Battery safety estimation device and battery safety estimation method

Abstract

A battery safety estimation device and a safety estimation method that can estimate the safety related to the heat generation of a battery based on the design parameters of the battery. The device includes a parameter acquisition unit (10) that acquires the design parameters (11) of the battery, a calculation unit (20) that calculates the voltage behavior of the battery based on the design parameters (11) according to a machine learning-complete logical model (41), and an output unit (30) that outputs the voltage behavior as information related to the safety of the battery related to heat generation.

Description

Technical Field

[0001] The present disclosure relates to a battery safety estimation device and a battery safety estimation method. Background Art

[0002] In recent years, high energy density has been required for secondary batteries used in electronic devices such as mobile phones and laptop computers, and power sources for vehicles. From this perspective, non-aqueous electrolyte secondary batteries capable of achieving high energy density are widely used. A non-aqueous electrolyte secondary battery includes, for example, a positive electrode, a negative electrode, a separator interposed therebetween, and a non-aqueous electrolyte. In order to improve the volumetric efficiency and achieve high energy density, a battery having a structure in which an electrode group formed by winding a positive electrode and a negative electrode with a separator therebetween is known.

[0003] In addition, the deterioration of the battery capacity during repeated charging and discharging of the battery is called the life characteristic, which is one of the important battery characteristics. In addition to the fact that the evaluation of the life characteristic takes a long time, it is necessary to evaluate while separating the deterioration factors under various conditions, and it takes a huge amount of time to improve the characteristic. In recent years, techniques for estimating life characteristics or control techniques for extending the life have been disclosed (for example, Patent Documents 1, 2, and 3).

[0004] On the other hand, as the energy density of non-aqueous electrolyte secondary batteries increases, there is a trade-off between energy density and safety, and the problems related to the safety of the battery become larger.

[0005] Generally, the safety of non-aqueous electrolyte secondary batteries is affected by the thermal stability of materials, the margin of design, and the appropriateness of the production process. The influence on safety depends to a large extent on the thermal stability of the positive electrode active material, particularly the thermal stability against oxygen release from the positive electrode active material in the charged state. In order to improve the thermal stability of the positive electrode active material, research on synthesis processes and material compositions is mostly carried out. In recent years, a technique for material design of an active material with higher thermal stability by means of first-principles calculation has been disclosed (for example, Patent Document 4).

[0006] Prior Art Documents

[0007] Patent Documents

[0008] Patent Document 1: Japanese Patent No. 5561268

[0009] Patent Document 2: International Publication No. 2014 / 155726

[0010] Patent Document 3: Japanese Unexamined Patent Application Publication No. 2013-217897

[0011] Patent Document 4: Japanese Unexamined Patent Application Publication No. 2017-162790 Summary of the Invention

[0012] Problems to be Solved by the Invention

[0013] However, as reported in Patent Document 4, although material design techniques for improving the safety of batteries have been reported, there is no report on estimation techniques for the safety related to battery heating.

[0014] Therefore, in a battery with an unknown design, it is also desirable to implement a battery safety estimation device or the like that can estimate the safety related to battery heating based on the design parameters of the battery.

[0015] Means for Solving the Problems

[0016] To solve the above problems, a battery safety estimation device according to one aspect of the present disclosure includes: a parameter acquisition unit that acquires design parameters of a battery; a calculation unit that calculates the voltage behavior of the battery based on a machine learning-complete logical model according to the design parameters; and an output unit that outputs the voltage behavior as information on the safety related to the heating of the battery.

[0017] In addition, a battery safety estimation method according to one aspect of the present disclosure includes: a parameter acquisition step of acquiring design parameters of a battery; a calculation step of calculating the voltage behavior of the battery based on a machine learning-complete logical model according to the design parameters; and an output step of outputting the voltage behavior as information on the safety related to the heating of the battery.

[0018] Advantages of the Invention

[0019] According to the present disclosure, it is possible to implement a battery safety estimation device or the like that can estimate the safety related to battery heating based on the design parameters of the battery. Brief Description of the Drawings

[0020] Figure 1 FIG. is a diagram for explaining problems in the estimation of safety related to battery heating.

[0021] Figure 2 FIG. is a functional block diagram showing an example of the structure of the battery safety estimation device according to the present embodiment.

[0022] Figure 3 FIG. is a flowchart showing a method for estimating the safety of a battery in the battery safety estimation device according to the present embodiment.

[0023] Figure 4 FIG. is a flowchart showing a method for constructing a machine learning-complete logical model in the battery safety estimation device according to the present embodiment.

[0024] Figure 5 It is a cross-sectional view showing a schematic structure of an evaluation battery for estimating safety.

[0025] Figure 6 It is a graph showing the relationship between the amount of gas generated in the evaluation battery and the elapsed time.

[0026] Figure 7 It is a graph showing the relationship between the voltage in the evaluation battery and the elapsed time.

[0027] Figure 8 It is a graph showing the relationship between the voltage behavior and the gas generation rate in the evaluation battery.

[0028] Figure 9 It is a graph showing the relationship between the estimated voltage behavior and the measured voltage behavior in the evaluation battery.

[0029] Explanation of reference numerals

[0030] 1 Battery case

[0031] 2 Sealing plate

[0032] 3 Insulating gasket

[0033] 4 Electrode group

[0034] 5 Positive electrode

[0035] 5a Positive electrode lead

[0036] 6 Negative electrode

[0037] 6a Negative electrode lead

[0038] 7 Separator

[0039] 8 Insulating ring

[0040] 10 Parameter acquisition unit

[0041] 11 Design parameter

[0042] 20 Calculation unit

[0043] 30 Output unit

[0044] 40 Storage unit

[0045] 41 Logic model

[0046] 42 Regression equation

[0047] 50 Learning data acquisition unit

[0048] 51 Learning design parameter

[0049] 52 Learning voltage behavior data

[0050] 53 Temperature data

[0051] 54 Model construction means

[0052] 60 Learning department

[0053] 100 Safety estimation device Detailed implementation manner

[0054] (Insight that forms the basis of the present disclosure)

[0055] Existing estimation techniques related to batteries are techniques that use measurable voltage, current, temperature, etc., or techniques that estimate the life characteristics of batteries using mapping data measured in advance under various conditions. In addition, it is difficult to apply existing estimation techniques aimed at estimating life characteristics to safety estimation. Therefore, there is a need to realize a technique that can estimate the safety related to the heat generation of a battery in a battery with an unknown combination of designs.

