A fully automatic online nondestructive testing method for lithium battery cells

Abstract

The present invention belongs to the technical field of lithium battery cell detection. The present invention discloses a fully automatic online nondestructive detection method for lithium battery cells, comprising: first, using a X-ray device to scan the battery cells layer by layer, obtain corresponding cross-sectional images and number them in sequence. Then, obtain a reference cross-sectional image corresponding to the battery cell model, perform difference analysis with all cross-sectional images, and mark the cross-sectional images with differences as suspicious. Then, from the suspicious cross-sectional images, select and form several suspicious image groups according to the numbering sequence. Perform abnormality analysis on the suspicious image groups, and mark the image groups with abnormalities as abnormal. Finally, for the abnormal image groups, generate a corresponding three-dimensional abnormality model based on the cross-sectional images therein, and identify the cause of the abnormality based on the model. This method can effectively distinguish between real defects and occasional abnormalities in the scanning process through steps such as position continuity analysis, and can accurately identify persistent real defects to reduce misjudgment.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery core detection, and more specifically, to a fully automatic online nondestructive detection method for lithium battery cores. Background Art

[0002] As a power source with high energy density, long life and environmental protection, lithium batteries are widely used in electric vehicles, consumer electronics, energy storage and other fields. With the rapid development of the lithium battery industry, the quality and safety requirements of lithium batteries are getting higher and higher. The quality of lithium batteries directly affects the performance and safety of electronic products, so it is very important to conduct comprehensive and accurate quality testing of lithium batteries.

[0003] Traditional lithium battery inspection methods mainly include appearance inspection, electrical performance testing, and X-ray inspection. Appearance inspection mainly relies on manual visual inspection, which is inefficient and prone to missed inspections. Although electrical performance testing can detect battery capacity, internal resistance and other parameters, it cannot detect structural defects inside the battery. In recent years, with the development of computer vision and artificial intelligence technology, battery inspection methods based on image processing and deep learning have gradually emerged. These methods can automatically identify various defects inside the battery by performing intelligent analysis of X-ray images.

[0004] However, the existing technology still has some limitations. First, during the X-ray scanning process, there may be insufficient energy in a single scan or temporary material fluctuations, resulting in abnormal features of the battery cell in the image, but in fact the battery cell is not abnormal. This situation is prone to misjudgment and reduces the accuracy of detection. Secondly, the existing technology usually models the entire battery cell by processing all images, and then identifies the cause of the abnormality. There is a problem with this method: when the abnormal features account for a small proportion of the whole, it may lead to inaccurate cause analysis. This is because the abnormal features may be masked by a large amount of normal data, causing the analysis results to deviate from the actual situation. Summary of the invention

[0005] In order to overcome the problem of easy misjudgment in the prior art, the present invention proposes a fully automatic online non-destructive testing method for lithium batteries to solve the above problem.

[0006] The present invention provides the following technical solutions:

[0007] A fully automatic online nondestructive testing method for lithium batteries, comprising:

[0008] Use a X-ray device to scan the battery cell layer by layer to obtain corresponding cross-sectional images, and number the cross-sectional images in sequence according to the scanning order;

[0009] Obtain the model of the battery cell, and obtain the corresponding reference cross-sectional view according to the model of the battery cell;

[0010] Perform difference analysis between the reference cross-sectional image of the battery cell and all cross-sectional images, and mark the cross-sectional images with differences as suspicious according to the difference analysis results;

[0011] From all cross-sectional images marked as suspicious, select cross-sectional images to form a number of suspicious image groups according to the serial number sequence of the cross-sectional images;

[0012] Performing anomaly analysis on the images in the suspicious image group, and marking the image group with abnormalities as abnormal according to the abnormality analysis result;

[0013] For the suspicious image group marked as abnormal, a corresponding three-dimensional abnormal model is generated according to the cross-sectional images therein; and the cause of the abnormality is identified based on the generated three-dimensional abnormal model.

