Computing device, calculation method, program, and machine learning model generation method

Abstract

This invention provides a computing device, a calculation method, a program, and a machine learning model generation method for inferring lattice volume from X-ray powder diffraction profiles. [Solution] A computing device 100 for inferring lattice volume from an X-ray powder diffraction profile, comprising: an information acquisition unit 110 that acquires information about the X-ray powder diffraction profile; and an inference unit 120 that includes a machine learning model that takes information about the X-ray powder diffraction profile as input and outputs an inferred lattice volume, and infers the lattice volume from the information about the X-ray powder diffraction profile acquired by the information acquisition unit 110.

Description

【Technical Field】 【0001】 The present invention relates to a computing device, a computing method, a program, and a machine learning model generation method for inferring lattice volume from an X-ray powder diffraction profile. 【Background Art】 【0002】 Indexing is a method of calculating the lattice constants (a, b, c, α, β, γ) of a crystal phase from an X-ray diffraction profile, and is used to obtain the lattice constants of a crystal phase not registered in a database. There is a close relationship between the lattice constants and the peak positions of the X-ray diffraction profile, and programs (for example, DICVOL, N-TREOR, ITO, etc.) for calculating the lattice constants from the X-ray diffraction profile are used. 【0003】 For these programs, depending on the symmetry of the crystal system, it has been a problem that they take a long time. Therefore, research on a program for determining lattice constants using AI has been conducted. 【0004】 Patent Document 1 discloses a method for generating a peak position extraction model for extracting peak positions from a measured diffraction pattern obtained by irradiating a powdery crystal with radiation by machine learning. The method described in Patent Document 1 includes a learning data generation step of generating a learning diffraction pattern from a known crystal structure, and a learning step of generating a peak position extraction model by machine learning using the learning diffraction pattern as learning data. 【0005】 In addition, Non-Patent Document 1 describes a technique related to the learning and application of a neural network model that takes an XRD profile as an input and outputs lattice constants. 【Prior Art Documents】 【Patent Documents】 【0006】 【Patent Document 1】 Japanese Patent Application Laid-Open No. 2020-134382 【Non-Patent Documents】 【0007】 [Non-Patent Document 1] Sathya R. Chitturi et. al., Automated prediction of lattice parameters from X-ray powder diffraction patterns, J. Appl. Chryst. (2021). 54, 1799-1810. [Overview of the Initiative] [Problems that the invention aims to solve] 【0008】 The method described in Patent Document 1 only determines the peak position and cannot determine the lattice volume or lattice constants. Furthermore, the method described in Non-Patent Document 1 fails to infer the angular parameters and therefore cannot determine the lattice constants for monoclinic and triclinic grids. 【0009】 The inventors discovered that by using a machine learning model to determine the lattice volume, even a list of d-values ​​from the X-ray powder diffraction profile information alone can be used to obtain the volume with high accuracy. Furthermore, they found that by using the lattice volume as an auxiliary constraint on conventional methods for determining the lattice constant, the time required to determine the lattice constant can be reduced, thus completing the present invention. 【0010】 This invention has been made in view of these circumstances, and aims to provide a computing device, a computing method, a program, and a machine learning model generation method for inferring lattice volume from X-ray powder diffraction profiles. [Means for solving the problem] 【0011】 (1) In order to achieve the above objective, the computing device of the present invention employs the following means. That is, one embodiment of the present invention is a computing device for inferring a lattice volume from an X-ray powder diffraction profile, characterized by comprising: an information acquisition unit that acquires information regarding the X-ray powder diffraction profile; and an inference unit that includes a machine learning model that takes information regarding the X-ray powder diffraction profile as input and outputs an inferred lattice volume, and infers the lattice volume from the information regarding the X-ray powder diffraction profile acquired by the information acquisition unit. 【0012】 (2) Furthermore, in a computing device according to one aspect of the present invention, the information relating to the X-ray powder diffraction profile is characterized in that it is a list of d values ​​of the X-ray powder diffraction profile. 【0013】 (3) Furthermore, in a computing device according to one aspect of the present invention, the machine learning model is characterized in that it is a neural network model. 