Liquid material prediction device, liquid material prediction method, and program

Abstract

The present invention provides a liquid material prediction device that can actually exclude compounds that can solidify and can screen for candidate liquid materials that function as liquids within a predetermined temperature range with low computational cost, high accuracy, and high efficiency. [Solution] A liquid material prediction device having a control unit that predicts liquid materials that function as liquids in a predetermined temperature range, wherein the control unit inputs the structure of the material to be predicted into a trained model trained using training data including the structure, melting point, and boiling point of the material to be learned, predicts the melting point and boiling point of the material to be predicted, outputs a first group of candidate materials that are predicted to be in a liquid state in the predetermined temperature range, predicts multiple physical properties of the first group of candidate materials, including kinematic viscosity, by machine learning and / or simulation, and outputs a second group of candidate materials in which the predicted values ​​of the multiple physical properties in the predetermined temperature range are predicted to meet a standard value.

Description

Technical Field

[0001] The present disclosure relates to a liquid material prediction device, a liquid material prediction method, a program, and a compound for liquid materials.

Background Art

[0002] Methods for generating design data of articles having predictable properties by machine learning and computer simulation have been reported (see, for example, Patent Document 1 and the like).

[0003] However, screening from a huge group of compounds still has the problem of high computational load. In addition, when the usage conditions of a compound have a temperature range, it is important to provide a material that functions as a liquid within the temperature range. Although it is possible to estimate the liquid state from the boiling point and freezing point, the fluidity required for liquid materials cannot be considered solely based on the phase change temperatures. Therefore, in the screening of liquid materials, it is required to predict the kinematic viscosity in a predetermined temperature range with high accuracy, but conventional methods have the problem of low prediction accuracy for temperature-dependent kinematic viscosity.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of Invention

Problems to be Solved by the Invention

[0005] An object of the present disclosure is to provide a liquid material prediction device capable of screening candidates for liquid materials that function as liquids in a predetermined temperature range with high accuracy and high efficiency at low computational cost.

Means for Solving the Problems

[0006] <1> A liquid material prediction device having a control unit that predicts a liquid material that functions as a liquid in a predetermined temperature range, The control unit, By inputting the structure of the target substance into a trained model that has been trained using training data including the structure, melting point, and boiling point of the target substance, the melting point and boiling point of the target substance are predicted, and a first group of candidate substances that are predicted to be in a liquid state within the predetermined temperature range are output. The system predicts multiple physical properties, including kinematic viscosity, of a first candidate group of materials using machine learning and / or simulation, and outputs a second group of materials in which the predicted values ​​of the multiple physical properties in a predetermined temperature range are expected to meet a standard value. Liquid material prediction device. <2> The simulation includes quantum chemical calculations and / or molecular dynamics calculations. <1> The liquid material prediction device described above. <3> The temperature-dependent predicted value of the kinematic viscosity is a value output based on a function that includes the absolute viscosity and density at a reference temperature calculated by molecular dynamics calculations. <1> or <2> The liquid material prediction device described above. <4> A liquid material prediction device having a control unit that predicts a liquid material that functions as a liquid in a predetermined temperature range, The control unit, The system predicts multiple physical properties of a target substance, including its kinematic viscosity, using machine learning and / or simulation. The temperature-dependent predicted value of the kinematic viscosity is output as a function that includes the self-diffusion coefficient at a reference temperature calculated by molecular dynamics calculations, and also includes the self-diffusion coefficient and density at an arbitrary temperature within the predetermined temperature range. The system then outputs a group of candidate substances in which the predicted values ​​of the multiple physical properties within the predetermined temperature range are expected to satisfy the reference values. Liquid material prediction device. <5> The temperature-dependent predicted value of the kinematic viscosity is a value output as a function that includes the self-diffusion coefficient at a reference temperature calculated by molecular dynamics calculations, and also includes the self-diffusion coefficient and density at an arbitrary temperature within the predetermined temperature range. <1> from <4> A liquid material prediction device as described in any one of the items. <6> The plurality of physical properties further include thermal properties and / or electrical properties. <1> from <5> A liquid material prediction device as described in any one of the items. <7> The aforementioned thermal properties include thermal conductivity and / or specific heat. <6> The liquid material prediction device described above. <8> The aforementioned electrical properties include one or more properties selected from the group consisting of dielectric constant, dielectric loss tangent, electrical conductivity, volume resistivity, and dielectric breakdown voltage. <6> or <7> The liquid material prediction device described above. <9> The application of the liquid material is selected from the group consisting of a heat transfer medium, a solvent, a cleaning agent, a propellant, a fire extinguishing agent, a lubricant, and a dielectric. <1> from <8> A liquid material prediction device as described in any one of the items. <10> The heat transfer medium is either a cooling heat transfer medium or a heating heat transfer medium. <9> The liquid material prediction device described above. <11> The solvent is either a reaction solvent or an extraction solvent. <9> The liquid material prediction device described above. <12> A liquid material prediction method that uses a computer to predict a liquid material that functions as a liquid in a predetermined temperature range, The aforementioned computer, The process involves inputting the structure of a target substance into a trained model, which has been trained using training data including the structure, melting point, and boiling point of the target substance, thereby predicting the melting point and boiling point of the target substance, and outputting a first group of candidate substances that are predicted to be in a liquid state within a predetermined temperature range. A step of predicting multiple physical properties, including kinematic viscosity, of a first candidate group of materials by machine learning and / or simulation, and outputting a second group of materials in which the predicted values ​​of the multiple physical properties in a predetermined temperature range are predicted to meet a reference value, A liquid material prediction method that performs the following. <13> A liquid material prediction method that uses a computer to predict a liquid material that functions as a liquid in a predetermined temperature range, The aforementioned computer, The process involves predicting multiple physical properties of a target substance, including its kinematic viscosity, using machine learning and / or simulation, wherein the temperature-dependent predicted value of the kinematic viscosity is output as a function that includes the self-diffusion coefficient at a reference temperature calculated by molecular dynamics calculations, and the self-diffusion coefficient and density at an arbitrary temperature within the predetermined temperature range, and outputting a group of candidate substances in which the predicted values ​​of the multiple physical properties within the predetermined temperature range are predicted to satisfy a reference value. A liquid material prediction method that performs the following. <14> A program for predicting liquid materials that function as liquids within a predetermined temperature range, On the computer, The process involves inputting the structure of a target substance into a trained model, which has been trained using training data including the structure, melting point, and boiling point of the target substance, thereby predicting the melting point and boiling point of the target substance, and outputting a first group of candidate substances that are predicted to be in a liquid state within a predetermined temperature range. A step of predicting multiple physical properties, including kinematic viscosity, of a first candidate group of materials by machine learning and / or simulation, and outputting a second group of materials in which the predicted values ​​of the multiple physical properties in a predetermined temperature range are predicted to meet a reference value, A program to execute. <15> A program for predicting liquid materials that function as liquids within a predetermined temperature range, On the computer, The process involves predicting multiple physical properties of a target substance, including its kinematic viscosity, using machine learning and / or simulation, wherein the temperature-dependent predicted value of the kinematic viscosity is output as a function that includes the self-diffusion coefficient at a reference temperature calculated by molecular dynamics calculations, and the self-diffusion coefficient and density at an arbitrary temperature within the predetermined temperature range, and outputting a group of candidate substances in which the predicted values ​​of the multiple physical properties within the predetermined temperature range are predicted to satisfy a reference value. A program to execute. <16> The aforementioned <12> or <13> A compound for liquid materials that is predicted by the liquid material prediction method described above and functions as a liquid in the predetermined temperature range. [Effects of the Invention]

[0007] According to the present disclosure, it is possible to provide a liquid material prediction apparatus capable of screening candidates for liquid materials that function as liquids in a predetermined temperature range with high accuracy and high efficiency at low calculation cost. [Brief Description of the Drawings]

