JP2024011665A - Design support device, design support method, and design support program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently optimize a characteristic of a product or the like and a design variable constituting an objective variable in a design process of the product or the like.

SOLUTION: A design support device 10 acquires a recommendation design parameter group according to optimization of a target directivity acquisition function constructed by applying prediction models for predicting observation values of characteristic items, acquires the n-th observation value probability distribution by inputting the n-th (n is an integer 1 to (K-1)) recommendation design parameter group, acquires the n-th conditional expectation value on the basis of the n-th observation value probability distribution, reconstructs each prediction model according to the first to n-th prediction actual data composed from the first to n-th recommendation design parameter group and the conditional expectation value, reconstructs the target directivity acquisition function on the basis of the prediction model, and acquires the (n+1)-th recommendation design parameter group according to optimization of the target directivity acquisition function, in order to output the first to K-th recommendation design parameter group acquired sequentially.

SELECTED DRAWING: Figure 2

COPYRIGHT: (C)2024,JPO&INPIT

Description

One aspect of the present disclosure relates to a design support device, a design support method, and a design support program.

Product design methods using machine learning are being researched. As a field of product design, for example, in the design of functional materials, the characteristics of materials are estimated by machine learning using learning data consisting of pairs of raw material blend ratios and characteristics for experimental and manufactured materials. A model has been constructed to predict properties for raw material mixture ratios that have not yet been tested. By planning experiments based on predictions of such properties, it becomes possible to efficiently optimize parameters such as material properties and raw material blending ratios, thereby improving development efficiency. Furthermore, Bayesian optimization is known to be effective as such an optimization method, and design devices that output design values using Bayesian optimization are known (for example, see Patent Document 1). ).

JP2020-52737A

On the other hand, in the development of products such as materials, multiple objective variables (characteristics) are given and optimization of multiple objective variables is performed in order to improve the multiple properties that change according to the design variables. . This is called multi-objective optimization, and for example, a method such as multi-objective Bayesian optimization is applied. Particularly in the field of product design, goal-oriented multi-objective Bayesian optimization aimed at achieving target values for each objective variable is useful. In this optimization process, we repeat the process of conducting experiments using the recommended design variables, evaluating the results of the experiments, and feeding back the evaluation results to provide explanations for achieving the target values of each objective variable. Optimization of variables and objective variables is achieved. In such optimization, recommendations for design variables for one experiment are obtained, and unless experiments and evaluations based on the recommendations are performed, recommendations for the next experiment cannot be obtained. In the optimization process, if it is necessary to repeat the process of obtaining recommendations, experimenting, and evaluating many times until the objective variable reaches the target value, it is necessary to perform experiments and evaluation each time a recommendation is obtained. However, it was inefficient. Such problems are not limited to material design, but are common to product design in general.

The present invention has been made in view of the above problems, and is aimed at optimizing the characteristics and design variables of products, etc., which constitute objective variables in the manufacturing process of products, work-in-progress, semi-finished products, parts, or prototypes. , the purpose is to realize it efficiently.

The design support device according to the first aspect of the present disclosure determines design parameters in designing a product, work in process, semi-finished product, part, or prototype manufactured based on a design parameter group consisting of a plurality of design parameters. In order to apply the method of optimizing design parameters by repeating the production of products, work-in-progress, semi-finished products, parts, or prototypes based on the determined design parameters, products, work-in-process, semi-finished products, parts or A design support device that calculates a plurality of design parameter groups that satisfy the target values set for each of a plurality of characteristic items that indicate the characteristics of a prototype, and is used for processing manufactured products, work-in-progress, semi-finished products, parts, or A data acquisition unit that acquires a plurality of performance data consisting of a group of design parameters and observed values of each of a plurality of characteristic items regarding the prototype; A model construction unit that constructs a prediction model to predict as an alternative indicator based on actual data, and a unit that receives design parameter groups as input and outputs index values of design parameter groups related to improvement of characteristics shown in all characteristic items. an acquisition function construction unit that constructs a goal-oriented acquisition function that is an acquisition function of a design parameter group acquisition unit that acquires, as a first recommended design parameter group, a recommended design parameter group obtained by optimizing a goal-oriented acquisition function constructed using a prediction model constructed based on the data; an output unit that outputs the recommended design parameter group acquired by the parameter group acquisition unit; and an output unit that outputs the recommended design parameter group acquired by the parameter group acquisition unit; Obtaining the observed value probability distribution in which the probability distribution of the observed values of each characteristic item obtained by inputting the recommended design parameter group (a given integer of 2 or more)) into the prediction model is obtained as the nth observed value probability distribution. and a conditional expected value acquisition unit that acquires, as the n-th conditional expected value, a conditional expected value given a given condition regarding the observed value based on the n-th observed value probability distribution. In preparation, the model construction unit rebuilds a prediction model for each characteristic item based on the actual data and the first to nth predicted actual data, and The predicted performance data consists of the l-th recommended design parameter group and the l-th conditional expected value of each characteristic item, and the acquisition function construction unit generates a goal-oriented model based on the prediction model reconstructed by the model construction unit. The acquisition function is reconstructed, and the design parameter group acquisition unit acquires the recommended design parameter group obtained by optimizing the goal-oriented acquisition function reconstructed by the acquisition function construction unit as the (n+1)th recommended design parameter group. The output unit performs (K-1) times of obtaining the observed value probability distribution, obtaining the conditional expected value, reconstructing the prediction model, reconstructing the goal-oriented acquisition function, and obtaining the recommended design parameter group. Through repetition, the first to Kth recommended design parameter groups sequentially acquired by the design parameter group acquisition unit are output.

The design support method according to the first aspect of the present disclosure is used to determine design parameters in the design of a product, work-in-progress, semi-finished product, part, or prototype manufactured based on a design parameter group consisting of a plurality of design parameters. In order to apply the method of optimizing design parameters by repeating the production of products, work-in-progress, semi-finished products, parts, or prototypes based on the determined design parameters, products, work-in-process, semi-finished products, parts or A design support method in a design support device that calculates a plurality of design parameter groups that satisfy set target values for each of a plurality of characteristic items that indicate the characteristics of a prototype. A data acquisition step of acquiring a plurality of performance data consisting of a group of design parameters and observed values of each of a plurality of characteristic items regarding the product, part, or prototype, and a probability distribution of observed values of the characteristic items based on the group of design parameters. or a model construction step in which a predictive model is constructed based on actual data to predict as an approximation or alternative indicator thereof, and a model construction step that takes the design parameter group as input and outputs the index value of the design parameter group regarding the improvement of characteristics shown in all characteristic items. an acquisition function construction step for constructing a goal-oriented acquisition function that is a single acquisition function, and a design parameter group acquisition step for acquiring a design parameter group obtained by optimizing the goal-oriented acquisition function as a recommended design parameter group. design parameter group acquisition, in which a recommended design parameter group obtained by optimizing a goal-oriented acquisition function constructed using a predictive model constructed based on actual data is obtained as a first recommended design parameter group; Step and predict the n-th (n=1, 2, ..., K-1, (K is a given integer greater than or equal to 2)) recommended design parameter group obtained by optimizing the goal-oriented acquisition function. An observed value probability distribution acquisition step of obtaining the probability distribution of observed values of each characteristic item obtained by inputting it into the model as an nth observed value probability distribution; a conditional expected value obtaining step of obtaining a conditional expected value given a given condition regarding the conditional expected value as an n-th conditional expected value, and based on the actual data and the first to nth predicted actual performance data, In the model reconstruction step of reconstructing a prediction model for each characteristic item, the lth (l=1, 2, ..., n) predicted performance data is the lth recommended design parameter group and each characteristic item. an acquisition function reconstruction step for reconstructing a goal-oriented acquisition function based on the predictive model reconstructed in the model reconstruction step; a design parameter group reacquisition step of acquiring a recommended design parameter group obtained by optimizing the goal-oriented acquisition function reconstructed in the reconstruction step as an (n+1)th recommended design parameter group; and an observed value probability distribution acquisition step. , a repetition step of repeating the conditional expected value acquisition step, the model reconstruction step, the acquisition function reconstruction step, and the design parameter group reacquisition step (K-1) times, and the design parameter group acquisition step and the repetition step sequentially. and an output step of outputting the obtained first to Kth recommended design parameter groups.

