Battery performance estimation method, battery performance estimation device, and battery performance estimation program

Abstract

According to the present invention, the performance of a battery manufactured using carbon as a negative electrode active material is estimated with high accuracy from the physical properties of the carbon.　A method for estimating the performance of a battery manufactured using carbon as a negative electrode active material from the physical properties of the carbon is characterized in that the performance of the battery is estimated using a machine learning model obtained on the basis of training data including the following physical properties (a) and (b) of carbon, and values related to the battery performance measured for the battery manufactured using the carbon as the negative electrode active material. (a) The ratio (ID / IG) of the peak top intensity (ID) of the D band calculated from the Raman spectrum to the peak top intensity (IG) of the G band or a value relating to the ratio (b) The width of the G band calculated from the Raman spectrum or a value relating to the width

Description

Battery performance estimation method, battery performance estimation device, and battery performance estimation program

[0001] The present invention relates to a battery performance estimation method, a battery performance estimation device, and a battery performance estimation program.

[0002] There is a need to reduce the amount of time required for battery development and investment in experimental equipment, as well as to establish benchmarks to determine whether the performance of manufactured batteries is appropriate.

[0003] Therefore, in order to know the state of the battery, attempts have been made to measure the Raman spectrum or X-ray diffraction spectrum of the carbon used as the negative electrode active material and to investigate the degree of graphitization, lattice spacing, and crystallite size of the carbon (Patent Document 1).

[0004] However, when estimating battery performance from the results of the analysis of the degree of graphitization, lattice spacing, and crystallite size of carbon as described above, there is a problem in that it is impossible to eliminate unexpected noise. In addition, it is not clear which specific parameters from the various analytical results of carbon should be used to accurately estimate battery performance, and a method for accurately predicting battery performance has not yet been established.

[0005] Patent No. 6489529

[0006] The present invention has been made in view of the above-mentioned problems, and aims to estimate with high accuracy the performance of a battery manufactured by using carbon as a negative electrode active material, based on the physical property values ​​of the carbon used as the negative electrode active material.

[0007] The present invention was completed only after the inventors, through extensive research to solve the above-mentioned problems, found that by using the above-mentioned (a) and (b) as physical property values ​​of carbon and estimating battery performance using a machine learning model obtained based on training data including these (a) and (b) and values ​​related to battery performance measured for batteries manufactured using the carbon as a negative electrode active material, it is possible to estimate battery performance with a sufficiently higher accuracy than conventional methods.

[0008] That is, the battery performance estimation method according to the present invention is a method for estimating the performance of a battery manufactured using carbon as a negative electrode active material from the physical property values ​​of the carbon, and is characterized in that the battery performance is estimated using a machine learning model obtained based on training data including the following physical property values ​​of the carbon (a) and (b) and values ​​related to the battery performance measured for a battery manufactured using the carbon as a negative electrode active material: (a) the ratio (ID / IG) of the peak top intensity (ID) of the D band to the peak top intensity (IG) of the G band calculated from the Raman spectrum, or a value related to said ratio; and (b) the width of the G band calculated from the Raman spectrum, or a value related to said width.

[0009] In order to estimate the battery performance from multiple angles with high accuracy, it is preferable that the physical property data further include (c) a value relating to the lattice spacing in the C-axis direction calculated from the X-ray diffraction spectrum.

[0010] In one specific embodiment of the present invention, the (c) is obtained by reducing the dimension of a plurality of variables calculated from an X-ray diffraction spectrum.

[0011] An example of the battery performance estimated by the present invention is the charge / discharge capacity or C-rate characteristics of the battery.

[0012] The teaching data may further include values ​​relating to the environmental temperature and the number of charge / discharge cycles of the battery in addition to the above-mentioned values ​​relating to the physical properties of carbon or the battery performance.

[0013] According to the present invention, it is possible to estimate with high accuracy the performance of a battery manufactured using carbon used as a negative electrode active material, based on the physical property values ​​of the carbon.