[0056] Figure 1 This is a diagram for explaining the problems in the estimation of the safety related to the heat generation of a battery. As Figure 1 shown, the correlation between the safety related to the heat generation of a battery and the temperature behavior of the battery is high. Therefore, in order to estimate the safety related to the heat generation of a battery, it is direct and general to estimate the temperature behavior.

[0057] Among them, when the inventors estimated the safety related to the heat generation of a battery, they found the following insights. The temperature behavior can be easily measured using a thermocouple or the like. On the other hand, the time resolution of the change in the temperature behavior is poor, and the error factors caused by the environment and measurement conditions are large. Therefore, in a logical model that performs machine learning with the design parameters of the battery as explanatory variables and the temperature behavior of the battery as the target variable, the error factors of the temperature behavior are large, and the temperature behavior cannot be estimated with high precision.

[0058] On the other hand, the gas generation rate of the battery has a high correlation with the temperature behavior of the battery. In addition, the voltage behavior of the battery also has a high correlation with the temperature behavior and the gas generation rate of the battery. Therefore, by estimating the voltage behavior of the battery, the temperature behavior of the battery can be estimated, and further, the stability related to the heat generation of the battery can be estimated. In addition, the time resolution of the change in the voltage behavior of the battery is high, and the measurement is also easy.

[0059] In the present disclosure, there is provided a battery safety estimation device and the like that can estimate the safety related to the heat generation of a battery based on the design parameters of the battery.

[0060] The outline of one aspect of the present disclosure is as follows.

[0061] The safety estimation device for a battery according to one aspect of the present disclosure includes: a parameter acquisition unit that acquires design parameters of the battery; a calculation unit that calculates the voltage behavior of the battery based on a machine learning-complete logical model according to the design parameters; and an output unit that outputs the voltage behavior as information on the safety related to the heat generation of the battery.

[0062] Thereby, the voltage behavior is estimated based on the design parameters of the battery, and the voltage behavior is output as information on the safety of the battery. Thus, the safety estimation device for a battery according to this aspect can estimate the safety related to the heat generation of the battery based on the design parameters of the battery even for a battery with an unknown combination of designs.

[0063] In addition, for example, the safety estimation device further includes a learning data acquisition unit that acquires learning design parameters of a learning battery and learning voltage behavior data indicating the voltage behavior of the learning battery; and a learning unit that constructs the machine learning-complete logical model by using the learning design parameters as explanatory variables, the learning voltage behavior data as target variables, and causing the logical model to perform machine learning.

[0064] Thereby, a machine learning-complete logical model after performing machine learning with the voltage behavior that is easy to measure and has a high time resolution of change as the target variable is constructed. In addition, the correlation between the voltage behavior and the temperature of the battery is high. Thus, the safety estimation device for a battery according to this aspect can construct a machine learning-complete logical model with high estimation accuracy when estimating the safety related to the heat generation of the battery based on the design parameters of the battery.

[0065] In addition, for example, as a method for constructing the machine learning-complete logical model in the machine learning, the learning unit may use the gradient boosting method.

[0066] The machine learning-complete logical model constructed using the gradient boosting method can estimate the voltage behavior of the battery based on the design parameters of the battery with higher estimation accuracy.

[0067] In addition, for example, the calculation unit may calculate the temperature of the battery based on the correlation between the voltage behavior of the learning battery and the temperature of the learning battery, and the output unit further outputs the temperature according to the voltage behavior of the battery calculated by the calculation unit.

[0068] Thereby, the temperature of the battery is output by means of the voltage behavior of the battery calculated based on the design parameters of the battery. Thus, the safety related to the heat generation of the battery can be estimated more directly.

[0069] In addition, for example, among the design parameters, at least one of (i) the size of the electrode, which forms a part of the battery, (ii) the density of the electrode, (iii) the size of the separator, (iv) the amount of the electrolyte, (v) the composition of the material of the electrode or the electrolyte, (vi) the physical properties of the material of the electrode or the electrolyte, and (vii) the capacity of the battery is included.

[0070] Accordingly, among the design parameters of the battery, parameters that contribute significantly to the estimation of the voltage behavior of the battery are included. Therefore, the voltage behavior of the battery can be estimated with higher estimation accuracy based on the design parameters of the battery.

[0071] In addition, a method for estimating the safety of a battery according to one aspect of the present disclosure includes: a parameter acquisition step of acquiring the design parameters of the battery; a calculation step of calculating the voltage behavior of the battery based on a machine learning completed logical model according to the design parameters; and an output step of outputting the voltage behavior as information related to the safety of the battery related to heat generation.

[0072] Accordingly, the voltage behavior of the battery is estimated based on the design parameters of the battery, and the voltage behavior is output as information related to the safety of the battery. Accordingly, the method for estimating the safety of the battery according to this aspect can estimate the safety related to the heat generation of the battery based on the design parameters of the battery even in a battery with an unknown combination of designs.

[0073] In addition, for example, the safety estimation method further includes a learning data acquisition step of acquiring learning design parameters of a learning battery and learning voltage behavior data indicating the voltage behavior of the learning battery; and a learning step of constructing the machine learning completed logical model by using the learning design parameters as explanatory variables, the learning voltage behavior data as target variables, and causing the logical model to perform machine learning.

[0074] Accordingly, a machine learning completed logical model is constructed by performing machine learning with the voltage behavior of the battery, which is easy to measure and has a high time resolution of change, as the target variable. In addition, the voltage behavior of the battery has a high correlation with the temperature of the battery. Accordingly, the method for estimating the safety of the battery according to this aspect can construct a machine learning completed logical model with high estimation accuracy when estimating the safety related to the heat generation of the battery based on the design parameters of the battery.

