Method for generating training data, method for generating a trained model, estimation method, training data generation program, trained model generation program, estimation program, and laminated film

Abstract

To provide a method for generating training data, a method for generating a trained model, an estimation method, a training data generation program, a trained model generation program, an estimation program, and a laminated film that enable high-precision design of optical layers for optical properties. [Solution] A method for generating training data for a computer to estimate the laminated structure of a laminated film comprising multiple optical layers, comprising the steps of: obtaining the refractive index distribution and layer thickness distribution of the laminated film; generating refractive index parameters that can be converted with the refractive index distribution; generating layer thickness parameters that can be converted with the layer thickness distribution; obtaining the optical properties of the laminated film made by the layer thickness parameters and refractive index parameters; and generating training data consisting of a pair of layer thickness parameters and refractive index parameters and the optical properties of the laminated film made by said layer thickness parameters and refractive index parameters.

Description

[Technical Field]

[0001] The present invention relates to a method for generating training data, a method for generating a trained model, an estimation method, a training data generation program, a trained model generation program, an estimation program, and a laminated film. [Background technology]

[0002] The optical properties of a laminated film are adjusted by laminating multiple optical layers that have at least some different optical properties (see, for example, Patent Document 1). Patent Document 1 uses a genetic algorithm to optimize the refractive index and layer thickness of each layer as variables. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2004-37612 [Overview of the project] [Problems that the invention aims to solve]

[0004] Incidentally, it is known that the prediction accuracy of a trained model decreases with increasing explanatory variables and differences in scale between explanatory variables, and this is also true for optimization algorithms such as genetic algorithms. However, in the optical layer optimization described in Patent Document 1, the number of variables increases by twice the number of layers, and it is not possible to interpolate the difference in scale between refractive index and layer thickness. Therefore, the applicability of the optimization method is substantially limited to laminated films with a small number of layers in order to avoid the decrease in prediction accuracy that occurs with the expansion of the data space. Thus, there is a need for a learning method that has high prediction accuracy even for laminated films with a large number of layers.

[0005] The present invention has been made in view of the above, and aims to provide a method for generating training data, a method for generating a trained model, an estimation method, a training data generation program, a trained model generation program, an estimation program, and a laminated film that can perform high-precision design of an optical layer with respect to optical properties. [Means for solving the problem]

[0006] To solve the above-mentioned problems and achieve the objective, the present invention provides a learning data generation method for a computer to generate learning data for estimating the laminated structure of a laminated film in which a plurality of optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction differ, the method comprising: a distribution acquisition step of acquiring the refractive index distribution and layer thickness distribution of the laminated film; a refractive index parameter generation step of generating refractive index parameters that can be interconverted with the refractive index distribution of the laminated film for the refractive index of each optical layer of the laminated film; a layer thickness parameter generation step of generating layer thickness parameters that can be interconverted with the layer thickness distribution of the laminated film for the layer thickness distribution of optical layers made of the same type of material; a characteristic acquisition step of acquiring the optical properties of the laminated film produced by the layer thickness parameters and the refractive index parameters; and a learning data generation step of generating learning data in which the layer thickness parameters and the refractive index parameters are paired with the optical properties of the laminated film produced by the layer thickness parameters and the refractive index parameters.

[0007] Furthermore, in the learning data generation method according to the present invention, the number of optical layers is 51 or more.

[0008] Furthermore, in the learning data generation method according to the present invention, the refractive index parameter is a parameter whose dimensions have been reduced so as to be interconvertible with the refractive index distribution of the laminated film, and the layer thickness parameter is a parameter whose dimensions have been reduced so as to be interconvertible with the layer thickness distribution of the laminated film.

[0009] Furthermore, in the learning data generation method according to the present invention, the total number of variables for the layer thickness parameter and the refractive index parameter is 9 or more and 102 or less.

[0010] Furthermore, in the learning data generation method according to the present invention, the layer thickness parameter is a parameter that can represent the layer thickness distribution of the laminated film as a curve.

[0011] Furthermore, the learning data generation method according to the present invention includes, in the above invention, a refractive index parameter generation step, which includes a condition acquisition step of acquiring a number of components indicating the number of types of raw materials constituting the resin layer of the laminated film and a regular arrangement indicating the regularity of the lamination based on the types of raw materials; a classification step of grouping a plurality of optical layers based on the regular arrangement acquired in the condition acquisition step; a refractive index calculation step of calculating the refractive index of the optical layer of each group using the least squares method with respect to the refractive index distribution; and a refractive index parameter setting step of setting the number of components, regular arrangement, and refractive index of the optical layer of each group as refractive index parameters, which minimize the difference between the refractive index of the optical layer of each group calculated and the refractive index distribution calculated from the regular arrangement and the refractive index distribution acquired in the distribution acquisition step.

[0012] Furthermore, the learning data generation method according to the present invention includes, in the above invention, a layer thickness parameter generation step, a condition acquisition step, a condition acquisition step, a periodic array setting step, a periodic array indicating the periodicity of lamination based on the type of raw material and layer thickness of the optical layer, a classification step, a group of optical layers based on the set periodic array, a coefficient value calculation step, a coefficient value representing the layer thickness distribution of the optical layer of each group, calculated by mathematical optimization using the layer thickness distribution, and a layer thickness parameter setting step, a periodic array and coefficient value for each group that minimize the discrepancy between the calculated coefficient value for each group, the layer thickness distribution calculated from the periodic array, and the layer thickness distribution obtained in the distribution acquisition step, as layer thickness parameters.

[0013] Furthermore, in the learning data generation method according to the present invention, the layer thickness parameter includes a function type that indicates the thickness distribution of the optical layer, and a coefficient related to the function type.

[0014] Furthermore, in the learning data generation method according to the present invention, the layer thickness parameter includes a function type that indicates the thickness distribution of a plurality of optical layers constituting a regular arrangement, and a coefficient related to the function type.

[0015] Furthermore, in the learning data generation method according to the present invention, the refractive index distribution and layer thickness distribution of the laminated film are generated based on data points of the constituent elements generated based on the calculation conditions.

[0016] Furthermore, the method for generating a trained model according to the present invention is a method for generating a trained model for a computer to generate a trained model for estimating the laminated structure of a laminated film in which a plurality of optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction are different, and the method includes a learning step of generating a trained model by learning with the following variables: a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film for the refractive index of each optical layer, and a layer thickness parameter that can be interconverted with the layer thickness distribution of the laminated film for the layer thickness distribution of optical layers made of the same type of material, and the optical properties of the laminated film created by the layer thickness parameter and the refractive index parameter as the target variable.

[0017] Furthermore, the estimation method according to the present invention is an estimation method in which a computer estimates the laminated structure of a laminated film in which a plurality of optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction are different, and includes an estimation step in which the layer thickness parameter and the refractive index parameter to be estimated are input to a trained model generated by learning, in which the refractive index of each optical layer is a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film, and the layer thickness parameter that can be interconverted with the layer thickness distribution of the laminated film for the layer thickness distribution of optical layers made of the same type of material is an explanatory variable, and the optical properties of the laminated film created by the layer thickness parameter and the refractive index parameter are the objective variable, and the optical properties of the laminated film to be estimated are input, and the optical properties of the laminated film to be estimated are obtained by inputting the layer thickness parameter and the refractive index parameter to be estimated, and obtaining the optical properties of the laminated film composed of the said layer thickness parameter and refractive index parameter, a selection step in which the layer thickness parameter and the refractive index parameter that show the optimal optical properties are selected from the optical properties of the laminated film obtained in the estimation step, and a conversion step in which the layer thickness parameter and the refractive index parameter selected in the selection step are converted into the layer thickness distribution and the refractive index distribution.

