Spectrum simulation device, learning system, spectrum simulation method, and program

Abstract

To provide a spectrum simulation device, a learning system, a spectrum simulation method, and a program which can easily simulate an NMR spectrum with regard to a cathode material used in a lithium-ion battery.SOLUTION: A spectrum simulation device comprises a local structure acquisition unit 112 and an NMR shift calculation unit 113. The local structure acquisition unit 112 acquires a local structure surrounding each Li atom included in the crystal structure of a cathode material used in a lithium-ion battery. The NMR shift calculation unit 113 calculates an NMR shift with regard to each Li atom by using an NMR shift calculation model 113a that outputs data indicating an NMR shift upon receiving data indicating the local structure surrounding the Li atom, and calculates an NMR shift of the aggregate of Li atoms of the entire crystal structure on the basis of the NMR shift calculated with regard to each Li atom.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to a spectrum simulation device, a learning system, a spectrum simulation method, and a program. [Background technology]

[0002] To improve the performance of cathode materials used in lithium-ion batteries, it is important to understand the local structure within the cathode material. In particular, it is necessary to understand the local structure around the Li atom in the cathode material. The solid-state NMR (Nuclear Magnetic Resonance) spectrum of a Li atom reflects the shape of the occupied atoms around the Li atom, which is useful for understanding the local structure within the cathode material.

[0003] Cited Document 1 discloses a spectrum analyzer that analyzes an observed NMR spectrum. In order to analyze the NMR spectrum, this spectrum analyzer requires preparing a sample and observing the NMR spectrum using an NMR device. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2019-090747 Summary of the Invention [Problem to be solved by the invention]

[0005] If the solid-state NMR spectrum of Li atoms in a cathode material could be obtained by simulation, sample preparation and NMR measurement could be omitted, and the characteristics of the cathode material could be easily improved. However, the spectrum analyzer disclosed in Patent Document 1 analyzes observed NMR spectra, and there is a problem in that the NMR spectrum cannot be analyzed without an observed NMR spectrum.

[0006] The present disclosure has been made in consideration of the above, and aims to provide a spectrum simulation device, a learning system, a spectrum simulation method, and a program that can easily simulate NMR spectra for positive electrode materials used in lithium-ion batteries. [Means for solving the problem]

[0007] In order to achieve the above object, the spectrum simulation device according to the present disclosure comprises: a local structure acquisition unit that acquires the local structure around each Li atom contained in the crystal structure of a positive electrode material used in a lithium ion battery; and an NMR shift calculation unit that, upon receiving data indicating the local structure around the Li atoms, calculates the NMR shift for each Li atom using an NMR shift calculation model that outputs data indicating an NMR shift, and calculates the NMR shift of an aggregate of Li atoms in the entire crystal structure based on the NMR shift calculated for each Li atom.

[0008] The apparatus may further include a broadening unit that corrects the NMR shift calculated by the NMR shift calculation unit using a broadening function to calculate an NMR spectrum.

[0009] a comparison unit that receives data indicating a previously measured NMR spectrum and compares the received measured NMR spectrum with the NMR spectrum calculated by the broadening unit; The comparison unit may output data indicating the difference between the measured NMR spectrum and the NMR spectrum calculated by the broadening unit, or image data in which the measured NMR spectrum and the NMR spectrum calculated by the broadening unit are superimposed.

[0010] In order to achieve the above object, the learning system according to the present disclosure includes: an NMR shift DB storing training data including data showing the local structure around the Li atom and the NMR shift of the Li atom calculated by first-principles calculation; and a learning unit that performs learning to obtain an NMR shift calculation model that outputs data indicating an NMR shift when it receives data indicating the local structure around the Li atom based on the teacher data stored in the NMR shift DB.

[0011] In order to achieve the above object, the spectrum simulation method according to the present disclosure includes: a local structure acquisition step of acquiring a local structure around each Li atom included in a crystal structure of a positive electrode material used in a lithium ion battery; and an NMR shift calculation step of calculating the NMR shift for each Li atom using an NMR shift calculation model that receives data indicating the local structure around the Li atom and outputs data indicating an NMR shift, and calculating the NMR shift of an aggregate of Li atoms in the entire crystal structure based on the NMR shift calculated for each Li atom.

