A defect detection system based on lithium battery CT images

Abstract

The present invention relates to the technical field of lithium battery defect detection and discloses a defect detection system based on lithium battery CT images, comprising the following steps: step 1, manually marking the surface and serial number of the slice; step 2, collecting overhang images and training a deep learning algorithm model for positive and negative pole detection; step 3, obtaining required CT slice images; step 4, obtaining JR local images; step 5, obtaining all positive and negative poles on each JR local image; step 6, matching the positive and negative poles detected on each JR and calculating the distance between the matched positive and negative poles; step 7, judging whether an overhang defect is included based on the result in step 6. The present invention performs fully automatic identification during online operation, reducing labor costs, and accurately detects the positions of the positive and negative poles, and the calculated overhang value is more accurate than that obtained manually.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery defect detection, and more particularly, to a defect detection system based on lithium battery CT images. Background Art

[0002] Overhang affects the electrochemical performance of batteries. Theoretically, a battery with perfect overlap of the positive and negative electrodes and no overhang would achieve optimal performance. However, due to engineering precision, this cannot be achieved, and the battery can easily suffer from lithium deposition. Therefore, battery overhang must be managed and controlled to ensure it is within appropriate specifications to maintain battery performance.

[0003] However, currently, when detecting overhang defects, most battery manufacturers use CT machines to scan and obtain CT data, and then use existing software on the market, such as VG Studio, to manually analyze the CT data. During manual analysis, it is necessary to slice the right / top / front surfaces of each set of CT data, select the slices at the corresponding positions on the corresponding surfaces, and then manually pull the wires between adjacent anodes and cathodes to obtain all overhang values. The smallest overhang is then judged to see if it meets the requirements. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a defect detection system based on lithium battery CT images.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A defect detection system based on lithium battery CT images includes the following steps:

[0007] Step 1: Manually mark the slice surface and serial number;

[0008] Step 2: Collect overhang images and train a deep learning algorithm model for yin and yang pole detection;

[0009] Step 3: Obtain the required CT slice image;

[0010] Step 4: Get the JR partial image;

[0011] Step 5: Obtain all positive and negative poles on each JR local image;

[0012] Step 6: Match the positive and negative poles detected on each JR and calculate the distance between the matched positive and negative poles;

[0013] Step 7: Determine whether an overhang defect is present based on the result in step 6.

[0014] As a further solution of the present invention: Step 1 and Step 2 are both performed in an offline state; Step 3, Step 4, Step 5, Step 6, and Step 7 are all performed in an online state.

[0015] As a further solution of the present invention: in step 1, manually marking the surface and serial number of the slice means: configuring the surface and serial number of the slice to be calculated for each lithium battery model, and recording them as Side-Index and Cut-Index respectively.

[0016] As a further solution of the present invention: in step 2, collecting overhang images and training a deep learning algorithm model for yin and yang pole detection means collecting multiple overhang CT slice images, annotating them offline, and training a deep learning model to obtain the following two models:

[0017] Model 1: A model that can detect the location of each JR, denoted as JR-BBox-Model;

[0018] Model 2: A model that can predict all positive pole positions and negative pole positions on a single JR image, denoted as PN-Pt-Model.

[0019] As a further solution of the present invention: in step three, obtaining the required CT slice image means: inputting the CT data obtained by scanning, directly parsing the original data, and obtaining the slice image data of the corresponding position according to the surface index Side-Index and the slice index Cut-Index of the corresponding surface.

[0020] As a further solution of the present invention: in step 4, obtaining the JR local image refers to: using the JR-BBox-Model model to predict the slice image, obtaining the positions of N-JR JRs, and performing image subtraction at each position to obtain N-JR JR local images.

[0021] As a further solution of the present invention: in step five, obtaining all positive and negative poles on each JR local image means: using the PN-Pt-Model model to predict the JR local image to obtain all positive pole set P-Pt and negative pole set N-Pt.

[0022] As a further solution of the present invention: in step 6, the positive and negative poles detected on each JR are matched, and the distance between the matched positive and negative poles is calculated to be:

[0023] Step 1: Sort the set P-Pt from small to large to obtain P-Pt-Sort;

[0024] Step 2: Traverse P-Pt-Sort and perform the following operations for each point p:

[0025] S1. Find the set of all points in N-Pt that meet the specified horizontal pixel range N-Pt-InRange;

[0026] S2. If N-Pt-InRange is empty, it is considered an abnormal detection point and no judgment is made. Otherwise, the point n with the smallest vertical (y-axis) coordinate in N-Pt-InRange is selected to obtain the point pair (p, n);

[0027] S3. Calculate the vertical distance d = py - ny between points p and n, and finally obtain the (p, n, d) set of all points p, denoted as PN-Match.