[0008] [Figure 1] Figure 1 is a block diagram showing an example of the overall configuration of a liquid material prediction apparatus. [Figure 2] Figure 2 is a block diagram showing an example of a computer. [Figure 3] Figure 3 is a block diagram showing an example of the functional configuration of a liquid material prediction apparatus. [Figure 4] Figure 4 is a flowchart showing an example of a learning process related to the liquid material prediction method of the present embodiment. [Figure 5] Figure 5 is a flowchart showing an example of the first prediction process related to the liquid material prediction method of the present embodiment. [Figure 6] Figure 6 is a flowchart showing an example of the second prediction process related to the liquid material prediction method of the present embodiment. [Figure 7] Figure 7 is a graph showing the relationship between the predicted values by the boiling point prediction model for the learning target substance group and the literature values. [Figure 8] Figure 8 is a graph showing the relationship between the predicted values by the melting point prediction model for the learning target substance group and the literature values. [Figure 9] Figure 9 is a graph showing the calculated values of absolute viscosity at 300K for the first candidate substance group. [Figure 10] Figure 10 is a graph showing the predicted values of absolute viscosity at 300K for the first candidate substance group. [Figure 11] Figure 11 is a graph showing the predicted values of kinematic viscosity of substances A to C in the first candidate substance group in a comparative example. [Figure 12] Figure 12 is a graph showing the predicted values of kinematic viscosity of substances A to C in the first candidate substance group in an example. [Figure 13]Figure 13 is a graph showing the predicted density values ​​of the second candidate group of substances in the examples. [Figure 14] Figure 14 is a graph showing the predicted kinematic viscosity of the second candidate group of materials in the examples. [Figure 15] Figure 15 is a graph showing the predicted relative permittivity of the second candidate material group in the examples. [Figure 16] Figure 16 is a graph showing the predicted boiling points of the second candidate group of substances in the examples. [Figure 17] Figure 17 is a graph showing the predicted thermal conductivity values ​​for the second candidate material group in the example. [Figure 18] Figure 18 is a graph showing the predicted specific heat values ​​for the second candidate group of materials in the examples. [Modes for carrying out the invention]

[0009] Hereinafter, embodiments of this disclosure will be described with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0010] (Liquid material prediction device) The liquid material prediction device of this disclosure is a liquid material prediction device having a control unit that predicts a liquid material that functions as a liquid in a predetermined temperature range, The control unit, By inputting the structure of the target substance into a trained model that has been trained using training data including the structure, melting point, and boiling point of the target substance, the melting point and boiling point of the target substance are predicted, and a first group of candidate substances that are predicted to be in a liquid state within the predetermined temperature range are output. Machine learning and / or simulation are used to predict multiple physical properties of the first candidate group of materials, including kinematic viscosity, and a second candidate group of materials is output in which the predicted values ​​of the multiple physical properties in a predetermined temperature range are expected to meet a reference value.

[0011] The first embodiment of this disclosure is an example of an information processing system that predicts liquid materials that function as liquids in a predetermined temperature range. Hereinafter, the information processing system according to this embodiment will be referred to as the "liquid material prediction device." The substance to be predicted will be referred to as the "predicted substance."

[0012] In this embodiment, the substance to be predicted can be any substance. The properties to be predicted include melting point, boiling point, and kinematic viscosity, and may also include other properties depending on the application. As an example, the properties to be predicted may include thermophysical properties and / or electrical properties.

[0013] Conventionally, techniques for predicting the physical properties of materials using methods such as machine learning are known. In various applications, when searching for liquid materials that function as liquids within a desired temperature range, machine learning is used to predict the melting and boiling points of target materials, and then the materials that are above their melting point and below their boiling point within the set temperature range are screened.

[0014] However, the fluidity required for liquid materials cannot be considered solely by their melting and boiling points. Therefore, in screening liquid materials, it is necessary to predict the kinematic viscosity in a given temperature range with high accuracy. Temperature-dependent kinematic viscosity can be predicted using molecular dynamics simulations. However, when the inventors actually predicted temperature-dependent kinematic viscosity, they found errors and discrepancies with experimental values ​​regarding the trend of kinematic viscosity in the low-temperature range. Consequently, it was found that conventional kinematic viscosity prediction methods have the problem of not being able to screen compounds that function as liquid materials.

[0015] According to the liquid material prediction device of this embodiment, by further predicting the temperature-dependent kinematic viscosity of candidate substances screened by predicting their boiling point and melting point, it is possible to screen for candidate liquid materials that function as liquids in a predetermined temperature range with low computational cost, high accuracy, and high efficiency.

[0016] <Overall Structure> The overall configuration of the liquid material prediction device according to this embodiment will be described with reference to Figure 1. Figure 1 is a block diagram showing an example of the overall configuration of the liquid material prediction device.

[0017] As shown in Figure 1, the liquid material prediction device 1000 includes a learning device 10, a prediction device 20, and a terminal device 40. The learning device 10, the prediction device 20, and the terminal device 40 are connected via a communication network N to enable data communication. The communication network N may be, for example, a LAN (Local Area Network), a VPN (Virtual Private Network), or the Internet.

[0018] The learning device 10 is an example of an information processing device such as a personal computer, workstation, or server that learns a predictive model. The predictive model is an example of a machine learning model that predicts predetermined physical properties. The predetermined physical properties include melting point and boiling point, and may also include any other physical properties as needed.

[0019] The prediction model may be a machine learning model that has learned the relationship between the descriptor of the substance to be trained and the predetermined physical properties of that substance. Hereafter, the substance to be trained will also be called the "training target substance." The training target substance can be any substance, just like the prediction target substance.

[0020] The learning device 10 may train a predictive model using multiple datasets. The learning device 10 may set different fidelity levels for each of the multiple datasets. Each dataset may include identification information of the substance to be learned and physical property values ​​that indicate the physical properties of the substance. The learning device 10 may learn the relationship between descriptors and physical properties while taking into account the fidelity levels set for each dataset.

[0021] The prediction model can be any type of machine learning model, but it is preferably a multiple regression model that avoids multicollinearity. Examples of machine learning models include random forest regression, Gaussian process regression, and linear regression.

[0022] The learning device 10 may calculate a descriptor of the target substance based on the identification information of the target substance. The learning device 10 may generate training data for training a prediction model based on the descriptor and physical properties of the target substance. The learning device 10 may generate a prediction model by learning the relationship between the descriptor and physical properties based on the training data. The learning device 10 may output the trained prediction model to the prediction device 20.

[0023] The prediction device 20 is an example of an information processing device such as a personal computer, workstation, or server that predicts the physical properties of a target substance. The prediction device 20 may predict the physical properties of the target substance based on a trained prediction model. The trained prediction model may be generated by the learning device 10. The prediction device 20 may calculate a descriptor of the target substance based on the identification information of the target substance. The prediction device 20 may predict the physical properties of the target substance by inputting the descriptor of the target substance into the trained prediction model.

[0024] Terminal device 40 is an example of an information processing terminal such as a personal computer, smartphone, or tablet terminal operated by the user of the liquid material prediction device 1000. Terminal device 40 may accept input of identification information of the substance to be predicted and transmit it to the prediction device 20. Terminal device 40 may receive prediction results from the prediction device 20 and present them to the user.

[0025] The overall configuration of the liquid material prediction device 1000 shown in Figure 1 is just one example, and various system configurations are possible depending on the application and purpose. For example, one or more of the learning device 10, prediction device 20, and terminal device 40 may be included in multiple units of the liquid material prediction device 1000. For example, the learning device 10 or prediction device 20 may be implemented by multiple computers, or as a cloud computing service. For example, the liquid material prediction device 1000 may be implemented by a standalone computer. The classification of devices such as the learning device 10, prediction device 20, and terminal device 40 shown in Figure 1 is just one example.

[0026] <Hardware Configuration> The hardware configuration of the liquid material prediction device 1000 will be explained with reference to Figure 2. The learning device 10, prediction device 20, and terminal device 40 are implemented, for example, by a computer. Figure 2 is a block diagram showing an example of the computer's hardware configuration.

[0027] As shown in Figure 2, the computer 500 includes a CPU (Central Processing Unit) 501, ROM (Read Only Memory) 502, RAM (Random Access Memory) 503, HDD (Hard Disk Drive) 504, input device 505, display device 506, communication interface 507, and external interface 508. The CPU 501, ROM 502, and RAM 503 form what is known as a computer. Each piece of hardware in the computer 500 is interconnected via a bus line 509. The input device 505 and display device 506 may also be used by connecting them to the external interface 508.