[0012] In order to achieve the above object, the program according to the present disclosure is Computer a local structure acquisition unit that acquires the local structure around each Li atom contained in the crystal structure of the positive electrode material used in lithium-ion batteries; When data indicating the local structure around the Li atoms is received, the NMR shift calculation unit calculates the NMR shift for each Li atom using an NMR shift calculation model that outputs data indicating an NMR shift, and functions as an NMR shift calculation unit that calculates the NMR shift of the collection of Li atoms in the entire crystal structure based on the NMR shift calculated for each Li atom. [Effects of the Invention]

[0013] According to the present disclosure, NMR spectra can be easily simulated for positive electrode materials used in lithium-ion batteries. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a block diagram showing a spectrum simulation device according to an embodiment; [Figure 2] 1 is a diagram illustrating a spectrum simulation device according to an embodiment. [Figure 3] 1 is a flowchart showing a spectrum simulation process 1 according to an embodiment. [Figure 4] FIG. 2 is a diagram showing a crystal structure according to an embodiment. [Figure 5] FIG. 2 is a diagram showing NMR shifts according to an embodiment. [Figure 6] FIG. 1 is a diagram illustrating a learning system according to an embodiment. [Figure 7] 1 is a block diagram showing a learning device according to an embodiment; [Figure 8] FIG. 2 is a diagram showing a crystal structure DB according to an embodiment. [Figure 9] FIG. 2 is a diagram showing an NMR shift DB according to the embodiment. [Figure 10] 10 is a flowchart illustrating a learning process according to an embodiment. [Figure 11] FIG. 10 is a diagram illustrating a spectrum simulation device according to a modified example. [Figure 12] 10 is a flowchart showing a spectrum simulation process 2 according to a modified example. [Figure 13] FIG. 10 is a diagram showing a simulated NMR spectrum according to a modified example. [Figure 14] FIG. 10 is a diagram showing a measured spectrum according to a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0015] A spectrum simulation device, a spectrum simulation method, and a program according to embodiments of the present disclosure will be described below with reference to the drawings.

[0016] The spectrum simulation device 100 according to this embodiment is intended to understand the local structure around Li atoms in a cathode material in order to improve the characteristics of the cathode material used in a lithium ion battery, and receives data indicating the crystal structure of the cathode material used in a lithium ion battery, simulates the NMR (Nuclear Magnetic Resonance) spectrum of an aggregate of Li atoms in the received crystal structure, and outputs the simulated NMR spectrum. Note that the NMR spectrum has NMR shift on the horizontal axis and relative intensity on the vertical axis.

[0017] As shown in FIG. 1, the spectrum simulation device 100 includes a control unit 110 that simulates an NMR spectrum, a receiving unit 120 that receives data, and an output unit 130 that outputs data.

[0018] The control unit 110 has a processor 140 that executes programs, a main memory unit 150 that is used as a work area for the processor 140, and an auxiliary memory unit 160 that stores various data and programs used in the processing of the processor 140. Both the main memory unit 150 and the auxiliary memory unit 160 are connected to the processor 140 via a bus 170.

[0019] Processor 140 includes an MPU (Micro Processing Unit). Processor 140 executes programs stored in auxiliary storage unit 160 to realize various functions of spectrum simulation device 100.

[0020] The main memory unit 150 includes a RAM (Random Access Memory). Programs are loaded into the main memory unit 150 from the auxiliary memory unit 160. The main memory unit 150 is used as a working area for the processor 140.

[0021] The auxiliary storage unit 160 includes a flash memory or a nonvolatile memory, such as an EEPROM (Electrically Erasable Programmable Read-Only Memory). In addition to programs, the auxiliary storage unit 160 stores various data used in the processing of the processor 140. The auxiliary storage unit 160 supplies the processor 140 with data used by the processor 140 in accordance with instructions from the processor 140, and stores the data supplied from the processor 140. The auxiliary storage unit 160 also stores a trained NMR shift calculation model 113a that outputs data indicating an NMR shift when it receives data indicating the local structure around a Li atom.

[0022] The reception unit 120 receives data input by user operation including a mouse, touch panel, or keyboard, and includes a serial port, a USB (Universal Serial Bus) port, and a LAN (Local Area Network) port, and outputs the input data to the processor 140.

[0023] The output unit 130 is an output device including a display or a printer, an output device capable of outputting data to another computer system or a control device, or a combination thereof.

[0024] As shown in Figure 2, the control unit 110 functions as a crystal structure receiving unit 111, a local structure acquisition unit 112, an NMR shift calculation unit 113, and a broadening unit 114 by the processor 140 executing a program stored in the auxiliary memory unit 160.

[0025] The crystal structure receiving unit 111 receives data indicating the crystal structure of a positive electrode material used in a lithium ion battery. When a random solid solution structure is given as the crystal structure, the data indicating the crystal structure may be data indicating a SQS (Special Quasirandom Structure) structure consisting of several thousand atoms.

