Battery cell detection method and system based on neural network

Abstract

The present invention relates to the technical field of battery cell detection, and discloses a battery cell detection method and system based on a neural network, including: collecting the overall image of the battery cell, obtaining the position information of different parts of the battery cell according to the overall image of the battery cell; using X-ray rays to collect the overall image of the known defective battery cell and the overall image of the battery cell to be tested, and processing the collected images in combination with the position information of different parts of the battery cell to obtain defect maps of different parts of the battery cell and images of different parts of the battery cell to be tested; using the defect maps of different parts as training sets of different parts, respectively, using the training sets of different parts to train visual neural networks respectively, and inputting the images of different parts of the battery cell to be tested as test sets into the corresponding trained visual neural networks to obtain the defect conditions of the corresponding parts of the battery cell to be tested. The present invention improves the detection effect of unqualified batteries while eliminating missed detection and reducing labor costs.

Description

Technical Field

[0001] The present invention relates to the technical field of battery cell detection, and in particular to a battery cell detection method and system based on a neural network. Background Art

[0002] The tab is a metal conductor that leads the positive and negative poles of the battery from the battery cell. Like the pole piece, it is an indispensable part of the battery cell. At present, the tabs in the battery cell are usually very thin, only 4μm to more than 10μm. Therefore, during the winding process of the battery cell, the tabs on the pole piece are easily folded due to the high-speed movement of the battery cell due to the influence of airflow and rollers; at the same time, during the winding process of the battery cell, the pole piece will be affected by tension, correction and other factors, resulting in large-area wrinkles on the pole piece inside the battery cell. The folding of the tabs and the large-area wrinkles of the pole piece will lead to serious battery cell defects and cause safety hazards. Therefore, the battery cell needs to be tested during the production process of the battery cell to eliminate unqualified conditions. However, the folding of the tabs is the most difficult unqualified condition to detect in the battery cell winding process, and although the probability of large-area wrinkles of the pole piece is small, the location is completely inside the battery cell, which is also difficult to detect.

[0003] Therefore, in order to better detect the condition of the tabs and pole pieces in the battery cells and exclude unqualified battery cells, the prior art uses three methods: color code sensors, CCD cameras, and manual visual inspection to detect the tabs and pole pieces. Among them, the detection using color code sensors and CCD cameras is carried out simultaneously during the battery cell winding process, which has the disadvantages of relatively fixed and single detection positions, inability to cover the entire winding process, and missed detection; and, because the detection is carried out during the winding process, the state of the battery cell at output cannot be guaranteed. In order to reduce the probability of missed detection of tab folding, manual visual inspection is usually added after the battery cell winding process. The inspection personnel manually look through the battery cell tabs to find out whether there is a fold. Although this can further reduce the probability of missed detection, it is inefficient, has high labor costs, and still has the probability of missed detection, and it is impossible to perform manual visual inspection of the pole pieces inside the battery cell.

[0004] In summary, although the existing technology has adopted a variety of detection schemes to reduce the probability of missed detection, it is still impossible to eliminate missed detection. Moreover, each test in various detection schemes, especially manual visual inspection, requires a lot of manpower costs, with high detection costs, low efficiency and poor results. Summary of the invention

[0005] To this end, the technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide a battery cell detection method and system based on a neural network, which can improve the detection effect of unqualified battery cells while eliminating missed detections and reducing labor costs.

[0006] In order to solve the above technical problems, the present invention provides a battery cell detection method based on a neural network, comprising:

[0007] Use X-rays to collect the overall image of the battery cell, and obtain the position information of different parts of the battery cell based on the overall image of the battery cell;

[0008] The parameters of the X-rays are set according to the parameter formulas for collecting data from different parts of the battery cell, and the X-rays set corresponding to the different parts of the battery cell are used to collect the overall image of the known defective battery cell, and the overall image of the known defective battery cell collected is processed in combination with the position information of the different parts of the battery cell to obtain defect maps of the different parts of the battery cell; the defect maps of the different parts of the battery cell are used as training sets of the different parts of the battery cell, and the visual neural network is trained using the training sets of the different parts of the battery cell to obtain a trained visual neural network model of the different parts of the battery cell;

[0009] The parameters of the X-ray are set according to the parameter formula when collecting images of different parts of the battery cell, and the X-rays set corresponding to the different parts of the battery cell are used to collect the overall image of the battery cell to be tested, and the collected overall image of the battery cell to be tested is processed in combination with the position information of the different parts of the battery cell to obtain the images of the different parts of the battery cell to be tested; the images of the different parts of the battery cell to be tested are respectively used as test sets of the different parts of the battery cell to be tested, and the test sets of the different parts of the battery cell to be tested are input into the trained visual neural network model of the different parts of the battery cell, and it is judged whether the different parts of the battery cell to be tested have defects according to the trained visual neural network model of the different parts of the battery cell.

[0010] In one embodiment of the present invention, the position information of different parts of the battery cell is obtained according to the overall image of the battery cell, including: the position information of the positive electrode tab, the position information of the negative electrode tab, and the position information of the battery cell body;

[0011] The method includes setting the parameters of the X-ray rays according to the parameter formula when collecting data from different parts of the battery cell, using the X-ray rays set corresponding to the different parts of the battery cell to collect the overall image of the known defective battery cell, processing the collected overall image of the known defective battery cell in combination with the position information of the different parts of the battery cell to obtain defect images of the different parts of the battery cell; using the defect images of the different parts of the battery cell as training sets of the different parts of the battery cell, respectively, and using the training sets of the different parts of the battery cell to train the visual neural network to obtain a visual neural network model of the different parts of the battery cell after training, including:

[0012] The parameters of the X-ray are set according to the parameter formula when collecting the positive electrode lug part, the X-ray set corresponding to the positive electrode lug part is used to collect the overall image of the known defective battery cell, and the overall image of the known defective battery cell collected is processed in combination with the position information of the positive electrode lug to obtain a defect map of the positive electrode lug part where the positive electrode lug is located in the center of the image; the defect map of the positive electrode lug part is used as a training set of the positive electrode lug part, and the training set of the positive electrode lug part is used to train the visual neural network to obtain a trained visual neural network model of the positive electrode lug part;

[0013] The parameters of the X-ray are set according to the parameter formula when collecting the negative electrode tab part, the X-ray set corresponding to the negative electrode tab part is used to collect the overall image of the known defective battery cell, and the overall image of the known defective battery cell collected is processed in combination with the position information of the negative electrode tab to obtain a defect map of the negative electrode tab part where the negative electrode tab is located in the center of the image; the defect map of the negative electrode tab part is used as a training set of the negative electrode tab part, and the training set of the negative electrode tab part is used to train the visual neural network to obtain a trained visual neural network model of the negative electrode tab part;

[0014] The parameters of the X-ray rays are set according to the parameter formula when collecting the main body of the battery cell, the X-ray rays set corresponding to the main body of the battery cell are used to collect the overall image of the known defective battery cell, and the overall image of the known defective battery cell collected is processed in combination with the position information of the battery cell body to obtain a defect map of the main body of the battery cell where the battery cell body is located in the center of the image; the defect map of the main body of the battery cell is used as a training set of the main body of the battery cell, and the training set of the main body of the battery cell is used to train the visual neural network to obtain a trained visual neural network model of the main body of the battery cell.