[0018] Furthermore, the estimation method according to the present invention is an estimation method in which a computer estimates the laminated structure of a laminated film in which a plurality of optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction are different, and includes an estimation step in which the optical properties to be estimated are input to a trained model generated by learning, in which the refractive index of each optical layer is a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film, and the thickness of the layer thickness distribution of optical layers made of the same type of material is a thickness parameter that can be interconverted with the thickness of the laminated film, and the optical properties of the laminated film created by the thickness of the layer and refractive index parameters are input to obtain the optical properties to be estimated and obtain the thickness of the layer and refractive index parameters corresponding to the optical properties, and a conversion step in which the thickness of the layer and refractive index parameters obtained in the estimation step are converted into the thickness of the layer and refractive index distribution.

[0019] Furthermore, the learning data generation program according to the present invention is a learning data generation program that causes a computer to generate learning data for estimating the laminated structure of a laminated film in which a plurality of optical layers are stacked, wherein the optical properties of adjacent optical layers in the stacking direction are different, and the program causes the computer to execute the following steps: a distribution acquisition step to acquire the refractive index distribution and layer thickness distribution of the laminated film; a refractive index parameter generation step to generate refractive index parameters that can be interconverted with the refractive index distribution of the laminated film for the refractive index of each optical layer of the laminated film; a layer thickness parameter generation step to generate layer thickness parameters that can be interconverted with the layer thickness distribution of the laminated film for the layer thickness distribution of optical layers made of the same type of material; a characteristic acquisition step to acquire the optical properties of the laminated film made by the layer thickness parameters and the refractive index parameters; and a learning data generation step to generate learning data in which the layer thickness parameters and the refractive index parameters are paired with the optical properties of the laminated film made by the layer thickness parameters and the refractive index parameters.

[0020] Furthermore, the trained model generation program according to the present invention is a trained model generation program that causes a computer to generate a trained model for estimating the laminated structure of a laminated film in which a plurality of optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction are different, and the program causes the computer to execute a learning step in which the computer generates a trained model by learning with the following variables: a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film for the refractive index of each optical layer, and a layer thickness parameter that can be interconverted with the layer thickness distribution of the laminated film for the layer thickness distribution of optical layers made of the same type of material, and the optical properties of the laminated film created by the layer thickness parameter and the refractive index parameter as the target variable.

[0021] Further, the estimation program according to the present invention is an estimation program for causing a computer to estimate a laminated structure of a laminated film in which a plurality of optical layers are laminated and the optical properties of the optical layers adjacent to each other in the lamination direction are different in at least a part thereof. The refractive index parameters that can be mutually converted with the refractive index distribution of the laminated film for the refractive index of each optical layer, and the layer thickness parameters that can be mutually converted with the layer thickness distribution of the laminated film for the layer thickness distribution of the optical layers made of the same kind of material are explanatory variables, and an estimated step of inputting the layer thickness parameters and the refractive index parameters of the estimation target into a learned model generated by learning with the optical properties of the laminated film produced by the layer thickness parameters and the refractive index parameters as objective variables to obtain the optical properties of the laminated film constituted by the layer thickness parameters and the refractive index parameters, and a selection step of selecting the layer thickness parameters and the refractive index parameters showing the optimum optical properties from the optical properties of the laminated film obtained by the estimation step are executed by the computer.

[0022] Further, the estimation program according to the present invention is an estimation program for causing a computer to estimate a laminated structure of a laminated film in which a plurality of optical layers are laminated and the optical properties of the optical layers adjacent to each other in the lamination direction are different in at least a part thereof. The refractive index parameters that can be mutually converted with the refractive index distribution of the laminated film for the refractive index of each optical layer, and the layer thickness parameters that can be mutually converted with the layer thickness distribution of the laminated film for the layer thickness distribution of the optical layers made of the same kind of material are objective variables, and an estimated step of inputting the optical properties of the estimation target into a learned model generated by learning with the optical properties of the laminated film produced by the layer thickness parameters and the refractive index parameters as explanatory variables to obtain the layer thickness parameters and the refractive index parameters corresponding to the optical properties is executed by the computer.

[0023] In addition, the laminated film according to the present invention is obtained by using at least one of the learning data generation method according to claim 1, the learned model generation method according to claim 11, the estimation method according to claim 12 or 13, the learning data generation program according to claim 14, the learned model generation program according to claim 15, and the estimation program according to claim 16 or 17.

Advantages of the Invention

[0024] According to the present invention, the design of the optical layer with respect to the optical characteristics can be performed with high accuracy.

Brief Description of the Drawings

[0025] [Figure 1] FIG. 1 is a diagram showing a schematic configuration of an estimation system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram for explaining the layer thickness distribution. [Figure 3] FIG. 3 is a diagram showing an example of the spectral spectrum of the laminated film. [Figure 4] FIG. 4 is a diagram for explaining the designed optimal thickness distribution. [Figure 5] FIG. 5 is a block diagram showing a configuration of a learning device included in an estimation system according to an embodiment of the present invention. [Figure 6] FIG. 6 is a block diagram showing a configuration of an estimation device included in an estimation system according to an embodiment of the present invention. [Figure 7] FIG. 7 is a diagram for explaining the flow of estimation processing performed by an estimation system according to an embodiment of the present invention. [Figure 8] FIG. 8 is a flowchart showing the flow of learning data generation processing performed by a learning device according to an embodiment of the present invention. [Figure 9] FIG. 9 is a diagram for explaining the refractive index distribution and the layer thickness distribution. [Figure 10] FIG. 10 is a flowchart for explaining the estimated value calculation processing performed by an estimation device according to an embodiment of the present invention. [Figure 11] Figure 11 is a flowchart showing the flow of the refractive index parameter generation process performed by the learning device according to Modification Example 1. [Figure 12] Figure 12 is a diagram illustrating an example of information (calculation condition table) related to the generation of refractive index parameters. [Figure 13] Figure 13 is a diagram illustrating another example of information (calculation condition table) related to the generation of refractive index parameters. [Figure 14] Figure 14 is a flowchart showing the flow of the layer thickness parameter generation process performed by the learning device according to Modification Example 1. [Figure 15] Figure 15 is a diagram illustrating the layer thickness distribution of a 400-layer laminated film consisting of two types of optical layers (layer A and layer B) with different compositions. [Figure 16] Figure 16 is a diagram illustrating an example of information (coefficient values) related to the generation of layer thickness parameters. [Figure 17] Figure 17 is a diagram illustrating another example of information (coefficient values) related to the generation of layer thickness parameters. [Figure 18] Figure 18 is a diagram illustrating parameter dimensionality reduction. [Figure 19] Figure 19 illustrates the relationship between the number of explanatory variables and the average reflectance of the laminated film. [Figure 20] Figure 20 is a flowchart illustrating the estimated value calculation process performed by the estimation device according to Modification Example 2. [Modes for carrying out the invention]

[0026] The following describes in detail, with reference to the drawings, embodiments of the training data generation method, trained model generation method, estimation method, training data generation program, trained model generation program, estimation program, and laminated film according to the present invention. However, the present invention is not limited to these embodiments. Furthermore, the individual embodiments of the present invention are not independent but can be combined and implemented as appropriate.

[0027] (Embodiment) Figure 1 is a diagram showing a schematic configuration of an estimation system according to one embodiment of the present invention. The estimation system 1 shown in these figures comprises a learning device 2 that creates training data and generates a trained model using the created training data, an estimation device 3 that estimates the optical characteristics and design information of the target to be estimated using the trained model generated by the learning device 2, a display device 4 that displays information including the estimation results of the estimation device 3, and an input device 5.