【0014】 (4) Furthermore, a computing device according to one aspect of the present invention is characterized by further comprising: a range determination unit that determines a search range for the grid volume based on the grid volume inferred by the inference unit; and a grid constant determination unit that determines the grid constant based on the search range for the grid volume. 【0015】 (5) Furthermore, in a computing device according to one aspect of the present invention, the training data for generating the machine learning model is characterized in that it includes information about the crystal system. 【0016】 (6) Furthermore, in a computing device according to one aspect of the present invention, the machine learning model includes a plurality of models, and each of the plurality of models is generated by the training data which includes information on different crystal systems. 【0017】 (7) In addition, in a computing device according to one aspect of the present invention, the inference unit is characterized in that it uses at least one model from the plurality of models to infer the lattice volume for the crystal system corresponding to the model. 【0018】 (8) Further, in the computing device according to one aspect of the present invention, the training data for generating the machine learning model includes a list of the d-values that have been pre-processed to reduce a plurality of peaks with a difference in peak positions of not more than a predetermined threshold to a smaller number of peaks. 【0019】 (9) Further, in the computing device according to one aspect of the present invention, the machine learning model is generated by training data generated based on training data generation conditions. 【0020】 (10) Further, in the computing device according to one aspect of the present invention, the training data does not include information on atomic positions. 【0021】 (11) Further, in the computing device according to one aspect of the present invention, the training data generation conditions include a condition to generate training data having a lattice constant not exceeding a predetermined lattice volume. 【0022】 (12) Further, one aspect of the calculation method of the present invention is a method for inferring a lattice volume from an X-ray powder diffraction profile, including: a step of acquiring information regarding the X-ray powder diffraction profile; and a step of inferring the lattice volume from the acquired information regarding the X-ray powder diffraction profile using a machine learning model that takes the information regarding the X-ray powder diffraction profile as an input and outputs the inferred lattice volume. 【0023】 (13) Further, one aspect of the program of the present invention is a program for inferring a lattice volume from an X-ray powder diffraction profile, causing a computer to execute: a process of acquiring information regarding the X-ray powder diffraction profile; and a process of inferring the lattice volume from the acquired information regarding the X-ray powder diffraction profile using a machine learning model that takes the information regarding the X-ray powder diffraction profile as an input and outputs the inferred lattice volume. 【0024】 (14) Further, the method for generating a machine learning model according to one aspect of the present invention is a method for generating a machine learning model that outputs a lattice volume inferred from information regarding an X-ray powder diffraction profile when information regarding the X-ray powder diffraction profile is input, the method comprising: setting teacher data conditions; setting the structure of the machine learning model; preparing a plurality of teacher data that input information regarding the X-ray powder diffraction profile and output a lattice volume corresponding to the information regarding the X-ray powder diffraction profile based on the teacher data conditions; and optimizing the machine learning model based on the plurality of teacher data. BRIEF DESCRIPTION OF THE DRAWINGS 【0025】 [Figure 1] FIG. 8 is a block diagram showing an example of the configuration of a computing device according to Embodiment 1. [Figure 2] FIG. 11 is a flowchart showing an example of the operation of the computing device according to Embodiment 1. [Figure 3] FIG. 14 is a flowchart showing an example of a method for generating a machine learning model. [Figure 4] (a) and (b) are schematic block diagrams each showing an example of the configuration of a machine learning model included in an inference unit. [Figure 5] FIG. 20 is a block diagram showing an example of the configuration of a computing device according to Embodiment 2. [Figure 6] FIG. 23 is a flowchart showing an example of the operation of the computing device according to Embodiment 2. [Figure 7] FIG. 26 is a conceptual diagram showing an example of the configuration of a system. [Figure 8] FIG. 29 is a block diagram showing an example of the configuration of a control device and a computing device. [Figure 9] FIG. 32 is a block diagram showing a modified example of the configuration of a control device and a computing device. [Figure 10] FIG. 35 is a table showing the mean absolute error for each crystal system of the test data of Example 1. [Figure 11]This table shows the lattice volume of each crystal phase in Example 2, its inferred value, the time required to determine the lattice constant using the method of the present invention, and the time required to determine the lattice constant using the conventional method. [Modes for carrying out the invention] 【0026】 Next, embodiments of the present invention will be described with reference to the drawings. To facilitate understanding of the description, the same reference numerals are used for identical components in each drawing, and redundant descriptions are omitted. 