The design support program according to the first aspect of the present disclosure uses a computer to design a product, a work in progress, a semi-finished product, a part, or a prototype to be manufactured based on a design parameter group consisting of a plurality of design parameters. Products, work-in-progress, semi-finished products are applied to the method of optimizing design parameters by repeatedly determining parameters and producing products, work-in-progress, semi-finished products, parts, or prototypes based on the determined design parameters. , a design support program for functioning as a design support device for determining a plurality of design parameter groups that satisfy the target values set for each of a plurality of characteristic items indicating the characteristics of a part or a prototype, which has already been produced. A data acquisition function that acquires multiple performance data consisting of a group of design parameters and observed values of each of multiple characteristic items regarding products, work in progress, semi-finished products, parts, or prototypes, and a data acquisition function that acquires multiple pieces of performance data consisting of a group of design parameters and observed values of each of multiple characteristic items, and A model construction function that builds a prediction model based on actual data to predict the observed value of an item as a probability distribution, its approximation, or an alternative indicator, and a design that uses design parameter groups as input to improve the characteristics shown in all characteristic items. An acquisition function construction function that constructs a goal-oriented acquisition function, which is a single acquisition function whose output is the index value of a parameter group, and a design parameter group obtained by optimizing the goal-oriented acquisition function are acquired as a recommended design parameter group. A design parameter group acquisition function that uses, as a first recommended design parameter group, a recommended design parameter group obtained by optimizing a goal-oriented acquisition function constructed using a prediction model constructed based on actual data. a design parameter group acquisition function, an output function that outputs a recommended design parameter group acquired by the design parameter group acquisition function, and an n-th (n=1, 2, ..., K-1, (K is a given integer greater than or equal to 2)) are input into the prediction model. Based on the observed value probability distribution acquisition function that obtains the observed value probability distribution as a value probability distribution, and the nth observed value probability distribution, the conditional expected value given the given condition regarding the observed value is set as the nth conditional expected value. The model construction function reconstructs the prediction model for each characteristic item based on the actual data and the first to nth predicted actual data, and = 1, 2, ..., n) consists of the l-th recommended design parameter group and the l-th conditional expected value of each characteristic item, and the acquisition function construction function uses the model construction function to Based on the reconstructed prediction model, the goal-oriented acquisition function is reconstructed, and the design parameter group acquisition function obtains the recommended design parameter group obtained by optimizing the goal-oriented acquisition function reconstructed by the acquisition function construction function. , as the (n+1)th recommended design parameter group, and the output functions are acquisition of observed value probability distribution, acquisition of conditional expected value, reconstruction of prediction model, reconstruction of goal-oriented acquisition function, and recommended design. By repeating the parameter group acquisition (K-1) times, the first to Kth recommended design parameter groups sequentially acquired by the design parameter group acquisition unit are output.

According to this aspect, a predictive model and a goal-oriented acquisition function are constructed based on performance data, and a recommended design parameter group can be obtained by optimizing the goal-oriented acquisition function. By inputting the n-th recommended design parameter group into the prediction model, the probability distribution of the observed values of each characteristic item can be obtained. It is possible to predict the distribution of characteristics of products etc. produced based on Therefore, the conditional expected value of the distribution of the characteristics of the product, etc. can be regarded as the observed value of the characteristics of the product, etc. based on the recommended design parameter group. Then, a prediction model and a goal-oriented acquisition function are constructed by using the predicted performance data consisting of pairs of recommended design parameters and conditional expected values together with the actual data, and the (n+1th ), it becomes possible to obtain a set of recommended design parameters. Therefore, by repeating the process from obtaining a group of recommended design parameters to optimizing a goal-oriented acquisition function a desired number of times, a desired number of groups of recommended design parameters can be obtained without fabricating products or evaluating their characteristics. becomes possible.

In the design support apparatus according to the second aspect, in the design support apparatus according to the first aspect, the condition in the conditional expected value may be that at least one characteristic item does not satisfy the target value.

According to this aspect, the expected value based on the distribution of characteristics that do not satisfy the target value among the predicted characteristics distribution of a product etc. manufactured based on the nth recommended design parameter group is Since it is regarded as the observed value of the product, etc., the (n+1)th design parameter group is calculated based on the assumption that the product, etc. manufactured based on the nth design parameter group has not achieved the target in the characteristic item. You can get recommendations.

In the design support device according to the third aspect, in the design support device according to the first or second aspect, the goal-oriented acquisition function is a probability that the target values of all characteristic items are achieved, and is based on the prediction model. It is also possible to include at least a goal achievement probability term representing the overall achievement probability, which is a probability calculated using the design parameter group as a variable.

According to this aspect, the overall achievement probability is reflected in the index value output from the goal-oriented acquisition function, so optimization using the index value output from the goal-oriented acquisition function as the objective variable It is possible to obtain a group of design parameters that have a high possibility of achieving the goals related to the items.

In the design support apparatus according to the fourth aspect, in the design support apparatus according to the third aspect, the overall achievement probability is the sum of the achievement probabilities for the target value of each characteristic item, and The probability may be based on a probability distribution of observed values obtained by inputting a group of design parameters into a prediction model for each characteristic item.

According to this aspect, since the prediction model is configured to output a probability distribution of observed values, it is possible to obtain the probability of achieving the target value of each characteristic item according to the design parameter group. The overall achievement probability calculated by the sum of the achievement probabilities of the target values of each characteristic item is included in the goal achievement probability term of the goal-oriented acquisition function, so the overall achievement probability is included in the index value from the goal-oriented acquisition function. Appropriately reflected.

In the design support apparatus according to the fifth aspect, in the design support apparatus according to the fourth aspect, the goal achievement probability term may be composed of the overall achievement probability or the logarithm of the overall achievement probability.

According to this aspect, since the goal achievement probability term is composed of the overall achievement probability or the logarithm of the overall achievement probability, the overall achievement probability is appropriately reflected in the index value from the goal-oriented acquisition function.

In the design support apparatus according to the sixth aspect, in the design support apparatus according to any one of the first to fifth aspects, j pieces (j is 1 or more) of the first to K recommended design parameter groups. a designation reception unit that accepts a designation of not adopting the recommended design parameter group (an integer of , the (K+1)th to (K+j)th recommended design parameters sequentially obtained by the design parameter group obtaining unit by further repeating the reconstruction of the goal-oriented acquisition function and the obtaining of the recommended design parameter group j times. The group may be further output.

According to this aspect, for j design parameter groups that are clearly not suitable for manufacturing a product, etc., among the obtained first to Kth recommended design parameter groups, for example, the technology of the product, etc. We can accept specifications from experts in the field. Depending on the acceptance of the specification, the process from obtaining the observed value probability distribution to optimizing the goal-oriented acquisition function is repeated j times, thereby imparting proficiency regarding the production and characteristics of products, etc. to the prediction model and goal-oriented acquisition function. It is possible to obtain a desired number of recommended design parameter groups while reflecting the know-how of those involved.

In the design support apparatus according to the seventh aspect, in the design support apparatus according to any one of the first to sixth aspects, the prediction model is a regression model whose input is a group of design parameters and whose output is a probability distribution of observed values. The predictive model may be a model or a classification model, and the model construction unit may construct a predictive model by machine learning using at least performance data.

