengths, the defects and their properties in the material are identified; represents the ratio of the difference between the bandgap of the material and the defect energy level to the thermal motion energy; is the exponential term in the photoluminescence intensity formula, which describes the relationship between the photoluminescence intensity and the bandgap of the material, the defect energy level, and the temperature;

[0032] wherein, the bandgap energy E g of the material affects the interaction between electrons and photons; the defect energy E def is the energy level related to the material defect, which reflects the internal defects of the cell; the temperature T affects the temperature of the PL signal, and the PL signal is measured at low temperature to enhance the signal intensity;

[0033] EL - Electroluminescence technology:

[0034] The EL - electroluminescence technology applies current to the battery cells, excites them, and detects the optical signals they emit; the intensity and distribution of the EL signals are closely related to the defects of the battery cells and are used to detect the internal defects of the battery cells;

[0035] The solar cell is placed in the EL excitation device and an appropriate voltage is applied; by detecting the optical signal emitted by the cell and analyzing its intensity and distribution, cracks, broken grids, scratches, depressions, poor soldering, and hot spot defects in the cell are identified;

[0036] The EL signal intensity I EL is related to the current I current and the material properties of the cell; the EL signal is described by the following formula:

[0037]

[0038] In the formula, I EL is the electroluminescence intensity; I current is the current intensity applied to the cell; B is a constant that depends on the optoelectronic properties of the material; γ is the dependence exponent of the current on the electroluminescence intensity; represents the γ - th power of the current intensity I current applied to the cell; according to the change of the EL signal, the defects generated due to current non - uniformity in the cell are detected;

[0039] Among them, the constant B is the electroluminescence characteristic constant, which reflects the luminous efficiency of the material; the exponent γ characterizes the relationship between the current intensity and the luminous intensity, and the exponent γ is linear.

[0040] Furthermore, the specific process of step two is as follows:

[0041] In the hybrid multimodal data analysis algorithm, traditional image processing techniques are used to perform denoising, edge detection, and feature extraction on the collected near-infrared images, visible light images, PL imaging, and EL imaging images, and key feature information of surface and internal defects of solar cells is extracted from different perception modes;

[0042] The system uses denoising, edge detection, and feature extraction to process the collected images, and adopts multimodal data fusion technology to enhance the detection ability of the system;

[0043] Denoising

[0044] Image denoising is a basic step in image processing, aiming to remove noise components from the image, improve the image quality, and ultimately improve the accuracy of subsequent edge detection and feature extraction;

[0045] The collected images are denoised using Gaussian filtering; Gaussian filtering is a common image smoothing method that reduces high-frequency noise by assigning a weighted average to each pixel in the image;

[0046] The formula is as follows:

[0047]

[0048] In the formula, g(x,y) is the Gaussian filtering function; σ is the standard deviation, which controls the smoothing degree of the filter; x, y are the coordinates of the pixel points in the image; the image is convolved through Gaussian filtering to effectively remove the noise in the image; the selection of the standard deviation σ directly affects the denoising effect; a larger σ value smooths the image more strongly but will cause loss of image details; represents the normalized distance from a certain point (x,y) to the center point, and the larger the distance, the smaller the weight; is used to calculate the weight of each point in the spatial domain of the Gaussian kernel; in order to improve the detection accuracy of battery cell defects, the standard deviation needs to be adjusted according to the type and intensity of the image noise;

[0049] Edge detection

[0050] Edge detection is used to identify the edges in the image, that is, the prominent positions of the brightness and color changes in the image; in defect detection, the edges correspond to the contours of the defects: cracks, broken grids, scratches, dents, poor soldering, hot spots;

[0051] Use the Canny edge detection algorithm to perform edge detection and find the edges in the image through a series of steps; the basic steps include Gaussian filtering for denoising, calculating the gradient intensity and direction of the image, using non-maximum suppression to refine the edges, and applying double-threshold detection for edge detection;

[0052] The core formula of the Canny algorithm is for image gradient calculation:

[0053]

[0054] In the formula, G is the gradient amplitude, representing the degree of image change, measuring the degree of pixel intensity change in the image, and reflecting the existence of edges; I x and I y are the gradients of the image in the x and y directions respectively, calculated through convolution operations; through the Canny algorithm, we clearly extract the defect edges on the surface and inside of the battery cell, such as cracks, broken grids, scratches, depressions, virtual soldering, and hot spots;

[0055] Feature extraction

[0056] Feature extraction is a key step in image processing, aiming to extract key features representing battery defects from the denoised and edge-detected images; these features include shape features, texture features, and color features;

[0057] For the detection of battery cell defects, shape features are the most intuitive indicators; cracks are linear, while virtual soldering appears as irregular regions; we extract the shape features of defects by calculating the area, perimeter, and shape factor indicators of the contour;

[0058] The shape factor SF is a measure reflecting the complexity of an object's shape, and the formula is:

[0059]

[0060] In the formula, the shape factor SF is an indicator measuring the complexity of the defect shape, and a smaller value indicates a more regular defect shape; P is the perimeter of the defect area; A is the area of the defect area; through this formula, we judge the shape complexity of the defect based on the perimeter and area of the contour; cracks have a larger shape factor, while virtual soldering and hot spots have smaller shape factors;

[0061] Multi-modal data fusion

[0062] Multi-modal data fusion is the process of combining image data from different sources to extract more defect information; the defect information captured by each sensing technology is different; the purpose of data fusion is to improve the detection ability of the system and reduce false detections and missed detections of single modes;

[0063] A common multimodal data fusion method is weighted fusion. The basic idea is to assign a weight to each modality and weight the results according to the reliability and importance of each sensing technology. The results of the four sensing technologies are D NIR , D Visible , D PL , D EL , and their weighted fusion formula is as follows:

[0064] D final = w1·D NIR + w2·D Visible + w3·D PL + w4·D EL

[0065] In the formula, D final is the final detection result after fusion, representing the type and location of the defect; D NIR , D Visible , D PL , D EL are the detection results of each modality; w1, w2, w3, w4 are the weight coefficients of each modality, satisfying w1 + w2 + w3 + w4 = 1, and the weights w1, w2, w3, w4 are dynamically adjusted according to the accuracy of each technology and its performance in specific defect detection.

