Model generation method and abnormality estimation system

Abstract

The invention provides a model generation method for generating an abnormality estimation model and an abnormality estimation system using the abnormality estimation model. The abnormality estimation model can estimate an abnormality of a battery stack, namely a test object, from X-ray image data. An abnormality estimation system (1) is provided with: an input data reception unit (2) that receives, as input data, X-ray CT image data (7D) of a battery stack, i.e., a test object (7); a training data storage unit (6) that stores, as training data, X-ray CT image data of a battery stack, i.e., a sample, and abnormal data of the same sample; a model generation unit (3) that generates an abnormality estimation model for the battery stack by means of machine learning using the training data stored in the training data storage unit (6); a model estimation unit (4) that estimates an abnormality of the test subject (7) from the input data received by the input data reception unit (2) using the abnormality estimation model generated by the model generation unit (3); and a display unit (5) for displaying the estimation result of the model estimation unit (4).

Description

Technical Field

[0001] The present invention relates to a model generation method and an abnormality estimation system. More specifically, the present invention relates to a model generation method for generating an abnormality estimation model for a battery stack and an abnormality estimation system for estimating abnormalities in the battery stack using the abnormality estimation model. Background Art

[0002] In recent years, efforts have been actively made to realize a low-carbon society or a decarbonized society. In order to reduce CO2 emissions and improve energy efficiency in vehicles, research and development related to secondary batteries, especially all-solid-state batteries, is being carried out.

[0003] To improve battery performance and yield, it is essential to evaluate the stacked structure of all-solid-state batteries and the micron-level material dispersion within specific layers. For example, the X-ray computed tomography (CT) device described in Patent Document 1 can non-destructively observe the internal structure of the test object, making it a useful method for evaluating the internal structure of all-solid-state batteries.

[0004] [Prior Art Literature]

[0005] (Patent Document)

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2020-187024 Summary of the Invention

[0007] [Problems to be solved by the invention]

[0008] However, X-ray images obtained by X-ray CT devices have lower resolution than SEM images obtained by scanning electron microscopes (hereinafter referred to as "SEM"), and the types of elements that can be analyzed are also limited, so it is impossible to infer internal abnormalities of all-solid-state batteries with high precision.

[0009] In order to solve the above-mentioned problems, the present invention aims to provide a model generation method for generating an abnormality inference model and an abnormality inference system using the abnormality inference model. The abnormality inference model can use X-ray image data to infer abnormalities of a battery stack, that is, a test object, and thus the present invention helps to improve energy efficiency.

[0010] [Technical means to solve the problem]

[0011] (1) The model generation method of the present invention generates an abnormality inference model that uses X-ray image data (e.g., X-ray CT image data 7D described later) of a battery stack, i.e., a test object (e.g., test object 7 described later) as input and uses abnormality data of the aforementioned test object as output. The model generation method is characterized in that the aforementioned abnormality inference model is generated by machine learning using X-ray image data of a battery stack, i.e., a sample (e.g., sample S described later) and abnormality data of the aforementioned sample as training data.

[0012] (2) In this case, it is preferable that the abnormality data of the sample include information obtained by observing a cut surface (for example, a cut surface CS described later) of the sample after the X-ray image data is acquired.

[0013] (3) In this case, it is preferred that the X-ray image data of the sample include information obtained by irradiating the sample fixed to a columnar jig (for example, jig 92 described later) with X-rays, and the abnormality data of the sample include information obtained by observing the cut surface after cutting the sample by irradiating it with an ion beam (for example, focused ion beam B described later) while the sample is fixed to the jig.

[0014] (4) In this case, it is preferable that the abnormality data of the sample include information obtained by alternately and repeatedly cutting the sample and observing the cut surface.

