Battery evaluation device, machine learning device, battery evaluation program, battery evaluation method, machine learning program, and machine learning method

Abstract

The present invention calculates the deterioration state of a battery with high precision from EIS data of the battery, and is provided with: a correlation data storage unit for storing correlation data indicating the correlation between DRT data relating to relaxation time distribution obtained from the EIS data of the battery and the deterioration state of the battery; a DRT data acquisition unit that acquires DRT data of the test specimen battery to be evaluated; and a deterioration state calculation unit that calculates the deterioration state of the specimen on the basis of the DRT data acquired by the DRT data acquisition unit and the related data.

Description

Technical Field

[0001] The present invention relates to a battery evaluation device, a machine learning device, a battery evaluation program, a battery evaluation method, a machine learning program, and a machine learning method. Background Art

[0002] In a wide range of fields such as battery management systems (BMS) in electric vehicles or battery regeneration, evaluating the life of a battery is an important issue. For example, it is also considered to maximize the utilization rate by cycling the use of batteries, such as using a new product battery for automotive use, a battery with a capacitance less than 80% of the new product for power storage business use, and recycling a battery with a capacitance less than 30% of the new product and reusing the electrode material, etc.

[0003] As an index for evaluating the life of a battery, for example, the state of health (SOH: State of Health) or the remaining useful life (RUL: Remaining Useful Life) of the battery is used. SOH can be expressed as the ratio of the current capacitance Capacity of the battery Now to the capacitance Capacity of the new product battery New (Capacity Now / Capacity New ). In addition, RUL can be expressed as the ratio of the current cycle number Cycle number of the battery Now to the cycle number Cycle number when the capacitance of the battery drops to 80% of the new product 80％Capacity (Cycle number Now / Cyclenumber 80％Capacity ).

[0004] Here, the deterioration of the battery is caused by changes in multiple impedances inside the battery (mainly an increase in the resistance value) due to different deterioration causes. The impedance of the battery exhibits different resistance values and phase delays for each frequency of the inflowing current (or applied voltage), and thus is generally explained based on a combination of multiple RC circuits (equivalent circuit model). In addition, EIS (Electrochemical Impedance Spectroscopy) data is a combination of the applied frequency and the real part / imaginary part values (impedance values) of the corresponding impedance.

[0005] And, for example, as shown in Non-Patent Document 1, in a conventional machine learning method for estimating the SOH of a battery using EIS data, there are a method of using a Nyquist plot created from EIS data as an explanatory variable and a method of using a spectral matrix (a matrix composed of frequency and the real part / imaginary part of impedance) as an explanatory variable.

[0006] The method of using the Nyquist diagram as an explanatory variable has a problem in that, since information on the frequency corresponding to each impedance value is lost, it is impossible to learn the characteristics of the time constant of the internal impedance that causes deterioration. In addition, since the resistance value of the RC circuit in the equivalent circuit model corresponds to the radius of the arc on the Nyquist diagram, there is a problem that it is difficult to detect changes in the RC circuit with a relatively small resistance value.

[0007] On the other hand, the method of using the spectral matrix as an explanatory variable is superior to the method of using the Nyquist diagram as an explanatory variable in that it also learns information related to frequency. However, impedance changes caused by battery deterioration occur in a wide frequency range. Therefore, it is difficult to separate measurement noise and impedance changes caused by changes in the ambient temperature during measurement from the measured EIS data, and there is a problem that the estimation accuracy of SOH is not high when learning only information caused by battery deterioration based on limited data. Prior Art Documents Non-Patent Documents

[0008] Non-Patent Document 1: Yunwei Zhang, et al., “Identifying degradation patterns of lithium ion batteries from impedance spectroscopy using machine learning”, NATURE COMMUNICATIONS volume 11, Article number 1706 (2020), published on April 6, 2020 Summary of the Invention Problems to be Solved by the Invention

[0009] Therefore, the present invention has been completed to solve the above problems, and its object is to accurately calculate the degradation state of a battery. Means for Solving the Problems

[0010] That is, the battery evaluation device of the present invention is characterized by including: a correlation data storage unit that stores correlation data indicating the correlation between DRT data related to the relaxation time distribution obtained from the EIS data of a battery and the degradation state of the battery; a DRT data acquisition unit that acquires the DRT data of a test piece battery to be evaluated; and a degradation state calculation unit that calculates the degradation state of the test piece battery based on the DRT data acquired by the DRT data acquisition unit and the correlation data.

