Method for automated gas leak detection in battery manufacturing process using data from optical gas imaging system

Abstract

An automated system providing gas leak detection during battery manufacturing includes a battery system. The video of gas leaks occurring during a manufacturing phase of the battery system includes gas leaks as gas vapors. And determining the position of gas leakage. A leak intensity value for the gas leak is identified to determine whether the gas leak is slight and less than or within a predetermined window or threshold that allows the gas leak to be accepted without repair, or whether the gas leak requires further action that includes repairing the battery system.

Description

BACKGROUND OF THE DISCLOSURE

[0001] The present disclosure relates to electric vehicle battery systems and the identification of gas leaks that occur during the manufacture of rechargeable energy storage systems (RESS), battery cells, battery modules, or battery packs for electric vehicles.

[0002] An electric vehicle (EV) includes a battery electric vehicle (BEV), a hybrid vehicle, and / or a fuel cell vehicle. The EV includes one or more electric motors and a RESS or a battery system, and the battery system includes one or more battery cells, battery modules, and / or battery packs. The RESS, battery cells, battery modules, and / or battery packs are typically housed in an airtight sealed enclosure. Current methods for detecting gas leaks in the enclosure during battery manufacture include gas sniffing, such as helium sniffing, and these methods are performed manually, so they are time-consuming and inaccurate. In addition, helium sniffing can only identify the approximate area where the leak is located. Since the specific leak location must be identified to perform leak correction, such as welding repair, this method results in too much time required for leak detection and leak repair.

[0003] Accordingly, while current systems and methods for identifying gas leaks that occur during battery manufacture achieve their intended purposes, there is still a need for a new and improved system and method for automatically detecting gas leaks during battery manufacture. SUMMARY OF THE INVENTION

[0004] According to several aspects, an automated system for providing gas leak detection during battery manufacture includes a battery system that defines a RESS, battery cell, battery module, or battery pack. A video of a gas leak that occurs during a manufacturing phase of the battery system includes the gas leak as a gas vapor. The location of the gas leak is determined. A leak intensity value of the gas leak is identified to determine whether the gas leak is minor and less than or within a predetermined window or threshold that allows acceptance of the gas leak without repair, or whether the gas leak requires further action including repairing the battery system.

[0005] In another aspect of the present disclosure, a background frame with no gas leak is saved. The method of inter-frame differencing of consecutive frames of the video is compared with the background frame to define the frame in which the gas leak is initially detected.

[0006] In another aspect of the present disclosure, a classifier model is applied to identify at least one frame in the video that has a gas leak.

[0007] In another aspect of the present disclosure, a pixel threshold is applied to identify at least one frame in the video that has a gas leak.

[0008] In another aspect of the present disclosure, a pixel-to-physical location mapping is applied to identify the gas leak location.

[0009] In another aspect of the present disclosure, a contour and a boundary shape are applied to the frame initially detected with a gas leak. Multiple boundary shape coordinates are applied to the contour and the boundary shape to identify the physical location mapping of the gas leak.

[0010] In another aspect of the present disclosure, an image processor is applied to the image data defining the location of the gas leak to identify the leak intensity value of the gas leak.

[0011] In another aspect of the present disclosure, a convolutional neural network (CNN) model and / or a recurrent neural network (RNN) model are applied to the image data defining the location of the gas leak to identify the leak intensity value of the gas leak.

[0012] In another aspect of the present disclosure, object detection is performed during the detection phase to identify the gas leak; and a boundary enclosing the gas leak is defined to define the location or orientation of the gas leak.

[0013] In another aspect of the present disclosure, a video of the gas leak image data is prepared. Frame-by-frame analysis is performed on the image data to identify the frames in which the gas leak occurs.

[0014] According to several aspects, a method for performing automatic battery gas leak detection includes: collecting a video of a battery system having a plurality of video frames during the manufacturing phase of the battery system; detecting a gas leak occurring from the battery system, where the gas leak exists in at least one of the video frames; determining the physical location of the gas leak on the battery system; and identifying the leak intensity value of the gas leak.

[0015] In another aspect of the present disclosure, the method further includes: identifying the frames of the video in which the gas leak exists; and subtracting the background frame of the video in which the gas leak does not exist from the frames of the video in which the gas leak exists to obtain an inter-frame difference having the image data defining the gas leak.

