Prediction apparatus, method for prediction, program, and experimental system

Abstract

To provide a technique for reconciling physical property values acquired in different systems.SOLUTION: One aspect of the present disclosure relates to a prediction apparatus including: an acquisition unit for acquiring material information of a material fabricated in a first system according to first fabrication conditions; and a processing unit for deriving the physical property predicted value of the fabricated material from the acquired material information by using a physical property prediction model trained to derive a physical property predicted value from material information. The physical property prediction model is updated by a physical property evaluation value by a second system with respect to the material.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a prediction device, a prediction method, a program, and an experimental system. [Background technology]

[0002] In recent years, autonomous experiments using machine learning and robotics technologies have been put into practical use. Using an experimental system that enables such autonomous experiments can eliminate the tedious work of conducting experiments and prevent mistakes such as repeating the same experiment or making mistakes. It can also perform experiments that are difficult for some people, such as those involving the handling of hazardous substances or in adverse environments. Furthermore, it can conduct experiments much faster than human labor, and by applying data science techniques such as Bayesian optimization, it is possible to conduct efficient experiments in a short period of time.

[0003] Patent Document 1 proposes a fabrication evaluation system that can be used in such an experimental system. In this fabrication evaluation system, an estimation device sequentially searches for optimal sample fabrication conditions using a Bayesian optimization method, and fabricates a sample according to each fabrication condition. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2021 / 044913 Summary of the Invention [Problem to be solved by the invention]

[0005] Samples prepared according to each preparation condition are measured, and the next preparation condition is determined based on the measurement results of the sample. Chemical manufacturers manufacture and develop compounds according to specifications requested by customers, but they must manufacture and develop compounds so that they satisfy the physical property values ​​required by the specifications. These physical property values ​​are predicted within the chemical manufacturer based on the measurement results obtained from the prepared samples. Typically, such predicted physical property values ​​are derived using a property prediction model.

[0006] On the other hand, the physical properties of compounds delivered by chemical manufacturers to their customers may also be evaluated by the customers. The predicted physical properties obtained within the chemical manufacturer do not necessarily match the evaluated physical properties, and the discrepancy between the predicted and evaluated physical properties may be significant. In this case, it may be necessary to adjust the property prediction model used within the chemical manufacturer.

[0007] In view of the above problems, one object of the present disclosure is to provide a technique for matching physical property values ​​obtained in different systems. [Means for solving the problem]

[0008] One aspect of the present disclosure relates to a prediction device having an acquisition unit that acquires material information of a material created in a first system according to first production conditions, and a processing unit that derives predicted physical property values ​​of the created material from the acquired material information using a physical property prediction model trained to derive predicted physical property values ​​from the material information, wherein the physical property prediction model is updated by physical property evaluation values ​​for the material by a second system. [Effects of the Invention]

[0009] According to the present disclosure, a technique can be provided for matching physical property values ​​obtained in different systems. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a schematic diagram illustrating an experimental procedure according to one embodiment of the present disclosure. [Figure 2] FIG. 2 is a schematic diagram illustrating an experimental system according to one embodiment of the present disclosure. [Figure 3] FIG. 3 is a block diagram illustrating a hardware configuration of a prediction device according to an embodiment of the present disclosure. [Figure 4] FIG. 4 is a block diagram illustrating a functional configuration of a prediction device according to an embodiment of the present disclosure. [Figure 5] FIG. 5 is a flowchart illustrating an experiment process according to one embodiment of the present disclosure. [Figure 6] FIG. 6 is a flowchart illustrating a retraining process for a property prediction model according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0012] In the following embodiment, a prediction device that predicts the physical property values ​​of a substance using a physical property prediction model is disclosed.

[0013] [Summary of the Disclosure] In an experimental process according to an embodiment of the present disclosure, a manufacturer side 1 receives a request from a customer side 2 and then manufactures and develops a substance such as a compound to meet the requested specifications. For example, as shown in Fig. 1, the manufacturer side 1 manufactures samples under various manufacturing conditions, measures the manufactured samples, and derives the physical property values ​​of the manufactured samples based on the measurement results.

[0014] In the illustrated example, a substance is produced according to production conditions, and substance information is obtained as a measurement result of the produced substance. Then, predicted physical property values ​​are obtained from the substance information using a physical property prediction model 50. The physical property prediction model 50 may be realized as a machine learning model trained to derive predicted physical property values ​​from the substance information, or may be realized as a physical property prediction formula, such as a multiple regression formula, derived from data on the substance information and physical property values.