[0011] In the case of this battery evaluation device, since the degradation state of the battery is calculated using correlation data indicating the correlation between DRT data related to the relaxation time distribution obtained from the EIS data of the battery and the degradation state of the battery, the degradation state of the battery can be calculated with high precision. That is, in the present invention, instead of directly using the EIS data, which has a complex relationship with the degradation state of the battery, to calculate the degradation state of the battery, the DRT data related to the relaxation time distribution, which has a simple relationship with the degradation state of the battery, is used to calculate the degradation state of the battery. Therefore, the degradation state of the battery can be calculated with high precision. Specifically, the DRT data is a set of the time constants of the RC circuits in the equivalent circuit model and the values corresponding to the resistance components thereof. Different from the Nyquist plot, it can also include the time constant characteristics of the impedance for learning. Furthermore, different from the spectral matrix, since the resistance component corresponding to each time constant is calculated, even a small change in resistance can be easily detected as long as the time constants corresponding to the degradation causes are different. Therefore, the influence of the ambient temperature and the measurement noise can be reduced, and the degradation state of the battery can be calculated with high precision.

[0012] As a specific implementation of the correlation data, preferably, the correlation data is data obtained by machine learning using a data set including the DRT data obtained from the EIS data and the degradation state of the battery.

[0013] As an index for evaluating the life of the battery, the state of health (SOH) or the remaining useful life (RUL) of the battery is mostly used. Therefore, preferably, the degradation state in the correlation data is the SOH indicating the state of health of the battery or the RUL indicating the remaining useful life, and the degradation state calculation unit calculates the SOH or RUL of the test piece based on the DRT data obtained by the DRT data acquisition unit and the correlation data.

[0014] There is a range with poor precision of the relaxation time in the DRT data depending on its frequency range. Therefore, preferably, the correlation data indicates the correlation between the DRT data in a part of the time constant range and the degradation state of the battery. With this structure, the degradation state of the battery can be calculated with high precision using the correlation data as high-precision data. For example, consider removing the data in the low time constant range of 0.1 Hz and using the relaxation time in the time constant range other than this to create the correlation data.

[0015] Specifically, preferably, the relevant data represents the correlation between the peak intensity, peak intensity ratio, peak position, or peak width of the DRT spectrum included in the DRT data and the battery degradation state.

[0016] As a specific embodiment of the DRT data acquisition unit, preferably, the DRT data acquisition unit includes: an EIS data calculation unit that calculates the EIS data of the test piece battery; and a DRT data calculation unit that calculates the DRT data based on the EIS data calculated by the EIS data calculation unit. In addition, the DRT data acquisition unit may include a DRT data calculation unit that acquires EIS data calculated externally and calculates the DRT data based on the acquired EIS data. In addition, the DRT data acquisition unit may acquire DRT data calculated externally.

[0017] Specifically, preferably, the EIS data calculation unit calculates the EIS data based on an alternating current signal input to the test piece battery and the response signal of the test piece battery measured at this time.

[0018] In addition, preferably, the DRT data acquisition unit further includes: an alternating current signal input unit that inputs an alternating current signal to the test piece battery; and a response signal measurement unit that measures the response signal of the test piece battery.

[0019] Preferably, the battery evaluation device of the present invention further includes a parameter calculation unit that calculates parameters representing an equivalent circuit model of the battery based on the DRT data acquired by the DRT data acquisition unit. If it is this structure, the parameters calculated by the parameter calculation unit are associated with the degradation information calculated by the degradation information calculation unit, thereby enabling analysis of the battery life.