[0066] Furthermore, the specific process of step three is as follows:

[0067] Adopt the multimodal data fusion method, comprehensively analyze the data of different technologies through the decision-level fusion strategy to improve the accuracy of defect detection; in addition, by combining the convolutional neural network technology to perform in-depth learning analysis, classification, and localization on the data, the defect types of battery cells can be identified, and corresponding labels can be assigned to different types of defects to accurately distinguish surface defects and internal defects;

[0068] The following is a detailed expansion of the implementation process of this step, starting from the multimodal data fusion method:

[0069] Multimodal data fusion combines information from different sensors, including near-infrared lasers, array visible light, PL-photoluminescence, EL-electroluminescence, and utilizes their complementarity to improve detection accuracy; the fusion strategy is processed hierarchically based on different data levels, and multimodal fusion is achieved through decision-level fusion;

[0070] Decision-level fusion is the process of fusing the detection results of different sensing modalities; after multiple individual classifiers and detectors complete defect recognition, the decisions of each classifier are combined; decision-level fusion is based on the majority voting method and the weighted voting method to determine the final classification result;

[0071] Next is the convolutional neural network analysis:

[0072] Based on multi-modal data fusion, the convolutional neural network is used for deep learning analysis of data, which can more accurately identify and locate defects in battery cells;

[0073] Convolutional neural network model

[0074] The convolutional neural network consists of the following key layers:

[0075] Convolutional layer: Extract local features from the image and use multiple convolutional layers to learn different features;

[0076] Pooling layer: Reduce the dimension of the feature map and reduce the computational complexity; Common pooling methods include max pooling and average pooling;

[0077] Fully connected layer: Integrate the extracted features and output the defect category;

[0078] We input the multi-modal fusion image data into the convolutional neural network model. After multiple convolutional and pooling processes, the output defect category is represented as:

[0079] y = f(W·Flatten(Pooling(Convolution(I fused )))+b)

[0080] In the formula, y is the output of the model, indicating that the defect categories are crack, broken grid, scratch, depression, false soldering, and hot spot; I fused is the image after multi-modal data fusion; W is the weight matrix, b is the bias term; f is the activation function, used to convert the output of the network into class probabilities; Convolution is the convolutional layer, which extracts local features in the image and uses multiple convolutional kernels to learn different features; Pooling is the pooling layer, which reduces the dimension of the feature map and reduces the amount of calculation; Flatten flattens the feature map output by the pooling layer into a one-dimensional vector for processing by the fully connected layer;

[0081] Classification and localization

[0082] The convolutional neural network can not only classify defects but also accurately locate defects in the image; By using the region proposal network, the convolutional neural network generates candidate regions for each defect and further performs fine classification on these candidate regions;

[0083] The candidate region R output by the network candidate :

[0084] R candidate =(x1,y1,x2,y2)

[0085] Where, (x1, y1) and (x2, y2) respectively represent the coordinates of the upper left corner and the lower right corner of the candidate region; the label output by the network is whether there is a defect in each candidate region and the specific type of the defect;

[0086] In this way, the convolutional neural network can not only identify the type of the defect, but also calibrate the position of the defect, so as to provide accurate positioning information for subsequent production line adjustment and optimization;

[0087] Loss Function and Training

[0088] The convolutional neural network model is trained using the cross-entropy loss function;

[0089] y true is the true label, and y pred is the label predicted by the model. Then the cross-entropy loss function L is:

[0090]

[0091] Where, y true,i is the value of the i-th category of the true label, which is 0 or 1; y pred,i is the probability of the i-th category of the predicted label;

[0092] The weights and biases of the network are optimized through the backpropagation algorithm, so that the model gradually converges to the optimal parameters, thereby realizing the accurate classification and positioning of the defect type.

[0093] Furthermore, the specific process of step four is as follows:

[0094] In this step, the system provides feedback on the defect detection results obtained from multi-modal data fusion and adjusts the production line in real time through the automatic control module; the goal of this process is to improve the automation level of the production line, reduce human intervention, and ensure that measures are taken in a timely manner to adjust the production process when defects are detected, thereby improving production efficiency and product quality;

[0095] Workflow of the Feedback and Automatic Control Module

[0096] The automatic control module plays a core role in this step and relies on the defect detection results to implement production line adjustment; the basic process includes the following steps:

[0097] Defect Identification and Marking: The system first accurately identifies and marks the position and type of the defect on the battery cell according to the detection results obtained from multi-modal data fusion; this process is realized by combining the image processing system and the defect identification module of the convolutional neural network;

[0098] Defect Classification and Location: For each type of marked defect, including cracks, broken networks, scratches, dents, poor soldering, and hot spots, there is its specific location on the battery cell, and it is transmitted to the automated control system through a real-time feedback mechanism;

[0099] Production Line Adjustment: Based on the defect information, the automated control system will automatically adjust various links of the production line: repair equipment, laser cutting module, surface treatment, and different adjustment measures will be taken according to the defect type;

[0100] Formulas and Parameters for Adjusting the Production Line: During this process, the system dynamically adjusts through real-time feedback data; in order to achieve efficient feedback and adjustment, the following parameters and formulas need to be considered for production line adjustment:

[0101] First, Production Line Speed Adjustment:

[0102] According to the severity of the defect type and the requirements of the repair process, the speed of the production line needs to be adjusted; the original production line speed is V0, and the severity of the detected defect is S, then the system optimizes the production process by adjusting the speed V; the adjusted speed is calculated by the following formula:

[0103] V = V0·(1 - k·S)

[0104] In the formula, V is the production line speed adjusted by the system; V0 is the original production line speed; k is the adjustment coefficient, which depends on the defect type and the response time of the repair equipment; S is the defect severity, with a value range of 0 to 1, indicating the degree of defect severity. Larger cracks result in larger S values;

[0105] This formula indicates that when the defect is severe, the production line speed will be reduced to provide more time for the repair process, thus avoiding the defect affecting the next production step;

[0106] Second, Operation of the Automated Repair System:

[0107] When the system detects cracks, poor soldering, and other defects on the battery cell, the automated repair equipment will repair the battery cell; the laser power of the laser repair equipment is P, and the time required for the repair process is t repair , then the energy E required for the repair process repair can be calculated by the following formula:

[0108] E repair = P·t repair

[0109] In the formula, E repair is the energy required for the repair process; P is the laser power of the laser repair equipment; t repair is the time of the repair process;

[0110] The system automatically adjusts P and t according to the size and depth of the crack repair values;

[0111] Third, equipment recalibration:

[0112] If the defect is determined to be caused by equipment precision problems, the automated control system will issue a command to recalibrate the equipment; if the precision error of the equipment is Δ∈, the calibration process of the equipment is adjusted through the following formula:

[0113] Δ∈ adjusted = Δ∈·(1 - α)

[0114] In the formula, Δ∈ adjusted is the calibrated equipment precision; Δ∈ is the current precision error of the equipment; α is the calibration coefficient, indicating the reduction ratio of the error after calibration, and α ranges between 0 and 1;

[0115] This formula shows that after adjusting the equipment precision, the error will be reduced, ensuring the improvement of precision in the production process; Integrated feedback system and real-time reporting

[0116] The automated control module is combined with the feedback system to generate real-time defect reports and transmit them to managers and operators; The report includes the following information:

[0117] Detected defect types: crack, broken network, scratch, dent, cold solder joint, hot spot;

[0118] Exact location of the defect: specific coordinates;

[0119] Size and severity of the defect: crack length, cold solder joint area;

[0120] Production line adjustment measures: speed adjustment, rework processing;

[0121] Working status of the repair equipment: power settings and working hours of the laser repair machine.

[0122] These reports are displayed in real time through the visualization platform, helping production managers to promptly grasp the problems in the production process and make corresponding decisions quickly.

[0123] Beneficial effects

[0124] By combining multi-modal sensing technology, the present invention overcomes the limitations of single sensing technology in the prior art and effectively improves the accuracy and reliability of solar cell defect detection. There are the following specific advantages:

[0125] 1. Comprehensiveness

[0126] The present invention uses four different sensing technologies (near-infrared laser, array visible light, PL photoluminescence, EL electroluminescence) to comprehensively cover the surface and internal defects of battery cells, ensuring the comprehensiveness of the detection range.

[0127] 2. High precision

[0128] The multi-modal data fusion algorithm combines the data characteristics of various technologies, which can accurately distinguish various surface and internal defects, improving the accuracy and precision of detection.

[0129] 3. Intelligence

[0130] By using a convolutional neural network for deep learning analysis, effective information can be automatically extracted from a large amount of data, identifying and classifying defect types, greatly reducing manual intervention and improving detection efficiency.

[0131] 4. Real-time feedback and automatic adjustment

[0132] Through the defect marking and feedback mechanism on the production line, the production process can be adjusted in real time, optimizing the production process, reducing the generation of defective products, and improving the automation level and production efficiency of the production line. Brief description of the drawings

[0133] Figure 1 Overall flowchart of the solar cell defect detection system based on multi-modal perception technology

[0134] Figure 2 Defect recognition flowchart of solar cells based on multi-modal data fusion

[0135] Figure 3 Defect marking and production line adjustment flowchart based on dynamic feedback adjustment algorithm Detailed implementation manners

[0136] Next, a clear and complete description of the technical solutions in the embodiments of the present invention will be provided. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. Based on these embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0137] In this case, as Figure 2 :

[0138] First, multi-modal data of solar cells are collected in real time through four types of sensing technologies, and image processing and defect detection are performed.

[0139] 1. Near-infrared laser defect detection

[0140] Assume that the near-infrared laser image is I NIR, The image size is 256×256; let the variance σ of the Gaussian noise 2 = 0.1.

[0141] First, perform Gaussian filtering denoising on the near-infrared image, and the denoising can be carried out through the following formula:

[0142] I NIR，denoised = GaussianFilter(I NIR , σ)

[0143] I NIR，denoised is the denoised near-infrared image.

[0144] Next, use the Canny edge detection algorithm to extract the edges of the image, and set the thresholds low_threshold = 50 and high_threshold = 50.

[0145] I NIR，edge = Canny(I NIR，denoised , low_threshold, high_threshold)

[0146] I NIR，edge is the edge image, which shows the edge information of the defects in the image.

[0147] Then, extract features such as shape and area from the edge image, and the extracted feature vector is F NIR ,

[0148] F NIR = ExtractFeatures(I NIR，edge )

[0149] The feature vector F NIR , which contains information such as the shape and area of the defects.

[0150] Finally, use a pre-trained convolutional neural network to classify the feature vector, and the categories output by the network are

[0151] y NIR = 0 (crack), y NIR = 0 (no defect); c NIR = 0 (broken grid), c NIR = 0 (no defect);

[0152] s NIR = 0 (scratch), s NIR = 0 (no defect); d NIR = 0 (dent), d NIR = 0 (no defect);

[0153] f NIR = 0 (dry joint), fNIR = 0 (no defect):

[0154] y NIR = CNN(F NIR )

[0155] c NIR = CNN(F NIR )

[0156] s NIR = CNN(F NIR )

[0157] d NIR = CNN(F NIR )

[0158] f NIR = CNN(F NIR )

[0159] The result obtained is:

[0160] y NIR = 0

[0161] c NIR = 1

[0162] s NIR = 1

[0163] d NIR = 1

[0164] f NIR = 1

[0165] The result shows that there is a crack in the cell.