[0014] Preferably, the step of difference analysis comprises:

[0015] Acquire a cross-sectional image and a reference cross-sectional image, and pre-process the cross-sectional image and the reference cross-sectional image;

[0016] aligning the preprocessed cross-sectional image with the reference cross-sectional image;

[0017] For each corresponding pixel in the aligned cross-sectional image and the reference cross-sectional image, a difference image is obtained by subtracting the pixel value in the reference cross-sectional image from the pixel value in the cross-sectional image;

[0018] For each pixel in the difference image, perform the following pixel conversion to obtain a binary image:

[0019] If the pixel value in the difference image is not within the preset difference range, the pixel value is set to P1; if the pixel value in the difference image is within the preset difference range, the pixel value is set to P2; and P1 is not equal to P2;

[0020] The ratio of pixels with a pixel value of P1 in the binary image to the total number of pixels is calculated and recorded as the abnormal ratio;

[0021] If the abnormal proportion is greater than the preset proportion threshold, the difference analysis result is set to exist a difference; otherwise, the difference analysis result is set to not exist a difference.

[0022] Preferably, the step of acquiring the suspicious image group includes:

[0023] S41, establishing a temporary image group, and acquiring all cross-sectional images marked as suspicious;

[0024] S42, selecting the first cross-sectional image as a starting image and adding it to a temporary image group;

[0025] S43, checking the next cross-sectional image of the starting image to determine whether the interval between the next cross-sectional image and the starting image is within a preset range;

[0026] S44, if it is within the range, add it to the temporary image group and set it as a new starting image; otherwise, copy the temporary image group to the candidate group and clear it, use the image as the starting image, and add it to the temporary image group;

[0027] S45, repeating steps S43 and S44 until all acquired cross-sectional images marked as suspicious are processed;

[0028] S46 , for each copied candidate group, check whether the number of images therein is greater than or equal to a preset group threshold, and retain the candidate groups that exceed the threshold as suspicious image groups.

[0029] Preferably, the abnormality analysis step includes:

[0030] Obtaining a binary image obtained when performing a difference analysis on each cross-sectional image of the group in the suspicious image;

[0031] Acquire abnormal position data corresponding to each cross-sectional image according to the binary image;

[0032] Perform positional consistency analysis based on the abnormal position data corresponding to each cross-sectional image;

[0033] If the positional consistency analysis result is consistent, the abnormality analysis result is set to have an abnormality, otherwise, the abnormality analysis result is set to have no abnormality.

[0034] Preferably, the step of acquiring the abnormal position corresponding to each cross-sectional image according to the binary image comprises:

[0035] Obtain a binary image, and use a connected domain analysis algorithm to analyze pixels with a pixel value of P1 in the binary image to obtain all connected domains;

[0036] For each connected domain, obtain its area information;

[0037] Set a minimum area threshold, remove connected domains with an area smaller than the threshold, and record the remaining connected domains as abnormal areas;

[0038] All abnormal regions and corresponding positioning information are acquired to form abnormal position data corresponding to the cross-sectional image, wherein the positioning information includes a center point and a boundary box.

[0039] Preferably, the step of position consistency analysis includes:

[0040] Obtain the number of images N in the suspicious image group, and set a continuity threshold n, where n is an integer less than N;

[0041] Obtaining abnormal position data of each image in the suspicious image group;

[0042] Performing a coherence calculation on each abnormal region in the 1st to Nnth cross-sectional images in the suspicious image group to obtain a coherence number;

[0043] After each coherence calculation, the calculated coherence number is obtained to determine whether it is greater than or equal to the coherence number threshold n;

[0044] If yes, stop the coherence calculation and set the coherence analysis result to coherent;

[0045] If not, determine whether the coherence calculation of all abnormal areas is completed. If not, continue the coherence calculation; if so, set the coherence analysis result to incoherent.

[0046] Preferably, the coherent calculation steps are as follows:

[0047] S51, recording the abnormal region for which the continuity number is to be calculated as the current region, and initializing the continuity number to 1;

[0048] S52, recording the cross-sectional image of the current region as the initial image;

[0049] S53, calculating the distance index between all abnormal regions and the current region in the next cross-sectional image of the initial image according to the positioning coordinates;

[0050] S54, obtaining the minimum value of the calculated distance index, and recording the corresponding abnormal area as the nearest area;

[0051] S55, determining whether the calculated minimum value of the distance index is less than a preset distance index threshold;

[0052] If not, obtain the current coherence number value as the coherence number of the abnormal position, and end this coherence calculation;

[0053] If yes, the continuity number is increased by 1, the nearest area is recorded as the new current area, and it is determined whether the cross-sectional image where the current area is located is the last image; if not, go to S52, if yes, obtain the value of the current continuity number as the continuity number of the abnormal position, and end this continuity calculation.