[0020] In addition, the machine learning device of the present invention generates a machine learning model for the battery evaluation device, performs machine learning on a data set including DRT data related to the relaxation time distribution obtained from the EIS data of the battery and the degradation state of the battery, and generates a machine learning model representing the correlation between the DRT data and the degradation state of the battery.

[0021] In addition, the battery evaluation program of the present invention is characterized in that the computer is provided with the following functions: a function as a correlation data storage unit for storing correlation data representing the correlation between DRT data related to the relaxation time distribution obtained from the EIS data of the battery and the deterioration state of the battery; a function as a DRT data acquisition unit for acquiring the DRT data of the battery to be evaluated; and a function as a deterioration state calculation unit for calculating the deterioration state of the battery based on the DRT data acquired by the DRT data acquisition unit and the correlation data.

[0022] In addition, the battery evaluation program can be sent electronically or recorded on a program recording medium such as a CD, DVD, or flash drive.

[0023] In addition, the battery evaluation method of the present invention is characterized in that correlation data representing the correlation between DRT data related to the relaxation time distribution obtained from the EIS data of the battery and the deterioration state of the battery is acquired, the DRT data of the battery to be evaluated is acquired, and the deterioration state of the battery is calculated based on the acquired DRT data and the correlation data.

[0024] In addition, the battery evaluation program can be sent electronically or recorded on a program recording medium such as a CD, DVD, or flash drive.

[0025] In addition, the machine learning program of the present invention generates a machine learning model for a battery evaluation device, and is characterized in that the computer is provided with a function of performing machine learning on a data set including DRT data related to the relaxation time distribution obtained from the EIS data of the battery and the deterioration state of the battery, and generating a machine learning model representing the correlation between the DRT data and the deterioration state of the battery.

[0026] In addition, the machine learning method generates a machine learning model for a battery evaluation device, and is characterized in that machine learning is performed on a data set including DRT data related to the relaxation time distribution obtained from the EIS data of the battery and the deterioration state of the battery, and a machine learning model representing the correlation between the DRT data and the deterioration state of the battery is generated.

[0027] According to the present invention configured as described above, the deterioration state of the battery can be calculated with high accuracy based on the EIS data of the battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a diagram schematically showing the structure of a battery evaluation device according to an embodiment of the present invention. Figure 2 is a functional block diagram of the arithmetic device of the above embodiment. Figure 3 Fig. (a) shows a graph of EIS data, and Fig. (b) shows a graph of DRT data. Figure 4 is a functional block diagram of the machine learning device of the above-described embodiment. Figure 5 is a flowchart showing the machine learning method and the battery evaluation method of the above-described embodiment. Figure 6 Fig. (a) shows the simulation results of the conventional example (machine learning model using EIS data), and Fig. (b) shows the simulation results of the present embodiment (machine learning model using DRT data). Figure 7 is a functional block diagram of the arithmetic device of the modified embodiment. Detailed Embodiment

[0029] Hereinafter, an embodiment of the battery evaluation device of the present invention will be described with reference to the drawings. In addition, for any of the following figures, for ease of understanding, appropriate omissions or exaggerated schematic descriptions are made. For the same components, the same reference numerals are given and the description is appropriately omitted.

[0030] <Basic Structure of Battery Evaluation Device 100> As Figure 1 shown, the battery evaluation device 100 of the present embodiment includes: a current input unit 2 that applies a prescribed current input to a test piece battery W (hereinafter simply referred to as test piece W) as an evaluation object; a voltage detection unit 3 that detects an output voltage V(t) of the output terminal of the test piece W; and an arithmetic device 4 that calculates deterioration information of the test piece W based on the input current I(t) and the output voltage V(t) of the test piece W.

[0031] The current input unit 2 can control the input current I(t) applied to the test piece W. In addition, the current input unit 2 can also adjust the frequency of the input current I(t) applied to the test piece W. The current input unit 2 can also charge and discharge the test piece W. In addition, the battery evaluation device 100 may include an AC signal input unit that inputs an AC signal (current or voltage) to the test piece W instead of the current input unit 2.

