Experiment condition generation device, program, and experiment condition generation method

Abstract

Provided is an experiment condition generation device that generates an experiment condition that is used for training data of a prediction model that predicts an experiment result from an experiment condition, the experiment condition generation device comprising: a starting point acquisition unit that acquires, from among searched conditions that are a plurality of experimented experiment conditions, an experiment condition of a search starting point that serves as a starting point of a search; a search region setting unit that sets, on the basis of the search starting point, a search region that is at least a part of a search space defined by the experiment condition and indicates a range for use in generating the experiment condition; and an experiment condition generation unit that generates, from the search region, an experiment condition different from the searched conditions.

Description

Experimental condition generation apparatus, program, and experimental condition generation method

[0001] The present invention relates to an experimental condition generation device, a program, and an experimental condition generation method.

[0002] Patent Document 1 describes a system comprising: "Map display means 21 displays a two-dimensional map in which frames representing combinations of two experimental conditions extracted from a planning table are arranged in a matrix. Frame display means 22 displays the progress for each combination of the two experimental conditions by controlling the display state of each frame on the two-dimensional map displayed by the map display means 21 based on the progress of the experiment by the experiment control means 13. Expanded display means 23 accepts the designation of a frame on the two-dimensional map according to the operation on that frame, and displays the progress of the experiment corresponding to the designated frame in an expanded manner." (Abstract). Patent Document 2 states that "the present invention relates to a system (100) for managing sample test results and their respective result context information in a laboratory environment. The system comprises at least one analyzer (20) configured to perform at least one test on a sample, and a control device (10) connected to the at least one analyzer (20) for data exchange, wherein the control device (10) stores and displays the sample test results and their respective result context information upon request, dynamically controls at least one effective value of at least one item of the result context information with respect to a predetermined threshold, and is configured to initiate at least one action as soon as the effective value matches the predetermined threshold according to a predetermined execution plan. Furthermore, the present invention relates to an appropriate control device and method for managing sample test results and their respective result context information in a laboratory environment" (abstract).Patent Document 3 describes that "an experimental design program acquires experimental content data indicating the content of the experiment, equipment type data indicating the types of equipment used in each stage of the experiment, available equipment data indicating available equipment, and at least one of data indicating a utility function related to the equipment type data and data indicating a utility function related to the available equipment data, and performs at least one of the following: solving a mixed integer programming problem by applying the branch and bound method to determine the available equipment data that minimizes the time required for the experiment, assuming that the equipment indicated by one piece of equipment type data is used in each stage of the experiment; and solving a mixed integer programming problem by applying the branch and bound method to determine the equipment type data that minimizes the time required for the experiment, assuming that the equipment indicated by one piece of available equipment data is available for use in the laboratory." (Abstract). Patent Document 4 describes that "the experimental condition determination device 10 includes a choice identification unit 104 that identifies a plurality of experimental condition options in each process based on the evaluation value of an evaluation function that takes experimental conditions as input and outputs an evaluation value, and an experimental condition acquisition unit 106 that acquires a combined experimental condition, which is a combination of experimental conditions of a plurality of processes, from the plurality of experimental condition options based on the processing amount for each experimental condition and the evaluation value obtained by inputting the experimental condition options into the evaluation function" (abbreviation). Patent Document 5 describes a manufacturing evaluation system comprising: a manufacturing apparatus for manufacturing a sample; a measuring apparatus for measuring material information representing the physical properties or structure of the sample manufactured by the manufacturing apparatus; and an estimation apparatus (1) connected to the manufacturing apparatus and the measuring apparatus, wherein the estimation apparatus (1) includes an estimation unit that estimates the manufacturing conditions to optimize the material information based on a dataset including the manufacturing conditions of the sample and the material information of the sample; the manufacturing apparatus manufactures the sample according to the manufacturing conditions estimated by the estimation unit; the estimation apparatus (1) includes a data addition unit that adds the material information of the sample measured by the measuring apparatus and the manufacturing conditions of the sample to the dataset; and the estimation unit sequentially estimates the manufacturing conditions based on the dataset to which the manufacturing conditions and material information have been added. (Abstract)[Prior Art Document] [Patent Document] [Patent Document 1] Japanese Patent Application Laid-Open No. 2009-180576 [Patent Document 2] Japanese National Publication of International Patent Application No. 2009-539078 [Patent Document 3] Japanese Patent Application Laid-Open No. 2022-170525 [Patent Document 4] Japanese Patent Application Laid-Open No. 2024-128357 [Patent Document 5] WO2021 / 044913A1. General Statement

[0003] In a first aspect of the present invention, there is provided an experimental condition generation device that generates experimental conditions used as training data for a prediction model that predicts experimental results from experimental conditions. The experimental condition generation device includes a starting point acquisition unit, a search area setting unit, and an experimental condition generation unit. The starting point acquisition unit acquires an experimental condition as a search starting point that serves as a starting point for search from among searched conditions, which are a plurality of already-conducted experimental conditions. Based on the search starting point, the search area setting unit sets a search area that is at least a part of a search space defined by experimental conditions and indicates a range used for generating experimental conditions. The experimental condition generation unit generates an experimental condition different from the searched conditions from the search area.

[0004] In the above description, the starting point acquisition unit may acquire, as the search starting point, an experimental condition that satisfies a predetermined condition from among the searched conditions.

[0005] In the above description, the search area setting unit may set the search area within a predetermined distance from the search starting point in the search space.

[0006] In the above description, the experimental condition may be composed of a plurality of element conditions. The search area setting unit may set the search area in a range in which at least some element conditions overlap with the search starting point in the search space.

[0007] In the above description, the experimental condition may be composed of a plurality of element conditions. The search area setting unit may set the search area to a numerical value within a predetermined range from the search starting point for those element conditions that take continuous values.

[0008] In the above description, the experimental condition may be composed of a plurality of element conditions. Based on the search starting point, the search area setting unit may set a plurality of discrete options for the element conditions as the search area.

[0009] In the above, the experimental condition generation device may further include a learning unit. The learning unit may machine-learn a predictive model based on the experimental conditions generated by the experimental condition generation unit and the experimental results of said experimental conditions.

[0010] In the above, the experimental condition generation device may further include a starting material identification unit and a similar material identification unit. The starting material identification unit may identify a starting material, which is the material used in the experimental conditions of the starting point of the search. The similar material identification unit may identify a similar material, which is a material whose characteristics are similar to those of the starting material. The search area setting unit may generate a search area that includes experimental conditions using similar materials.

[0011] In the above, the experimental condition generation device may further include a prediction unit. The prediction unit may input experimental conditions into a prediction model and output prediction results. The experimental condition generation unit may generate experimental conditions based on the prediction results predicted by the prediction unit.

[0012] In the above, the experimental condition generation device may further include a learning unit. The learning unit generates training data using the explored conditions included in the search domain. The learning unit may train a prediction model for the search domain based on the training data. The prediction unit may input experimental conditions using the prediction model for the search domain and output prediction results. The experimental condition generation unit may generate experimental conditions based on the prediction results predicted by the prediction unit using the prediction model for the search domain.

[0013] In the above, the experimental condition generation device may further include a specified condition input unit. The specified condition input unit may input a specified range, which is a range specified by the user for some of the elemental conditions of the experimental conditions. The experimental condition generation unit may generate experimental conditions from the specified range for the elemental conditions.

[0014] In the above, the experimental condition generation device may include a specified condition input unit. The specified condition input unit may input specified options, which are choices specified by the user for some of the elemental conditions of the experimental conditions. The experimental condition generation unit may generate the experimental conditions after selecting the elemental conditions from the specified options.

[0015] In the above, when the element conditions that are user-specified are designated conditions and the element conditions other than the designated conditions are free conditions, the experimental condition generation unit may have a candidate generation unit, an evaluation unit, and an output unit. The candidate generation unit may generate multiple sets of two or more experimental conditions in which the free conditions are the same but the designated conditions are different. The evaluation unit may comprehensively evaluate each of the multiple sets of experimental conditions based on the prediction results predicted by the prediction unit for each experimental condition included in the set of experimental conditions. The output unit may output a recommended set of experimental conditions based on the evaluation of the sets of experimental conditions by the evaluation unit.

[0016] In a second embodiment of the present invention, a program is provided which is executed by a computer and causes the computer to function as the experimental condition generation device described above.