[0016] In another aspect of the present disclosure, the method further includes: passing the inter-frame difference having the image data defining the gas leak through a classification network;

[0017] And generating a gas leak signal.

[0018] In another aspect of the present disclosure, the method further includes: separately identifying multiple inter-frame differences having the image data defining the gas leak; performing a summation of the multiple inter-frame differences; and determining the sum of the total pixel values in the multiple inter-frame differences to apply a pixel threshold to identify the image frames of the gas leak.

[0019] In another aspect of the present disclosure, the method further includes: fitting a contour to the image data defining the gas leak within the inter-frame difference; and applying a boundary shape to the contour, where the known coordinates of the boundary shape identify the physical location of the gas leak.

[0020] In another aspect of the present disclosure, the method further includes: performing a summation on a plurality of inter-frame differences, wherein the summation is equal to the sum of the pixel values in the plurality of inter-frame differences; wherein, if the sum of the pixel values is greater than a predetermined threshold among a plurality of predetermined thresholds, the leakage intensity value defines one of low-intensity leakage, medium-intensity leakage, or high-intensity leakage.

[0021] In another aspect of the present disclosure, the method further includes: saving a calibration file having a plurality of I i xy values, where the I i xy values define coordinates in respective video frames; identifying the location of the gas leakage as the Lxy value in an image of one of the video frames; calculating an i value, where i is equal to the parameter of the minimum value (argmin) multiplied by the absolute value of Lxy - I i xy; and using the calibration file to map at least one of the values of I i xy to the gas leakage location.

[0022] According to several aspects, a method for performing automatic gas leakage detection during the manufacture of a battery for a vehicle battery system defining a RESS, a battery cell, a battery module, or a battery pack includes: collecting image data of a video of the vehicle battery system, the video having a plurality of video frames saved during a manufacturing stage of the battery system; detecting a gas leakage that occurs and is present in at least one of the video frames, wherein the gas leakage is a gas vapor escaping from the surface of the battery system into the atmosphere; applying one of object detection analysis, frame-by-frame analysis, or a classifier model to determine the location of the gas leakage; and identifying a leakage intensity value of the gas leakage to determine whether the gas leakage is minor and less than or within a predetermined window or threshold allowing acceptance of the gas leakage without repair, or whether the gas leakage requires further action including repairing the battery system.

[0023] In another aspect of the present disclosure, the method further includes: defining a boundary surrounding the gas leakage to identify the location or orientation of the gas leakage; and identifying the coordinates of the boundary to identify the location of the gas leakage.

[0024] In another aspect of the present disclosure, the method further includes: during frame-by-frame analysis, capturing at least one frame of a visual image of the video containing the gas leakage; and subtracting a background frame of the video without the gas leakage from at least one of the plurality of video frames of the video with the gas leakage to obtain an inter-frame difference having image data defining the gas leakage.

[0025] Other application areas will become apparent from the description provided herein. It should be understood that these descriptions and specific examples are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Description of the Drawings

[0026] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the present disclosure in any way.

[0027] Figure 1 is a flowchart of steps taken to detect a gas leak according to an exemplary aspect;

[0028] Figure 2 is a flowchart of a frame-by-frame analysis method for identifying a gas leak;

[0029] Figure 3 is a system flowchart of multiple automated steps for detecting and identifying a gas leak by applying video imaging and analyzing multiple video images using frame-by-frame analysis;

[0030] Figure 4 is video frame comparison, subtracting the image data of the background frame from the frame showing no gas leak to obtain the inter-frame difference representing no gas leak;

[0031] Figure 5 is video frame comparison, subtracting the image data of the background frame from the frame showing a gas leak to obtain the inter-frame difference representing a gas leak;

[0032] Figure 6 is the inter-frame difference of video frames without gas leak, which generates a no-gas-leak signal through a classification network;

[0033] Figure 7 is the inter-frame difference of video frames with gas leak, which generates a gas-leak signal through a classification neural network;

[0034] Figure 8 is summing the inter-frame differences of video frames to obtain the sum of pixel values in the inter-frame differences of video frames;

[0035] Figure 9 is a graph of pixel value differences of inter-frame differences showing no gas leak, medium gas leak, and high gas leak;