【0027】 [Embodiment] (Embodiment 1) [Computing device] Embodiment 1 describes the case in which the lattice volume is inferred. Figure 1 is a block diagram showing an example of the configuration of the computing device 100 according to Embodiment 1. The computing device 100 may be connected to the X-ray diffractometer 200 via a control device 300 that controls the X-ray diffractometer 200, which will be described later, or directly to the X-ray diffractometer 200. 【0028】 The computing device 100 infers the lattice volume from the X-ray powder diffraction profile. The computing device 100 is composed of a computer with a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), and memory connected to a bus. The computing device 100 may be a PC terminal or a server on the cloud. Furthermore, not only the entire device, but also some of the devices or some of the functions within the device may be provided on the cloud. The input device 510 and the display device 520 are connected to the CPU of the computing device 100 via appropriate interfaces. The input device 510 is, for example, a keyboard and mouse, and provides input to the computing device 100. The display device 520 is, for example, a display, and displays the inferred lattice volume, lattice constants, crystal system, etc. 【0029】 The computing device 100 includes an information acquisition unit 110 and an inference unit 120. Each unit can send and receive information via the control bus L. 【0030】 The information acquisition unit 110 acquires information regarding the X-ray powder diffraction profile. The information acquisition unit 110 may acquire information regarding the X-ray powder diffraction profile directly from the X-ray diffractometer 200 or via the control device 300. Alternatively, it may acquire information regarding the X-ray powder diffraction profile stored in a storage device or the like after measurement by the X-ray diffractometer 200. 【0031】 Information regarding the X-ray powder diffraction profile includes the X-ray powder diffraction profile itself, a list of d-values ​​for the X-ray powder diffraction profile, and pairs of the list of d-values ​​for the X-ray powder diffraction profile and their peak intensities. Note that when referring to a list of d-values ​​for the X-ray powder diffraction profile, peak intensity information is not included. The list of d-values ​​can be one created from the X-ray powder diffraction profile using an existing peak search method. It is preferable that the information regarding the X-ray powder diffraction profile be a list of d-values ​​for the X-ray powder diffraction profile, as the lattice volume can be inferred from the list of d-values ​​alone. 【0032】 The inference unit 120 includes a machine learning model that takes information about the X-ray powder diffraction profile as input and outputs an inferred lattice volume, and infers the lattice volume from the information about the X-ray powder diffraction profile acquired by the information acquisition unit 110. Various forms are acceptable for the inferred lattice volume. For example, the inferred lattice volume may be the volume itself or the cube root of the volume. Alternatively, the configuration may infer a range of these values. Furthermore, it is assumed that the machine learning model has already been generated and stored at the stage when the inference unit 120 infers the lattice volume. Details of the method for generating the machine learning model and the training data for generating the machine learning model will be described later. 【0033】 As described later, if a machine learning model includes multiple models, the inference unit 120 preferably uses at least one of the multiple models to infer the lattice volume for the crystal system corresponding to the model. Inferring the lattice volume for the crystal system corresponding to the model means inferring the lattice volume assuming that the information regarding the input X-ray powder diffraction profile corresponds to the crystal system of the model. A single model may include multiple crystal systems. Furthermore, inferring the lattice volume for the crystal system corresponding to the model also includes using a model that does not limit the information regarding the crystal system. The model used by the inference unit 120 may be selected when crystal system information is provided, or it may be specified by the user. If crystal system information is not provided or not specified by the user, the inference unit 120 may infer the lattice volume for all models sequentially or in parallel. 