[0014] FIG. 1 is a schematic diagram of an overall battery performance estimation device according to one embodiment of the present invention. FIG. 2 is a schematic diagram showing the procedure for estimating battery performance using the battery performance estimation device according to the present embodiment and measurement results. FIG. 3 is a schematic diagram showing a machine learning device according to another embodiment of the present invention. FIG. 4 is a diagram showing experimental results confirming the effectiveness of the battery performance estimation method according to the present invention. FIG. 5 is a diagram showing experimental results confirming the effectiveness of the battery performance estimation method according to the present invention. FIG. 6 is a diagram showing experimental results confirming the effectiveness of the battery performance estimation method according to the present invention.

[0015] An embodiment of the present invention will be described below with reference to the drawings.

[0016] <Configuration of the battery performance estimation device according to this embodiment> As shown in FIG. 1 , the battery performance estimation device 100 according to this embodiment includes, for example, a measurement unit 1 that measures the physical property values ​​of carbon, and an information processing unit 2 that receives the physical property values ​​of carbon output from the measurement unit 1 and estimates the performance of a battery manufactured using the carbon as a negative electrode material.

[0017] The measurement unit 1 includes a Raman spectrometer 11 that measures the Raman spectrum of carbon and an X-ray diffraction unit 12 that measures the X-ray diffraction spectrum of carbon. The Raman spectrometer 11 and the X-ray diffraction unit 12 may be commercially available Raman spectrometers, X-ray diffraction devices, etc. In addition to the above, the measurement unit 1 may also include, for example, a temperature sensor that measures the environmental temperature at the time when the Raman spectrum or X-ray diffraction spectrum of carbon is measured.

[0018] The measurement unit 1 is configured to be able to transmit and receive data to and from the information processing unit 2 via wire or wirelessly.

[0019] The information processing unit 2 is, for example, a general-purpose computer having analog electrical circuits including buffers, amplifiers, etc., digital electrical circuits including a CPU, memory, DSP, etc., and an A / D converter and the like interposed between them.

[0020] The information processing unit 2 is configured to function as a data receiving unit 21 that receives signals (e.g., Raman spectra and X-ray diffraction spectra) output from the Raman spectroscopy unit 11 and the X-ray diffraction unit 12, and an estimation unit 22 that estimates battery performance based on the signals received by the data receiving unit 21, by having the CPU and its peripheral devices cooperate in accordance with a predetermined program stored in memory.

[0021] The information processing unit 2 according to this embodiment also functions as a memory unit 23 that stores and accumulates teacher data created based on data accepted by the data accepting unit 21, and a machine learning model generating unit 24 that generates a machine learning model by performing machine learning based on the teacher data accumulated in the memory unit 23. The estimation unit 22 is configured to estimate a value related to battery performance using the machine learning model generated by the machine learning model generating unit 24.

[0022] <Battery performance estimation method using battery performance estimation device according to this embodiment> A method for estimating battery performance using the battery performance estimation device 100 configured in this manner involves, for example, acquiring training data for generating a machine learning model (S1, S2), creating a machine learning model based on the training data (S3), and further estimating battery performance from the physical property values ​​of an actual carbon sample based on this machine learning model (S4, S5), as shown in Fig. 2. Each of these steps will be described in detail below.

[0023] The training data used in this embodiment is a set of data in which the physical property values ​​of carbon, the physical property values ​​of carbon measured by the measurement unit 1 of a battery manufactured using the carbon as a negative electrode active material, and values ​​related to the performance of a battery actually manufactured using the carbon (also referred to as battery performance data) are linked to each other.

[0024] In this embodiment, the physical property values ​​of carbon are calculated based on the Raman spectrum and the X-ray diffraction spectrum measured by the measurement unit 1. The calculation of the physical property values ​​can be performed, for example, by the data receiving unit 21. In this case, the data receiving unit 21 can be said to be a pre-processing unit.