[0015] Thereafter, the manufacturing condition determination model 60 is used to search for the next manufacturing conditions based on the specifications from the customer side 2, the current manufacturing conditions, and the predicted physical property values ​​output from the physical property prediction model 50, and these are used to manufacture the next material. Here, the manufacturing condition determination model 60 can be realized as a Bayesian optimization model. The above-mentioned experimental cycle is repeatedly executed until the experiment termination condition is satisfied.

[0016] When the customer side 2 receives a manufactured and developed material from the manufacturer side 1, the customer side 2 independently measures the delivered material and evaluates the material's physical properties. The physical property evaluation values ​​evaluated by the customer side 2 can be fed back to the manufacturer side 1. At this time, there may be a discrepancy between the physical property prediction values ​​obtained by the manufacturer side 1 and the physical property evaluation values ​​fed back from the customer side 2. If there is a significant discrepancy, the physical property prediction model 50 currently used by the manufacturer side 1 needs to be retrained to reduce the discrepancy.

[0017] In the following example, as shown in FIG. 2, the experimental system 10 on the manufacturer side 1 has a manufacturing device 30 and a prediction device 100, and the prediction device 100 uses a physical property prediction model 50 and a manufacturing condition determination model 60.

[0018] The production device 30 produces a substance such as a compound in accordance with the production conditions acquired from the production condition determination model 60. The production device 30 can be realized, for example, as an experimental robot that performs an experiment in the experimental system 10 and produces a sample or the like. The produced substance is measured, and the measurement results are provided to the prediction device 100 as physical information of the substance.

[0019] The prediction device 100 first inputs material information of the produced material into a physical property prediction model 50 to obtain predicted physical property values. In the illustrated embodiment, the physical property prediction model 50 is installed in the prediction device 100, but the present disclosure is not limited to this, and the physical property prediction model 50 may be stored in a server or the like (not shown) external to the prediction device 100. In this case, the prediction device 100 may, for example, transmit physical information to the server via a network (not shown) and receive processing results of the physical property prediction model 50 from the server.

[0020] Thereafter, the prediction device 100 inputs the predicted physical property values ​​acquired from the physical property prediction model 50 into the manufacturing condition determination model 60 to acquire the next manufacturing conditions. In the illustrated embodiment, the manufacturing condition determination model 60 is installed in the prediction device 100, but the present disclosure is not limited to this, and the manufacturing condition determination model 60 may be stored in a server or the like (not shown) external to the prediction device 100. In this case, the prediction device 100 may, for example, transmit the predicted physical property values ​​to the server via a network (not shown), and receive the processing results of the manufacturing condition determination model 60 from the server.

[0021] The prediction device 100 notifies the production device 30 of the next production conditions, and the production device 30 produces a substance such as a compound according to the notified next production conditions. This experiment cycle is repeated until the experiment end condition is satisfied.

[0022] When the material thus produced meets the specifications requested by the customer, the produced material is delivered from the manufacturer side 1 to the customer side 2. After that, the physical property evaluation values ​​of the delivered material, which are independently evaluated by the customer side 2, are fed back to the manufacturer side 1, and the manufacturer side 1 compares the physical property evaluation values ​​fed back from the customer side 2 for the delivered material with the physical property prediction values ​​of the material according to the physical property prediction model 50.

[0023] If the discrepancy between the predicted physical property values ​​obtained from the physical property prediction model 50 and the physical property evaluation values ​​at the customer side 2 is equal to or greater than a predetermined threshold, the manufacturer side 1 determines that the physical property prediction model 50 needs to be retrained, and the physical property prediction model 50 is retrained using the physical property evaluation values ​​fed back from the customer side 2 so that the discrepancy between the predicted physical property values ​​and the physical property evaluation values ​​becomes less than the predetermined threshold. On the other hand, if the discrepancy between the predicted physical property values ​​obtained from the physical property prediction model 50 and the physical property evaluation values ​​at the customer side 2 is less than the predetermined threshold, the manufacturer side 1 determines that the currently used physical property prediction model 50 is consistent with the physical property evaluation by the customer side 2, and continues to use the physical property prediction model 50. This allows the physical property prediction model 50 used at the manufacturer side 1 to match or approximate the physical property evaluation by the customer side 2.