[0017] In a third aspect of the present invention, an experimental condition generation method is provided, which is performed by an experimental condition generation device which may be any of the above-described devices, and comprises a starting point input step, a search area setting step, and an experimental condition generation step. In the starting point input step, the device may accept the acquisition of an experimental condition for the search starting point from among a plurality of previously experimented experimental conditions which are searched conditions. In the search area setting step, based on the search starting point, a search area may be set in the search space defined by the experimental conditions, which is at least a part of the search space and indicates the range used for generating experimental conditions. In the experimental condition generation step, experimental conditions different from the searched conditions may be generated from the search area.

[0018] The above summary of the invention does not enumerate all of its features. Furthermore, subcombinations of these features may also constitute an invention.

[0019] This shows the configuration of the experimental condition generation device 10 according to this embodiment. This shows the flow of experimental condition generation by the experimental condition generation device 10 according to this embodiment. This shows an example of training data according to this embodiment. This shows an example of composition conditions according to this embodiment. This shows an example of process conditions according to this embodiment. This shows an example of experimental results according to this embodiment. This shows an example of the search space shown in two dimensions. This shows an example of the subflow of S200. This shows an example of the search area 910 set on the search space according to Figure 7. This shows another example of the subflow of S200. This shows an example of the feature space shown in two dimensions. This shows an example of the search area set based on the search space and the subflow of Figure 10. This shows an example of the subflow of S300. This shows an example of recommended experimental conditions output within the search area according to Figure 9. This shows an example of the subflow of S300 according to the first modification of this embodiment. This shows an example of experimental condition candidates according to the first modification of this embodiment. This shows another example of experimental condition candidates according to the first modification of this embodiment. This shows an example of the subflow of S300 according to the second modification of this embodiment. This shows an example of a set of experimental condition candidates according to the second modification of this embodiment. This shows an example of a set of experimental condition candidates according to the second modification of this embodiment. The flow diagram for generating experimental conditions according to a third modification of this embodiment is shown. An example of a computer 2200 in which multiple aspects of the present invention may be embodied in whole or in part is shown.

[0020] The present invention will be described below through embodiments, but these embodiments are not intended to limit the scope of the claims. Furthermore, not all combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0021] Figure 1 shows the configuration of the experimental condition generation device 10 according to this embodiment. Generally, in order to obtain desired experimental results, it is necessary to conduct experiments under a large number of experimental conditions, but actual experiments consume a lot of resources such as materials, personnel, and time. If experimental conditions can be input to a prediction model that predicts experimental results, and the prediction model can then predict the experimental results, the resources can be reduced.

[0022] On the other hand, in order to train a high-quality predictive model, appropriate experimental conditions and the corresponding experimental results are necessary as training data. The experimental condition generation device 10 according to this embodiment generates experimental conditions to be used as training data for a predictive model that predicts experimental results from the experimental conditions.

[0023] The experimental condition generation device 10 includes a target acquisition unit 105, a starting point acquisition unit 110, a starting point material identification unit 120, a similar material identification unit 130, a search area setting unit 140, a specified condition input unit 150, a prediction unit 152, a learning unit 154, an experimental condition generation unit 160, and an experimental data acquisition unit 170.

[0024] The experimental condition generation device 10 may be a computer such as a PC (personal computer), tablet computer, smartphone, workstation, server computer, or general-purpose computer, and may also be a computer system in which multiple computers are connected.

[0025] Alternatively, the experimental condition generation device 10 may be a dedicated computer designed according to its respective purpose, or it may be dedicated hardware realized by a dedicated circuit. The experimental condition generation device 10 may be implemented by a single device (computer), or it may be realized by multiple devices with assigned roles. In the experimental condition generation device 10, although not specifically described below, memory / hard disk etc. are provided and the information necessary for processing is stored as appropriate, and information is transmitted between each processing module of the experimental condition generation device 10, such as the target acquisition unit 105, starting point acquisition unit 110, starting point material identification unit 120, similar material identification unit 130, search area setting unit 140, specified condition input unit 150, prediction unit 152, learning unit 154, experimental condition generation unit 160 and experimental data acquisition unit 170.

[0026] The target acquisition unit 105 acquires the experimental target from the user.

[0027] The starting point acquisition unit 110 acquires at least one experimental condition from among multiple experimental conditions that have already been tested, which will serve as the starting point for the search. The starting point acquisition unit 110 may acquire an experimental condition that satisfies predetermined conditions as the starting point for the search. The starting point acquisition unit 110 may also receive input from the user specifying the experimental condition to be used as the starting point for the search.

[0028] The starting material identification unit 120 identifies the starting material, which is the material used in the experimental conditions of the search starting point. The starting material identification unit 120 identifies the material that is specified to be used in the experimental conditions of the search starting point as the starting material.

[0029] The similar material identification unit 130 identifies similar materials that have characteristics similar to the starting material. The specific method for identifying similar materials will be described later.

[0030] The search area setting unit 140 sets a search area that is at least a part of the search space defined by the experimental conditions, based on the search starting point acquired by the starting point acquisition unit 110, and indicates the range used to generate the experimental conditions. For example, the search area setting unit 140 may set the search area within a predetermined distance from the search starting point in the search space. Alternatively, for example, the search area setting unit 140 may generate a search area that includes experimental conditions using similar materials.

[0031] The specified condition input unit 150 accepts input of a specified range, which is a range specified by the user for some of the elemental conditions of the experimental conditions. The specified condition input unit 150 accepts input of a specified option, which is a selection of options specified by the user for some of the elemental conditions of the experimental conditions.

[0032] The prediction unit 152 inputs experimental conditions into the prediction model and outputs prediction results. The prediction model outputs the experimental results that are expected to occur when the experiment is conducted under the input experimental conditions.

[0033] The learning unit 154 trains the prediction model that the prediction unit 152 uses for prediction. The learning unit 154 may train the prediction model using all or part of the explored conditions. The learning unit 154 may train a prediction model for the search domain based on the explored conditions that are included in the search domain. The learning unit 154 may generate training data to be used for training.

[0034] The experimental condition generation unit 160 generates experimental conditions different from the previously explored conditions from the search area set by the search area setting unit 140. For example, the experimental condition generation unit 160 may generate experimental conditions randomly from the search area. For example, the experimental condition generation unit 160 may generate experimental conditions based on the prediction results predicted by the prediction unit 152. The experimental condition generation unit 160 may generate experimental conditions from a specified range for element conditions.

[0035] The experimental condition generation unit 160 may output a set of experimental conditions. In this case, the experimental condition generation unit 160 may have a candidate generation unit 162, an evaluation unit 164, and an output unit 166, and output a set of experimental conditions using these units.

[0036] The candidate generation unit 162 generates experimental condition candidates, which are candidates for the experimental conditions that the experimental condition generation unit 160 will ultimately output. The candidate generation unit 162 may further generate sets of experimental condition candidates. Here, when the element conditions that are input by the user are designated as specified conditions, and the element conditions other than the specified conditions are designated as free conditions, the candidate generation unit 162 generates multiple sets of two or more experimental condition candidates in which the free conditions are the same, but the specified conditions are different.

[0037] The evaluation unit 164 evaluates the experimental condition candidates. The evaluation unit 164 further evaluates each of the multiple experimental condition candidate sets comprehensively based on the prediction results predicted by the prediction unit 152 for each experimental condition included in the set of experimental condition candidates. The evaluation unit 164 may perform the evaluation based on the target acquired by the target acquisition unit 105.

[0038] The output unit 166 outputs recommended experimental conditions based on the evaluation of the candidate experimental conditions. The output unit 166 further outputs a set of recommended experimental conditions based on the evaluation of the set of candidate experimental conditions by the evaluation unit 164. For example, the output unit 166 outputs a set of experimental conditions with a superior evaluation.

[0039] The experimental data acquisition unit 170 acquires experimental results obtained by actually executing the experimental conditions output by the output unit 166. The experimental data acquisition unit 170 may provide the learning unit 154 with pairs of experimental conditions and corresponding experimental results.

[0040] Thus, the experimental condition generation device 10 of this embodiment sets a search area from the search starting point and generates new experimental conditions from the search area. This allows the experimental condition generation device 10 to propose experimental conditions related to the search starting point that should be executed in the future. Subsequently, the predictive model can be trained or updated using the proposed experimental conditions.

[0041] In order to train a highly accurate predictive model, a large amount of training data and / or a complex model are generally required, resulting in a significant increase in computational and memory resources. Particularly in materials development, it is necessary to find experimental conditions that satisfy the target (for example, a material composition that satisfies the desired performance, or suitable conditions for the process of creating a sample from the material) from a vast number of options across a wide range of items such as composition and process. According to the experimental condition generation device 10 of this embodiment, by limiting the range of experimental conditions (search range), experimental conditions that satisfy the target can be found quickly and efficiently with relatively little computational and / or memory resources.