[0054] Preferably, the step of calculating the distance index comprises:

[0055] Obtain two abnormal areas for which the deviation index needs to be calculated, which are respectively recorded as area A and area B;

[0056] Get the location information corresponding to area A and area B, and calculate the distance index using the following formula:

[0057] ,

[0058] In the formula, represents the distance index, Represents the distance between the center points of area A and area B, Indicates the preset maximum allowed distance, represents the overlap coefficient, and represents the weight factor, and ;

[0059] The calculation formula of the overlap coefficient is as follows:

[0060] ,

[0061] In the formula, Indicates the horizontal overlap, Indicates the vertical overlap, represents the function of taking a smaller value, represents the area of ​​region A, represents the area of ​​region B; the calculation formula for the horizontal overlap is: ; The calculation formula for the longitudinal overlap is: ;

[0062] In the formula, , , and are respectively the horizontal position of the right edge, the horizontal position of the left edge, the vertical position of the bottom edge, and the vertical position of the top edge of the bounding box of region A; , , and are respectively the horizontal position of the right edge, the horizontal position of the left edge, the vertical position of the bottom edge, and the vertical position of the top edge of the bounding box of region B; Indicates the function of taking a larger value.

[0063] The present invention provides a fully automatic online nondestructive testing method for lithium batteries, which has the following beneficial effects:

[0064] By scanning the battery cells layer by layer, obtaining continuous cross-sectional images, and performing difference analysis, construction of suspicious image groups, and abnormality analysis, real defects and occasional abnormalities during the scanning process can be effectively distinguished. By analyzing continuous cross-sectional images and performing positional consistency analysis, persistent real defects can be accurately identified, while excluding false abnormalities caused by insufficient energy in a single scan or temporary material fluctuations, greatly reducing the possibility of misjudgment.

[0065] By specifically analyzing the abnormal area instead of modeling the entire battery cell, this method avoids the problem of abnormal features being ignored because they account for too small a proportion of the overall feature. By calculating the distance index and overlap coefficient, the changes in the abnormal area in continuous cross-sections can be accurately tracked, thereby improving the accuracy of the abnormal cause analysis. In addition, this method only generates the corresponding three-dimensional abnormal model for the image group marked as abnormal, rather than fully modeling the entire battery cell. This method not only greatly improves the modeling efficiency, but also ensures the accuracy of the abnormal area model, thereby improving the accuracy of the abnormal cause analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 The present invention is a schematic diagram of the process of a fully automatic online nondestructive testing method for lithium batteries. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] Example 1

[0069] See also Figure 1 In this embodiment, a fully automatic online nondestructive testing method for lithium batteries includes:

[0070] S1. Scan the battery cell layer by layer using a X-ray device to obtain corresponding cross-sectional images, and number the cross-sectional images in sequence according to the scanning order;

[0071] In this embodiment, the use of a X-ray device to scan the battery cell layer by layer can be carried out in the following steps: First, fix the battery cell to be inspected on the detection platform of the X-ray scanning device. Then, set the scanning parameters such as energy, scanning layer spacing, etc., which can be set accordingly according to the specifications of the battery cell. Next, start the scanning device and scan layer by layer from the bottom or top of the battery cell. Generate a cross-sectional image of each layer. At the same time, the obtained cross-sectional images are numbered. The numbering rule can be: the cross-sectional image of the first layer is numbered 1, the cross-sectional image of the second layer is numbered 2, and so on.

[0072] S2. Obtain the model of the battery cell, and obtain a corresponding reference cross-sectional view according to the model of the battery cell;

[0073] In this embodiment, the model of the battery cell can be obtained by scanning a barcode or the like, and the reference cross-sectional view can be an ideal cross-sectional view designed for the battery cell of this model, or a cross-sectional view of a battery cell of this model that has passed testing, etc.

