Composition search device, composition search method, and program

Abstract

This composition search device comprises: a composition generation unit that generates a candidate composition of a target substance; a physical property prediction unit that predicts a physical property of the target substance on the basis of the candidate composition; an energy calculation unit that calculates energy corresponding to the candidate composition, on the basis of a loss function that minimizes a loss when a predicted value of the physical property is within a target range; and a result output unit that outputs a prescribed number of candidate compositions having small energy.

Description

Composition search device, composition search method, and program

[0001] The present disclosure relates to a composition searching device, a composition searching method, and a program.

[0002] In materials development, combinatorial optimization techniques such as simulated annealing (SA) are used. For example, Patent Literature 1 discloses a material property estimation method that determines an assumed value of a first property, obtains an analysis result of a second property when the first property is the assumed value, calculates an evaluation function value by comparing the analysis result of the second property with an actual measurement result, and specifies the optimal assumed value of the first property according to a global optimization technique.

[0003] JP 2010-25641 A

[0004] However, conventional techniques are unable to obtain multiple solutions. For example, if there is an allowable range for a specific variable, obtaining multiple solution patterns would enable efficient material development.

[0005] One aspect of the present disclosure aims to output multiple compositions whose physical properties fall within a target range.

[0006] The present disclosure has the following configuration.

[0007] <1> A composition searching device comprising: a composition generating unit configured to generate candidate compositions of a target substance produced using two or more raw materials; a property predicting unit configured to predict a property of the target substance based on the candidate compositions; an energy calculating unit configured to calculate an energy corresponding to the candidate composition based on a loss function that minimizes loss when the predicted value of the property is within a target range; and a result output unit configured to output a predetermined number of the candidate compositions with small energies.

[0008] <2> The composition search device according to <1> above, wherein the loss function outputs zero when the predicted value is within the target range, and outputs an absolute value of a difference between the predicted value and a boundary value of the target range when the predicted value is outside the target range.

[0009] <3> The composition search device according to <1> or <2>, wherein the energy calculation unit is configured to calculate the energy based on the loss function and a second loss function that minimizes the loss when the candidate composition is within a constraint range.

[0010] <4> The composition searching device according to <3> above, wherein the second loss function outputs zero when the candidate composition is within the constraint range, and outputs an absolute value of a difference between the candidate composition and a boundary value of the constraint range when the candidate composition is outside the constraint range.

[0011] <5> The composition searching device according to any one of <1> to <4> above, wherein the composition generating unit is configured to generate a new candidate composition by changing the blending amounts of two materials included in the candidate composition for which the energy has been calculated.

[0012] <6> The composition searching device described in <5> above, wherein the composition generating unit is configured to generate the new candidate composition by a generating method randomly selected from: a first generating method of increasing the blending amount of a first material substance and decreasing the blending amount of a second material substance; a second generating method of exchanging the blending amounts of the first material substance and the second material substance; a third generating method of averaging the largest blending amount and the smallest blending amount; and a fourth generating method of exchanging the smallest blending amount excluding zero with a blending amount that is zero.

[0013] <7> The composition searching device according to any one of <1> to <6> above, wherein a series of processes including generating the candidate compositions, predicting the physical properties, and calculating the energy, and a process of storing the candidate composition having the smallest calculated energy in the series of processes are repeatedly executed the predetermined number of times, and the result output unit is configured to output the stored candidate compositions.

[0014] <8> The composition searching device according to any one of <1> to <6> above, wherein the composition searching device repeatedly executes a series of processes including generating the candidate compositions, predicting the physical properties, calculating the energy, and storing the candidate compositions and the energy, and the result output unit is configured to output the predetermined number of the candidate compositions in ascending order of the stored energy.

[0015] <9> A composition search method in which a computer executes the following steps: generating candidate compositions of a target substance produced using two or more raw materials; predicting physical properties of the target substance based on the candidate compositions; calculating energies corresponding to the candidate compositions based on a loss function that minimizes loss when the predicted values ​​of the physical properties are within a target range; and outputting a predetermined number of the candidate compositions with small energies.

[0016] <10> A program for causing a computer to execute the following steps: generating candidate compositions of a target substance produced using two or more raw materials; predicting physical properties of the target substance based on the candidate compositions; calculating energies corresponding to the candidate compositions based on a loss function that minimizes loss when the predicted values ​​of the physical properties are within a target range; and outputting a predetermined number of the candidate compositions with small energies.

[0017] According to one aspect of the present disclosure, multiple compositions whose physical properties fall within a target range can be output.

[0018] FIG. 1 is a diagram showing an example of a loss function according to the prior art. FIG. 2 is a diagram showing an example of a loss function according to an embodiment. FIG. 3 is a block diagram showing an example of the overall configuration of a composition search system. FIG. 4 is a block diagram showing an example of the hardware configuration of a computer. FIG. 5 is a block diagram showing an example of the functional configuration of a composition search device. FIG. 6 is a flowchart showing a first example of a composition search method. FIG. 7 is a diagram showing an example of a condition input screen. FIG. 8 is a flowchart showing a second example of a composition search method.

[0019] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0020] [Embodiment] One embodiment of the present disclosure is an information processing system that searches for the composition of a substance manufactured using multiple substances. Hereinafter, the information processing system according to this embodiment will be referred to as a "composition search system." Furthermore, a substance to be designed will be referred to as a "target substance," and a substance used to manufacture the target substance will be referred to as a "material substance."

[0021] The target substance may be, for example, a semiconductor material. The semiconductor material is a composite material containing multiple resins and at least one of an additive and a filler. Examples of the semiconductor material include a resist material, an adhesive, a pressure-sensitive adhesive, and a sealing material.

[0022] Combinatorial optimization methods are sometimes used in materials development, allowing developers to efficiently search for promising combinations that have not been predicted by the developer from among a vast number of possible combinations of substance compounds.

[0023] An example of a combinatorial optimization method is a global optimization method known as simulated annealing or simulated annealing. Simulated annealing is a method specialized for combinatorial optimization, regardless of whether it involves multiple variables or a single variable. Simulated annealing outputs only one optimal solution under the same initial conditions. On the other hand, in materials development, there may be an acceptable range for variables. Obtaining multiple solution patterns that satisfy the acceptable range allows for efficient development. For example, since experiments are essential in materials development, it is efficient to conduct experiments in parallel using multiple combination patterns.