【0034】 Figure 2 is a flowchart illustrating an example of the operation of the computing device 100 according to Embodiment 1. Figure 2 shows an example of operation when only the lattice volume is inferred. First, the computing device 100 acquires information about the X-ray powder diffraction profile using the information acquisition unit 110 (step S1). Next, the inference unit 120 infers the lattice volume from the information about the X-ray powder diffraction profile acquired by the information acquisition unit 110 (step S2). The inference unit 120 may output the inferred lattice volume as needed. In this way, the lattice volume can be inferred from the information about the X-ray powder diffraction profile. The inferred lattice volume can be used to determine the lattice constant. 【0035】 Next, we will explain the method for generating a machine learning model. Figure 3 is a flowchart showing an example of a machine learning model generation method. First, the training data conditions are set (step T1). Training data conditions are the conditions for the training data used in machine learning, such as the scan range, the range of lattice constants, the wavelength of X-rays, etc. When generating training data, the training data conditions are the conditions for generating the training data. 【0036】 Next, the structure of the machine learning model is configured (step T2). Configuring the structure of the machine learning model involves specifying the type of machine learning model to apply and the specific structure of that model. For example, if a neural network model is used as the machine learning model, this would involve setting the number of layers, the number of nodes, the types of layers, etc. The training data conditions or the structure of the machine learning model may be configured to be arbitrarily set by the user through selection, specification, input, etc. Note that the configuration of the training data conditions and the configuration of the machine learning model can be done in either order or simultaneously. 【0037】 Next, multiple training data sets are prepared (step T3). The preparation of multiple training data sets involves preparing multiple training data sets that take information about the X-ray powder diffraction profile as input and output the lattice volume corresponding to the information about the X-ray powder diffraction profile, based on the training data conditions. 【0038】 Then, the machine learning model is optimized based on multiple training data sets (step T4). The optimization of the machine learning model can be performed using general methods for that model, based on the type of machine learning model being applied and its specific structure. New training data may be generated in parallel with the optimization of the machine learning model. Afterward, the optimized machine learning model or its parameters are saved, and the process terminates. Parameters and other information may be output as needed. If multiple models are to be generated, the entire process, starting from setting the training data conditions, is repeated. 【0039】 The machine learning model is preferably a neural network model. This allows for higher accuracy in the inferred grid volume. 【0040】 The training data used to generate machine learning models preferably includes information about the crystal system. This makes it possible to generate different machine learning models for each crystal system, or to generate machine learning models that infer and output the crystal system. Crystal system information refers to information indicating the crystal system of the sample from which the training data was measured or generated, and its format may be anything. 【0041】 Figures 4(a) and 4(b) are schematic block diagrams showing examples of the configuration of a machine learning model included in the inference unit 120, respectively. The machine learning model may include one model, as shown in Figure 4(a), or it may include multiple models, as shown in Figure 4(b). When the machine learning model includes multiple models, it is preferable that each of the multiple models is generated from training data containing information on different crystal systems. Alternatively, each of the multiple models may be generated from training data consisting only of information on different crystal systems. This makes it possible to generate different machine learning models for each crystal system, improving the accuracy of lattice volume inference. 【0042】 Note that the machine learning model 121 for the common block in Figure 4(a) is a machine learning model generated using training data that does not include information about the crystal system, i.e., training data that is not limited to a specific crystal system. Also, the machine learning models 122 to 128 for the XX block in Figure 4(b) are machine learning models generated using training data that is limited to that particular crystal system, i.e., training data that includes information about that particular crystal system. In Figure 4(b), an example of an inference unit 120 containing different machine learning models for each of the seven crystal systems is shown, but it is also possible to combine several crystal systems into an inference unit 120 containing fewer machine learning models, or to include at least one machine learning model for the common block and one machine learning model for each crystal system block. Furthermore, depending on the method for determining the lattice constant, the inference unit 120 may contain different machine learning models with different point clouds or space groups. 