[0028] In this embodiment, the optical properties estimated by the estimation device 3 are the spectral spectra of a laminated film formed by stacking multiple optical layers that differ in at least some of their optical properties. In this laminated film, the optical properties change depending on the stacking configuration of the optical layers (number of layers and layer thickness). Examples of optical properties include reflectance and transmittance, and the spectral spectrum shows the reflectance / transmittance of light with respect to wavelength. As design information for the laminated film, each optical layer is assigned a number (layer number) in the order of stacking, and the layer thickness is set corresponding to the layer number.

[0029] In this embodiment, the thickness distribution and refractive index distribution of each optical layer constituting the laminated film are used as data for generating training data. Figure 2 illustrates the distribution of layer thickness. By plotting the layer thickness against the layer number and generating, for example, an approximation curve, the relationship between layer number and layer thickness can be obtained, as shown in Figure 2. The example shown in Figure 2 illustrates a case where the layer thickness increases as the layer number increases. Similarly, the refractive index distribution is also represented by a curve (approximation curve) that shows the relationship between the layer number and the refractive index. Thus, the refractive index distribution and layer thickness distribution of the laminated film are generated based on data points of the constituent elements that are generated based on the calculation conditions.

[0030] Figure 3 shows an example of the spectral spectrum of a laminated film. When designing a laminated film, for example, a spectral spectrum like the one shown in Figure 3 is set as an optical property. This spectral spectrum shows the reflectance of light at a given wavelength, which is determined by the refractive index and thickness of each optical layer. The conditions for obtaining a laminated film with this spectral spectrum include setting the thickness (thickness distribution), type, and number of layers of the optical layers.

[0031] Figure 4 illustrates the designed optimal thickness distribution. For example, suppose the initial layer thickness distribution is the distribution shown in Figure 2, and the target spectral distribution is the spectrum shown in Figure 3. In this case, the estimation system 1 generates a trained model by having the learning device 2 learn the refractive index distribution and layer thickness distribution of the laminated film, as well as the optical properties of those distributions. The estimation device 3 then uses this trained model to estimate the refractive index distribution and layer thickness distribution that exhibit the target spectral distribution. For example, Figure 4 shows the curve L1 (solid line) which represents the estimated layer thickness distribution. Note that in Figure 4, the dashed curve L2 represents the initial layer thickness distribution shown in Figure 2. In this way, the estimation system 1 uses the trained model to estimate the refractive index distribution and layer thickness distribution that exhibit the desired optical properties.

[0032] The learning device 2 is electrically connected to the estimation device 3. The learning device 2 selectively extracts training data and generates and outputs a trained model by training using the extracted training data. Figure 5 is a block diagram showing the configuration of the learning device in the estimation system according to an embodiment of the present invention. The learning device 2 has a training data generation unit 21, a learning unit 22, a control unit 23, and a storage unit 24. In this embodiment, the training data generation device is composed of at least the training data generation unit 21 and the storage unit 24.

[0033] The learning data generation unit 21 includes a distribution acquisition unit 211, a refractive index parameter generation unit 212, a layer thickness parameter generation unit 213, a characteristic acquisition unit 214, and a dataset generation unit 215.

[0034] The distribution acquisition unit 211 acquires the refractive index distribution and layer thickness distribution of the laminated film, as well as the optical properties of the distribution, from the storage unit 24.

[0035] The refractive index parameter generation unit 212 generates refractive index parameters for each optical layer of the laminated film that are interconvertible with the refractive index distribution of the laminated film.

[0036] The layer thickness parameter generation unit 213 generates layer thickness parameters that can be interconverted between the layer thickness distribution of an optical layer made of the same type of material and the layer thickness distribution of a laminated film.

[0037] The characteristic acquisition unit 214 acquires the optical properties of the laminated film produced by the layer thickness parameter and the refractive index parameter.

[0038] The dataset generation unit 215 generates training data consisting of a layer thickness parameter and a refractive index parameter, and the optical properties of the laminated film produced by said layer thickness parameter and refractive index parameter.

[0039] The learning unit 22 uses the training data generated by the training data generation unit 21 to perform training and generate a trained model. The training performed by the learning unit 22 can employ known machine learning methods such as Bayesian optimization, genetic algorithms, linear regression, nonlinear regression, logistic regression, k-nearest neighbors, support vector machines, decision trees, random forests, gradient boosting trees, k-means algorithms, principal component analysis, and deep learning. The trained model is, for example, a neural network consisting of an input layer, hidden layers, and output layers, with each layer having one or more nodes. Information such as network parameters in the trained model is stored in the memory unit 24. Network parameters include information about the weights and biases between layers of the neural network.

[0040] Furthermore, when the learning unit 22 generates a trained model, for example, by learning using regularization, it is given multiple candidate values ​​for the hyperparameters of the regression model and performs learning for each of the given candidate values. After that, it calculates the prediction error using cross-validation or holdout validation with the training data for the model obtained by learning with each candidate value, and selects the regression model that gives the smallest prediction error. The selected regression model is output as the trained model. Here, hyperparameters refer to, for example, the number of layers in the neural network or the regularization coefficients.

[0041] The control unit 23 comprehensively controls the operation of the learning device 2.

[0042] The memory unit 24 stores data including various programs for operating the learning device 2, and various parameters necessary for the operation of the learning device 2. The various programs include a learning data generation program that generates training data for generating a trained model, and a trained model generation program that uses the training data to train and generate a trained model. The memory unit 24 also has a dataset storage unit 241 that stores multiple datasets that constitute the training data. Furthermore, the memory unit 24 generates refractive index parameters and layer thickness parameters within a pre-set range as needed, and stores the corresponding refractive index distribution and layer thickness distribution. Here, it is preferable to set the parameter range based on the purpose of generating a trained model. For example, assuming actual film production, constraints such as the types of raw materials available and the upper and lower limits of the number of layers that would make it possible to manufacture the device in a way that is realistic from the standpoint of economy and processing accuracy can be set as the parameter range. As for the parameter generation method, for example, random generation or grid generation can be used.

[0043] The memory unit 24 is composed of a ROM (Read Only Memory) on which various programs are pre-installed, and RAM (Random Access Memory), HDD (Hard Disk Drive), SSD (Solid State Drive), etc., which store calculation parameters and data for each process.

[0044] Various programs can be recorded on computer-readable recording media such as HDDs, flash memory, CD-ROMs, DVD-ROMs, and Blu-ray® discs and widely distributed. Furthermore, the learning device 2 can acquire various programs via a communication network. This communication network can be comprised of existing public telephone networks, LANs (Local Area Networks), WANs (Wide Area Networks), etc., and can be wired or wireless.

[0045] The learning device 2 having the above functional configuration is a computer composed of one or more hardware components such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), and an FPGA (Field Programmable Gate Array).

[0046] The estimation device 3 is electrically connected to the learning device 2 and the display device 4. The estimation device 3 uses a trained model acquired from the learning device 2 to output a layer thickness distribution and refractive index distribution that represent the required optical properties. Figure 6 is a block diagram showing the configuration of the estimation device in the estimation system according to an embodiment of the present invention. The estimation device 3 has a calculation unit 31, a control unit 32, and a storage unit 33.

[0047] The calculation unit 31 calculates the optical properties of the laminated film, the optimal layer thickness distribution, and the refractive index distribution, which are estimated using the conditions of the target to be estimated obtained from the input device 5 and the trained model obtained from the learning device 2.