[0166] 2. Defect Detection of Area Array Visible Light Images

[0167] Assume that the near-infrared laser image is I Visible , and the image size is 512×512; let the variance σ of the Gaussian noise 2 = 0.1.

[0168] First, perform Gaussian filtering denoising on the near-infrared image, and the denoising can be carried out through the following formula:

[0169] I Visible，denoised = GaussianFilter(I Visible , σ)

[0170] I Visible，denoised is the denoised near-infrared image.

[0171] Next, use the Canny edge detection algorithm to extract the edges of the image, setting the thresholds low_threshold = 50 and high_threshold = 50.

[0172] I Visible，edge = Canny(I Visible，denoised , low_threshold, high_threshold)

[0173] I Visible，edge is the edge image, which shows the edge information of the defects in the image.

[0174] Then, extract features such as shape and area from the edge image. The extracted feature vector is F Visible ,

[0175] F Visible = ExtractFeatures(I Visible，edge )

[0176] The feature vector F Visible , which contains information such as the shape and area of the defects.

[0177] Finally, use a pre-trained convolutional neural network to classify the feature vector. The output class of the network is y Visible = 0 (crack), y Visible = 0 (no defect); c Visible = 0 (broken grid), c Visible = 0 (no defect); s Visible = 0 (scratch), s Visible = 0 (no defect); d Visible = 0 (dent), d Visible = 0 (no defect); f Visible = 0 (dry joint), f Visible = 0 (no defect):

[0178] y Visible = CNN(F Visible )

[0179] c Visible = CNN(F Visible )

[0180] s Visible = CNN(F Visible )

[0181] d Visible = CNN(F Visible )

[0182] f Visible = CNN(F Visible )

[0183] The result obtained is:

[0184] y Visible = 0

[0185] c Visible = 1

[0186] s Visible = 1

[0187] d Visible = 1

[0188] f Visible = 1

[0189] The result shows that there are cracks in the cell.

[0190] 3. PL - Photoluminescence Defect Detection

[0191] Assume that the near - infrared laser image is I PL , and the image size is 256×256; let the variance of Gaussian noise be σ 2 = 0.1.

[0192] First, perform Gaussian filtering denoising on the near - infrared image, and the denoising can be carried out through the following formula:

[0193] I PL，denoised = GaussianFilter(I PL , σ)

[0194] I PL，denoised is the denoised near - infrared image.

[0195] Next, use the Canny edge detection algorithm to extract the edges of the image, and set the thresholds low_threshold = 100, high_threshold = 200.

[0196] I PL，edge = Canny(I PL，denoised , low_threshold, high_threshold)

[0197] I PL，edge is the edge image, which shows the edge information of the defects in the image.

[0198] Then, extract features such as shape and area from the edge image, and the extracted feature vector is F PL ,

[0199] F PL = ExtractFeatures(I PL，edge )

[0200] Feature vector F PL , which contains information such as the shape and area of the defect.

[0201] Finally, use a pre-trained convolutional neural network to classify the feature vector, and the class output by the network is y PL = 0 (crack), y PL = 0 (no defect); c PL = 0 (broken grid), c PL = 0 (no defect); s PL = 0 (scratch), s PL = 0 (dent), d PL = 0 (no defect); f PL = 0 (dry joint), f PL = 0 (no defect): PL = 0 (no defect):

[0202] y PL = CNN(F PL )

[0203] c PL = CNN(F PL )

[0204] s PL = CNN(F PL )

[0205] d PL = CNN(F PL )

[0206] f PL = CNN(F PL )

[0207] The result is obtained:

[0208] y PL = 1

[0209] c PL = 1

[0210] s PL = 1

[0211] d PL = 1

[0212] f PL = 1

[0213] The result shows that there are no defects in the cell.

[0214] 4. EL - Electroluminescence Defect Detection

[0215] The near - infrared laser image is I EL, The image size is 256×256; let the variance of Gaussian noise be σ 2 = 0.1. First, perform Gaussian filtering denoising on the near-infrared image, and the denoising can be carried out through the following formula:

[0216] I EL，denoised = GaussianFilter(I EL , σ)

[0217] I EL，denoised is the denoised near-infrared image.

[0218] Next, use the Canny edge detection algorithm to extract the edges of the image, and set the thresholds low_threshold = 50 and high_threshold = 50.

[0219] I EL，edge = Canny(I EL，denoised , low_threshold, high_threshold)

[0220] I EL，edge is the edge image, which shows the edge information of the defects in the image.

[0221] Then, extract features such as shape and area from the edge image, and the extracted feature vector is F EL ,

[0222] F EL = ExtractFeatures(I EL，edge )

[0223] The feature vector F EL , which contains information such as the shape and area of the defects.

[0224] Finally, use a pre-trained convolutional neural network to classify the feature vector, and the output category of the network is y EL = 0 (crack), y EL = 0 (no defect); c EL = 0 (broken grid), c EL = 0 (no defect); s EL = 0 (scratch), s EL = 0 (no defect); d EL = 0 (dent), d EL = 0 (no defect); f EL = 0 (dry joint), f EL = 0 (no defect):

[0225] y EL = CNN(F EL )

[0226] c EL = CNN(F EL )

[0227] s EL = CNN(F EL )

[0228] d EL = CNN(F EL )

[0229] f EL = CNN(F EL )

[0230] The result obtained is:

[0231] y EL = 0

[0232] c EL = 1

[0233] s EL = 1

[0234] d EL = 1

[0235] f EL = 1

[0236] The result shows that there is a crack in this cell.