[0015] (5) The abnormality estimation system of the present invention (for example, the abnormality estimation system 1 described later) is characterized by comprising: an input data receiving unit (for example, the input data receiving unit 2 described later) that receives X-ray image data (for example, X-ray CT image data 7D described later) of a battery stack, that is, a test object (for example, the test object 7 described later) as input data; a model generating unit (for example, the model generating unit 3 described later) that generates an abnormality estimation model by machine learning using the X-ray image data of the battery stack, that is, a sample (for example, the sample S described later) and the abnormality data of the aforementioned sample as training data; and a model estimating unit (for example, the model estimating unit 4 described later) that uses the aforementioned abnormality estimation model to infer the abnormality of the aforementioned test object based on the aforementioned input data.

[0016] (6) The abnormality estimation system of the present invention (for example, the abnormality estimation system 1 described later) is characterized by comprising: an input data receiving unit (for example, the input data receiving unit 2 described later) that receives X-ray image data (for example, X-ray CT image data 7D described later) of a battery stack, that is, a test object (for example, the test object 7 described later) as input data; and a model estimation unit (for example, the model estimation unit 4 described later) that uses an abnormality estimation model to estimate the abnormality of the test object based on the above-mentioned input data, wherein the abnormality estimation model is generated by machine learning using X-ray image data of a battery stack, that is, a sample (for example, the sample S described later), and abnormality data of the above-mentioned sample as training data.

[0017] (Effects of the Invention)

[0018] (1) According to the model generation method of the present invention, an abnormality estimation model is generated by machine learning using X-ray image data of a battery stack, i.e., a sample, and abnormality data of the same sample as training data. Thus, by inputting X-ray image data of a battery stack, i.e., a test subject, as input data, an abnormality estimation model can be generated that estimates abnormalities of the test subject. Furthermore, by utilizing the abnormality estimation model using X-ray image data of the battery stack as input data, abnormalities of the test subject can be non-destructively and rapidly estimated, thereby improving battery performance or yield, thereby contributing to improved energy efficiency.

[0019] (2) According to the model generation method of the present invention, by using abnormality data containing information obtained by observing the cross-section of a sample as training data, an abnormality estimation model can be used to associate the X-ray image data of the test subject with abnormalities that cannot be determined by analyzing the X-ray image data alone. Thus, according to the present invention, an abnormality estimation model with high abnormality estimation accuracy can be generated.

[0020] (3) The model generation method of the present invention uses, as training data, X-ray image data containing information obtained by irradiating a sample fixed to a cylindrical jig with X-rays, and abnormality data containing information obtained by observing the cut surface after the sample is cut by irradiating an ion beam while the sample is fixed to the jig. Specifically, the present invention allows the coordinate positions in the X-ray image data to be aligned with the coordinate positions in the abnormality data, thereby generating an abnormality estimation model with high accuracy in estimating abnormalities.

[0021] (4) By alternately and repeatedly cutting the sample and observing the cut surface, three-dimensional information about the interior of the sample can be obtained. According to the model generation method of the present invention, by using abnormal data containing this three-dimensional information as training data, an abnormality estimation model with high anomaly estimation accuracy can be generated.

[0022] (5) According to the abnormality estimation system of the present invention, the input data receiving unit receives X-ray image data of a battery stack, i.e., a test object, as input data. The model generation unit generates an abnormality estimation model through machine learning using the sample X-ray image data and abnormality data as training data. The model estimation unit uses the abnormality estimation model generated by the model generation unit to estimate abnormalities of the test object based on the input data received by the input data receiving unit. Thus, according to the present invention, abnormalities of the test object can be estimated non-destructively and quickly, thereby improving battery performance or yield, thereby contributing to improved energy efficiency.

[0023] (6) According to the abnormality estimation system of the present invention, the input data receiving unit receives X-ray image data of a battery stack, i.e., a test object, as input data, and the model estimation unit uses an abnormality estimation model to estimate abnormalities of the test object based on the input data received by the input data receiving unit. The abnormality estimation model is generated by machine learning using sample X-ray image data and abnormality data as training data. Thus, according to the present invention, abnormalities of the test object can be estimated non-destructively and rapidly, thereby improving battery performance or yield, thereby contributing to improved energy efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a diagram showing the structure of an abnormality estimation system according to one embodiment of the present invention.