[0048] Figure 7 is a diagram illustrating the flow of the estimation process performed by the estimation system according to an embodiment of the present invention. The calculation unit 31 acquires a trained model 100 that has been trained using the training data set IP. Using this trained model 100, the calculation unit 31 outputs the estimated optical properties under the conditions of the target to be estimated, for example, as estimated physical property values ​​OP. The training data set IP is data consisting of a layer thickness parameter and a refractive index parameter, and the optical properties of the laminated film produced by said layer thickness parameter and refractive index parameter.

[0049] The control unit 32 comprehensively controls the operation of the estimation device 3. The control unit 32 includes a display control unit 321 that displays the calculation results (estimation results) of the calculation unit 31 on the display device 4. In addition to the estimation results, the display control unit 321 may also display information about the estimation targets, such as each condition, on the display device 4.

[0050] The memory unit 33 stores data including various programs for operating the estimation device 3, and various parameters necessary for the operation of the estimation device 3. The various programs include estimation programs that are executed using trained models. The memory unit 33 is composed of a ROM with various programs pre-installed, and RAM, HDD, SSD, etc., for storing calculation parameters and data for each process.

[0051] Various programs can be recorded on computer-readable recording media such as HDDs, flash memory, CD-ROMs, DVD-ROMs, and Blu-ray® discs and widely distributed. Furthermore, the estimation device 3 can acquire various programs via a communication network.

[0052] The estimation device 3, having the above functional configuration, is a computer composed of one or more hardware components such as a CPU, GPU, ASIC, and FPGA.

[0053] The display device 4 is a display made of liquid crystal or organic EL (Electro-Luminescence), and is electrically connected to the estimation device 3. The display device 4 acquires and displays display data output from the estimation device 3 under the control of the display control unit 321. The display device 4 may also have an audio output function such as a speaker.

[0054] Input device 5 accepts various types of information, including settings related to the process of estimating characteristic values, and outputs the received information to learning device 2 and estimation device 3. Input device 5 is configured using a user interface such as a keyboard, mouse, microphone, and touch panel.

[0055] Next, the processes of the learning device 2 and the estimation device 3 will be explained. In the estimation system 1, the learning unit 22 generates a trained model by learning using the training data generated by the training data generation unit 21, and the estimation device 3 uses the trained model to output the layer thickness distribution and refractive index distribution that represent the set optical characteristics.

[0056] First, let's explain the training data generated by the training data generation unit 21. Figure 8 is a flowchart showing the flow of the learning data generation process performed by a learning device according to one embodiment of the present invention.

[0057] The training data generation unit 21 acquires the refractive index distribution and the layer thickness distribution (step 101). In this process, the training data generation unit 21 acquires the refractive index distribution and the layer thickness distribution by referring to the storage unit 24 or by reading data from an external server.

[0058] Here, we will explain the refractive index distribution and layer thickness distribution. Figure 9 is a diagram illustrating the refractive index distribution and layer thickness distribution. As shown in Figure 9, the refractive index distribution is registered by associating it with the optical layer. Similarly, the layer thickness distribution is registered by associating it with the optical layer and its thickness. Each distribution has as many refractive indices and thicknesses as there are optical layers. The number of optical layers is preferably 51 or more. Having 51 or more optical layers allows for a design that satisfies the desired optical performance even in a laminated system consisting of layers with low refractive indices, and allows for a wide range of material combinations to be selected, which is preferable. Here, a laminated system consisting of layers with low refractive indices refers to a laminated system in which the refractive indices of all optical layers constituting the laminated film are 2.0 or less. The reasons why 51 layers are preferred are explained below.

[0059] The optical performance of a laminated film depends on the refractive index of the optical layers that make up the laminated film and the thickness distribution of those optical layers. For example, when optimizing layer thickness using silicon dioxide (refractive index: approximately 1.4) and niobium oxide (refractive index: approximately 2.3), using niobium oxide with a refractive index of approximately 2.3 allows for the design of a laminated film that satisfies the desired optical performance even with a small number of layers.

[0060] However, the higher the refractive index of a layer, the more sensitive its optical performance becomes to changes in layer thickness. Therefore, optical performance is more susceptible to change during the manufacturing or processing stages, leading to reduced handling properties. In contrast, if it becomes possible to design a laminated system consisting of layers with a low refractive index, the range of raw materials used in the laminated film can be expanded, and handling can be improved. In a laminated system consisting of layers with a low refractive index, the design of the number of optical layers and the thickness distribution of the layers becomes important. From the above viewpoint, it is preferable that the number of optical layers be 51 or more.

[0061] After obtaining the refractive index distribution and layer thickness distribution, the refractive index parameter generation unit 212 generates refractive index parameters using the refractive index distribution (step S102). The refractive index parameter generation unit 212 generates refractive index parameters that can be converted to and from the refractive index distribution of the laminated film for the refractive index of each optical layer of the laminated film. For example, when the refractive index of each optical layer of the laminated film is input, the refractive index parameter generation unit 212 calculates refractive index parameters generated by processing such as scaling, using the value obtained by dividing the refractive index of each optical layer by the median value in the refractive index distribution of the laminated film and the median value.

[0062] The layer thickness parameter generation unit 213 generates layer thickness parameters using the layer thickness distribution (step S103). The layer thickness parameter generation unit 213 generates layer thickness parameters that can be converted between the layer thickness distribution of the laminated film and the layer thickness distribution of optical layers made of the same type of material. The layer thickness parameters are parameters that can represent the layer thickness distribution of the laminated film as a curve, for example. When the layer thickness of each optical layer of the laminated film is input, the layer thickness parameter generation unit 213 calculates layer thickness parameters that are generated by processing such as scaling, for example, by dividing the layer thickness of each optical layer by the median value in the layer thickness distribution of the laminated film and the median value.

[0063] The characteristic acquisition unit 214 acquires the optical properties of the laminated film fabricated using the layer thickness parameter and the refractive index parameter (step S104). The characteristic acquisition unit 214 acquires the optical properties from the corresponding combination of refractive index distribution and layer thickness distribution for the pair of layer thickness parameter and refractive index parameter. In this case, the optical properties are calculated by optical calculations using the refractive index distribution and layer thickness distribution. Known methods can be used for the optical calculations. For example, the spectral spectrum of a laminated film in any wavelength range can be calculated using the calculation formula described on pages 44-49 of the first edition of Macleod's "Theory of Optical Thin Films". Furthermore, the color tone under any light source can be calculated by using the above spectral spectrum calculation formula and the calculation formula and light source intensity described in the JIS Z 8722 standard (2009).

[0064] The dataset generation unit 215 generates training data consisting of a layer thickness parameter and a refractive index parameter, and the optical properties of the laminated film produced by said layer thickness parameter and refractive index parameter (step S105).

[0065] The learning unit 22 generates a trained model by learning using the generated training data. The learning unit 22 generates the trained model by learning using, for example, Bayesian optimization, genetic algorithms, linear regression, nonlinear regression, or deep learning models. The machine learning algorithm is selected according to the purpose for which the trained model will be used. The advantages of using each algorithm are described below.

[0066] <Bayesian optimization and genetic algorithms> This method is preferable when you want to efficiently find a laminated film configuration that satisfies the desired optical performance. Because the learning cost is lower compared to deep learning and analysis can be performed in a short time, it is preferable to use Bayesian optimization and genetic algorithms rather than deep learning models when the dimensionality of the target variable is small. For example, the components of a laminated film that maximizes the average reflectivity can be obtained through prediction and retraining using a pre-trained model.

[0067] <Linear Regression / Nonlinear Regression> This method is preferably used when analyzing training data and identifying important constituent variables of a laminated film that are necessary to satisfy the desired optical performance, such as by extracting variable importance. For example, a Lasso model can be generated with the constituent variables of the laminated film as explanatory variables and the average reflectance as the dependent variable, and the variables contributing to the average reflectance can be identified by extracting explanatory variables with large regression coefficients.