[0015] In one embodiment of the present invention, the parameters of the X-rays are set according to the parameter formula when collecting different parts of the battery cell, and the X-rays set corresponding to the different parts of the battery cell are used to collect the overall image of the battery cell to be tested, and the collected overall image of the battery cell to be tested is processed in combination with the position information of the different parts of the battery cell to obtain the images of the different parts of the battery cell to be tested; the images of the different parts of the battery cell to be tested are respectively used as test sets of the different parts of the battery cell to be tested, and the test sets of the different parts of the battery cell to be tested are input into the trained visual neural network model of the different parts of the battery cell, and the visual neural network model of the different parts of the battery cell is used to judge whether there are defects in the different parts of the battery cell to be tested, including:

[0016] The parameters of the X-ray are set according to the parameter formula when collecting the positive electrode tab part, and the X-ray set corresponding to the positive electrode tab part is used to collect the overall image of the battery cell to be tested, and the collected overall image of the battery cell to be tested is processed in combination with the position information of the positive electrode tab to obtain the image of the positive electrode tab part of the battery cell to be tested; the image of the positive electrode tab part of the battery cell to be tested is input into the trained visual neural network model of the positive electrode tab part as a test set, and the positive electrode tab part of the battery cell to be tested is judged whether there is a defect according to the trained visual neural network model of the positive electrode tab part, wherein the number of layers of the positive electrode tab of the battery cell to be tested that is folded into the battery cell body is obtained according to the trained visual neural network model of the positive electrode tab part;

[0017] The parameters of the X-ray are set according to the parameter formula when collecting the negative electrode tab part, and the X-ray set corresponding to the negative electrode tab part is used to collect the overall image of the battery cell to be tested, and the collected overall image of the battery cell to be tested is processed in combination with the position information of the negative electrode tab to obtain the image of the negative electrode tab part of the battery cell to be tested; the image of the negative electrode tab part of the battery cell to be tested is input into the trained visual neural network model of the negative electrode tab part as a test set, and the negative electrode tab part of the battery cell to be tested is judged according to the trained visual neural network model of the negative electrode tab part whether there is a defect, wherein the number of layers of the negative electrode tab of the battery cell to be tested that is folded into the battery cell body is obtained according to the trained visual neural network model of the negative electrode tab part;

[0018] The parameters of the X-ray are set according to the parameter formula when collecting the main body of the battery cell, and the X-ray set corresponding to the main body of the battery cell is used to collect the overall image of the battery cell to be tested, and the collected overall image of the battery cell to be tested is processed in combination with the position information of the battery cell body to obtain the image of the main body of the battery cell to be tested; the image of the main body of the battery cell to be tested is input as a test set into the trained visual neural network model of the main body of the battery cell, and it is judged whether the main body of the battery cell to be tested has defects based on the trained visual neural network model of the main body of the battery cell.

[0019] In one embodiment of the present invention, the parameter formula for collecting the negative electrode tab portion and the parameter formula for collecting the positive electrode tab portion are both:

[0020] x×h×ρ=k+b,

[0021] Wherein, x is the number of layers of the positive electrode tab or the negative electrode tab, h is the thickness of the positive electrode tab or the negative electrode tab, ρ is the density of the positive electrode tab or the negative electrode tab, k represents the voltage parameter of the X-ray ray, k = 30KV-70KV, where the adjustment span of k is 5KV; b represents the current parameter of the X-ray ray, b = 240μA-300μA, where the adjustment span of b is 10μA;

[0022] The parameter formula for collecting the battery body is:

[0023] x1×h1×ρ1+x2×h2×ρ2+x3×h3×ρ3=k′+b′, wherein x1 is the number of layers of the negative pole ear folded into the battery body, x2 is the number of layers of the positive pole ear folded into the battery body, and x3 is the number of layers of the pole piece in the battery cell; h1 is the thickness of the negative pole ear, h2 is the thickness of the positive pole ear, and h3 is the thickness of the pole piece in the battery cell; ρ1 is the density of the negative pole ear, ρ2 is the density of the positive pole ear, and ρ3 is the density of the pole piece in the battery cell, k′ represents the voltage parameter of the X-ray ray, k′=30KV-70KV, and the adjustment span of k′ is 5KV; b′ represents the current parameter of the X-ray ray, b′=240μA-300μA, and the adjustment span of b′ is 10μA.

[0024] In one embodiment of the present invention, when the visual neural network is trained using training sets of different parts of the battery cell, the training model includes an image generator, a calculation sequence, and a supervision method;

[0025] During the training process, different image generators, calculation orders and supervision methods are randomly combined to generate corresponding solutions. The optimal solution is selected from all solutions and the image generator, calculation order and supervision method in the training model are adjusted according to the optimal solution until the optimal solution converges to complete the training.

[0026] In one embodiment of the present invention, the calculation sequence includes forward calculation, reverse calculation and double reverse calculation.

[0027] The forward calculation calculates the current value based on the value at the previous moment, the reverse calculation calculates the current value based on the values ​​at the previous two moments, and the double reverse calculation calculates the current value by combining the value at the previous moment and the values ​​at the previous two moments.

[0028] In one embodiment of the present invention, the supervision method includes full unsupervised, full supervised, interval supervision, node supervision and half-supervised.

[0029] The whole process is unsupervised, which means that during the training process, the training results of any step will not be used to judge the labeled pictures;

[0030] The full supervision means that during the training process, the training results of any step will be used to judge the labeled pictures, and whether to abandon this training mode is determined according to the results of each step, thereby reducing the overall training time and difficulty;

[0031] The interval supervision is to participate in the judgment every other training result during the training process, thereby reducing the single-step training time and increasing the training efficiency;

[0032] The node supervision is to switch the calculation order or the image generator during the training process, and after the transformation, the node supervision determines the labeled picture according to the continuous multiple training results after the switching;

[0033] The half-way supervision is to judge multiple consecutive training results in the halfway point of training. If the training effect continues to decline, a backtracking judgment is performed to find the starting point of the decline, and the training mode is switched to continue training; if the training effect does not decline, the current training is continued, and this supervision is repeated in the subsequent half-way point for judgment.

[0034] In one embodiment of the present invention, when randomly combining different image generators, calculation orders and supervision methods, a sequential traversal method is used to combine all combinations of image generators, calculation orders and supervision methods and encode them, and the corresponding combination is selected through the encoding during training.

