Prediction method and prediction system

Abstract

To improve efficiency in design of material that is required to satisfy a plurality of characteristics.SOLUTION: A method according to an embodiment of the present invention is a method executed by a prediction system, and the method includes acquiring the composition of and a manufacturing condition for the material; and inputting the acquired composition of and the manufacturing condition for the material to a machine learning model that predicts a plurality of characteristics of the material from the composition of and the manufacturing condition for the material, to predict the plurality of characteristics of the material. The machine learning model is on the basis of the mixture of continuous probability distribution.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a prediction method and a prediction system.

Background Art

[0002] Conventionally, in material development, a method of actually prototyping a material and evaluating the characteristics of the material has been taken. Currently, a method of predicting the characteristics of a material using machine learning, called materials informatics, is also being taken.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] For example, in dental glass ceramics, a plurality of characteristics regarding color and transparency must be satisfied, and it is desired that the design of a material that requires satisfying a plurality of characteristics can be made more efficient.

[0005] Therefore, an object of the present invention is to improve the efficiency of designing a material that requires satisfying a plurality of characteristics.

Means for Solving the Problems

[0006] A method according to an embodiment of the present invention is a method executed by a prediction system, including: obtaining a composition and manufacturing conditions of a material; inputting the obtained composition and manufacturing conditions of the material into a machine learning model that predicts a plurality of characteristics of the material from the composition and manufacturing conditions of the material, and predicting the plurality of characteristics of the material, wherein the machine learning model is based on a mixture of continuous probability distributions.

Effects of the Invention

[0007] In the present invention, it is possible to improve the efficiency of designing materials that are required to satisfy a plurality of characteristics.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Embodiments for Carrying Out the Invention

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0010] <Explanation of Terms> · In this specification, "material" may be any material. For example, "material" is a medical material (e.g., dental material). For example, "material" is any one of ceramics, glass, glass ceramics, polymer materials, composite resins, glass ionomers, metals (e.g., dental ceramics, dental glass ceramics, dental polymer materials, dental composite resins, dental glass ionomers, dental metals, dental impression materials). · In this specification, "property" may be any property. For example, "property" is a property related to the color and transparency of a material. For example, the properties related to the color and transparency of a material include at least one of the values in the CIE Lab color system (e.g., L * value, a * value, b * value), total light transmittance (TT value), parallel light component (PT value), diffused (scattered) light component (DIF), Translucency Parameter (TP value), haze, Contrast Ratio (CR value). That is, as the properties related to the color and transparency of a material, all of the values in the CIE Lab color system (e.g., L * value, a * value, b * value), total light transmittance (TT value), parallel light component (PT value), diffused (scattered) light component (DIF), Translucency Parameter (TP value), haze, Contrast Ratio (CR value) may be used, or some of them may be used.

[0011] <Overall Configuration> FIG. 1 is a diagram showing the overall configuration according to an embodiment of the present invention. The user 30 operates the prediction device 10 and the learning device 20. In FIG. 1, the prediction device 10 and the learning device 20 are described as separate devices, but the prediction device 10 and the learning device 20 may be implemented as one device.

[0012] <<Prediction Device>> The prediction device 10 is a device that predicts a plurality of characteristics of a material, as well as the composition and manufacturing conditions of the material (e.g., heat treatment conditions (e.g., heating rate, holding temperature, holding time, slow cooling rate, slow cooling time)). The prediction device 10 is composed of one or more computers. The prediction device 10 can transmit and receive data with the learning device 20 via an arbitrary network.

[0013] Specifically, the prediction device 10 can predict a plurality of characteristics of the material from the composition and manufacturing conditions of the material. Also, the prediction device 10 can predict the composition and manufacturing conditions of the material from a plurality of characteristics of the material. Note that the device for predicting a plurality of characteristics of the material and the device for predicting the composition and manufacturing conditions of the material may be the same device or different devices.

[0014] <<Learning Device>> The learning device 20 is a device that generates a machine learning model for predicting a plurality of characteristics of a material from the composition and manufacturing conditions of the material. The learning device 20 is composed of one or more computers. The learning device 20 can transmit and receive data with the prediction device 10 via an arbitrary network.