[0068] <Deep Learning Models> This method is preferable when you want to efficiently find a laminated film configuration that satisfies the desired optical performance. Bayesian optimization and genetic algorithms are preferable to Bayesian optimization and genetic algorithms when there are dozens or more objective variables, because the computation time required to reach the desired optical performance increases exponentially as the dimensionality of the objective variable increases. For example, the reflectance for each 1 nm wavelength is used as the explanatory variable, and the components of the laminated film are used as the objective variable. The components of the laminated film for the desired spectral spectrum (wavelength-reflectance graph) are then output.

[0069] Estimation device 3 estimates the layer thickness distribution and refractive index distribution that represent the set optical properties. Figure 10 is a flowchart illustrating the estimation process performed by the estimation device according to this embodiment. In the example shown in Figure 10, the optical properties are set in advance, and the trained model outputs the estimated optical properties based on the input of the layer thickness parameter and the refractive index parameter.

[0070] The calculation unit 31 inputs the layer thickness parameter and refractive index parameter into the trained model generated by the learning device 2 (step S201). At this time, the calculation unit 31 reads out multiple sets of layer thickness parameters and refractive index parameters stored in the storage unit 24, for example, and inputs the parameters into the trained model for each set.

[0071] Then, the calculation unit 31 selects a layer thickness parameter and a refractive index parameter from the estimated optical characteristics output from the trained model that show or show values ​​close to the preset optical characteristics (step S202).

[0072] Subsequently, the calculation unit 31 converts the selected layer thickness parameter and refractive index parameter into a layer thickness distribution and a refractive index distribution (step S203). If the refractive index parameter and layer thickness parameter are values ​​generated by the scaling process, for example, the refractive index of each layer and the median refractive index, the calculation unit 31 uses this median to convert them into a refractive index distribution and a layer thickness distribution. This yields a layer thickness distribution and refractive index distribution that are estimated to exhibit the set optical properties. The user designs the laminated film based on this layer thickness distribution and refractive index distribution. Once the laminated film design is complete, the laminated film is manufactured based on that design.

[0073] In the embodiment described above, a trained model is generated by learning using layer thickness parameters that can be converted into a layer thickness distribution of the laminated film, and refractive index parameters that can be converted into a refractive index distribution. The laminated film is then designed using this trained model. According to this embodiment, the design of the optical layer with respect to optical properties can be performed with high accuracy.

[0074] (Variation 1) Next, a modified example 1 of this embodiment will be described. In this modified example 1, the method for calculating the layer thickness parameter and the refractive index parameter differs from that of the embodiment described above. The following describes the process that differs from the embodiment.

[0075] Figure 11 is a flowchart showing the flow of the refractive index parameter generation process performed by the learning device according to Modification 1. Below, we will describe an example in which multiple optical layers with different compositions are stacked in a regular manner. Furthermore, the layer thickness is assumed to increase / decrease gradually with respect to the stacking order.

[0076] The refractive index parameter generation unit 212 obtains the number of components and the regular arrangement as calculation conditions (step S111). The number of components indicates, for example, the number of different optical layers (types of optical layers). The regular arrangement indicates the regularity of the optical layers in the above number of components. For example, if a laminated film consists of three types of optical layers with different compositions, and the layers of the laminated film are classified as A layer, B layer, and C layer according to their composition, the laminated configurations can be (ABC)m, (ABCB)m, (ACBC)m, (BACA)m, (ABABC)m, (ACACB)m, (BCBCA)m, (ABCBCB)m, (ACBCBC)m, (BACACA)m, (BCACAC)m, (CABABA)m, (CBABAB)m, etc. Here, the regular arrangement corresponds to the arrangement in parentheses (the smallest repeating unit). m is, for example, an arbitrary natural number, and here it corresponds to the number of repetitions of the regular arrangement. The refractive index parameter generation unit 212 acquires information on the refractive index of each optical layer, such as the range of refractive indices that each optical layer can take, in addition to the number and regular arrangement of the components mentioned above.

[0077] Figure 12 is a diagram illustrating an example of information (calculation conditions table) related to the generation of refractive index parameters. Figure 12 shows the number of components, the regular arrangement, and the range of possible refractive indices for each optical layer in a laminated film. Here, optical layers are named according to their type, and are referred to as A layer, B layer, C layer, D layer, ... The number of components indicates the number of types of optical layers, and here it is the number of types of optical layers (A layer, B layer, C layer, D layer, ...) contained in each laminated film. The regular arrangement indicates the arrangement of optical layers in a regular pattern. The refractive index is the range of possible refractive indices for each optical layer (A layer, B layer, C layer, D layer, ...), and here it is shown as [minimum value, maximum value]. Note that the refractive index of non-existent optical layers (C layer and D layer if data number is 1) is set to [0,0].

[0078] If the refractive index distribution shown in Figure 9 is used directly as the refractive index parameter, the number of candidate parameters (e.g., regular arrangement patterns) increases exponentially as the number of components and layers increases, and the number of parameters also increases. In this modified example 1, the refractive index distribution is clustered to calculate the number of groups with different refractive indices, and this is used as an estimate of the number of components, thereby reducing the dimensionality of the refractive index distribution.

[0079] Returning to Figure 11, the refractive index parameter generation unit 212 groups the optical layers (step S112). The refractive index parameter generation unit 212 separates the various optical layers (layer A, layer B, layer C, layer D, ...) by type.

[0080] Then, the refractive index parameter generation unit 212 calculates the refractive index of the optical layer for each group (step S113). In this modified example 1, the refractive index parameter generation unit 212 calculates the refractive index of each layer using the least squares method. In this case, the refractive index parameter generation unit 212 calculates the refractive index of the target optical layer by using the refractive index of an optical layer other than the target optical layer (for example, if layer A is the target, an optical layer other than A (layer B, etc.)) as the minimum value.

[0081] The refractive index parameter generation unit 212 calculates the refractive index of various optical layers (layer A, layer B, layer C, layer D, ...) for data number 1. In this process, the refractive index parameter generation unit 212 calculates the refractive index for each optical layer within the range of possible refractive indices. The refractive index parameter generation unit 212 similarly calculates the refractive index for each data number. As a result, different refractive index distributions are obtained for each target optical layer for each data number, and furthermore, refractive index distributions are obtained for each target optical layer within the range of possible refractive indices.

[0082] After calculating the refractive index, the refractive index parameter generation unit 212 sets the refractive index parameters (step S114). Specifically, the refractive index parameter generation unit 212 calculates Y1 using the following equation (1), where Y1 is the difference between the refractive index (calculated value) calculated in step S113 and the refractive index (input value) obtained from the refractive index distribution in step S101. Y1 = Σ|refractive index (calculated value) - refractive index (input value)| ... (1)

[0083] The refractive index parameter generation unit 212 extracts the refractive index distribution that minimizes the deviation Y1, and uses the refractive index related to this refractive index distribution as the refractive index parameter. The refractive index parameter generation unit 212 extracts the refractive index distribution for each data number and uses it as the refractive index parameter. In this way, the refractive index of each optical layer is set for each data number. As a result, for each data number, a refractive index parameter consisting of the number of components, the regular arrangement, and the refractive index of each optical layer is set.

[0084] Note that the representation of the calculation conditions table is not limited to the format shown in Figure 12, where the refractive index is matched numerically. Figure 13 is a diagram illustrating another example of refractive index parameters. For example, as shown in Figure 13, the refractive index may be matched according to the number in the raw material table. In the calculation conditions table shown in Figure 13, the refractive index of each optical layer is indicated by a number. For example, number 0 corresponds to a refractive index of 0, number 1 corresponds to a refractive index of 1.1, number 2 corresponds to a refractive index of 1.2, and so on. When using this calculation conditions table, the refractive index parameter may be expressed as a number or as a converted refractive index.