[0025] Figure 2A This is a diagram showing a first display example of the abnormality estimation result of the test subject.

[0026] Figure 2B This is a diagram showing a second display example of the abnormality estimation result of the test subject.

[0027] Figure 3 It is a flowchart illustrating the specific process of the training data generation method.

[0028] Figure 4A This is a diagram schematically showing a process for preparing a sample of a battery stack.

[0029] Figure 4B FIG2 is a diagram schematically showing the structure of a sample holder.

[0030] Figure 4C It is a diagram schematically showing the flow of the cutting process and the observation process.

[0031] Figure 5 This figure compares an X-ray CT image and a SEM image.

[0032] Figure 6A This is a diagram showing an example of an X-ray CT image of a battery stack.

[0033] Figure 6B This is a diagram showing an example of a SEM image of a battery stack. DETAILED DESCRIPTION

[0034] Hereinafter, the structure of a battery stack abnormality estimation system according to one embodiment of the present invention and the flow of a method for generating an abnormality estimation model used in the abnormality estimation system will be described with reference to the drawings.

[0035] Figure 1 This diagram illustrates the structure of an abnormality estimation system 1 according to this embodiment. Abnormality estimation system 1 uses a battery stack (more specifically, a fully solid battery or a semi-solid battery comprising a negative electrode current collector, a negative electrode layer, a solid electrolyte layer such as a fully solid or semi-solid electrolyte layer such as a gel, a positive electrode layer, and a positive electrode current collector) whose internal state is unknown as a test object 7, and estimates abnormalities in the test object 7 based on X-ray image data of the test object 7 generated by an X-ray CT device 81.

[0036] The X-ray CT apparatus 81 performs CT (Computed Tomography) on the test object 7 using X-rays. The X-ray CT apparatus 81 comprises an X-ray source for irradiating the test object 7 with X-rays; an X-ray detector, positioned across the test object 7 from the X-ray source, for detecting the X-rays that pass through the test object 7; and an image processing computer for generating an X-ray image based on the X-rays detected by the X-ray detector. The computer then performs predetermined image processing on the X-ray image to produce a slice-like image (hereinafter also referred to as "X-ray CT image data") of the surface along the stacking direction of the battery stack, i.e., the test object 7.

[0037] Abnormality estimation system 1 is a computer composed of the following hardware: a processing unit such as a central processing unit (CPU), auxiliary storage such as a hard disk drive (HDD) or solid state drive (SSD) for storing various programs, main storage such as random access memory (RAM) for temporarily storing data required by the processing unit while executing programs, and a display unit for visually displaying the processing results of the processing unit in a manner that can be visually recognized by the user. In abnormality estimation system 1, this hardware structure realizes various functions, such as an input data receiving unit 2, a model generation unit 3, a model estimation unit 4, a display unit 5, and a training data storage unit 6.

[0038] The input data accepting unit 2 accepts X-ray CT image data 7D of the test subject 7 generated by the X-ray CT apparatus 81 as input data.

[0039] The model generation unit 3 reads multiple sets of training data stored in the training data storage unit 6 and uses this data to perform machine learning on the neural network, thereby generating an abnormality estimation model that uses X-ray CT image data of the test object 7 as input and outputs abnormality data of the test object 7. As will be described later, the abnormality data output from the abnormality estimation model includes information related to abnormalities in the shape or composition of the test object 7 (more specifically, information related to the presence or absence of abnormalities such as cracks, pores, Li precipitation, and foreign matter adhesion, as well as the locations of these abnormalities and the three-dimensional structure of the abnormalities).

[0040] The training data storage unit 6 stores a plurality of sets of training data used in the machine learning of the model generation unit 3 as described above. Figures 3 to 5 The detailed description uses data generated for each of a plurality of samples, that is, battery stacks prepared in advance other than the test object 7. Each training data set includes at least X-ray CT image data of each sample and abnormality data of the same sample.