[0237] 5. Multimodal data fusion

[0238] First, images of different modalities (near-infrared, visible light, PL, EL) are used as inputs for feature fusion, and the feature vector F of each modality is calculated by weighted average fused :

[0239] F fused = w NIP ·F NIR + w Visible ·F Visible + w PL ·F PL + ·F EL

[0240] where w NIR , w Visible , w PL , w EL are weight coefficients representing the relative importance of each modality.

[0241] Second, the fused feature F fused is also input into a convolutional neural network for defect classification:

[0242] y fused = CNN(Ffused )

[0243] c fused = CNN(F fused )

[0244] s fused = CNN(F fused )

[0245] d fused = CNN(F fused )

[0246] f fused = CNN(F fused )

[0247] Get the result

[0248] y fused = 0

[0249] c fused = 1

[0250] s fused = 1

[0251] d fused = 1

[0252] f fused = 1

[0253] The result shows that there is a crack in the cell.

[0254] Finally, through decision-level fusion, the result is obtained that the defect existing in the solar panel is a crack.

[0255] In this case, such as Figure 3 :[[]]END]]

[0256] The defect detection results from multi-modal data fusion, after processing different modal images (near-infrared, visible light, PL, EL) through a convolutional neural network, the obtained defect types and locations.

[0257] The detected defect type is "crack", the defect severity score is 0.85, and the location is the coordinate (x, y) = (120, 150).

[0258] The current manufacturing process data on the production line, the real-time data of the production line equipment are as follows:

[0259] The current production speed: v current = 5m / s

[0260] The current position of the cell: x current = 120, y current = 150

[0261] Production line temperature: T current = 300K

[0262] Production line status: normal operating status

[0263] The current control system parameters are:

[0264] Adjustment factor: w speed = 0.5

[0265] Temperature control parameter: w temperature = 0.2

[0266] Defect severity weight: w crack = 0.3

[0267] Next, start the adjustment of the production line:

[0268] 1. Defect marking and classification

[0269] In the result of multi-modal data fusion, through the CNN model, we obtained the location and type of the defect. For example, the defect in the near-infrared image is "crack", and the defect location is (120, 150), and the defect severity score is 0.85. This information will be marked as a defect on the cell:

[0270] Defect marking: location (120, 150), defect type is "crack", severity score is 0.85.

[0271] Feedback signal: Feedback = (120, 150, defect, 0.85)

[0272] 2. Feedback and input to the control module

[0273] Combine the defect marking results (such as the location, type, and severity score of the crack) with the real-time data of the production line to form the feedback input. For example, in the real-time data of the production line, the current position of the cell is the same as the defect location, that is, x current = x crack , y current = y crack , and the production line speed is 5 m / s.

[0274] Current production line position: x current = 120, y current = 150

[0275] Current production speed: v current = 5 m / s

[0276] Defect severity score: S crack = 0.85

[0277] 3. Production line adjustment decision

[0278] Based on the severity score of the defect, the control module will decide whether to adjust production parameters such as production speed and temperature to reduce or avoid the recurrence of similar defects.

[0279] First, adjust the production speed.

[0280] Δv = w speed ×(1 - S crack )×v current

[0281] where w speed = 0.5 is the weight coefficient for production speed adjustment, and S crack = 0.85 is the severity score of the crack.

[0282] Δv = 0.5×(1 - 0.85)×5 = 0.375m / s

[0283] The new production speed is:

[0284] v new = v current -Δv = 5 - 0.375 = 4.625m / s

[0285] Next, adjust the temperature.

[0286] ΔT = w temperature ×(1 - S crack )×T current

[0287] where w temperature = 0.2 is the weight coefficient for temperature adjustment, and T current = 300K is the current temperature.

[0288] Calculated as:

[0289] ΔT = 0.2×(1 - 0.85)×300 = 9K

[0290] The new temperature is:

[0291] T new = T current -ΔT = 300 - 9 = 291K

[0292] 4. Adjust the operation of the production line

[0293] According to the calculation results, the feedback signal will trigger the automation control module to make corresponding adjustments.

[0294] Adjust the production speed to v new = 4.625m / s

[0295] Adjust the temperature of the production line to Tnew = 291K

[0296] The control system will automatically adjust the equipment settings, ensuring that the defect marking information is consistent with the production line status through real-time feedback, thereby optimizing the production process and reducing defect generation.

[0297] Finally, record the defect marking and production adjustment:

[0298] Defect type: Crack

[0299] Defect location: (120, 150)

[0300] Defect severity score: 0.85

[0301] Adjusted production speed: 4.625 m / s

[0302] Adjusted temperature: 291K

[0303] Through the above steps, the feedback and automation control module can timely adjust the production line parameters, thereby reducing the defect rate caused by defects and further improving the production efficiency and quality of solar cells.

[0304] This implementation case demonstrates the specific operation of adjusting the production process through the feedback and automation control module according to the defect detection results after multi-modal data fusion, combined with the real-time data of the production line. The input of the algorithm includes defect type, location, severity score, and production line status data; the output of the algorithm includes control instructions, such as adjusted production speed and temperature.