[0026] The local structure acquisition unit 112 acquires the local structure around each Li atom included in the crystal structure. The range for acquiring the local structure is preferably at least 0.4 nm (4 angstroms) around each Li atom, and more preferably 0.5 nm (5 angstroms). If the range for acquiring the local structure is 0.5 nm around the Li atom, it includes part of the 3rd site around the Li atom, so the range for acquiring the local structure is preferably 0.5 nm around the Li atom. Furthermore, the range of the local structure around the Li atom is preferably the same as the range of the local structure around the Li atom included in the training data used to create the NMR shift calculation model 113a. Furthermore, the local structure around the Li atom may be considered up to 0.5 nm around, and the local structure may be characterized using a one-hot vector for the atomic species at each site. Note that the one-hot vector is a discrete vector expressed as a value of 0 or 1.

[0027] The NMR shift calculation unit 113 receives data indicating the local structure around a Li atom and calculates the NMR shift for each Li atom using an NMR shift calculation model 113a that outputs data indicating the NMR shift. The NMR shift calculation unit 113 then calculates the NMR shift for the entire Li atom ensemble in the crystal structure based on the NMR shifts calculated for each Li atom. The NMR shift calculation model 113a receives data indicating the local structure around a Li atom and outputs data indicating the NMR shift of the Li atom without using first-principles calculation. The NMR shift calculation model 113a is a trained model that outputs the NMR shift of the Li atom with a smaller computational effort than first-principles calculation. First-principles calculation is a calculation method based on quantum mechanics that does not depend on experimental values ​​other than fundamental physical constants. The data indicating the local structure around the Li atom is an explanatory variable, and the data indicating the NMR shift is a target variable.

[0028] The broadening unit 114 corrects the NMR shift calculated by the NMR shift calculation unit 113 using a broadening function and calculates the NMR spectrum. In a stoichiometric compound, there is no uncertainty in the local structure of the Li atom, but broadening occurs in the measured spectrum due to noise from the NMR measurement device, etc. This broadening is not taken into account in the calculated NMR shift, so the NMR shift is corrected using a broadening function. In general, the shape of an NMR shift spectrum is often fitted using a Voigt function, etc., so the broadening function is preferably a Voigt function or a similar function.

[0029] Next, a spectrum simulation process 1 executed by the spectrum simulation device 100 having the above configuration will be described.

[0030] In response to a user's instruction to start the process, the spectrum simulation device 100 starts the spectrum simulation process 1 shown in Fig. 3. The spectrum simulation process 1 executed by the spectrum simulation device 100 will be described below with reference to a flowchart.

[0031] When the spectrum simulation process 1 is started, the crystal structure receiving unit 111 receives data indicating the crystal structure of the positive electrode material used in the lithium ion battery input to the receiving unit 120 (step S101), and stores the data in the main memory unit 150. The positive electrode material is, for example, LiNi 0.95 Co 0.05 The cathode material has a structure in which a Ni layer, an O layer, and a Li layer are stacked, and Ni and Co atoms are randomly dissolved in the Ni layer. When a random solid solution structure is given as the crystal structure, the data showing the crystal structure may be data showing an SQS structure consisting of several thousand atoms.

[0032] Next, the local structure acquisition unit 112 acquires the local structure around each Li atom included in the crystal structure accepted by the crystal structure acceptance unit 111 (step S102) and stores it in the main memory unit 150. The number of Li atoms for which the local structure is acquired is not particularly limited, but is preferably 100 to 1000. The range for acquiring the local structure is at least 0.4 nm around each Li atom, preferably 0.5 nm. For example, when focusing on the Li atom indicated by the arrow in Figure 4, the range in which the local structure is acquired is represented as the interior of a sphere centered on the Li atom. Furthermore, the range of the local structure around the Li atom is preferably the same as the range of the local structure around the Li atom included in the training data used to create the NMR shift calculation model 113a. Furthermore, the local structure around the Li atom is considered up to 0.5 nm around it, and the local structure may be characterized using a one-hot vector for the atomic species at each site. The local structure around a Li atom is expressed by the number of Ni and Co atoms occupying the sites adjacent to one Li atom, for example, 6 for the 1st site, 6 for the 2nd site, and 10 for the 3rd site, with Ni or Co occupying each site. A LiNiO2-based structure has Li atom sites, O atom sites, and Ni atom sites, but in a Co-substituted system, random occupation occurs only between Ni and Co, and the occupation of Li atom and O atom sites does not change at each Li atom site, so only the occupation state of the Ni site may be used as an explanatory variable.