[0041] The model estimation unit 4 uses the abnormality estimation model generated by the model generation unit 3 to estimate an abnormality of the test subject 7 based on the input data of the test subject 7 accepted by the input data acceptance unit 2. When the input data accepted by the input data acceptance unit 2 is input to the abnormality estimation model, the model estimation unit 4 transmits the abnormality data output from the abnormality estimation model as an estimation result of the abnormality of the test subject 7 to the display unit 5.

[0042] The display unit 5 displays the estimation result transmitted from the model estimation unit 4 in a manner that the operator can visually see it.

[0043] Figure 2A and Figure 2B : is a diagram showing an example of display of the estimated result of abnormality of the test subject 7. Figure 2A and Figure 2B As shown, when the abnormality data includes information regarding the location of the abnormality, the display unit 5 may also highlight the X-ray CT image of the test subject 7 received by the input data receiving unit 2, along with the location and type of the abnormality. This allows the operator to easily identify the location and type of abnormality within the test subject 7, which cannot be visually confirmed. Furthermore, according to this embodiment, by using the abnormality estimation model to estimate the abnormality of the test subject 7 based on the X-ray CT image data 7D, the operator can confirm the presence of abnormalities in the internal shape and composition of the test subject 7, which are difficult to determine based solely on analysis of the X-ray CT image data 7D, without destroying the test subject 7.

[0044] Next, for the flow of a method for generating training data required to generate the abnormality estimation model of the battery stack as described above, refer to Figures 3 to 5 Provide detailed explanation.

[0045] Figure 3 is a flowchart illustrating the specific process of the training data generation method. More specifically, Figure 3 The process of generating training data for one sample is shown in FIG. That is, the multiple sets of training data required to generate anomaly estimation models by machine learning can be generated by repeatedly performing the training on different samples. Figure 3 The training data is generated using the method shown.

[0046] Figures 4A to 4C This is a diagram for schematically explaining the flow of each step of the training data generation method.

[0047] First, in step ST1 , the operator prepares a sample of a battery stack, and then the process proceeds to step ST2 .

[0048] Figure 4A FIG is a diagram schematically illustrating the process of preparing a sample of a battery stack in step ST1. In step ST1, the operator uses, for example, a focused ion beam (FIB) device 82, such as Figure 4AAs shown in FIG, a focused ion beam B is irradiated along the stacking direction onto a sheet of battery stack, and a portion of the battery stack is excised as a sample S for generating training data, described below. As described in detail below, while the position of the sample S is fixed using a jig, the sample S is preferably excised from the battery stack so that one side has a length of approximately tens of μm to several centimeters, allowing for observation using at least an X-ray CT scanner and a scanning electron microscope.

[0049] Next, in step ST2 , the operator attaches the sample S prepared in step ST1 to the sample holder 9 , generates X-ray CT image data of the sample S using the X-ray CT apparatus, and then proceeds to step ST3 .

[0050] Figure 4B This figure schematically illustrates the structure of the sample holder 9. The sample holder 9 comprises a base 91, a columnar jig 92 mounted on the base 91, and a non-exposed seal 93 that covers the jig 92 and the sample S secured thereto. Multiple needle-like components are mounted on the tip of the jig 92, and the bottom of the sample S is secured to the surface of these needle-like components. More specifically, the bottom of the sample S is bonded to the surface of the needle-like components by, for example, a deposited film formed using the FIB apparatus 82.

[0051] In step ST2, the operator Figure 4B As shown in FIG, the sample S is fixed to the front end of the jig 92. In step ST2, the operator uses, for example, Figure 1 The X-ray CT apparatus 81 shown in FIG. 8 irradiates the sample S fixed to the jig 92 with X-rays R to generate X-ray CT image data of the sample S. More specifically, Figure 4B As shown by the middle arrow, the X-ray source and the X-ray detector of the X-ray CT device 81 are moved along a circle centered on the sample S and set within a plane perpendicular to the jig 92, while continuously performing X-ray CT imaging on the sample S, thereby generating X-ray CT image data of the sample S.

[0052] Next, in step ST3 , the operator alternately repeats the cutting process using the FIB device 82 and the observation process of the cut surface of the sample S using the observation device 83 at least once, preferably multiple times, on the sample S fixed to the jig 92 , thereby generating abnormality data of the sample S.