[0075] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0076] In addition, the embodiments described below all represent general or specific examples. The numerical values, shapes, materials, components, arrangement positions of the components, connection methods, steps, order of steps, etc. shown in the following embodiments are examples and are not intended to limit the present disclosure. In addition, among the components in the following embodiments, the components not described in the independent claims are described as optional components.

[0077] (Embodiment)

[0078] [Structure of Battery Safety Estimation Device]

[0079] Figure 2 It is a functional block diagram showing an example of the structure of the battery safety estimation device 100 of the present embodiment.

[0080] As Figure 2 shown, the safety estimation device 100 of the present embodiment includes a parameter acquisition unit 10, a calculation unit 20, an output unit 30, a storage unit 40, a learning data acquisition unit 50, and a learning unit 60.

[0081] The safety estimation device 100 is a safety estimation device that estimates the safety of a battery. Specifically, the safety estimation device 100 estimates the safety related to the heat generation when a short circuit occurs in the battery.

[0082] The parameter acquisition unit 10 acquires the design parameter 11 of the battery. The parameter acquisition unit 10 is, for example, an input interface such as a keyboard, and acquires the design parameter 11 of the battery by inputting the design parameter 11 of the battery. In addition, the parameter acquisition unit 10 may be a communication interface with the outside, or may acquire the design parameter 11 of the battery by reading the data sheet of the design parameter 11 of the battery.

[0083] The design parameter 11 of the battery is the material, characteristics, shape, size, etc. of each element constituting the battery. As the design parameter 11 of the battery, from the viewpoint of improving the estimation accuracy of the battery safety, for example, (i) the size of the electrode, (ii) the density of the electrode, (iii) the size of the separator, (iv) the amount of the electrolyte, (v) the composition of the material of the electrode or the electrolyte, (vi) the physical properties of the material of the electrode or the electrolyte, and (vii) the capacity of the battery, which are part of the battery.

[0084] The calculation unit 20 receives the design parameter 11 of the battery from the parameter acquisition unit 10. The calculation unit 20 calculates the voltage behavior based on the machine learning complete logic model 41 such as a model formula according to the design parameter 11 of the battery. The voltage behavior is, for example, a value calculated by the average value, deviation, differentiation, or integration of the voltage.

[0085] In addition, furthermore, the calculation unit 20 can also calculate the temperature of the battery based on the regression formula 42 representing the correlation between the voltage behavior of the battery and the temperature of the battery, according to the calculated voltage behavior of the battery. The temperature of the battery is the temperature of the battery after a certain period of time has elapsed since the short circuit occurred, for example, the temperature 10 seconds after the short circuit occurred.

[0086] In addition, the calculation unit 20 can also calculate the pressure change of the battery based on the machine learning-complete logic model 41 according to the design parameters 11 of the battery. The pressure change of the battery refers to, for example, the gas generation rate when the pressure is converted to gas volume by the equation of state.

[0087] The calculation unit 20 can also store the received design parameters 11 and information such as the calculated voltage behavior of the battery in the storage unit 40. For example, information related to the battery including the design parameters 11 and the calculated voltage behavior of the battery are stored in the storage unit 40 as a corresponding table.

[0088] The calculation unit 20 is composed of, for example, a CPU (Central Processing Unit), a RAM (Random Access Memory), and a ROM (Read Only Memory), etc.

[0089] The output unit 30 receives the calculation result of the voltage behavior of the battery from the calculation unit 20. The output unit 30 outputs the voltage behavior of the battery as information related to the safety of the battery's heat generation. The output unit 30 is composed of, for example, a CPU, a RAM, a ROM, and a display, etc., and displays the information related to the safety of the battery's heat generation as an output. In addition, the output unit 30 can also be a structure including a communication interface for sending the information related to the safety of the battery's heat generation as an output to an external storage device or an external terminal.

[0090] In addition, furthermore, the output unit 30 can also output the calculation result of the temperature of the battery. In addition, the output unit 30 can also output the calculation result of the temperature of the battery as information related to the safety of the battery's heat generation.

[0091] In addition, the output unit 30 can also output the machine learning-complete logic model 41 and the regression formula 42 stored in the storage unit 40. In addition, when the design parameters 11 and information such as the calculated voltage behavior of the battery are stored in the storage unit 40, the output unit 30 can also output the stored design parameters 11 and information such as the calculated voltage behavior of the battery.

[0092] The machine learning completed logical model 41 and the regression formula 42 used by the calculation unit 20 are stored in the storage unit 40. In addition, the voltage behavior of the battery and the temperature of the battery calculated by the calculation unit 20 can also be stored in the storage unit 40.

[0093] The machine learning completed logical model 41 is a machine learning completed logical model constructed by performing machine learning on a logical model constructed by the following model construction means. The model construction means includes model construction means such as the gradient boosting method, the support vector regression method, and the random forest regression method. From the viewpoint of improving the estimation accuracy of the battery safety, the machine learning completed logical model 41 is, for example, a machine learning completed logical model after performing machine learning using the gradient boosting method. In the storage unit 40, one machine learning completed logical model 41 can be stored, or multiple machine learning completed logical models 41 obtained by performing machine learning using different model construction means can be stored.

[0094] The regression formula 42 is a regression formula representing the correlation between the voltage behavior of the battery and the temperature of the battery.

[0095] In the storage unit 40, a logical model for performing machine learning in the learning unit 60 described later and a logical model in the middle of machine learning can also be stored.

[0096] In addition, a safety determination criterion corresponding to the voltage behavior of the battery can also be stored in the storage unit 40. The safety determination criterion is, for example, a criterion for determining high safety if the voltage behavior is below a threshold value.

[0097] The storage unit 40 is composed of a rewritable non-volatile memory such as a hard disk drive or a solid state drive.