[0024] In this embodiment, a loss function that enables multiple solutions to be output in the simulated annealing method is introduced. The difference between the loss function of the conventional technology and the loss function of this embodiment will be described with reference to FIGS. 1 and 2.

[0025] Fig. 1 is a diagram showing an example of a loss function according to the prior art. Fig. 1 shows a graph showing the relationship between the search target and energy, with the horizontal axis representing the search target and the vertical axis representing energy. In the simulated annealing method, the energy is calculated based on the loss function, and the smaller the energy, the more appropriate the solution is evaluated to be.

[0026] In simulated annealing, if the energy decreases between the current state and the next state, a transition to the next state is always made, and if the energy increases between the current state and the next state, a transition to the next state is made probabilistically. For example, in FIG. 1, a transition from state C1 to state C2 is always made, but a transition from state C2 to state C0 is made probabilistically. In simulated annealing, even if the energy increases, a transition occurs probabilistically, so the possibility of falling into a local optimum is reduced, and a global optimum with the minimum energy can be found.

[0027] As shown in Figure 1, a conventional loss function has a single solution with minimum energy. Figure 1 shows that the minimum energy E0 is at the state C0. Therefore, with a conventional loss function, only the state C0 is output as the optimal solution.

[0028] 2 is a diagram illustrating an example of a loss function according to the embodiment. As shown in FIG. 2, the loss function according to the embodiment is configured to minimize energy within a certain range. FIG. 2 shows that the minimum energy E0 is achieved in both state C0A and state C0B. Therefore, the loss function according to the embodiment can output a plurality of states C0A and C0B as optimal solutions.

[0029] The present embodiment aims to output a plurality of compositions whose physical properties fall within a target range. To this end, the present embodiment predicts the physical properties of a target substance based on candidate compositions of the target substance, calculates the energies corresponding to the candidate compositions based on a loss function that minimizes the loss when the predicted values ​​of the physical properties fall within the target range, and outputs a predetermined number of candidate compositions with low energies.

[0030] In one aspect, this embodiment uses a loss function that provides the minimum energy if the predicted value of the physical property is within the target range, so that multiple compositions whose physical properties fall within the target range can be output. In another aspect, this embodiment can output multiple promising compositions at once, so that material development can proceed efficiently.

[0031] <Overall Configuration> The overall configuration of the composition searching system according to this embodiment will be described with reference to Fig. 3. Fig. 3 is a block diagram showing an example of the overall configuration of the composition searching system.

[0032] 3, the composition search system 1000 includes a composition search device 10 and a terminal device 20. The composition search device 10 and the terminal device 20 are connected to each other so as to be able to communicate data with each other via a communication network N. The communication network N may be, for example, a local area network (LAN), a virtual private network (VPN), or the Internet.

[0033] The composition searching device 10 is an example of an information processing device such as a personal computer, workstation, or server that searches for the composition of a target substance. The composition searching device 10 receives search conditions for the target substance from the terminal device 20 and searches for the composition of the target substance based on the search conditions. The composition searching device 10 transmits the search results for the composition of the target substance to the terminal device 20.

[0034] The terminal device 20 is an example of an information processing terminal such as a personal computer, smartphone, or tablet terminal operated by a user of the composition search system 1000. The terminal device 20 transmits search conditions specified by the user to the composition search device 10. The terminal device 20 receives search results from the composition search device 10 and presents them to the user.

[0035] The overall configuration of the composition search system 1000 shown in FIG. 3 is one example, and various system configurations are possible depending on the application and purpose. For example, the composition search system 1000 may include multiple composition search devices 10 and one or more terminal devices 20. For example, the composition search device 10 may be implemented by multiple computers, or may be implemented as a cloud computing service. For example, the composition search system 1000 may be implemented by a standalone computer. The division of devices such as the composition search device 10 and terminal device 20 shown in FIG. 3 is one example.

[0036] <Hardware Configuration> The hardware configuration of the composition search system 1000 will be described with reference to Fig. 4. The composition search device 10 and the terminal device 20 are realized by, for example, a computer. Fig. 4 is a block diagram showing an example of the hardware configuration of a computer.

[0037] 4, the computer 500 includes a CPU (Central Processing Unit) 501, a ROM (Read Only Memory) 502, a RAM (Random Access Memory) 503, a HDD (Hard Disk Drive) 504, an input device 505, a display device 506, a communication I / F (Interface) 507, and an external I / F 508. The CPU 501, the ROM 502, and the RAM 503 form a so-called computer. The hardware components of the computer 500 are connected to each other via a bus line 509. The input device 505 and the display device 506 may be connected to the computer 500 via the external I / F 508 for use.

[0038] The CPU 501 is a computing device that reads programs and data from a storage device such as the ROM 502 or the HDD 504 onto the RAM 503 and executes processing to realize overall control and functions of the computer 500. The computer 500 may have a GPU (Graphics Processing Unit) in addition to or instead of the CPU 501.

[0039] The ROM 502 is an example of a non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. The ROM 502 functions as a main storage device that stores various programs, data, etc. required for the CPU 501 to execute various programs installed in the HDD 504. Specifically, the ROM 502 stores boot programs such as a Basic Input Output System (BIOS) and an Extensible Firmware Interface (EFI) that are executed when the computer 500 starts up, as well as data such as OS (Operating System) settings and network settings.

[0040] The RAM 503 is an example of a volatile semiconductor memory (storage device) in which programs and data are erased when the power is turned off. The RAM 503 is, for example, a dynamic random access memory (DRAM) or a static random access memory (SRAM). The RAM 503 provides a working area in which various programs installed in the HDD 504 are expanded when executed by the CPU 501.

[0041] The HDD 504 is an example of a non-volatile storage device that stores programs and data. The programs and data stored in the HDD 504 include an OS, which is basic software that controls the entire computer 500, and applications that provide various functions on the OS. Note that the computer 500 may use a storage device that uses flash memory as a storage medium (e.g., an SSD (Solid State Drive)) instead of the HDD 504.