[0053] Figure 4C Schematically depicts the flow of the cutting process and the observation process in step ST3. In the cutting process in step ST3, the operator uses the FIB device 82, such as Figure 4CAs shown in FIG, the sample S is irradiated with a focused ion beam B along a predetermined cutting plane to cut the sample S. The cutting plane of the sample S is set in parallel with the stacking direction of the sample S so that the battery stack, that is, the internal stacking structure of the sample S, is exposed. In addition, in this embodiment, the following case is described, in which the positions of the multiple cutting planes of the sample S are set as shown in FIG. Figure 4C The dashed lines schematically illustrate a case where the positions of the multiple cross-sections of the sample S are pre-set to be equally spaced in a direction perpendicular to the cross-sections; however, the present invention is not limited thereto. The positions of the multiple cross-sections of the sample S may also be manually or mechanically set based on the X-ray CT image data of the sample S acquired in step ST2, so as to include locations where abnormalities are predicted to exist in the sample S.

[0054] Furthermore, in the observation process in step ST3, the operator generates abnormality data of the sample S by observing the cut surface of the sample S revealed in the cutting process using an observation device 83. In this embodiment, a case where the observation device 83 is a device formed by combining a scanning electron microscope (SEM) that observes a magnified image of the cut surface using an electron beam, an energy dispersive X-ray spectrometer (EDS) that performs elemental analysis of the cut surface by detecting characteristic X-rays generated on the cut surface irradiated with the electron beam, and a time-of-flight secondary ion mass spectrometer (TOF-SIMS) that performs elemental or molecular analysis of the cut surface by detecting secondary ions generated when the cut surface is irradiated with pulsed ions is described. However, the present invention is not limited to this. In addition to these devices, for example, a Raman spectrometer can be used to analyze the chemical bond state or crystal structure of the cut surface, an electron beam backscatter diffraction device can be used to analyze the crystal structure, and a soft X-ray spectrometer can be used to analyze the elements or chemical bond state.

[0055] In step ST3 , the operator uses the observation device 83 to obtain an SEM image of the cross section of the sample S and element information in the cross section, and manually or mechanically analyzes the SEM image and element information to generate abnormality data of the cross section.

[0056] Figure 5This figure compares the X-ray CT image acquired in step ST2 and the SEM image acquired in step ST3. Figure 5 , a schematic diagram of an SEM image of a predetermined cross-section CS of a sample S is shown on the left, and a schematic diagram of an X-ray CT image of the same cross-section is shown on the right.

[0057] By using an X-ray CT device, an image of the internal cross section of the sample S can be obtained without destroying the sample S. In contrast, when using a scanning electron microscope, a portion of the sample S needs to be destroyed to expose the internal cross section CS. Figure 5 As shown in , a higher-resolution image of the cross-section CS can be obtained than with an X-ray CT image. Thus, in step ST3, by analyzing the SEM image, it is possible to confirm the presence of cracks (i.e., pores) at locations that cannot be accurately identified using X-ray CT images alone. Furthermore, in step ST3, by analyzing the elemental information obtained using EDS or TOF-SIMS, for example, it is possible to confirm the presence of elements adhering as foreign matter that cannot be identified using X-ray CT images alone, or to confirm the occurrence of Li precipitation at locations that cannot be identified using X-ray CT images alone.

[0058] In step ST3, the operator generates abnormal data of sample S by alternating between the cutting process and the observation process multiple times, including the presence or absence of abnormalities such as cracks, pores, Li precipitation, and foreign matter adhesion, as well as the occurrence locations of these abnormalities and the abnormal three-dimensional structure, which cannot be easily determined by analyzing the X-ray CT image alone.

[0059] Furthermore, in step ST3, the operator preferably repeats the aforementioned cutting and observation steps while keeping the sample S fixed to the jig 92 after the X-ray CT image data has been acquired, that is, without moving the sample S relative to the jig 92 from the time the X-ray CT image data is acquired. This allows the coordinate position of the sample S in the X-ray CT image data to be aligned with the coordinate position in the abnormality data of the sample S. Furthermore, by repeating the cutting and observation steps alternately multiple times in this manner, abnormality data containing three-dimensional information about the interior of the sample S can be generated.