[0074] S3, using the reference cross-sectional image of the battery cell and all cross-sectional images to perform difference analysis, and marking the cross-sectional images with differences as suspicious according to the difference analysis results;

[0075] The steps of the difference analysis include:

[0076] Acquire a cross-sectional image and a reference cross-sectional image, and pre-process the cross-sectional image and the reference cross-sectional image;

[0077] aligning the preprocessed cross-sectional image with the reference cross-sectional image;

[0078] For each corresponding pixel in the aligned cross-sectional image and the reference cross-sectional image, a difference image is obtained by subtracting the pixel value in the reference cross-sectional image from the pixel value in the cross-sectional image;

[0079] For each pixel in the difference image, perform the following pixel conversion to obtain a binary image:

[0080] If the pixel value in the difference image is not within the preset difference range, the pixel value is set to P1; if the pixel value in the difference image is within the preset difference range, the pixel value is set to P2; and P1 is not equal to P2;

[0081] The ratio of pixels with a pixel value of P1 in the binary image to the total number of pixels is calculated and recorded as the abnormal ratio;

[0082] If the abnormal proportion is greater than the preset proportion threshold, the difference analysis result is set to exist a difference; otherwise, the difference analysis result is set to not exist a difference.

[0083] In this embodiment, the process of difference analysis is as follows. First, a cross-sectional image and a reference cross-sectional image are obtained, and both are preprocessed. The preprocessing step includes operations such as image denoising and contrast enhancement, which are intended to improve the accuracy of subsequent analysis. Specifically, Gaussian filtering can be used for denoising, and a histogram equalization method can be used to enhance image contrast. After the preprocessing is completed, the next step is to align the processed cross-sectional image and the reference cross-sectional image. This step uses image registration techniques, such as feature point matching and affine transformation methods. The purpose of image alignment is to ensure that the same parts of the two images can correspond accurately, laying the foundation for subsequent pixel-level comparison.

[0084] After alignment is completed, each corresponding pixel in the cross-sectional image and the reference cross-sectional image is processed. The specific operation is to subtract the pixel value of the corresponding position in the reference cross-sectional image from the pixel value in the cross-sectional image to obtain a difference image. For example, if the pixel value in the cross-sectional image is 200, and the corresponding pixel value in the reference cross-sectional image is 180, then the value of the pixel in the difference image is 20.

[0085] Next, each pixel in the difference image is converted to generate a binary image. First, a difference range is preset, such as [-20, 20]. Then, for each pixel in the difference image, if its value is not within the preset difference range, it is set to P1 (such as 255); if it is within the preset range, it is set to P2 (such as 0). In this way, the area with large differences can be intuitively marked. After the binarization process is completed, the proportion of pixels with a pixel value of P1 in the binary image to the total number of pixels is calculated, and this proportion is recorded as the abnormal proportion. For example, if there are a total of 10,000 pixels in the binary image, and 1,500 of them have a pixel value of P1, then the abnormal proportion is 15%.

[0086] Finally, a ratio threshold is preset, such as 5%. If the calculated abnormal ratio is greater than this preset threshold, the difference analysis result is judged as having a difference; otherwise, it is judged as not having a difference. This method can effectively identify cross-sectional images that are significantly different from the reference cross-sectional image. These differences may indicate anomalies or defects inside the battery cell.

[0087] S4, selecting cross-sectional images from all cross-sectional images marked as suspicious according to the numbering sequence of the cross-sectional images to form a plurality of suspicious image groups;

[0088] S41, establishing a temporary image group, and acquiring all cross-sectional images marked as suspicious;

[0089] S42, selecting the first cross-sectional image as a starting image and adding it to a temporary image group;

[0090] S43, checking the next cross-sectional image of the starting image to determine whether the interval between the next cross-sectional image and the starting image is within a preset range;

[0091] S44, if it is within the range, add it to the temporary image group and set it as a new starting image; otherwise, copy the temporary image group to the candidate group and clear it, use the image as the starting image, and add it to the temporary image group;

[0092] S45, repeating steps S43 and S44 until all acquired cross-sectional images marked as suspicious are processed;

[0093] S46 , for each copied candidate group, check whether the number of images therein is greater than or equal to a preset group threshold, and retain the candidate groups that exceed the threshold as suspicious image groups.

[0094] In this embodiment, first, a temporary image group is established, and all cross-sectional images marked as suspicious are obtained. These suspicious cross-sectional images are identified in the previous difference analysis step, and they may contain abnormal information inside the battery cell.