[0025] In this embodiment, the following (a) to (c) are used as the physical property values: (a) the ratio (ID / IG) of the peak top intensity of the D band (ID) to the peak top intensity of the G band (IG) calculated from the Raman spectrum, or a value related to said ratio; (b) the width of the G band calculated from the Raman spectrum, or a value related to said width; and (c) a value related to the lattice spacing in the C-axis direction calculated from the X-ray diffraction spectrum.

[0026] The Raman spectrum of carbon obtained by Raman spectroscopy using excitation light with a wavelength of 532 nm contains the D band (1350 cm ‐１ 1370cm or more ‐１ The peaks included in the G band (1570 cm ‐１ 1620cm or more ‐１ It is known that the D band contains peaks included in the following bands. The ratio (ID / IG) of the maximum value of the peak intensity included in these D bands (also referred to as peak top intensity or ID) to the maximum value of the peak intensity included in the G band (also referred to as peak top intensity or IG), and the bandwidth of the G band (the half width of the peak present in the G band) are known to be indicators of the crystallinity of carbon, etc.

[0027] Specific examples of values ​​relating to the lattice spacing in the C-axis direction calculated from the X-ray diffraction spectrum include d (002), d (004), d (006), L (002), L (004), L (006), L (110) and L (112), which are values ​​relating to the lattice constant (d) and crystallite size (L) obtained by X-ray diffraction. In this embodiment, rather than using the physical values ​​relating to these lattice constants as they are, the multicollinearity existing between the values ​​relating to these lattice constants (d) and crystallite size (L) is utilized to perform dimensional reduction on the multiple variables (for example, the eight variables described above) relating to the lattice constant (d) and crystallite size (L) to obtain one-dimensional variables, and the values ​​obtained are used as physical values.

[0028] In this embodiment, the values ​​of (a), (b), and (c) are determined as follows. Raman spectroscopy and X-ray diffraction analysis are performed on a certain carbon, and the ratio (ID / IG) and the bandwidth of the G band calculated from the obtained Raman spectroscopy spectrum, along with the eight variables calculated from the X-ray diffraction spectrum, are calculated to obtain a total of 10 physical property values. Meanwhile, values ​​related to the performance of a battery manufactured using this carbon (e.g., the charge capacity of a lithium-ion secondary battery) are measured. This process is repeated for various carbons and batteries using these carbons, and a new axis is set using the Partial Least Square (PLS) method to maximize the covariance between all measured battery performance values ​​and the aforementioned 10 physical property values. The eight-dimensional variables calculated from the X-ray diffraction spectrum are projected onto the new axis set in this way, and the dimension is reduced to a one-dimensional variable, which is then used as the physical property value (c). Furthermore, for the above-mentioned (a) and (b), the variables obtained by projecting each onto the new axis obtained as described above are used as independent physical property values ​​(a) and (b), respectively. Note that, although the PLS method has been described here as a method for setting a new axis that maximizes the covariance between the value related to battery performance and the physical property value, other methods such as Principal Component Regression (PCR) may be used instead of the PLS method.

[0029] Examples of the values ​​related to battery performance include various indices that are generally measured when performing quality control of batteries, such as the charge capacity and irreversible capacity of the battery, input performance of the C rate characteristics, etc. These values ​​related to battery performance may be measured for a battery actually manufactured using the carbon whose physical properties have been measured as the negative electrode active material and input by the user to the information processing unit 2, or may be directly input to the information processing unit 2 from a measuring device that measures values ​​related to battery performance.

[0030] The memory unit 23 stores and accumulates, as training data, a set of data sets containing, as components, the physical property values ​​of the carbon calculated in this manner and values ​​related to the battery performance measured for a battery manufactured using the carbon as a negative electrode active material.

[0031] When the training data is accumulated in the memory unit 23 as described above, the machine learning model generation unit 24 generates a machine learning model regarding the correlation between the physical properties of carbon and battery performance based on the training data accumulated in the memory unit 23.