[0035] The present invention also provides a battery cell detection system based on a neural network, comprising:

[0036] An image acquisition module is used to acquire the overall image of the battery cell, and to use X-rays set at different parts of the battery cell to acquire the overall image of the known defective battery cell and the overall image of the battery cell to be tested;

[0037] A defect map extraction module is used to combine the position information of different parts of the battery cell to obtain defect maps of different parts of the known defective battery cell and images of different parts of the battery cell to be tested;

[0038] A training module, used to use defect images of different parts of the battery cell as training sets of different parts of the battery cell, and use the training sets of different parts of the battery cell to train the visual neural network to obtain a trained visual neural network model of different parts of the battery cell;

[0039] The detection module is used to use the images of different parts of the battery cell to be tested as test sets of the different parts of the battery cell to be tested, input the test sets of the different parts of the battery cell to be tested into the trained visual neural network model of the different parts of the battery cell, and judge whether there are defects in the different parts of the battery cell to be tested according to the trained visual neural network model of the different parts of the battery cell.

[0040] In one embodiment of the present invention, the image acquisition module includes:

[0041] A position determination module is used to determine the position of the imaging target surface according to the model parameters of the battery cell; when using X-ray rays to collect images, a photoelectric sensor is used to determine whether the battery cell is placed on the imaging target surface. If so, the X-ray ray is used to collect images. If not, the battery cell position is adjusted and re-determined whether the battery cell is placed on the imaging target surface;

[0042] The image storage module determines whether the contour of the pole ear or the large-area shape contour of the battery cell can be clearly identified in the collected image. If so, the current image is stored. If not, the X-ray parameters are adjusted and images of different parts of the battery cell are re-collected.

[0043] The above technical solution of the present invention has the following advantages compared with the prior art:

[0044] The present invention collects and obtains images of different parts of the battery cell to be tested through X-rays, and then uses the visual neural network model trained on different parts of the battery cell to determine whether there are defects in different parts of the battery cell to be tested. It can effectively target two types of bad conditions, namely, tab folding and large-area wrinkles on the pole piece, solve the problem that bad tabs and pole pieces cannot be fully detected during the production of the battery cell and after the battery cell is wound, and completely eliminate bad batteries with tab folding and large-area wrinkles, reducing the cost of manual inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0046] Figure 1 is a flow chart of the method of the present invention.

[0047] Figure 2 It is a schematic diagram of a battery cell with folded tabs and large-area wrinkles.

[0048] Figure 3 It is a schematic diagram of the shooting method in the present invention.

[0049] Figure 4 It is a diagram showing the effect of folding and imaging the copper pole ear in an embodiment of the present invention.

[0050] Figure 5 It is a diagram showing the effect of copper tab folding identification in an embodiment of the present invention.

[0051] Figure 6 It is a diagram showing the effect of folding and imaging the aluminum pole ear in an embodiment of the present invention.

[0052] Figure 7 It is a diagram showing the effect of identifying the folding of the aluminum tab in an embodiment of the present invention.

[0053] Figure 8 This is a diagram showing the effect of large-area wrinkle imaging of the pole piece in an embodiment of the present invention.

[0054] Fig. 9 This is a diagram showing the effect of large-area wrinkle recognition of the pole piece in an embodiment of the present invention.

[0055] Fig.10 It is a system structure diagram of the present invention.

[0056] Fig.11 is a schematic diagram of scaling an image in an embodiment of the present invention.

[0057] Fig.12 It is a schematic diagram of the first scanning method in an embodiment of the present invention.

[0058] Fig.13 Schematic diagram of the second scanning method in an embodiment of the present invention.

[0059] Fig.14 Schematic diagram of the third scanning method in the embodiment of the present invention.

[0060] Fig.15 Schematic diagram of the fourth scanning method in the embodiment of the present invention.

[0061] Fig.16 It is a schematic diagram of the forward calculation sequence in an embodiment of the present invention.

[0062] Fig.17 It is a schematic diagram of the reverse calculation sequence in an embodiment of the present invention.

[0063] Fig.18 Schematic diagram of the double reverse calculation sequence in an embodiment of the present invention.

[0064] Explanation of the markings in the specification: 1. Battery cell body; 2. Aluminum pole ear; 3. Copper pole ear; 4. Folded pole ear; 5. Pole piece in the core body; 6. Wrinkles on the pole piece; 7. X-ray receiving target surface; 8. X-ray coverage surface; 9. Folded copper pole ear; 10. Folding line of the identified copper pole ear; 11. Folded aluminum pole ear; 12. Folding line of the identified aluminum pole ear; 13. Large-area wrinkles; 14. Identified large-area wrinkles. DETAILED DESCRIPTION

[0065] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0066] Embodiment 1

[0067] Reference Figure 1 As shown, the present invention discloses a battery cell detection method based on a neural network, comprising the following steps:

[0068] S1: Use X-rays to capture the overall image of the battery cell, and obtain position information of different parts of the battery cell based on the overall image of the battery cell. In this embodiment, the battery cell is divided into three parts: the positive electrode tab, the negative electrode tab, and the battery cell body. The position information of different parts of the battery cell obtained based on the overall image of the battery cell includes the position information of the positive electrode tab, the position information of the negative electrode tab, and the position information of the battery cell body.

[0069] like Figure 2 As shown, the wound battery cell is divided into three parts: a battery cell body 1, a positive electrode ear and a negative electrode ear. In this embodiment, mark 5 represents the electrode piece constituting the battery cell body 1, the positive electrode ear is an aluminum ear 2, and the negative electrode ear is a copper ear 3. The winding action refers to the process of winding the electrode piece into a battery cell, and does not include other actions such as pre-pressing at the same work station.

[0070] It should also be noted that when collecting the overall image of the battery cell, X-rays or other rays can be used. There is no requirement for the intensity of the rays. It is sufficient that the collected image can identify the positive ear, negative ear, and various parts of the battery cell body. When identifying each part, the edge-grabbing algorithm can be used to extract the contour in the collected image, thereby identifying the ear contour or the large-area shape contour of the battery cell body. First, the image of the entire battery cell is collected for pre-positioning. After identifying the positive ear, negative ear, and battery cell body, the position information of different parts can be obtained, so as to quickly locate different parts in the subsequent collected images and crop different parts according to the position information.

[0071] S2: The parameter formula for collecting the negative electrode tab and the parameter formula for collecting the positive electrode tab are both:

[0072] x×h×ρ=k+b,

[0073] Wherein, x is the number of layers of the positive electrode tab or the negative electrode tab, h is the thickness of the positive electrode tab or the negative electrode tab, ρ is the density of the positive electrode tab or the negative electrode tab, k represents the voltage parameter of the X-ray ray, k = 30KV-70KV, where the adjustment span of k is 5KV; b represents the current parameter of the X-ray ray, b = 240μA-300μA, where the adjustment span of b is 10μA;

[0074] The parameter formula for collecting the battery body is:

[0075] x1×h1×ρ1+x2×h2×ρ2+x3×h3×ρ3=k′+b′, wherein x1 is the number of layers of the negative pole ear folded into the battery body, x2 is the number of layers of the positive pole ear folded into the battery body, and x3 is the number of layers of the pole piece in the battery cell; h1 is the thickness of the negative pole ear, h2 is the thickness of the positive pole ear, and h3 is the thickness of the pole piece in the battery cell; ρ1 is the density of the negative pole ear, ρ2 is the density of the positive pole ear, and ρ3 is the density of the pole piece in the battery cell, k′ represents the voltage parameter of the X-ray ray, k′=30KV-70KV, and the adjustment span of k′ is 5KV; b′ represents the current parameter of the X-ray ray, b′=240μA-300μA, and the adjustment span of b′ is 10μA.