[0033] Next, when the NMR shift calculation unit 113 receives data indicating the local structure around the Li atoms, it calculates the NMR shift of each Li atom using the NMR shift calculation model 113a, which outputs data indicating NMR shifts (step S103). This allows the NMR shift of one Li atom included in the crystal structure to be obtained. For example, when the NMR shift calculation unit 113 receives data indicating the number of Ni and Co atoms occupying the adjacent sites of a certain Li atom as follows, as data indicating the local structure around the Li atoms: Ni 1st 5, Co 1st 1, Ni 2nd 6, Co 2nd 0, Ni 3rd 10, and Co 3rd 0, it outputs data indicating an NMR shift of 611.5 ppm.

[0034] Next, the NMR shift calculation unit 113 determines whether or not the NMR shifts of all the Li atoms have been calculated for the local structure of the Li atoms acquired in step S102 (step S104).

[0035] If it is determined that the NMR shifts of all Li atoms have not been calculated (step S104; No), the NMR shift calculation unit 113 returns to step S103, repeats steps S103 to S104, and calculates the NMR shifts of all Li atoms for the local structure of the Li atoms obtained in step S102.

[0036] When it is determined that the NMR shifts of all Li atoms have been calculated (step S104; Yes), the NMR shift calculation unit 113 sums up the NMR shifts of each Li atom calculated in step S104 by NMR shift to calculate the NMR shift of the aggregate of Li atoms in the entire crystal structure (step S105). As a result, a histogram of the NMR shifts of the aggregate of Li atoms in the entire crystal structure summed by NMR shift is obtained, as shown in FIG.

[0037] Next, the broadening unit 114 corrects the NMR shift calculated by the NMR shift calculation unit 113 using a broadening function and calculates the NMR spectrum (step S106). In a stoichiometric compound, there is no uncertainty in the local structure of the Li atom, but broadening occurs in the measured spectrum due to noise from the NMR measurement device, etc. This broadening is not taken into account in the calculated NMR shift, so the NMR shift is corrected using a broadening function. In general, the shape of the NMR shift spectrum is often fitted using a Voigt function, etc., so a similar function is preferable. This results in the NMR spectrum shown in Figure 13, for example.

[0038] Next, the broadening unit 114 outputs the calculated NMR spectrum (step S107), after which the spectrum simulation process 1 ends.

[0039] Next, we will explain the learning system 200 for obtaining the NMR shift calculation model 113a. It is also possible to calculate NMR spectra using first-principles calculations. However, in element substitution systems and non-stoichiometric compounds involving Li atom elimination, which are important for practical applications, the number of combinations of occupied atom patterns is enormous, making this impractical due to the enormous amount of calculation required. For this reason, we use training data created using first-principles calculations to create the NMR shift calculation model 113a, which outputs the NMR shift of the Li atom with a smaller calculation load than first-principles calculations. As shown in Figure 6, the learning system 200 includes a learning device 300 that performs learning and a data server unit 400 that stores data used for learning. The learning device 300 includes a control unit 310 that performs learning processing, a reception unit 320 that receives input data, an output unit 330 that outputs data, and a communication unit 500 that communicates with the data server unit 400.

[0040] 7, the control unit 310 has a processor 340 that executes the learning process, a main memory unit 350 that is used as a work area for the processor 340, and an auxiliary memory unit 360 that stores various data and programs used in the processing by the processor 340. Both the main memory unit 350 and the auxiliary memory unit 360 are connected to the processor 340 via a bus 370.

[0041] The receiving unit 320, the output unit 330, the processor 340, the main memory unit 350, and the auxiliary memory unit 360 have the same configurations as the receiving unit 120, the output unit 130, the processor 140, the main memory unit 150, and the auxiliary memory unit 160, respectively, provided in the above-mentioned spectrum simulation device 100.

[0042] By executing the program stored in the auxiliary memory unit 360, the control unit 310 functions as an NMR simulation implementation unit 311 that simulates an NMR spectrum using first-principles calculations and a learning unit 312 that performs learning to obtain the NMR shift calculation model 113a, as shown in Figure 6.