[0042] The input device 505 includes a touch panel, operation keys and buttons, a keyboard and mouse, a microphone for inputting sound data such as voice, and the like, which are used by the user to input various signals.

[0043] The display device 506 is composed of a display such as a liquid crystal display or organic electroluminescence (EL) display for displaying a screen, a speaker for outputting sound data such as voice, and the like.

[0044] The communication I / F 507 is an interface that connects to a communication network and enables the computer 500 to perform data communication.

[0045] The external I / F 508 is an interface with external devices, such as a drive device 510.

[0046] The drive device 510 is a device for loading a recording medium 511. The recording medium 511 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 511 may also include semiconductor memories that record information electrically, such as ROMs and flash memories. This allows the computer 500 to read from and / or write to the recording medium 511 via the external I / F 508.

[0047] The various programs to be installed in the HDD 504 are installed, for example, by setting the distributed recording medium 511 in a drive device 510 connected to the external I / F 508 and reading the various programs recorded on the recording medium 511 by the drive device 510. Alternatively, the various programs to be installed in the HDD 504 may be installed by being downloaded from the communication network N or another network different from the communication network N via the communication I / F 507.

[0048] <Functional Configuration> The functional configuration of the composition searching device 10 will be described with reference to Fig. 5. Fig. 5 is a block diagram showing an example of the functional configuration of the composition searching device.

[0049] 5 , composition searching device 10 includes condition acquisition unit 110, composition generation unit 120, property prediction unit 130, function determination unit 140, loss calculation unit 150, energy calculation unit 160, composition storage unit 170, and result output unit 180. Composition searching device 10 functions as condition acquisition unit 110, composition generation unit 120, property prediction unit 130, function determination unit 140, loss calculation unit 150, energy calculation unit 160, composition storage unit 170, and result output unit 180 by executing a program installed in advance.

[0050] For example, the condition acquisition unit 110, composition generation unit 120, property prediction unit 130, function determination unit 140, loss calculation unit 150, energy calculation unit 160, and result output unit 180 are realized by processing that is executed by the CPU 501 in accordance with a program loaded from the HDD 504 onto the RAM 503 shown in Fig. 4. For example, the composition storage unit 170 is realized by the HDD 504 shown in Fig. 4.

[0051] The condition acquisition unit 110 acquires search conditions. The search conditions are information indicating conditions related to the search for the composition of a target substance. The search conditions include at least the physical properties of the target substance, target ranges of the physical properties, a list of raw materials, constraint ranges for the raw materials, the number of output compositions, and simulation settings.

[0052] The physical properties of the target substance are information indicating one or more physical properties that the target substance should satisfy. The physical properties of the target substance may be determined arbitrarily depending on, for example, the type of target substance or the purpose of material development. In this embodiment, the physical properties of the target substance may include, for example, a refractive index, an absorbance, a functional group concentration, or a functional group ratio.

[0053] The target range of a physical property is information indicating the range of values ​​that the physical property of the target substance should satisfy. The target range may include at least one boundary value (upper limit or lower limit). If the target range includes only one boundary value (upper limit or lower limit), the other boundary value (lower limit or upper limit) may be the minimum or maximum value that the physical property can take.

[0054] The list of materials is information indicating a plurality of materials used to produce a target substance. The number of materials may be determined arbitrarily depending on the type of target substance, etc. The list of materials may include the physical properties of each material. The physical properties of the material may include the same physical properties as the target substance. In this embodiment, the physical properties of the material may include, for example, specific gravity, volume ratio, weight ratio, molecular weight, refractive index, absorbance, functional group concentration, and functional group ratio.

[0055] The constraint range of a material is information indicating the conditions for the blending amount (e.g., weight %) of each material in the composition of the target material. The constraint range may include at least one boundary value (upper limit or lower limit). When the target range includes only one boundary value (upper limit or lower limit), the other boundary value (lower limit or upper limit) may be the minimum value (e.g., 0 wt %) or maximum value (e.g., 100 wt %) that the physical property can take.

[0056] The constraint range of the material may include constraint ranges for multiple material materials. The constraint range for the multiple material materials may include at least one of an upper limit value or a lower limit value for a calculation result using the blending amounts of the multiple material materials. The calculation result may include, for example, a sum, a difference, a product, a ratio, etc. For example, the constraint range of the material may be the range of the sum of a first material material and a second material material. Furthermore, the constraint range of the material may be the range of the ratio of the first material material to the second material material.

[0057] The number of output compositions is information indicating the number of compositions to be output as a search result. The number of output compositions may be arbitrarily determined by the user. The number of output compositions may be, for example, two or more, or may be around ten.

[0058] The simulation settings may include parameters related to the simulated annealing method. Examples of the parameters related to the simulated annealing method include a maximum temperature, a minimum temperature, a number of steps, a random value, a weight, and a verbose level. The parameters related to the simulated annealing method may be predetermined.

[0059] The maximum temperature and minimum temperature are parameters that control the temperature schedule. The maximum temperature is set at the beginning of the annealing process. The minimum temperature is set at the end of the annealing process.

[0060] The number of steps is the number of times the energy calculation is repeated. In this embodiment, the number of steps may be, for example, about 500,000. The weight is a parameter used to weight each loss when calculating a loss score by adding multiple losses. The redundancy is a parameter that controls how many steps the energy is calculated at.

[0061] The condition acquisition unit 110 may receive information or a signal indicating the search conditions from the terminal device 20. The condition acquisition unit 110 may accept input of the search conditions via the input device 505 of the composition search device 10. The condition acquisition unit 110 may read out search conditions that are pre-stored in a storage device such as the HDD 504 of the composition search device 10.

[0062] The composition generating unit 120 generates a composition of the target substance. The composition may be expressed as an array of blending amounts of each of the raw materials. The composition generating unit 120 may generate an initial composition and candidate compositions. The composition generating unit 120 may generate the initial composition and candidate compositions based on the search conditions acquired by the condition acquiring unit 110.

[0063] The initial composition is the initial value of the composition of the target substance. The composition generator 120 may generate the initial composition by setting the blending amount of each ingredient substance to zero. For example, if the target substance includes four ingredients, the initial composition may be [0,0,0,0].