[0032] The voltage detection unit 3 detects the output voltage V(t) when the input current I(t) is applied by the current input unit 2. These current input unit 2 and voltage detection unit 3 are used for electrochemical impedance spectroscopy (EIS) measurement. In addition, the battery evaluation device 100 may include a response signal measurement unit that measures the response signal of the test piece W instead of the voltage detection unit 3.

[0033] In addition, in the EIS measurement, an AC signal (voltage or current) is input to the battery, and the response signal (current or voltage) of the battery is measured, and EIS data such as impedance is obtained from the ratio of these signals (current / voltage).

[0034] As Figure 2 shown, the arithmetic device 4 includes: a correlation data storage unit 41 that stores correlation data indicating the correlation between data representing the relaxation time distribution obtained from the EIS data of the battery (hereinafter referred to as DRT (Distribution of Relaxation Times) data) and the degradation state of the battery; a DRT data acquisition unit 42 that acquires the DRT data of the test piece W; and a degradation state calculation unit 43 that calculates the degradation state of the test piece W based on the DRT data acquired by the DRT data acquisition unit 42 and the correlation data.

[0035] In addition, the arithmetic device 4 is composed of a computer having a CPU, a memory, an input / output interface, an AD converter, a display device 5 such as a display, etc. And, the arithmetic device 4 functions as the above-described units 41 to 43 by the cooperation of the CPU and peripheral devices based on the battery evaluation program stored in the memory. In addition, the arithmetic device 4 may be composed of a physically integrated computer or may be composed of physically separate computers respectively.

[0036] Hereinafter, each of the units 41 to 43 will be described. The correlation data storage unit 41 stores correlation data indicating the correlation between DRT data related to the relaxation time distribution obtained from the EIS data of the battery and the degradation state of the battery.

[0037] Here, the EIS data is data obtained by EIS measurement, and is the complex impedance of the electrochemical system obtained by performing discrete Fourier transform on the response signal of the battery when an AC signal is input to the battery. As Figure 3 shown in (a) of, for example, this EIS data can be an EIS spectrum represented by a coordinate graph with the real part of the complex impedance on the horizontal axis and the imaginary part of the complex impedance on the vertical axis.

[0038] In addition, the DRT data is the relaxation time distribution (also referred to as the DRT spectrum) obtained by performing DRT conversion on the EIS data as the frequency field. As Figure 3 shown in (b) of, for example, it can be represented by a coordinate graph with the relaxation time (time constant) on the horizontal axis and the value corresponding to the impedance on the vertical axis.

[0039] The related data is data obtained by performing machine learning using a data set including DRT data obtained from EIS data and the degradation state of the battery. In addition, the degradation state in the related data is the state of health of the battery (hereinafter, SOH) or the remaining useful life (hereinafter, RUL). Further, the related data may also be data representing the correlation between the DRT data within a time constant range representing a part of the DRT data and the degradation state of the battery. For example, it is possible to consider removing the data within the low time constant range of 0.1 Hz and using the relaxation time within the other time constant ranges to create the related data.

[0040] In addition, the related data may also be data representing the correlation between the feature data and the degradation state of the battery, where the feature data represents the features of the DRT spectrum included in the DRT data (such as peak intensity (peak height), peak position, peak intensity ratio, or peak width, etc.). Further, the related data may also be data representing the correlation between the parameters (DC resistance R0, inductance L, resistance R, CPE constant (pseudo-capacitance) Q, or CPE coefficient α, etc.) and the degradation state of the battery, where the parameters represent the equivalent circuit model of the battery obtained from the DRT data. In addition, the equivalent circuit model is a circuit model in which a plurality of parallel circuits composed of a resistance R and a CPE (Constant-Phase Element, a virtual component having the characteristics of both capacitance and resistance) and a DC resistance R0 and an inductance L are connected in series.