[0024] The prediction device 100 may be realized by a computing device such as a server, a personal computer (PC), a smartphone, or a tablet, and may have a hardware configuration such as that shown in Fig. 3. That is, the prediction device 100 includes a drive device 101, a storage device 102, a memory device 103, a processor 104, a user interface (UI) device 105, and a communication device 106, which are interconnected via a bus B.

[0025] The programs or instructions that realize the various functions and processes in the prediction device 100 may be stored in a removable storage medium such as a CD-ROM (Compact Disk-Read Only Memory) or a flash memory. When the storage medium is set in the drive device 101, the programs or instructions are installed from the storage medium to the storage device 102 or the memory device 103 via the drive device 101. However, the programs or instructions do not necessarily have to be installed from the storage medium, and may be downloaded from any external device via a network or the like.

[0026] The storage device 102 is realized by a hard disk drive or the like, and stores installed programs or instructions as well as files, data, etc. used to execute the programs or instructions.

[0027] The memory device 103 is realized by a random access memory, a static memory, or the like, and when a program or instruction is activated, reads and stores the program, instruction, data, or the like from the storage device 102. The storage device 102, the memory device 103, and the removable storage medium may be collectively referred to as a non-transitory storage medium.

[0028] The processor 104 may be realized by one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), processing circuitry, etc., which may be composed of one or more processor cores, and performs various functions and processes of the prediction device 100 in accordance with programs, instructions, data such as parameters required to execute the programs or instructions, etc. stored in the memory device 103.

[0029] The user interface (UI) device 105 may be composed of input devices such as a keyboard, a mouse, a camera, a microphone, etc., output devices such as a display, a speaker, a headset, a printer, etc., and input / output devices such as a touch panel, and realizes an interface between a user and the prediction device 100. For example, a user operates the prediction device 100 by operating a GUI (Graphical User Interface) displayed on a display or a touch panel using a keyboard, a mouse, etc.

[0030] The communication device 106 is realized by various communication circuits that execute wired and / or wireless communication processing with external devices, the Internet, a LAN (Local Area Network), a cellular network, or other communication networks.

[0031] However, the above-described hardware configuration is merely an example, and the prediction device 100 according to the present disclosure may be realized by any other appropriate hardware configuration.

[0032] [Prediction device] Next, a description will be given of the prediction device 100 according to an embodiment of the present disclosure. Fig. 4 is a block diagram showing the functional configuration of the prediction device 100 according to an embodiment of the present disclosure.

[0033] 4, the prediction device 100 includes an acquisition unit 110 and a processing unit 120. The prediction device 100 may be realized by a program that causes one or more processors 104 to function as the acquisition unit 110 and / or the processing unit 120.

[0034] The acquisition unit 110 acquires material information of a material prepared in a first system according to first preparation conditions. Specifically, on the manufacturer side 1, a preparation device 30 prepares a sample of a compound or the like according to certain preparation conditions, and the prepared sample is measured. Then, the acquisition unit 110 acquires the measurement results of the prepared sample.

[0035] For example, in the development of a sheet-shaped sealing material (such as an adhesive for a foldable device), raw materials include, for example, epoxy resin, hardener, solvent, and other additives (such as tackifiers and leveling agents as appropriate), and the manufacturing apparatus 30 blends and stirs the raw material liquid and raw material powder, and then coats and dries the resulting mixture into a sheet.

[0036] Then, to measure the physical properties of the produced sheet-like sealing material, for example, a rheometer, DMS (dynamic viscoelasticity measurement), TMA (thermomechanical analysis), DSC (differential scanning calorimetry), etc. are performed to measure the complex viscosity, elastic modulus, Tg (glass transition temperature), reactivity, etc. The acquisition unit 110 acquires the measurement results of the complex viscosity, elastic modulus, Tg (glass transition temperature), reactivity, etc. as material information, and passes the acquired measurement results to the processing unit 120.

[0037] The processing unit 120 derives predicted physical property values ​​of the material produced from the acquired material information, using the material property prediction model 50 that has been trained to derive predicted physical property values ​​from material information.

[0038] For example, the physical property prediction model 50 is trained to receive material information of a prepared sample as an input and output predicted physical property values ​​of the sample. For example, in the case of a sheet-shaped sealing material, the physical property prediction model 50 may be trained to receive measurement results of complex viscosity, elastic modulus, Tg (glass transition temperature), reactivity, etc. as input and output predicted physical property values ​​such as number of possible folds m [10,000 times] and touch pen resistance n [N].