[0043] The NMR simulation implementation unit 311 acquires a crystal structure for one simulation from the crystal structure DB 410 and performs a numerical simulation to acquire an NMR shift based on this crystal structure. Since the crystal structure preferably includes the local structure around the Li atom for which the NMR shift is to be predicted, it is preferable to provide it near the composition for which a prediction model is to be constructed. Furthermore, the range of the local structure around the Li atom is preferably the same as the range of the local structure around the Li atom for which the NMR shift is calculated in the spectrum simulation process 1. Furthermore, the local structure may be characterized using a one-hot vector for the atomic species at each site. Next, the NMR simulation implementation unit 311 calculates the NMR shift of the Li atom using first-principles calculations based on the local structure around the Li atom. While the calculation of the NMR shift is not particularly limited, for example, if the positive electrode material contains a 3d transition metal such as Ni, Fe, or Mn, a shift due to the magnetism of the 3d transition metal may be calculated. Next, the NMR simulation implementation unit 311 acquires the local structure including the occupied atoms around the Li atom based on the calculated NMR shift of the Li atom. This allows the calculated NMR shifts of individual Li atoms to be obtained, along with the local structure around the Li atoms (such as the occupied atoms at each site), which serves as a feature of the NMR shift calculation model 113a, a prediction model. The range for acquiring the local structure is at least 0.4 nm around each Li atom, preferably 0.5 nm. If the range for acquiring the local structure is 0.5 nm around the Li atom, this range includes a portion of the 3rd site around the Li atom, so the range for acquiring the local structure is preferably 0.5 nm around the Li atom. Furthermore, a LiNiO2-based Co-substituted system has a Li atom site, an O atom site, and a Ni atom site. The local structure around the Li atom is expressed by the number of Ni and Co atoms occupying adjacent sites to one Li atom, for example, 6 for the 1st site, 6 for the 2nd site, and 10 for the 3rd site, with Ni or Co occupying each site. The NMR simulation implementation unit 311 outputs data indicating the calculated NMR shifts of the Li atoms and the local structure including the occupied atoms around the Li atom to the NMR shift DB 420.

[0044] The learning unit 312 is implemented by the NMR simulation implementation unit 311 and performs learning to obtain the NMR shift calculation model 113a based on the training data stored in the NMR shift DB 420. The NMR shift calculation model 113a may include, for example, an input layer and an output layer, and preferably includes at least one convolutional layer and at least one pooling layer between the input layer and the output layer. When the NMR shift calculation model 113a receives data indicating the local structure around the Li atom stored in the NMR shift DB 420, the learning unit 312 optimizes a function included in the NMR shift calculation model 113a to obtain the NMR shift calculation model 113a that outputs the training data indicating the NMR shift stored in the NMR shift DB 420. The NMR shift calculation model 113a is a model that can obtain the NMR shift from the local structure around the Li atom with less calculation effort than the simulation implemented by the NMR simulation implementation unit 311.

[0045] The communication unit 500 communicates with the data server unit 400 via wired or wireless communication. The communication unit 500 receives a signal from the data server unit 400 and outputs data indicated by this signal to the processor 340. The communication unit 500 also transmits a signal indicating the data output from the processor 340 to the data server unit 400.

[0046] The data server unit 400 includes a crystal structure DB 410 that stores data indicating the crystal structure of the positive electrode material used in the lithium ion battery, and an NMR shift DB 420 that stores data indicating the local structure around the Li atom and data indicating the NMR shift.

[0047] 8, the crystal structure DB 410 stores data indicating compositions for calculating NMR shifts by first-principles calculations and data indicating crystal structures. Here, the crystal structure DB 410 includes data indicating crystal structures containing elements included in the crystal structure to be simulated in the spectrum simulation process 1, and data indicating crystal structures that are the same as or similar to the crystal structure to be simulated. The positive electrode material used in lithium-ion batteries is not particularly limited as long as it is usable as a positive electrode material for lithium-ion batteries, and includes, for example, lithium cobalt oxide, an oxide in which some of the cobalt in lithium cobalt oxide is replaced with nickel or manganese, lithium manganese oxide, and lithium iron phosphate.

[0048] 9, the NMR shift DB 420 stores training data including the NMR shift of the Li atom calculated by first-principles calculation and data indicating the local structure around the Li atom. Note that the NMR shift DB 420 stores the local structure around the Li atom, for example, in the form of a one-hot vector of the atomic species at each site or in the form of a graph structure of the local structure around the Li atom.

[0049] Next, the learning process executed by the learning device 300 having the above configuration will be described.

[0050] In response to a user's instruction to start the process, the learning device 300 starts the learning process shown in Fig. 10. The learning process executed by the learning device 300 will be described below with reference to a flowchart.

[0051] When the learning process starts, the NMR simulation execution unit 311 acquires data indicating a crystal structure for performing one simulation from the crystal structure DB 410 (step S201). This crystal structure preferably includes a local structure around the Li atom for which the NMR shift is to be predicted, and therefore is preferably provided in the vicinity of the composition for which a prediction model is to be constructed.

[0052] Next, the NMR simulation unit 311 calculates the NMR shift of each Li atom by first-principles calculation based on the crystal structure provided in step S201 (step S203). The calculation of the NMR shift is not particularly limited, but for example, when the positive electrode material contains a 3d transition metal such as Ni, Fe, or Mn, the shift due to the magnetism of the 3d transition metal may be calculated.