[0032] After the machine learning model is generated in this manner, the Raman spectrum and X-ray diffraction spectrum of the carbon for which the battery performance is to be estimated are actually measured in the measurement unit, and the data accepting unit 21 that accepts this measurement data calculates the physical property values ​​(a) to (c) of the carbon described above. The physical property values ​​of the carbon calculated in this manner are sent to the estimation unit 22 that estimates the battery performance. The estimation unit 22 estimates the battery performance based on the machine learning model generated by the machine learning model generating unit 24 and the physical property values, and outputs a value related to the battery performance (also referred to as battery performance estimation data) as the estimation result.

[0033] <Effects of this embodiment> According to the battery performance estimation device 100 of this embodiment configured as described above, a machine learning model is generated using the following (a) and (b) as physical property values ​​of carbon, so that values ​​related to battery performance can be estimated with higher accuracy than conventional estimation methods. (a) A value related to the ratio (ID / IG) of the peak top intensity (ID) of the D band to the peak top intensity (IG) of the G band calculated from the Raman spectrum. (b) A value related to the width of the G band calculated from the Raman spectrum.

[0034] Furthermore, in addition to (a) and (b), the following (c) is also used as the physical property value of carbon, so that values ​​related to battery performance can be estimated with even higher accuracy: (c) A value related to the lattice spacing in the C-axis direction calculated from the X-ray diffraction spectrum.

[0035] By using the physical property values ​​of carbon calculated as described above, it is possible to minimize errors in values ​​related to battery performance estimated from these physical property values, thereby making it possible to estimate battery performance more accurately than before.

[0036] Other Embodiments of the Present Invention The present invention is not limited to the above-described embodiments. In the above-described embodiments, values ​​projected onto new axes determined based on the PLS method were used as (a) and (b), respectively. However, the ratio (ID / IG) or the width of the G band without such processing may also be used as (a) and (b). Furthermore, dimension reduction is not necessarily required for (c). It is also possible to use a value related to the lattice spacing in the C-axis direction calculated from the X-ray diffraction spectrum as is, or to reduce the dimension of only a portion of it and use multiple types as the physical property value of (C). Furthermore, (C) does not necessarily have to be used as a physical property value.

[0037] In the above-described embodiment, all battery performance values ​​used as training data were measured under standard conditions, such as 25°C, 1 atmosphere, and a charge rate of 0.1 C. However, for example, the charge capacity retention rate of a lithium-ion secondary battery when charged at a specific charge rate and temperature may be plotted as a function of the charge rate during battery charging, assuming that the charge capacity is 100%. An approximate equation may be derived based on this plot, and one or more coefficients included in this approximate equation may be further included in the training data. Using such training data, the C-rate characteristics of the battery can be estimated. Furthermore, coefficients obtained by gradually changing the temperature during battery charging (i.e., the temperature during battery use, i.e., the ambient temperature) may be similarly derived and included in the training data. If the ambient temperature is included as training data, it is also possible to estimate battery performance values ​​(such as charge capacity) at a certain ambient temperature by inputting information about the ambient temperature in addition to the carbon physical property values.

[0038] For example, in the above-described embodiment, the battery performance estimation device was described as having both a Raman spectroscopy section and an X-ray diffraction section in the measurement unit, but it is sufficient that the device has at least a Raman spectroscopy section, and the X-ray diffraction section is not necessarily a required configuration.

[0039] In the above-described embodiment, the case where the measurement unit directly outputs the Raman spectrum or the X-ray diffraction spectrum measured by the measurement unit to the information processing unit has been described. However, the measurement unit is not necessarily an essential component, and the user may manually input measurement data about carbon that has been measured in advance by another device or already calculated physical property values ​​into the information processing unit.

[0040] Furthermore, measuring equipment such as a temperature sensor for measuring the physical properties of carbon and values ​​related to the measurement conditions may be provided in the measurement unit, or measurements made using independent measuring equipment may be input into a separate information processing device.