[0015] <Functional Block> Hereinafter, the functional blocks of the prediction device 10 will be described with reference to FIG. 2, and the functional blocks of the learning device 20 will be described with reference to FIG. 3.

[0016] [Prediction Device] FIG. 2 is a functional block diagram of a prediction device 10 according to an embodiment of the present invention. The prediction device 10 can include a composition / manufacturing condition acquisition unit 101, a property acquisition unit 102, a prediction unit 103, a visualization unit 104, and a machine learning model storage unit 105. Further, the prediction device 10 can function as the composition / manufacturing condition acquisition unit 101, the property acquisition unit 102, the prediction unit 103, and the visualization unit 104 by executing a program.

[0017] The composition / manufacturing condition acquisition unit (simply referred to as the acquisition unit) 101 acquires the composition of the material and the manufacturing conditions. For example, the composition / manufacturing condition acquisition unit 101 acquires the composition of the material and the manufacturing conditions input by the user 30 to the prediction device 10 or the like.

[0018] The property acquisition unit (simply referred to as the acquisition unit) 102 acquires a plurality of properties of the material. For example, the property acquisition unit 102 acquires a plurality of properties of the material input by the user 30 to the prediction device 10 or the like.

[0019] Note that the property acquisition unit 102 may acquire a plurality of properties of the material by acquiring a designation of a region in a space representing a plurality of properties of the material (for example, a designation of a region on a screen as shown in FIG. 14) and specifying a plurality of properties of the material corresponding to the region.

[0020] [Properties of the material] Here, the properties of the material will be described. The plurality of properties of the material may be a plurality of parameters (for example, hue, lightness, chroma) for determining one property (for example, color) of the material, or a plurality of parameters (for example, parameters for determining color and parameters for determining transparency) for determining a plurality of properties (for example, color and transparency) of the material.

[0021] For example, the "properties" are properties related to the color and transparency of the material. For example, the properties related to the color and transparency of the material are values in the CIE Lab color space system (specifically, the L * value, a * value, b *The value of L when the background of the material is black * The value of a * The value of b * (Values), there are eight in total, namely the total light transmittance (TT value) and the Translucency Parameter (TP value).

[0022] The prediction unit 103 predicts a plurality of characteristics of the material, as well as the composition and manufacturing conditions of the material. Note that the prediction unit 103 may output (for example, display on the prediction device 10, etc., transmit data to other devices, etc.) the predicted plurality of characteristics of the material and the predicted composition and manufacturing conditions of the material.

[0023] Hereinafter, it will be described in detail by dividing it into [Prediction of a plurality of characteristics of the material] and [Prediction of the composition and manufacturing conditions of the material].

[0024] [Prediction of a plurality of characteristics of the material] The prediction unit 103 inputs the composition and manufacturing conditions of the material acquired by the composition and manufacturing condition acquisition unit 101 into the machine learning model stored in the machine learning model storage unit 105 (specifically, a machine learning model that predicts a plurality of characteristics of the material from the composition and manufacturing conditions of the material), and predicts a plurality of characteristics of the material.

[0025] In this way, a plurality of characteristics of the material manufactured according to the composition and manufacturing conditions specified by the user 30, which are obtained (predicted by the prediction device 10) without actually prototyping the material, can be used for material design.

[0026] [Prediction of the composition and manufacturing conditions of the material] The prediction unit 103 uses the machine learning model stored in the machine learning model storage unit 105 (specifically, a machine learning model that predicts a plurality of characteristics of the material from the composition and manufacturing conditions of the material) for inverse analysis, and predicts the composition and manufacturing conditions of the material from the plurality of characteristics of the material acquired by the characteristic acquisition unit 102.

[0027] In this way, the composition and manufacturing conditions of the material for realizing the plurality of characteristics specified by the user 30, which are obtained (predicted by the prediction device 10) without actually prototyping the material, can be used for material design.

[0028] [Inverse Analysis Using Machine Learning Model] Here, the inverse analysis using a machine learning model will be described. In the inverse analysis using a machine learning model (specifically, a machine learning model that predicts a plurality of characteristics of a material from the composition and manufacturing conditions of the material), the composition and manufacturing conditions of the material that satisfy a plurality of characteristics of a predetermined material (that is, the input values of the machine learning model) are searched for.