[0064] The candidate composition is a candidate for the composition to be searched for, and is a composition for which energy is calculated. The composition generation unit 120 may generate multiple candidate compositions. The composition generation unit 120 may repeat generating candidate compositions for the number of steps. The composition generation unit 120 may generate a new candidate composition based on the composition generated previously. If an initial composition was generated previously, the composition generation unit 120 may generate a candidate composition in which the blending amounts of each ingredient are equal. If the target substance contains four ingredients, the candidate composition generated after the initial composition may be, for example, [25, 25, 25, 25].

[0065] When a candidate composition has been previously generated, the composition generator 120 may generate a new candidate composition by changing the blending amounts of two materials contained in the previously generated candidate composition (in other words, the candidate composition for which the energy has been calculated). Specifically, the composition generator 120 may generate a new candidate composition by a generation method randomly selected from four generation methods.

[0066] The first generation method is a generation method in which the amount of the first material is increased and the amount of the second material is decreased. The first material and the second material may be selected randomly. The amount of increase and decrease in the amount of the increase and decrease may be determined randomly.

[0067] The second production method is a production method in which the blending amounts of the first material and the second material are exchanged. The first material and the second material may be selected randomly.

[0068] The third generation method is to average the largest and smallest blending amounts. When there are multiple largest or smallest blending amounts, the material whose blending amounts are averaged may be selected randomly.

[0069] The fourth generation method is a generation method in which the smallest non-zero blend amount is exchanged with a blend amount of 0. When there are multiple smallest blend amounts or multiple blend amounts of 0, the material substances to be exchanged may be selected randomly.

[0070] The physical property prediction unit 130 predicts the physical properties of the target substance. The physical property prediction unit 130 may predict the physical properties of the target substance based on the candidate compositions generated by the composition generation unit 120. The physical property prediction unit 130 may predict a physical property included in the search conditions acquired by the condition acquisition unit 110. The physical property prediction unit 130 may predict multiple physical properties included in the search conditions.

[0071] The property prediction unit 130 may predict the properties of the target substance based on a property prediction function corresponding to the properties. The property prediction function may be stored in advance in a storage device of the composition searching apparatus 10. The property prediction function may be included in the search conditions acquired by the condition acquisition unit 110.

[0072] In this embodiment, the physical property prediction unit 130 may predict the refractive index of the target substance by calculating a physical property prediction function defined by equation (1).

[0073]

[0074] However, n ｍｉｘ is the refractive index of the target substance, and n ｉ is the refractive index of the i-th material, and v ｉ is the volume ratio of the i-th material.

[0075] In this embodiment, the physical property prediction unit 130 may predict the absorbance of the target substance by calculating a physical property prediction function defined by equation (2).

[0076]

[0077] However, A ｍｉｘ is the absorbance of the target substance, and A ｉ is the absorbance of the i-th material, and v ｉ is the volume ratio of the i-th material.

[0078] In this embodiment, the property prediction unit 130 may predict the functional group concentration of the target substance by calculating the property prediction function defined by equation (3).

[0079]

[0080] However, c ｆ is the molar concentration of functional group f, and ρ ｍｉｘis the specific gravity of the target substance, and w ｉ is the weight ratio of the i-th material, and M ｉ is the molecular weight of the i-th material, and m ｆ－ｉ is the number of functional groups f in the i-th material. Examples of functional groups f include acryloyl groups, amide groups, amino groups, aldehyde groups, isocyanate groups, imine groups, ether groups, ester groups, epoxy groups, carboxy groups, glycidyl groups, cyano groups, nitro groups, hydroxy groups, phenyl groups, maleimide groups, and mercapto groups.

[0081] The specific gravity ρ in formula (3) ｍｉｘ can be calculated using formula (4). ｉ can be calculated by equation (5).

[0082]

[0083] However, ρ ｉ is the specific gravity of the i-th material, and v ｉ is the volume ratio of the i-th material.

[0084] The physical property prediction unit 130 outputs a prediction result of the physical property of the target substance. The prediction result may include a predicted value of the physical property of the target substance. When multiple physical properties are predicted, the prediction result may be expressed as an array of predicted values ​​for each physical property. As an example, if the predicted value of the refractive index is 1.57 and the predicted value of the absorbance is 0.04, the prediction result may be [1.57, 0.04].

[0085] The function determination unit 140 determines a loss function. The loss function may include a first loss function related to physical properties and a second loss function related to composition. The function determination unit 140 may determine the loss function based on the search conditions acquired by the condition acquisition unit 110. For example, the function determination unit 140 may determine the first loss function based on a target range of physical properties included in the search conditions. Furthermore, the function determination unit 140 may determine the second loss function based on a constraint range of a material included in the search conditions.

[0086] The first loss function is a function that minimizes the loss when the predicted value of the physical property is within a target range. The first loss function may be a function that outputs zero when the predicted value of the physical property is within the target range, and outputs the absolute value of the difference between the predicted value of the physical property and the boundary value of the target range when the predicted value of the physical property is outside the target range.

[0087] Specifically, the first loss function may be a function f(x) defined as follows: f(x)=|min(median value+threshold value, current value)−max(median value−threshold value, current value)|

[0088] Here, the median is the median (average) between the upper and lower limits of the target range. The threshold is the absolute value of the difference between the median and the boundary value (upper or lower limit). The current value is the predicted value of the physical property based on the current candidate composition.

[0089] When multiple physical properties are predicted, the median may be expressed as an array of medians for each physical property. The threshold may also be expressed as an array of thresholds for each physical property. For example, if the target range for refractive index is 1.50 to 1.60 and the target range for absorbance is 0.01 to 0.03, the median may be [1.55, 0.02] and the threshold may be [0.05, 0.01].

[0090] The second loss function is a function that minimizes the loss when the composition is within the constraint range. The second loss function may be a function that outputs zero when the composition is within the constraint range and outputs the absolute value of the difference from the boundary value of the constraint range when the composition is outside the constraint range.