[0085] Next, we will explain the generation of layer thickness parameters. Figure 14 is a flowchart showing the flow of the layer thickness parameter generation process performed by the learning device related to the modified example.

[0086] The layer thickness parameter generation unit 213 obtains the number of components and their orderly arrangement (step S121). The layer thickness parameter generation unit 213 obtains the number of components and their orderly arrangement from the refractive index parameters generated by the refractive index parameter generation unit 212.

[0087] Then, the layer thickness parameter generation unit 213 sets a periodic array (step S122). Here, the periodic array refers to the regularity of the stacking of optical layers, taking into account the difference in layer thickness in optical layers of the same component, in addition to the number of components mentioned above. As an example to explain the significance of the periodic array, Figure 15 shows the layer thickness distribution of a 400-layer laminated film consisting of two types of optical layers (layer A and layer B) with different compositions. The stacking configuration of this laminated film is (AB)m (where AB in parentheses is a regular array and m is a natural number representing the number of repetitions), and in Figure 15, the distribution L 10 This shows the thickness distribution of layer A at layer number 1 + 4n (where n is a non-negative integer), and the distribution L 11 This shows the thickness distribution of layer B at layer number 2+4n, and the distribution L 12 This shows the thickness distribution of layer A at layer number 3+4n, and distribution L 13 This shows the thickness distribution of layer B at layer number 4+4n. Distribution L 10 and distribution L 12The difference in layer thickness is approximately 100 nm, and the distribution L 11 and distribution L 13 The difference in layer thickness is approximately 100 nm. Thus, in order to generate training data for laminated films designed with optical layers that have different layer thickness distribution trends, even if they have the same optical components, a periodic array is set.

[0088] The layer thickness parameter generation unit 213 sets a periodic array that indicates the periodicity of the lamination based on the type of raw material and the layer thickness of the optical layer. For example, in the case of a laminated film with a regular arrangement of AB, possible candidates for setting the periodic array include (A1B1)m, (A1B1A2B2)m, (A1B1A1B2)m, (A1B1A2B1)m, and (A1B1A2B2A3B3)m. Here, A1, A2, and A3 are groups of optical layers in which the component A is the same, but the trend of the layer thickness distribution differs from A1 to A3, and B1, B2, and B3 are groups of optical layers in which the component B is the same, but the trend of the layer thickness distribution differs from B1 to B3.

[0089] The layer thickness parameter generation unit 213 groups the optical layers (step S123). The layer thickness parameter generation unit 213 separates the various optical layers (A layer, B layer, C layer, D layer, ...) by type based on a periodic array. Similar to the calculation of refractive index parameters, the layer thickness parameter generation unit 213 groups the layers according to a periodic array corresponding to the data number in the calculation conditions table. For example, if data number 2 (A1B1A2B2) in Figure 16 is obtained, the layer thickness parameter generation unit 213 divides the optical layers into four groups: the optical layer corresponding to the A1 layer, the optical layer corresponding to the B1 layer, the optical layer corresponding to the A2 layer, and the optical layer corresponding to the B2 layer.

[0090] After grouping, the layer thickness parameter generation unit 213 calculates coefficient values ​​(step S124). The layer thickness parameter generation unit 213 calculates coefficient values ​​that represent the layer thickness distribution of the optical layer for each group using mathematical optimization based on the layer thickness distribution.

[0091] Figure 16 is a diagram illustrating an example of information (coefficient values) related to the generation of layer thickness parameters. Through the coefficient value calculation process, data is generated for each optical layer of each data number, corresponding to the function name and coefficient value for each type of optical layer. Here, the coefficient values ​​are represented by coefficient 0, coefficient 1, coefficient 2, ... For example, in a linear function, coefficient 0 represents the intercept and coefficient 1 represents the slope. In this way, the parameters of the function are assigned to coefficient 0, coefficient 1, coefficient 2, ... respectively. Note that, for example, in a linear function, if there are no coefficients such as coefficient 2, ..., these coefficients are fixed to 0. When generating a layer thickness parameter using coefficient values ​​as shown in Figure 16, the layer thickness parameter will include a function type that indicates the thickness distribution of the optical layer, and a coefficient related to that function type.

[0092] Note that in Figure 16, the coefficient values ​​are set for each optical layer (layer A (A1 layer), layer B (B1 layer), layer C, layer D, ...), but this is not the only example. Figure 17 is a diagram illustrating another example of information (coefficient values) related to the generation of layer thickness parameters. As shown in Figure 17, the coefficient values ​​may be expressed as parameters that show the relationships within a periodic array, with one period of the periodic array as one unit, and the function for each unit expressed as an nth-degree function, a sine function, or an exponential function. When generating a layer thickness parameter using coefficient values ​​as shown in Figure 17, the layer thickness parameter will include a function type that represents the thickness distribution of multiple optical layers constituting a periodic array, and a coefficient related to the function type.

[0093] Returning to Figure 14, the layer thickness parameter generation unit 213 sets the layer thickness parameters (step S125). The layer thickness parameter generation unit 213 sets the coefficient values ​​of each data number calculated in step S124, the periodic array corresponding to the data number that minimizes the difference between the layer thickness distribution calculated from the periodic array set in step S122 and the layer thickness distribution obtained in step S101, and the coefficient values ​​of each group as layer thickness parameters. When the difference between the layer thickness (calculated value) obtained from the coefficient values ​​and periodic array calculated in step S124 and the layer thickness (input value) obtained from the layer thickness distribution in step S101 is Y2, the layer thickness parameter generation unit 213 calculates Y2 using the following equation (2). Y2 = Σ|Layer Thickness (Calculated Value) - Layer Thickness (Input Value)| ... (2) The layer thickness parameter generation unit 213 then extracts the layer thickness distribution that minimizes the displacement Y2 and sets the coefficient value related to this layer thickness distribution as the layer thickness parameter. The layer thickness parameter generation unit 213 generates the layer thickness parameter by setting the coefficient value for each optical layer.

[0094] In the layer thickness parameters of the laminated film obtained in this way, if there are groups with the same components and substantially equivalent layer thickness parameters for the optical layers, these groups may be considered to represent the same layer thickness distribution, and the layer thickness parameters may be regenerated. For example, if the periodic array is (A1B1A2B2)m, and the layer thickness parameters for the optical layers in A1 and A2 are equal, and the layer thickness parameters for the optical layers in B1 and B2 are equal, the periodic array may be set to (A1B1)m, and the layer thickness parameters may be regenerated.

[0095] As explained above, the refractive index parameter and layer thickness parameter are generated. FIG. 18 is a diagram for explaining dimensionality reduction of parameters. As shown in FIG. 18, refractive index parameters composed of the number of components, regular arrangement, and refractive index for each type of optical layer, and layer thickness parameters composed of the total number of layers, function, and coefficient values (coefficient 0, coefficient 1, coefficient 2, ···) for each type of optical layer are generated. For example, in the case of a 100-layer laminated film in which layers A and B are alternately laminated and the layer thickness distribution is expressed by a linear function (see FIG. 9), the refractive index distribution and the layer thickness distribution are each expressed by 100 data. However, in the example shown in FIG. 18, it is expressed by 5 data of the refractive index distribution from the total number of layers to the refractive index of layer B, and 7 data of the layer thickness distribution of 6 variables from the periodic arrangement to the function of the layer thickness distribution of layer B1 and the total number of layers (1 variable).