[0098] The learning data acquisition unit 50 acquires the learning design parameters 51 of the battery and the learning voltage behavior data 52. In addition, temperature data 53 representing the temperature of the battery can also be acquired. The learning data acquisition unit 50 is composed of, for example, an input interface such as a keyboard or a communication interface. In addition, the learning data acquisition unit 50 can also include a voltage sensor such as a voltmeter and a temperature sensor such as a thermocouple or a temperature measuring body, and acquire the learning voltage behavior data 52 and the temperature data 53 measured by the voltage sensor and the temperature sensor. The learning data acquisition unit 50 stores the acquired learning design parameters 51, learning voltage behavior data 52, and temperature data 53 in the storage unit 40. In addition, the learning data acquisition unit 50 also acquires a model construction means 54 for machine learning and sends the model construction means 54 to the learning unit 60.

[0099] The learning design parameter 51 of the battery is the same parameter as the design parameter 11 described above. The learning design parameter 51 is data measured by an actual battery (e.g., an evaluation battery).

[0100] The learning voltage behavior data 52 is measured data representing the voltage behavior of the battery corresponding to the learning design parameter 51 of the battery. The learning voltage behavior data 52 is, for example, measured data such as the voltage behavior after a short circuit occurs.

[0101] The temperature data 53 is measured data representing the temperature of the battery corresponding to the learning design parameter 51 of the battery. The temperature data 53 is, for example, the temperature of the battery after a certain period of time such as 10 seconds after the start of a short circuit.

[0102] As the model construction means 54, generally, model construction means for machine learning is used. Examples of the model construction means 54 can include gradient boosting method, support vector regression method, and random forest regression method, etc. It can be these individual model construction means, or a model construction means that combines multiple models. Among them, from the perspective of improving the estimation accuracy of the battery's safety, the gradient boosting method can also be used.

[0103] The learning unit 60 uses the learning design parameter 51 as the explanatory variable and the learning voltage behavior data 52 as the target variable, and makes the logical model constructed using the obtained model construction means 54 perform machine learning, thereby constructing the machine learning completed logical model 41. The learning unit 60 stores the constructed machine learning completed logical model 41 in the storage unit 40.

[0104] In addition, the learning unit 60 can also derive the regression formula 42 based on the learning voltage behavior data 52 and the temperature data 53. The learning unit 60 stores the derived regression formula 42 in the storage unit 40.

[0105] The learning unit 60 is composed of, for example, a CPU, a RAM, and a ROM, etc.

[0106] The machine learning completed logical model 41 and the regression formula 42 can be constructed by the learning unit 60, or the pre-constructed machine learning completed logical model 41 and the regression formula 42 can be stored in the storage unit 40.

[0107] [Operation of the Battery Safety Estimation Device]

[0108] Next, the operation of the battery safety estimation device 100 configured as described above will be described.

[0109] First, the method by which the battery safety estimation device 100 estimates the safety of the battery will be described. Figure 3It is a flowchart showing the method for estimating the safety of a battery in the battery safety estimation device 100.

[0110] As Figure 3 shown, first, the parameter acquisition unit 10 acquires the design parameters 11 of the battery (S11). The parameter acquisition unit 10 sends the acquired design parameters 11 to the calculation unit 20. The parameter acquisition unit 10 can acquire the design parameters 11 of one battery or the design parameters 11 of multiple batteries respectively.

[0111] Next, based on the design parameters 11 sent from the parameter acquisition unit 10 and the machine learning completed logic model 41 stored in the storage unit 40, the calculation unit 20 calculates the voltage behavior of the battery (S12). The calculation unit 20 sends the calculated voltage behavior of the battery to the output unit 30. In addition, the calculation unit 20 can also store the calculated voltage behavior of the battery in the storage unit 40. When the parameter acquisition unit 10 acquires multiple design parameters 11, the calculation unit 20 calculates the voltage behavior of the battery according to the multiple design parameters 11 respectively.

[0112] The voltage behavior of the battery has a high correlation with the temperature of the battery. Therefore, by calculating the voltage behavior of the battery, the safety related to the heat generation of the battery can be estimated.

[0113] Next, based on the regression formula 42 of the voltage behavior and the temperature of the battery, the calculation unit 20 calculates the temperature of the battery according to the calculated voltage behavior of the battery (S13). The calculation unit 20 sends the calculated temperature of the battery to the output unit 30. The calculation unit 20 can also store the calculated temperature of the battery in the storage unit 40.

[0114] Next, the output unit 30 outputs the voltage behavior of the battery received from the calculation unit 20 as information on the safety related to the heat generation of the battery (S14). The output unit 30 can also directly output the voltage behavior of the battery as information on the safety related to the heat generation of the battery. Or, the output unit 30 can output the information on the safety related to the heat generation of the battery as information based on the voltage behavior of the battery. As the information based on the voltage behavior of the battery, for example, the determination result based on the safety determination criterion can be cited.

[0115] Next, the output unit 30 outputs the temperature of the battery received from the calculation unit 20 (S15). In addition, the order of step S14 and step S15 can be reversed or they can be simultaneous.

[0116] In addition, in step S14, the output unit 30 can also output the temperature of the battery as information on the safety related to the heat generation of the battery instead of the voltage behavior of the battery.

[0117] In this way, the safety estimation device 100 estimates the voltage behavior of the battery that has a high correlation with the temperature of the battery based on the design parameters of the battery. Thereby, the safety related to the heat generation of the battery is estimated. Thus, the safety estimation device 100 can estimate the safety related to the heat generation of the battery based on the design parameters of the battery even in a battery with an unknown combination of designs.

[0118] Next, a method for constructing the machine learning complete logic model 41 of the battery safety estimation device 100 will be described.

[0119] Figure 4 It is a flowchart showing the method of constructing the machine learning complete logic model 41 in the battery safety estimation device 100.