According to this aspect, since the prediction model is constructed as a predetermined regression model or classification model, it is possible to obtain a prediction model that can obtain a probability distribution of observed values of a characteristic item, an approximation thereof, or an alternative index.

In the design support device according to the eighth aspect, in the design support device according to the seventh aspect, the prediction model includes a posterior distribution of predicted values based on Bayesian theory, a distribution of predicted values of predictors forming an ensemble, and a regression model. Probability distribution of observed values, its approximation, or alternative index using any one of theoretical formulas for prediction intervals and confidence intervals, Monte Carlo dropout, and prediction distributions of multiple predictors constructed under different conditions. It may also be a machine learning model that predicts.

According to this aspect, a prediction model capable of predicting a probability distribution of observed values of a characteristic item based on a group of design parameters, an approximation thereof, or an alternative index is constructed.

According to one aspect of the present disclosure, it is possible to efficiently optimize the characteristics and design variables of products, etc. that constitute objective variables in the manufacturing process of products, work-in-progress, semi-finished products, parts, or prototypes. becomes.

1 is a diagram showing an overview of a material design process to which a design support device according to an embodiment is applied; FIG. 1 is a block diagram showing an example of a functional configuration of a design support device according to an embodiment. FIG. FIG. 1 is a hardware block diagram of a design support device according to an embodiment. FIG. 3 is a diagram showing an example of a group of design parameters related to manufactured materials. It is a figure which shows the example of the observed value regarding the produced material. 3 is a flowchart showing a process of optimizing property items and design parameters in material design. 3 is a flowchart illustrating an example of the content of a design support method in a design support apparatus according to an embodiment. FIG. 3 is a diagram schematically showing the flow of data acquired at each stage of design support processing. FIG. 3 is a diagram for explaining a conditional expected value acquisition process. FIG. 2 is a diagram showing the configuration of a design support program.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In addition, in the description of the drawings, the same or equivalent elements are given the same reference numerals, and redundant description will be omitted.

FIG. 1 is a diagram illustrating an overview of a material design process, which is an example of a design process for a product, work-in-progress, semi-finished product, part, or prototype to which a design support apparatus according to an embodiment is applied. In the following, "products, work-in-progress, semi-finished products, parts, or prototypes" will be referred to as "products, etc." The design support apparatus 10 of this embodiment can be applied to the process of designing any product, etc., which has a plurality of characteristic items indicating the characteristics of the product, etc., and a target value for each characteristic item. The design support device 10 repeatedly determines design parameters (design variables) and characteristics (objective variables) of products, etc. by repeatedly determining design parameters and producing products, work-in-progress, semi-finished products, parts, or prototypes based on the determined design parameters. ) can be applied to the optimization method. Specifically, in addition to the development and design of materials, the design support device 10 can be applied to, for example, the design of products such as automobiles and drugs, and the optimization of the molecular structure of drugs. In this embodiment, as described above, the design support processing by the design support apparatus 10 will be explained using an example of material design as an example of the design of a product or the like.

As shown in FIG. 1, the design support processing by the design support apparatus 10 is applied to material production and experiments in a plant, a laboratory A, etc., as an example. That is, according to the set design parameter group x, a material is produced in the plant, laboratory A, etc., and based on the produced material, observed values y of a plurality of characteristic items indicating the characteristics of the material are obtained. Note that material preparation and experiments in the plant and laboratory A may be simulations. In this case, the design support device 10 provides the design parameter group x for executing the next simulation.

The design support device 10 optimizes a plurality of characteristic items and design parameters based on performance data consisting of a design parameter group x and observed values y of a plurality of characteristic items of a material manufactured based on the design parameter group x. conduct. Specifically, the design support device 10 determines design parameters that are likely to yield more suitable characteristics for the next fabrication and experiment, based on the design parameter group x and observed value y regarding the material that has already been fabricated. Output the group x.

For example, the design support apparatus 10 of this embodiment is applied for the purpose of tuning a plurality of design variables to achieve a plurality of target characteristics in the design of a material product. As an example of the design of a material product, when a certain material is manufactured by mixing a plurality of polymers and additives, the design support device 10 uses a group of design parameters such as the blending amount of each polymer and additive as design variables, and calculates the characteristics. The observed values of the elastic modulus and thermal expansion coefficient are used as objective variables to tune a group of design parameters to achieve the target values of multiple characteristic items.

FIG. 2 is a block diagram showing an example of the functional configuration of the design support apparatus according to the embodiment. The design support device 10 of the present embodiment is configured to satisfy target values set for each of a plurality of property items indicating the characteristics of the material in the design of a material produced based on a design parameter group consisting of a plurality of design parameters. This is a device that calculates multiple design parameters. As shown in FIG. 2, the design support apparatus 10 may include a functional unit configured in the processor 101, a design parameter storage unit 21, and an observed value storage unit 22. Each functional unit will be described later.

FIG. 3 is a diagram showing an example of the hardware configuration of the computer 100 that constitutes the design support apparatus 10 according to the embodiment. Note that the computer 100 can constitute the design support apparatus 10.

As an example, the computer 100 includes a processor 101, a main storage device 102, an auxiliary storage device 103, and a communication control device 104 as hardware components. The computer 100 constituting the design support apparatus 10 may further include an input device 105 such as a keyboard, touch panel, or mouse, and an output device 106 such as a display.

Processor 101 is a computing device that executes an operating system and application programs. Examples of processors include a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit), but the type of processor 101 is not limited to these. For example, processor 101 may be a combination of sensors and dedicated circuitry. The dedicated circuit may be a programmable circuit such as an FPGA (Field-Programmable Gate Array), or may be another type of circuit.

The main storage device 102 is a device that stores programs for realizing the design support device 10 and the like, calculation results output from the processor 101, and the like. The main storage device 102 includes, for example, at least one of ROM (Read Only Memory) and RAM (Random Access Memory).

The auxiliary storage device 103 is generally a device that can store a larger amount of data than the main storage device 102. The auxiliary storage device 103 is configured by a nonvolatile storage medium such as a hard disk or flash memory. The auxiliary storage device 103 stores a design support program P1 and various data for causing the computer 100 to function as the design support device 10 or the like.

The communication control device 104 is a device that performs data communication with other computers via a communication network. The communication control device 104 is configured by, for example, a network card or a wireless communication module.

Each functional element of the design support device 10 is realized by loading a corresponding program P1 onto the processor 101 or the main storage device 102 and causing the processor 101 to execute the program. The program P1 includes codes for implementing each functional element of the corresponding server. The processor 101 operates the communication control device 104 according to the program P1 to read and write data in the main storage device 102 or the auxiliary storage device 103. Through such processing, each functional element of the corresponding server is realized.

The program P1 may be provided after being permanently recorded on a tangible recording medium such as a CD-ROM, a DVD-ROM, or a semiconductor memory. Alternatively, at least one of these programs may be provided via a communication network as a data signal superimposed on a carrier wave.

Referring again to FIG. 2, the design support apparatus 10 includes a data acquisition section 11, a model construction section 12, an acquisition function construction section 13, a design parameter group acquisition section 14, an output section 15, an observed value probability distribution acquisition section 16, and a condition It includes an expected value acquisition section 17 and a specification reception section 18. The design parameter storage unit 21 and the observation value storage unit 22 may be configured in the design support device 10, as shown in FIG. 2, or may be configured as another device that can be accessed from the design support device 10. .

The data acquisition unit 11 acquires a plurality of performance data regarding manufactured materials. The performance data consists of pairs of design parameters and observed values of each of a plurality of characteristic items. The design parameter storage unit 21 is a storage unit that stores a group of design parameters in actual performance data, and may be configured, for example, in the main storage device 102, the auxiliary storage device 103, or the like. The observed value storage unit 22 is a storage means that stores observed values in actual performance data.