[0028] As a further solution of the present invention: in step seven, the judgment of whether an overhang defect is included means: if d<Thr exists in the PN-Match set, it is judged that an overhang defect exists and is judged as NG, otherwise it is OK.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. Fully automatic recognition during online operation reduces labor costs.

[0031] 2. The position detection of the positive and negative poles is accurate, and the calculated overhang value is more accurate than manual detection.

[0032] 3. When running online, there is no need to use paid software such as vgstudio, saving costs.

[0033] 4. Only slices at specific locations are required instead of all slices, which improves the running rhythm.

[0034] 5. The software calculation speed is better than manual speed, which improves the operation rhythm. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a system flow chart of a defect detection system based on lithium battery CT images of the present invention;

[0036] Figure 2 This is an example of a CT image of JR in the present invention. DETAILED DESCRIPTION

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making any creative efforts shall fall within the scope of protection of the present invention.

[0038] Reference Figure 1 - Figure 2 The following examples were obtained:

[0039] Example 1

[0040] Notes on the symbols in the text:

[0041] Overhang: The length of the negative electrode exceeds the length of the positive and negative electrode plates.

[0042] JR: Bare cell. Generally, a battery has 2-4 JRs. In the following text, it is assumed that the number of JRs is N-JR.

[0043] vgi file: A CT data format file obtained by scanning with a CT machine. There is currently no ready-made tool on the market, and it generally requires paid software from VG company to open it.

[0044] Image coordinate system: The coordinates mentioned later in this article are based on the image coordinate system. The upper left corner of the image is the origin of the coordinate system, the right is the positive x direction, and the downward is the positive y direction.

[0045] Box position: Use BBox to represent the bounding box, represented by x (x coordinate of the upper left corner), y (y coordinate of the upper left corner), w (width), and h (height).

[0046] Point position: Use Pt to represent a point, x (x coordinate), y (y coordinate).

[0047] Set: Use the symbol Set to represent it.

[0048] Positive electrode: Indicated by P.

[0049] Negative electrode: Indicated by N.

[0050] Model: Indicated by Model.

[0051] JR location: Indicated using JR-BBox.

[0052] JR position detection model: represented by JR-BBox-Model.

[0053] Positive pole position: Indicated by P-Pt.

[0054] Negative pole position: Indicated by N-Pt.

[0055] Positive and negative pole detection model: represented by PN-Pt-Model.

[0056] Ct data plane sequence number: expressed using Side-Index.

[0057] CT data slice number: Use Cut-Index.

[0058] Overhang distance threshold: Thr.

[0059] A defect detection system based on lithium battery CT images includes the following steps:

[0060] Step 1: Manually mark the slice surface and serial number;

[0061] Step 2: Collect overhang images and train a deep learning algorithm model for yin and yang pole detection;

[0062] Step 3: Obtain the required CT slice image;

[0063] Step 4: Get the JR partial image;

[0064] Step 5: Obtain all positive and negative poles on each JR local image;

[0065] Step 6: Match the positive and negative poles detected on each JR and calculate the distance between the matched positive and negative poles;

[0066] Step 7: Determine whether an overhang defect is present based on the result in step 6.

[0067] Steps 1 and 2 are performed offline; steps 3, 4, 5, 6, and 7 are performed online.

[0068] In step 1, manually marking the slice surface and serial number means: configuring the surface and serial number of the slice to be calculated for each lithium battery model, and recording them as Side-Index and Cut-Index respectively.

[0069] In step 2, collecting overhang images and training a deep learning algorithm model for yin and yang pole detection involves collecting multiple overhang CT slice images, labeling them offline, and training the deep learning model to obtain the following two models:

[0070] Model 1: A model that can detect the location of each JR, denoted as JR-BBox-Model;

[0071] Model 2: A model that can predict all positive pole positions and negative pole positions on a single JR image, denoted as PN-Pt-Model.

[0072] In step 3, obtaining the required CT slice image means: inputting the scanned CT data, directly parsing the original data, and obtaining the slice image data of the corresponding position according to the surface index Side-Index and the slice index Cut-Index of the corresponding surface.

[0073] In step 4, obtaining the JR local image means: using the JR-BBox-Model model to predict the slice image, obtaining the locations of N-JR JRs, and performing image deduction at each location to obtain N-JR JR local images.

[0074] In step 5, obtaining all positive and negative poles on each JR local image means: using the PN-Pt-Model model to predict the JR local image to obtain all positive pole sets P-Pt and negative pole sets N-Pt.