[0091] Specifically, the second loss function may be a function f(v) defined as follows: f(v)=|min(median value+threshold value, current value)−max(median value−threshold value, current value)|

[0092] Here, the median is the median (average) between the upper and lower limits of the constraint range, the threshold is the absolute value of the difference between the median and the boundary value (upper or lower limit), and the current value is the current candidate composition.

[0093] The median may be expressed as an array of medians for each of the ingredients. The thresholds may be expressed as an array of thresholds for each of the ingredients. For example, if the target substance includes four ingredients, and the first ingredient has a constraint range of 27 to 33, the second ingredient has a constraint range of 16 to 24, the third ingredient has a constraint range of 16 to 24, and the fourth ingredient has a constraint range of 27 to 33, the medians may be [30, 20, 20, 30], and the thresholds may be [3, 4, 4, 3].

[0094] The loss calculation unit 150 calculates a loss score corresponding to the candidate composition. The loss calculation unit 150 may calculate the loss score based on the loss function determined by the function determination unit 140. The loss calculation unit 150 may calculate the loss score using a first loss function based on the prediction result of the physical property predicted by the physical property prediction unit 130. The loss calculation unit 150 may calculate the loss score using a second loss function based on the current candidate composition.

[0095] The loss calculation unit 150 may calculate the loss score based on the first loss function and the second loss function. The loss calculation unit 150 may calculate the loss score by weighting and adding the first loss calculated using the first loss function and the second loss calculated using the second loss function. The loss calculation unit 150 may weight and add the first loss and the second loss using weights included in the search conditions acquired by the condition acquisition unit 110.

[0096] If multiple properties are predicted, the first loss may be represented as an array of losses for each property. For example, if the calculated loss for refractive index is 0 and the calculated loss for absorbance is 0.01, the first loss may be [0, 0.01].

[0097] The second loss may be expressed as an array of the losses for each of the materials. As an example, if the calculated loss for a first material is 2, the calculated loss for a second material is 1, the calculated loss for a third material is 1, and the calculated loss for a fourth material is 2, the first loss may be [2, 1, 1, 2].

[0098] When multiple physical properties are predicted, the loss calculation unit 150 may calculate the loss score by adding the sum of each element of the first loss and the sum of each element of the second loss. When performing weighted addition, the loss calculation unit 150 may multiply each element of the first loss and each element of the second loss by a weight and then calculate the sum of each element.

[0099] The energy calculation unit 160 calculates the energy corresponding to the candidate composition. The energy calculation unit 160 may calculate the energy based on the loss score calculated by the loss calculation unit 150. In this embodiment, the energy calculation unit 160 may calculate the energy by multiplying the loss score by a predetermined coefficient. The predetermined coefficient may be 1000, for example.

[0100] The composition storage unit 170 stores candidate compositions and the energies corresponding to the candidate compositions. The composition storage unit 170 may store candidate compositions generated by the composition generation unit 120 and energies calculated by the energy calculation unit 160 in association with each other. The composition storage unit 170 may store only the candidate composition with the smallest energy. The composition storage unit 170 may store multiple candidate compositions with the smallest energy. The composition storage unit 170 may store all candidate compositions for which the energies have been calculated.

[0101] The result output unit 180 outputs the search results for the composition of the target substance. The search results may include a predetermined number of candidate compositions. The search results may include a predetermined number of candidate compositions with low energy calculated by the energy calculation unit 160. The number of candidate compositions included in the search results may be the number of output compositions included in the search conditions acquired by the condition acquisition unit 110.

[0102] The result output unit 180 may generate search results based on candidate compositions read from the composition storage unit 170. When only candidate compositions with the smallest energy are stored in the composition storage unit 170, the result output unit 180 may include all candidate compositions read from the composition storage unit 170 in the search results. When all candidate compositions for which energies have already been calculated are stored in the composition storage unit 170, the result output unit 180 may include a predetermined number of candidate compositions with the smallest energy among the candidate compositions read from the composition storage unit 170 in the search results.

[0103] 5 is merely an example, and there are, of course, various examples of functional configurations depending on the application and purpose. The division of processing units or storage units, such as the condition acquisition unit 110, composition generation unit 120, property prediction unit 130, function determination unit 140, loss calculation unit 150, energy calculation unit 160, composition storage unit 170, and result output unit 180 shown in FIG. 5, is merely an example.

[0104] For example, at least two of the condition acquisition unit 110, composition generation unit 120, property prediction unit 130, function determination unit 140, loss calculation unit 150, energy calculation unit 160, and result output unit 180 may be integrated into one processing unit. For example, at least one of the condition acquisition unit 110, composition generation unit 120, property prediction unit 130, function determination unit 140, loss calculation unit 150, energy calculation unit 160, and result output unit 180 may be divided into multiple processing units.

[0105] For example, the composition storage unit 170 may be divided into multiple storage units. For example, the composition storage unit 170 may not be included in the composition search device 10, but may be included in an external information processing device or storage device. Note that "external" means not included in the composition search system 1000.

[0106] <Processing Procedure> A composition search method executed by the composition search system 1000 will be described with reference to Fig. 6. Fig. 6 is a flowchart showing a first example of the composition search method.

[0107] In step S1, a user of the composition search system 1000 inputs search conditions for a target substance into the terminal device 20. The terminal device 20 accepts the search conditions input by the user. For example, the terminal device 20 may display a condition input screen for inputting search conditions on the display device 506 and accept the search conditions input by the user into the condition input screen. The terminal device 20 transmits information or a signal indicating the accepted search conditions to the composition search device 10.

[0108] The composition search device 10 receives information indicating search conditions from the terminal device 20. The condition acquisition unit 110 of the composition search device 10 acquires the search conditions from the information indicating the search conditions received by the composition search device 10. The condition acquisition unit 110 sends the search conditions to the composition generation unit 120 and the function determination unit 140.

[0109] 7 is a diagram showing an example of a condition input screen 600. As shown in FIG. 7, the condition input screen 600 has a resin selection section 601, a constraint input section 602, an optical property input section 603, a functional group property input section 604, a setting input section 605, and a start button 606.

[0110] The resin selection unit 601 receives a selection of a resin, which is an example of a material. The resin selection unit 601 may receive a selection of a resin to be excluded. The resin selection unit 601 may receive a selection of a resin to be included.