[0036] Figure 10 is a flowchart of a gas leak image, coordinate mapping of the gas leak, and a calibration file applied to map the gas leak location;

[0037] Figure 11 is a diagram of steps taken to map the physical location of the gas leak;

[0038] Figure 12 is a flowchart of a gas leak image, a contour fitted to the gas leak image, and a boundary shape after fitting to the contour;

[0039] Figure 13 is a flowchart of the summing step for identifying the sum of pixel values in the inter-frame differences;

[0040] Figure 14 is a histogram for identifying the gas leakage intensity value;

[0041] Figure 15 is a flowchart for identifying the gas leakage flow intensity value by applying a convolutional neural network; and

[0042] Figure 16 is a flowchart for identifying the gas leakage flow intensity value by applying a recurrent neural network and a convolutional neural network together. Detailed implementation manners

[0043] The following description is merely exemplary in nature and is not intended to limit the present disclosure, application, or uses.

[0044] When a component, element, or layer is referred to as being "on", "engaged to", "connected to", or "coupled to" another element or layer, it may be directly on, engaged, connected, or coupled to another component, element, or layer, or there may be intervening elements or layers. In contrast, when an element is referred to as being "directly on", "directly engaged to", "directly connected to", or "directly coupled to" another element or layer, there may be intervening elements or layers. Other words used to describe the relationship between elements should be interpreted in a similar manner, such as "between" relative to "directly between", "adjacent" relative to "directly adjacent", etc. As used herein, the terms "and / or" and "one or two" include any and all combinations of one or more of the associated listed items.

[0045] Reference Figure 1 In, the automatic gas leakage detection system 10 during the battery manufacturing process uses image data and / or video data from an optical gas imaging system to automatically detect and identify the specific locations and intensities of gas leakage from a rechargeable energy storage system (RESS), cell, module, and / or battery pack (collectively referred to as a battery system hereinafter) during the manufacturing process of the battery system. The automatic gas leakage detection system 10 during the battery manufacturing process can be applied to automatically perform gas leakage detection on the housing of the battery system, which exists in the form of a RESS, battery cell, battery module, or battery pack during the manufacturing stage of the battery system.

[0046] In this example, the battery cell 12 of the battery system 14 that defines the RESS, battery cell, battery module, or battery pack has one or more features that include terminals. In the illustrated example, a gas leak 16 occurs during the manufacturing phase of the battery cell 12, where the gas leak 16 is a gas vapor detected using an optical gas imaging system 17. The illustrated gas vapor escapes from the surface 18 of the battery cell 12 into the atmosphere. According to several aspects, the battery system 14 is installed or intended to be installed in a vehicle 19, such as but not limited to a battery electric vehicle, a gas / electric hybrid vehicle, a sport utility vehicle, a truck, a van, etc.

[0047] Continuing to refer Figure 1 , using the optical gas imaging system 17, a target detection 22 is performed in the detection phase 20 to identify the gas leak 16. To define the location or orientation of the gas leak 16, a boundary 24 that encloses the gas leak 16 is defined. The coordinates of the boundary 24 are identified or known. According to several aspects, to provide the ability to determine whether the gas leak 16 is minor and less than or within a predetermined window or threshold that allows acceptance of the gas leak 16 without repair, or to determine whether the gas leak 16 is severe and requires further action including repairing the battery system 14, the gas leak 16 can be assigned a leak intensity value 26, which will be described in more detail with reference to Figure 3 , Figure 13 , Figure 14 , Figure 15 and Figure 16 below.

[0048] Referring Figure 2 and referring again Figure 1 , instead of performing the detection phase 20 to identify a specific target that can define the gas leak 16, a video 28 of the same battery cell 12 can be generated, thereby performing a frame-by-frame analysis 30 of the gas leak 16. The frame-by-frame analysis 30 allows at least one frame 31 containing a visual image 32 of the gas leak to be captured.