[0042] Figure 2 shows the flow of experimental condition generation by the experimental condition generation device 10 according to this embodiment. The experimental condition generation device 10 generates experimental conditions by executing each of the processes S10 to S500, for example. Some of S10 to S500 may be omitted. Other operations may be performed in addition to S10 to S500.

[0043] First, in S10, the target acquisition unit 105 acquires an experimental target from a user. The target may include item information that is the target object and corresponds to the objective variable of the prediction model (e.g., strength), and directionality information related to the direction of the experiment (e.g., exceeding a threshold, aiming for as large a value as possible, falling within or approaching a predetermined range of values, etc.).

[0044] In S50, the learning unit 154 generates a prediction model. The prediction model receives an experimental condition as input and outputs an experimental result corresponding to the experimental condition. The learning unit 154 generates the prediction model by performing machine learning or the like using training data.

[0045] Figure 3 shows an example of training data according to the present embodiment. The training data includes a large number of pairs of experimental conditions and experimental results. The experimental result may be a result obtained by actually performing an experiment according to the experimental conditions, and / or a result obtained by at least partially simulating the experiment on a computer using the experimental conditions. The learning unit 154 may acquire completed experimental conditions and corresponding experimental results, and generate training data based thereon.

[0046] The experimental conditions may include composition conditions and process conditions. In the example of Figure 3, the training data includes the experimental condition indicated by ID0001, the experimental condition indicated by ID0002, and so on. As ID0001, it is shown that a pair including an experimental condition having composition condition XXX1 and process condition YYY1, and an experimental result ZZZ1 is included. As ID0002, it is shown that a pair including an experimental condition having composition condition XXX2 and process condition YYY2, and an experimental result ZZZ2 is included in the training data.

[0047] Figure 4 shows an example of composition conditions according to the present embodiment. The composition conditions may include information specifying materials used in the experiment. For example, the composition conditions may include information such as the type, name, chemical or physical structure, chemical composition, physical properties, content, content ratio, and content percentage of materials used in the experiment.

[0048] The compositional conditions may be combinations of the content of materials used or potentially used in the experiment. The compositional conditions may specify the content of each type of material (e.g., polymer, solvent, additive, etc.).

[0049] For example, Figure 4 shows the compositional conditions for the materials contained in the coating liquid. In Figure 4, the compositional conditions for ID 0001 are shown to be 100% by weight of polymer A and 0% by weight of polymer B, and 100% by weight of solvent X and 0% by weight of solvent Y. The compositional conditions for ID 0002 are shown to be 0% by weight of polymer A and 100% by weight of polymer B, and 100% by weight of solvent X and 0% by weight of solvent Y.

[0050] Figure 4 shows an example of how compositional conditions can be expressed, and the expression of compositional conditions is not limited to this. For example, one elemental condition may specify the type of material, and another elemental condition may specify the content of that material. For example, one elemental condition may limit the type of polymer (for example, polymer A is used if the value is 1, polymer B if the value is 2, and polymer C if the value is 3), and another elemental condition may specify the content of that polymer.

[0051] Figure 5 shows an example of process conditions according to this embodiment. Process conditions may include information about the conditions of the operation (also called the "process") performed in the experiment. For example, process conditions may include the type of operation, time-related conditions (e.g., heating time), temperature-related conditions (e.g., heating temperature, cooling temperature, mixing temperature, etc.), size and / or weight-related conditions (e.g., film thickness, volume, etc.), pressure-related conditions (e.g., pressurized pressure), conditions related to the premise of the treatment (e.g., type of substrate), etc.

[0052] For example, in Figure 5, the process conditions for ID 0001 are those related to the application and drying of the coating solution, where a metal is used as the substrate for coating, a coating film with a thickness of 5 μm is formed, the coating solution is prepared at 90°C, and the post-application treatment (e.g., curing and / or drying) is performed at 110°C. The process conditions for ID 0002 show that a metal is used as the substrate, a coating film with a thickness of 5 μm is formed, and a preparation temperature of 110°C and a processing temperature of 120°C are used.

[0053] Each condition included in the experimental conditions is referred to as an "elemental condition." In other words, the experimental conditions consist of multiple elemental conditions. For example, ID0001 includes the following elemental conditions: polymer A content, polymer B content, solvent X content, solvent Y content, substrate type, film thickness, mixing temperature, and processing temperature.

[0054] The elemental conditions may include continuous variables, ordinal variables, or nominal variables. For example, the content of polymer A and the content of polymer B are continuous variables. For example, the type of substrate is a nominal variable.

[0055] Figure 6 shows an example of experimental results according to this embodiment. For example, in Figure 6, the experimental results include the physical properties of the coating film. The experimental results for ID 0001 include the adhesion of the coating film (10 N / mm), the heat resistance temperature of the coating film (200°C), and the elongation of the coating film (150%). The experimental results for ID 0002 include the adhesion of the coating film (8 N / mm), the heat resistance temperature of the coating film (210°C), and the elongation of the coating film (155%). Each result included in the experimental results is referred to as an "elemental result".

[0056] The type of predictive model is not particularly limited. For example, predictive models may include various regression analysis methods such as multiple regression analysis, ridge regression, partial least squares regression, multilayer perceptrons, neural networks such as CNN (Convolutional Neural Network) and RNN (Recurrent Neural Network), support vector machines using arbitrary kernel functions such as Gaussian kernels, random forests modeled as regression trees, models using hidden Markov models, statistical models, and probabilistic models. When the information indicating quality is a qualitative variable, various discriminant analysis methods such as logistic regression, random forests, and neural networks may be used.

[0057] Next, in S100, the starting point acquisition unit 110 acquires the experimental conditions for the search starting point. The starting point acquisition unit 110 may select one experimental condition from among the experimental conditions that have already been experimented on and for which experimental results have been obtained, and use this as the search starting point. The starting point acquisition unit 110 may select one or more search starting points. The search starting point may be an experimental condition included in the training data used in S50.

[0058] As shown in Figures 3 to 5, the experimental conditions include multiple elemental conditions. If each elemental condition is used as a coordinate axis, a multidimensional space containing many experimental conditions can be defined, and this will be used as the search space. For example, if the experimental conditions include 10 elemental conditions, a 10-dimensional space can be defined as the search space. The search space may consist of elemental conditions from the experimental conditions that become explanatory variables of the prediction model.

[0059] Here, elemental conditions that are not quantified or that correspond to nominal variables (for example, the type of substrate) may be quantified using known methods and reflected in the search space. For example, in the search space, substrate "metal" = 1, substrate "glass" = 2, substrate "plastic" = 3, etc.

[0060] Figure 7 shows an example of a search space represented in two dimensions. Figure 7 may represent a search space (e.g., 10 dimensions) compressed to two dimensions using principal component analysis, with experimental conditions shown in two dimensions. For example, the horizontal axis in Figure 7 may represent the first principal component, and the vertical axis may represent the second principal component. The search space is the region enclosed by the thick line. For the sake of explanation, the region corresponding to the search space is shown as a rectangle, but the shape of the search space is not limited to this.

[0061] The search space contains numerous explored conditions 702 (shown as black circles in the figure). The starting point acquisition unit 110 acquires a search starting point 704 (shown as a white circle in the figure) from among the explored conditions 702, which will serve as the starting point for subsequent searches.

[0062] The starting point acquisition unit 110 acquires experimental conditions that satisfy predetermined conditions as the search starting point. For example, the starting point acquisition unit 110 may acquire from the searched conditions 702 those in which the experimental results satisfy predetermined conditions as the search starting point.

[0063] For example, the starting point acquisition unit 110 uses the experimental condition in which a specified element (e.g., adhesion) is best among the experimental results as the starting point for the search. For example, the starting point acquisition unit 110 uses the experimental condition in which the best result is obtained when multiple specified elements (e.g., adhesion and heat resistance temperature) among the experimental results are evaluated using a weighted average or evaluation function as the starting point for the search. For example, the starting point acquisition unit 110 uses the experimental condition in the search space that is furthest from the nearest other experimental condition (i.e., there are no nearby experimental conditions) as the starting point for the search.

[0064] Alternatively, the starting point acquisition unit 110 may receive input from the user specifying the experimental conditions that will serve as the starting point for the search. The starting point acquisition unit 110 may also receive input specifying the experimental conditions for a pre-specified search starting point.