[0111] The constraint input unit 602 accepts input of a constraint range for the resin selected by the resin selection unit 601. The constraint input unit 602 may accept input of upper and lower limit values ​​for the blending amount of the resin. If multiple resins are selected by the resin selection unit 601, the constraint input unit 602 may accept input of upper and lower limit values ​​for the blending amount for each resin.

[0112] The optical property input unit 603 accepts input of a target range of an optical property, which is an example of a physical property. The optical property input unit 603 may accept input of an upper limit value and a lower limit value of a refractive index, which is an example of an optical property. The optical property input unit 603 may accept input of an upper limit value and a lower limit value of an absorbance, which is an example of an optical property. The optical property input unit 603 may accept input of calculation conditions for the optical property. The calculation conditions for the optical property may include, for example, wavelength.

[0113] The functional group characteristic input unit 604 accepts input of a target range of a functional group characteristic, which is an example of a physical property. The functional group characteristic input unit 604 may accept input of an upper limit value and a lower limit value of a functional group concentration, which is an example of a functional group characteristic. The functional group characteristic input unit 604 may accept input of an upper limit value and a lower limit value of a functional group ratio, which is an example of a functional group characteristic.

[0114] The setting input section 605 receives input of setting information related to composition search. The setting input section 605 may include the number of resins to be included in the target substance or the number of output compositions.

[0115] The start button 606 is a button for starting a composition search. When the user presses the start button 606, the terminal device 20 accepts the search conditions entered on the condition input screen 600. Upon accepting the input of the search conditions, the terminal device 20 transmits information or a signal indicating the search conditions to the composition search device 10.

[0116] Returning to FIG. 6 , in step S2, the function determination unit 140 of the composition search device 10 receives the search conditions from the condition acquisition unit 110. The function determination unit 140 determines a first loss function and a second loss function based on the search conditions. The function determination unit 140 sends the first loss function and the second loss function to the loss calculation unit 150.

[0117] Specifically, the function determination unit 140 determines a first loss function for a physical property based on a target range of the physical property included in the search conditions. In this embodiment, the function determination unit 140 calculates a median and a threshold based on the target range and sets them as the function f(x), thereby determining the first loss function.

[0118] Furthermore, the function determination unit 140 determines a second loss function for the composition based on the constraint range of the composition included in the search conditions. In this embodiment, the function determination unit 140 calculates a median and a threshold based on the constraint range and sets them as the function f(v), thereby determining the second loss function.

[0119] In step S3, the composition generator 120 of the composition search device 10 receives the search conditions from the condition acquirer 110. The composition generator 120 generates an initial composition based on the search conditions. Specifically, the composition generator 120 generates an array indicating the blending amounts of each material substance and sets all elements of the array to zero.

[0120] The composition generator 120 also counts up the number of repeats. The number of repeats is the number of times the process from step S3 to step S10 has been repeated. The number of repeats can also be considered the number of times the initial composition has been generated. Here, the composition generator 120 sets the number of repeats to 1.

[0121] In step S4, the composition generator 120 of the composition searching apparatus 10 generates candidate compositions based on the initial composition. Specifically, the composition generator 120 sets each element of the initial composition to a value obtained by dividing 100 by the number of materials. The composition generator 120 sends the candidate compositions to the property predictor 130.

[0122] The composition generator 120 also counts up the number of steps. The number of steps is the number of times the processes from step S4 to step S8 are repeated. The number of steps can also be considered the number of times candidate compositions are generated after one initial composition is generated. Here, the composition generator 120 sets the number of steps to 1.

[0123] In step S5, the property prediction unit 130 of the composition searching apparatus 10 receives the candidate compositions from the composition generation unit 120. The property prediction unit 130 predicts the properties of the target substance based on the candidate compositions. In this embodiment, the property prediction unit 130 predicts the refractive index of the target substance by calculating Equation (1). The property prediction unit 130 also predicts the absorbance of the target substance by calculating Equation (2). The property prediction unit 130 sends the candidate compositions and the prediction results to the loss calculation unit 150.

[0124] In step S6, the loss calculation unit 150 of the composition searching apparatus 10 receives the first loss function and the second loss function from the function determination unit 140. The loss calculation unit 150 also receives the candidate compositions and the prediction results from the property prediction unit 130.

[0125] The loss calculation unit 150 calculates the first loss by inputting the prediction result into a first loss function. The loss calculation unit 150 calculates the second loss by inputting the candidate composition into a second loss function. The loss calculation unit 150 performs a weighted addition of the first loss and the second loss based on the weights included in the search conditions. This calculates a loss score corresponding to the candidate composition. The loss calculation unit 150 sends the loss score to the energy calculation unit 160.

[0126] In step S7, the energy calculation unit 160 of the composition searching apparatus 10 receives the loss score from the loss calculation unit 150. The energy calculation unit 160 calculates the energy corresponding to the candidate composition based on the loss score. Specifically, the energy calculation unit 160 multiplies the loss score by 1000.

[0127] The energy calculation unit 160 stores candidate compositions and their energies in association with each other. If a candidate composition is already stored, the energy calculation unit 160 compares the stored energy with the energy corresponding to the current candidate composition. If the energy corresponding to the current candidate composition is smaller than the stored energy, the energy calculation unit 160 discards the stored candidate composition and its energy and stores the current candidate composition and its energy in association with each other.

[0128] In step S8, the composition search device 10 determines whether a predetermined number of steps have been executed. Specifically, the composition search device 10 determines whether the current number of steps is equal to or greater than the number of steps included in the search conditions. If it is determined that the predetermined number of steps have been executed (YES), the composition search device 10 proceeds to step S9. On the other hand, if it is determined that the predetermined number of steps have not been executed (NO), the composition search device 10 returns to step S4.

[0129] Returning to step S4, the composition generator 120 generates a new candidate composition based on the current candidate composition. The composition generator 120 selects a generation method to be used for generating the candidate composition from among the first to fourth generation methods. The composition generator 120 generates a new candidate composition from the current candidate composition using the selected generation method. The composition generator 120 also counts up the number of steps.