[0028] The CPU 501 is a processing unit that reads programs and data from a storage device such as the ROM 502 or HDD 504 onto the RAM 503 and executes processing, thereby realizing the overall control and functions of the computer 500. The computer 500 may have a GPU (Graphics Processing Unit) in addition to or instead of the CPU 501.

[0029] ROM502 is an example of non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. ROM502 functions as the main memory, storing various programs and data necessary for the CPU501 to execute the programs installed on HDD504. Specifically, ROM502 stores boot programs such as BIOS (Basic Input Output System) and EFI (Extensible Firmware Interface) that are executed when the computer 500 starts up, as well as OS (Operating System) settings, network settings, and other data.

[0030] RAM503 is an example of volatile semiconductor memory (storage device) whose programs and data are erased when the power is turned off. RAM503 includes, for example, DRAM (Dynamic Random Access Memory) and SRAM (Static Random Access Memory). RAM503 provides a working area that is expanded when various programs installed on HDD504 are executed by CPU501.

[0031] HDD504 is an example of a non-volatile storage device that stores programs and data. The programs and data stored in HDD504 include the operating system (OS), which is the basic software that controls the entire computer 500, and applications that provide various functions on the OS. Note that computer 500 may use a storage device that uses flash memory as its storage medium (e.g., SSD: Solid State Drive) instead of HDD504.

[0032] The input device 505 includes a touch panel used by the user to input various signals, operation keys and buttons, a keyboard and mouse, and a microphone for inputting sound data such as voice.

[0033] The display device 506 consists of a display such as a liquid crystal or organic EL (Electro-Luminescence) that displays a screen, and a speaker that outputs sound data such as audio.

[0034] Communication I / F 507 is an interface that connects to a communication network and allows computer 500 to perform data communication.

[0035] External I / F 508 is an interface for external devices. Examples of external devices include the drive device 510.

[0036] The drive device 510 is a device for setting the recording medium 511. The recording medium 511 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 511 may also include semiconductor memory that records information electrically, such as ROMs and flash memory. This allows the computer 500 to read and / or write to the recording medium 511 via the external I / F 508.

[0037] The various programs to be installed on the HDD 504 are installed, for example, when the distributed recording medium 511 is set in a drive device 510 connected to an external I / F 508, and the various programs recorded on the recording medium 511 are read by the drive device 510. Alternatively, the various programs to be installed on the HDD 504 may be installed by downloading them via the communication I / F 507 from a communication network or another network different from the communication network.

[0038] <Functional Configuration> The functional configuration of the liquid material prediction device 1000 will be explained with reference to Figure 3. Figure 3 is a block diagram showing an example of the functional configuration of the liquid material prediction device.

[0039] <<Learning device>> As shown in Figure 3, the learning device 10 includes a data storage unit 101, a descriptor calculation unit 110, a learning data generation unit 120, and a model generation unit 130. The learning device 10 functions as the data storage unit 101, descriptor calculation unit 110, learning data generation unit 120, and model generation unit 130 when a pre-installed learning program is executed.

[0040] For example, the data storage unit 101 is implemented by the HDD 504 shown in Figure 2. For example, the descriptor calculation unit 110, the learning data generation unit 120, and the model generation unit 130 are implemented by a process in which a program loaded from the HDD 504 shown in Figure 2 onto the RAM 503 is executed by the CPU 501.

[0041] The data storage unit 101 stores a dataset. The dataset is a collection of training data used to build a predictive model. The data storage unit 101 may have multiple datasets stored in it beforehand. The dataset may include identification information of the target substance and physical property values ​​of the target substance.

[0042] Identification information is information that can identify a substance. Examples of identification information may include the compound name, CAS number, structural formula, descriptor, SMILES (Simplified Molecular Input Line Entry System) information, ECFP (Extended Connectivity Circular Fingerprints) information, etc. Examples of descriptors include Mordred descriptors, RDKit descriptors, ISIDA descriptors, CircuS descriptors, WHALES descriptors, CDK descriptors, and CATS descriptors. Identification information is not limited to these; any information that can identify a substance can be used.

[0043] The dataset may include, for example, data available from publicly accessible databases such as PubChem (National Center for Biotechnology Information, NCBI), ChemSpider (RSC), SMARTS (Patcore, Inc.), and CAS REGISTRY, and may include experimental data, simulation data, or literature data.

[0044] The experimental data may include experimental values ​​obtained by measuring the physical properties of the target substance as physical property values ​​of the target substance. Multiple experimental data may be multiple experimental data using different experimental methods. The experimental methods may include, for example, experiments using existing substances, or experiments using novel substances synthesized from multiple substances.

[0045] Simulation data may include calculated values ​​obtained by simulation as physical properties of the target substance. Multiple simulation data may be multiple simulation data using different simulation methods. The simulation method may be, for example, quantum scientific calculation or molecular dynamics calculation. Simulation data obtained by quantum scientific calculation may include multiple simulation data with different numbers of basis function parameters. Simulation data obtained by molecular dynamics calculation may include multiple simulation data with different total number of atoms in the system.

[0046] Literature data may include literature values ​​described in publicly available literature as physical properties of the substance being studied. Literature values ​​may be experimental or calculated. Multiple literature data may include data from different literatures.

[0047] The descriptor calculation unit 110 calculates the descriptor of the target substance. The descriptor calculation unit 110 may calculate the descriptor of the target substance based on the identification information of the target substance. The descriptor calculation unit 110 may calculate the descriptor of the target substance based on the identification information of the target substance included in the dataset.

[0048] The learning data generation unit 120 generates learning data. The learning data is electronic data used to train a prediction model. The learning data generation unit 120 may generate learning data that includes identification information of the target substance, descriptors of the target substance, and physical properties of the target substance. The identification information of the learning data may be identification information included in the dataset. The descriptors of the learning data may be descriptors calculated by the descriptor calculation unit 110. The physical properties of the learning data may be physical properties included in the dataset. In other words, the learning data generation unit 120 may generate learning data by adding descriptors calculated by the descriptor calculation unit 110 to the dataset stored in the data storage unit 101.

[0049] The training data generation unit 120 may generate training data by performing preprocessing on the acquired dataset, such as multicollinearity avoidance, outlier removal, removal of zero-variance features, and standardization, in order to improve the reliability of predictions. The training data generation unit 120 may also store the generated training data in the data storage unit 101.

[0050] The model generation unit 130 generates prediction models with melting point, boiling point, and any other physical property as the target variable. The model generation unit 130 may also generate prediction models based on the training data generated by the training data generation unit 120. The model generation unit 130 may also generate prediction models based on the training data read from the data storage unit 101. The model generation unit 130 may also generate prediction models by learning the relationship between descriptors and physical properties based on the descriptors and physical property values ​​included in the training data.

[0051] The model generation unit 130 outputs a trained prediction model. The model generation unit 130 may also transmit the trained prediction model to the prediction device 20. The trained prediction model may be stored in the model storage unit 201 of the prediction device 20.

[0052] <<First Prediction Device>> As shown in Figure 3, the prediction device 20 (first prediction device) comprises a model storage unit 201, an information acquisition unit 210, a descriptor calculation unit 220, and a first physical property prediction unit 230. The prediction device 20 functions as the model storage unit 201, information acquisition unit 210, descriptor calculation unit 220, and first physical property prediction unit 230 when a pre-installed prediction program is executed.

[0053] For example, the model storage unit 201 is implemented by the HDD 504 shown in Figure 2. For example, the information acquisition unit 210, the descriptor calculation unit 220, and the first physical property prediction unit 230 are implemented by a process in which a program loaded from the HDD 504 shown in Figure 2 onto the RAM 503 is executed by the CPU 501.

[0054] The model storage unit 201 stores a trained prediction model. The model storage unit 201 may already have a pre-trained prediction model stored in it. The prediction model stored in the model storage unit 201 may be generated by the learning device 10.

[0055] The information acquisition unit 210 acquires identification information of the substance to be predicted. The information acquisition unit 210 may also receive identification information of the substance to be predicted transmitted by the terminal device 40. The information acquisition unit 210 may also accept input of identification information of the substance to be predicted via the input device 505 of the prediction device 20.