[0096] Here, the relationship between the number of explanatory variables based on the total number of optical layers and the prediction accuracy will be explained. The number of explanatory variables indicates the number of explanatory variables in the learning data whose dispersion in the column direction is not zero. Here, as an example, for a laminated film in which two types of optical layers are alternately laminated, learning data with dimensionality reduction was randomly generated according to the following conditions, and the prediction accuracy was confirmed from the output results of the trained model that learned this learning data. Using the trained model, explanatory variables that satisfy the desired optical performance (in this explanation, an average reflectance of 100% at wavelengths of 400 nm to 800 nm) were estimated, and the actual optical performance in the explanatory variables was obtained by optical calculation and used as the prediction accuracy. Also, the mean squared error (MSE) was used as the generalization error of the model to calculate the coefficient of determination R 2 was calculated. The generalization error was calculated by performing grid search cross-validation (Cross Validation: CV) under the following learning model and number of folds conditions. · Explanatory variables Layer thickness parameters Number of variables: Variable Value of variable: Generated so that the layer thickness is 0 nm or more Refractive index parameters Number of variables: Two Value of variable: Constant · Objective variable Optical performance Average reflectance at wavelengths of 400 nm to 800 nm • Number of training data points generated: 100,000 Grid search CV Learning models: Lasso regression, Ridge regression, Elastic Net, PLS regression, Support Vector Regression Number of folds: 5

[0097] Figure 19 illustrates the relationship between the number of explanatory variables and the actual optical performance (average reflectance (%)). Figure 19 also shows the coefficient of determination R in the training data. 2 Measurement points with a value of 0.6 or higher are shown as circles (●), and measurement points with a value less than 0.6 are shown as squares (◆). In the graph shown in Figure 19, it was found that the average reflectance increases sharply when the number of explanatory variables increases from 4 to 9. Thus, when the number of explanatory variables is less than 9, there is little room to adjust the layer thickness distribution and refractive index distribution of the laminated film, and there is a risk that a design that satisfies the desired optical performance cannot be found. Also, as shown in Figure 19, by keeping the number of explanatory variables to 102 or less, it can be said that a model with high generalization performance can be obtained regardless of the number of optical layers. On the other hand, when the number of explanatory variables exceeds 102, the coefficient of determination R 2 The value falls below 0.6, indicating that the generalization performance of the trained model is poor. In this result, the average reflectance is expressed by the refractive index distribution and the layer thickness distribution, so it is thought that a similar effect can be obtained even if the layer thickness parameter is fixed and the refractive index parameter is changed. Therefore, it is preferable that the number of explanatory variables in the training data be between 9 and 102.

[0098] In the modified example 1 described above, a trained model is generated by learning using layer thickness parameters that can be converted into a layer thickness distribution of the laminated film, and refractive index parameters that can be converted into a refractive index distribution. The laminated film is then designed using this trained model. According to this modified example 1, the design of the optical layer with respect to optical properties can be performed with high accuracy.

[0099] Furthermore, in Modification 1, dimensionality reduction is performed on the layer thickness parameter and refractive index parameter, so it is possible to generate a model with suppressed degradation in generalization performance for a trained model using the layer thickness parameter and refractive index parameter.

[0100] Furthermore, setting the function's order is crucial for reducing the number of parameters in generating layer thickness parameters. The setting of the function's order varies depending on the purpose of the trained model, but if you want to analyze the trained model and find design guidelines for layer thickness distributions that affect optical performance, you can determine the order based on the number of extreme values ​​of the layer thickness distribution from experimental data.

[0101] Furthermore, the refractive index parameter and the layer thickness parameter can be appropriately combined with those of the embodiment. For example, the refractive index parameter may be generated using the method according to Modification 1, and the layer thickness parameter may be generated using the method according to the embodiment.

[0102] (Modification 2) Next, a modified example 2 of this embodiment will be described. In this modified example 2, the method for estimating the layer thickness distribution and refractive index distribution differs from that of the embodiment described above. The following describes the process that differs from the embodiment. In this modified example 2, the learning unit 22 will be described using a trained model generated by learning with optical properties as explanatory variables and layer thickness parameters and refractive index parameters (see Figure 18) as objective variables.

[0103] Figure 20 is a flowchart illustrating the estimated value calculation process performed by the estimation device according to the modified example 2. The calculation unit 31 inputs the optical properties of the target to be estimated into the trained model generated by the learning device 2 and obtains the layer thickness parameter and refractive index parameter (step S211).

[0104] Then, the calculation unit 31 converts the layer thickness parameter and refractive index parameter output from the trained model into a layer thickness distribution and a refractive index distribution (step S212). This yields a layer thickness distribution and refractive index distribution that are estimated to exhibit the set optical properties. The user then designs the laminated film based on this layer thickness distribution and refractive index distribution.

[0105] In the modified example 2 described above, a trained model is generated by learning using layer thickness parameters that can be converted into a layer thickness distribution of the laminated film, and refractive index parameters that can be converted into a refractive index distribution. The laminated film is then designed using this trained model. According to this modified example 2, the design of the optical layer with respect to optical properties can be performed with high accuracy.

[0106] Furthermore, in Modification 2, by simply inputting the desired optical properties into a trained model, the refractive index parameter and layer thickness parameter that represent those optical properties can be obtained. Therefore, the design of the optical layer for the optical properties can be performed in fewer steps compared to the embodiment.

[0107] (Other embodiments) While embodiments for carrying out the present invention have been described so far, the present invention should not be limited to the embodiments described above. For example, the estimation device may also include the function of a learning unit. In this case, in addition to calculating the target variable to be estimated, the estimation device sequentially updates the trained model. Thus, the laminated film is obtained using at least one of the above-described training data generation method, trained model generation method, estimation method, training data generation program, trained model generation program, and estimation program. [Explanation of Symbols]

[0108] 1. Estimation System 2 Learning device 3 Estimation device 4 Display device 5 Input devices 21. Training Data Generation Unit 22 Learning Department 23, 32 Control Unit 24, 33 Storage section 31 Calculation Section 100 pre-trained models 211 Distribution acquisition part 212 Refractive Index Parameter Generation Unit 213 Layer Thickness Parameter Generation Unit 214 Characterization Unit 215 Dataset Generation Unit 241 Dataset Storage Unit 321 Display Control Unit

Claims

1. A method for generating training data for a computer to generate training data for estimating the laminated structure of a laminated film in which multiple optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction differ, A distribution acquisition step to acquire the refractive index distribution and layer thickness distribution of the laminated film, A refractive index parameter generation step generates refractive index parameters that can be interconverted with the refractive index distribution of the laminated film for the refractive index of each optical layer of the laminated film, A layer thickness parameter generation step that generates layer thickness parameters that can be interconverted between the layer thickness distribution of the laminated film and the layer thickness distribution of an optical layer made of the same type of material, A characteristic acquisition step to acquire the optical properties of a laminated film made according to the layer thickness parameter and the refractive index parameter, A learning data generation step that generates learning data comprising the layer thickness parameter and the refractive index parameter, and the optical properties of the laminated film produced by the layer thickness parameter and refractive index parameter, A method for generating training data that includes [the specified element].

2. The number of layers in the aforementioned plurality of optical layers is 51 or more. The method for generating training data according to claim 1.

3. The refractive index parameter is a dimensionally reduced parameter that is interconvertible with the refractive index distribution of the laminated film. The aforementioned layer thickness parameter is a dimensionally reduced parameter that is interconvertible with the layer thickness distribution of the laminated film. The method for generating training data according to claim 1.

4. The total number of variables for the layer thickness parameter and the refractive index parameter is between 9 and 102. The method for generating training data according to claim 1.