[0130] Thereafter, the composition searching apparatus 10 repeats the process from step S5 to step S8 for the new candidate composition. In this manner, the composition searching apparatus 10 repeatedly executes a series of processes including generating candidate compositions, predicting physical properties, calculating loss scores, and calculating energy, until it determines in step S8 that a predetermined number of steps have been executed.

[0131] In step S9, the energy calculation unit 160 of the composition searching apparatus 10 stores the candidate compositions and energies it holds in the composition storage unit 170. Since the energy calculation unit 160 holds candidate compositions with low energy each time it calculates an energy, the composition storage unit 170 stores the candidate composition with the smallest energy among the candidate compositions whose energies have been calculated during the execution of a predetermined number of steps.

[0132] In step S10, the composition search device 10 determines whether the process from step S3 to step S8 has been repeated a predetermined number of times. Specifically, the composition search device 10 determines whether the current number of repetitions is equal to or greater than the number of output compositions included in the search conditions. If it is determined that the process has been repeated the predetermined number of times (YES), the composition search device 10 proceeds to step S11. On the other hand, if it is determined that the process has not been repeated the predetermined number of times (NO), the composition search device 10 returns to step S3.

[0133] When the process returns to step S3, the composition generator 120 generates a new initial composition and counts up the number of repeats.

[0134] Thereafter, the composition searching apparatus 10 executes the process from step S4 to step S10 again for a new initial composition. In this manner, the composition searching apparatus 10 generates an initial composition until it determines that the process has been repeated a predetermined number of times in step S10, and for that initial composition, it repeatedly executes a series of processes including the generation of candidate compositions, prediction of physical properties, calculation of loss scores, and calculation of energy for the number of steps.

[0135] In step S11, the result output unit 180 of the composition search device 10 outputs the search results for the composition of the target substance. The result output unit 180 reads all candidate compositions stored in the composition storage unit 170 and includes them in the search results. In this embodiment, the composition storage unit 170 stores candidate compositions for the number of output compositions included in the search conditions, and therefore the search results include candidate compositions with the number of output compositions having low energy. The result output unit 180 transmits the search results for the composition of the target substance to the terminal device 20.

[0136] The terminal device 20 receives the search results for the composition of the target substance from the composition search device 10. The terminal device 20 presents the search results for the composition of the target substance to the user. The terminal device 20 may display candidate compositions included in the search results on the display device 506. The terminal device 20 may display the candidate compositions together with the energies corresponding to the candidate compositions.

[0137] A user of the composition search system 1000 may refer to the candidate compositions displayed on the display device 506 of the terminal device 20 and use them in material development. For example, the user may produce a target substance using raw materials based on multiple candidate compositions. The user may also conduct an experiment using the produced target substance. The user may reconsider the search conditions based on the experimental results and request the composition search device 10 to perform a composition search based on the new search conditions.

[0138] Effects of the Embodiment The composition searching device 10 according to the present embodiment predicts the physical properties of a target substance based on candidate compositions of the target substance produced using two or more raw materials, calculates the energy corresponding to the candidate compositions based on a loss function that minimizes the loss when the predicted values ​​of the physical properties are within a target range, and outputs a predetermined number of candidate compositions with low energy. In one aspect, the present embodiment can output multiple compositions whose physical properties are within a target range. In another aspect, the present embodiment can efficiently advance the design of materials.

[0139] The loss function may output zero when the predicted value of the physical property is within the target range, and may output the absolute value of the difference from the boundary value of the target range when the predicted value of the physical property is outside the target range. In one aspect, according to this embodiment, the energy increases as the physical property deviates from the target range, making it possible to realize a search that is less likely to fall into a local optimum solution.

[0140] The composition searching apparatus 10 may calculate the energy based on the loss function and a second loss function that minimizes the loss when the candidate composition is within the constraint range. According to one aspect, this embodiment can output a plurality of compositions that fall within the constraint range among compositions whose physical properties fall within the target range.

[0141] The second loss function may output zero when the candidate composition is within the constraint range, and may output the absolute value of the difference between the candidate composition and the boundary value of the constraint range when the candidate composition is outside the constraint range. In one aspect, according to this embodiment, the energy increases as the composition deviates from the constraint range, making it possible to realize a search that is less likely to fall into a local optimum.

[0142] The composition searching device 10 may generate a second candidate composition by changing the amounts of two substances contained in the first candidate composition. The composition searching device 10 may generate the second candidate composition by a randomly selected method from among a first generation method that increases the amount of the first substance and decreases the amount of the second substance, a second generation method that exchanges the amounts of the first substance and the second substance, a third generation method that averages the largest and smallest amounts, and a fourth generation method that exchanges the smallest non-zero amount with a zero amount. In one aspect, this embodiment allows for random generation of a new composition while retaining the characteristics of the original composition.

[0143] The composition searching apparatus 10 may repeat a predetermined number of steps, including generating candidate compositions, predicting physical properties, and calculating energies, and storing the candidate composition with the smallest calculated energy among the repeated steps, and output the stored candidate compositions. In one aspect, this embodiment can output multiple compositions with physical properties that fall within a target range.

[0144] [Modification] In the embodiment, a configuration was described in which a predetermined number of candidate compositions are output by repeating the optimization of candidate compositions over a predetermined number of steps a predetermined number of times. In the modification, a configuration will be described in which multiple candidate compositions are output during a single optimization of candidate compositions over a predetermined number of steps. Below, a composition searching system 1000 according to the modification will be described, focusing on the differences from the embodiment.

[0145] <Processing Procedure> The composition search method according to the modified example will be described with reference to Fig. 8. Fig. 8 is a flowchart showing a second example of the composition search method.

[0146] The processes from step S11 to step S17 are the same as the processes from step S1 to step S7 of the composition searching method according to the embodiment (see FIG. 6 ). That is, the composition searching device 10 acquires search conditions from the terminal device 20, generates candidate compositions, predicts the physical properties of the target substance, calculates loss scores, and calculates energy.

[0147] In step S18, the energy calculation unit 160 of the composition searching apparatus 10 associates the candidate compositions generated in step S14 with the energies calculated in step S17 and stores them in the composition storage unit 170. That is, in this modification, the energy calculation unit 160 associates the energy with all candidate compositions and stores them in the composition storage unit 170 each time it calculates an energy.