[0041] The DRT data acquisition unit 42 acquires the DRT data of the test piece W. In the present embodiment, it has: an EIS data calculation unit 42a that calculates the EIS data of the test piece W; and a DRT data calculation unit 42b that calculates the DRT data based on the EIS data calculated by the EIS data calculation unit 42a. In addition, the DRT data acquisition unit 42 may be configured to include an AC signal input unit and / or a response signal measurement unit.

[0042] The EIS data calculation unit 42a calculates the EIS data based on the input current I(t), which is the AC signal input by the current input unit 2, and the output voltage V(t), which is the response signal detected by the voltage detection unit 3.

[0043] Here, the acquisition timing of the EIS data (the timing of EIS measurement) will be described. First, place the test specimen W in a thermostatic bath at a specified temperature and perform constant current (CC) charging and constant voltage (CV) charging. Additionally, the constant current is, for example, 1C. Here, 1C is the magnitude of the current required to fully charge the theoretical capacity of the battery within 1 hour. For example, for a 1Ah battery, it is a charging speed of 1A. After that, starting from the completion of charging, place the test specimen W in an environment at a specified temperature (e.g., 25°C) to reach a stable state, make the temperature of the battery the above-specified temperature, perform EIS measurement, and obtain EIS data. Additionally, the battery may not be placed in the thermostatic bath.

[0044] The DRT data calculation unit 42b converts the EIS data calculated by the EIS data calculation unit 42a through DRT conversion using a radial basis function RBF (Radial Basis Function) such as a Gaussian function as the kernel function to calculate DRT data.

[0045] The degradation state calculation unit 43 calculates the SOH or RUL as the degradation state of the battery based on the DRT data calculated by the DRT data calculation unit 42b and the correlation data.

[0046] Here, when the correlation data is data representing the correlation between the characteristic data and the degradation state of the battery, and the characteristic data represents the characteristics of the spectrum of the DRT data (such as peak intensity, peak intensity ratio, peak position, or peak width, etc.), the degradation state calculation unit 43 extracts the characteristics of the spectrum from the DRT data and calculates the SOH or RUL as the degradation state of the battery based on the extracted characteristic data and the correlation data.

[0047] Additionally, when the correlation data is data representing the correlation between the parameters (DC resistance R0, inductance L, resistance R, CPE constant (pseudo-capacitance) Q, or CPE coefficient α, etc.) and the degradation state of the battery, and the parameter represents the equivalent circuit model of the battery obtained from the DRT data, the degradation state calculation unit 43 calculates the parameters (DC resistance R0, inductance L, resistance R, CPE constant (pseudo-capacitance) Q, or CPE coefficient α, etc.) based on the DRT data, and calculates the SOH or RUL as the degradation state of the battery based on the calculated parameters and the correlation data. Additionally, the parameters (DC resistance R0, inductance L, resistance R, CPE constant (pseudo-capacitance) Q, or CPE coefficient α, etc.) are used to fit the decomposed sub-peaks using the transfer function of the equivalent circuit model to calculate the respective parameters R0, L, R, Q, α of the equivalent circuit model.

[0048] <Machine learning device 10> The machine learning model as the above-mentioned correlation data is generated by the machine learning device 10 shown below.

[0049] The machine learning device 10 generates a machine learning model for the battery evaluation device 100, performs machine learning on a data set including DRT data obtained from the EIS data of the battery and the degradation state of the battery, and generates a machine learning model representing the correlation between the DRT data and the degradation state of the battery.

[0050] Specifically, as Figure 4 shown, the machine learning device 10 includes: a data set receiving unit 10a that receives a data set including DRT data and the degradation state of the battery; and a machine learning unit 10b that performs machine learning on the data set received by the data set receiving unit 10a and generates a machine learning model representing the correlation between the DRT data and the degradation state of the battery.

[0051] The data set receiving unit 10a receives, for example, a data set including a plurality of DRT data and degradation states obtained from batteries with different temperatures or different degradation states.