[0120] As Figure 4 shown, first, the learning data acquisition unit 50 acquires the learning design parameters 51 of the battery, the learning voltage behavior data 52 indicating the voltage behavior of the battery, the temperature data 53 of the battery, and the model construction means 54 for machine learning (S21). This battery is, for example, a learning battery prepared for machine learning. The learning data acquisition unit 50 stores the learning design parameters 51, the learning voltage behavior data 52, and the temperature data 53 in the storage unit 40. The learning data acquisition unit 50 acquires, for example, combinations of the learning design parameters 51 and the learning voltage behavior data 52 for 100 or more, more preferably 300 or more learning batteries. The number of learning batteries for acquiring the learning design parameters 51 and the learning voltage behavior data 52 only needs to be a number above the estimation accuracy for the purpose of the constructed machine learning complete logic model 41.

[0121] In addition, the learning data acquisition unit 50 sends the acquired model construction means 54 to the learning unit 60.

[0122] Next, the learning unit 60 constructs a logic model such as a model formula for calculating the learning voltage behavior data 52 based on the learning design parameters 51 by the received model construction means 54. The learning unit 60 uses the learning design parameters 51 and the learning voltage behavior data 52 stored in the storage unit 40 as teacher data, uses the learning design parameters 51 as explanatory variables, and uses the learning voltage behavior data 52 as target variables, and makes the constructed logic model perform machine learning (S22). Thereby, the learning unit 60 constructs the machine learning complete logic model 41 (S23). The learning unit 60, for example, sets parameters in a variety of logic models and comprehensively makes the logic model perform machine learning for combinations of these parameters. The learning unit 60 stores the constructed machine learning complete logic model 41 in the storage unit 40.

[0123] The voltage behavior of the battery is easy to measure and has a high time resolution of change. Therefore, it is useful to use the learning voltage behavior data 52 of the voltage behavior of the battery as the teacher data for constructing the machine learning completed logic model 41 with high estimation accuracy.

[0124] When using the machine learning completed logic model 41 constructed in step S23 to confirm the accuracy of estimating the safety of the battery, for example, the learning design parameters 51 and the learning voltage behavior data 52 in multiple batteries not used for machine learning are used. In this case, for example, the correlation coefficient R between the voltage behavior estimated by the machine learning completed logic model 41 and the learning voltage behavior data 52 of the battery 2 is preferably 0.4 or more, and more preferably 0.7 or more.

[0125] Next, the learning unit 60 derives a regression formula 42 (S24) based on the learning voltage behavior data 52 and the temperature data 53 stored in the storage unit 40. The regression formula 42 is derived, for example, as a regression formula for the linear approximation of the learning voltage behavior data 52 and the temperature data 53. The learning unit 60 stores the constructed regression formula 42 in the storage unit 40.

[0126] In this way, in the safety estimation device 100, the voltage behavior of the battery that is easy to measure and has a high time resolution of change is used as the target variable to construct the machine learning completed logic model. In addition, the voltage behavior of the battery has a high correlation with the temperature of the battery. Thus, when the safety estimation device 100 estimates the safety related to the heat generation of the battery based on the design parameters of the battery, it can construct a machine learning completed logic model with high estimation accuracy.

[0127] (Example)

[0128] Hereinafter, an example of manufacturing an evaluation battery and estimating the safety related to the heat generation of the battery will be described. In addition, the example shown below is an example, and the present disclosure is not limited to the following examples.

[0129] [Structure of the battery]

[0130] First, an evaluation battery for safety estimation will be described. Figure 5 is a cross-sectional view schematically showing the structure of an evaluation battery for safety estimation.

[0131] As Figure 5 shown, the evaluation battery for safety estimation includes a battery case 1, an electrode group 4 housed in the battery case 1, and insulating rings 8 disposed above and below the electrode group 4, respectively. The battery case 1 has an opening at the upper part, and the opening is sealed by a sealing plate 2.

[0132] The electrode group 4 has a structure in which the positive electrode 5 and the negative electrode 6 are wound around the separator 7 multiple times in a spiral shape. A positive electrode lead 5a made of, for example, aluminum is led out from the positive electrode 5, and a negative electrode lead 6a made of, for example, nickel is led out from the negative electrode 6. The positive electrode lead 5a is connected to the sealing plate 2 of the battery case 1. The negative electrode lead 6a is connected to the bottom of the battery case 1. In addition, although not shown, an electrolytic solution is injected into the battery case 1 together with the electrode group 4.

[0133] [Method for manufacturing an evaluation battery]

[0134] Next, the method for manufacturing the evaluation battery will be described. The evaluation battery is manufactured by the following manufacturing method.

[0135] (1) Positive electrode

[0136] For 100 parts by mass of LiNi 1 / 3 Co 1 / 3 Mn 1 / 3 O2, 2 parts by mass of acetylene black, 2 parts by mass of polyvinylidene fluoride, and an appropriate amount of N-methyl-2-pyrrolidone (NMP) are mixed with a mixer to prepare a positive electrode mixture paste. The positive electrode mixture paste is coated on both sides of a current collector sheet made of an Al foil with a thickness of 15 μm, dried, and rolled to obtain a strip-shaped positive electrode. The strip-shaped positive electrode is cut into a size corresponding to the battery case of a cylindrical 18650, and an aluminum lead is welded to obtain a positive electrode for the evaluation battery. The thickness of the positive electrode is 128 μm.

[0137] (2) Negative electrode

[0138] Graphite particles with an average particle size of 20 μm are used as the negative electrode active material. For 100 parts by mass of the negative electrode active material, 1 part by mass of carboxymethyl cellulose as a thickener, 1 part by mass of styrene-butadiene rubber as a binder, and an appropriate amount of pure water are mixed with a mixer to prepare a negative electrode mixture paste. The negative electrode mixture paste is coated on both sides of a current collector sheet made of an electrolytic copper foil with a thickness of 8 μm, dried, and rolled to obtain a strip-shaped negative electrode.

[0139] The coating amount of the negative electrode mixture paste is determined as follows. That is, the negative electrode charge capacity and the positive electrode charge capacity when the fully charged state is specified as 4.2 V satisfy the relationship:

[0140] (Negative electrode charge capacity) / (Positive electrode charge capacity) = 1.1.

[0141] The strip-shaped negative electrode is cut into a size corresponding to the battery case of a cylindrical 18650, and a nickel lead is welded to obtain a negative electrode for the evaluation battery.