Claims

1. A method for detecting defects in solar cells based on multi-modal perception technology, characterized in that Including the following steps: Step 1: Utilize four sensing technologies, including near-infrared laser, area array visible light, PL - photoluminescence, and EL - electroluminescence technologies, to collect real-time multimodal data of solar cells; these technologies are used to identify surface and internal defects of solar cells; Step 2: Apply traditional image processing algorithms to the analysis of hybrid multimodal data, processing the collected near-infrared images, visible light images, PL images, and EL images; including denoising, edge detection, and feature extraction to obtain key defect features of surface and internal defects from different sensing modes; Step 3: Adopt a multimodal data fusion method and combine a decision-level fusion strategy to comprehensively analyze data from different technologies, improving the detection accuracy; by integrating multiple data layers, the system gives full play to the advantages of each technology and minimizes errors caused by the limitations of a single technology; on this basis, apply a convolutional neural network (CNN) for deep learning analysis, classification, and localization to identify defect types such as cracks, broken grids, scratches, depressions, poor soldering, and hot spots, and assign corresponding labels to different defect types to ensure the accurate distinction between surface defects and internal defects; Step 4: According to the defect detection results of multimodal data fusion, mark the defects and adjust the production line through a feedback and automation control module.

2. The method for detecting defects of solar cells based on multi-modal perception technology according to claim 1, wherein, The specific process of Step 1 is as follows: By implementing four sensing technologies: near-infrared laser, area array visible light, PL - photoluminescence, and EL - electroluminescence, the system collects real-time multimodal data of solar cells on the production line and analyzes it through subsequent intelligent algorithms; by precisely adjusting the sensor acquisition parameters and applying these technologies, the system can identify and locate surface and internal defects of solar cells, thus achieving efficient and accurate detection of solar cells; The four types of sensing technologies are as follows: Near-infrared laser technology: The near-infrared laser technology irradiates the surface of the solar cell with a laser and measures the intensities of the reflected and transmitted lasers to detect internal defects of the solar cell; the near-infrared laser can penetrate the surface layer of the cell to detect internal defects such as cracks, broken grids, scratches, depressions, poor soldering, and hot spots; A group of near-infrared laser sensors is installed on the production line of the battery wafers. The laser beam penetrates the battery wafer and is reflected, and the sensor measures the intensity of the reflected light; based on the change in the reflected light, it is speculated whether there are defects in the battery wafer; The intensity of the reflected light is described by the following formula: I ref = I0·e -α·d Where, I ref is the reflected light intensity, I0 is the initial laser intensity, α is the absorption coefficient of the near-infrared laser in the cell material, d is the thickness of the cell, and e -α·d represents the attenuation ratio of the laser after passing through the cell thickness d, and is a key factor for calculating the reflected light intensity I ref ; among them, the absorption coefficient α affects the attenuation degree of the laser and varies with different materials, and the thickness d of the cell directly affects the laser transmission ability; Area array visible light technology: The area array visible light imaging technology illuminates the surface of the solar cell and uses an imaging device to capture the reflected light image to identify surface defects; subsequently, image processing technology is applied to analyze the light intensity change of each pixel, and the smallest flaws on the surface of the solar cell can be detected; Install a high-resolution area array visible light camera to perform real-time imaging on the surface of the battery wafer; the resolution of this camera is 3000×2000 pixels, and each pixel corresponds to a tiny area on the surface of the battery wafer; through the collected images, traditional image processing algorithms are used to analyze the images to identify cracks, broken grids, scratches, depressions, poor soldering, and hot spot defects on the surface; The image processing of area array visible light imaging identifies edges through image gradient calculation; the image f(x,y) represents the surface image of the collected solar cell, and edge detection is performed through the gradient of the image; Gradient calculation formula: In the formula, is the gradient of the image in the x and y directions, representing the degree of change in image brightness; edge detection determines whether it is a defect area by calculating the magnitude of the gradient ; through the image gradient, the defect features on the surface of the battery cell can be accurately extracted; Among them, f(x,y) is the image grayscale value, representing the light intensity at the (x,y) coordinate position on the surface of the solar cell; the gradient reflects the brightness changes in different regions of the image, helping to identify edges and defects; PL - Photoluminescence technology: The PL - Photoluminescence technology detects internal defects of a solar cell by exciting the solar cell and measuring the emitted optical signal; changes in the PL signal intensity and wavelength reveal electrical defects inside the solar cell: lattice defects, structural defects, and material inhomogeneities; The intensity I of the PL signal PL is related to the concentration of defects, the intensity of the excitation source, and the material properties of the cell; the PL signal intensity is expressed as: Where, I PL (T) is the photoluminescence intensity at temperature T; A is a constant that depends on the type and structure of the material; E g is the bandgap of the material; E def is the defect energy level; k is the Boltzmann constant; by comparing the PL signal intensities at different wavelengths, the defects and their properties in the material are identified; represents the ratio of the difference between the bandgap of the material and the defect energy level to the thermal motion energy; is the exponential term in the photoluminescence intensity formula, which describes the relationship between the photoluminescence intensity and the bandgap of the material, the defect energy level, and the temperature; Among them, the bandgap energy E of the material g affects the interaction between electrons and photons; the defect energy E def is the energy level related to material defects, reflecting the internal defects of the cell; the temperature T affects the temperature of the PL signal, and the PL signal is measured at low temperature to enhance the signal intensity; EL - Electroluminescence technology: The EL - Electroluminescence technology applies current to the battery cells, excites them, and detects the optical signals they emit; the intensity and distribution of the EL signal are closely related to the defects of the battery cells and are used to detect internal defects of the battery cells; Place the solar cell in the EL excitation device and apply an appropriate voltage; by detecting the optical signal emitted by the solar cell and analyzing its intensity and distribution, cracks, broken grids, scratches, dents, false soldering, and hot spot defects in the solar cell are identified; EL signal intensity I EL is related to the current I current and the material properties of the cell; the EL signal is described by the following formula: Where, I EL is the electroluminescence intensity; I current is the current intensity applied to the cell; B is a constant that depends on the optoelectronic properties of the material; γ is the dependence exponent of the current on the electroluminescence intensity; represents the γ-th power of the current intensity I current applied to the cell; According to the change of the EL signal, defects generated due to current non-uniformity in the cell are detected; Among them, the constant B is the electroluminescence characteristic constant, which reflects the luminous efficiency of the material; the exponent γ characterizes the relationship between the current intensity and the luminous intensity, and the exponent γ is linear.