[0052] The machine learning unit 10b may also extract features of the spectrum of the DRT data (such as peak intensity or peak position, peak intensity ratio or peak width, etc.) and generate a machine learning model representing the correlation between the extracted feature data and the degradation state of the battery. In addition, the machine learning unit may also calculate parameters representing the equivalent circuit model of the battery (DC resistance R0, inductance L, resistance R, CPE constant (pseudo-capacitance) Q, CPE coefficient α, etc.) from the DRT data and generate a machine learning model representing the correlation between the calculated parameters and the degradation state of the battery.

[0053] <Machine Learning Method and Battery Evaluation Method> Refer to Figure 5 A machine learning method and a battery evaluation method using the machine learning device and the battery evaluation device of the present embodiment will be described. In addition, hereinafter, the machine learning method and the battery evaluation method are performed continuously, but they may also be performed separately.

[0054] First, the battery is placed in a thermostat at a specified temperature, and constant current (CC) charging and constant voltage (CV) charging are performed (step S1). In addition, this CC / CV charging can be performed by controlling a current input unit 2 and other charge and discharge devices by a control unit (not shown).

[0055] After that, starting from the completion of charging, the test piece W is placed in an environment at a specified temperature (for example, 25°C) to reach a stable state, the temperature of the battery is set to the above-specified temperature, and EIS measurement is performed to obtain EIS data (step S2). In addition, this EIS measurement can be performed by controlling a current input unit 2 and other charge and discharge devices by a control unit (not shown).

[0056] Next, constant current (CC) discharge or constant voltage (CV) discharge is performed, and the capacitance (battery capacity) output from the battery is measured during this discharge process (step S3). In addition, this CC / CV discharge can be performed by controlling a charging and discharging device such as the current input unit 2 by a control unit (not shown).

[0057] The above steps S1 to S3 are repeated until the capacitance output from the battery during the above discharge process becomes equal to or less than a specified ratio (e.g., 80%) of the value measured for the first time, and EIS data is obtained.

[0058] Next, DRT conversion is performed on the obtained multiple EIS data, and multiple DRT data are calculated (step S4).

[0059] Then, the DRT data and the battery capacity data (SOH or RUL) are used as explanatory variables and target variables respectively, and input into an initially set machine learning model prepared in the machine learning device 10. The machine learning device 10 trains the machine learning model using the input DRT data and battery capacity data (SOH or RUL) (step S5). Through the above, the machine learning device 10 generates a machine learning model representing the correlation between the DRT data and the degradation state of the battery.

[0060] Next, the test piece W is placed in a thermostat at a specified temperature, and constant current (CC) charging and constant voltage (CV) charging are performed (step S6). In addition, this CC / CV charging can be performed by controlling a charging and discharging device such as the current input unit 2 by a control unit (not shown).

[0061] After that, starting from the completion of charging, the test piece W is placed in an environment at a specified temperature (e.g., 25°C) to reach a stable state, the temperature of the battery is made the specified temperature, and EIS measurement is performed. The EIS data calculation unit 42a obtains the EIS data (step S7). In addition, this EIS measurement is performed by controlling the current input unit 2 by a control unit (not shown).

[0062] Next, the DRT data calculation unit 42b performs DRT conversion on the obtained EIS data of the test piece W, and calculates the DRT data of the test piece W (step S8).

[0063] Then, the degradation state calculation unit 43 inputs the calculated DRT data of the test piece W into the machine learning model, and calculates the predicted value of the SOH or RUL of the test piece W (step S9). This predicted value is displayed (output) on a display device 5 such as a display.

[0064] <Effects of the present embodiment> In the battery evaluation device 100 according to the present embodiment, since the deterioration state of the battery is calculated using the correlation data indicating the correlation between the DRT data obtained from the EIS data of the battery and the deterioration state of the battery, the deterioration state of the battery can be calculated with high precision. That is, in the present invention, instead of directly using the EIS data with a complex relationship with the deterioration state of the battery to calculate the deterioration state of the battery, the DRT data with a simple relationship with the deterioration state of the battery is used to calculate the deterioration state of the battery. Therefore, the deterioration state of the battery can be calculated with high precision. Specifically, the DRT data is a set of the time constants of the RC circuits in the equivalent circuit model and the values corresponding to the resistance components thereof. Different from the Nyquist plot, it can also include the time constant characteristics of the impedance for learning. Furthermore, different from the spectral matrix, since the resistance component corresponding to each time constant is calculated, even a small resistance change can be easily detected as long as the time constants corresponding to the deterioration causes are different. Therefore, the influence of the ambient temperature and the measurement noise can be reduced, and the deterioration state of the battery can be calculated with high precision.