[0095] Next, the first of these suspicious cross-sectional images is selected as the starting image and added to the temporary image group. This starting image will serve as the reference point for constructing the suspicious image group. Then, the next cross-sectional image of the starting image is checked to determine whether the interval between it and the starting image is within the preset range. This preset range can be set according to actual conditions, for example, it can be set to 2, indicating that the difference in the numbers of the two images does not exceed 2. If the interval between the next cross-sectional image and the starting image is within the preset range, it is added to the temporary image group and set as the new starting image. Doing so can help identify abnormal areas that appear continuously in the battery cell, which are more likely to represent real defects. If the interval exceeds the preset range, the current temporary image group is copied as a candidate group, and then the temporary image group is cleared and the image currently being checked is used as the new starting image. This step can process discontinuous suspicious areas separately. Repeat the above checking and adding steps until all cross-sectional images marked as suspicious are processed.

[0096] Finally, for each copied candidate group, we check whether the number of images in it exceeds the preset group threshold. This threshold can be set according to actual needs. For example, it can be set to 5, which means that only groups containing 5 or more images are considered valid suspicious image groups. This step can filter out isolated anomalies that may be caused by accidental factors in the scanning process (such as insufficient energy in a single scan). Only groups containing enough consecutive images are considered valid suspicious image groups, which improves the accuracy of identifying real defects.

[0097] S5, performing anomaly analysis on the images in the suspicious image group, and marking the image group with abnormalities as abnormal according to the abnormality analysis result;

[0098] The steps of the abnormality analysis include:

[0099] Obtaining a binary image obtained when performing a difference analysis on each cross-sectional image of the group in the suspicious image;

[0100] Acquire abnormal position data corresponding to each cross-sectional image according to the binary image;

[0101] Perform positional consistency analysis based on the abnormal position data corresponding to each cross-sectional image;

[0102] If the positional consistency analysis result is consistent, the abnormality analysis result is set to have an abnormality, otherwise, the abnormality analysis result is set to have no abnormality.

[0103] The step of acquiring the abnormal position corresponding to each cross-sectional image according to the binary image comprises:

[0104] Obtain a binary image, and use a connected domain analysis algorithm to analyze pixels with a pixel value of P1 in the binary image to obtain all connected domains;

[0105] For each connected domain, obtain its area information;

[0106] Set a minimum area threshold, remove connected domains with an area smaller than the threshold, and record the remaining connected domains as abnormal areas;

[0107] All abnormal regions and corresponding positioning information are acquired to form abnormal position data corresponding to the cross-sectional image, wherein the positioning information includes a center point and a boundary box.

[0108] The steps of position consistency analysis include:

[0109] Obtain the number of images N in the suspicious image group, and set a continuity threshold n, where n is an integer less than N;

[0110] Obtaining abnormal position data of each image in the suspicious image group;

[0111] Performing a coherence calculation on each abnormal region in the 1st to Nnth cross-sectional images in the suspicious image group to obtain a coherence number;

[0112] After each coherence calculation, the calculated coherence number is obtained to determine whether it is greater than or equal to the coherence number threshold n;

[0113] If yes, stop the coherence calculation and set the coherence analysis result to coherent;

[0114] If not, determine whether the coherence calculation of all abnormal areas is completed. If not, continue the coherence calculation; if so, set the coherence analysis result to incoherent.

[0115] The coherent calculation steps are as follows:

[0116] S51, recording the abnormal region for which the continuity number is to be calculated as the current region, and initializing the continuity number to 1;

[0117] S52, recording the cross-sectional image of the current region as the initial image;

[0118] S53, calculating the distance index between all abnormal regions and the current region in the next cross-sectional image of the initial image according to the positioning coordinates;

[0119] S54, obtaining the minimum value of the calculated distance index, and recording the corresponding abnormal area as the nearest area;

[0120] S55, determining whether the calculated minimum value of the distance index is less than a preset distance index threshold;

[0121] If not, obtain the current coherence number value as the coherence number of the abnormal position, and end this coherence calculation;

[0122] If yes, the continuity number is increased by 1, the nearest area is recorded as the new current area, and it is determined whether the cross-sectional image where the current area is located is the last image; if not, go to S52, if yes, obtain the value of the current continuity number as the continuity number of the abnormal position, and end this continuity calculation.