[0056] The descriptor calculation unit 220 calculates the descriptor of the target substance. The descriptor calculation unit 220 may calculate the descriptor of the target substance based on the identification information of the target substance. The descriptor calculation unit 220 may calculate the descriptor of the target substance based on the identification information of the target substance acquired by the information acquisition unit 210. The descriptor calculation unit 220 may calculate the descriptor in the same manner as the descriptor calculation unit 110 of the learning device 10.

[0057] The first physical property prediction unit 230 predicts the melting point, boiling point, and any other physical properties as needed of the target substance. The first physical property prediction unit 230 may predict the physical properties of the target substance based on a prediction model generated by the learning device 10. The first physical property prediction unit 230 may predict the physical properties of the target substance based on a prediction model read from the model storage unit 201. The first physical property prediction unit 230 may predict the physical properties of the target substance based on descriptors calculated by the descriptor calculation unit 220. The first physical property prediction unit 230 may predict the physical properties of the target substance by inputting the descriptors of the target substance into the prediction model.

[0058] The first physical property prediction unit 230 may output the predicted physical properties. The first physical property prediction unit 230 may transmit the predicted physical properties to the terminal device 40. The first physical property prediction unit 230 may present the predicted physical properties to the user. The first physical property prediction unit 230 may display the predicted results on the display device 506 of the prediction device 20. The first physical property prediction unit 230 may output the information used for the prediction along with the predicted physical properties. The information used for the prediction may, as an example, include at least one of the identification information of the substance to be predicted or the descriptor of the substance to be predicted.

[0059] <<Second Prediction Device>> As shown in Figure 3, the prediction device 30 (second prediction device) comprises a data storage unit 301, a physical property calculation unit 310, a physical property prediction unit 320, and a second physical property prediction unit 330. The prediction device 30 functions as the data storage unit 301, physical property calculation unit 310, physical property prediction unit 320, and second physical property prediction unit 330 when a pre-installed prediction program is executed.

[0060] For example, the data storage unit 301 is implemented by the HDD 504 shown in Figure 2. For example, the physical property calculation unit 310, the physical property prediction unit 320, and the second physical property prediction unit 330 are implemented by a process in which a program loaded from the HDD 504 shown in Figure 2 onto the RAM 503 is executed by the CPU 501. Identification information of the first candidate group of materials is stored in the data storage unit 301.

[0061] The physical property calculation unit 310 acquires identification information of the first candidate group of substances and calculates multiple physical properties, including absolute viscosity and density at a reference temperature Ts, for each substance in the first candidate group of substances by machine learning and / or simulation. The physical property calculation unit 310 may receive identification information of the first candidate group of substances from the predicted substances transmitted by the terminal device 40. The physical property calculation unit 310 may accept input of identification information of the first candidate group of substances via the input device 505 of the prediction device 30. The physical property calculation unit 310 may predict the physical properties of the first candidate group of substances based on the prediction model read from the data storage unit 101. The physical property calculation unit 310 may predict the physical properties of the first candidate group of substances based on descriptors calculated by the descriptor calculation unit 220. The physical property calculation unit 310 may predict the physical properties of the first candidate group of substances by inputting the descriptors of the first candidate group of substances into the prediction model.

[0062] The physical property prediction unit 320 predicts multiple physical properties, including kinematic viscosity at an arbitrary temperature T, for each substance in the first candidate substance group, based on the calculated absolute viscosity and density at a reference temperature Ts. The physical property prediction unit 320 may also calculate a function for predicting kinematic viscosity at an arbitrary temperature T based on the acquired absolute viscosity and density at the reference temperature. The physical property prediction unit 320 may also calculate a function for predicting kinematic viscosity at an arbitrary temperature T based on the acquired self-diffusion coefficient at the reference temperature, and the self-diffusion coefficient and density at an arbitrary temperature in a predetermined temperature range. The physical property prediction unit 320 predicts kinematic viscosity at an arbitrary temperature T based on the function calculated for each substance in the first candidate substance group.

[0063] The second physical property prediction unit 330 receives a plurality of physical property values, including the predicted value of kinematic viscosity at an arbitrary temperature T, which was predicted by the physical property prediction unit 320. The second physical property prediction unit 330 determines whether the predicted values ​​of the plurality of physical properties in a predetermined temperature range meet the reference value, and outputs a second group of candidate materials that are predicted to meet the reference value.

[0064] The second physical property prediction unit 330 may output the predicted physical properties. The second physical property prediction unit 330 may transmit the predicted physical properties to the terminal device 40. The second physical property prediction unit 330 may present the predicted physical properties to the user. The second physical property prediction unit 330 may display the predicted results on the display device 506 of the prediction device 20. The second physical property prediction unit 330 may output the information used for the prediction along with the predicted physical properties. The information used for the prediction may, for example, include at least one of the identification information of the substance to be predicted or the descriptor of the substance to be predicted.

[0065] <Processing Procedure> The liquid material prediction method performed by the liquid material prediction device 1000 will be explained with reference to Figures 4 and 5. The step of outputting the first group of candidate substances in the liquid material prediction method includes a learning process performed by the learning device 10 (see Figure 4) and a first prediction process performed by the prediction device 20 (see Figure 5). Subsequently, the step of outputting the second group of candidate substances includes a second prediction process performed by the prediction device 30 (see Figure 6).

[0066] <<Learning Process>> Figure 4 is a flowchart showing an example of the learning process according to the first embodiment. The learning process is the process of learning a predictive model. The learning process is performed by the learning device 10.

[0067] In step S11, the descriptor calculation unit 110 of the learning device 10 reads the dataset from the data storage unit 101. The descriptor calculation unit 110 obtains identification information of the learning target substances included in the dataset. Based on the identification information obtained from the dataset, the descriptor calculation unit 110 calculates the descriptors of the learning target substances. The descriptor calculation unit 110 sends the descriptors of the learning target substances included in the dataset to the learning data generation unit 120.

[0068] In step S12, the learning data generation unit 120 of the learning device 10 receives descriptors of the learning target substances included in the dataset from the descriptor calculation unit 110.

[0069] The training data generation unit 120 adds descriptors of the target substance to the dataset and generates training data. The training data generation unit 120 sends the training data to the model generation unit 130.

[0070] In step S13, the model generation unit 130 of the learning device 10 receives learning data from the learning data generation unit 120. Based on the learning data, the model generation unit 130 generates a trained predictive model by learning the relationship between descriptors and physical properties of the target substance.

[0071] In step S14, the model generation unit 130 of the learning device 10 outputs the trained prediction model generated in step S13. The learning device 10 transmits the trained prediction model output from the model generation unit 130 to the prediction device 20.

[0072] The prediction device 20 receives a trained prediction model from the learning device 10. The prediction device 20 stores the received trained prediction model in the model storage unit 201.

[0073] As described above, in the learning process of the step that outputs the first group of candidate substances, a trained model is obtained using training data including the structure, melting point, and boiling point of the substance to be trained.

[0074] <<First Prediction Process>> Figure 5 is a flowchart showing an example of the prediction process in the step of outputting the first group of candidate substances in the liquid material prediction method. The first prediction process is a process of predicting physical properties based on a trained prediction model. The first prediction process is performed by the prediction device 20.

[0075] In step S21, the user of the liquid material prediction device 1000 inputs identification information of the substance to be predicted into the terminal device 40. The terminal device 40 transmits the identification information of the substance to be predicted to the prediction device 20.

[0076] The prediction device 20 receives identification information of a target substance for prediction from the terminal device 40. The information acquisition unit 210 of the prediction device 20 acquires the identification information of the target substance for prediction received by the prediction device 20. The information acquisition unit 210 sends the acquired identification information of the target substance for prediction to the descriptor calculation unit 220.

[0077] In step S22, the descriptor calculation unit 220 of the prediction device 20 receives the identification information of the target substance for prediction from the information acquisition unit 210. The descriptor calculation unit 220 calculates a descriptor of the target substance for prediction based on the identification information of the target substance for prediction. The descriptor calculation unit 220 sends the descriptor of the target substance for prediction to the first physical property prediction unit 230.