[0075] In step 6, the positive and negative poles detected on each JR are matched, and the distance between the matched positive and negative poles is calculated:

[0076] Step 1: Sort the set P-Pt from small to large to obtain P-Pt-Sort;

[0077] Step 2: Traverse P-Pt-Sort and perform the following operations for each point p:

[0078] S1. Find the set of all points in N-Pt that meet the specified horizontal pixel range N-Pt-InRange;

[0079] S2. If N-Pt-InRange is empty, it is considered an abnormal detection point and no judgment is made. Otherwise, the point n with the smallest vertical (y-axis) coordinate in N-Pt-InRange is selected to obtain the point pair (p, n);

[0080] S3. Calculate the vertical distance d = py - ny between points p and n, and finally obtain the (p, n, d) set of all points p, denoted as PN-Match.

[0081] In step 7, determining whether an overhang defect is included means: if d < Thr exists in the PN-Match set, it is determined that an overhang defect exists and is judged as NG; otherwise, it is OK.

[0082] Example 2

[0083] A defect detection system based on lithium battery CT images consists of offline and online steps:

[0084] Offline steps:

[0085] Step 1: Manually mark the slice surface and serial number;

[0086] Because the slice positions to be calculated for each lithium battery model may be different, for example, model 1 is the 100th slice on the front surface, and model 2 is the 400th slice on the front surface, you need to configure the surface and sequence number of the slice to be calculated for each model in advance, which are recorded as Side-Index and Cut-Index.

[0087] Step 2: Collect overhang images and train a deep learning algorithm model for yin and yang pole detection;

[0088] We collected a moderate amount of overhang CT slice images, annotated them offline, and trained the deep learning model to obtain the following two models:

[0089] 1. A model that can detect the location of each JR, denoted as JR-BBox-Model;

[0090] 2. A model that can predict all positive pole positions and negative pole positions on a single JR image is denoted as PN-Pt-Model.

[0091] Among them, when the number of CT slice images with appropriate overhang needs to be determined, 1000 images are generally selected.

[0092] Online steps:

[0093] Step 1: Obtain the required CT slice image;

[0094] Input the CT data obtained by scanning, directly parse the original data, and obtain the slice image data of the corresponding position according to the surface index Side-Index and the slice index Cut-Index of the corresponding surface.

[0095] Step 2: Get the JR partial image;

[0096] The JR-BBox-Model model is used to predict the slice image to obtain the locations of N-JR JRs, and the image is subtracted at each location to obtain N-JR JR local images.

[0097] Step 3: Get all the positive and negative poles on each JR local image;

[0098] The PN-Pt-Model model is used to predict the JR local image to obtain the set of all positive poles P-Pt and the set of negative poles N-Pt.

[0099] Step 4: Match the positive and negative poles detected on each JR, and calculate the distance between the matched positive and negative poles;

[0100] Use the following method to match the positive and negative poles:

[0101] 1. Sort the set P - Pt in ascending order to obtain P - Pt - Sort;

[0102] 2. Traverse P - Pt - Sort, and perform the following operations for each point p:

[0103] A. Find all the point sets N - Pt - InRange in N - Pt that satisfy the specified horizontal (x - axis) pixel range (for example, when the horizontal pixel width of the anode and cathode is G, it can be set to 2 * G);

[0104] B. If N - Pt - InRange is empty, it is considered an abnormal detection point and no judgment is made. Otherwise, select the point n with the smallest vertical (y - axis) coordinate in N - Pt - InRange to obtain the point pair (p, n);

[0105] C. Calculate the vertical distance d = py - ny (which can be negative, that is, the negative pole exceeds the positive pole) between points p and n, and finally obtain the set of all points p's (p, n, d), denoted as PN - Match;

[0106] Step 5: Make a judgment on whether there is an overhang defect;

[0107] If there exists d < Thr in the PN - Match set, it is judged that there is an overhang defect and judged as NG, otherwise OK.

[0108] Example 3

[0109] A defect detection system based on CT images of lithium batteries consists of an offline step and an online step:

[0110] Offline step: [[ID=3,5]]

[0111] Step 1: Manually label the plane and serial number of the slice;

[0112] Since the slice positions that need to be calculated for each lithium battery model may be different, for example, the 100th slice on the front side for model one and the 400th slice on the front side for model two, it is necessary to configure in advance the plane and serial number of the slices that need to be calculated for each model, denoted as Side - Index and Cut - Index.

[0113] Step 2: Collect overhang pictures and train a deep - learning algorithm model for detecting positive and negative poles;

[0114] We collected a moderate amount of overhang CT slice images, annotated them offline, and trained the deep learning model to obtain the following two models:

[0115] 1. A model that can detect the location of each JR, denoted as JR-BBox-Model;

[0116] 2. A model that can predict all positive pole positions and negative pole positions on a single JR image is denoted as PN-Pt-Model.

[0117] Among them, when the number of CT slice images with appropriate overhang needs to be determined, 1000 images are generally selected.