[0049] Referring Figure 3 and referring again Figure 2, According to another aspect, frame-by-frame analysis 30 can be performed to identify and quantify visual images 32 of gas leakage from the battery system 14. In the startup step 34, a background frame of a video showing no gas leakage from the battery system over time, for example, is saved. In the difference step 36, a consecutive frame difference method is performed on the video of the battery system to identify whether a difference such as gas leakage has occurred at a certain point after the background frame is identified. In the identification step 38, at least one individual frame from the consecutive frame difference method step detecting gas leakage is identified. In the location identification step 40, the location where gas leakage occurs in the battery system is identified. After the location identification step 40, a leakage intensity step 42 is performed to distinguish the leakage intensity, which is estimated based on a comparison with imaging of known gas leakage intensities.

[0050] After the identification step 38, a leakage classification step 44 is performed by applying a classifier model. Meanwhile, a thresholding step 46 is performed to identify the pixel threshold of the image of the detected gas leakage.

[0051] After the location identification step 40, further differentiation of the leakage location is performed through a mapping step 48, in which pixel-to-physical location mapping is performed. While performing the mapping step 48, an exhaust gas shape identification step 50 is performed, in which the contour and boundary shape of the detected gas leakage are performed. After the exhaust gas shape identification step 50, a boundary step 52 is performed, in which the shape coordinates of the boundary shape of the detected gas leakage are defined and mapped to the physical location of the gas leakage.

[0052] After the leakage intensity step 42, an image processing step 54 is performed to apply the image data of the gas leakage to known patterns or known images of gas leakage saved in a database as an initial step for quantifying the gas leakage intensity. Then this data can be compared with previously saved data to help determine whether the detected gas leakage is occurring, and this gas leakage has a predetermined amount compared to a gas leakage less than a predetermined threshold. A gas leakage less than the predetermined threshold is considered acceptable and thus does not require repair of the battery system. While performing the image processing step 54, the image data of the gas leakage can be input into different models to further quantify the gas leakage intensity. For example, a convolutional neural network (CNN) model and / or a recurrent neural network (RNN) model can be used. The CNN model is considered more effective than the RNN model and is very suitable for imaging and video processing because the CNN model learns to identify patterns across space, while the RNN model can be used to solve temporal data problems. The RNN is very suitable for text and speech analysis, and the RNN model has recurrent connections, while the CNN model does not necessarily have recurrent connections. The RNN model also allows arbitrary input lengths and output lengths.

[0053] ReferenceFigure 4 and Figure 5 , the inter-frame difference method can be applied to identify gas leakage. More specifically referring to Figure 4 , first, frames 58 without leakage are collected from the image data of the battery cell (such as the battery cell 12 described in reference Figure 1 ), and the background frame 60 is subtracted from them. The background frame 60 can be stored in a database or memory and retrieved from the database or memory, and an image showing the battery cell 12 without gas leakage is provided. Therefore, the first inter-frame difference 62 in this example does not include the image data defining gas leakage.

[0054] Referring to Figure 5 , the inter-frame difference method is further applied to the frame with leakage 64, and the background frame 60 is also subtracted from this frame. The second inter-frame difference 66 in this example includes the image data 68 defining gas leakage.

[0055] Referring to Figure 6 , by applying a classifier model (such as the classifier neural network 70) to the first inter-frame difference 62, it is possible to identify the image frames for identifying gas leakage. In this example, a no-leakage signal 72 is generated by the classifier neural network 70.

[0056] Referring to Figure 7 and referring again to Figure 6 , the classifier neural network 70 is applied to the second inter-frame difference 66. In this example, a gas-leakage signal 74 is generated by the classifier neural network 70, which is produced by the classification of the image data 68 defining gas leakage.

[0057] Referring to Figure 8 , the image frames for identifying gas leakage can also be identified by applying pixel threshold processing as follows. For example, a summation 76 is performed on multiple inter-frame differences in the second inter-frame differences 66 to 66n. The summation 76 produces a sum 78 of the total pixel values in the multiple inter-frame differences in the second inter-frame differences 66 to 66n.

[0058] Referring to Figure 9 and referring again to Figure 8, the sum 78 of the total pixel values of a plurality of inter-frame differences among the second inter-frame differences 66 to 66n can be presented in a curve graph 80 that compares the sum of pixel values 82 with the frame number 84 of image data of, for example, a battery cell 12 collected over a period of time. The curve 86 shows the pixel summation data frame by frame. The first data point 88 along the curve 86 represents the first non-gas-leakage frame 90, and the result shown by the first non-gas-leakage frame 90 is similar to the result shown by the first inter-frame difference 62. The second data point 92 can be represented as the second gas-leakage frame 94, which has a sum of pixel values representing a low gas leakage 96. The third data point 98 at the peak of the curve 86 can be represented as the third gas-leakage frame 100, which has a sum of pixel values representing a high gas leakage 102.