[0039] It should be noted that the physical property prediction model 50 does not need to directly output the number of times m [ten thousand times] that can be bent and the touch pen resistance n [N] from the measurement results of complex viscosity, elastic modulus, Tg (glass transition temperature), reaction rate, etc., but may be configured, for example, to first derive a predicted value of the stress exerted when bending from the measurement results of complex viscosity, elastic modulus, Tg (glass transition temperature), reaction rate, etc., and then output the number of times m [ten thousand times] that can be bent from the derived predicted stress value.

[0040] The physical property prediction model 50 can be realized as a machine learning model such as a neural network model, but the physical property prediction model 50 according to the present disclosure is not limited to this and may be realized by any model capable of deriving a predicted physical property value from material information, such as an equation derived by statistical analysis such as a multiple regression equation. Furthermore, the physical property prediction model 50 may adopt a value obtained by subtracting a target value of the specifications from the predicted physical property value as the predicted physical property value.

[0041] The physical property prediction model 50 can be trained using a training data set consisting of material information and physical property values. For example, when the physical property prediction model 50 is realized as a neural network model, material information from the training data set is input to the physical property prediction model 50 to be trained, and processing results are obtained from the physical property prediction model 50 to be trained. Then, parameters of the physical property prediction model 50 are adjusted to reduce the error between the processing results and the physical property values ​​of the training data set. Such parameter adjustment is continued until a predetermined training end condition is satisfied. Here, the training process may be performed by a training device (not shown) within the manufacturer 10 or by a vendor external to the manufacturer 10.

[0042] Next, the processing unit 120 inputs the specifications instructed by the customer side 20, the manufacturing conditions applied to the manufacturing of the sample, the predicted physical property values ​​obtained from the physical property prediction model 50, etc. into the manufacturing condition determination model 60, and obtains the manufacturing conditions to be applied to manufacturing the next sample from the manufacturing condition determination model 60. For example, the manufacturing conditions may indicate the types of raw materials and their blending amounts (blending ratios).

[0043] Here, the fabrication condition determination model 60 may be obtained by, for example, Bayesian optimization. For more details, see, for example, International Publication No. 2021 / 044913. However, the fabrication condition determination model 60 according to the present disclosure may be realized by any other machine learning model or determination procedure capable of determining fabrication conditions based on specifications specified by the customer side 2, fabrication conditions applied to fabricate the sample, predicted physical property values ​​obtained from the physical property prediction model 50, etc.

[0044] When the next manufacturing conditions are acquired from the manufacturing condition determination model 60, the processing unit 120 provides the acquired next manufacturing conditions to the manufacturing device 30. The manufacturing device 30 manufactures the next sample in accordance with the next manufacturing conditions acquired from the prediction device 100.

[0045] The fabrication device 30 and the prediction device 100 repeat the above-described experiment cycle until the experiment end condition is satisfied.

[0046] Here, the physical property prediction model 50 can be updated by a physical property evaluation value for a substance obtained by a second system. That is, the manufacturer side 1 delivers a substance such as a developed compound to the customer side 2. The customer side 2 can then independently evaluate the physical properties of the delivered substance and feed back the physical property evaluation value to the manufacturer side 1. The manufacturer side 1 compares the physical property evaluation value fed back from the customer side 2 for the delivered substance with the physical property prediction value obtained from the physical property prediction model 50 for the substance, and determines whether the discrepancy between the physical property evaluation value and the physical property prediction value is equal to or greater than a predetermined threshold.

[0047] If the discrepancy between the physical property evaluation value and the physical property predicted value is less than a predetermined threshold, the manufacturer side 1 determines that the currently used physical property prediction model 50 is consistent with the physical property evaluation of the customer side 2, and continues to use the physical property prediction model 50. On the other hand, if the discrepancy between the physical property evaluation value and the physical property predicted value is equal to or greater than a predetermined threshold, the manufacturer side 1 determines that the currently used physical property prediction model 50 is inconsistent with the physical property evaluation of the customer side 2, and retrains the physical property prediction model 50 using the physical property evaluation value fed back from the customer side 2.

[0048] For example, the retraining of the physical property prediction model 50 may be performed by a training device (not shown) within the manufacturer side 1 or by a vendor outside the manufacturer side 1. For the retraining, physical property evaluation values ​​obtained from the customer side 2 are used.