【0043】 The training data used to generate the machine learning model preferably includes a list of d-values ​​that have been preprocessed to reduce the number of peaks by a number of peaks where the difference in peak positions is less than or equal to a predetermined threshold. This allows the training data to be adjusted to the accuracy of the peak search, similar to when performing a peak search on measured data. As a result, the accuracy of inferring the grid volume for the list of d-values ​​based on measured data is improved. The preprocessing may involve, for example, setting the average position or centroid position of multiple peaks where the difference in peak positions is less than or equal to a predetermined threshold as a single peak, or determining more than one peak position from three or more peaks. The preprocessing is preferably performed automatically based on pre-set preprocessing rules. 【0044】 The training data used to generate machine learning models may be extracted using a database, but it can also be generated mechanically by mechanically creating sets of lattice constants without using a database. In other words, it is preferable that machine learning models be generated using training data generated based on training data generation conditions. This ensures a larger number of training data points and prevents bias in the number of training data points between space groups and lattice constants. As a result, it is possible to generate machine learning models that enable highly accurate inference even for samples with unusual crystal systems. 【0045】 It is preferable that the training data does not include information about atomic positions. This allows for efficient calculation of peak positions and facilitates the creation of a list of d-values. In particular, it makes it easier to mechanically generate sets of lattice constants, and theoretically possible training data can be generated in a short time. 【0046】 The training data generation conditions preferably include a condition that training data having grid constants less than or equal to a predetermined grid volume is generated. By restricting the training data generation conditions in this way, the training data necessary for generating a machine learning model can be generated efficiently. As a result, highly accurate inference becomes possible in a short time. The predetermined grid volume may be specified by the user. 【0047】 (Embodiment 2) [Computing device] Embodiment 2 describes a case in which the lattice constant is determined based on the inferred lattice volume. Since the inference of the lattice volume is the same as in Embodiment 1, the subsequent determination of the lattice constant will be described. Figure 5 is a block diagram showing an example of the configuration of the computing device 100 according to Embodiment 2. As shown in Figure 5, it is preferable that the computing device 100 includes a range determination unit 130 and a lattice constant determination unit 140 in addition to the information acquisition unit 110 and the inference unit 120. The computing device 100 with the configuration of Figure 5 may also be described as a computing device that determines the lattice constant from the X-ray powder diffraction profile. 【0048】 The range determination unit 130 determines the search range for the lattice volume based on the lattice volume inferred by the inference unit 120. The search range for the lattice volume may be a range obtained by adding or subtracting a predetermined constant from the inferred lattice volume, or the search range may be determined each time based on the inferred lattice volume and information on the crystal system. Furthermore, the search range for the lattice volume may be configured to be arbitrarily set by the user through selection or instruction. If the inference unit 120 outputs a predetermined range, the calculation device 100 of this embodiment may not include the range determination unit 130, but instead include a lattice constant determination unit 140. 【0049】 The lattice constant determination unit 140 determines the lattice constant based on the search range of the lattice volume. The lattice constant determination unit 140 may also determine the lattice constant based on information about the crystal system in addition to the search range of the lattice volume. If the machine learning model of the inference unit 120 includes multiple models and there are multiple inferred lattice volumes, it is preferable for the lattice constant determination unit 140 to search for a lattice constant for each inferred lattice volume. This process may be performed sequentially or in parallel. If multiple lattice constant candidates are obtained, the lattice constant may be determined by the program by evaluating the degree of agreement with the X-ray powder diffraction profile. Alternatively, multiple candidates may be displayed so that the user can select one. As information for the user to select, an evaluation of the degree of agreement with the X-ray powder diffraction profile may be displayed. The lattice constant determination unit 140 may have a function to determine the lattice constant itself, or it may be a function that causes an external device or software to determine the lattice constant. For example, the lattice constant determination unit 140 may input the necessary information to software such as DICVOL, N-TREOR, or ITO to determine the lattice constant. 