[0142] (3) Non-aqueous electrolyte

[0143] The volume ratio of mixed EC (ethylene carbonate): EMC (ethyl methyl carbonate): DMC (dimethyl carbonate) is 1:1:8. LiPF6 is dissolved in the mixture at a concentration of 1.2 mol / L to obtain a non-aqueous electrolyte for evaluating a battery.

[0144] (4) Electrode assembly

[0145] The positive electrode and the negative electrode obtained in the above manner are wound with a separator composed of a microporous polyethylene film with a thickness of 16 μm to form a spiral electrode assembly. The obtained electrode assembly is housed in a cylindrical 18650 battery case, and the negative electrode and the positive electrode leads are connected. Then, the above non-aqueous electrolyte is added to the battery case in such a way that the design capacity is 1.59 - 1.72 g per 1 Ah. After the electrode assembly is impregnated with the non-aqueous electrolyte under vacuum, the battery case is sealed with a sealing plate. The sealing plate for sealing has a safety valve and has the function of operating when the internal pressure of the battery reaches the upper limit value and discharging the gas generated inside.

[0146] After the above process, the Figure 5 cylindrical battery shown is completed, and thus a battery for evaluation is obtained.

[0147] [Method for confirming the initial battery capacity]

[0148] In an environment of 25°C, the obtained battery for evaluation is charged to 4.1 V at a current value equivalent to 0.2C, and then aged in a constant temperature bath at 45°C for 3 days. Then, the charged battery for evaluation is discharged to 3 V at a current value equivalent to 0.2C in an environment of 25°C.

[0149] Furthermore, the discharged battery for evaluation is subjected to constant voltage charging under the conditions of a maximum current value of 0.2C, a charging voltage value of 4.2V, and a charging termination current of 0.05C, and discharged under the conditions of a discharging current of 0.2C and a discharging termination voltage of 3.0V to confirm the initial battery capacity.

[0150] [Method for obtaining design parameters]

[0151] To obtain explanatory variables used in machine learning, using the same method as [Method for fabricating evaluation battery] above, the composition of (I) the positive electrode active material, (II) the negative electrode active material, (III) the (i) thickness, (ii) length, (iii) width, (iv) density, (v) utilization capacity of the active material, (vi) binder composition of the positive electrode plate and the negative electrode plate, (IV) the (i) thickness, (ii) length, (iii) width, (iv) porosity of the separator, and (V) the (i) composition, (ii) amount of the electrolyte as design parameters are changed, and then the evaluation battery is fabricated. In addition, for each fabricated evaluation battery, the initial battery capacity is confirmed as a design parameter by the same method as [Method for confirming initial battery capacity]. Furthermore, in evaluation batteries with the same structure, the depth of charge and the test temperature during the test are changed as test conditions to conduct the evaluation. 480 evaluation batteries are fabricated, and for the 480 evaluation batteries, a data set including battery design parameters with test conditions is obtained.

[0152] [Method for obtaining voltage behavior, temperature, and pressure]

[0153] In the fabricated evaluation battery, constant voltage charging is performed under the conditions of a maximum current value of 0.2C, a charging voltage value of 4.2V, and a charging termination current of 0.05C. Next, nickel leads for voltage measurement are welded to the terminals of the positive electrode and the negative electrode respectively. A thermocouple is fixed near the center in the height direction of the evaluation battery with the welded leads using heat-resistant tape, and the evaluation battery with the fixed thermocouple is placed in the test tank of a nail penetration testing machine. A nail penetration testing machine capable of performing a nail penetration test in a closed test tank equipped with a pressure sensor for measuring the internal pressure is used as the nail penetration testing machine.

[0154] The temperature in the test tank is controlled to be 65°C, and a round nail with an iron shaft diameter of 3mmφ penetrates the evaluation battery at a speed of 80mm / second, thereby causing an internal short circuit. The voltage of the evaluation battery, the temperature of the evaluation battery, and the pressure inside the test tank after the occurrence of the internal short circuit are measured. Furthermore, since the pressure inside the test tank also changes according to the temperature, in order to standardize the values under each evaluation battery and each test condition, the measured pressure inside the test tank is converted into the gas volume at 1 atmosphere and 20°C using the equation of state, and the gas volume generated after the occurrence of the short circuit is calculated. Figure 6 It is a graph showing the amount of gas generated after the occurrence of a short circuit in the evaluation battery. In addition, Figure 7 It is a graph showing the voltage behavior after the occurrence of a short circuit in the evaluation battery. Figure 6 and Figure 7 are graphs when the time when the nail penetrates the evaluation battery and the internal short circuit occurs is set to 0. In addition, in Figure 6 andFigure 7 Among them, data of 480 evaluation batteries produced in the above [Method for obtaining design parameters] are plotted.

[0155] [Method for constructing model formula and calculating estimation accuracy]

[0156] Using the data obtained in [Method for obtaining design parameters] and [Method for obtaining voltage behavior, temperature and pressure], a model formula is constructed through machine learning.

[0157] In machine learning, three model construction methods, namely gradient boosting method, support vector regression method and random forest regression method, are used. Through machine learning using each method, the following logical models are constructed respectively as model formulas according to the three methods. This logical model can estimate and calculate target variables such as the voltage behavior of the battery, the temperature of the battery and the change in the pressure of the battery based on the design parameters including test conditions as explanatory variables. For example, using the gradient boosting method, various parameters such as Learning rate and Max features are set, and comprehensive machine learning is performed on combinations of these parameters. These model construction methods of machine learning are all methods that include the non-linear effect from the design parameters including test conditions as explanatory variables to the target variables, and can efficiently perform learning even in the dataset of 480 evaluation batteries.