[0060] Back to Figure 3 In step ST4, the operator associates the X-ray CT image data of the sample S generated in step ST2 with the abnormality data of the sample S generated in step ST3 and stores them in the training data storage unit 6 as a set of training data.

[0061] As described above, simply analyzing X-ray CT image data can identify the location and general shape of certain abnormalities, but it cannot determine the specific nature of the abnormality. However, the contrast of X-ray CT images of a battery stack is believed to correlate with the specific nature of the abnormality. In other words, it is believed that there is a correlation between the X-ray CT image data of sample S and the abnormality data of the same sample S. Therefore, the model generator 3 uses this correlation between X-ray CT image data and abnormality data as training data to perform machine learning, thereby generating a battery stack abnormality estimation model.

[0062] According to the model generation method and the abnormality estimation system 1 of this embodiment, the following effects are achieved.

[0063] (1) According to the model generation method of this embodiment, an abnormality estimation model is generated by machine learning using X-ray CT image data of a battery stack, i.e., a sample S, and abnormality data of the same sample S as training data. Thus, by inputting X-ray CT image data 7D of a battery stack, i.e., a test object 7, as input data, an abnormality estimation model can be generated that estimates abnormalities in the test object 7. Furthermore, by using the abnormality estimation model using X-ray CT image data 7D of the battery stack as input data, abnormalities in the test object 7 can be estimated non-destructively and rapidly, thereby improving battery performance or yield, thereby contributing to improved energy efficiency.

[0064] (2) According to the model generation method of this embodiment, by using abnormality data including information obtained by observing the cross-section CS of the sample S as training data, the abnormality estimation model can be used to associate the X-ray CT image data 7D of the test subject 7 with abnormalities that cannot be determined by analyzing the X-ray CT image data alone. Thus, according to this embodiment, an abnormality estimation model with high abnormality estimation accuracy can be generated.

[0065] Figure 6A FIG. 1 is a diagram showing an example of an X-ray CT image of a battery stack. Figure 6B This is a diagram showing an example of a SEM image of a battery stack. Figure 6A and Figure 6B Each shows a portion of the battery stack including the Cu current collector foil 61 , the Li layer 62 , and the solid electrolyte layer 63 .

[0066] like Figure 6A As shown in FIG, in the X-ray CT image, an abnormality is confirmed, which is a darker contrast than the surrounding area in the solid electrolyte layer 63 as shown by symbol 64. It can be inferred that the portion shown by symbol 64 has a lower density than the surrounding solid electrolyte layer, but only based on Figure 6AThe X-ray CT image shown in cannot distinguish whether it is pores or Li precipitation. Figure 6B As shown in , SEM images are clearer than X-ray CT images and can also confirm the concavity and convexity in the depth direction of the cross section, so it is possible to determine not only the presence of abnormalities in the parts indicated by symbols 65 and 66, but also their specific contents. Figure 6B In the example shown in FIG, , the portion indicated by reference numeral 65 shows asperities in the depth direction, so the abnormality in the portion indicated by reference numeral 65 can be determined to be a crack. Furthermore, the abnormality in the portion indicated by reference numeral 66 can be determined to be Li precipitation. Thus, in the model generation method of this embodiment, by observing the cross-section CS of the sample S using a scanning electron microscope and generating abnormality data based on the resulting SEM image, the abnormality estimation model can be used to associate the X-ray CT image data 7D of the test object 7 with abnormalities that cannot be easily identified by analyzing the X-ray CT image data alone.

[0067] (3) According to the model generation method of this embodiment, X-ray image data containing information obtained by irradiating a sample S fixed to a columnar jig 92 with X-rays and abnormality data containing information obtained by observing a cut surface CS after the sample S is cut by irradiating the sample S with a focused ion beam B while the sample S is fixed to the jig 92 are used as training data. Specifically, according to this embodiment, the coordinate positions in the X-ray CT image data and the coordinate positions in the abnormality data can be aligned, thereby generating an abnormality estimation model with high accuracy in estimating abnormalities.