【0050】 Figure 6 is a flowchart illustrating an example of the operation of the computing device 100 according to Embodiment 2. Figure 6 shows an example of the operation when determining the lattice constant after inferring the lattice volume. First, the computing device 100 acquires information about the X-ray powder diffraction profile using the information acquisition unit 110 (step U1). Next, the computing device 100 uses the inference unit 120 to infer the lattice volume from the information about the X-ray powder diffraction profile acquired by the information acquisition unit (step U2). The inference unit 120 may output the inferred lattice volume as needed. The operation up to this point is the same as in Embodiment 1. 【0051】 Next, the computing device 100 determines the search range for the lattice volume using the range determination unit 130 (step U3). The range determination unit 130 may output the determined search range as needed. Then, the computing device 100 determines the lattice constants using the lattice constant determination unit 140 (step U4). The lattice constant determination unit 140 may output the determined lattice constants as needed. In this way, the lattice volume can be inferred from information regarding the X-ray powder diffraction profile and the lattice constants can be determined. 【0052】 [Overall System] The computing device 100 or computing method of the present invention can acquire powder X-ray diffraction profiles and infer lattice constants or determine lattice volumes independently of the X-ray diffractometer 200 and control device 300. Therefore, the computing device 100 does not need to be used simultaneously with the X-ray diffractometer 200 or control device 300. On the other hand, it can also be integrated into a system with the X-ray diffractometer 200 and control device 300. Figure 7 is a conceptual diagram showing an example of the configuration of a system 400 including the computing device 100 and the X-ray diffractometer 200. The system 400 has the computing device 100, the X-ray diffractometer 200, and the control device 300. 【0053】 Note that in Figure 7, the computing device 100 and the control device 300 are depicted as the same PC. However, the computing device 100 may be configured as a different device from the control device 300. The following section will explain the case where the computing device 100 and the control device 300 are configured as different devices. 【0054】 [X-ray diffractometer] The X-ray diffractometer 200 comprises an optical system that incidents X-rays onto a sample and detects reflected X-rays generated from the sample. The X-ray diffractometer 200 includes at least an X-ray generator 210 that generates X-rays from an X-ray focal point, i.e., an X-ray source; a sample stage 240 on which the sample is placed and which controls the rotation of the sample; and a detector 260 that detects the X-rays. The X-ray diffractometer 200 may also include an incident optical unit 220, a goniometer 230, or an exit optical unit 250. The X-ray generator 210, incident optical unit 220, goniometer 230, sample stage 240, exit optical unit 250, and detector 260 that constitute the X-ray diffractometer 200 can be general-purpose components, so a detailed explanation is omitted. Note that the configuration shown in Figure 7 is just one example, and various other configurations can be adopted. 【0055】 [Control device] The control device 300 is connected to the X-ray diffractometer 200 and controls the X-ray diffractometer 200, as well as processing, storing, and displaying the acquired data. 【0056】 Figure 8 is a block diagram showing an example configuration of the control device 300 and the computing device 100. The control device 300 is composed of a computer with a CPU, ROM, RAM, and memory connected to a bus. The control device 300 may be a PC terminal or a server on the cloud. Furthermore, not only the entire system, but also some of the system or some of the functions within the system may be located on the cloud. The control device 300 is connected to the X-ray diffractometer 200 to receive information. 【0057】 The control device 300 comprises a control unit 310, a device information storage unit 320, a measurement data storage unit 330, and a display unit 340. Each unit can send and receive information via the control bus L. When the computing device 100 and the control device 300 have different configurations, the input device 510 and the display device 520 are connected to the CPU of the control device 300 via an appropriate interface. In this case, the input device 510 and the display device 520 may be different from those connected to the computing device 100. 【0058】 The control unit 310 controls the operation of the X-ray diffractometer 200. The device information storage unit 320 stores device information acquired from the X-ray diffractometer 200. The device information may include information about the X-ray diffractometer 200 such as the device name, type of radiation source, wavelength, and background. 