[0065] Next, in S200, the search area setting unit 140 sets the search area based on the search starting point acquired in S100.

[0066] Figure 8 shows an example of a subflow of S200. The search area setting unit 140 may execute S200 by executing a subflow including S210 in Figure 8.

[0067] In S210, the search area setting unit 140 sets a search area in the search space that is within a predetermined distance from the search starting point set in S100. The predetermined distance may be a distance predetermined before processing, or it may be a distance entered by the user.

[0068] In addition to / or instead of the above, the search area setting unit 140 may be a search area that includes only specific options that have a common part with the search starting point set in S100 in the search space.

[0069] Figure 9 shows an example of a search area 910 set in the search space shown in Figure 7. The search area setting unit 140 sets the search area 910 within a range of distance D centered on the search starting point 704. In Figure 9, for the sake of explanation, the search area 910 is shown to be set on a two-dimensional plane, but in reality, the search area 910 may be set in a multi-dimensional space. If multiple experimental conditions are specified as search starting points, the search area 910 may be set within a range of distance D from each of the multiple search starting points. In this case, the distance D may be set to a different value for each search starting point, or to the same value. The search area setting unit 140 may also set the search area 910 as the union or intersection of the ranges of distance D from multiple search starting points.

[0070] In addition to / or alternative to setting the search area by distance, the search area setting unit 140 may set the search area in the search space to a range in which the search starting point and at least some of the element conditions overlap. For example, the search area may be the range in which some or all of the materials used in the composition conditions and / or process conditions of the search starting point are used.

[0071] As an example, in ID0001 of Figure 4, polymer A and solvent X are used in amounts exceeding 0%. Therefore, if the search starting point is ID0001 of Figure 4, the search area setting unit 140 may set the search area to a range in which polymer A is used in amounts greater than 0% but less than or equal to 100%, and / or in which solvent X is used in amounts greater than 0% but less than or equal to 100%.

[0072] The search area setting unit 140 may set the search area to a range of values ​​within a predetermined range from the search starting point for element conditions that take continuous values. For example, the search area setting unit 140 may set the search area to a range within a predetermined range starting from the amount of at least some of the materials used, as indicated by the composition conditions at the search starting point. As an example, if polymer A is used at 50% and polymer B is used at 50% at the search starting point, the search area setting unit 140 may set the search area to a range where polymer A is used at 50% ± 20% and / or polymer B is used at 50% ± 20%.

[0073] For example, the search area setting unit 140 may set the search area to a predetermined range starting from the numerical values ​​of at least some of the conditions indicated by the process conditions of the search starting point. As an example, if the search starting point includes the conditions of a mixing temperature of 90°C and a processing temperature of 110°C, as shown in ID 0001 of Figure 5, the search area setting unit 140 may set the range of a mixing temperature of 90±30°C and / or a processing temperature of 110±30°C as the search area.

[0074] As an example, in Figure 5, ID0001 and ID0002 specify metal as the base material. If the search starting point is ID0001 or ID0002 in Figure 5, the search area setting unit 140 may set the area in which metal is used as the base material as the search area.

[0075] In addition to setting the search domain to a numerical value within a predetermined range from the search starting point, the search domain setting unit 140 may, based on the search starting point, set a discrete number of options for the element conditions as the search domain.

[0076] The search area setting unit 140 may set the search area after temporarily excluding variables that are difficult to define by distance (for example, nominal variables such as the type of substrate in process conditions). For example, the search area setting unit 140 may set the search area in a search space that excludes the "substrate" variable. The search area setting unit 140 may configure the search area by setting discrete options for the excluded variables.

[0077] The search area setting unit 140 may set discrete options for variables other than those that are difficult to define by distance. For example, if the search starting point includes the conditions of a mixing temperature of 90°C and a processing temperature of 110°C, the search area setting unit 140 may set multiple options as the search area in which the mixing temperature is selected from 70°C, 80°C, 90°C, 100°C, and 110°C, and the processing temperature is selected from 90°C, 100°C, 110°C, 120°C, and 130°C.

[0078] Figure 10 shows another example of the subflow of S200. The search area setting unit 140 may execute S200 by executing at least a portion of the subflows S220 to S240 in Figure 10.

[0079] First, in S220, the starting material identification unit 120 identifies the starting material, which is the material used in the experimental conditions of the search starting point. For example, the starting material identification unit 120 identifies the material used at the search starting point from the composition conditions of the search starting point.

[0080] The starting material identification unit 120 identifies one, more, or all of the materials indicated by the composition conditions that were used at the search starting point as the starting material. The starting material identification unit 120 may identify a predetermined type of material, or a type of material specified by the user, from among the materials used at the search starting point as the starting material. For example, if only polymers are specified, the starting material identification unit 120 may identify only the polymers used at the search starting point as the starting material.

[0081] For example, in ID0001 of Figure 4, polymer A and solvent X are used in amounts exceeding 0%. Therefore, if the search starting point is ID0001 of Figure 4, the starting material identification unit 120 identifies polymer A and / or solvent X as the starting materials. For example, in ID0002 of Figure 4, polymer B and solvent X are used in amounts exceeding 0%. Therefore, if the search starting point is ID0002 of Figure 4, the starting material identification unit 120 identifies polymer B and / or solvent X as the starting materials.

[0082] In some cases, process conditions as well as composition conditions can be used to identify materials. In such cases, the starting material identification unit 120 may identify the material used at the starting point of the search from the process conditions at the starting point of the search, in addition to / or instead of the composition conditions at the starting point of the search.

[0083] As an example, in Figure 5, ID0001 and ID0002 specify metal as the base material. If the search starting point is ID0001 or ID0002 in Figure 5, the starting material identification unit 120 identifies metal as the starting material.

[0084] Next, in S230, the similar material identification unit 130 identifies similar materials that have characteristics similar to the starting material. The similar material identification unit 130 may identify similar materials based on the characteristic quantities of the materials.

[0085] The feature quantities of a material may be a single numerical value or a combination of numerical values ​​(e.g., a vector or matrix) represented by a continuous, ordinal, or nominal variable that indicates one or more characteristics of the material. Each of the numerical values ​​included in the feature quantity is also called a "feature element."

[0086] The material characteristics may be state variables representing the physical and / or chemical properties of the material, other characteristics of the material, or numerical values ​​representing the performance achieved by the material. For example, material characteristics may include the constituent elements, density, hardness, molecular weight, molecular formula, molecular structure, crystal structure, strength, and / or viscosity of the material. In particular, material characteristics that affect or may affect the experimental results may be selected.

[0087] The similar material identification unit 130 may identify similar materials using a feature space where each feature element included in the feature quantity is used as a coordinate axis. For example, the similar material identification unit 130 identifies materials that satisfy predetermined conditions regarding the distance from the starting material in the feature space, among the materials available for use in the experiment (for example, materials that are listed in advance), as similar materials.

[0088] For example, the similar material identification unit 130 may identify a predetermined number of materials as similar materials in order of proximity to the starting material. For example, the similar material identification unit 130 may identify materials as similar materials whose distance from the starting material is below a threshold.

[0089] Figure 11 shows an example of a feature space represented in two dimensions. Figure 11 may represent a feature space (e.g., 5-dimensional) compressed to two dimensions using principal component analysis or the like, showing the experimental conditions in two dimensions. For example, the horizontal axis of Figure 11 may represent the first principal component and the vertical axis may represent the second principal component.

[0090] The case in which the starting material identification unit 120 identifies polymer A as the starting material in S220 will be explained. The similar material identification unit 130 searches for other materials within a threshold distance 1110 from polymer A 1102, which is the starting material. For example, the similar material identification unit 130 identifies polymer B 1104 as a material within a threshold distance 1110 and designates it as a similar material. Polymers C 1106 and D are outside the threshold distance and therefore do not need to be considered similar materials. If the similar material identification unit 130 does not find a predetermined number of other materials within a threshold distance 1110, it may expand the threshold distance until a predetermined number of materials are found.

[0091] If multiple materials are specified as starting materials, the similar material identification unit 130 may set a range within a threshold distance 1110 from each of the multiple materials. In this case, the threshold distance 1110 may be the same or different. The similar material identification unit 130 may identify similar materials from the union or common part of each range.

[0092] Next, in S240, the search area setting unit 140 sets a search area that includes experimental conditions that use some or all of the similar materials identified in S230. For example, if polymer B and solvent Y are identified as similar materials in S230, the search area setting unit 140 sets the search area to include polymer B and solvent Y.