[0053] Next, the NMR simulation implementation unit 311 acquires the local structure including the occupied atoms around the Li atom based on the calculated NMR shift of the Li atom (step S204). This acquires the local structure around the Li atom (e.g., the occupied atoms at each site) that serves as a feature of the NMR shift calculation model 113a, which is a prediction model, along with the calculated NMR shift of each Li atom. The range for acquiring the local structure is at least 0.4 nm around each Li atom, preferably 0.5 nm. The NMR simulation implementation unit 311 outputs data indicating the NMR shift of the Li atom calculated in step S203 and data indicating the local structure including the occupied atoms around the Li atom acquired in step S204 to the NMR shift DB 420. The NMR shift DB 420 stores the local structure around the Li atom, for example, as a one-hot vector of the atomic species at each site or as a graph structure.

[0054] Next, the NMR simulation execution unit 311 determines whether the NMR shifts of all Li atoms have been calculated (step S205). The amount of data is preferably about 100 Li atoms for multiple regression with a low degree of freedom of the model, and about 1000 Li atoms for neural networks with a high degree of freedom.

[0055] If it is determined that there are still Li atoms whose shifts have not been calculated (step S205; No), the process returns to step S203, and steps S203 to S205 are repeated to calculate the NMR shifts of the Li atoms that have not yet been calculated.

[0056] When it is determined that the NMR shifts of all Li atoms have been calculated (step S205; Yes), the NMR simulation execution unit 311 determines whether or not the amount of data required for learning has been acquired (step S206).

[0057] If it is determined that the amount of data required for learning has not been acquired (step S206; No), the process returns to step S201, and steps S201 to S206 are repeated to calculate the NMR shifts of Li atoms contained in the crystal structure that have not been calculated so far. Through these processes, an NMR shift DB 420 is obtained, which stores the NMR shifts in correspondence with the local structures around the Li atoms.

[0058] If it is determined that the amount of data required for learning has been acquired (step S206; Yes), the learning unit 312 performs learning to obtain an NMR shift calculation model 113a based on the training data stored in the NMR shift DB 420 (step S207). When the NMR shift calculation model 113a receives data indicating a local structure including occupied atoms around the Li atom stored in the NMR shift DB 420, the learning unit 312 optimizes the bias, weight, or function included in the NMR shift calculation model 113a to obtain the NMR shift calculation model 113a that outputs training data indicating the NMR shift of the Li atom stored in the NMR shift DB 420. Furthermore, learning is repeatedly performed using all data stored in the NMR shift DB 420. Algorithms for obtaining the NMR shift calculation model 113a include linear regression, multidimensional function fitting, decision tree, support vector machine, and neural network methods. In either case, the NMR shift calculation model 113a is a model that outputs the NMR shift of a Li atom when it receives data indicating a local structure including occupied atoms around the Li atom, and is a model that outputs the NMR shift of a Li atom with a smaller calculation amount than first-principles calculation. The use of a model with a large degree of freedom, such as a neural network model, can improve prediction accuracy.

[0059] Next, the learning unit 312 outputs the obtained NMR shift calculation model 113a to the data server unit 400 (step S208). As a result, the NMR shift calculation model 113a is stored in the data server unit 400. Thereafter, the learning process ends.

[0060] The spectrum simulation device 100 having the above configuration can easily simulate the NMR spectrum of a cathode material used in a lithium-ion battery by calculating the NMR shift of the Li atom using the NMR shift calculation model 113a obtained by the learning device 300. Specifically, by predicting the NMR shift relative to the local structure using the NMR shift calculation model 113a, which uses the results of first-principles calculations as training data, the NMR spectrum can be simulated accurately while reducing computational costs. Furthermore, since the NMR shift of the Li atom is primarily due to the local structure adjacent to the Li atom, the complexity of the model can be reduced while maintaining accuracy by limiting the range of local structure considered in the NMR shift calculation model 113a to 4 to 0.5 nm adjacent to the Li atom. Furthermore, since broadening occurs in the measured NMR spectrum even in stoichiometric compounds, spectral simulations that reproduce this effect can be performed by correcting the simulated NMR shift using a broadening function.

[0061] In contrast, there is an approach to calculate NMR spectra from first-principles calculations for the local structure around a specific Li atom. However, in element-substituted systems and non-stoichiometric compounds with Li atom elimination, which are important for applications, the local structure patterns around the Li atom become enormously complex, and it is necessary to consider several thousand atoms to obtain sufficient spectral statistics. Handling several thousand atoms in first-principles calculations is difficult in terms of the increased computational cost. For this reason, previous research has used NMR spectrum simulations for crystal structures with more than 1,000 atoms using NMR shifts estimated from experimental results rather than calculations.