【0059】 The measurement data storage unit 330 stores measurement data, including two-dimensional diffraction images acquired from the X-ray diffractometer 200. Along with the measurement data, it may also store necessary information such as the type of radiation source, wavelength, background, and other information related to the X-ray diffractometer 200, as well as the shape, arrangement, types and composition of constituent elements, and absorption coefficient of the sample. The display unit 340 displays the measurement data on the display device 520. This allows the user to confirm the measurement data. The user can also give instructions and specifications to the control device 300, the calculation device 100, etc., based on the measurement data. 【0060】 Figure 9 is a block diagram showing a modified configuration of the control device 300 and the computing device 100. As shown in Figure 9, the computing device 100 may be configured as a part of the functions included in the control device 300. Alternatively, the computing device 100 and the control device 300 may be configured as an integrated device. 【0061】 [Measurement method] A sample is placed in the X-ray diffractometer 200, and based on the control of the control device 300, X-rays are incident on the sample, and the diffracted X-rays generated from the sample are detected. If necessary, the sample stage or goniometer is also driven under predetermined conditions. This allows the X-ray powder diffraction profile to be obtained. The X-ray diffractometer 200 transmits the device information and the acquired X-ray powder diffraction profile as measurement data to the control device 300. 【0062】 By using the system 400 described above, it is possible to measure the X-ray powder diffraction profile and infer the lattice volume from the information in that profile. Furthermore, the lattice constant can be determined using the inferred lattice volume. 【0063】 [Examples] (Example 1) Example 1 involved constructing a computing device that included multiple machine learning models configured as described above, using randomly generated training data. Specifically, a neural network model consisting of 10 fully connected layers was applied as the machine learning model. Furthermore, cells were randomly generated for each crystal system, and preprocessed by combining peaks adjacent to a list of d-values, along with a set of lattice volumes, were used as training data to generate different neural network models for each crystal system. Two million data points were generated for each crystal system. Note that neural network models were generated for each of the six types of crystal systems, including Trigonal under Hexagonal. 【0064】 Next, a list of d-values ​​of cells randomly generated for each crystal system (test data), separate from the training data used to generate the neural network model, was input into the computing device, and the error between the inferred cube root of the lattice volume and the actual cube root of the lattice volume was checked. Figure 10 is a table showing the median absolute error (MedAE) for each crystal system in the test data of Example 1. Trigonal and hexagonal were checked using separate test data. After inputting 300 test data for each crystal system and examining the median absolute error, it was confirmed that although there were slight differences between crystal systems, even the crystal system with the largest error was only about 1 Å. This confirmed that the method of the present invention can accurately infer the lattice volume. 【0065】 (Example 2) In Example 2, the lattice volume of the actual material was inferred and the lattice constant was determined using the same computing device as in Example 1. 【0066】 First, two cards were randomly selected from the PDF-5+Organic subfiles whose crystal system is Triclinic, and these were designated as crystalline phase A and crystalline phase B of the organic material. Next, the list of d values ​​for the cards was input into the calculator, and the Triclinic block was made to infer the cube root of the lattice volume. Then, the inferred cube root of the lattice volume was set to (cubic root of lattice volume ± 0.5).3 Å 3 Using the search range, we restricted the crystal system to Triclinic using DICVOL and searched for and determined the lattice constants. Furthermore, the range was 0–2500 Å. 3 Using the conventional method's search range, the lattice constant was searched and determined by restricting the crystal system to Triclinic using DICVOL. Figure 11 is a table showing the lattice volume of each crystal phase in Example 2, its inferred value, the time required to determine the lattice constant using the method of the present invention, and the time required to determine the lattice constant using the conventional method. 【0067】 As shown in Figure 11, it was confirmed that when the method of the present invention is applied to infer the lattice volume and use that as a constraint to determine the lattice constant, results with the same level of accuracy can be obtained in a shorter time compared to the time required to determine the lattice constant using the conventional method. Furthermore, because the method of the present invention can infer the lattice volume in a sufficiently short time, the lattice constant can be determined in a shorter time than the conventional method, even when there is no information on the crystal system of the sample. 【0068】 Based on the above results, it has been confirmed that the computing device, method, and program of the present invention can infer the lattice volume from the X-ray powder diffraction profile. Furthermore, it has been confirmed that the lattice constant can be determined in a short time by utilizing the inferred lattice volume. [Explanation of symbols] 【0069】 100 Computing equipment 110 Information Acquisition Department 120 Reasoning Department 121-128 Machine Learning Models 130 Range Determination Unit 140 Lattice constant determination unit 200 X-ray diffractometer 210 X-ray generating unit 220 Incident Optical Unit 230 Goniometer 240 Sample stage 250 Output-side optical unit 260 detectors 300 Control device 310 Control Unit 320 Device information storage unit 330 Measurement data storage unit 340 Display section 400 System 510 Input device 520 Display device