[0093] The search area setting unit 140 may set a search area that includes experimental conditions that use a starting material in addition to similar materials. For example, if polymer A and solvent X are identified as starting materials in S220, and polymer B and solvent Y are identified as similar materials in S230, the search area setting unit 140 sets a search area that includes either polymer A or polymer B, and either solvent X or solvent Y.

[0094] The search area setting unit 140 may exclude experimental conditions from the search area that do not include either the similar material or the starting material. For example, if polymer A is identified as the starting material in S220 and polymer B is identified as the similar material in S230, experimental conditions that do not include either polymer A or polymer B may be excluded from the search area.

[0095] The search area setting unit 140 may set the search area in the same manner as described in Figures 8 and 9, including similar materials and / or a starting material. For example, the search area setting unit 140 may set the search area to include similar materials and / or a starting material, and within a predetermined range from the search starting point.

[0096] Figure 12 shows an example of a search area set based on the search space and the subflow in Figure 10. The "minimum value" and "maximum value" in the "search space" on the left side of Figure 12 indicate the range of possible values ​​for each element condition in the search space, and / or the maximum range to be searched for in the experimental conditions. Such minimum and maximum values ​​may be set in advance, or may be set by the user. The search area setting unit 140 may set the minimum and maximum values ​​of the search space to include all existing searched conditions.

[0097] The elemental conditions included in the compositional conditions (e.g., "Polymer A," "Polymer B," etc. in the table) may represent the proportion of each material that can be present within a group of similar materials. Instead of a proportion, the content itself may be used. The elemental conditions included in the process conditions (e.g., "Mixing Temperature," "Processing Temperature," etc. in the table) may indicate the range of values ​​that each elemental condition can take during the experimental process.

[0098] For example, the minimum possible value for the elemental condition related to polymer A included in the composition conditions in the search space is 0. This means that polymer A is not included at all in the material used in the experiment (for example, the same meaning as 0 wt% shown in Figure 4). The maximum possible value for the elemental condition related to polymer A is 1. This means that, among the materials used in the experiment, only polymer A is included as a polymer, and other polymers such as polymer B are not included (for example, the same meaning as 100 wt% shown in Figure 4). The minimum and maximum values ​​for the elemental conditions related to polymers B and C are the same as for polymer A. The sum of the elemental conditions (i.e., content) of polymers A, B and C, etc., or of a group of materials of the same type such as solvent X and solvent Y, may be 1.

[0099] In the search space, similar to polymers, the minimum value of the elemental conditions related to solvent X and solvent Y may be 0, and the maximum value may be 1. The sum of the elemental conditions for solvent X and solvent Y may also be 1.

[0100] In the search space, the elemental condition related to polymer / solvent content represents the ratio of polymer to solvent content (e.g., weight ratio or volume ratio). The minimum value of the elemental condition related to polymer / solvent content is 0.1, and the maximum value is 0.5. This means that only the range in which the ratio of polymer content to solvent content is between 0.1 (i.e., 10%) and 0.5 (i.e., 50%) is included in the search space.

[0101] Therefore, the search area cannot be set in the range where the polymer / solvent ratio is less than 0.1 or greater than 0.5. For example, if it is known in advance that the product cannot be manufactured due to process problems when the polymer / solvent ratio is less than 0.1 or greater than 0.5, the range of polymer / solvent ratio in the search space is limited in advance to 0.1 to 0.5.

[0102] Similarly, in the search space, the minimum value of the elemental condition related to the mixing temperature is 30°C, and the maximum value is 150°C. The minimum value of the elemental condition related to the processing temperature is 80°C, and the maximum value is 200°C. The minimum value of the elemental condition related to the film thickness is 10 μm, and the maximum value is 10 μm. When the minimum and maximum values ​​are the same, it means that in the search space, that elemental condition can only take on one value, which is the minimum and maximum value. Therefore, only one value is set for that elemental condition in the search domain, and the experimental conditions generated later will also take on only that one value.

[0103] The "starting point" included in the "search area" on the right side of Figure 12 corresponds to the experimental conditions acquired as the starting point in S100. In other words, the experimental conditions acquired by the starting point acquisition unit 110 as the starting material are shown to be that the polymer contains only polymer A, the solvent contains only solvent X, the ratio of polymer amount to solvent is 0.3, the mixing temperature is 80°C, the processing temperature is 140°C, and the film thickness is 10 μm.

[0104] The "minimum value" and "maximum value" of the "search area" on the right side of Figure 12 indicate the range of the search area set by the search area setting unit 140 in S200. The search area setting unit 140 sets the search area within the search space so as to include the search starting point. With respect to the composition conditions, the search area setting unit 140 sets the search area so as to include at least one of polymer A, which is the starting material, and polymer B, which is a similar material. The search area setting unit 140 sets the search area so as not to include polymers other than polymer A, which is the starting material, and polymer B, which is a similar material (for example, polymer C).

[0105] As an example, the search area setting unit 140 sets the range of elemental conditions for polymer A to 0 to 1, the range of elemental conditions for polymer B to 0, and the range of elemental conditions for polymer C to 0. As a result, the experimental conditions included in the search area will only include the starting material and similar materials.

[0106] Similarly, the search area setting unit 140 sets the range of elemental conditions for solvent X, which is the starting material, to 0 to 1, the range of elemental conditions for solvent Y, which is a similar material, and sets the range of elemental conditions for solvent Z, which is neither the starting material nor a similar material, to 0.

[0107] The search area setting unit 140 sets the search area for process conditions, centering on the search starting point and including values ​​within a predetermined range from the search starting point. For example, the search area setting unit 140 sets the range of the polymer / solvent amount element condition to 0.2 to 0.4, which is within ±0.1 of the search starting point. For example, the search area setting unit 140 sets the range of the mixing temperature element condition to 70 to 90°C, which is within ±10°C of the search starting point. For example, the search area setting unit 140 sets the range of the processing temperature element condition to 110 to 170°C, which is within ±30°C of the search starting point. For example, the search area setting unit 140 sets the film thickness to 10, which is the same as the search starting point.

[0108] In S300, following S200, the experimental condition generation unit 160 generates experimental conditions different from the already explored conditions from the search area set by the search area setting unit 140. For example, the experimental condition generation unit 160 may generate experimental conditions included in the search area using a known search method or the like.

[0109] Figure 13 shows an example of a subflow for S300. The experimental condition generation unit 160 may execute S300 by executing at least a portion of the subflows S310 to S370 in Figure 13.

[0110] In S310, the candidate generation unit 162 generates candidate experimental conditions (hereinafter also referred to as "candidate experimental conditions") within the search area. For example, in the initial S310, the candidate generation unit 162 may randomly generate candidate experimental conditions so that they are included in the search area.

[0111] In S330, following S310, the evaluation unit 164 evaluates the experimental condition candidates generated in S310. The evaluation unit 164 may evaluate the experimental condition candidates using a prediction model. For example, the evaluation unit 164 provides the experimental condition candidates to the prediction unit 152. The prediction unit 152 inputs the experimental condition candidates into the prediction model learned in S50 and has it predict the experimental results of the experimental condition candidates.

[0112] The evaluation unit 164 acquires and evaluates the experimental results obtained from the prediction model by the prediction unit 152. For example, the evaluation unit 164 may evaluate the experimental results by quantifying one or more element results included in the experimental results into one or more numerical values. For example, the evaluation unit 164 may input the element results included in the experimental results into a pre-set evaluation function and acquire the output of the evaluation function as the evaluation.

[0113] An evaluation function may be a function that takes one or more element results as input and outputs a corresponding evaluation value. The evaluation value may be a numerical value indicating the quality of the experimental result, and for example, the closer the input element result is to a desirable value, the higher the output value may be.

[0114] The evaluation unit 164 may evaluate the experimental results using an evaluation function based on the item information and directional information acquired by the target acquisition unit 105 in S10. For example, the evaluation unit 164 may set an evaluation function such that the evaluation increases as the element result variable included in the item information moves in the direction included in the directional information (for example, the higher the intensity, the larger the evaluation value; the lower the defect occurrence rate, the smaller the evaluation value).

[0115] The evaluation unit 164 may evaluate candidate experimental conditions by considering their similarity to other experimental conditions in addition to the predicted experimental results. For example, the evaluation unit 164 may include evaluation terms such that the evaluation decreases the closer the candidate experimental conditions are to the explored conditions, other candidate experimental conditions proposed so far, and / or other candidate experimental conditions that are currently planned to be proposed (e.g., distance in the search space). As an example, the evaluation unit 164 may include an evaluation term in the evaluation function, in addition to the evaluation term related to the predicted experimental results, which is the sum of the reciprocals of the distances to each of the explored conditions and / or other candidate experimental conditions.