FIG. 4 is a diagram showing an example of a design parameter group stored in the design parameter storage unit 21. As shown in FIG. 4, the design parameter storage unit 21 stores a group of design parameters x _t for material production from the first time (t=1) to the T-th time (t=T). The design parameter group x includes, for example, P design parameters p (p=1 to P). The design parameter group x may include, for example, the blending amount of the raw material A, the blending amount of the raw material B, and the design parameter P, and may constitute vector data with the number of dimensions P corresponding to the number of design parameters. In addition to the exemplified design parameters, the design parameters may be, for example, non-vector data such as molecular structures and images. Furthermore, when dealing with the problem of selecting an optimal molecule from a plurality of molecule types, the design parameters may be data indicating choices among the plurality of molecules.

FIG. 5 is a diagram showing an example of observed values y stored in the observed value storage unit 22. As shown in FIG. 5, the observed value storage unit 22 stores a plurality of characteristic items m (m= The observed values y _{m, t} of 1 to M) are stored. The property items m may include, for example, glass transition temperature, adhesive strength, and property items M. Further, a target value y _{m (target)} is set for each characteristic item m. A pair of the design parameter group x _t and the observed value y _m,t constitutes actual data.

The design support device 10 obtains a recommended design parameter group X ₁ for the T+1st material fabrication based on the performance data from the first (t=1) to the Tth (t=T) material fabrication (in addition, A group of design parameters in actual data is written with a lowercase letter x, and a group of recommended design parameters is written with a capital letter X). The recommended _{design parameter group )} is a parameter group that approaches

The model construction unit 12 constructs a prediction model based on actual data. The prediction model is a model that predicts the observed value y _m of the characteristic item m as a probability distribution or its approximation or alternative index based on the design parameter group x. The model constituting the prediction model may be any model that can predict the observed value y _m as a probability distribution, its approximation, or an alternative index, and its type is not limited. A prediction model that predicts the observed value y _m as an alternative indicator of the probability distribution is, for example, the distribution of the predicted values of the predictors that make up the ensemble (random forest), the distribution obtained by Monte Carlo dropout (neural network), or the distribution under different conditions. The probability distribution of observed values is predicted using the prediction distribution of multiple predictors (any machine learning method) as an alternative indicator.

For example, the prediction model may be a regression model that takes the design parameter x as input and outputs the probability distribution of observed values y _m . When the prediction model is a regression model, the prediction model may be configured by any one of regression models such as Gaussian process regression, random forest, and neural network, for example. The model construction unit 12 may construct a predictive model using a well-known machine learning method using track record data. The model construction unit 12 may construct a prediction model using a machine learning method that applies performance data to a prediction model and updates parameters of the prediction model.

In addition, the prediction model is based on the posterior distribution of the predicted value based on Bayesian theory, the distribution of the predicted value of the predictors forming the ensemble, the theoretical formula for the prediction interval and confidence interval of the regression model, Monte Carlo dropout, and multiple models under different conditions. It may be a machine learning model that predicts the probability distribution of observed values, its approximation, or alternative index using any one of the prediction distributions of individually constructed predictors. Predictions of probability distributions of observed values or alternative indicators thereof can be obtained by model-specific techniques. The probability distribution of observed values or its approximation or substitute index is based on the posterior distribution of predicted values in the case of Gaussian process regression and Bayesian neural networks, or based on the distribution of predictions of the predictors forming the ensemble in the case of random forests. In the case of linear regression, it can be obtained based on the prediction interval and confidence interval, and in the case of neural network, it can be obtained based on Monte Carlo dropout. However, the method for calculating the distribution of observed values for each machine learning model or its substitute index is not limited to the above method.

Also, any model may be extended to a model that can predict the probability distribution of observed values or an alternative indicator thereof. For example, a model that uses the distribution of predicted values of each model, obtained by constructing multiple datasets using the bootstrap method etc. and constructing a predictive model for each, as a substitute indicator for the probability distribution of observed values. is given as an example. However, the method of extending the machine learning model to a model that can predict the probability distribution of observed values or its substitute index is not limited to the above method.

Further, the prediction model may be constructed using linear regression, PLS regression, Gaussian process regression, bagging ensemble learning such as random forest, boosting ensemble learning such as gradient boosting, support vector machine, neural network, etc.

In a prediction model constructed as a Gaussian process regression, by inputting into the model a group of design parameters x in actual data that constitute explanatory variables of training data, observed values y that constitute objective variables, and design parameters x to be predicted, The probability distribution of the observed values is predicted.

Further, the model construction unit 12 may tune the hyperparameters of the prediction model using a well-known hyperparameter tuning method. That is, the model construction unit 12 generates a hyperlink of a prediction model constructed by Gaussian process regression using a vector representing a design parameter group x, which is an explanatory variable in actual data, and maximum likelihood estimation using an observed value y, which is an objective variable. Parameters may be updated.

Further, the prediction model may be constructed using a classification model. When the predictive model is a classification model, the model construction unit 12 can construct the predictive model using a machine learning method that can evaluate a well-known probability distribution using actual data.

In this way, by the model construction unit 12 constructing a prediction model using a predetermined regression model or classification model, it becomes possible to obtain a probability distribution of observed values of characteristic items based on an arbitrary design parameter group x.

In addition, the prediction model is a single-task model that predicts the observed value of one characteristic item as a probability distribution, its approximation, or an alternative indicator, or a single-task model that predicts the observed value of multiple characteristic items as a probability distribution, its approximation, or alternative indicator. It may also be a multitasking model. In this way, by constructing a prediction model using a multi-task model or a single-task model that is appropriately configured according to the nature of the characteristic item, it is possible to improve the accuracy of prediction of observed values by the prediction model.

The acquisition function construction unit 13 constructs a goal-oriented acquisition function, which is a single acquisition function, which receives the design parameter group as input and outputs the index value of the design parameter group regarding improvement of characteristics shown in all characteristic items. The configuration of the goal-oriented acquisition function is not limited as long as it is a function that outputs index values for optimization of design parameter groups and characteristic items.

The goal orientation acquisition function in this embodiment may include at least a goal achievement probability term. The goal achievement probability term represents the overall achievement probability, which is the probability that the goal values of all characteristic items will be achieved. The overall achievement probability is calculated based on the prediction model using the design parameter group as variables.

Specifically, the acquisition function construction unit 13 constructs a goal-oriented acquisition function A(x) as shown in equation (1) below.
A(x)=g(P(x))...(1)
In equation (1), g(P(x)) is a goal achievement probability term. That is, the goal orientation acquisition function A(x) includes at least the goal achievement probability term g(P(x)). Note that the goal orientation acquisition function in equation (1) is just an example, and the goal orientation acquisition function may include terms other than the goal achievement probability term.

The goal achievement probability term includes the overall achievement probability P(x). For example, the overall achievement probability P(x) may be defined as shown in equation (2) below, assuming that the goal achievement events of each characteristic item are independent of each other.
P(x)=Π _1<=m<=M Pm(x) ...(2)
That is, the overall achievement probability P(x) is the sum of the achievement probabilities Pm(x) of each characteristic item m (m=1 to M). Since the prediction model can predict the probability distribution of the observed values of the characteristic items based on the design parameter group x, the achievement probability Pm(x) of each characteristic item can be calculated using the design parameter group x using the prediction model of each characteristic item. It can be expressed as a function with input variables. Alternatively, the overall achievement probability P(x) may be expressed as a function using a design parameter group x based on a prediction model of all characteristic items as an input variable, without calculating the achievement probability of each characteristic item. .