[0059] Reference Figure 10 , by applying a direct pixel-to-physical location mapping method, the location of gas leakage occurring, for example, from the battery cell 12 is determined. A two-dimensional (2D) image 104 identifies a slight or low gas leakage 106 near one of three potential leakage locations 108, 110, 112, and the three potential leakage locations 108, 110, 112 are also respectively defined as horizontal position I 1 xy, I 2 xy and I 3 xy, and is approximately located at position 110I according to this example 2 xy. The horizontal position I 1 xy, I 2 xy and I 3 xy values are provided as examples, and the programmer can decide to include more or fewer than three positions. Any Ixy leakage location can be shown on the second gas-leakage frame 94, such as the low gas leakage 106. The values of the Ixy positions can be compared with the values stored in a calibration file 114, and the calibration file 114 provides baseline values of X, Y, and Z coordinates for the first I 1 xy value 116, the second I 2 xy value 118, and up to the nth I N xy value 120.

[0060] Reference Figure 11 And referring again to Figure 10 , three steps can be used to map any Ixy gas leakage location. In the first leakage location step 122, a gas leakage location Lxy is identified in an image (such as a photographic image of a battery). In the second leakage location step 124, the value of i is calculated to be equal to the parameter (argmin) N of the minimum value multiplied by the absolute value of Lxy - I i xy. In the third leakage location step 126, using reference Figure 10The identified calibration file 114 maps the I i xy values to physical locations.

[0061] Reference Figure 12 , by applying the contour and boundary shape methods, the location of gas leakage occurring, for example, from the battery cell 12 is determined. First, in the first fitting step 128, an inter-frame difference such as the second inter-frame difference 66 described in the reference Figure 5 has a contour 130 that fits to the image data 68 defining the gas leakage. Then, in a subsequent second fitting step 132, a boundary shape 134 such as a rectangular shape fits onto the contour 130. The coordinates of the boundary shape 134 are known and can thus be used to identify the physical location of the gas leakage. According to other aspects, in addition to the rectangular boundary shape 134, other boundary shapes can be used, including but not limited to circular, elliptical, etc.

[0062] Reference Figure 13 , the leakage intensity can be determined by applying image processing to, for example, classify the leakage intensity as low-intensity, medium-intensity, or high-intensity leakage. This is achieved by performing a summation 136 on a plurality of inter-frame differences including the second inter-frame difference 66 to the inter-frame difference 66N. The sum 136 is equal to the sum of the pixel values 138 in the inter-frame differences. If the sum of the pixel values 140 in the inter-frame differences is greater than one of a plurality of predetermined thresholds 142, the leakage intensity 144 can be determined as low-intensity leakage 146, medium-intensity leakage 148, or high-intensity leakage 150 by the difference between the sum of the pixel values 140 and the predetermined threshold 142.

[0063] Reference Figure 14 , by applying statistical methods, the leakage intensity can be identified by using the statistical data of a plurality of pixel value histograms superimposed on the graph 152. These can include, for example, identifying a first pixel value histogram 154 with a net leakage intensity lower than the second pixel value histogram 156, which identifies a net higher leakage intensity indicated by a higher statistical average pixel value.

[0064] Reference Figure 15 and Figure 16 , a convolutional neural network (CNN) model and / or a recurrent neural network (RNN) model can also be used to identify the gas leakage intensity. Specifically referring to Figure 15, data from different images of the gas leakage frames is fed into a CNN classifier that determines the leakage rate. In a first detection step 158, the image data defining a first non-gas leakage 160 is fed into a CNN classifier 162, which outputs a no-leakage or low-leakage signal 164. In a second detection step 166, the image data defining a second gas leakage 168 is fed into the CNN classifier 162, which outputs a medium-leakage signal 170. In a third detection step 172, the image data defining a third gas leakage 174 is fed into the CNN classifier 162, which outputs a high-leakage signal 176.