[0065] Here, Figure 6 The prediction results of the SOH using the machine learning model (existing example) machine-learned with the EIS data and the prediction results of the SOH using the machine learning model (present embodiment) machine-learned with the DRT data are shown. From these results, it can be seen that, as in the present embodiment, by using the DRT data to calculate the deterioration state of the battery, the deterioration state of the battery can be calculated with high precision.

[0066] <Modified Embodiment of the Present Invention> In addition, the present invention is not limited to the above embodiment.

[0067] For example, as Figure 7 shown, in addition to the deterioration state calculation unit 43, the battery evaluation device 100 may further include a parameter calculation unit 44 that calculates parameters (such as the DC resistance R0, inductance L, resistance R, CPE constant (pseudo-capacitance) Q, or CPE coefficient α) representing the equivalent circuit model of the battery based on the DRT data obtained by the DRT data acquisition unit 42. By calculating the parameters (such as the DC resistance R0, inductance L, resistance R, CPE constant (pseudo-capacitance) Q, or CPE coefficient α) representing the equivalent circuit model in this way, it can be used for the analysis of the internal changes of the battery, and in addition, the life of the battery can be analyzed.

[0068] In addition, it may be configured to perform calibration processing considering the test environment or installation environment (temperature, humidity, air pressure, etc.) of the battery (specimen W) to be evaluated. For example, calibration such as multiplying or adding a coefficient to the deterioration state obtained by the deterioration state calculation unit 43 or the machine learning model may be performed based on the difference between the acquisition environment of the data set for machine learning and the test environment or installation environment of the specimen W.

[0069] In addition, the data set for machine learning may be a data set obtained under different conditions of the ambient environment (e.g., temperature, humidity, or air pressure, etc.) around the battery, in addition to the data sets obtained at different temperatures and different deterioration states of the battery.

[0070] Furthermore, the machine learning device 10 may be configured to have an EIS data calculation unit for calculating EIS data, may be configured to have a DRT data calculation unit for calculating DRT data based on the EIS data, or may be configured to have a data set generation unit for generating a data set including DRT data and the deterioration state.

[0071] The AC signal that can be input in the EIS measurement of the battery evaluation method or machine learning method in the above-described embodiment may also use an M-sequence signal. In this case, a charge-discharge device such as the power input unit 2 has an M-sequence signal generation unit. By this M-sequence signal generation unit, according to the set measurement frequency range, it is divided into one or more M-sequences for measurement. In addition, in the case of dividing into multiple M-sequences, the frequency range is divided by 1000 times each time from the lower limit of the set frequency to the upper limit of the frequency range. For example, if the measurement frequency range is 1 mHz to 10 kHz, it is considered to be divided into three M-sequence signals, which are respectively divided into 1 mHz to 1 Hz, 1 Hz to 1 kHz, and 1 kHz to 10 kHz. By using the M-sequence signal in this way, compared with the case of measuring impedance using an AC signal that is a single sine wave, the time for EIS measurement can be significantly shortened (about 1 / 4).

[0072] In addition, the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the gist thereof, of course. Industrial Applicability

[0073] According to the present invention, the deterioration state of the battery can be calculated with high accuracy based on the EIS data of the battery. Description of Reference Numerals:

[0074] 100: Battery evaluation device; W: Specimen; 41: Related data storage unit; 42: DRT data acquisition unit; 43: Deterioration state calculation unit; 42a: EIS data calculation unit; 42b: DRT data calculation unit; 44: Parameter calculation unit; 10: Machine learning device.