[0068] (4) By alternately and repeatedly cutting the sample and observing the cut surface, three-dimensional information about the interior of the sample can be obtained. According to the model generation method of this embodiment, by using abnormal data containing this three-dimensional information as training data, an abnormality estimation model with high anomaly estimation accuracy can be generated.

[0069] (5) According to the abnormality estimation system 1 of this embodiment, the input data accepting unit 2 accepts X-ray CT image data of the battery stack, i.e., the test object 7, as input data. The model generating unit 3 generates an abnormality estimation model through machine learning using the X-ray CT image data of the sample S and the abnormality data as training data. The model estimating unit 4 uses the abnormality estimation model generated by the model generating unit 3 to estimate an abnormality of the test object 7 based on the input data accepted by the input data accepting unit 2. Thus, according to this embodiment, an abnormality of the test object 7 can be estimated non-destructively and quickly, thereby improving battery performance or yield, thereby contributing to improved energy efficiency.

[0070] (6) According to the abnormality estimation system 1 of this embodiment, the input data receiving unit 2 receives X-ray CT image data of the battery stack, i.e., the test object 7, as input data. The model estimation unit 4 uses an abnormality estimation model generated by machine learning using the X-ray CT image data and abnormality data of the sample S as training data to estimate abnormalities in the test object 7 based on the input data received by the input data receiving unit 2. Thus, according to this embodiment, abnormalities in the test object 7 can be estimated non-destructively and quickly, thereby improving battery performance or yield, thereby contributing to improved energy efficiency.

[0071] While one embodiment of the present invention has been described above, the present invention is not limited thereto and the detailed structure may be appropriately modified within the scope of the gist of the present invention.

[0072] Reference numerals

[0073] 1: Abnormal inference system

[0074] 2: Input data reception department

[0075] 3: Model generation department

[0076] 4: Model estimation department

[0077] 5: Display

[0078] 6: Training data storage

[0079] 7: Test Subject

[0080] 7D: X-ray CT image data

[0081] 81: X-ray CT device

[0082] 82: FIB device

[0083] 83: Observation Device

[0084] S: Sample

[0085] CS: cut surface

[0086] 92: Jig

Claims

1. A model generation method for generating an abnormality estimation model that uses X-ray image data of a battery stack, i.e., a test object, as input and outputs abnormality data of the test object, wherein: The abnormality estimation model is generated by machine learning using X-ray image data of a battery stack, ie, a sample, and abnormality data of the sample as training data.

2. The model generation method according to claim 1, wherein: The abnormality data of the sample includes information obtained by observing a cross-section of the sample after acquiring X-ray image data.

3. The model generation method according to claim 2, wherein: The X-ray image data of the sample includes information obtained by irradiating the sample fixed to the columnar jig with X-rays. The abnormality data of the sample includes information obtained by observing the cut surface after the sample is cut by irradiating the sample with an ion beam while the sample is fixed to the jig.

4. The model generation method according to claim 3, wherein: The internal abnormality information of the sample includes information obtained by alternately and repeatedly cutting the sample and observing the cut surface.

5. An abnormality estimation system, characterized in that: have: an input data receiving unit for receiving X-ray image data of a battery stack, i.e., a test object, as input data; a model generation unit that generates an abnormality estimation model by machine learning using X-ray image data of a battery stack, i.e., a sample, and abnormality data of the sample as training data; and The model estimation unit estimates an abnormality of the test object based on the input data using the abnormality estimation model.

6. An abnormality estimation system, characterized in that: have: An input data receiving unit receives X-ray image data of a battery stack, ie, a test object, as input data; and The model estimation unit estimates an abnormality of the test object based on the input data using an abnormality estimation model generated by machine learning using X-ray image data of a battery stack, i.e., a sample, and abnormality data of the sample as training data.

Images