[0041] In addition to the components described above, the training data may further include, for example, measurement data that is the basis for the physical properties of carbon, measurement conditions when measuring battery performance (such as information on humidity, model number and year of manufacture of the measuring device, etc.), information on the structure and components of the battery, and battery quality information such as the number of charge / discharge cycles of the battery at the time the battery performance was measured to obtain the training data.

[0042] A part of the information processing unit may be, for example, a machine learning device configured as an independent server device or the like that can communicate with multiple measuring instruments via the Internet, and may collect training data from measuring units or the like used by an unspecified number of users, and distribute measurement values ​​or machine learning models calculated using a machine learning model to each of multiple measuring units.

[0043] In this case, the machine learning device may include, for example, a device as shown in Figure 3, which includes a data receiving unit that receives training data, a memory unit, and a machine learning model generation unit, accumulates training data output from multiple measurement units, generates a machine learning model, and outputs the generated machine learning model for each measurement unit.

[0044] In addition, some or all of the above-described embodiments and modified embodiments may be combined as appropriate, and it goes without saying that various modifications are possible within the scope of the spirit thereof.

[0045] The effects of the battery performance estimation method according to the present invention will be explained below using more specific examples, but the present invention is not limited to these.

[0046] Various commercially available carbon powders were used as the negative electrode active material. Specifically, approximately 20 types of carbon materials suitable for use as negative electrode active materials in lithium secondary batteries were prepared. These included spherical or crushed natural graphite and spherical or crushed artificial graphite, each with different manufacturers, product numbers, and average particle sizes. Raman and X-ray diffraction spectra were measured for each of the carbons used. The Raman spectroscopy analysis was performed using a Horiba, Ltd. microscopic laser Raman spectroscopy system under the following measurement conditions: laser power: 2 mW, exposure time: 60 seconds, and number of accumulations: 2. The X-ray diffraction analysis was performed using a Rigaku Corporation SmartLab system using Cu Kα1 radiation.

[0047] The physical properties (a) and (b) described above for various carbons were calculated from the Raman spectra measured for various carbons. Furthermore, the physical property (c) described above for various carbons was calculated from the X-ray diffraction spectra measured for various carbons. Specifically, as described in the above-described embodiment, a new axis was set that maximizes the covariance between the measured values ​​of battery performance and the physical properties of carbon (eight variables calculated from the X-ray diffraction spectrum, the ratio (ID / IG) calculated from the Raman spectroscopy spectrum, and the width of the G band, a total of 10 variables). All eight variables calculated from the X-ray diffraction spectrum were projected onto the set new axis and reduced to a one-dimensional variable. Alternatively, the eight variables calculated from the X-ray diffraction spectrum were not reduced in dimension, and d(002), d(004), d(006), L(002), L(004), L(006), L(110), and L(112) were used as they were and used as the physical property (c). For (a) and (b), independent values ​​were used for the variables obtained by projecting the variables onto the new axes obtained as described above.

[0048] Lithium-ion secondary batteries were fabricated using the various carbons described above as negative electrode active materials, and performance values ​​of these lithium-ion secondary batteries were measured. The battery performance values ​​measured included charge / discharge capacity and C-rate characteristics. The battery charge / discharge capacity and C-rate characteristics were measured as follows: A mixed powder containing a 1:1 ratio of carbon (the negative electrode active material) and solid electrolyte was compressed into pellets and stacked on a pellet-shaped solid electrolyte layer formed from the solid electrolyte. A test battery cell was then fabricated using lithium metal as the counter electrode. The battery was then measured using a charge / discharge measurement system manufactured by Scribner Associates, Inc. The charge / discharge capacity was measured for three cycles at a 10-hour charge rate (a charge rate requiring 10 hours for charging). The C-rate characteristics were evaluated by varying the charge rate, measuring the charge capacity at each charge rate, and examining the charge capacity retention rate at each charge rate, with the charge capacity at a certain charge rate (here, 0.1 C) defined as 100%.

[0049] The thus measured physical property values ​​(a) to (c) of carbon and values ​​related to battery performance were provided to the machine learning model generation unit as a set of training data, and the machine learning model generation unit was caused to generate a machine learning model.