[0049] Note that the retraining of the physical property prediction model 50 does not necessarily have to be performed based on whether the discrepancy between the physical property evaluation value and the physical property predicted value is equal to or greater than a predetermined threshold, and other conditions may be applied. For example, when the number of fed-back physical property evaluation values ​​reaches a predetermined number or more, the physical property prediction model 50 may be retrained using these physical property evaluation values.

[0050] According to the prediction device 100 described above, the property prediction model 50 used on the manufacturer side 1 can be made to match or approximate the property evaluation by the customer side 2, and substances such as compounds that are consistent with the property evaluation by the customer side 2 can be manufactured and developed.

[0051] [Experimental treatment] Next, an experimental process according to an embodiment of the present disclosure will be described. The experimental process is executed by the experimental system 10, and more specifically, may be realized by the fabrication device 30 and the prediction device 100 of the experimental system 10.

[0052] In step S101, the production apparatus 30 produces a substance in a first system according to production conditions. For example, the production apparatus 30 produces a sample according to production conditions (e.g., types of raw materials used to produce the sample, their blending amounts, blending ratios, etc.) notified by the prediction apparatus 100.

[0053] In step S102, prediction apparatus 100 acquires material information of the produced material. Specifically, the sample produced by production apparatus 30 is measured by a measurement device (not shown), and the measurement results are provided to prediction apparatus 100 as material information.

[0054] In step S103, the prediction device 100 derives predicted physical property values ​​from the substance information using the physical property prediction model 50. Specifically, the physical property prediction model 50, which has been trained to output predicted physical property values ​​from substance information, is made available to the experimental system 10, and the prediction device 100 inputs the substance information acquired in step S102 into the physical property prediction model 50 and acquires the predicted physical property values ​​from the physical property prediction model 50.

[0055] In step S104, the prediction device 100 determines whether an experiment termination condition has been satisfied. For example, the experiment termination condition may be that a predetermined number of experiments have been performed, or that the acquired predicted physical property values ​​have converged to a predetermined value. If the experiment termination condition has not been satisfied (S104: NO), the experiment processing proceeds to step S105. On the other hand, if the experiment termination condition has been satisfied (S104: YES), the experiment processing ends.

[0056] In step S105, the prediction device 100 determines the next fabrication conditions based on the specifications, the applied fabrication conditions, and the predicted physical property values ​​using the fabrication condition determination model 60. For example, the fabrication condition determination model 60 may be a model obtained by Bayesian optimization. Then, the experimental process returns to step S101, and the fabrication device 30 fabricates a sample according to the next fabrication conditions.

[0057] [Retraining Process] Next, a retraining process of the physical property prediction model 50 according to an embodiment of the present disclosure will be described. The retraining process may be performed by the experimental system 10 on the manufacturer side 1 or a training device (not shown) provided by a vendor outside the manufacturer side 1. Here, it is assumed that the experimental system 10 retrains the physical property prediction model 50.

[0058] In step S201, the experimental system 10 stores the physical property evaluation values ​​acquired from the customer side 2.

[0059] In step S202, the experimental system 10 determines whether a predetermined trigger condition is satisfied. For example, the predetermined trigger condition may be that the deviation between the predicted physical property value from the physical property prediction model 50 and the evaluated physical property value obtained from the customer side 2 is equal to or greater than a predetermined threshold, or that the number of evaluated physical property values ​​obtained from the customer side 2 is equal to or greater than a predetermined number. If the predetermined trigger condition is not satisfied (S202: NO), the experimental system 10 returns to step S201 and waits for the next evaluated physical property value to be obtained. On the other hand, if the predetermined trigger condition is satisfied (S202: YES), the experimental system 10 proceeds to step S203.

[0060] In step S203, the experimental system 10 retrains the property prediction model 50 using the stored property evaluation values.

[0061] According to the above-described experimental system 10, the physical property prediction model 50 used on the manufacturer side 1 can be made to match or approximate the physical property evaluation by the customer side 2, and substances such as compounds that are consistent with the physical property evaluation by the customer side 2 can be manufactured and developed.