The goal achievement probability term g(P(x)) includes the overall achievement probability P(x). For example, the goal achievement probability term g(P(x)) may consist of the overall achievement probability P(x) as shown in equation (3), or may consist of the overall achievement probability P(x) as shown in equation (4). It may be made up of the logarithm of the achievement probability P(x).
g(P(x))=P(x)...(3)
g(P(x))=log(P(x))...(4)
Further, the goal achievement probability term may be a term in which the overall achievement probability P(x) or the logarithm of the overall achievement probability P(x) is further multiplied by a coefficient, or may be a term in which another factor is added. may include terms. For an example of a goal-oriented acquisition function constructed in this way, its optimization can be treated as a maximization problem.

In this example of the present embodiment, since the prediction model is configured to output the probability distribution of observed values, the probability of achieving the target value Pm(x) for each characteristic item m according to the design parameter group is obtained. be able to. Then, since the overall achievement probability calculated by the sum of the achievement probability Pm(x) of the target value of each characteristic item m is included in the goal achievement probability term of the goal-oriented acquisition function, the index value from the goal-oriented acquisition function The overall achievement probability is appropriately reflected.

The design parameter group acquisition unit 14 acquires a design parameter group obtained by optimizing the goal orientation acquisition function. Specifically, as an example, the design parameter group acquisition unit 14 may acquire at least one design parameter group that optimizes (for example, maximizes) the output of the goal-oriented acquisition function. Specifically, the design parameter group acquisition unit 14 performs optimization using the index value output from the goal-oriented acquisition function A(x) constructed by the acquisition function construction unit 13 as the objective variable, and obtains the optimal solution. Obtain a set of design parameters x.

The output unit 15 outputs the design parameter group acquired by the design parameter group acquisition unit 14. That is, the output unit 15 converts the design parameter group obtained based on the performance data in the first (t=1) to T-th (t=T) material fabrication into the design parameter group for the (T+1)-th material fabrication. It can be output as a design parameter group x.

Although the manner of output is not limited, the output unit 15 outputs the design parameter group candidates by displaying them on a predetermined display device or storing them in a predetermined storage means, for example.

FIG. 6 is a flowchart showing the process of optimizing property items and design parameter groups in material design.

In step S1, a set of design parameters is obtained. The design parameter group acquired here is for initial material production (experiment), and may be a design parameter group set arbitrarily, or set based on experiments that have already been conducted. It may also be a set of design parameters.

In step S2, material preparation is performed. In step S3, observed values of characteristic items of the fabricated material are acquired. A pair of the design parameter group as the manufacturing condition in step S2 and the observed value of each characteristic item acquired in step S3 constitutes performance data.

In step S4, it is determined whether a predetermined termination condition is satisfied. The predetermined termination condition is a condition for optimizing the design parameter group and the observed values of the characteristic items, and may be set arbitrarily. The termination conditions for optimization may be, for example, reaching a predetermined number of times of fabrication (experiment) and obtaining observed values, reaching a target value of observed values, and convergence of optimization. If it is determined that the predetermined termination condition is satisfied, the optimization process is terminated. If it is not determined that the predetermined termination condition is satisfied, the process proceeds to step S5.

In step S5, design support processing is performed by the design support device 10. The design support process is a process that outputs a group of design parameters for the next material production. The process then returns to step S1 again.

Note that in the first cycle of the processing cycle composed of steps S1 to S5, if a plurality of pairs of observed values of design parameter groups and characteristic items are obtained as initial data, the processing of steps S1 to S4 is omitted. . If initial data cannot be obtained, a group of design parameters obtained by an arbitrary method such as experimental design or random search is obtained in step S1. In the second and subsequent processing cycles, in step S1, the design parameter group output in step S5 is acquired.

Next, design support processing by the design support apparatus 10 in this embodiment will be described in detail with reference to FIG. 1 and FIGS. 7 to 9. FIG. 7 is a flowchart showing an example of the content of the design support method in the design support apparatus according to the embodiment, and shows the process of step S5 in FIG. 6. The design support method is executed by loading the design support program P1 into the processor 101 and executing the program to realize each of the functional units 11 to 18. FIG. 8 is a diagram schematically showing the flow of data acquired at each stage of the design support process. FIG. 9 is a diagram for explaining the conditional expected value acquisition process.

The design support apparatus 10 of this embodiment obtains a plurality of design parameter groups based on performance data. Specifically, as an example, _the design support device _{10 calculates} ( K sets (K is a given integer of 2 or more) of recommended design parameter groups X ₁ , X ₂ , . _. . Note that in the optimization process to which the design support apparatus 10 of this embodiment is applied, it is not excluded that K becomes 1.

In step S11, the data acquisition unit ₁₁ acquires a track record consisting of a pair of a design parameter group x _t (t is an integer from 1 to T) and an observed value y _m,t of each characteristic item, as shown by the symbol v 11. Data d _t (performance data d ₁ , d ₂ , ..., d _T ) is acquired.

In step S12, the model construction unit 12 constructs a prediction model for each characteristic item m based on the performance data _dt .

In step S13, the acquisition function construction unit 13 constructs a goal-oriented acquisition function based on the prediction model constructed in step S12.

In step S14, the design parameter group acquisition unit 14 acquires a recommended design parameter group by optimizing the goal-oriented acquisition function constructed in step S13. Here, the design parameter group acquisition unit 14 obtains a recommended design obtained by optimizing a goal-oriented acquisition function constructed using a prediction model constructed based on actual data _dt , as shown by code pr11. The parameter group is obtained as a first recommended design parameter group _X1 , as indicated by the symbol _v12 .

In step S15, a variable n is set to 1. Then, in step S16, the observed value probability distribution acquisition unit 16 selects each of the nth (n=1, 2,..., K-1) recommended design parameter groups acquired by the optimization of the goal orientation acquisition function. The probability distribution of the observed values of each characteristic item m obtained by inputting it into the prediction model of the characteristic item m is obtained as the n-th observed value probability distribution f _n .

Here, since the variable n is 1, the observed value probability distribution acquisition unit ₁₆ inputs the first recommended design parameter group , obtain the first observed value probability distribution _f1 . The first observed value probability distribution f ₁ is a simultaneous distribution of all the characteristic items m, and represents the prediction of the results of material production, experiments, etc. performed based on the first recommended design parameter group X ₁ . Further, the first observed value probability distribution _f1 may be composed of a probability distribution of each of the characteristic items m. That is, the first observed value probability distribution f ₁ can be considered as a pseudo experimental result based on the first recommended design parameter group X ₁ .

In step S17, the conditional expected value acquisition unit 17 acquires the conditional expected value given the given condition regarding the observed value as the nth conditional expected value based on the nth observed value probability distribution. do. Here, the n-th conditional expected value will be explained with further reference to FIG. In FIG. 9, for convenience of explanation, the n-th observed value probability distribution f _n is schematically shown assuming that there are two characteristic items, characteristic items A and B. Further, it is assumed that the probability density function corresponding to the n-th observation value probability distribution f _n is 0 outside the region Y _n and a finite value of 0 or more inside the region Y n. In other words, the region Y _n is a schematic visualization of the n-th observed value probability distribution f _n , and includes a random variable y A that is the observed value of characteristic item _A and a random variable that is the observed value of characteristic item B. The probability distribution of the combination with y _B is shown.

Here, if the target values of the random variables y _A and y _B are respectively less than or equal to y _{A (TARGET)} and less than y _{B (TARGET)} , then the random variables y _A and y _B are The area Y _n corresponding to the probability distribution of combinations of can be divided into an area Y _{n (OK)} corresponding to observed values that have achieved the target and an area Y _{n (NG)} corresponding to observed values that have not reached the target. can. That is, the region Y _{n (NG)} is a region in which at least one characteristic item in the region Y _n does not satisfy the target value.

If the given condition in the conditional expected value is that at least one characteristic item does not satisfy the target value, the conditional expected value acquisition unit 17, by considering only the region Y _{n (NG)} , Obtain the nth conditional expectation value _Y'n . Specifically, the conditional expected value acquisition unit 17 normalizes the probability distribution f _n so that the integral value of the probability distribution f _n in the region Y _{n (NG)} becomes 1, and obtains the standardized probability distribution f By calculating the expected values of the random variable y _A and the random variable y _B in the region Y _(NG) using _n , the n-th conditional expected value Y' _n is obtained.