[0118] Online steps:

[0119] Step 1: Obtain the required CT slice image;

[0120] Input the CT data obtained by scanning, directly parse the original data, and obtain the slice image data of the corresponding position according to the surface index Side-Index and the slice index Cut-Index of the corresponding surface.

[0121] Step 2: Get Figure 2 Partial image of JR shown; Figure 2 is an example of a CT image of JR, where the points in the upper row are positive poles and the points in the lower row are negative poles;

[0122] The JR-BBox-Model model is used to predict the slice image to obtain the locations of N-JR JRs, and the image is subtracted at each location to obtain N-JR JR local images.

[0123] Step 3: Get all the positive and negative poles on each JR local image;

[0124] The PN-Pt-Model model is used to predict the JR local image to obtain the set of all positive poles P-Pt and the set of negative poles N-Pt.

[0125] Step 4: Match the positive and negative poles detected on each JR and calculate the distance between the matched positive and negative poles;

[0126] Use the following method to match the positive and negative poles:

[0127] 1. Sort the set P-Pt from small to large to obtain P-Pt-Sort;

[0128] 2. Traverse P-Pt-Sort and perform the following operations for each point p:

[0129] A. Find all the point sets N-Pt-InRange in N-Pt that satisfy the specified horizontal (x-axis) pixel range (for example, when the horizontal pixel widths of the anode and cathode are G, it can be set to 2*G).

[0130] B. If N-Pt-InRange is empty, it is considered an abnormal detection point and no judgment is made. Otherwise, select the point n with the smallest vertical (y-axis) coordinate in N-Pt-InRange to obtain the point pair (p, n).

[0131] C. Calculate the vertical distance d = py - ny between points p and n (it can be negative, that is, the negative electrode exceeds the positive electrode), and finally obtain the set of (p, n, d) for all points p, denoted as PN-Match; py is the coordinate position of point p on the y-axis, and ny is the coordinate position of point n on the y-axis.

[0132] Step Five: Make a judgment on whether there is an overhang defect.

[0133] If there exists d < Thr in the PN-Match set, it is judged that there is an overhang defect and judged as NG, otherwise OK.

[0134] In the description of the present invention, it should be understood that the terms "upper", "lower", "left", "right", etc. indicating the orientation or position relationship are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, as well as a specific orientation structure and operation. Therefore, it cannot be understood as a limitation to the present invention. In addition, "first" and "second" are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise specified, the meaning of "plurality" is two or more.

[0135] In the description of the present invention, it should be noted that unless otherwise clearly specified and limited, the terms "installation", "connection", "connection", etc. should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0136] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that improvements and modifications that do not depart from the principles of the present invention by ordinary skill in the art should also be considered within the scope of protection of the present invention.

Claims

1. A defect detection system based on lithium battery CT images, characterized in that: The following steps are involved: Step 1: Configure the surface and serial number of the slice to be calculated for each lithium battery model, and record them as Side-Index and Cut-Index respectively; Step 2: Collect multiple overhang CT slice images, perform offline annotation and deep learning model training, and obtain the following two models: Model 1: A model that can detect the location of each JR, denoted as JR-BBox-Model; Model 2: A model that can predict all positive pole positions and negative pole positions on a single JR image, denoted as PN-Pt-Model; Step 3: Input the scanned CT data, directly parse the original data, and obtain the slice image data of the corresponding position according to the surface index Side-Index and the slice index Cut-Index of the corresponding surface; Step 4: Use the JR-BBox-Model model to predict the slice image, obtain the locations of N-JR JRs, and perform image deduction at each location to obtain N-JR JR local images; Step 5: Use the PN-Pt-Model model to predict the JR local image and obtain the set of all positive poles P-Pt and the set of negative poles N-Pt; Step 6: Step 1: Sort the set P-Pt from small to large to obtain P-Pt-Sort; Step 2: Traverse P-Pt-Sort and perform the following operations for each point p: S1. Find the set of all points in N-Pt that meet the specified horizontal pixel range N-Pt-InRange; S2. If N-Pt-InRange is empty, it is considered an abnormal detection point and no judgment is made. Otherwise, the point n with the smallest vertical (y-axis) coordinate in N-Pt-InRange is selected to obtain the point pair (p, n); S3. Calculate the vertical distance d = py - ny between points p and n, and finally obtain the (p, n, d) set of all points p, denoted as PN-Match; Step 7: If d < Thr exists in the PN-Match set, it is determined that an overhang defect exists and is judged as NG; otherwise, it is OK.

2. The defect detection system based on lithium battery CT images according to claim 1, characterized in that: Steps 1 and 2 are performed offline; steps 3, 4, 5, 6, and 7 are performed online.

Images