[0076] Since the ranges of the voltage parameters and current parameters of X-ray rays are known, and the number of layers of the positive electrode ear or the negative electrode ear, the thickness of the positive electrode ear or the negative electrode ear, the density of the positive electrode ear or the negative electrode ear, the number of layers of the negative electrode ear folded into the battery cell body, and the number of layers of the positive electrode ear folded into the battery cell body can all be obtained, the voltage parameters and current parameters of the X-ray rays when collecting the negative electrode ear or the negative electrode ear part or the battery cell body part can be obtained according to the parameter formula. For example, when the current parameter setting value of the X-ray ray is 300μA, the voltage parameter of the X-ray ray is preliminarily set, and then adjusted up or down by 5KV each time according to the pre-judgment result of whether the image is valid.

[0077] S3: Place a known defective battery cell at an X-ray imaging location, set X-ray parameters according to the parameter formula for collecting data from different parts of the battery cell, and use X-rays set at corresponding different parts of the battery cell to collect an overall image of the known defective battery cell, and process the collected overall image of the known defective battery cell in combination with the position information of different parts of the battery cell to obtain defect maps of different parts of the battery cell; use the defect maps of different parts of the battery cell as training sets for different parts of the battery cell, respectively, and use the training sets for different parts of the battery cell to train a visual neural network to obtain a trained visual neural network model for different parts of the battery cell.

[0078] like Figure 3 As shown, place the battery cell at the X-ray imaging location. Figure 3 The mark 7 in the figure indicates the X-ray receiving target surface, and the mark 8 indicates the X-ray coverage surface. The X-ray used in this embodiment is X-ray. The X-rays of different intensities are used to collect the overall image of the battery cell multiple times, wherein the X-rays of different intensities are set based on obtaining images of the three parts of the battery cell main body, the positive pole ear part and the negative pole ear part. That is to say, when it is necessary to obtain the image of the battery cell main body, it is necessary to set the parameters of the X-rays according to the parameter formula for collecting the battery cell main body.

[0079] In this embodiment, the three parts of the battery cell body, the positive electrode tab and the negative electrode tab are imaged separately because the densities of the positive electrode tab, the negative electrode tab and the battery cell body are different, and X-rays of different intensities are required for shooting. Therefore, imaging can be any number of times, any angle, and any method, as long as the image can be distinguished by the human eye as an unqualified area. The above-mentioned X-ray shooting with specific parameters can reflect the details of different parts of the battery cell, so it can accurately capture the morphology of different parts of the battery cell.

[0080] The specific reason for using X-rays is that since the tabs are arranged in multiple layers and the battery body is obtained by winding, in order to determine whether the battery has tab folding and large wrinkles, it is necessary to be able to use X-rays to penetrate the battery. X-rays are widely used in medical, industrial, detection and other fields due to their strong penetrating ability.

[0081] X-ray imaging can adjust its intensity according to the needs of different parts of the battery cell, using X-rays to irradiate the battery cell from one side and receive imaging information from the other side. The denser part has fewer X-rays penetrating it, resulting in a darker imaging effect, while the less dense part has more X-rays penetrating it, resulting in a brighter imaging effect. The recognition of X-ray imaging images usually uses a contrast threshold method, but this method requires that the grayscale values ​​at the same position must be uniform and the grayscale values ​​at different positions must be different, and the image acquisition using X-rays cannot guarantee complete uniformity. Therefore, the present invention uses X-rays of different intensities to irradiate different materials at different positions, so that it can produce images that meet the subsequent processing conditions, and then uses the visual neural network model trained on different parts of the battery cell to detect whether different parts of the battery cell are qualified, and finally realizes the effective identification, classification and marking of unqualified batteries.

[0082] S3-1: The known defective battery cell is placed at the X-ray imaging position, and the parameters of the X-ray are set according to the parameter formula for collecting the positive electrode tab portion in reference to step S2, and the X-ray set corresponding to the positive electrode tab portion is used to collect the overall image of the known defective battery cell;

[0083] Setting the parameters of the X-ray according to the parameter formula for collecting the negative electrode tab portion in step S2, and using the X-ray set corresponding to the negative electrode tab portion to collect the overall image of the known defective battery cell;

[0084] The parameters of the X-ray ray are set according to the parameter formula when collecting the main body of the battery cell in step S2, and the overall image of the known defective battery cell is collected using the X-ray ray set corresponding to the main body of the battery cell.

[0085] S3-2: Before processing the collected overall image of the battery cell to obtain defect images of different parts of the battery cell, pre-judgment is performed on whether the image acquisition is effective, so as to adjust the parameters of the X-ray, which includes:

[0086] Use the traditional edge-grabbing algorithm to extract the contours in the collected image, and determine whether the contour of the pole ear or the large-area shape contour of the battery cell can be identified. If so, pre-process the collected image. If not, adjust the X-ray parameters to re-collect images of different parts of the battery cell. Because the pole ears have different sizes, the edge-grabbing algorithm will also recognize contours at different positions. In this embodiment, the preset contour position is compared with the contour position extracted by the edge-grabbing algorithm. If the position comparison is successful, the contour is complete, that is, the recognition is successful, and the next step of pre-processing is continued. In this embodiment, if the contour of the pole ear or the large-area shape of the battery cell cannot be identified, an error message is issued while re-collecting the image. The image is not processed in the pre-discrimination, and it is not determined whether the battery cell is defective. The pre-discrimination is a process of screening images to improve the efficiency of subsequent image processing and determination of whether the battery cell is defective.

[0087] S3-3: The overall image of the known defective battery cell collected is processed in combination with the position information of the positive electrode tab to obtain a defect map of the positive electrode tab portion where the positive electrode tab is located in the center of the image;

[0088] The overall image of the known defective battery cell collected is processed in combination with the position information of the negative electrode tab to obtain a defect map of the negative electrode tab portion where the negative electrode tab is located in the center of the image;

[0089] The overall image of the known defective battery cell collected is processed in combination with the position information of the battery cell body to obtain a defect map of the battery cell body portion where the battery cell body is located in the center of the image.