[0148] In step S19, the composition searching device 10 determines whether a predetermined number of steps have been executed. If it is determined that the predetermined number of steps have been executed (YES), the composition searching device 10 proceeds to step S20. On the other hand, if it is determined that the predetermined number of steps have not been executed (NO), the composition searching device 10 returns to step S14.

[0149] Returning to step S14, the composition generator 120 generates a new candidate composition based on the current candidate composition, and counts up the number of steps.

[0150] Thereafter, the composition searching apparatus 10 repeats the process from step S15 to step S19 for the new candidate composition. In this manner, the composition searching apparatus 10 repeatedly executes a series of processes including generating candidate compositions, predicting physical properties, calculating loss scores, calculating energies, and storing candidate compositions and energies, until it determines in step S19 that a predetermined number of steps have been executed.

[0151] In step S20, the result output unit 180 of the composition search device 10 outputs the search results for the composition of the target substance. The result output unit 180 reads out the candidate compositions stored in the composition storage unit 170, starting with the candidate composition with the lowest energy, and includes them in the search results. The result output unit 180 transmits the search results for the composition of the target substance to the terminal device 20.

[0152] The terminal device 20 receives the search results for the composition of the target substance from the composition search device 10. The terminal device 20 presents the search results for the composition of the target substance to the user. The terminal device 20 may display candidate compositions included in the search results on the display device 506. The terminal device 20 may display the candidate compositions together with the energies corresponding to the candidate compositions.

[0153] Effect of the Modification The composition searching device 10 according to this modification repeatedly executes a series of processes including generating candidate compositions, predicting physical properties, calculating energies, and storing the candidate compositions and energies, and outputs a predetermined number of candidate compositions in ascending order of stored energy. In one aspect, this modification can output multiple compositions whose physical properties fall within a target range with a small amount of calculation.

[0154] [Supplementary Note] Each function of the above-described embodiments can be realized by one or more processing circuits. Here, the term "processing circuit" in this specification includes a processor programmed to execute each function by software, such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit) implemented by an electronic circuit, as well as devices such as an ASIC (Application Specific Integrated Circuit), DSP (Digital Signal Processor), FPGA (Field Programmable Gate Array), and conventional circuit modules designed to execute each of the above-described functions.

[0155] Although the embodiments of the present disclosure have been described in detail above, the embodiments disclosed herein are illustrative in all respects and are not limiting. The embodiments can be modified and improved in various ways without departing from the scope and spirit of the appended claims. The matters described in the above embodiments can be configured in other ways as long as they are not inconsistent, and can be combined as long as they are not inconsistent.

[0156] This application claims priority from Japanese Patent Application No. 2024-145435, filed on August 27, 2024, with the Japan Patent Office, the entire contents of which are incorporated herein by reference.

[0157] 10: Composition search device 20: Terminal device 110: Condition acquisition unit 120: Composition generation unit 130: Physical property prediction unit 140: Function determination unit 150: Loss calculation unit 160: Energy calculation unit 170: Composition storage unit 180: Result output unit 1000: Composition search system

Claims

1. A composition searching device comprising: a composition generation unit configured to generate candidate compositions of a target substance manufactured using two or more raw materials; a property prediction unit configured to predict the properties of the target substance based on the candidate compositions; an energy calculation unit configured to calculate the energy corresponding to the candidate compositions based on a loss function that minimizes the loss when the predicted value of the property is within a target range; and a result output unit configured to output a predetermined number of the candidate compositions with small energy.

2. A composition search device according to claim 1, wherein the loss function outputs zero when the predicted value is within the target range, and outputs the absolute value of the difference between the predicted value and the boundary value of the target range when the predicted value is outside the target range.

3. A composition search device according to claim 1 or 2, wherein the energy calculation unit is configured to calculate the energy based on the loss function and a second loss function that minimizes the loss when the candidate composition is within a constraint range.

4. A composition search device according to claim 3, wherein the second loss function outputs zero when the candidate composition is within the constraint range, and outputs the absolute value of the difference between the candidate composition and the boundary value of the constraint range when the candidate composition is outside the constraint range.

5. A composition searching device according to any one of claims 1 to 4, wherein the composition generating unit is configured to generate a new candidate composition by changing the blending amounts of two materials contained in the candidate composition for which the energy has been calculated.

6. A composition searching device as defined in claim 5, wherein the composition generating unit is configured to generate the new candidate composition by a generating method randomly selected from: a first generating method of increasing the blending amount of a first material substance and decreasing the blending amount of a second material substance; a second generating method of exchanging the blending amounts of the first material substance and the second material substance; a third generating method of averaging the largest blending amount and the smallest blending amount; and a fourth generating method of exchanging the smallest blending amount excluding zero with a blending amount that is zero.

7. A composition searching device according to any one of claims 1 to 6, wherein a series of processes including generating the candidate compositions, predicting the physical properties, and calculating the energies, and a process of storing the candidate composition having the smallest calculated energy among the series of processes, are repeated the predetermined number of times, and the result output unit is configured to output the stored candidate compositions.

8. A composition searching device according to any one of claims 1 to 6, which repeatedly executes a series of processes including generating the candidate compositions, predicting the physical properties, calculating the energy, and storing the candidate compositions and the energy, and the result output unit is configured to output the predetermined number of candidate compositions in ascending order of the stored energy.

9. A composition search method in which a computer executes the following steps: generating candidate compositions for a target substance manufactured using two or more raw materials; predicting physical properties of the target substance based on the candidate compositions; calculating the energy corresponding to the candidate compositions based on a loss function that minimizes the loss when the predicted value of the physical property is within a target range; and outputting a predetermined number of the candidate compositions with small energy.

10. A program for causing a computer to execute the following steps: generating candidate compositions for a target substance to be manufactured using two or more raw materials; predicting the physical properties of the target substance based on the candidate compositions; calculating the energy corresponding to the candidate compositions based on a loss function that minimizes the loss when the predicted value of the physical properties is within a target range; and outputting a predetermined number of the candidate compositions with small energy.

Images