3. A method for detecting defects of solar cells based on multi-modal perception technology according to claim 1, characterized in that The specific process of step 2 is as follows: Use traditional image processing techniques in the hybrid multimodal data analysis algorithm to perform denoising, edge detection, and feature extraction on the collected near - infrared images, visible light images, PL imaging, and EL imaging images, and extract key feature information of the surface and internal defects of the solar cell from different perception modes; The system uses denoising, edge detection, and feature extraction to process the collected images and adopts multimodal data fusion technology to enhance the detection ability of the system; Denoising Image denoising is a basic step in image processing, aiming to remove noise components from the image, improve image quality, and ultimately improve the accuracy of subsequent edge detection and feature extraction; Use Gaussian filtering to perform denoising processing on the collected images; Gaussian filtering is a common image smoothing method that reduces high - frequency noise by assigning weighted average values to each pixel in the image; The formula is as follows: In the formula, g(x, y) is the Gaussian filtering function; σ is the standard deviation, which controls the smoothing degree of the filter; x and y are the coordinates of the pixel points in the image; convolving the image through Gaussian filtering can effectively remove the noise in the image; the selection of the standard deviation σ directly affects the denoising effect; a larger σ value smoother the image more strongly, but will cause the loss of image details; represents the normalized distance from a certain point (x, y) to the center point, and the larger the distance, the smaller the weight; is used to calculate the weight of each point in the spatial domain of the Gaussian kernel; To improve the detection accuracy of solar cell defects, the standard deviation needs to be adjusted according to the type and intensity of the image noise; Edge detection Edge detection is used to identify edges in the image, that is, the prominent positions where brightness and color change in the image; In defect detection, the edge corresponds to the contour of the defect: crack, broken network, scratch, dent, false soldering, hot spot; Use the Canny edge detection algorithm to perform edge detection, and find edges in the image through a series of steps; the basic steps include Gaussian filtering for denoising, calculating the gradient intensity and direction of the image, using non - maximum suppression to refine the edges, and applying double - threshold to detect edges; The core formula of the Canny algorithm is image gradient calculation: Where G is the gradient magnitude, representing the degree of image change, measuring the degree of pixel intensity change in the image, and reflecting the presence of edges; I x and I y are the gradients of the image in the x and y directions respectively, calculated through convolution operations; through the Canny algorithm, we clearly extract the defect edges on the surface and inside of the battery cell, such as cracks, broken grids, scratches, depressions, virtual soldering, and hot spots; Feature extraction Feature extraction is a key step in image processing, aiming to extract key features representing battery defects from the denoised and edge - detected images; these features include shape features, texture features, and color features; For the detection of battery cell defects, shape features are the most intuitive indicators; cracks are linear, while virtual welding appears as irregular regions; we extract the shape features of defects by calculating the area, perimeter, and shape factor metrics of the contour; The shape factor SF is a measure reflecting the complexity of an object's shape, and the formula is: In the formula, the shape factor SF is an indicator for measuring the complexity of the defect shape. The smaller the value, the more regular the defect shape; P is the perimeter of the defect area; A is the area of the defect area; through this formula, we judge the shape complexity of the defect based on the perimeter and area of the contour; the shape factor of cracks is larger, while the shape factors of virtual soldering and hot spots are smaller; Multi-modal data fusion Multi-modal data fusion is the process of combining image data from different sources to extract more defect information; the defect information captured by each sensing technology is different; the purpose of data fusion is to improve the detection ability of the system and reduce false detections and missed detections of single modalities; A common multi-modal data fusion method is weighted fusion. The basic idea is to assign a weight to each modality and weight the results according to the reliability and importance of each sensing technology; the results of the four sensing technologies are D NIR , D Visible , D PL , D EL , and their weighted fusion formula is as follows: D final = w1·D NIR + w2·D Visible + w3·D PL + w4·D EL where D final is the final detection result after fusion, representing the type and location of the defect; D NIR , D Visible , D PL , D EL are the detection results of each modality; w1, w2, w3, w4 are the weight coefficients of each modality respectively, satisfying w1 + w2 + w3 + w4 = 1, and the weights w1, w2, w3, w4 are dynamically adjusted according to the accuracy of each technology and its performance in specific defect detection.