Referring again to FIG. 8, under the probability distribution shown in FIG. 9, the area Y ₁ corresponding to the first observation value probability distribution f ₁ is the observation value where the goal has been achieved, as indicated by the symbol v ₁₃ . It is divided into a region Y _{1 (OK)} corresponding to the value and a region Y _{1 (NG)} corresponding to the observed value that has not reached the target. The conditional expected value acquisition unit 17 takes into account the region Y _{1 (NG)} of the region Y ₁ , as indicated by symbol pr13, and obtains the first observed value probability distribution, as indicated by symbol _v14 . By calculating the expected value of the random variable at _f1 , the first conditional expected value _Y'1 is obtained. For convenience of explanation, a probability distribution with a finite value of 0 or more is assumed only in the region Y ₁ , but if another probability distribution such as a multivariate normal distribution is assumed, the region Y ₁ and the region Y _{1 (OK)} The first conditional expected value Y′ ₁ may be obtained by directly evaluating the conditional expected values in all regions that do not satisfy the target value, without obtaining the region Y1 _(NG) .

In step S18, the model construction unit 12 reconstructs a prediction model for each characteristic item based on the performance data and the first to nth predicted performance data. Here, the l-th (l=1, 2, . . . , n) predicted performance data consists of the l-th recommended design parameter group and the l-th conditional expected value of each characteristic item.

Specifically, the model construction unit 12 generates the first predicted performance data based on the first recommended design parameter group _X1 and the first conditional expected value _Y'1 , as indicated by the symbol _v15 . Get D ₁ . That is, the first predicted performance data _D1 can be regarded as performance data obtained by the T-th next material production. In this way, performance data can be obtained without going through actual material production.

As indicated by the symbol v ₂₁ , the model construction unit 12 calculates the characteristic item m based on the performance data d _t (performance data d ₁ , d ₂ ,..., d _T ) and the first predicted performance data D ₁ . Rebuild each prediction model.

In step S19, the acquisition function construction unit 13 reconstructs the goal-oriented acquisition function based on the prediction model reconstructed in step S18.

In step S20, the design parameter group acquisition unit 14 acquires the (n+1)th recommended design parameter group X _n+1 by optimizing the goal orientation acquisition function reconstructed in step S19. Here, the design parameter group acquisition unit 14 generates a goal-oriented acquisition function constructed using a prediction model constructed based on the performance data _dt and the first predicted performance data _D1 , as shown by code pr21. The recommended design parameter group obtained by the optimization is obtained as a second recommended design parameter group _X2 , as indicated by the symbol _v22 .

Since the purpose of the design support process of this embodiment is to obtain K recommended design parameter groups, it is determined in step S21 whether the variable (n+1) is equal to a predetermined number K. If it is determined that the variable (n+1) is equal to the predetermined number K, the process proceeds to step S23. On the other hand, if the variable (n+1) is not determined to be equal to the predetermined number K, the process proceeds to step S22.

In step S22, the value of variable n is incremented by 1, and the process returns to step S16. With continued reference to FIG. 8, the process when the process returns to step S16 will be described.

In step S16, since the variable n is 2, the observed value probability distribution acquisition unit 16 inputs the second recommended design parameter group X ₂ into the prediction model of each characteristic item m, as indicated by the symbol pr22. As a result, the second observed value probability distribution _f2 is obtained. The second observed value probability distribution _f2 is a simultaneous distribution of all the characteristic items m, and represents predictions of the results of material production, experiments, etc. performed based on the second recommended design parameter group _X2 . That is, the second observed value probability distribution f ₂ can be regarded as a pseudo experimental result based on the second recommended design parameter group X ₂ .

In step S17, as indicated by the symbol _v23 , a region Y2 corresponding to the second observed value probability distribution _f2 is divided into a region _Y2 corresponding to the observed value whose target has been achieved _(OK) and a region Y2 corresponding to the observed value whose target has not been achieved. It is divided into regions Y2 _(NG) corresponding to the observed values. The conditional expected value acquisition unit 17 takes into account the region Y _{2 (NG)} of the region Y ₂ , as shown in symbol pr23, and obtains the second observed value probability distribution, as shown in symbol _v24 . By calculating the expected value of the random variable at _f2 , a second conditional expected value _Y'2 is obtained.

In step S18, the model construction unit 12 generates the second predicted performance data D based on the second recommended design parameter group _X2 and the second conditional expected value _Y'2 , as indicated by the symbol _v25 . Get ₂ . Then, the model construction unit 12 determines the characteristic items based on the performance data d _t (performance data d ₁ , d ₂ ,..., d _T ), the first predicted performance data D ₁ , and the second predicted performance data D ₂ . Rebuild each prediction model of m.

Then, in step S19, the acquisition function construction unit 13 reconstructs the goal-oriented acquisition function based on the prediction model reconstructed in step S18. Subsequently, in step S20, the design parameter group acquisition unit 14 constructs a prediction model based on the performance data _dt , the first predicted performance data _D1 , and the second predicted performance data _D2 . A recommended design parameter group obtained by optimizing the objective-oriented acquisition function is obtained as a third recommended design parameter group _X3 .

In this way, the recommended design parameter group X can be obtained each time the processing of steps S16 to S20 is repeated. Then, after repeating the processing of steps S16 to S20 K-1 times (n=K-1), as shown by the symbol v _K5 , the (K-1)th recommended design parameter group X _K-1 The (K-1)th predicted performance data D K- ₁ is obtained based on the (K-1)th conditional expected value Y' _K-1 .

As indicated by the symbol v _K1 , the model construction unit 12 generates actual performance data d _t (actual data d ₁ , d ₂ ,..., d _T ) and first to (K-1) predicted actual data D ₁ , Based on D ₂ ,...D _K-1 , a prediction model for each characteristic item m is reconstructed (step S18).

The acquisition function construction unit 13 reconstructs the goal-oriented acquisition function based on the reconstructed prediction model (step S19). Subsequently, the design parameter group acquisition unit 14 converts the recommended design parameter group obtained by optimizing the reconstructed goal-oriented acquisition function into a K-th recommended design parameter group _XK , as indicated by the symbol _vK2 . (step S20).

At this stage, the variable n is (K-1), so it is determined in step S21 that the variable (n+1) is equal to the predetermined number K. Therefore, the process proceeds to step S23.

In step S23, the output unit 15 outputs the first to Kth recommended design parameter groups X ₁ , X ₂ , . . . , X _K sequentially acquired by the design parameter group acquisition unit 14.

In this way, by repeating the process from obtaining a group of recommended design parameters to optimizing a goal-oriented acquisition function a desired number of times, a desired number K of recommended design parameters can be obtained without fabricating products or evaluating their characteristics. It becomes possible to obtain a group.

In the conventional process of optimizing explanatory variables and objective variables through repeated feedback of experiment and evaluation results, recommendations for design variables for one experiment are obtained, and experiments and evaluations are conducted based on those recommendations. Otherwise, you will not be able to get recommendations for the next experiment. In the optimization process, if it is necessary to repeat the process of obtaining recommendations, experimenting, and evaluating many times until the objective variable reaches the target value, the experiment and evaluation must be performed each time a recommendation is obtained. However, it was inefficient. In the design support processing of this embodiment, recommendations for a given number K of recommended design parameter groups are obtained for one experiment and evaluation, so for example, K experiments can be performed simultaneously or in parallel. . Therefore, optimization of explanatory variables and objective variables can be efficiently performed.