[0123] The step of calculating the distance index comprises:

[0124] Obtain two abnormal areas for which the deviation index needs to be calculated, which are respectively recorded as area A and area B;

[0125] Get the location information corresponding to area A and area B, and calculate the distance index using the following formula:

[0126] ,

[0127] In the formula, represents the distance index, Represents the distance between the center points of area A and area B, Indicates the preset maximum allowed distance, represents the overlap coefficient, and represents the weight factor, and ;

[0128] The calculation formula of the overlap coefficient is as follows:

[0129] ,

[0130] In the formula, Indicates the horizontal overlap, Indicates the vertical overlap, represents the function of taking a smaller value, represents the area of ​​region A, represents the area of ​​region B; the calculation formula for the horizontal overlap is: ; The calculation formula for the longitudinal overlap is: ;

[0131] In the formula, , , and are respectively the horizontal position of the right edge, the horizontal position of the left edge, the vertical position of the bottom edge, and the vertical position of the top edge of the bounding box of region A; , , and are respectively the horizontal position of the right edge, the horizontal position of the left edge, the vertical position of the bottom edge, and the vertical position of the top edge of the bounding box of region B; Indicates the function of taking a larger value.

[0132] In this embodiment, the purpose of performing abnormal analysis on the images in the suspicious image group is to further confirm whether there are real abnormalities in the suspicious image group, and mark the image group confirmed to have abnormalities as an abnormal image group. First, obtain the binary images obtained in the previous difference analysis of each cross-sectional image in the suspicious image group. These binary images contain the information of the abnormal area preliminarily identified. Then, use the connected domain analysis algorithm to analyze the pixels with pixel values ​​of P1 (indicating that abnormalities may exist) in the binary image to obtain all connected domains. This step can help identify potential abnormal areas.

[0133] For each connected domain, its area information is obtained. Then, a minimum area threshold is set to remove connected domains with an area smaller than the threshold. This step can help filter out some erroneous results that may be caused by noise or minor scanning errors. The remaining connected domains are recorded as abnormal areas, and the positioning information of these abnormal areas is obtained, including the center point and the bounding box. This information will be used for subsequent position consistency analysis. It should be noted that the center point and the bounding box can usually be represented by pixel coordinates, and the area can be represented by the number of pixels. For example, the center point of an abnormal area may be represented as (x, y), where x and y are pixel coordinates; the bounding box can be represented by the pixel coordinates of the four corners of the top, bottom, left and right, such as (x1, y1, x2, y2), etc.; the area can be directly represented by the number of pixels contained in the connected domain. This representation method is not only intuitive, but also convenient for subsequent calculation and analysis. By using pixel-level representation, we can accurately locate and quantify abnormal areas, providing an accurate data basis for subsequent consistency analysis.

[0134] Positional coherence analysis is used to determine whether an abnormality persists in multiple consecutive cross-sectional images. The specific steps are as follows: First, the number of images N in the suspicious image group is obtained, and a coherence threshold n (n is less than N) is set. Then, the coherence calculation is performed on each abnormal area in the 1st to Nnth cross-sectional images in the suspicious image group.

[0135] The process of coherence calculation is as follows: the abnormal area for which the coherence number is to be calculated is recorded as the current area, and the initial coherence number is 1. Then, the distance index between the current area and all abnormal areas in the next cross-sectional image is calculated. It should be noted that the calculation of the distance index takes into account the center point distance and the degree of regional overlap, where the center point distance can be calculated by the Euclidean distance, and the distance between the centers of the two abnormal areas is reflected by the ratio between the remaining maximum allowable distances. The smaller the ratio value, the closer the distance. The maximum allowable distance can be set based on experience. The overlap coefficient reflects the degree of overlap between the bounding boxes of the two abnormal areas, which is calculated as the ratio of the overlapping area to the area of ​​the smaller bounding box. The positions of the edges of the bounding boxes used in the calculation can use the corresponding coordinate values, such as the right edge position corresponds to the horizontal coordinate of the right corner of the bounding box; the corresponding weight factors of the two can be adjusted according to the actual effect.

[0136] The calculation of the distance index takes into account the distance between the center points of the abnormal regions and the degree of overlap of the regions, which helps to determine whether the abnormal regions in different cross-sectional images belong to the same continuous defect.

[0137] If the calculated minimum distance index is less than the preset threshold, it is considered that a continuous abnormal area has been found, the coherence number is increased by 1, and the next cross-sectional image is checked. If the minimum distance index is greater than the threshold, or the last image has been checked, the current coherence calculation is terminated.

[0138] After each coherence calculation, determine whether the calculated coherence number is greater than or equal to the coherence number threshold n. If so, stop the calculation and set the coherence analysis result as coherent. If the coherence calculation of all abnormal areas is completed, but no case of coherence number greater than or equal to n is found, the coherence analysis result is set to incoherent.