[0078] In step S23, the first physical property prediction unit 230 of the prediction device 20 receives the descriptor of the target substance for prediction from the descriptor calculation unit 220. The first physical property prediction unit 230 reads a learned prediction model from the model storage unit 201. The first physical property prediction unit 230 inputs the descriptor of the target substance for prediction into the read prediction model. The prediction model predicts a predetermined physical property based on the input descriptor of the target substance for prediction, and outputs a predicted value of the predetermined physical property. The first physical property prediction unit 230 acquires the predicted value output from the prediction model. In the present embodiment, the first physical property prediction unit 230 predicts the predetermined physical property based on the difference learning model by executing a physical property prediction process (see FIG. 9).

[0079] In step S24, the first physical property prediction unit 230 of the prediction device 20 outputs the prediction result of the physical property. The prediction result of the physical property includes the predicted values of physical properties including the melting point and boiling point acquired in step S23. Based on the predicted values, a substance group that satisfies predicted melting point Tm < Tx and Ty < predicted boiling point Tb is selected and output as the first candidate substance group, which is the group of substances predicted to be in a liquid state in the predetermined temperature range Tx to Ty [K]. The prediction result of the physical property may include information used for the prediction of the physical property. The information used for the prediction may include, for example, at least one of the identification information of the target substance for prediction and the descriptor of the target substance for prediction.

[0080] The prediction device 20 transmits the first candidate substance group to the data storage unit 301 as a physical property prediction result. The data storage unit 301 receives the physical property prediction result from the prediction device 20. Alternatively, the prediction device 20 may transmit the result to the terminal device 40. The terminal device 40 may receive the physical property prediction result from the prediction device 20. The terminal device 40 may present the received physical property prediction result to the user. The terminal device 40 may display the predicted value included in the prediction result on the display device 506. The terminal device 40 may display the information used for the prediction together with the predicted value.

[0081] As described above, in the first prediction process of the step of outputting the first candidate substance group, the melting point and boiling point of the target substance for prediction are predicted by inputting the structure of the target substance for prediction into the trained model. Then, as substances predicted to be in a liquid state in the predetermined temperature range Tx to Ty [K], the first candidate substance group satisfying predicted melting point Tm < Tx and Ty < predicted boiling point Tb is selected and output.

[0082] <<Second Prediction Process>> Figure 6 is a flowchart showing an example of the prediction process in the step of outputting the second candidate substance group in the liquid material prediction method. The second prediction process is a process of predicting physical properties by machine learning and / or simulation. The second prediction process is executed by the prediction device 30.

[0083] In step S31, the data storage unit 301 transmits the identification information of the first candidate substance group to the physical property calculation unit 310. The physical property calculation unit 310 receives the identification information of the first candidate substance group.

[0084] In step S32, the physical property calculation unit 310 calculates several physical properties, including absolute viscosity and density at a reference temperature Ts, in order to predict the kinematic viscosity of each substance in the first candidate group of substances by machine learning and / or simulation. The simulation includes quantum chemical calculations and / or molecular dynamics calculations. The physical property calculation unit 310 sends the calculated kinematic viscosity and other physical properties to the physical property prediction unit 320. Preferably, the multiple physical properties further include the self-diffusion coefficient, and may further include physical properties selected as appropriate according to the desired application of the liquid material.

[0085] In step S33, it is preferable that the physical property prediction unit 320 receives multiple physical properties from the physical property calculation unit 310, including absolute viscosity and density at a reference temperature Ts for each substance in the first candidate group of substances. Based on the calculated values ​​of absolute viscosity and density at the reference temperature Ts, the physical property prediction unit 320 predicts multiple physical properties, including kinematic viscosity at an arbitrary temperature T. Specifically, it can predict the predicted values ​​of absolute viscosity μ(T) and density ρ(T) at an arbitrary temperature T, and then predict the predicted value of kinematic viscosity ν(T) at an arbitrary temperature T. By determining the kinematic viscosity ν(T) at an arbitrary temperature T within a predetermined temperature range Tx~Ty[K], a temperature-dependent predicted value of kinematic viscosity can be obtained.

[0086] By applying this method, it becomes possible to predict the temperature dependence of the kinematic viscosity of a liquid with a relatively low computational load. Therefore, even by using the prediction device 30 incorporating this method alone, it is possible to efficiently perform temperature-dependent predictions of kinematic viscosity.

[0087] The temperature-dependent predicted value of the kinematic viscosity is preferably a value output based on a function that includes the absolute viscosity and density at a reference temperature calculated by molecular dynamics calculations. Specifically, it can be output based on the calculated absolute viscosity and density at a reference temperature Ts, using the following equations (4) and (5).

[0088] Furthermore, it is more preferable that the temperature-dependent predicted value of the kinematic viscosity is a value output as a function that includes the self-diffusion coefficient at a reference temperature calculated by molecular dynamics calculations, and also includes the self-diffusion coefficient and density at an arbitrary temperature within the predetermined temperature range. Specifically, it can be output based on the calculated absolute viscosity and density at a reference temperature Ts, using the following equations (4) and (5).

[0089] As is clear from the examples described later, the calculated absolute viscosity value obtained using equation (1) below is almost always larger than the measured value, indicating a tendency for the calculated value to be overestimated. By correcting with equation (4) below, which is a correction formula that includes density, or the self-diffusion coefficient and density as variables, the correlation with the measured absolute viscosity is improved, and the absolute viscosity and kinematic viscosity at an arbitrary temperature T can be predicted with greater accuracy.

[0090] When performing molecular dynamics calculations in the physical properties calculation unit 310, the calculations can be performed under the following conditions.

[0091] -MD calculation for absolute viscosity- For a system consisting of a single molecule with approximately 10,000 atoms (more specifically, between 9,000 and 11,000 atoms), a 5 ns equilibration calculation is performed using an NPT ensemble at a reference temperature Ts [K] (e.g., 300 K) and a pressure of 1 [atm]. Subsequently, a 5 ns production run is performed using an NVE ensemble. The force field is a modified GAFF2 force field for alkanes fluorides (Reference 1: J. Trag and D. Zahn, J. Mol. Model. 25, 39 (2019)), and the atomic charge is the RESP charge. In the NPT ensemble, a nose-hoover thermal bath is used for temperature control, a nose-hoover barostat for pressure control, and LAMMPS is used to execute the MD simulation. The Green-Kubo linear response theory (Reference 2: Green, M. S, J. Chem. Phys. 22, 398-413 (1954); Reference 3: Kubo, R, J. Phys. Soc. 12, 570-586 (1957)) is applied to the results of the MD simulation, and the absolute viscosity μ is calculated using the following equation (1) with respect to shear stress and other factors. Thus, the absolute viscosity μ(Ts) at the reference temperature Ts can be calculated for each material in the first candidate material group.

[0092]

number

[0093] -MD calculations for density and self-diffusion coefficients- For a system consisting of a single molecule with approximately 10,000 atoms (more specifically, between 9,000 and 11,000 atoms), a 5 ns production run is performed using an NPT ensemble at a reference temperature Ts [K] (e.g., 300 K) and a pressure of 1 [atm]. The force field is a modified GAFF2 force field for alkanes fluorides (Reference 1), and the atomic charge is RESP charge. A nose-Hoover thermal bath is used for temperature control, a nose-Hoover barostat for pressure control, and LAMMPS is used to perform the MD simulation. Furthermore, from the results of the MD simulation, the density ρ(Ts) and self-diffusion coefficient D(Ts) at the reference temperature Ts can be calculated for each substance in the first candidate group using equations (2) and (3) below. The reference temperature Ts can be set within a predetermined temperature range Tx~Ty [K], for example, 300 K.

[0094]

number

[0095]

number

[0096] -MD calculations for density and self-diffusion coefficient at an arbitrary temperature T- In this calculation, a 5 ns production run is performed on a system consisting of approximately 10,000 atoms (more specifically, between 9,000 and 11,000 atoms) composed of single molecules, using an NPT ensemble at a temperature T [K] and a pressure of 1 [atm]. The force field is the GAFF2 force field modified for alkanes fluorides (Reference 1), and the atomic charge is the RESP charge. A nose-Hoover thermal bath is used for temperature control, a nose-Hoover barostat for pressure control, and LAMMPS is used to execute the MD simulation. Furthermore, the density ρ(T) and self-diffusion coefficient D(T) at an arbitrary temperature T [K] are calculated for each substance in the first candidate group at the results of the MD simulation using the above equation.