[0158] Specifically, the obtained dataset of 480 evaluation batteries is randomly divided into 6 groups. The design parameters including test conditions of the dataset of evaluation batteries belonging to any 5 of these groups, that is, the dataset of 400 evaluation batteries, are used as explanatory variables to be used as teacher data for machine learning, and a model formula of machine learning completion of the logical model completed by machine learning is derived. This machine learning completion logical model estimates the voltage behavior of the battery, the temperature of the battery and the change in the pressure of the battery as target variables through the above machine learning method. In other words, 400 evaluation batteries are used as learning batteries. Then, the dataset of evaluation batteries belonging to the group not used as teacher data, that is, the dataset of 80 evaluation batteries, is used as verification data, and the design parameters including evaluation conditions are substituted into the previously derived model formula of machine learning completion to estimate the target variables. The correlation coefficient R in the linear regression of the estimated target variable and the measured target variable 2 is calculated as the estimation accuracy. In addition, in the following calculation results of the estimation accuracy, the value representing the estimation accuracy in the case of using the model construction method with the highest estimation accuracy among the three model construction methods is shown.

[0159] [Comparative Example 1]

[0160] In Comparative Example 1, the temperature of the battery is used as the target variable to estimate the safety of the battery.

[0161] By the method described in [the method of model construction and calculation method of estimation accuracy], a machine learning-complete model formula with the battery temperature 10 seconds after the short circuit occurs as the target variable is derived, and the estimation accuracy is calculated. The correlation coefficient R between the estimated target variable and the measured target variable 2 is 0.143, and a result indicating low estimation accuracy is obtained.

[0162] As Figure 6 and Figure 7 shown, the time when the volume of the generated gas and the voltage change significantly is within 1 second after the short circuit occurs. In contrast, the change in temperature is small immediately after the short circuit occurs, and it is after 10 seconds that there is a tendency to show differences among the evaluation batteries. Therefore, it is considered that the time resolution of the temperature change is poor and it is not suitable as a target variable for estimating the behavior after the short circuit occurs.

[0163] [Example 1]

[0164] In Example 1, the voltage behavior of the battery is used as the target variable to estimate the safety of the battery.

[0165] (1) Correlation between temperature and pressure

[0166] First, in order to confirm whether measurement data other than the temperature of the battery can be used as substitute data for the temperature of the battery, the correlation between the temperature of the battery and the pressure change of the battery was confirmed. As Figure 6 shown, the amount of gas in the test tank, that is, the pressure, starts to change immediately after the short circuit occurs, so the time resolution is also high. The amount of gas in most of the evaluation batteries changes in the time period from the short circuit occurrence to 1 second later. In addition, since the gas generation amount increases at a substantially constant speed before reaching the maximum value, the change amount of the gas from the start of gas generation to 1 second after the short circuit occurs is used as the data representing the pressure change, and the gas generation speed is calculated. If the measurement results of 480 evaluation batteries are used to calculate the correlation coefficient between the gas generation speed and the temperature 10 seconds after the short circuit occurs, the correlation coefficient R 2 is 0.76, and a result with a high correlation between the gas generation speed and the temperature is obtained. That is, the slower the gas generation speed of the evaluation battery, the lower the temperature of the battery and the safer it is. In order to estimate the safety related to the heat generation of the battery, the gas generation speed of the battery can also be estimated.

[0167] (2) Correlation between voltage and pressure

[0168] Next, the correlation between the time-dependent change in the pressure at the time of short circuit of the battery and the voltage behavior of the battery will be described. As Figure 7As shown, the voltage of the evaluation battery changes significantly immediately after a short circuit occurs, so the time resolution is high. In addition, similar to the gas generation rate, the voltage of most evaluation batteries changes during the period from the occurrence of the short circuit to 1 second later. Therefore, the voltage behavior is calculated using the voltage results from the occurrence of the short circuit to 1 second later. Figure 8 It is a graph showing the relationship between the voltage behavior from the occurrence of the short circuit to 1 second later and the gas generation rate from the occurrence of the short circuit to 1 second later in the evaluation battery. As Figure 8 shown, it shows a high correlation between the calculated voltage behavior up to 1 second later and the gas generation rate up to 1 second later. If the correlation coefficient when calculating the linear regression of the voltage behavior up to 1 second later and the gas generation rate up to 1 second later using the measurement results of 480 evaluation batteries is calculated, the correlation coefficient R 2 is 0.92.

[0169] In addition, if the correlation coefficient when calculating the linear regression of the voltage behavior and the temperature 10 seconds after the occurrence of the short circuit is calculated, the correlation coefficient R 2 is 0.84.

[0170] (3) Calculation of estimation accuracy

[0171] Next, the results of estimating the safety of the battery by using the voltage behavior of the battery as the target variable will be described. Through the method described in [Construction of the model formula and calculation method of estimation accuracy], a model formula completed by machine learning with the voltage behavior of the battery as the target variable is derived, and the estimation accuracy is calculated. Figure 9 It is a graph showing the relationship between the voltage behavior estimated by substituting the design parameters of the verification data into the model formula completed by machine learning in the evaluation battery and the measured voltage behavior.

[0172] As Figure 9 shown, the correlation coefficient R 2 between the estimated voltage behavior and the measured voltage behavior is 0.75, and a result showing high estimation accuracy is obtained. That is, through the model formula derived by machine learning with the voltage behavior of the battery as the target variable, the safety related to the heat generation of the battery can be predicted with high accuracy based on the design parameters of the battery.

[0173] [Example 2]

[0174] In Example 2, the voltage behavior of the battery was used as the target variable, and the model construction means shown in Table 1 was used to estimate the safety of the battery.

[0175] In addition to using the model construction means shown in Table 1, a machine learning-complete model formula with the voltage behavior of the battery as the target variable was derived by the same method as [the method for constructing the model formula and calculating the estimation accuracy], and the estimation accuracy was calculated. The results of the calculated estimation accuracy are shown in Table 1. In Table 1, the model construction means used and the results of the estimation accuracy are recorded in descending order of the estimation accuracy, starting from high accuracy.

[0176]

Table 1

[0177]

[0178] As shown in Table 1, regardless of which model construction means is used, the estimation accuracy is higher than when the temperature at the time of short circuit occurrence is used as the target variable in Comparative Example 1. In addition, it can be seen that when the gradient boosting method is used as the model construction means, the estimation accuracy is higher than when other model construction means are used.