Claims

1. A battery evaluation device, comprising: A related data storage unit that stores related data, where the related data represents the correlation between DRT data related to the relaxation time distribution obtained from the EIS data of the battery and the degradation state of the battery; A DRT data acquisition unit that acquires the DRT data of the test-piece battery to be evaluated; And A degradation state calculation unit that calculates the degradation state of the test-piece battery based on the DRT data acquired by the DRT data acquisition unit and the related data.

2. The battery evaluation device according to claim 1, wherein The related data is data obtained by performing machine learning using a data set that includes the DRT data obtained from the EIS data and the degradation state of the battery.

3. The battery evaluation device according to claim 1 or 2, wherein The degradation state in the related data is SOH representing the state of health of the battery or RUL representing the remaining useful life, The degradation state calculation unit calculates the SOH or RUL of the test-piece based on the DRT data acquired by the DRT data acquisition unit and the related data.

4. The battery evaluation device according to any one of claims 1 to 3, wherein The related data represents the correlation between the DRT data within a part of the time constant range and the degradation state of the battery.

5. The battery evaluation device according to any one of claims 1 to 4, wherein The related data represents the correlation between the peak intensity, peak intensity ratio, peak position, or peak width of the DRT spectrum included in the DRT data and the battery degradation state.

6. The battery evaluation device according to any one of claims 1 to 5, wherein The DRT data acquisition unit has: An EIS data calculation unit that calculates the EIS data of the test-piece battery; And A DRT data calculation unit that calculates the DRT data based on the EIS data calculated by the EIS data calculation unit.

7. The battery evaluation device according to claim 6, wherein The EIS data calculation unit calculates the EIS data based on the alternating current signal input to the test-piece battery and the response signal of the test-piece battery measured at this time.

8. The battery evaluation device according to any one of claims 1 to 7, wherein The DRT data acquisition unit further includes: An alternating current signal input unit that inputs an alternating current signal to the test-piece battery; and A response signal measurement unit that measures the response signal of the test-piece battery.

9. The battery evaluation device according to any one of claims 1 to 8, wherein The battery evaluation device further includes a parameter calculation unit that calculates parameters representing the equivalent circuit model of the battery based on the DRT data acquired by the DRT data acquisition unit.

10. A machine learning device that generates a machine learning model for a battery evaluation device, wherein Machine learning is performed on a data set including DRT data related to the distribution of relaxation time obtained from the EIS data of a battery and the degradation state of the battery to generate a machine learning model representing the correlation between the DRT data and the degradation state of the battery.

11. A battery evaluation program, wherein the battery evaluation program causes a computer to have the following functions: a function of a correlation data storage unit that stores correlation data representing the correlation between DRT data related to the distribution of relaxation time obtained from the EIS data of a battery and the degradation state of the battery; a function of a DRT data acquisition unit that acquires the DRT data of a battery to be evaluated; and a function of a degradation state calculation unit that calculates the degradation state of the battery based on the DRT data acquired by the DRT data acquisition unit and the correlation data.

12. A battery evaluation method, wherein correlation data representing the correlation between DRT data related to the distribution of relaxation time obtained from the EIS data of a battery and the degradation state of the battery is acquired, the DRT data of a battery to be evaluated is acquired, and the degradation state of the battery is calculated based on the acquired DRT data and the correlation data.

13. A machine learning program that generates a machine learning model for a battery evaluation device, wherein the machine learning program causes a computer to have a function of performing machine learning on a data set including DRT data related to the distribution of relaxation time obtained from the EIS data of a battery and the degradation state of the battery to generate a machine learning model representing the correlation between the DRT data and the degradation state of the battery.

14. A machine learning method that generates a machine learning model for a battery evaluation device, wherein machine learning is performed on a data set including DRT related to the distribution of relaxation time obtained from the EIS data of a battery and the degradation state of the battery to generate a machine learning model representing the correlation between the DRT data and the degradation state of the battery.

Images