Referring again to FIG. 2, the design support apparatus 10 may further include a designation reception section 18. The designation receiving unit 18 receives a designation for not adopting j recommended design parameter groups (j is an integer of 1 or more) among the first to Kth recommended design parameter groups. Specifically, the designation receiving unit 18 receives the number of recommended design parameter groups (j pieces (j is an integer of 1 or more)) of recommended design parameter groups to be rejected from the first to Kth recommended design parameter groups. Good too. Further, the designation receiving unit 18 may receive a designation input for a recommended design parameter group to be rejected from among the first to Kth recommended design parameter groups.

The specification receiving unit 18 is configured to specify the user's specifications for _the first to Kth recommended design parameter groups X ₁ , X ₂ , ..., Accepts input of designation of rejection.

For example, a user (for example, an expert in the production of the material, etc.) refers to the presented first to K recommended design parameter groups X ₁ , X ₂ , ..., X _K , and based on each recommended design parameter group, If it is determined that the material produced using the above method will not achieve the target value of the characteristic item, it is possible to input a designation indicating that the recommended design parameter group will not be adopted in the next production of the material.

When the designation of not adopting j recommended design parameter groups is accepted, the design support device 10 selects the first to Kth recommended design parameter groups including the rejected recommended design parameter groups. Based on this, the acquisition of the observed value probability distribution, the acquisition of the conditional expected value, the reconstruction of the prediction model, the reconstruction of the goal-oriented acquisition function, and the acquisition of the recommended design parameter group are further repeated j times.

The output unit 15 further outputs the j recommended design parameter groups sequentially acquired by the design parameter group acquisition unit 14 as (K+1)th to (K+j)th recommended design parameter groups.

The number (j) of design parameter groups that are obviously not suitable for manufacturing a product, etc., among the obtained first to Kth recommended design parameter groups, can be specified by, for example, a person skilled in the technical field of the product, etc. I can accept it. In response to acceptance of the designation, the process from obtaining the observed value probability distribution to optimizing the goal-oriented acquisition function is repeated j times. Then, among the obtained K+j recommended design parameter groups, K design parameter groups excluding j clearly unsuitable design parameter groups are adopted for the next manufacturing process. This optimization process is based on the assumption that none of the K recommended design parameters, including the J clearly unfavorable design parameters, achieves the goal. Since it is reflected in the function, the knowledge of users such as experts is reflected without any special processing. Then, by repeating the optimization process, it is possible to obtain a desired number of recommended design parameter groups while reflecting know-how in manufacturing products and evaluating characteristics in the prediction model and goal-oriented acquisition function.

Next, a design support program for causing a computer to function as the design support apparatus 10 of this embodiment will be described. FIG. 10 is a diagram showing the configuration of the design support program.

The design support program P1 includes a main module m10 that comprehensively controls design support processing in the design support apparatus 10, a data acquisition module m11, a model construction module m12, an acquisition function construction module m13, a design parameter group acquisition module m14, and an output module m15. , an observed value probability distribution acquisition module m16, a conditional expected value acquisition module m17, and a specification reception module m18. Each module m11 to m18 includes a data acquisition section 11, a model construction section 12, an acquisition function construction section 13, a design parameter group acquisition section 14, an output section 15, an observed value probability distribution acquisition section 16, and a conditional expected value acquisition section. 17 and the specification reception unit 18 are realized.

Note that the design support program P1 may be transmitted via a transmission medium such as a communication line, or may be stored in a recording medium M1 as shown in FIG.

According to the design support device 10, the design support method, and the design support program P1 of the present embodiment described above, a prediction model and a goal-oriented acquisition function are constructed based on performance data, and by optimizing the goal-oriented acquisition function, a recommendation is made. A set of design parameters can be obtained. By inputting the first to nth recommended design parameter groups into the prediction model, the probability distribution of the observed values of each characteristic item can be obtained, so the nth recommended design It is possible to predict the distribution of characteristics of manufactured products based on a group of parameters. Therefore, the conditional expected value of the distribution of the characteristics of the product, etc. can be regarded as the observed value of the characteristics of the product, etc. based on the recommended design parameter group. Then, a prediction model and a goal-oriented acquisition function are constructed by using the predicted performance data consisting of pairs of recommended design parameters and conditional expected values together with the actual data, and the (n+1th ), it becomes possible to obtain a set of recommended design parameters. Therefore, by repeating the process from obtaining a group of recommended design parameters to optimizing a goal-oriented acquisition function a desired number of times, a desired number of groups of recommended design parameters can be obtained without fabricating products or evaluating their characteristics. becomes possible.

The present invention has been described above in detail based on the embodiments thereof. However, the present invention is not limited to the above embodiments. The present invention can be modified in various ways without departing from the gist thereof.

In the design support apparatus 10 of this embodiment, the condition for the conditional expected value is that at least one characteristic item does not satisfy the target value, but the condition may be set in more detail. For example, a condition in the conditional expected value may be that all characteristic items do not satisfy the target value. Further, for example, the condition for the conditional expected value may be that some of the characteristic items satisfy the target value and the remaining characteristic items do not satisfy the target value.

P1...Design support program, m10...Main module, m11...Data acquisition module, m12...Model construction module, m13...Acquisition function construction module, m14...Design parameter group acquisition module, m15...Output module, m16...Observation value probability distribution acquisition Module, m17... Conditional expected value acquisition module, m18... Specification reception module, 10... Design support device, 11... Data acquisition section, 12... Model construction section, 13... Acquisition function construction section, 14... Design parameter group acquisition section, 15... Output section, 16... Observed value probability distribution acquisition section, 17... Conditional expected value acquisition section, 18... Specification reception section, 21... Design parameter storage section, 22... Observed value storage section.

Claims (10)