[0116] Next, in S350, the experimental condition generation unit 160 determines whether or not the termination conditions for generating candidate experimental conditions are met. For example, the experimental condition generation unit 160 may determine that the termination conditions are met if one or more of the following conditions are met: the evaluation of the candidate experimental conditions obtained in S330 meets the criteria; the loop processing of S310 to S350 is completed a predetermined number of times; the loop processing of S310 to S350 is executed for a predetermined time; or no improvement in the evaluation obtained in S330 is observed for a predetermined number of times or for a predetermined time.

[0117] If the experimental condition generation unit 160 satisfies the termination conditions, it proceeds to S370. If the experimental condition generation unit 160 does not satisfy the termination conditions, it returns to S310. In the second and subsequent S310s, the candidate generation unit 162 may generate experimental condition candidates within the search domain using known search methods (e.g., gradient method, genetic algorithm, mixed integer programming) to improve the evaluation in S330.

[0118] In S370, the output unit 166 outputs recommended experimental conditions based on the evaluation of the experimental condition candidates so far. For example, the output unit 166 outputs a predetermined number (for example, one or more) of experimental condition candidates in order of their evaluation in S330 as recommended experimental conditions (also called "recommended experimental conditions").

[0119] The output unit 166 may output recommended experimental conditions on the display screen and / or to a storage medium. In this case, the user may conduct the experiment according to the outputted recommended experimental conditions.

[0120] The output unit 166 may transmit recommended experimental conditions to an experimental control system connected to the experimental apparatus. In this case, the experimental control system may control the experimental apparatus using a computer and perform the experiment according to the recommended experimental conditions. The experimental control system may obtain experimental results from the experimental apparatus or the user. The experiment may be an actual experiment or a simulation performed on a computer.

[0121] Figure 14 shows an example of recommended experimental conditions output within the search area shown in Figure 9. As shown in the figure, the output unit 166 outputs multiple recommended experimental conditions 1402 (shown as squares in the figure) within the search area 910.

[0122] In step S400, following step S300, the experimental data acquisition unit 170 acquires experimental results obtained by executing the recommended experimental conditions generated in step S300. The experimental results may be the results of an actual experiment or the results of an experiment simulated on a computer. For example, the experimental data acquisition unit 170 may acquire experimental results from a user or an experiment control system. As a result, the experimental data acquisition unit 170 acquires pairs of recommended experimental conditions and experimental results obtained under those recommended experimental conditions.

[0123] In S500, following S400, the learning unit 154 updates the prediction model. The learning unit 154 may retrain the prediction model using the recommended experimental conditions and experimental results pairs obtained in S400, in addition to / instead of the training data used in S50. As a result, the learning unit 154 machine-learns the prediction model based on the experimental conditions generated by the experimental condition generation unit 160 and the experimental results of those experimental conditions.

[0124] As described above, this embodiment sets a search area based on the search starting point and searches for recommended experimental conditions within the search area. This makes it possible to efficiently improve the accuracy of the prediction model compared to conducting a large number of additional experiments without a plan. In particular, since the learning model is updated using the experimental conditions within the search area, the prediction accuracy of the prediction model in the search area can be rapidly improved without using a lot of computing or memory resources. Especially in materials development, it is necessary to find experimental conditions that satisfy the target (for example, a material composition that satisfies the desired performance, or suitable conditions for a process to create a sample from the material) from a vast number of choices across a wide range of items such as composition and process. With the experimental condition generation device 10 according to this embodiment, by limiting the range of experimental conditions (search area), experimental conditions that satisfy the target can be found quickly and efficiently with relatively few computing and / or memory resources.

[0125] Figure 15 shows an example of a subflow of S300 according to a first modified example of this embodiment. In this modified example, S300 may be executed by executing at least a portion of the subflows S305 to S370.

[0126] In S305, the specified condition input unit 150 inputs a specified range, which is a range specified by the user for some of the elemental conditions of the experimental conditions, and / or a specified option, which is a choice specified by the user. The specified condition input unit 150 may also receive input from the user for both the specified range and the specified option.

[0127] The specified condition input unit 150 may accept input from the user of a specified range for element conditions relating to continuous or ordinal variables. For example, the specified condition input unit 150 may accept input from the user of a range for element conditions relating to the content of polymer A included in the composition conditions (e.g., 0.1 to 0.5).

[0128] The specified condition input unit 150 may accept input of specified options from the user for element conditions relating to continuous variables, ordinal variables, or nominal variables. The specified options may be a set of multiple options. For example, the specified condition input unit 150 may accept input from the user of multiple options (e.g., metal, glass, and plastic) for element conditions relating to the substrate included in the process conditions.

[0129] Next, in S311, the candidate generation unit 162 generates experimental condition candidates within the search area, taking into consideration the specified range and / or specified options obtained in S305. That is, the candidate generation unit 162 generates experimental condition candidates such that the experimental condition candidates are included in the search area, the experimental condition candidates are included in the specified range, and / or the experimental condition candidates are selected from the specified options.

[0130] For example, the candidate generation unit 162 generates candidate experimental conditions that satisfy a specified range (e.g., polymer A content = 0.1 to 0.5) within the search area. For example, the candidate generation unit 162 generates candidate experimental conditions that satisfy a specified option (e.g., substrate = metal, glass, or plastic) within the search area.

[0131] If the specified range and / or specified options are not included in the search area (for example, if part or all of the specified range is outside the search area, and part or all of the specified options are outside the search area), the candidate generation unit 162 may ignore the parts of the specified range and / or specified options that are outside the search area. Alternatively, with respect to element conditions related to the specified range and / or specified options, the candidate generation unit 162 may prioritize the specified range and / or specified options and allow the generation of experimental condition candidates outside the search area.

[0132] The candidate generation unit 162 may generate candidate experimental conditions in the same manner as in S310 of Figure 13, except that it is constrained by a specified range and / or specified options. For example, candidate experimental conditions may be generated randomly or using a search method within the range of the candidate generation unit 162, the search area, and the specified range and / or specified options.

[0133] The process in S330 may be performed after S311. The processes in S330, S350, and S370 may be performed in the same manner as described in Figure 13.

[0134] Figure 16 shows an example of candidate experimental conditions according to the first modified example of this embodiment. In the example in Figure 16, the specified range for film thickness is 5 to 10 μm, and the specified range for mixing temperature is 50 to 110°C. The experimental condition generation unit 160 generates candidate experimental conditions ID0XX1 and ID0XX2 to satisfy the specified ranges.

[0135] Figure 17 shows another example of experimental condition candidates according to the first modification of this embodiment. In the example in Figure 17, (metal, glass, plastic) is specified as the substrate specification option. The experimental condition generation unit 160 selects from the specified options for the element condition "substrate" and generates experimental condition candidates ID0YY1 and ID0YY2.

[0136] In this modified example, in S300, the experimental condition generation unit 160 generates experimental conditions after selecting elemental conditions from a specified range or specified options. This allows for more detailed specifications of the generated experimental conditions, in addition to the search area, in response to user requests.

[0137] Figure 18 shows an example of a subflow of S300 according to a second modified example of this embodiment. In this modified example, S300 may be executed by performing at least a portion of the subflows S306 to S372.

[0138] In S306, the specified condition input unit 150 receives user specifications from the user. The user specifications include a set of which element conditions will be specified conditions and the specific details of the specified conditions. Among the experimental conditions, the element conditions for which user specifications are entered are specified conditions, and the element conditions other than the specified conditions are free conditions.

[0139] For example, the specified condition input unit 150 receives input from the user that the film thickness included in the process conditions is a specified condition, and that the set of film thicknesses of 5 μm and 10 μm is the specified content. In this case, the film thickness is the specified condition, and the other elemental conditions are free conditions.

[0140] In S312, the candidate generation unit 162 generates a set of candidate experimental conditions. The set of candidate experimental conditions may include two or more candidate experimental conditions. The set of candidate experimental conditions has the same conditions for the free conditions, but different conditions for the specified conditions.

[0141] Figures 19A and 19B show an example of a set of candidate experimental conditions according to a second modification of this embodiment. The set of first and second candidate experimental conditions in the figures is the set of candidate experimental conditions. The first and second candidate experimental conditions differ in the specified condition of film thickness (5 μm and 10 μm). The first and second candidate experimental conditions have the same content in terms of elemental conditions other than the free condition of film thickness.