5. The aforementioned layer thickness parameter is a parameter that can represent the layer thickness distribution of the laminated film as a curve. The method for generating training data according to claim 1.

6. The refractive index parameter generation step is: A condition acquisition step to obtain the number of components indicating the number of types of raw materials constituting the resin layer of the laminated film, and the regular arrangement indicating the regularity of the lamination based on the types of raw materials, A classification step in which multiple optical layers are grouped based on the regular arrangement obtained in the condition acquisition step, A refractive index calculation step in which the refractive index of the optical layer of each group is calculated by the least squares method using the refractive index distribution, A refractive index parameter setting step in which the refractive index parameter is set to the number of components, the order of elements, and the refractive index of the optical layer of each group that minimize the difference between the refractive index distribution calculated from the refractive index distribution of each group and the refractive index distribution obtained in the distribution acquisition step, and the refractive index distribution of each group, calculated from the refractive index distribution of each group and the order of elements, A method for generating training data according to claim 1, including the following:

7. The layer thickness parameter generation step is: A condition acquisition step to obtain the number of components and regular arrangement based on the aforementioned refractive index parameter, A periodic array setting step, which sets a periodic array that indicates the periodicity of the lamination based on the type of material and layer thickness of the optical layer, A classification step of grouping multiple optical layers based on the set periodic arrangement, A coefficient value calculation step in which coefficient values ​​representing the thickness distribution of the optical layer for each group are calculated by mathematical optimization using the said thickness distribution, A layer thickness parameter setting step in which the coefficient values ​​of each group calculated and the layer thickness distribution calculated from the periodic array are set as layer thickness parameters, the periodic array that minimizes the difference between the layer thickness distribution obtained in the distribution acquisition step and the layer thickness distribution obtained in the distribution acquisition step, A method for generating training data according to claim 1, including the following:

8. The layer thickness parameter includes a function type that indicates the thickness distribution of the optical layer, and a coefficient related to the function type. The method for generating training data according to claim 7.

9. The layer thickness parameter includes a function type that indicates the thickness distribution of multiple optical layers constituting a regular arrangement, and a coefficient related to the function type. The method for generating training data according to claim 7.

10. The refractive index distribution and layer thickness distribution of the laminated film are generated based on data points of the constituent elements generated based on the calculation conditions. The method for generating training data according to claim 1.

11. Computers A method for generating a trained model to generate a trained model for estimating the laminated structure of a laminated film in which multiple optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction differ, A learning step to generate a trained model by learning, where the explanatory variables are a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film for the refractive index of each optical layer, and a layer thickness parameter that can be interconverted with the layer thickness distribution of the laminated film for the layer thickness distribution of optical layers made of the same material, and the objective variable is the optical properties of the laminated film produced by the layer thickness parameter and the refractive index parameter. A method for generating a pre-trained model that includes this.

12. A computer estimation method for estimating the laminated structure of a laminated film in which multiple optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction differ, Estimation step: Inputting the layer thickness parameter and refractive index parameter to be estimated into a trained model generated by learning, in which the refractive index of each optical layer is a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film, and the layer thickness parameter that can be interconverted with the layer thickness distribution of the laminated film for the layer thickness distribution of optical layers made of the same material is an explanatory variable, and the optical properties of the laminated film created by the layer thickness parameter and refractive index parameter are the objective variable, thereby obtaining the optical properties of the laminated film composed of the said layer thickness parameter and refractive index parameter. A selection step in which, from among the optical properties of the laminated film obtained by the estimation step, a layer thickness parameter and a refractive index parameter that exhibit the optimal optical properties are selected, A conversion step that converts the layer thickness parameter and refractive index parameter selected in the selection step into the layer thickness distribution and the refractive index distribution, An estimation method that includes [this].

13. A computer estimation method for estimating the laminated structure of a laminated film in which multiple optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction differ, An estimation step in which the optical properties to be estimated are input to a trained model generated by learning, in which the refractive index of each optical layer is a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film, and the thickness of the optical layers made of the same material is a thickness parameter that can be interconverted with the thickness of the laminated film, and the optical properties of the laminated film created by the thickness parameter and the refractive index parameter are used as independent variables, and the thickness parameter and refractive index parameter and corresponding thickness parameter and refractive index parameter are obtained. A conversion step that converts the layer thickness parameter and refractive index parameter obtained in the estimation step into the layer thickness distribution and the refractive index distribution, An estimation method that includes [this].

14. A training data generation program that causes a computer to generate training data for estimating the laminated structure of a laminated film in which multiple optical layers are stacked, wherein the optical properties of adjacent optical layers in the stacking direction are different. A distribution acquisition step to acquire the refractive index distribution and layer thickness distribution of the laminated film, A refractive index parameter generation step generates refractive index parameters that can be interconverted with the refractive index distribution of the laminated film for the refractive index of each optical layer of the laminated film, A layer thickness parameter generation step that generates layer thickness parameters that can be interconverted between the layer thickness distribution of the laminated film and the layer thickness distribution of an optical layer made of the same type of material, A characteristic acquisition step to acquire the optical properties of a laminated film made according to the layer thickness parameter and the refractive index parameter, A learning data generation step that generates learning data comprising the layer thickness parameter and the refractive index parameter, and the optical properties of the laminated film produced by the layer thickness parameter and refractive index parameter, A training data generation program that causes the aforementioned computer to execute.

15. A trained model generation program that causes a computer to generate a trained model for estimating the laminated structure of a laminated film in which multiple optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction differ, A learning step to generate a trained model by learning, where the explanatory variables are a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film for the refractive index of each optical layer, and a layer thickness parameter that can be interconverted with the layer thickness distribution of the laminated film for the layer thickness distribution of optical layers made of the same material, and the objective variable is the optical properties of the laminated film produced by the layer thickness parameter and the refractive index parameter. A trained model generation program that causes the aforementioned computer to execute.

16. An estimation program that causes a computer to estimate the laminated structure of a laminated film in which multiple optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction are different, Estimation step: Inputting the layer thickness parameter and refractive index parameter to be estimated into a trained model generated by learning, in which the refractive index of each optical layer is a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film, and the layer thickness parameter that can be interconverted with the layer thickness distribution of the laminated film for the layer thickness distribution of optical layers made of the same material is an explanatory variable, and the optical properties of the laminated film created by the layer thickness parameter and refractive index parameter are the objective variable, thereby obtaining the optical properties of the laminated film composed of the said layer thickness parameter and refractive index parameter. A selection step in which, from among the optical properties of the laminated film obtained by the estimation step, a layer thickness parameter and a refractive index parameter that exhibit the optimal optical properties are selected, An estimation program that causes the aforementioned computer to execute.

17. An estimation program that causes a computer to estimate the laminated structure of a laminated film in which multiple optical layers are stacked, wherein at least a portion of the optical properties of adjacent optical layers in the stacking direction are different, An estimation step in which the optical properties to be estimated are input to a trained model generated by learning, in which the refractive index of each optical layer is a refractive index parameter that can be interconverted with the refractive index distribution of the laminated film, and the thickness of the optical layers made of the same material is a thickness parameter that can be interconverted with the thickness of the laminated film, and the optical properties of the laminated film created by the thickness parameter and the refractive index parameter are used as independent variables, and the thickness parameter and refractive index parameter and corresponding thickness parameter and refractive index parameter are obtained. An estimation program that causes the aforementioned computer to execute.

18. A laminated film obtained using at least one of the training data generation method described in claim 1, the trained model generation method described in claim 11, the estimation method described in claim 12 or 13, the training data generation program described in claim 14, the trained model generation program described in claim 15, and the estimation program described in claim 16 or 17.

Images