[0062] In addition, the following supplementary notes are provided in relation to the above description. (Appendix 1) an acquisition unit that acquires material information of a material prepared in the first system under first preparation conditions; a processing unit that derives predicted physical property values ​​of the created material from the acquired material information by using a material property prediction model that has been trained to derive predicted physical property values ​​from material information; and A prediction device, wherein the physical property prediction model is updated by a physical property evaluation value for the substance by a second system. (Appendix 2) The prediction device according to claim 1, wherein the acquisition unit acquires the physical property evaluation value from the second system. (Appendix 3) 3. The prediction device of claim 2, wherein when a predetermined trigger condition is satisfied, the property prediction model is retrained using property evaluation values ​​obtained from the second system. (Appendix 4) A prediction device described in any of Appendices 1 to 3, wherein the processing unit uses a fabrication condition determination model trained to derive fabrication conditions from fabrication conditions, physical property values, and specifications, and obtains second fabrication conditions to be applied to the first system from the derived physical property prediction values. (Appendix 5) 5. The prediction device according to claim 4, wherein the production condition determination model is a Bayesian optimization model. (Appendix 6) acquiring material information of a material prepared in a first system under first preparation conditions; deriving predicted physical property values ​​of the created material from the acquired material information using a material property prediction model trained to derive predicted physical property values ​​from material information; and A computer-implemented prediction method, wherein the property prediction model is updated by property evaluation values ​​for the substance by a second system. (Appendix 7) acquiring material information of a material prepared in a first system under first preparation conditions; deriving predicted physical property values ​​of the created material from the acquired material information using a material property prediction model trained to derive predicted physical property values ​​from material information; on the computer, The property prediction model is updated by property evaluation values ​​for the substance obtained by a second system. (Appendix 8) a production device for producing a material in a first system according to production conditions; a prediction device that derives predicted physical property values ​​from material information of the prepared material; and The prediction device includes: an acquisition unit that acquires material information of a material prepared in the first system under first preparation conditions; a processing unit that derives predicted physical property values ​​of the created material from the acquired material information by using a material property prediction model that has been trained to derive predicted physical property values ​​from material information; and An experimental system in which the property prediction model is updated by property evaluation values ​​for the substance obtained by a second system.

[0063] Although the examples of the present disclosure have been described in detail above, the present disclosure is not limited to the specific embodiments described above, and various modifications and variations are possible within the scope of the gist of the present disclosure as set forth in the claims. [Explanation of symbols]

[0064] 1. Manufacturer side 2. Customer side 10 Experimental System 30 Manufacturing Equipment 50 Physical property prediction model 60 Manufacturing condition determination model 100 Prediction Device 110 Acquisition Department 120 Processing section

Claims

1. an acquisition unit that acquires material information of a material prepared in the first system under the first preparation conditions; a processing unit that derives predicted physical property values ​​of the created material from the acquired material information by using a material property prediction model that has been trained to derive predicted physical property values ​​from material information; and A prediction device, wherein the physical property prediction model is updated by a physical property evaluation value for the substance by a second system.

2. The prediction device according to claim 1 , wherein the acquisition unit acquires the physical property evaluation value from the second system.

3. The prediction device according to claim 2 , wherein when a predetermined trigger condition is satisfied, the physical property prediction model is retrained using physical property evaluation values ​​obtained from the second system.

4. The prediction device according to claim 1 , wherein the processing unit acquires second fabrication conditions to be applied to the first system from the derived predicted physical property values ​​by utilizing a fabrication condition determination model trained to derive fabrication conditions from fabrication conditions, physical property values, and specifications.

5. The prediction device according to claim 4 , wherein the manufacturing condition determination model is a Bayesian optimization model.

6. acquiring material information of a material prepared in a first system under first preparation conditions; deriving predicted physical property values ​​of the created material from the acquired material information using a material property prediction model trained to derive predicted physical property values ​​from material information; and A computer-implemented prediction method, wherein the property prediction model is updated with property evaluation values ​​for the substance by a second system.

7. acquiring material information of a material prepared in a first system under first preparation conditions; deriving predicted physical property values ​​of the created material from the acquired material information using a material property prediction model trained to derive predicted physical property values ​​from material information; on the computer, The property prediction model is updated by property evaluation values ​​for the substance obtained by a second system.

8. a fabrication device for fabricating a material in accordance with fabrication conditions in a first system; a prediction device that derives predicted physical property values ​​from material information of the prepared material; and The prediction device includes: an acquisition unit that acquires material information of a material prepared in the first system under first preparation conditions; a processing unit that derives predicted physical property values ​​of the created material from the acquired material information by using a material property prediction model that has been trained to derive predicted physical property values ​​from material information; and An experimental system, wherein the property prediction model is updated by property evaluation values ​​for the substance by a second system.

Images