[0090] Process the collected images. In the pre-identified images, the positions of the tabs and the large surface are different, and each data bit is a 16-bit image. Too many image details are not conducive to accurate and fast judgment, so two pre-processing operations are required: to crop the different collected images so that the positive tabs, negative tabs, and the large area of ​​the battery body (that is, the electrode of the battery body) are located in the center of the image, and to scale the cropped images proportionally to reduce unnecessary details in the image while retaining the area to be identified. Further cropping the image to a size that fits the contour of the image can reduce the subsequent neural network judgment time, and proportional scaling minimizes the amount of data without losing the identifiable details of the image, thereby speeding up the training and judgment of the neural network. An example of the scaling principle is as follows. Fig.11As shown in the figure, the original 0-5 is reduced to 0-2, and the original two are reduced to one. Geometric scaling is to scale the data of each pixel from 0-65535 to 0-g, where the value of g is any number less than or equal to 65535. The scaling ratio is set according to the actual imaging situation. The final imaging effect is as follows: Figure 4 , Figure 6 and Figure 8 shown.

[0091] S3-4: using the defect image of the positive electrode tab as a training set of the positive electrode tab, and using the training set of the positive electrode tab to train a visual neural network to obtain a trained visual neural network model of the positive electrode tab;

[0092] Using the defect image of the negative electrode tab as a training set of the negative electrode tab, and using the training set of the negative electrode tab to train a visual neural network to obtain a trained visual neural network model of the negative electrode tab;

[0093] The defect image of the main part of the battery cell is used as a training set of the main part of the battery cell, and the training set of the main part of the battery cell is used to train the visual neural network to obtain a trained visual neural network model of the main part of the battery cell.

[0094] After the above-mentioned defect maps of different parts of the battery cell are obtained by processing the collected images in combination with the position information of different parts of the battery cell, the defect maps of different parts of the battery cell are imported into the visual neural network by scanning, and the defects in the defect maps of different parts of the battery cell are annotated and identified (for example, the defect map of the pole ear folded into the battery cell body is annotated and identified, and the number of layers of the pole ear folded into the battery cell body is defined), thereby forming a training set of different parts of the battery cell, wherein the defect feature sets corresponding to different battery cell models will be different.

[0095] For each image, the following four scanning methods need to be performed. The scanning order is random and the arrangement of the scanning order is determined according to the results of the supervision feedback during the training process of the attention mechanism algorithm. The specific four scanning methods are as follows: Fig.12 The partially overlapping scanning directions shown, such as Fig.13 The scanning directions shown are as follows: Fig.14 The tangent scanning direction shown and Fig.15 The scan directions are shown to be completely overlapped.

[0096] The visual neural network model used in the present invention is not limited in form, and can be of various architectures and various modes. During training, the weights of the visual neural network are updated according to the defect features.

[0097] The training module includes an image generator, a calculation sequence and a supervision method. During the training process, a training mode combiner is used to randomly combine different image generators, calculation sequences and supervision methods to generate corresponding solutions, and the optimal solution is selected from all solutions and the image generator, calculation sequence and supervision method in the training model are adjusted according to the optimal solution until the optimal solution converges to complete the training, thereby speeding up the training speed and improving the training effect.

[0098] The image generator used in this embodiment is a different attention box generator, which is used to generate different image labels. The image labeling rules are: the first level is the classification and marking of unqualified and qualified images, the second level is to mark the contour line of the unqualified area, and the third level is to mark the folding posture of the unqualified tab. The above levels are in no particular order, and more levels can be extended.

[0099] In this embodiment, the calculation sequence includes forward calculation, reverse calculation and double reverse calculation. Fig.16 As shown, the forward calculation calculates the current value based on the value at the previous moment, that is, f(T) = f(T-1), f(T) is the value at moment T, that is, the value to be calculated, f(T-1) is the current value, that is, the value at moment T-1 before moment T, and T represents the time series. Fig.17 As shown, the reverse calculation calculates the current value based on the values ​​of the last two moments, that is, f(T) = f(T-2), where f(T-2) is the value of the last two moments T-2. Fig.18 As shown, the double reverse calculation combines the value at the previous moment and the values ​​at the previous two moments to calculate the current value. In this embodiment, the double reverse calculation is the average of the value at the previous moment and the values ​​at the previous two moments, that is, T = ((T-1) + (T-2)) / 2, and the value at moment T is the weighted average of the value at moment T-1 and the value at moment T-2.

[0100] In this embodiment, the supervision methods include full unsupervised, full supervised, interval supervision, node supervision and half-way supervision. The full unsupervised means that during the training process, the training results of any step will not participate in the judgment of the labeled pictures; the full supervised means that during the training process, the training results of any step will participate in the judgment of the labeled pictures, and whether to abandon this training mode is determined according to the results of each step, thereby reducing the overall training time and difficulty; the interval supervision means that during the training process, every other training result participates in the judgment, thereby reducing the single-step training time and increasing the training efficiency; the node supervision means that during the training process, the training mode combiner will switch the calculation order or the image generator, and after the transformation, the node supervision judges the labeled pictures according to the continuous multiple training results after the switch; the half-way supervision means that in the half of the training, multiple continuous training results are judged. If the training effect continues to decline, a backtracking judgment is performed to find the starting point of the decline, and the training mode is switched to continue training; if the training effect does not decline, the current training is continued, and this supervision is repeated in the subsequent half to make a judgment.

[0101] In this embodiment, when randomly combining different image generators, calculation orders, and supervision methods, a sequential traversal method is used to combine all combinations of image generators, calculation orders, and supervision methods and encode them, and the corresponding combination is selected by the encoding during training. Encoding the training process can exclude the existing training path in the next training, and reduce the subsequent training time according to the weight information. This method can achieve faster training speed and higher training accuracy when more training content has been trained.

[0102] The training is completed when the objective function converges, and three trained visual neural network models corresponding to the battery cell body, the positive electrode tab and the negative electrode tab are obtained.

[0103] S4: setting X-ray parameters according to parameter formulas for collecting images of different parts of the battery cell, and using X-rays set corresponding to different parts of the battery cell to collect an overall image of the battery cell to be tested, processing the collected overall image of the battery cell to be tested in combination with position information of different parts of the battery cell to obtain images of different parts of the battery cell to be tested; using the images of different parts of the battery cell to be tested as test sets of different parts of the battery cell to be tested, respectively, inputting the test sets of different parts of the battery cell to be tested into a trained visual neural network model of different parts of the battery cell, and judging whether different parts of the battery cell to be tested have defects based on the trained visual neural network model of different parts of the battery cell.

[0104] S4-1: placing the battery cell to be tested at an X-ray imaging location, setting the parameters of the X-ray according to the parameter formula for collecting the positive electrode tab portion in reference to step S2, and using the X-ray set corresponding to the positive electrode tab portion to collect the overall image of the battery cell to be tested;

[0105] The parameters of the X-ray ray are set according to the parameter formula when collecting the negative electrode tab portion in reference step S2, and the overall image of the battery cell to be tested is collected using the X-ray ray set corresponding to the negative electrode tab portion.