[0062] (Variation) In the above embodiment, an example has been described in which the spectrum simulation device 100 outputs a calculated NMR spectrum. The spectrum simulation device 100 may also compare an NMR spectrum corrected using a broadening function with a previously measured NMR spectrum. The NMR spectrum is measured by preparing a sample of a cathode material used in a lithium-ion battery and measuring the prepared sample with an NMR device. In this case, as shown in FIG. 11 , the spectrum simulation device 100 includes a comparison unit 115 that receives data indicating the measured NMR spectrum and compares the measured NMR spectrum with the NMR spectrum corrected using the broadening function.

[0063] This spectrum simulation device 100 executes spectrum simulation process 2 shown in FIG. 12. Steps S301 to S306 of spectrum simulation process 2 are the same as steps S101 to S106 of spectrum simulation process 1 shown in FIG. 3. In step S307, the comparison unit 115 receives a measured NMR spectrum. A sample of this measured NMR spectrum has the same crystal structure as the crystal structure from which the calculated NMR spectrum was derived. Next, the comparison unit 115 compares the NMR spectrum corrected using a broadening function with the measured NMR spectrum (step S308). Next, the comparison unit 115 outputs the result (step S209). The comparison unit 115 may output data indicating the difference between the measured NMR spectrum and the NMR spectrum corrected using the broadening function. The data indicating the difference is obtained by adjusting the main peak intensities of the measured NMR spectrum and the NMR spectrum corrected using the broadening function to be the same, and then calculating the difference between the intensity of the measured NMR spectrum and the intensity of the NMR spectrum corrected using the broadening function at each NMR shift. Specifically, the comparison unit 115 may output data indicating the difference between the NMR spectrum corrected using the broadening function shown in FIG. 13 and the measured NMR spectrum shown in FIG. 14. The comparison unit 115 may also output image data in which the measured NMR spectrum and the NMR spectrum corrected using the broadening function are superimposed. In this case, the main peak intensities of the measured NMR spectrum and the NMR spectrum corrected using the broadening function are adjusted to be the same. Specifically, the comparison unit 115 may output image data in which the measured NMR spectrum shown in FIG. 14 is superimposed on the NMR spectrum corrected using the broadening function shown in FIG. 13.

[0064] The spectrum simulation device 100 is provided with a comparison unit 115, which allows the measured NMR spectrum to be compared with the calculated NMR spectrum as follows. Figure 13 shows an NMR spectrum simulated by the spectrum simulation device 100. This NMR spectrum is obtained by comparing the measured NMR spectrum with the calculated NMR spectrum of LiNiO2, LiNi 0.95 Co 0.05 O2, and LiNi 0.9 Co 0.1 This is an NMR spectrum of a cathode material used in lithium-ion batteries with an O2 crystal structure. In the simulation, a random solid solution of Ni and Co atoms was assumed, and the crystal structure was generated as an SQS structure. The local structure around the Li atom was taken into account up to 0.5 nm, and the local structure was characterized using a one-hot vector for the atomic species at each site. A linear model was also used as the prediction model. It can be seen that many spectra are generated due to the randomness of the Ni and Co arrangement. The peak with the highest intensity corresponds to the NMR shift that occurs in Li atoms with no Co present within 0.5 nm.

[0065] For comparison, Figure 14 shows the spectrum measured with the same composition. In LiNiO2, the randomness of the occupied atoms does not occur, resulting in a spectrum consisting of a single peak. 0.95 Co 0.05 O2, and LiNi 0.9 Co 0.1 In O2, a spectrum consisting of many peaks is obtained due to the randomness of Ni and Co occupancy. Comparing the simulated and measured spectra, LiNi 0.95 Co 0.05 O2, and LiNi 0.9 Co 0.1It can be seen that the shoulder in O2 also occurs in the random solid solution model. On the other hand, the peak near 0 ppm that only appears in the measured spectrum, and the peak near 450 ppm in LiNiO2, are thought to be due to element segregation or a different phase. As described above, by comparing the simulated NMR spectrum with the measured NMR spectrum, it is easy to link the NMR spectrum to the local structure around each Li atom.

[0066] In the above embodiment, the control unit 110 of the spectrum simulation device 100 is configured to include one processor 140, but multiple processors 140 may cooperate to execute the above functions. Furthermore, the control unit 110 may include multiple main storage units 150 and auxiliary storage units 160.