In the design of products, work-in-progress, semi-finished products, parts, or prototypes manufactured based on a design parameter group consisting of multiple design parameters, the determination of design parameters and the products, work-in-progress, semi-finished products based on the determined design parameters In order to apply the method of optimizing design parameters by repeating the production of parts or prototypes, the method is set for each of multiple characteristic items indicating the characteristics of products, work-in-progress, semi-finished products, parts, or prototypes. A design support device that calculates a plurality of the design parameter groups that satisfy a set target value,
a data acquisition unit that acquires a plurality of performance data including the design parameter group and observed values of each of the plurality of characteristic items regarding the manufactured product, the work-in-progress, the semi-finished product, the part, or the prototype; and,
a model construction unit that constructs a prediction model that predicts the observed value of the characteristic item as a probability distribution or its approximation or alternative index based on the design parameter group, based on the performance data;
an acquisition function construction unit that constructs a goal-oriented acquisition function that is a single acquisition function that receives the design parameter group as input and outputs an index value of the design parameter group regarding improvement of the characteristics shown in all the characteristic items;
A design parameter group acquisition unit that acquires the design parameter group obtained by optimizing the goal-oriented acquisition function as a recommended design parameter group, the design parameter group acquisition unit configured using the prediction model constructed based on the performance data. a design parameter group acquisition unit that acquires the recommended design parameter group obtained by optimizing the goal-oriented acquisition function as a first recommended design parameter group;
an output unit that outputs the recommended design parameter group acquired by the design parameter group acquisition unit;
The nth (n=1, 2, ..., K-1, (K is a given integer of 2 or more)) recommended design parameter group obtained by the optimization of the goal-oriented acquisition function is predicted by the an observation value probability distribution acquisition unit that acquires the probability distribution of observed values of each characteristic item obtained by inputting it into the model as an n-th observation value probability distribution;
a conditional expected value acquisition unit that acquires, as an n-th conditional expected value, a conditional expected value given a given condition regarding the observed value based on the n-th observed value probability distribution;
Equipped with
The model construction unit reconstructs the prediction model of each characteristic item based on the performance data and the first to nth predicted performance data, and ) is composed of the l-th recommended design parameter group and the l-th conditional expected value of each characteristic item,
The acquisition function construction unit reconstructs the goal-oriented acquisition function based on the prediction model reconstructed by the model construction unit,
The design parameter group acquisition unit acquires the recommended design parameter group obtained by optimizing the goal-oriented acquisition function reconstructed by the acquisition function construction unit as a (n+1)th recommended design parameter group,
The output unit acquires the observed value probability distribution, the conditional expected value, reconstructs the prediction model, reconstructs the goal orientation acquisition function, and obtains the recommended design parameter group 1) Outputting the first to Kth recommended design parameter groups sequentially acquired by the design parameter group acquisition unit by repeating the times;
Design support equipment. The condition in the conditional expected value is that at least one of the characteristic items does not satisfy the target value;
A design support device according to claim 1. The goal-oriented acquisition function is the probability that the target values of all the characteristic items will be achieved, and is calculated based on the prediction model using the design parameter group as a variable. containing at least a probability term;
A design support device according to claim 1 or 2. The overall achievement probability is the sum of the achievement probabilities for the target value of each characteristic item,
The probability of achieving the target value of each characteristic item is based on the probability distribution of the observed value obtained by inputting the design parameter group into the prediction model of each characteristic item.
The design support device according to claim 3. The target achievement probability term consists of the overall achievement probability or the logarithm of the overall achievement probability,
A design support device according to claim 4. further comprising a designation reception unit that accepts a designation of not adopting j recommended design parameter groups (j is an integer of 1 or more) among the first to Kth recommended design parameter groups;
The output unit further acquires the observed value probability distribution, acquires the conditional expected value, reconstructs the prediction model, reconstructs the goal-oriented acquisition function, and obtains the recommended design parameter group. further outputting the (K+1)-th to (K+j)-th recommended design parameter groups sequentially acquired by the design parameter group acquisition unit by repeating the times;
A design support device according to claim 1. The prediction model is a regression model or a classification model that uses the design parameter group as input and outputs the probability distribution of the observed values,
The model construction unit constructs the prediction model by machine learning using at least the performance data.
A design support device according to claim 1. The prediction model includes a posterior distribution of predicted values based on Bayesian theory, a distribution of predicted values of predictors forming an ensemble, theoretical formulas for the prediction interval and confidence interval of a regression model, Monte Carlo dropout, and multiple models under different conditions. A machine learning model that predicts the probability distribution of the observed value or its approximation or alternative index using any one of the prediction distributions of the constructed predictor,
A design support device according to claim 7. In the design of products, work-in-progress, semi-finished products, parts, or prototypes manufactured based on a design parameter group consisting of multiple design parameters, the determination of design parameters and the products, work-in-progress, semi-finished products based on the determined design parameters In order to apply the method of optimizing design parameters by repeating the production of parts or prototypes, the method is set for each of multiple characteristic items indicating the characteristics of products, work-in-progress, semi-finished products, parts, or prototypes. 1. A design support method in a design support device for determining a plurality of the design parameter groups that satisfy set target values, the method comprising:
a data acquisition step of acquiring a plurality of performance data including the design parameter group and observed values of each of the plurality of characteristic items regarding the manufactured product, the work-in-progress, the semi-finished product, the part, or the prototype; and,
a model construction step of constructing a prediction model based on the actual data to predict the observed value of the characteristic item as a probability distribution or its approximation or alternative index based on the design parameter group;
an acquisition function construction step of constructing a goal-oriented acquisition function that is a single acquisition function that receives the design parameter group as input and outputs an index value of the design parameter group regarding improvement of the characteristics shown in all the characteristic items;
a design parameter group acquisition step of acquiring the design parameter group obtained by optimizing the goal-oriented acquisition function as a recommended design parameter group, the design parameter group being constructed using the predictive model constructed based on the actual performance data; a design parameter group obtaining step of obtaining the recommended design parameter group obtained by optimizing the goal-oriented acquisition function as a first recommended design parameter group;
The nth (n=1, 2, ..., K-1, (K is a given integer of 2 or more)) recommended design parameter group obtained by the optimization of the goal-oriented acquisition function is predicted by the an observed value probability distribution obtaining step of obtaining the probability distribution of observed values of each characteristic item obtained by inputting it into the model as an n-th observed value probability distribution;
a conditional expected value obtaining step of obtaining a conditional expected value given a given condition regarding the observed value as the n-th conditional expected value based on the n-th observed value probability distribution;
a model reconstruction step of reconstructing the prediction model of each characteristic item based on the performance data and the first to nth predicted performance data, the lth (l=1, 2, . . . .
an acquisition function reconstruction step of reconstructing the goal-oriented acquisition function based on the prediction model reconstructed in the model reconstruction step;
a design parameter group reacquisition step of acquiring the recommended design parameter group obtained by optimizing the goal-oriented acquisition function reconstructed in the acquisition function reconstruction step as an (n+1)th recommended design parameter group;
a repetition step of repeating the observed value probability distribution acquisition step, the conditional expected value acquisition step, the model reconstruction step, the acquisition function reconstruction step, and the design parameter group reacquisition step (K-1) times;
an output step of outputting the first to Kth recommended design parameter groups sequentially acquired in the design parameter group acquisition step and the repeating step;
A design support method having In the design of products, work-in-progress, semi-finished products, parts, or prototypes that are manufactured based on a design parameter group consisting of multiple design parameters, a computer is used to determine the design parameters and to determine the products and work-in-progress based on the determined design parameters. , each of multiple characteristic items indicating the characteristics of a product, work-in-progress, semi-finished product, part, or prototype, in order to apply the method of optimizing design parameters by repeating the production of semi-finished products, parts, or prototypes. A design support program for functioning as a design support device for determining a plurality of design parameter groups that satisfy target values set for
A data acquisition function that acquires a plurality of performance data consisting of the design parameter group and observed values of each of the plurality of characteristic items regarding the manufactured product, the work-in-progress, the semi-finished product, the part, or the prototype. and,
a model construction function that constructs a prediction model that predicts the observed value of the characteristic item as a probability distribution or its approximation or alternative index based on the design parameter group, based on the performance data;
an acquisition function construction function that constructs a goal-oriented acquisition function that is a single acquisition function that receives the design parameter group as input and outputs an index value of the design parameter group regarding improvement of the characteristics shown in all the characteristic items;
A design parameter group acquisition function that acquires the design parameter group obtained by optimizing the goal-oriented acquisition function as a recommended design parameter group, the function being constructed using the predictive model constructed based on the actual performance data. a design parameter group acquisition function that acquires the recommended design parameter group obtained by optimizing the goal-oriented acquisition function as a first recommended design parameter group;
an output function that outputs the recommended design parameter group acquired by the design parameter group acquisition function;
The nth (n=1, 2, ..., K-1, (K is a given integer of 2 or more)) recommended design parameter group obtained by the optimization of the goal-oriented acquisition function is predicted by the an observation value probability distribution acquisition function that acquires the probability distribution of observed values of each characteristic item obtained by inputting it into the model as an n-th observation value probability distribution;
a conditional expected value acquisition function that acquires a conditional expected value given a given condition regarding observed values as an n-th conditional expected value based on the n-th observed value probability distribution;
Realize,
The model construction function reconstructs the prediction model of each characteristic item based on the performance data and the first to nth predicted performance data, and ) is composed of the l-th recommended design parameter group and the l-th conditional expected value of each characteristic item,
The acquisition function construction function reconstructs the goal-oriented acquisition function based on the prediction model reconstructed by the model construction function,
The design parameter group acquisition function acquires the recommended design parameter group obtained by optimizing the goal-oriented acquisition function reconstructed by the acquisition function construction function as a (n+1)th recommended design parameter group,
The output function includes (K- 1) Outputting the first to Kth recommended design parameter groups sequentially acquired by the design parameter group acquisition unit by repeating the times;
Design support program.

Images