[0106] S4-2: Before processing the collected overall image of the battery cell to obtain images of the negative electrode tab and the positive electrode tab, pre-determine whether the image acquisition is effective so as to adjust the parameters of the X-ray ray. For details, refer to step S3-2.

[0107] S4-3: Processing the collected overall image of the battery cell to be tested in combination with the position information of the positive electrode tab to obtain an image of the positive electrode tab portion of the battery cell to be tested;

[0108] The collected overall image of the battery cell to be tested is processed in combination with the position information of the negative electrode tab to obtain an image of the negative electrode tab portion of the battery cell to be tested; the above details refer to step S3-3.

[0109] S4-4: Inputting the image of the positive electrode tab of the battery cell to be tested as a test set into the trained visual neural network model of the positive electrode tab, and judging whether there is a defect in the positive electrode tab of the battery cell to be tested according to the trained visual neural network model of the positive electrode tab, wherein the number of layers of the positive electrode tab of the battery cell to be tested that is folded into the battery cell body is obtained according to the trained visual neural network model of the positive electrode tab;

[0110] The image of the negative electrode tab part of the battery cell to be tested is input as a test set into the trained visual neural network model of the negative electrode tab part, and the trained visual neural network model of the negative electrode tab part is used to judge whether there is a defect in the negative electrode tab part of the battery cell to be tested, wherein the number of layers of the negative electrode tab of the battery cell to be tested that is folded into the battery cell body is obtained according to the trained visual neural network model of the negative electrode tab part.

[0111] S4-5: Place the battery cell to be tested at the X-ray imaging location, set the parameters of the X-ray according to the parameter formula for collecting the battery cell main body in step S2, and use the X-ray set corresponding to the battery cell main body to collect the overall image of the battery cell to be tested.

[0112] S4-6: Before processing the collected overall image of the battery cell to obtain the image of the main body of the battery cell, pre-determine whether the image acquisition is effective so as to adjust the parameters of the X-ray ray. For details, refer to step S3-2.

[0113] S4-7: Processing the collected overall image of the battery cell to be tested in combination with the position information of the battery cell body to obtain an image of the battery cell body of the battery cell to be tested; the above details refer to step S3-3.

[0114] S4-8: Input the image of the main body of the battery cell to be tested as a test set into the trained visual neural network model of the main body of the battery cell, and judge whether the main body of the battery cell to be tested has defects based on the trained visual neural network model of the main body of the battery cell.

[0115] like Fig.10 As shown, the present invention discloses a battery cell detection system based on a neural network, including an image acquisition module, a defect map extraction module, a training module and a detection module.

[0116] The image acquisition module is used to acquire the overall image of the battery cell, and use X-rays set at different parts of the corresponding battery cell to acquire the overall image of the known defective battery cell and the overall image of the battery cell to be tested.

[0117] The defect map extraction module is used to combine the position information of different parts of the battery cell to obtain defect maps of different parts of the known defective battery cell and images of different parts of the battery cell to be tested.

[0118] The training module is used to use the defect images of different parts of the battery cell as training sets of different parts of the battery cell, and obtain the trained visual neural network models of different parts after using the training sets of different parts of the battery cell for training with the visual neural network.

[0119] The detection module is used to use the images of different parts of the battery cell to be tested as test sets of the different parts of the battery cell to be tested, input the test sets of the different parts of the battery cell to be tested into the trained visual neural network model of the different parts of the battery cell, and judge whether there are defects in the different parts of the battery cell to be tested according to the trained visual neural network model of the different parts of the battery cell.

[0120] In this embodiment, the image acquisition module includes a position determination module, a parameter formulation module, an image storage module and an image transmission module.

[0121] The position determination module is used to determine the x-axis and y-axis positions of the three imaging target surfaces of the battery body, the positive pole ear and the negative pole ear according to the model parameters of the battery cell; when using X-ray rays to collect images, a photoelectric sensor is used to determine whether the battery cell is placed on the imaging target surface. If so, the X-ray rays are used to collect images of different parts of the battery cell multiple times. If not, the battery cell position is adjusted and the battery cell is re-determined whether it is placed on the imaging target surface; it can ensure that the position of each imaging is the same and can match the subsequently generated model. The parameter formula module is used to combine the model parameters of the battery cell, the X-ray parameters k and b, the scaling parameter g, and the historical visual neural network model into independent formulas to facilitate other modules to determine the action. The image storage module determines whether the ear contour or the large-area shape contour of the battery cell can be clearly identified in the collected image. If so, the current image is stored. If not, the X-ray parameters are adjusted and images of different parts of the battery cell are re-collected. The image transmission module transmits the stored image to the defect feature extraction module.

[0122] The present invention discloses a storage medium on which a computer program is stored. When the computer program is executed by a processor, the battery cell detection method based on a neural network in the first embodiment is implemented.

[0123] The present invention discloses a device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the neural network-based cell detection method in Example 1 is implemented through a programming language.

[0124] Compared with the prior art, the advantages of the present invention are:

[0125] 1. After the battery cell is wound, before or after pre-pressing, the present invention collects and obtains images of different parts of the battery cell to be tested through X-rays, and then uses the visual neural network model trained for different parts of the battery cell to determine whether there are defects in different parts of the battery cell to be tested. It can effectively target two types of undesirable conditions, namely, tab folding and large-area wrinkles on the pole piece, solve the problem that defective tabs and pole pieces cannot be fully detected during the battery cell production process and after the battery cell is wound, and completely eliminate defective batteries with tab folding and large-area wrinkles that flow downstream after detection by the present invention, thereby reducing the cost of manual inspection.

[0126] 2. The present invention uses X-rays of different intensities to irradiate different materials at different positions, so that images that meet subsequent processing conditions can be produced, and then a convolutional neural network is used to detect different unqualified conditions in different parts of the battery cell, thereby improving the recognition effect of unqualified battery cells.

[0127] 3. The application scenarios of the present invention are not limited to after the battery cell is wound, and it can be implemented in any workstation related to the battery cell. The present invention can also be applied to the folding and wrinkle detection of electrode rolls, diaphragm rolls, tape rolls, foil rolls and other stacked conditions, and has a wide range of applications.

[0128] In order to further illustrate the beneficial effects of the present invention, experiments were conducted using the present invention. Figure 4 This is the copper tab folding image collected in the experiment. Figure 4 The mark 9 in the figure is a recognizable folded copper tab; Figure 5 To use the test results obtained by the present invention, Figure 5 The line 10 marked in the figure is the folding line of the copper electrode ear identified and marked by the method of the present invention. Figure 6 This is the folded image of the aluminum tab collected in the experiment. Figure 6 The mark 11 in the figure is a recognizable folded aluminum tab; Figure 7 To use the test results obtained by the present invention, Figure 7 The line 12 marked in the figure is the folding line of the aluminum pole ear identified and marked by the method of the present invention. Figure 8 This is the image of large-area wrinkles on the pole piece collected in the experiment. Figure 8 The mark 13 in the figure is the large area wrinkles of the pole piece that can be identified; Fig. 9 To use the test results obtained by the present invention, Fig. 9 The area 14 marked in the figure is a large area wrinkle area identified and marked by the method of the present invention. It can be seen from the experiment that the present invention can indeed realize the effective identification of defects of the battery cell tabs and pole pieces.