[0097] In the physical property prediction unit 320, a function for predicting the kinematic viscosity at an arbitrary temperature T is calculated based on the acquired absolute viscosity and density at a reference temperature. Specifically, when predicting the kinematic viscosity at an arbitrary temperature T, the calculation can be performed as follows.

[0098] - Prediction of absolute viscosity at arbitrary temperatures - Based on the absolute viscosity, density, and self-diffusion coefficient calculated by MD calculation, the absolute viscosity μ(T) at any temperature T within a given temperature range Tx~Ty[K] can be calculated using the following equation (2) or equation (3). In equations (2) and (3), the constant b1 can be determined by fitting it to measured data of the absolute viscosity μ(T) at any temperature T.

[0099]

number

[0100] - Prediction of kinematic viscosity at arbitrary temperatures - Based on the absolute viscosity μ(T) and density ρ(T) predicted by equations (1) to (4), the kinematic viscosity ν(T) at an arbitrary temperature T is calculated using equation (5) below.

[0101]

number

[0102] In step S34, the second physical property prediction unit 330 receives a plurality of physical property values, including the predicted value of kinematic viscosity ν(T) at an arbitrary temperature T predicted by the physical property prediction unit 320. The second physical property prediction unit 330 determines whether the predicted values ​​of the plurality of physical properties in a predetermined temperature range meet a reference value and outputs a second candidate group of substances that are predicted to meet the reference value. The prediction result may include a plurality of predicted values ​​of physical properties, including kinematic viscosity in a predetermined temperature range Tx~Ty[K] obtained in step S33. The information of the second candidate group of substances may include at least one of the identification information of the substance to be predicted or the descriptor of the substance to be predicted.

[0103] Here, the "reference value" in the predicted values ​​of multiple physical properties, including kinematic viscosity, and other physical properties can be appropriately set according to the desired application. Specifically, when the physical property is kinematic viscosity, it is preferable that the kinematic viscosity in a predetermined temperature range Tx~Ty[K] is less than or equal to the reference value, depending on the intended application. As reference values, for example, reference values ​​such as 20[cSt], 10[cSt], and 5[cSt] can be set in advance as judgment indicators. Note that the unit [cSt] is 10 -3 [m 2 It can be converted to [ / s]. By determining whether the kinematic viscosity meets the standard value, substances whose kinematic viscosity increases rapidly and therefore cannot function as a liquid in a given temperature range can be excluded.

[0104] The prediction device 30 transmits the prediction results of the physical properties to the terminal device 40. The terminal device 40 receives the second candidate group of substances from the prediction device 30. The terminal device 40 presents the received second candidate group of substances to the user. The terminal device 40 may display the predicted values ​​included in the prediction results on the display device 506. The terminal device 40 may also display the information used for the prediction along with the predicted values.

[0105] Based on the above, in the second prediction process of the step of outputting the second candidate material group, multiple physical properties, including the kinematic viscosity of the first candidate material group, are predicted by machine learning and / or simulation, and the second candidate material group is output in which the predicted values ​​of the multiple physical properties in the predetermined temperature range are predicted to meet the reference value.

[0106] The user of the liquid material prediction device 1000 may refer to the predicted physical properties displayed on the display device 506 of the terminal device 40 and use them for material development. For example, the user can select a liquid material that functions as a liquid in a predetermined temperature range from among multiple predictable substances as a liquid material for a specific application.

[0107] [Physical properties] Other physical properties besides kinematic viscosity can be appropriately set according to the desired application, and preferably further include thermal properties and / or electrical properties. The thermal properties preferably include thermal conductivity and / or specific heat. The electrical properties preferably include one or more properties selected from the group consisting of dielectric constant, dielectric loss tangent, electrical conductivity, volume resistivity, and dielectric breakdown voltage.

[0108] By further selecting substances from the second group of candidate materials that meet the criteria values ​​for physical properties other than kinematic viscosity, it is possible to screen for candidate liquid materials that not only function as liquids in a predetermined temperature range but also possess suitable properties for the desired application, with low computational cost, high accuracy, and high efficiency.

[0109] [Application] There are no particular restrictions on the applications of the liquid material, and it can be appropriately selected according to the purpose. Examples include heat transfer fluids, solvents, cleaning agents, propellants, fire extinguishing agents, dielectrics, and lubricants. These may be used individually or in combination of two or more. The heat transfer fluid is preferably either a cooling heat transfer fluid or a heating heat transfer fluid. The solvent is preferably either a reaction solvent or an extraction solvent.

[0110] Below are examples of suitable combinations of physical properties and their reference values ​​for various applications. These combinations are not limiting, and appropriate selections can be made depending on the purpose.

[0111] -Heating medium- There are no particular restrictions on the specified temperature range Tx~Ty[K] for the application of a heat transfer medium; it can be appropriately selected according to the purpose. For example, for cooling heat transfer mediums, ranges such as 173~473[K] and 298~423[K] are used.

[0112] Desired physical properties of a heat transfer medium include low kinematic viscosity, high thermal conductivity, high specific heat, and low dielectric constant within a given temperature range Tx~Ty[K]. Criteria for meeting the standard values ​​include, for example, a kinematic viscosity of 10[cSt] or less, a thermal conductivity of 0.05[W / m·K] or more, a specific heat of 1.0[J / (g·K)] or more, and a dielectric constant of 10 or less.

[0113] - Lubricant - There are no particular restrictions on the specified temperature range Tx~Ty[K] for lubricant applications; it can be appropriately selected according to the purpose, for example, 273~423[K].

[0114] Desired physical properties of a lubricant include high kinematic viscosity in a given temperature range Tx~Ty[K]. Criteria for meeting the standard value include, for example, a kinematic viscosity of 2~6000[cSt].

[0115] (Compound for liquid materials) The liquid material compound of this embodiment can be predicted by the liquid material prediction method of this embodiment, and can provide a liquid material compound that functions as a liquid in the predetermined temperature range. [Examples]

[0116] The present invention will be described more specifically below based on examples, but the present invention is not limited to the following examples.

[0117] <Output of the first candidate substance group> <<Machine Learning>> -Construction of a boiling point prediction model- To construct the boiling point prediction model, we collected the structures of the target substances, as well as their physical properties, including their melting and boiling points, from the PubChem database. For the features that quantify the structure and characteristics of the substances, we adopted Mordred descriptors, which can be converted into high-dimensional vectors with 1826 dimensions. In data preprocessing, to avoid multicollinearity, we removed one of any feature pairs in the target substances whose correlation coefficient was 0.9 or higher when comparing the two features. Furthermore, to exclude outliers, we performed a loop process to remove data outside the two-sided 3σ range for each feature distribution, keeping only the data within that range. Additionally, to remove features without a distribution, we removed features with zero variance, and then standardized the remaining feature data to mean 0 and variance 1. Using this training data, from which duplicate data and outliers had been removed, we constructed a boiling point prediction model using Random Forest regression.

[0118] -Construction of a melting point prediction model- Similar to the boiling point prediction model, a melting point prediction model was constructed using random forest regression on the aforementioned training data.

[0119] - Prediction of melting and boiling points - Figure 7 is a graph showing the relationship between the predicted boiling point values ​​from the boiling point prediction model and the literature values ​​for the group of substances being studied. Figure 8 is a graph showing the relationship between the predicted melting point values ​​from the melting point prediction model and the literature values ​​for the group of substances being studied. In the figures, the vertical axis represents the predicted value [K], the horizontal axis represents the literature value [K] from the PubChem database, and each dot represents each substance in the group of substances being studied.

[0120] As a result, the boiling point prediction model showed a correlation coefficient of 0.83, and the melting point prediction model showed a correlation coefficient of 0.91, indicating high prediction accuracy in both cases. Therefore, it was found that the predicted boiling and melting points obtained by machine learning models can replace experimental values ​​and can be used to predict the liquid state within a given temperature range.