[0065] Specific reference Figure 16 , after a gas leakage is first detected, the next N video images are analyzed to identify whether the gas leakage is low, medium, or high. In the example shown, a gas leakage is first detected in frame 1, and then any changes in the gas leakage are detected in subsequent frames 2 to 5. The CNN features 180 of the individual frames from frames 1 to 5 are passed to a first simple RNN 182. The output of the first simple RNN 182 is passed to a second simple RNN 184. The gas leakage intensity can be classified as one of low leakage, medium leakage, or high leakage for each frame, and the resulting leakage signal is identified at output 186.

[0066] In Figure 16 the example given, if the gas leakage detected in frame 1 is a low gas leakage and the image data does not change in subsequent frames such as from frame 2 to 5, the output 186 will identify a low gas leakage. If the gas leakage detected in frame 1 becomes a medium gas leakage, for example, by frame 3, and remains substantially the same thereafter, the output 186 will identify a medium gas leakage. If the intensity of the gas leakage detected in frame 1 continuously increases to a high gas leakage by frame 5, the output 186 will identify a high gas leakage.

[0067] Automatically detect and identify the intensity and specific location of gas leakage using images and / or video data from an optical gas imaging system. The automatic gas leakage detection system 10 in the battery manufacturing process can be applied to detect gas leakage in the enclosures of rechargeable energy storage systems (RESS), battery cells, battery modules, or battery packs during manufacturing.

[0068] The automatic gas leakage detection system 10 in the battery manufacturing process of the present disclosure offers several advantages. These include image processing and deep learning-based methods that automatically detect and identify the specific location and intensity of gas leakage using images and / or video data from an optical gas imaging system. The automatic gas leakage detection system 10 in the battery manufacturing process can be applied to detect gas leakage in the enclosures of rechargeable energy storage systems (RESS), battery cells, battery modules, or battery packs during manufacturing.

Claims

1. An automated system for providing gas leak detection during battery manufacturing, comprising: Battery system; a video of a gas leak occurring during a manufacturing stage of the battery system, the video of the gas leak including the gas leak as a gas vapor; The location of the identified gas leak; and A leak intensity value for the gas leak is identified to determine whether the gas leak is minor and less than or within a predetermined window or threshold that allows the gas leak to be accepted without repair, or whether the gas leak requires further action including repairing the battery system.

2. The automated system for providing gas leak detection during battery manufacturing according to claim 1, comprising: The background frames of the video without gas leaks are saved; as well as The differences between successive frames of the video when compared to background frames defining the frame in which the gas leak was initially detected.

3. The automated system for providing gas leak detection during battery manufacturing of claim 2, further comprising a classifier model applied to identify at least one frame of the video having the gas leak.

4. The automated system for providing gas leak detection during battery manufacturing of claim 2, further comprising a pixel threshold applied to identify at least one frame of the video having the gas leak.

5. The automated system for providing gas leak detection during battery manufacturing of claim 2, further comprising a pixel to physical location mapping applied to identify the location of the gas leak.

6. The automated system for providing gas leak detection during battery manufacturing of claim 2, comprising: an outline and boundary shape, the outline and boundary shape applied to the frame in which the gas leak was initially detected; as well as A plurality of bounding shape coordinates are applied to the contour and bounding shape to identify a physical location map of the gas leak.

7. An automated system for providing gas leak detection during battery manufacturing as in claim 1, comprising an image processor applied to image data defining the location of the gas leak to identify a leak intensity value for the gas leak.

8. The automated system for providing gas leak detection during battery manufacturing according to claim 1, comprising a convolutional neural network (CNN) model and / or a recurrent neural network (RNN) model, wherein the convolutional neural network (CNN) model and / or the recurrent neural network (RNN) model is applied to image data defining the location of the gas leak to identify a leak intensity value of the gas leak.

9. The automated system for providing gas leak detection during battery manufacturing of claim 1, comprising: object detection, the object detection being performed during a detection phase to identify the gas leak; as well as A defined boundary surrounds the gas leak to define a location or orientation of the gas leak.

10. The automated system for providing gas leak detection during battery manufacturing of claim 1, comprising: Image data of a video distinguished in a plurality of frames; as well as A frame-by-frame analysis is performed on the image data to identify at least one frame in a plurality of frames in which the gas leak occurs.

Images