[0142] The candidate generation unit 162 may generate sets of experimental condition candidates using the specified conditions as defined in S306, and the free conditions in the same manner as described in S310 of Figure 13. For example, the candidate generation unit 162 may set the free conditions randomly, or generate sets of experimental condition candidates using a search method to improve the evaluation in S332.

[0143] In S306, the specified conditions do not need to be specified. In this case, the candidate generation unit 162 may generate combinations of different values ​​for the specified conditions randomly or by a search method within the set of experimental condition candidates. The candidate generation unit 162 may generate the free conditions randomly or by a search method so that they have the same value within the set of experimental condition candidates.

[0144] Next, in S332, the evaluation unit 164 evaluates the set of experimental condition candidates generated in S312. The evaluation unit 164 may evaluate each of the experimental condition candidates included in the set using the same method as in S330 after obtaining predicted values ​​from the prediction model. By evaluating the set of experimental condition candidates using the above method, the evaluation unit 164 can evaluate, for example, the performance values ​​when different experimental conditions are applied to the same composition. In the second modified example, the target acquisition unit 105 may acquire a development target for each experimental condition included in the set when acquiring the target in S10 (for example, the first experimental condition candidate may aim to increase strength, and the second experimental condition candidate may aim to increase heat resistance, etc.).

[0145] Next, in S352, the experimental condition generation unit 160 determines whether the termination conditions for generating the set of candidate experimental conditions are met. For example, the experimental condition generation unit 160 may determine that the termination conditions are met if one or more of the following conditions are met: the evaluation of the set of candidate experimental conditions obtained in S332 meets the criteria; the loop processing of S312 to S352 is completed a predetermined number of times; the loop processing of S312 to S352 is executed for a predetermined time; or no improvement in the evaluation obtained in S332 is observed for a predetermined number of times or for a predetermined time.

[0146] If the experimental condition generation unit 160 satisfies the termination conditions, it proceeds to S372. If the experimental condition generation unit 160 does not satisfy the termination conditions, it returns to S312.

[0147] In S372, the output unit 166 outputs a recommended set of experimental conditions based on the evaluation of the sets of candidate experimental conditions so far. For example, the output unit 166 outputs a predetermined number (for example, one or more) sets of candidate experimental conditions in order of the highest evaluation in S332 as a recommended set of experimental conditions (also called a "recommended set of experimental conditions").

[0148] In this modified example, steps S400 and S500 may be performed in the same manner as described above. For example, an experiment may be conducted using the recommended set of experimental conditions, and the predictive model may be updated using the experimental results.

[0149] According to this modified version, a set of experimental conditions with identical elemental conditions other than the specified condition can be output, and the predictive model can be updated using the experimental results of that set of experimental conditions. In some cases, it may be desirable to obtain good experimental conditions at multiple levels for some elemental conditions. For example, it may be desirable to find experimental conditions that can be expected to yield good results in both cases of film thickness of 5 μm and 10 μm. According to this modified version, experimental conditions that can provide overall good experimental results at multiple levels can be proposed.

[0150] Figure 20 shows the flow of experimental condition generation according to a third modified example of this embodiment. In this modified example, the experimental condition generation device 10 may, for example, execute each of the processes S10 to S550. Some of S10 to S550 may be omitted. In addition to S10 to S550, other operations may be performed.

[0151] The experimental condition generation device 10 may perform steps S10 to S200 in the same manner as described in Figure 2.

[0152] In S250, following S200, the learning unit 154 generates learning data for the search domain using the explored conditions included in the search domain. For example, the learning unit 154 generates learning data for the search domain that has multiple pairs of explored conditions included in the search domain and corresponding experimental results. Based on the learning data for the search domain, the learning unit 154 learns a prediction model for the search domain. That is, the learning unit 154 learns a prediction model for the search domain separately from the prediction model learned in S50.

[0153] The learning unit 154 trains a prediction model for the search domain using only the search conditions that are included in the search domain set in S200. Here, the learning unit 154 may use all or part of the search conditions that are included in the search domain set in S200. The training method may be the same as the method used in S50.

[0154] S300 may be performed in the same manner as described in Figures 2, 13, 15, and 18. Here, the evaluation unit 164 may evaluate the candidate experimental conditions or sets of candidate experimental conditions in S330 and S332, etc., using a prediction model for the search area. That is, the prediction unit 152 may input experimental conditions using a prediction model for the search area and output prediction results. As a result, the experimental condition generation unit 160 generates experimental conditions based on the prediction results predicted by the prediction unit 152 using the prediction model for the search area.

[0155] According to this modification, a prediction model specifically tailored to the search domain is trained, and experimental conditions are generated using this prediction model. Since the generated experimental conditions will be included in the search domain, it is expected that the prediction accuracy will be higher when using the prediction model for the search domain compared to when using other prediction models. As a result, experimental conditions that are expected to yield good experimental results can be proposed with high accuracy using fewer computational and / or memory resources.

[0156] Various embodiments of the present invention may be described with reference to flowcharts and block diagrams, where a block may represent (1) a stage in a process in which an operation is performed, or (2) a section of a device having the role of performing the operation. Specific stages and sections may be implemented by dedicated circuits, programmable circuits supplied with computer-readable instructions stored on a computer-readable medium, and / or processors supplied with computer-readable instructions stored on a computer-readable medium. Dedicated circuits may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. Programmable circuits may include reconfigurable hardware circuits, including logic AND, logic OR, logic XOR, logic NAND, logic NOR, and other logic operations, memory elements such as flip-flops, registers, field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), etc.

[0157] Computer-readable media may include any tangible device capable of storing instructions to be executed by a suitable device, and as a result, computer-readable media having instructions stored therein will comprise a product containing instructions that can be executed to create means for performing operations specified in a flowchart or block diagram. Examples of computer-readable media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable media may include floppy disks (registered trademark), diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disk read-only memory (CD-ROM), digital versatile disk (DVD), Blu-ray (registered trademark) disk, memory stick, integrated circuit card, etc.

[0158] Computer-readable instructions may include assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, Java®, C++, and conventional procedural programming languages ​​such as the C programming language or similar programming languages.

[0159] Computer-readable instructions may be provided locally or via a wide area network (WAN) such as a local area network (LAN) or the Internet to the processor or programmable circuit of a programmable data processing device such as a general-purpose computer, a special-purpose computer, or another computer, and the computer-readable instructions may be executed to create means for performing operations specified in a flowchart or block diagram. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

[0160] Figure 21 shows an example of a computer 2200 in which multiple aspects of the present invention may be embodied in whole or in part. A program installed on the computer 2200 can cause the computer 2200 to function as an operation or one or more sections of an apparatus according to an embodiment of the present invention, or to execute such operation or one or more sections, and / or to cause the computer 2200 to execute a process or a stage of such process according to an embodiment of the present invention. Such a program may be executed by the CPU 2212 to cause the computer 2200 to perform a particular operation associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0161] The computer 2200 according to this embodiment includes a CPU 2212, RAM 2214, a graphics controller 2216, and a display device 2218, which are interconnected by a host controller 2210. The computer 2200 also includes input / output units such as a communication interface 2222, a hard disk drive 2224, a DVD-ROM drive 2226, and an IC card drive, which are connected to the host controller 2210 via an input / output controller 2220. The computer also includes legacy input / output units such as a ROM 2230 and a keyboard 2242, which are connected to the input / output controller 2220 via an input / output chip 2240.

[0162] The CPU 2212 operates according to programs stored in the ROM 2230 and RAM 2214, thereby controlling each unit. The graphics controller 2216 acquires image data generated by the CPU 2212 from a frame buffer provided in RAM 2214 or from itself, and displays the image data on the display device 2218.

[0163] The communication interface 2222 communicates with other electronic devices via a network. The hard disk drive 2224 stores programs and data used by the CPU 2212 in the computer 2200. The DVD-ROM drive 2226 reads programs or data from the DVD-ROM 2201 and provides them to the hard disk drive 2224 via the RAM 2214. The IC card drive reads programs and data from the IC card and / or writes programs and data to the IC card.

[0164] The ROM 2230 stores boot programs and / or programs that depend on the computer 2200's hardware, which are executed by the computer 2200 when activated. The input / output chip 2240 may also connect various input / output units to the input / output controller 2220 via parallel ports, serial ports, keyboard ports, mouse ports, etc.