[0121] <<Output of the first candidate material group>> A first candidate substance group predicted to be in a liquid state in a predetermined temperature range was output. Specifically, the same substance group as the learning target substance group was used as the prediction target substance. As substances predicted to be in a liquid state in the temperature range Tx to Ty: 193 K to 333 K, the first candidate substance group satisfying predicted melting point Tm < Tx and Ty < predicted boiling point Tb was selected and output.

[0122] <Output of Second Candidate Substance Group> First, the absolute viscosity, density, self-diffusion coefficient, and kinematic viscosity of the first candidate substance group were predicted by molecular dynamics calculation described below. Specifically, 300 K was set as the reference temperature to calculate the absolute viscosity and density, and the temperature-dependent value of kinematic viscosity was calculated based on a function including the absolute viscosity and density at the reference temperature, which was taken as the predicted value.

[0123] <<Molecular Dynamics (MD) Calculation>> - MD Calculation for Absolute Viscosity - In this calculation, an equilibration calculation of 5 ns was performed in the NPT ensemble at a temperature of 300 [K] and a pressure of 1 [atm] for a system composed of a single molecule with about 10,000 atoms (specifically, 9,000 atoms or more and 11,000 atoms or less). After that, an equilibration calculation of 5 ns was performed in the NPT ensemble. Thereafter, a 5 ns production run was performed in the NVE ensemble. As the force field, the GAFF2 force field modified for fluorinated alkanes (Reference 1) was applied, and as the atomic charge, the RESP charge was applied. In the NPT ensemble, the Nose-Hoover thermostat was used for temperature control, the Nose-Hoover barostat was used for pressure control, and LAMMPS was used to execute the MD simulation. The Green-Kubo linear response theory (References 2 and 3) was applied to the execution result of the MD simulation, and the absolute viscosity μ was calculated by the following formula (1) using the shear stress and the like. Through the above steps, the absolute viscosity μ(300) at 300 K was calculated for each substance in the first candidate substance group.

[0124] [Mathematical Formula]

[0125] Figure 9 is a graph showing the calculated absolute viscosity at 300K for the first candidate group of materials. Figure 10 is a graph showing the predicted absolute viscosity at 300K for the first candidate group of materials. In Figure 9, the vertical axis shows the calculated absolute viscosity [Pa·s], and in Figure 10, the vertical axis shows the predicted absolute viscosity [Pa·s]. In Figures 9 and 10, the horizontal axis shows the measured absolute viscosity [Pa·s], and each dot represents the value for each material in the first candidate group of materials.

[0126] The calculated absolute viscosity in Figure 9 is calculated using equation (1). However, the calculated absolute viscosity is almost always larger than the measured value, indicating a tendency for the calculated value to be overestimated. Therefore, as described later, the density and self-diffusion coefficient were calculated and the calculated absolute viscosity of each substance was corrected. The predicted absolute viscosity in Figure 10 is the value corrected using the correction formula: (a1-a2ρ(300))μ(300)+a3. Here, μ(300) is the calculated absolute viscosity of each substance at 300K, ρ(300) is the density of each substance at 300K, and a1, a2, and a3 are constants determined by fitting the measured absolute viscosity μ(300) at an arbitrary temperature of 300K to the measured data. As a result, it was found that correcting with a correction formula that includes density as a variable improves the correlation with the measured absolute viscosity, allowing for a more accurate prediction of the absolute viscosity at an arbitrary temperature T.

[0127] -MD calculations for density and self-diffusion coefficients- In this calculation, a 5 ns production run was performed on a system consisting of approximately 10,000 atoms (more specifically, between 9,000 and 11,000 atoms) composed of single molecules, using an NPT ensemble at a temperature of 300 [K] and a pressure of 1 [atm]. The force field used was the GAFF2 force field modified for alkanes fluorides (Reference 1), and the atomic charge was determined by the RESP charge. A nose-Hoover thermal bath was used for temperature control, a nose-Hoover barostat for pressure control, and LAMMPS was used to perform the MD simulation. Furthermore, the density ρ(300) and self-diffusion coefficient D(300) at 300 K for each substance in the first candidate group were calculated from the results of the MD simulation using the following equations (2) and (3).

[0128]

number

[0129]

number

[0130] -MD calculations for density and self-diffusion coefficient at an arbitrary temperature T- In this calculation, a 5 ns production run was performed using an NPT ensemble at temperature T [K] and pressure 1 [atm] for a system consisting of approximately 10,000 atoms (specifically, between 9,000 and 11,000 atoms) composed of single molecules. The force field used was the GAFF2 force field modified for alkanes fluorides (Reference 1), and the atomic charge was determined by the RESP charge. A nose-Hoover thermal bath was used for temperature control, a nose-Hoover barostat for pressure control, and LAMMPS was used to perform the MD simulation. Furthermore, the density ρ(T) and self-diffusion coefficient D(T) at an arbitrary temperature T [K] were calculated for each substance in the first candidate group at the above equation based on the results of the MD simulation.

[0131] - Prediction of absolute viscosity at arbitrary temperatures - Based on the absolute viscosity, density, and self-diffusion coefficient calculated by MD calculation, the absolute viscosity μ(T) at an arbitrary temperature T[K] was calculated using the following equation (2-1) or equation (3-1). In equations (2-1) and (3-1), the constant b1 can be determined by fitting it to the measured data of absolute viscosity μ(T) at an arbitrary temperature T.

[0132]

number

[0133] - Prediction of kinematic viscosity at arbitrary temperatures - Based on the absolute viscosity μ(T) and density ρ(T) predicted by equations (1) to (4), the kinematic viscosity ν(T) at an arbitrary temperature T was calculated using equation (5) below.

[0134]

number

[0135] <<Output of the second candidate material group>> Next, a second group of candidate materials was output, in which the predicted values ​​of multiple physical properties in the temperature range of 193K to 323K were predicted to meet the standard values. Specifically, in the temperature range of 193K to 323K, the kinematic viscosity was 10 [cSt] (= 10 -3 [m 2 A second group of candidate materials was selected and output, satisfying the following conditions: ≤ / s, thermal conductivity of 0.05 [W / m·K] or higher, specific heat of 1.0 [J / (g·K)] or higher, and relative permittivity of 10 or less.

[0136] Figure 11 is a graph showing the predicted kinematic viscosity of substances A to C of the first candidate substance group in the comparative example, based on the prior art. Figure 12 is a graph showing the predicted kinematic viscosity of substances A to C of the first candidate substance group in the example. In the figures, the vertical axis represents the predicted kinematic viscosity [cSt], the horizontal axis represents the temperature [K], each dot represents the measured kinematic viscosity of substances A to C selected from the first candidate substance group, and the solid and dashed lines represent the predicted kinematic viscosity of substances A to C.

[0137] Here, Figure 11 shows the calculated kinematic viscosity obtained by substituting the absolute viscosity calculated using equation (1) into equation (5). Figure 12 shows the predicted kinematic viscosity obtained by substituting the values ​​into equations (1) to (4) after performing fitting based on equations (1) to (4).

[0138] As a result, in the examples, by calculating and fitting based on the absolute viscosity, density, and self-diffusion coefficient calculated by MD calculation, it was possible to predict kinematic viscosity with higher accuracy, showing a high correlation with the measured values. This allows us to predict, for example, that substances A and B have a sharp increase in kinematic viscosity at 200K (-73℃), and are likely in a solidified state or unsuitable for liquid material applications, and thus exclude them from the prediction. Therefore, candidate liquid materials that function as liquids in a given temperature range can be screened with low computational cost, high accuracy, and high efficiency. On the other hand, it was found that in the conventional technique, there was a discrepancy between the measured values ​​(dots) and the simulation (line), making it impossible to make predictions that accurately reflected the actual situation.

[0139] <Prediction of additional physical properties> In this calculation, a 5 ns production run was performed on a 10,000-atom system composed of single molecules using an NPT ensemble at a temperature of 300 K and a pressure of 1 atm. The force field used was a GAFF2 force field modified for alkanes fluorides, and the atomic charge was determined by RESP charge. A nose-Hoover thermal bath was used for temperature control, a nose-Hoover barostat for pressure control, and LAMMPS was used to perform the MD simulation. Furthermore, from the results of the MD simulation, the density at 300 K, the kinematic viscosity at 300 K, and the relative permittivity at 300 K were calculated for each substance in the first candidate group using equation (2), equations (4-1) and (5), and equation (6) below.