Claims

[Claim 1] A computing device for inferring lattice volume from X-ray powder diffraction profiles, An information acquisition unit that acquires information regarding the X-ray powder diffraction profile, A computing device comprising: a machine learning model that takes information about an X-ray powder diffraction profile as input and outputs an inferred lattice volume; and an inference unit that infers the lattice volume from the information about the X-ray powder diffraction profile acquired by the information acquisition unit. [Claim 2] The computing apparatus according to claim 1, characterized in that the information relating to the X-ray powder diffraction profile is a list of d values ​​of the X-ray powder diffraction profile. [Claim 3] The computing device according to claim 1 or 2, characterized in that the machine learning model is a neural network model. [Claim 4] A range determination unit determines the search range for the grid volume based on the grid volume inferred by the inference unit, The computing device according to claim 1 or 2, further comprising a lattice constant determination unit that determines the lattice constant based on the search range of the lattice volume. [Claim 5] The computing apparatus according to claim 1 or 2, characterized in that the training data for generating the machine learning model includes information about the crystal system. [Claim 6] The aforementioned machine learning model includes multiple models, The computing device according to claim 5, characterized in that the plurality of models are generated by the training data which includes information on different crystal systems. [Claim 7] The computing device according to claim 6, characterized in that the inference unit uses at least one of the plurality of models to infer the lattice volume for the crystal system corresponding to the model. [Claim 8] The computing device according to claim 2, characterized in that the training data for generating the machine learning model includes a list of d values ​​that have been preprocessed to reduce the number of peaks from a plurality of peaks whose peak position difference is less than or equal to a predetermined threshold to a smaller number of peaks. [Claim 9] The computing device according to claim 1 or 2, characterized in that the machine learning model is generated using training data generated based on training data generation conditions. [Claim 10] The computing device according to claim 9, characterized in that the training data does not include information on atomic positions. [Claim 11] The computing device according to claim 9, characterized in that the training data generation conditions include the condition that training data having a lattice constant less than or equal to a predetermined lattice volume is generated. [Claim 12] A calculation method for inferring lattice volume from X-ray powder diffraction profiles, The steps include obtaining information regarding the X-ray powder diffraction profile, A calculation method characterized by including the step of using a machine learning model that takes information about an X-ray powder diffraction profile as input and outputs an inferred lattice volume, and inferring the lattice volume from the acquired information about the X-ray powder diffraction profile. [Claim 13] A program that infers lattice volume from X-ray powder diffraction profiles, A process for obtaining information regarding the X-ray powder diffraction profile, A program characterized by having a computer perform the following: use a machine learning model that takes information about an X-ray powder diffraction profile as input and outputs an inferred lattice volume, and infer the lattice volume from the acquired information about the X-ray powder diffraction profile. [Claim 14] A method for generating a machine learning model that, when given information about an X-ray powder diffraction profile, outputs a lattice volume inferred from the information about the X-ray powder diffraction profile, Steps to set training data conditions, The steps include setting the structure of the machine learning model, The steps include: preparing multiple training data sets based on the aforementioned training data conditions, inputting information about the X-ray powder diffraction profile and outputting the lattice volume corresponding to the information about the X-ray powder diffraction profile; A method for generating a machine learning model, comprising the step of optimizing the machine learning model based on the aforementioned plurality of training data.

Images