[0050] Next, a similar data set (referred to as test data) was obtained using the same procedure as for the training data, but for a different type of carbon than that used to obtain the training data.

[0051] Then, the physical property values ​​(a) to (c) of carbon contained in certain test data were provided to the estimation unit, and an investigation was conducted to determine whether the values ​​related to battery performance estimated by the estimation unit based on these physical property values ​​and the above-mentioned machine learning model matched the values ​​related to battery performance contained in the test data. The results are shown in Figures 4 to 7.

[0052] FIG. 4 shows the results of estimating the C-rate characteristics of a battery using (a) and (b) as the physical properties of carbon. The vertical axis of FIG. 4 represents A, one of the coefficients of a cubic equation that is an approximation formula when the C-rate characteristics are plotted against the actual measured values, and the horizontal axis represents the estimated coefficient A. If these are on a line with a slope of 1, this indicates that the actual measured and estimated values ​​of the C-rate characteristics of the battery are consistent. As shown in FIG. 4, when (a) and (b) were used as the physical properties of carbon, the distribution of dots indicating the correlation between the estimated values ​​of the C-rate characteristics of the battery output from the estimation unit and the actual measured values ​​included in the test data was concentrated near the line with a slope of 1, and there was almost no deviation between the estimated and actual measured values. This confirmed that the present invention can estimate the C-rate characteristics of a battery from the physical properties of carbon with high accuracy.

[0053] 5, when (a) and (b) and (c) obtained by reducing the dimension to one dimension were used as the physical property values ​​of carbon, there was almost no error between the battery performance value (battery performance estimation data) output from the estimation unit for the battery charge / discharge capacity and the actual measured values ​​for battery performance included in the test data, and it was confirmed that the battery charge / discharge capacity can also be estimated very accurately from the physical property values ​​of carbon. When these results are compared with a comparative example using only the physical property value (c) of carbon as shown in FIG. 6 (FIG. 6(b) uses d(002) as the physical property value, and FIG. 6(c) uses five variables, d(002), d(004), d(006), L(002), and L(004) as the physical property values), it can be seen that the estimation method according to the present invention (FIG. 6(a)) using (a), (b), and (c) as the physical property values ​​of carbon shows that the distribution is concentrated near a line with a slope of 1, clearly demonstrating improved estimation accuracy.

[0054] Furthermore, when (a), (b), and (c) are all used as the physical property values ​​of carbon, a more detailed study was conducted as shown in FIG. 7. As a result, it was found that the dimension reduction for the physical property (c) was not performed, and d (002), d (004), d (006), L (002), L (004), L (006), L (110) and L (112) were used as they were to perform multiple regression with 10 variables (FIG. 7(b)). It was found that the estimation accuracy was improved when d (002), d (004), d (006), L (002), L (004), L (006), L (110) and L (112) were used as a single variable by dimension reduction (FIG. 7(a)).

[0055] According to the present invention, it is possible to estimate with high accuracy the performance of a battery produced by using carbon as a negative electrode active material, based on the physical properties of the carbon used as the negative electrode active material.

[0056] REFERENCE SIGNS LIST 100: Battery performance estimation device 1: Measurement unit 11: Raman spectroscopy section 12: X-ray diffraction section 2: Information processing unit 21: Data reception section 22: Estimation section 23: Storage section 24: Machine learning model generation section

Claims

1. A method for estimating the performance of a battery manufactured using carbon as a negative electrode active material from the physical property values of the carbon, comprising: estimating the performance of the battery using a machine learning model obtained based on teacher data including the following (a) and (b) which are physical property values of the carbon and a value related to the battery performance measured for the battery manufactured using the carbon as a negative electrode active material. (a) The ratio (ID / IG) of the peak top intensity (ID) of the D band calculated from the Raman spectrum to the peak top intensity (IG) of the G band or a value related to the ratio (b) The width of the G band calculated from the Raman spectrum or a value related to the width

2. The battery performance estimation method according to claim 1, further including the following values as the physical property values. (c) A value related to the lattice spacing in the C-axis direction calculated from the X-ray diffraction spectrum

3. The battery performance estimation method according to claim 2, wherein (c) is obtained by reducing the dimensions of a plurality of variables calculated from the X-ray diffraction spectrum.