[0140]

number

[0141] Furthermore, using HSPiP, a practical Hansen solubility parameter software, we obtained predicted values ​​for the boiling point, thermal conductivity, and specific heat at 300K for each substance in the first candidate group.

[0142] Figures 13-18 are graphs comparing experimental and predicted values ​​for density, kinematic viscosity, relative permittivity, boiling point, thermal conductivity, and specific heat. In the figures, the vertical axis shows the predicted value for each parameter, and the horizontal axis shows the experimental value; all show a high correlation. The units in each figure are density [kg / m³]. 3 These are the viscosity [cSt], relative permittivity, boiling point [K], thermal conductivity [W / (m·K)], and specific heat [J / (g·K)].

[0143] By further selecting substances that meet additional physical property criteria from the first group of candidate substances that meet the viscosity criteria in a predetermined temperature range, it is possible to screen for candidate liquid materials that not only function as liquids in a predetermined temperature range but also possess suitable properties for desired applications, with low computational cost, high accuracy, and high efficiency.

[0144] Although the present invention has been described above based on various embodiments, the present invention is not limited to the requirements shown in the above embodiments. These points can be modified as long as they do not impair the spirit of the present invention, and can be appropriately determined according to their application.

[0145] <References> Reference 1: J. Trag and D. Zahn, Improved GAFF2 Parameters for Fluorinated Alkanes and Mixed Hydro- and Fluorocarbons, J. Mol. Model. 25, 39 (2019). Reference 2: Green, M. S, Markoff random processes and the statistical mechanics of time-dependent phenomena, ii. irreversible processes in fluids, J. Chem. Phys. 22, 398-413 (1954). Reference 3: Kubo, R, Statistical-mechanical theory of irreversible processes. i. general theory and simple applications to magnetic and conduction problems, J. Phys. Soc. Jpn. 12, 570-586 (1957). [Explanation of Symbols]

[0146] 10: Learning device 20: Prediction device (First prediction device) 30: Prediction device (second prediction device) 40: Terminal device 101: Data storage unit 110: Descriptor calculation section 120: Training Data Generation Unit 130: Model generation unit 201: Model Memory Unit 210: Information acquisition department 220: Descriptor calculation section 230: First Physical Property Prediction Unit 301: Data storage unit 310: Physical property calculation department 320: Material Properties Prediction Unit 330: Second Physical Property Prediction Unit 1000: Liquid material prediction device

Claims

1. A liquid material prediction device having a control unit that predicts a liquid material that functions as a liquid in a predetermined temperature range, The control unit, By inputting the structure of the target substance into a trained model that has been trained using training data including the structure, melting point, and boiling point of the target substance, the melting point and boiling point of the target substance are predicted, and a first group of candidate substances that are predicted to be in a liquid state within the predetermined temperature range are output. The system predicts multiple physical properties, including kinematic viscosity, of a first group of candidate materials using machine learning and / or simulation, and outputs a second group of candidate materials in which the predicted values ​​of the multiple physical properties in a predetermined temperature range are expected to meet a standard value. Liquid material prediction device.

2. The liquid material prediction apparatus according to claim 1, wherein the simulation includes quantum chemical calculations and / or molecular dynamics calculations.

3. The liquid material prediction device according to claim 1, wherein the temperature-dependent predicted value of the kinematic viscosity is a value output based on a function that includes the absolute viscosity and density at a reference temperature calculated by molecular dynamics calculations.

4. A liquid material prediction device having a control unit that predicts a liquid material that functions as a liquid in a predetermined temperature range, The control unit, The system predicts multiple physical properties of a target substance, including its kinematic viscosity, using machine learning and / or simulation. The temperature-dependent predicted value of the kinematic viscosity is output as a function that includes the self-diffusion coefficient at a reference temperature calculated by molecular dynamics calculations, and also includes the self-diffusion coefficient and density at an arbitrary temperature within a predetermined temperature range. The system then outputs a group of candidate substances in which the predicted values ​​of the multiple physical properties within the predetermined temperature range are expected to satisfy a reference value. Liquid material prediction device.

5. The liquid material prediction device according to claim 1 or 4, wherein the temperature-dependent predicted value of the kinematic viscosity is a value output as a function that includes the self-diffusion coefficient at a reference temperature calculated by molecular dynamics calculation, and also includes the self-diffusion coefficient and density at an arbitrary temperature in the predetermined temperature range.

6. The liquid material prediction apparatus according to claim 1, wherein the plurality of physical properties further include thermal properties and / or electrical properties.

7. The liquid material prediction apparatus according to claim 6, wherein the aforementioned thermal properties include thermal conductivity and / or specific heat.

8. The liquid material prediction apparatus according to claim 6, wherein the aforementioned electrical properties include one or more properties selected from the group consisting of dielectric constant, dielectric loss tangent, electrical conductivity, volume resistivity, and dielectric breakdown voltage.

9. The liquid material prediction apparatus according to claim 1, wherein the application of the liquid material is selected from the group consisting of a heat transfer medium, a solvent, a cleaning agent, a propellant, a fire extinguishing agent, a lubricant, and a dielectric.

10. The liquid material prediction apparatus according to claim 9, wherein the heat transfer medium is either a cooling heat transfer medium or a heating heat transfer medium.

11. The liquid material prediction apparatus according to claim 9, wherein the solvent is either a reaction solvent or an extraction solvent.

12. A liquid material prediction method that uses a computer to predict a liquid material that functions as a liquid in a predetermined temperature range, The aforementioned computer, The process involves inputting the structure of a target substance into a trained model, which has been trained using training data including the structure, melting point, and boiling point of the target substance, thereby predicting the melting point and boiling point of the target substance, and outputting a first group of candidate substances that are predicted to be in a liquid state within a predetermined temperature range. A step of predicting multiple physical properties, including kinematic viscosity, of a first candidate group of materials by machine learning and / or simulation, and outputting a second group of materials in which the predicted values ​​of the multiple physical properties in a predetermined temperature range are predicted to meet a reference value, A liquid material prediction method that performs the following.

13. A liquid material prediction method that uses a computer to predict a liquid material that functions as a liquid in a predetermined temperature range, The aforementioned computer, The process involves predicting multiple physical properties of a target substance, including its kinematic viscosity, using machine learning and / or simulation, wherein the temperature-dependent predicted value of the kinematic viscosity is output as a function that includes the self-diffusion coefficient at a reference temperature calculated by molecular dynamics calculations, and the self-diffusion coefficient and density at an arbitrary temperature within the predetermined temperature range, and outputting a group of candidate substances in which the predicted values ​​of the multiple physical properties within the predetermined temperature range are predicted to satisfy a reference value. A liquid material prediction method that performs the following.

14. A program for predicting liquid materials that function as liquids within a predetermined temperature range, On the computer, The process involves inputting the structure of a target substance into a trained model, which has been trained using training data including the structure, melting point, and boiling point of the target substance, thereby predicting the melting point and boiling point of the target substance, and outputting a first group of candidate substances that are predicted to be in a liquid state within a predetermined temperature range. A step of predicting multiple physical properties, including kinematic viscosity, of a first candidate group of materials by machine learning and / or simulation, and outputting a second group of materials in which the predicted values ​​of the multiple physical properties in a predetermined temperature range are predicted to meet a reference value, A program to execute.

15. A program for predicting liquid materials that function as liquids within a predetermined temperature range, On the computer, The process involves predicting multiple physical properties of a target substance, including its kinematic viscosity, using machine learning and / or simulation, wherein the temperature-dependent predicted value of the kinematic viscosity is output as a function that includes the self-diffusion coefficient at a reference temperature calculated by molecular dynamics calculations, and the self-diffusion coefficient and density at an arbitrary temperature within the predetermined temperature range, and outputting a group of candidate substances in which the predicted values ​​of the multiple physical properties within the predetermined temperature range are predicted to satisfy a reference value. A program to execute.

16. A compound for liquid materials that is predicted by the liquid material prediction method described in claim 12 or 13 and functions as a liquid in the predetermined temperature range.

Images