[0129] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0130] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0131] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0132] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0133] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.

Claims

1. A battery cell detection method based on a neural network, characterized in that: include: Using X-rays to collect an overall image of the battery cell, and obtaining position information of different parts of the battery cell according to the overall image of the battery cell, wherein the position information of different parts of the battery cell includes: position information of the positive electrode tab, position information of the negative electrode tab, and position information of the battery cell body; The parameters of the X-rays are set according to the parameter formulas for collecting data from different parts of the battery cell, and the X-rays set corresponding to the different parts of the battery cell are used to collect the overall image of the known defective battery cell, and the overall image of the known defective battery cell collected is processed in combination with the position information of the different parts of the battery cell to obtain defect maps of the different parts of the battery cell; the defect maps of the different parts of the battery cell are used as training sets of the different parts of the battery cell, and the visual neural network is trained using the training sets of the different parts of the battery cell to obtain a trained visual neural network model of the different parts of the battery cell; The parameters of the X-ray are set according to the parameter formula when collecting images of different parts of the battery cell, and the X-rays set corresponding to the different parts of the battery cell are used to collect the overall image of the battery cell to be tested, and the collected overall image of the battery cell to be tested is processed in combination with the position information of the different parts of the battery cell to obtain images of the different parts of the battery cell to be tested; the images of the different parts of the battery cell to be tested are respectively used as test sets of the different parts of the battery cell to be tested, and the test sets of the different parts of the battery cell to be tested are input into the trained visual neural network model of the different parts of the battery cell, and it is judged whether there are defects in the different parts of the battery cell to be tested according to the trained visual neural network model of the different parts of the battery cell; The parameter formulas for collecting the negative electrode tab and the positive electrode tab are both: x×h×ρ=k+b, Wherein, x is the number of layers of the positive electrode tab or the negative electrode tab, h is the thickness of the positive electrode tab or the negative electrode tab, ρ is the density of the positive electrode tab or the negative electrode tab, k represents the voltage parameter of the X-ray ray, k = 30KV-70KV, where the adjustment span of k is 5KV; b represents the current parameter of the X-ray ray, b = 240μA-300μA, where the adjustment span of b is 10μA; The parameter formula for collecting the battery body is: x1×h1×ρ1+x2×h2×ρ2+x3×h3×ρ3=k′+b′, Among them, x1 is the number of layers of the negative electrode tab folded into the battery body, x2 is the number of layers of the positive electrode tab folded into the battery body, and x3 is the number of layers of the electrode sheet in the battery cell; h1 is the thickness of the negative electrode tab, h2 is the thickness of the positive electrode tab, and h3 is the thickness of the electrode sheet in the battery cell; ρ1 is the density of the negative electrode tab, ρ2 is the density of the positive electrode tab, and ρ3 is the density of the electrode sheet in the battery cell. k′ represents the voltage parameter of the X-ray ray, k′=30KV-70KV, and the adjustment span of k′ is 5KV; b′ represents the current parameter of the X-ray ray, b′=240μA-300μA, and the adjustment span of b′ is 10μA.

2. The battery cell detection method based on neural network according to claim 1, It is characterized by: The method includes setting the parameters of the X-ray rays according to the parameter formula when collecting data from different parts of the battery cell, using the X-ray rays set corresponding to the different parts of the battery cell to collect the overall image of the known defective battery cell, processing the collected overall image of the known defective battery cell in combination with the position information of the different parts of the battery cell to obtain defect images of the different parts of the battery cell; using the defect images of the different parts of the battery cell as training sets of the different parts of the battery cell, respectively, and using the training sets of the different parts of the battery cell to train the visual neural network to obtain a visual neural network model of the different parts of the battery cell after training, including: The parameters of the X-ray are set according to the parameter formula when collecting the positive electrode lug part, the X-ray set corresponding to the positive electrode lug part is used to collect the overall image of the known defective battery cell, and the overall image of the known defective battery cell collected is processed in combination with the position information of the positive electrode lug to obtain a defect map of the positive electrode lug part where the positive electrode lug is located in the center of the image; the defect map of the positive electrode lug part is used as a training set of the positive electrode lug part, and the training set of the positive electrode lug part is used to train the visual neural network to obtain a trained visual neural network model of the positive electrode lug part; The parameters of the X-ray are set according to the parameter formula when collecting the negative electrode tab part, the X-ray set corresponding to the negative electrode tab part is used to collect the overall image of the known defective battery cell, and the overall image of the known defective battery cell collected is processed in combination with the position information of the negative electrode tab to obtain a defect map of the negative electrode tab part where the negative electrode tab is located in the center of the image; the defect map of the negative electrode tab part is used as a training set of the negative electrode tab part, and the training set of the negative electrode tab part is used to train the visual neural network to obtain a trained visual neural network model of the negative electrode tab part; The parameters of the X-ray rays are set according to the parameter formula when collecting the main body of the battery cell, the X-ray rays set corresponding to the main body of the battery cell are used to collect the overall image of the known defective battery cell, and the overall image of the known defective battery cell collected is processed in combination with the position information of the battery cell body to obtain a defect map of the main body of the battery cell where the battery cell body is located in the center of the image; the defect map of the main body of the battery cell is used as a training set of the main body of the battery cell, and the training set of the main body of the battery cell is used to train the visual neural network to obtain a trained visual neural network model of the main body of the battery cell.