[0139] The coherence-based analysis method can effectively identify abnormal areas that persist in multiple consecutive cross-sectional images, which are more likely to represent real defects in the battery cell. By setting an appropriate coherence threshold, the sensitivity and reliability of the detection can be balanced, so that smaller continuous defects can be captured while avoiding misjudging some accidental, discontinuous anomalies as defects.

[0140] S6. For the image group marked as abnormal, generate a corresponding three-dimensional abnormal model according to the cross-sectional images therein; and identify the cause of the abnormality based on the generated three-dimensional abnormal model.

[0141] In this embodiment, first, the images in the abnormal image group usually indicate the specific location where the abnormality occurs. Then, only the cross-sectional image of the abnormal location is reconstructed in three dimensions to generate a local three-dimensional abnormal model, rather than completely modeling the entire part. This method can focus more on the abnormal area and improve the accuracy and efficiency of modeling. It should be noted that the three-dimensional reconstruction can adopt existing image reconstruction technology. For example, a voxel-based three-dimensional reconstruction method can be used to stack and interpolate two-dimensional cross-sectional images to form three-dimensional voxel data; a point cloud-based reconstruction method can also be used to extract feature points from the cross-sectional image, construct point cloud data, and then reconstruct the three-dimensional model by surface fitting and other methods. The cause of the abnormality is identified based on the reconstructed three-dimensional abnormal model. Compared with identification on a complete part model, this local model can more accurately reflect the abnormal characteristics, which is conducive to improving the accuracy of identification. The specific identification method can adopt existing machine learning technologies, such as convolutional neural network (CNN), support vector machine (SVM), etc. For example, a three-dimensional model database containing various known abnormal causes can be established in advance, and the abnormal model to be identified can be matched with the samples in the database to determine the cause of the abnormality. Through the above method, the present invention can locate and identify anomalies more accurately, thereby improving the efficiency and accuracy of detection.

[0142] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0143] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0144] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A fully automatic online nondestructive testing method for lithium batteries, characterized in that: include: Use a X-ray device to scan the battery cell layer by layer to obtain corresponding cross-sectional images, and number the cross-sectional images in sequence according to the scanning order; Obtain the model of the battery cell, and obtain the corresponding reference cross-sectional view according to the model of the battery cell; Perform difference analysis between the reference cross-sectional image of the battery cell and all cross-sectional images, and mark the cross-sectional images with differences as suspicious according to the difference analysis results; From all cross-sectional images marked as suspicious, select cross-sectional images to form a number of suspicious image groups according to the serial number sequence of the cross-sectional images; Performing anomaly analysis on the images in the suspicious image group, and marking the image group with abnormalities as abnormal according to the abnormality analysis result; For the suspicious image group marked as abnormal, a corresponding three-dimensional abnormal model is generated according to the cross-sectional images therein; Based on the generated 3D anomaly model, identify the cause of the anomaly; The steps of the difference analysis include: Acquire a cross-sectional image and a reference cross-sectional image, and pre-process the cross-sectional image and the reference cross-sectional image; aligning the preprocessed cross-sectional image with the reference cross-sectional image; For each corresponding pixel in the aligned cross-sectional image and the reference cross-sectional image, a difference image is obtained by subtracting the pixel value in the reference cross-sectional image from the pixel value in the cross-sectional image; For each pixel in the difference image, perform the following pixel conversion to obtain a binary image: If the pixel value in the difference image is not within the preset difference range, the pixel value is set to P1; if the pixel value in the difference image is within the preset difference range, the pixel value is set to P2; and P1 is not equal to P2; The ratio of pixels with a pixel value of P1 in the binary image to the total number of pixels is calculated and recorded as the abnormal ratio; If the abnormal proportion is greater than the preset proportion threshold, the difference analysis result is set to exist; otherwise, the difference analysis result is set to not exist; The steps of the abnormality analysis include: Obtaining a binary image obtained when performing a difference analysis on each cross-sectional image of the group in the suspicious image; Acquire abnormal position data corresponding to each cross-sectional image according to the binary image; Perform positional consistency analysis based on the abnormal position data corresponding to each cross-sectional image; If the positional consistency analysis result is consistent, the abnormality analysis result is set to have an abnormality, otherwise, the abnormality analysis result is set to have no abnormality.