4. The battery performance estimation method according to claim 1 or 2, wherein the battery performance is the charge-discharge capacity or C-rate characteristic of the battery.

5. The battery performance estimation method according to claim 1 or 2, wherein the teacher data further includes a value related to the environmental temperature.

6. The battery performance estimation method according to claim 1 or 2, wherein the teacher data further includes a value related to the number of charge-discharge cycles of the battery.

7. A battery performance estimation device including an estimation unit that estimates the performance of a battery manufactured using carbon as a negative electrode active material from the physical property values of the carbon, wherein the estimation unit estimates the performance of the battery using a machine learning model obtained based on teacher data including the following (a) and (b) which are physical property values of the carbon and a value related to the battery performance measured for the battery manufactured using the carbon as a negative electrode active material. (a) The ratio (ID / IG) of the peak top intensity (ID) of the D band calculated from the Raman spectrum to the peak top intensity (IG) of the G band or a value related to the ratio (b) The width of the G band calculated from the Raman spectrum or a value related to the width

8. A battery performance estimation program for estimating the performance of a battery manufactured using carbon as a negative electrode active material from the physical property values of the carbon, A battery performance estimation program that causes a computer to function as an estimation unit for estimating the performance of a battery using a machine learning model obtained based on teacher data including the following (a) and (b) which are physical property values of carbon and values related to the battery performance measured for a battery manufactured using the carbon as a negative electrode active material. (a) The ratio (ID / IG) of the peak top intensity (ID) of the D band calculated from the Raman spectrum to the peak top intensity (IG) of the G band, or a value related to said ratio (b) The width of the G band calculated from the Raman spectrum, or a value related to said width

9. A machine learning device used in a battery performance estimation device for estimating the performance of a battery manufactured using carbon as a negative electrode active material from the physical property values of the carbon, A data reception unit that acquires teacher data including the following (a) and (b) which are physical property values of the carbon and values related to the battery performance measured for a battery manufactured using the carbon as a negative electrode active material, A machine learning device comprising a machine learning model generation unit that generates a machine learning model based on the teacher data acquired by the data reception unit. (a) The ratio (ID / IG) of the peak top intensity (ID) of the D band calculated from the Raman spectrum to the peak top intensity (IG) of the G band, or a value related to said ratio (b) The width of the G band calculated from the Raman spectrum, or a value related to said width

10. A machine learning method used for estimating the performance of a battery manufactured using carbon as a negative electrode active material from the physical property values of the carbon, acquiring teacher data including the following (a) and (b) which are physical property values of the carbon and values related to the battery performance measured for a battery manufactured using the carbon as a negative electrode active material, A machine learning method for generating a machine learning model based on the acquired teacher data. (a) The ratio (ID / IG) of the peak top intensity (ID) of the D band calculated from the Raman spectrum to the peak top intensity (IG) of the G band, or a value related to said ratio (b) The width of the G band calculated from the Raman spectrum, or a value related to said width

11. A program for a machine learning device used in a battery performance estimation device for estimating the performance of a battery manufactured using carbon as a negative electrode active material from the physical property values of the carbon, a data reception unit that acquires teacher data including the following (a) and (b) which are physical property values of the carbon and values related to the battery performance measured for a battery manufactured using the carbon as a negative electrode active material, A machine learning program that causes a computer to exhibit a function as a machine learning model generation unit that generates a machine learning model based on the teacher data acquired by the data reception unit. (a) The ratio (ID / IG) of the peak top intensity (ID) of the D band calculated from the Raman spectrum to the peak top intensity (IG) of the G band or a value related to the ratio (b) The width of the G band calculated from the Raman spectrum or a value related to the width