4. A method for detecting defects in solar cells based on multi-modal perception technology according to claim 1, characterized in that, The specific process of Step 3 is as follows: Adopt the multi-modal data fusion method, and comprehensively analyze the data of different technologies through the decision-level fusion strategy to improve the accuracy of defect detection; in addition, by combining convolutional neural network technology for in-depth learning analysis, classification, and localization of the data, the defect types of battery cells can be identified, and corresponding labels can be assigned to different types of defects to accurately distinguish surface defects and internal defects; The following is a detailed expansion of the implementation process of this step, starting from the multi-modal data fusion method: Multi-modal data fusion combines information from different sensors, including near-infrared lasers, array visible light, PL-photoluminescence, and EL-electroluminescence, and utilizes their complementarity to improve the detection accuracy; the fusion strategy is hierarchically processed based on different data levels, and multi-modal fusion is achieved through decision-level fusion; Decision-level fusion is the process of fusing the detection results of different sensing modalities; this method combines the decisions of each classifier after multiple individual classifiers and detectors complete defect recognition; decision-level fusion determines the final classification result based on the majority voting method and the weighted voting method; Next is the convolutional neural network analysis: Based on multi-modal data fusion, convolutional neural networks are used for in-depth learning analysis of data, which can more accurately identify and locate defects in battery cells; Convolutional neural network model A convolutional neural network consists of the following key layers as follows: Convolutional layer: Extract local features from the image and use multiple convolutional layers to learn different features; Pooling layer: Reduce the dimension of the feature map and reduce the computational complexity; common pooling methods include max pooling and average pooling; Fully connected layer: Synthesize the extracted features and output the defect category; We input the multi-modal fusion image data into the convolutional neural network model. After multiple layers of convolution and pooling processing, the output defect category is expressed as: y = f(W·Flatten(Pooling(Convolution(I fused )) + b) Where y is the output of the model, indicating that the defect categories are crack, broken grid, scratch, dent, false soldering, and hot spot; I fused is the image after multi-modal data fusion; W is the weight matrix, b is the bias term; f is the activation function, which is used to convert the output of the network into class probabilities; Convolution is the convolutional layer that extracts local features in the image and uses multiple convolutional kernels to learn different features; Pooling is the pooling layer that reduces the dimension of the feature map and the computational amount; Flatten flattens the feature map output by the pooling layer into a one-dimensional vector for processing by the fully connected layer; Classification and Localization The convolutional neural network can not only classify defects, but also accurately locate defects in the image; by using the region proposal network, the convolutional neural network generates candidate regions for each defect and further performs fine classification on these candidate regions; Candidate region R output by the network candidate : R candidate = (x1, y1, x2, y2) In the formula, (x1, y1) and (x2, y2) respectively represent the coordinates of the upper left corner and the lower right corner of the candidate region; the label output by the network is whether there is a defect in each candidate region and the specific type of the defect; In this way, the convolutional neural network can not only identify the type of defect, but also calibrate the location of the defect, thereby providing accurate location information for subsequent production line adjustment and optimization; Loss Function and Training The convolutional neural network model is trained using the cross-entropy loss function; y true is the true label, y pred is the label predicted by the model, then the cross-entropy loss function L is: where y true,i is the value of the i-th category of the true label, which is 0 or 1; y pred,i is the probability of the i-th category of the predicted label; The weights and biases of the network are optimized through the backpropagation algorithm, so that the model gradually converges to the optimal parameters, thereby achieving accurate classification and localization of defect types.

5. A method for detecting defects in solar cells based on multi-modal perception technology according to claim 1, characterized in that, The specific process of Step 4 is as follows: In this step, the system provides feedback on the defect detection results obtained from the multi-modal data fusion and adjusts the production line in real time through the automatic control module; the goal of this process is to improve the automation level of the production line, reduce human intervention, and ensure that measures are taken in a timely manner to adjust the production process when defects are detected, thereby improving production efficiency and product quality; Workflow of the Feedback and Automatic Control Module The automatic control module plays a core role in this step and relies on the defect detection results to implement production line adjustment; The basic process includes the following steps: Defect Identification and Marking: The system first accurately identifies and marks the location and type of defects on the battery cell according to the detection results obtained from the multi-modal data fusion; this process is realized by combining the image processing system and the defect identification module of the convolutional neural network; Defect Classification and Localization: Each marked defect type, including cracks, broken networks, scratches, dents, false soldering, and hot spots, has its specific location on the battery cell and is transmitted to the automatic control system through a real-time feedback mechanism; Production Line Adjustment: According to the defect information, the automatic control system will automatically adjust each link of the production line: repair equipment, laser cutting module, surface treatment, and take different adjustment measures according to the defect type; Formulas and Parameters for Adjusting the Production Line: In this process, the system dynamically adjusts through real-time feedback data; in order to achieve efficient feedback and adjustment, the production line adjustment needs to consider the following parameters and formulas: First, Production Line Speed Adjustment: According to the severity of the defect type and the requirements of the repair process, the speed of the production line needs to be adjusted; the original production line speed is V0, and the severity of the detected defect is S, then the system optimizes the production process by adjusting the speed V; the adjusted speed is calculated by the following formula: V = V0·(1 - k·S) Wherein, V is the production line speed after system adjustment; V0 is the original production line speed; k is the adjustment coefficient, which depends on the type of defect and the response time of the repair equipment; S is the defect severity, with a value range of 0 to 1, indicating the severity of the defect. The larger the crack, the larger the S value. This formula means that when the defect is severe, the production line speed will be reduced to provide more time for the repair process, thus avoiding the defect affecting the next production step. Second, the operation of the automated repair system: When the system detects cracks, poor soldering and other defects on the solar cell, the automated repair device will repair the solar cell; the laser power of the laser repair device is P, and the time required for the repair process is t repair , then the energy E required for the repair process repair can be calculated by the following formula: E repair = P·t repair where E repair is the energy required for the repair process; P is the laser power of the laser repair device; t repair is the time of the repair process; The system automatically adjusts the values of P and t according to the size and depth of the crack repair ; Third, equipment recalibration: If the defect is determined to be caused by equipment precision problems, the automated control system will issue a command to recalibrate the equipment. If the precision error of the equipment is Δ∈, the calibration process of the equipment is adjusted through the following formula: Δ∈ adjusted = Δ∈·(1 - α) Where, Δ ∈ adjusted is the calibrated device accuracy; Δ ∈ is the accuracy error of the current device; α is the calibration coefficient, representing the reduction ratio of the error after calibration, and α ranges between 0 and 1; This formula shows that after adjusting the equipment precision, the error will be reduced, ensuring the improvement of precision during the production process. Integrated feedback system and real-time reporting The automated control module is combined with the feedback system to generate a real-time defect report and transmit it to the management and operators. The report includes the following information: Detected defect types: crack, broken network, scratch, dent, false soldering, hot spot; Exact location of the defect: specific coordinates; Size and severity of the defect: crack length, false soldering area; Production line adjustment measures: speed adjustment, rework processing; Working status of the repair equipment: power settings and working hours of the laser repair machine.

Images