[0067] Spectrum simulation device 100, learning system 200, and learning device 300 can be realized using a normal computer system, rather than a dedicated system. For example, a computer program for executing the above-described operations may be stored and distributed on a computer-readable recording medium (such as a flexible disk, a CD-ROM (Compact Disc Read-Only Memory), or a DVD-ROM (Digital Versatile Disc Read-Only Memory)), and the learning system 1, spectrum simulation device 100, and learning device 300 that execute the above-described processes may be configured by installing the computer program on a computer. Alternatively, the computer program may be stored in a storage device of a server device on a communication network, and downloaded by a normal computer system to configure spectrum simulation device 100, learning system 200, and learning device 300.

[0068] In addition, when the functions of the spectrum simulation device 100, the learning system 200, and the learning device 300 are realized by sharing the functions between the OS and an application program, or by collaboration between the OS and an application program, only the application program portion may be stored on a recording medium or storage device.

[0069] It is also possible to superimpose a computer program on a carrier wave and distribute it via a communication network. For example, the computer program may be posted on a bulletin board system (BBS) on the communication network and distributed via the communication network. The computer program may then be started and executed under the control of the OS in the same way as other application programs, thereby performing the above-mentioned processing.

[0070] The present disclosure allows various embodiments and modifications without departing from the broad spirit and scope of the present disclosure. Furthermore, the above-described embodiments are intended to illustrate the present disclosure and do not limit the scope of the present disclosure. That is, the scope of the present disclosure is defined by the claims, not the embodiments. Various modifications made within the scope of the claims and the meaning of the disclosure equivalent thereto are considered to be within the scope of the present disclosure. [Explanation of symbols]

[0071] 100...spectrum simulation device, 110, 310...control unit, 111...crystal structure reception unit, 112...local structure acquisition unit, 113...NMR shift calculation unit, 113a...NMR shift calculation model, 114...broadening unit, 115...comparison unit, 120, 320...reception unit, 130, 330...output unit, 140, 340...processor, 150, 350...main memory unit, 160, 360...auxiliary memory unit, 170, 370...bus, 200...learning system, 300...learning device, 311...NMR simulation implementation unit, 312...learning unit, 400...data server unit, 410...crystal structure DB, 420...NMR shift DB, 500...communication unit

Claims

1. a local structure acquisition unit that acquires a local structure around each Li atom included in a crystal structure of a positive electrode material used in a lithium ion battery; an NMR shift calculation unit that, upon receiving data indicating a local structure around the Li atom, calculates the NMR shift for each Li atom using an NMR shift calculation model that outputs data indicating an NMR shift, and calculates the NMR shift of an aggregate of Li atoms in the entire crystal structure based on the NMR shift calculated for each Li atom; A spectrum simulation device comprising:

2. a broadening unit that corrects the NMR shift calculated by the NMR shift calculation unit using a broadening function to calculate an NMR spectrum; 2. The spectrum simulation device according to claim 1.

3. a comparison unit that receives data indicating a previously measured NMR spectrum and compares the received measured NMR spectrum with the NMR spectrum calculated by the broadening unit; the comparison unit outputs data indicating the difference between the measured NMR spectrum and the NMR spectrum calculated by the broadening unit, or image data in which the measured NMR spectrum and the NMR spectrum calculated by the broadening unit are superimposed.

3. The spectrum simulation device according to claim 2.

4. an NMR shift DB storing training data including data indicating a local structure around a Li atom and an NMR shift of the Li atom calculated by first-principles calculation; a learning unit that performs learning to obtain an NMR shift calculation model that outputs data indicating an NMR shift when data indicating a local structure around the Li atom is received based on teacher data stored in the NMR shift DB; and A learning system that includes:

5. a local structure acquisition step of acquiring a local structure around each Li atom included in a crystal structure of a positive electrode material used in a lithium ion battery; an NMR shift calculation step of calculating the NMR shift for each Li atom using an NMR shift calculation model that receives data indicating a local structure around the Li atom and outputs data indicating an NMR shift, and calculating the NMR shift of an aggregate of Li atoms in the entire crystal structure based on the NMR shift calculated for each Li atom; A spectrum simulation method comprising:

6. Computer a local structure acquisition unit that acquires the local structure around each Li atom included in the crystal structure of the positive electrode material used in the lithium ion battery; an NMR shift calculation unit that, upon receiving data indicating a local structure around the Li atom, calculates the NMR shift for each Li atom using an NMR shift calculation model that outputs data indicating an NMR shift, and calculates the NMR shift of an aggregate of Li atoms in the entire crystal structure based on the NMR shift calculated for each Li atom; A program that functions as a

Images