[0179] (Other Embodiments)

[0180] As described above, the battery safety estimation device and the battery safety estimation method of the present disclosure have been described based on the embodiments, but the present disclosure is not limited to these embodiments. As long as it does not deviate from the gist of the present disclosure, various modifications that can be conceived by those skilled in the art implemented in the embodiments, and other methods constructed by combining some of the constituent elements in the embodiments are also included in the scope of the present disclosure.

[0181] For example, in the above embodiment, the safety estimation device 100 includes the learning data acquisition unit 50 and the learning unit 60, and the learning unit 60 constructs the machine learning-complete logic model 41, but is not limited thereto. The safety estimation device 100 may not include the learning data acquisition unit 50 and the learning unit 60, but may acquire the pre-constructed machine learning-complete logic model 41 and store it in the storage unit 40.

[0182] In addition, for example, in the above embodiment, the safety estimation device 100 calculates and outputs the temperature of the battery, but is not limited thereto. The safety estimation device 100 may also be a device that only outputs the voltage behavior of the battery as information on the battery safety.

[0183] In addition, for example, the safety estimation device 100 includes the storage unit 40, but is not limited thereto. The safety estimation device 100 may also communicate with an external server or the like through a communication unit such as a wired or wireless communication interface, and use the external server or the like instead of the storage unit 40.

[0184] In addition, for example, the safety estimation device 100 may also include a verification unit that verifies the estimation accuracy of the machine learning completed logic model 41 using a part of each of the acquired learning design parameters 51 and learning voltage behavior data 52. The verification unit may also verify the estimation accuracy of the constructed machine learning completed logic model 41 Figure 4 after step S23 in. At this time, the safety estimation device 100 may repeatedly perform steps S21 to S23 while changing at least one of the learning design parameters 51, the learning voltage behavior data 52, and the model construction means 54 until a machine learning completed logic model 41 with a certain estimation accuracy is constructed. The verification unit is composed of, for example, a CPU, a RAM, a ROM, and the like.

[0185] In addition, for example, a plurality of machine learning completed logic models 41 may be stored in the storage unit 40, and the calculation unit 20 uses the plurality of machine learning completed logic models 41 to calculate the voltage behavior of the battery corresponding to each machine learning completed logic model 41. At this time, the output unit 30 may output the voltage behavior of the battery calculated by any one of the machine learning completed logic models as information on the safety related to the heat generation of the battery, and from the viewpoint of improving the estimation accuracy, the average value of the voltage behaviors of the battery calculated by each machine learning completed logic model 41 may also be output as information on the safety related to the heat generation of the battery.

[0186] In addition, for example, a method including steps (processes) performed by each component of the battery safety estimation device and the battery safety estimation method constituting the above-described embodiment can also be executed by a computer (computer system). Moreover, the present disclosure can implement the steps included in these methods as a program for causing a computer to execute. Further, the present disclosure can be implemented as a non-volatile computer-readable recording medium such as a CD-ROM recording this program.

[0187] For example, when the above-described embodiment is implemented by a program (software), each step is executed by executing the program by using hardware resources such as a CPU, a memory, and an input / output circuit of a computer. That is, the CPU acquires data from a memory, an input / output circuit, or the like, performs calculations, or outputs the calculation results to a memory, an input / output circuit, or the like, thereby executing each step.

[0188] Industrial Applicability

[0189] According to the battery safety estimation device and the like of the present disclosure, it is possible to estimate the safety related to heat generation even in a battery with an unknown design, and it is suitable for implementing battery control technologies and batteries with higher safety.

Claims

1. A safety estimation device for a battery, wherein, Comprising: A parameter acquisition unit that acquires design parameters of a battery; A calculation unit that completes a logical model based on machine learning and calculates the voltage behavior of the battery according to the design parameters; And An output unit that outputs the voltage behavior as information related to the safety of the battery's heat generation, The machine learning completed logical model is a logical model obtained by performing machine learning with the learning design parameters of the learning battery as explanatory variables and the learning voltage behavior data representing the voltage behavior of the learning battery after a short circuit as target variables.

2. The battery safety estimation device according to claim 1, wherein, Further comprising: A learning data acquisition unit that acquires the learning design parameters and the learning voltage behavior data; And A learning unit that constructs the machine learning completed logical model by using the learning design parameters as explanatory variables, the learning voltage behavior data as target variables, and making a logical model perform machine learning.

3. The battery safety estimation device according to claim 2, wherein The learning unit uses the gradient boosting method as the method for constructing the machine learning completed logical model in the machine learning.

4. The battery safety estimation device according to any one of claims 1 to 3, wherein The calculation unit further calculates the temperature of the battery according to the voltage behavior of the battery calculated by the calculation unit based on the correlation between the learning voltage behavior data and the temperature of the learning battery; The output unit also outputs the temperature.

5. The battery safety estimation device according to any one of claims 1 to 3, wherein Among the design parameters, at least one of (i) the size of the electrode, (ii) the density of the electrode, (iii) the size of the separator, (iv) the amount of the electrolyte, (v) the composition of the material of the electrode or the electrolyte, (vi) the physical properties of the material of the electrode or the electrolyte, and (vii) the capacity of the battery, which are part of the battery, is included.

6. A method for estimating the safety of a battery, wherein, Including: A parameter acquisition step that acquires design parameters of a battery; A calculation step that calculates the voltage behavior of the battery according to the design parameters based on a machine learning completed logical model; And An output step that outputs the voltage behavior as information related to the safety of the battery's heat generation, The machine learning completed logical model is a logical model obtained by performing machine learning with the learning design parameters of the learning battery as explanatory variables and the learning voltage behavior data representing the voltage behavior of the learning battery after a short circuit as target variables.

7. The method for estimating the safety of a battery according to claim 6, wherein, Further including: A learning data acquisition step that acquires the learning design parameters and the learning voltage behavior data; And A learning step that constructs the machine learning completed logical model by using the learning design parameters as explanatory variables, the learning voltage behavior data as target variables, and making a logical model perform machine learning.

Images