[0165] The program is provided on a computer-readable medium such as a DVD-ROM 2201 or an IC card. The program is read from the computer-readable medium and installed on a hard disk drive 2224, RAM 2214, or ROM 2230, which are examples of computer-readable mediums, and executed by the CPU 2212. The information processing described within these programs is read by the computer 2200, resulting in coordination between the program and the various types of hardware resources described above. The apparatus or method may be configured to realize the manipulation or processing of information in accordance with the use of the computer 2200.

[0166] For example, when communication is performed between a computer 2200 and an external device, the CPU 2212 may execute a communication program loaded into the RAM 2214 and, based on the processing described in the communication program, instruct the communication interface 2222 to perform communication processing. Under the control of the CPU 2212, the communication interface 2222 reads transmission data stored in a transmission buffer processing area provided in a recording medium such as the RAM 2214, hard disk drive 2224, DVD-ROM 2201, or IC card, transmits the read transmission data to the network, or writes received data received from the network to a reception buffer processing area provided on the recording medium.

[0167] The CPU 2212 may read all or necessary parts of a file or database stored on an external recording medium such as a hard disk drive 2224, a DVD-ROM drive 2226 (DVD-ROM 2201), or an IC card into the RAM 2214, and perform various types of processing on the data in the RAM 2214. The CPU 2212 then writes the processed data back to the external recording medium.

[0168] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and subjected to information processing. The CPU 2212 may perform various types of processing on the data read from the RAM 2214, including various types of operations, information processing, conditional judgments, conditional branching, unconditional branching, information retrieval / replacement, etc., as described throughout this disclosure and specified by the program instruction sequence, and write the results back to the RAM 2214. The CPU 2212 may also retrieve information in files, databases, etc., within the recording medium. For example, if a plurality of entries having attribute values ​​of a first attribute, each associated with an attribute value of a second attribute, are stored in the recording medium, the CPU 2212 may search among the plurality of entries for an entry that matches the condition for which the attribute value of the first attribute is specified, read the attribute value of the second attribute stored in that entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0169] The program or software module described above may be stored on or near the computer 2200 on a computer-readable medium. Alternatively, a recording medium such as a hard disk or RAM provided within a server system connected to a dedicated communication network or the Internet can be used as a computer-readable medium, thereby providing the program to the computer 2200 via the network. The computer-readable medium may be a non-temporary computer-readable medium. The computer-readable medium may record a program that causes the computer to execute the method of this embodiment.

[0170] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications or improvements can be made to the above embodiments. It will be clear from the claims that such modified or improved forms may also be included in the technical scope of the present invention.

[0171] It should be noted that the execution order of operations, procedures, steps, and stages in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not explicitly stated as "before," "prior to," etc., and can be implemented in any order unless the output of a previous process is used in a later process. Even if the operation flow in the claims, specifications, and drawings is described using phrases such as "first," "next," etc. for convenience, it does not mean that it is mandatory to perform them in that order. The notation "A and / or B" may mean "A, B, or A and B." The notation "A, B and / or C" may mean "any one of A, B, and C, or any combination of two or more of these."

[0172] 10 Experimental Condition Generation Device 105 Target Acquisition Unit 110 Starting Point Acquisition Unit 120 Starting Point Material Identification Unit 130 Similar Material Identification Unit 140 Search Area Setting Unit 150 Specified Condition Input Unit 152 Prediction Unit 154 Learning Unit 160 Experimental Condition Generation Unit 162 Candidate Generation Unit 164 Evaluation Unit 166 Output Unit 170 Experimental Data Acquisition Unit 702 Searched Conditions 704 Search Starting Point 910 Search Area 1102 Polymer A 1104 Polymer B 1106 Polymer C 1108 Polymer D 1110 Threshold Distance 1402 Recommended Experimental Conditions 2200 Computer 2201 DVD-ROM 2210 Host Controller 2212 CPU 2214 RAM 2216 Graphics Controller 2218 Display Device 2220 Input / Output Controller 2222 Communication Interface 2224 Hard Disk Drive 2226 DVD-ROM Drive 2230 ROM 2240 Input / Output Chip 2242 Keyboard

Claims

1. An experimental condition generation device that generates experimental conditions to be used as training data for a prediction model that predicts experimental results from experimental conditions, comprising: a starting point acquisition unit that acquires an experimental condition to be used as the starting point for a search from among a plurality of experimental conditions that have been tested and explored; a search area setting unit that sets a search area that indicates a range used for generating the experimental conditions, which is at least a part of the search space defined by the experimental conditions, based on the search starting point; and an experimental condition generation unit that generates experimental conditions different from the explored conditions from the search area.

2. The experimental condition generation apparatus according to claim 1, wherein the starting point acquisition unit acquires experimental conditions that satisfy predetermined conditions from among the searched conditions as the search starting point.

3. The experimental condition generation apparatus according to claim 1 or 2, wherein the search area setting unit sets the search area within a predetermined distance from the search starting point in the search space.

4. The experimental condition generation device according to any one of claims 1 to 3, wherein the experimental conditions are composed of a plurality of elemental conditions, and the search area setting unit sets the search area in the search space within a range in which the search starting point and at least some of the elemental conditions overlap.

5. The experimental condition generation device according to any one of claims 1 to 3, wherein the experimental conditions consist of a plurality of elemental conditions, and the search area setting unit sets the search area to a value within a predetermined range from the search starting point for the elemental conditions that take continuous values.

6. The experimental condition generation device according to any one of claims 1 to 3, wherein the experimental conditions are composed of a plurality of elemental conditions, and the search area setting unit sets a plurality of discrete options for the elemental conditions as the search area based on the search starting point.

7. The experimental condition generation device according to any one of claims 1 to 6, further comprising a learning unit that performs machine learning on a predictive model based on the experimental conditions generated by the experimental condition generation unit and the experimental results of said experimental conditions.

8. An experimental condition generation device according to any one of claims 1 to 7, comprising: a starting material identification unit that identifies a starting material which is a material used in the experimental conditions of the starting point of the search; and a similar material identification unit that identifies a similar material which is a material whose characteristics are similar to those of the starting material, wherein the search area setting unit generates a search area which includes experimental conditions which are used for the similar material.

9. The experimental condition generation device according to any one of claims 1 to 8, further comprising a prediction unit that inputs experimental conditions to the prediction model and outputs a prediction result, wherein the experimental condition generation unit generates the experimental conditions based on the prediction result predicted by the prediction unit.

10. The experimental condition generation device according to claim 9, further comprising: a learning unit that generates learning data for a search domain using the searched conditions included in the search domain; a learning unit that learns a prediction model for a search domain based on the learning data for the search domain; a prediction unit that inputs experimental conditions using the prediction model for the search domain and outputs a prediction result; and an experimental condition generation unit that generates the experimental conditions based on the prediction result predicted by the prediction unit using the prediction model for the search domain.

11. An experimental condition generation device according to any one of claims 1 to 10, further comprising a specified condition input unit for inputting a specified range which is a range specified by the user for some elemental conditions of the experimental conditions, wherein the experimental condition generation unit generates the experimental conditions from the specified range for the elemental conditions.

12. An experimental condition generation device according to any one of claims 1 to 10, further comprising a specified condition input unit for inputting specified options, which are selections made by the user for some of the elemental conditions of the experimental conditions, wherein the experimental condition generation unit generates the experimental conditions after selecting the elemental conditions from the specified options.

13. When, among the experimental conditions, the element conditions that are input by the user are designated as specified conditions, and the element conditions other than the specified conditions are designated as free conditions, the experimental condition generation unit comprises: a candidate generation unit that generates multiple sets of two or more experimental condition candidates in which the free conditions are the same and the specified conditions are different; an evaluation unit that comprehensively evaluates each of the multiple experimental condition candidate sets based on the prediction results predicted by the prediction unit for each experimental condition included in the set of experimental condition candidate sets; and an output unit that outputs a recommended set of experimental conditions based on the evaluation of the set of experimental condition candidate sets by the evaluation unit, the experimental condition generation device according to claim 9 or 10.

14. A program executed by a computer, which causes the computer to function as an experimental condition generating device according to any one of claims 1 to 13.

15. An experimental condition generation method comprising: an experimental condition generation device according to any one of claims 1 to 13, which performs the following steps: a starting point input step of receiving an experimental condition to be used as the starting point for a search from among a plurality of experimental conditions which are explored conditions; a search area setting step of setting a search area that is at least a part of the search space and indicates a range used for generating the experimental conditions in a search space defined by the experimental conditions based on the search starting point; and an experimental condition generation step of generating experimental conditions different from the explored conditions from the search area.

Images