3. The battery cell detection method based on neural network according to claim 2, characterized in that: The method comprises setting the parameters of the X-ray rays according to the parameter formula when collecting information of different parts of the battery cell, collecting the overall image of the battery cell to be tested by using the X-ray rays set corresponding to the different parts of the battery cell, processing the collected overall image of the battery cell to be tested in combination with the position information of the different parts of the battery cell to obtain images of the different parts of the battery cell to be tested; using the images of the different parts of the battery cell to be tested as test sets of the different parts of the battery cell to be tested, inputting the test sets of the different parts of the battery cell to be tested into the trained visual neural network model of the different parts of the battery cell, and judging whether the different parts of the battery cell to be tested have defects according to the trained visual neural network model of the different parts of the battery cell, including: The parameters of the X-ray are set according to the parameter formula when collecting the positive electrode tab part, and the X-ray set corresponding to the positive electrode tab part is used to collect the overall image of the battery cell to be tested, and the collected overall image of the battery cell to be tested is processed in combination with the position information of the positive electrode tab to obtain the image of the positive electrode tab part of the battery cell to be tested; the image of the positive electrode tab part of the battery cell to be tested is input into the trained visual neural network model of the positive electrode tab part as a test set, and the positive electrode tab part of the battery cell to be tested is judged whether there is a defect according to the trained visual neural network model of the positive electrode tab part, wherein the number of layers of the positive electrode tab of the battery cell to be tested that is folded into the battery cell body is obtained according to the trained visual neural network model of the positive electrode tab part; The parameters of the X-ray are set according to the parameter formula when collecting the negative electrode tab part, and the X-ray set corresponding to the negative electrode tab part is used to collect the overall image of the battery cell to be tested, and the collected overall image of the battery cell to be tested is processed in combination with the position information of the negative electrode tab to obtain the image of the negative electrode tab part of the battery cell to be tested; the image of the negative electrode tab part of the battery cell to be tested is input into the trained visual neural network model of the negative electrode tab part as a test set, and the negative electrode tab part of the battery cell to be tested is judged according to the trained visual neural network model of the negative electrode tab part whether there is a defect, wherein the number of layers of the negative electrode tab of the battery cell to be tested that is folded into the battery cell body is obtained according to the trained visual neural network model of the negative electrode tab part; The parameters of the X-ray are set according to the parameter formula when collecting the main body of the battery cell, and the X-ray set corresponding to the main body of the battery cell is used to collect the overall image of the battery cell to be tested, and the collected overall image of the battery cell to be tested is processed in combination with the position information of the battery cell body to obtain the image of the main body of the battery cell to be tested; the image of the main body of the battery cell to be tested is input as a test set into the trained visual neural network model of the main body of the battery cell, and it is judged whether the main body of the battery cell to be tested has defects based on the trained visual neural network model of the main body of the battery cell.

4. The battery cell detection method based on a neural network according to any one of claims 1 to 3, characterized in that: When the visual neural network is trained using training sets of different parts of the battery cell, the training model includes an image generator, a calculation sequence and a supervision method; During the training process, different image generators, calculation orders and supervision methods are randomly combined to generate corresponding solutions. The optimal solution is selected from all solutions and the image generator, calculation order and supervision method in the training model are adjusted according to the optimal solution until the optimal solution converges to complete the training.

5. The battery cell detection method based on neural network according to claim 4, characterized in that: The calculation sequence includes forward calculation, reverse calculation and double reverse calculation. The forward calculation calculates the current value based on the value at the previous moment, the reverse calculation calculates the current value based on the values ​​at the previous two moments, and the double reverse calculation calculates the current value by combining the value at the previous moment and the values ​​at the previous two moments.

6. The battery cell detection method based on neural network according to claim 5, characterized in that: The supervision methods include full unsupervised, full supervised, interval supervision, node supervision and half-supervised. The whole process is unsupervised, which means that during the training process, the training results of any step will not be used to judge the labeled pictures; The full supervision means that during the training process, the training results of any step will be used to judge the labeled pictures, and whether to abandon this training mode is determined according to the results of each step, thereby reducing the overall training time and difficulty; The interval supervision is to participate in the judgment every other training result during the training process, thereby reducing the single-step training time and increasing the training efficiency; The node supervision is to switch the calculation order or the image generator during the training process, and after the transformation, the node supervision determines the labeled picture according to the continuous multiple training results after the switching; The half-way supervision is to judge multiple consecutive training results in the halfway of training. If the training effect continues to decline, backtrack to find the starting point of the decline, and switch the training mode to continue training; If the training effect has not decreased, the current training will continue, and this supervision will be repeated in the subsequent half to make a judgment.

7. The battery cell detection method based on neural network according to claim 6, characterized in that: When randomly combining different image generators, calculation orders and supervision methods, a sequential traversal method is used to combine all combinations of image generators, calculation orders and supervision methods and encode them. During training, the corresponding combination is selected through the encoding.

8. A battery cell detection system based on a neural network, characterized in that: include: An image acquisition module is used to acquire the overall image of the battery cell, and to use X-rays set at different parts of the battery cell to acquire the overall image of the known defective battery cell and the overall image of the battery cell to be tested; The parameter formulas for collecting the negative electrode tab and the positive electrode tab are both: x×h×ρ=k+b, Wherein, x is the number of layers of the positive electrode tab or the negative electrode tab, h is the thickness of the positive electrode tab or the negative electrode tab, ρ is the density of the positive electrode tab or the negative electrode tab, k represents the voltage parameter of the X-ray ray, k = 30KV-70KV, where the adjustment span of k is 5KV; b represents the current parameter of the X-ray ray, b = 240μA-300μA, where the adjustment span of b is 10μA; The parameter formula for collecting the battery body is: x1×h1×ρ1+x2×h2×ρ2+x3×h3×ρ3=k′+b′, Among them, x1 is the number of layers of the negative pole ear folded into the battery body, x2 is the number of layers of the positive pole ear folded into the battery body, and x3 is the number of layers of the pole piece in the battery cell; h1 is the thickness of the negative pole ear, h2 is the thickness of the positive pole ear, and h3 is the thickness of the pole piece in the battery cell; ρ1 is the density of the negative pole ear, ρ2 is the density of the positive pole ear, and ρ3 is the density of the pole piece in the battery cell. k' represents the voltage parameter of the X-ray ray, k'=30KV-70KV, where the adjustment span of k' is 5KV; b' represents the current parameter of the X-ray ray, b'=240μA-300μA, where the adjustment span of b' is 10μA; The defect map extraction module is used to combine the position information of different parts of the battery cell to obtain defect maps of different parts of the known defective battery cell and images of different parts of the battery cell to be tested; the position information of different parts of the battery cell obtained according to the overall image of the battery cell includes the position information of the positive electrode tab, the position information of the negative electrode tab, and the position information of the battery cell body; A training module, used to use defect images of different parts of the battery cell as training sets of different parts of the battery cell, and use the training sets of different parts of the battery cell to train the visual neural network to obtain a trained visual neural network model of different parts of the battery cell; The detection module is used to use the images of different parts of the battery cell to be tested as test sets of the different parts of the battery cell to be tested, input the test sets of the different parts of the battery cell to be tested into the trained visual neural network model of the different parts of the battery cell, and judge whether there are defects in the different parts of the battery cell to be tested according to the trained visual neural network model of the different parts of the battery cell.

9. The battery cell detection system based on neural network according to claim 8, characterized in that: The image acquisition module comprises: A position determination module is used to determine the position of the imaging target surface according to the model parameters of the battery cell; when using X-ray rays to collect images, a photoelectric sensor is used to determine whether the battery cell is placed on the imaging target surface. If so, the X-ray ray is used to collect images. If not, the battery cell position is adjusted and re-determined whether the battery cell is placed on the imaging target surface; The image storage module determines whether the contour of the pole ear or the large-area shape contour of the battery cell can be clearly identified in the collected image. If so, the current image is stored. If not, the X-ray parameters are adjusted and images of different parts of the battery cell are re-collected.

Images