2. A fully automatic online nondestructive testing method for lithium batteries according to claim 1, characterized in that: The step of obtaining the suspicious image group comprises: S41, establishing a temporary image group, and acquiring all cross-sectional images marked as suspicious; S42, selecting the first cross-sectional image as a starting image and adding it to a temporary image group; S43, checking the next cross-sectional image of the starting image to determine whether the interval between the next cross-sectional image and the starting image is within a preset range; S44, if it is within the range, add it to the temporary image group and set it as a new starting image; otherwise, copy the temporary image group to the candidate group and clear it, use the image as the starting image, and add it to the temporary image group; S45, repeating steps S43 and S44 until all acquired cross-sectional images marked as suspicious are processed; S46 , for each copied candidate group, check whether the number of images therein is greater than or equal to a preset group threshold, and retain the candidate groups that exceed the threshold as suspicious image groups.

3. A fully automatic online nondestructive testing method for lithium batteries according to claim 1, characterized in that: The step of acquiring the abnormal position corresponding to each cross-sectional image according to the binary image comprises: Obtain a binary image, and use a connected domain analysis algorithm to analyze pixels with a pixel value of P1 in the binary image to obtain all connected domains; For each connected domain, obtain its area information; Set a minimum area threshold, remove connected domains with an area smaller than the threshold, and record the remaining connected domains as abnormal areas; All abnormal regions and corresponding positioning information are acquired to form abnormal position data corresponding to the cross-sectional image, wherein the positioning information includes a center point and a boundary box.

4. A fully automatic online nondestructive testing method for lithium batteries according to claim 3, characterized in that: The steps of position consistency analysis include: Obtain the number of images N in the suspicious image group, and set a continuity threshold n, where n is an integer less than N; Obtaining abnormal position data of each image in the suspicious image group; Performing a coherence calculation on each abnormal region in the 1st to Nnth cross-sectional images in the suspicious image group to obtain a coherence number; After each coherence calculation, the calculated coherence number is obtained to determine whether it is greater than or equal to the coherence number threshold n; If yes, stop the coherence calculation and set the coherence analysis result to coherent; If not, determine whether the coherence calculation of all abnormal areas is completed. If not, continue the coherence calculation; if so, set the coherence analysis result to incoherent.

5. A fully automatic online nondestructive testing method for lithium batteries according to claim 4, characterized in that: The coherent calculation steps are as follows: S51, record the abnormal area for which the coherence number is to be calculated as the current area, and initialize the coherence number to 1 S52, recording the cross-sectional image of the current region as the initial image; S53, calculating the distance index between all abnormal regions and the current region in the next cross-sectional image of the initial image according to the positioning coordinates; S54, obtaining the minimum value of the calculated distance index, and recording the corresponding abnormal area as the nearest area; S55, determining whether the calculated minimum value of the distance index is less than a preset distance index threshold; If not, obtain the current coherence number value as the coherence number of the abnormal position, and end this coherence calculation; If yes, the continuity number is increased by 1, the nearest area is recorded as the new current area, and it is determined whether the cross-sectional image where the current area is located is the last image; if not, go to S52, if yes, obtain the value of the current continuity number as the continuity number of the abnormal position, and end this continuity calculation.

6. A fully automatic online nondestructive testing method for lithium batteries according to claim 5, characterized in that: The distance index calculation step includes: Obtain two abnormal areas for which the deviation index needs to be calculated, which are respectively recorded as area A and area B; Get the location information corresponding to area A and area B, and calculate the distance index using the following formula: ， In the formula, represents the distance index, Represents the distance between the center points of area A and area B, Indicates the preset maximum allowed distance, represents the overlap coefficient, and represents the weight factor, and ; The calculation formula of the overlap coefficient is as follows: ， In the formula, Indicates the horizontal overlap, Indicates the vertical overlap, represents the function of taking a smaller value, represents the area of ​​region A, represents the area of ​​region B; the calculation formula for the horizontal overlap is: ; The calculation formula for the longitudinal overlap is: ; In the formula, , , and are respectively the horizontal position of the right edge, the horizontal position of the left edge, the vertical position of the bottom edge, and the vertical position of the top edge of the bounding box of region A; , , and are respectively the horizontal position of the right edge, the horizontal position of the left edge, the vertical position of the bottom edge, and the vertical position of the top edge of the bounding box of region B; Indicates the function of taking a larger value.

Images