[0029] The visualization unit 104 visualizes a plurality of characteristics of each material by arranging each material in a space representing a plurality of characteristics of the material. The space representing a plurality of characteristics of the material is a space of the number of dimensions after dimensionality reduction (principal component analysis) of a plurality of characteristics of the material (for example, eight characteristics) (for example, a two-dimensional space). Note that the space representing a plurality of characteristics of the material may be a two-dimensional space or a three-dimensional space.

[0030] The machine learning model storage unit 105 stores a machine learning model (specifically, a machine learning model that predicts a plurality of characteristics of a material from the composition and manufacturing conditions of the material) generated by the learning device 20.

[0031] [Learning Device] FIG. 3 is a functional block diagram of a learning device 20 according to an embodiment of the present invention. The learning device 20 includes a learning data acquisition unit 201, a learning unit 202, and a machine learning model storage unit 203. Further, the learning device 20 functions as the learning data acquisition unit 201 and the learning unit 202 by executing a program.

[0032] The learning data acquisition unit 201 acquires learning data used when generating a machine learning model (specifically, a machine learning model that predicts a plurality of characteristics of a material from the composition and manufacturing conditions of the material). Specifically, the learning data acquisition unit 201 acquires the composition and manufacturing conditions of the material, and a plurality of characteristics (that is, measured values) of the material actually manufactured (prototyped, etc.) under the composition and manufacturing conditions.

[0033] The learning unit 202 performs machine learning using the learning data acquired by the learning data acquisition unit 201, and generates a machine learning model for predicting a plurality of characteristics of the material from the composition and manufacturing conditions of the material.

[0034] Specifically, when the composition and manufacturing conditions of the material and a plurality of characteristics of the material are input into the model, the learning unit 202 estimates parameters using the EM algorithm (expectation maximization algorithm) based on the input data. More specifically, the model is trained by estimating each parameter so as to maximize the log-likelihood function.

[0035] The machine learning model storage unit 203 stores the machine learning model generated by the learning unit 202 (specifically, a machine learning model for predicting a plurality of characteristics of the material from the composition and manufacturing conditions of the material).

[0036] <Machine learning model> Here, the machine learning model will be described. In the present invention, the method of the mixture distribution model is used. The machine learning model used in the present invention is based on a mixture of continuous probability distributions (specifically, a regression model in which a plurality of multi-dimensional (multivariate) continuous probability distributions are mixed). That is, in the machine learning model used in the present invention, the continuous probability distributions having the same number of dimensions as the number of characteristics of the material are mixed by the number of clusters when the data of the characteristics are clustered. As a result, a probability density distribution of the data is obtained, and this is used as a regression model.

[0037] Hereinafter, as an example of a regression model in which a plurality of multi-dimensional continuous probability distributions are mixed, a mixture Gaussian regression model (hereinafter also referred to as GMR) and a mixture Student's t regression model (hereinafter also referred to as SMR) will be described. As will be described later, the mixture Student's t regression model has higher prediction accuracy than the mixture Gaussian regression model.

[0038] The following shows a multi-dimensional Gaussian distribution (normal distribution) and its mixture.

[0039] [Mathematics]

[0040] The following shows the multivariate Student's t-distribution and its mixture.

[0041] [Mathematics]

[0042] μ: Mean vector Σ: Covariance matrix Λ: Precision matrix ν: Degrees of freedom.

[0043] Note that the mixing ratio is as follows for both the multivariate Gaussian distribution and the multivariate Student's t-distribution.

[0044] [Mathematics]

[0045] In the machine learning model, · SMR Parameters: μ, Λ, π, ν (degrees of freedom) Hyperparameters: Number of Student's t-distributions, types of precision matrix (Λ) · GMR Parameters: μ, Σ, π Hyperparameters: Number of normal distributions, types of covariance matrix (Σ) That is.

[0046] <Processing method> Hereinafter, the prediction processing of a plurality of characteristics will be described with reference to FIG. 4, the prediction processing of the composition and manufacturing conditions will be described with reference to FIG. 5, and the learning processing will be described with reference to FIG. 6.

[0047] FIG. 4 is a flowchart of the prediction processing of a plurality of characteristics according to an embodiment of the present invention.

[0048] In step 11 (S11), the composition and manufacturing condition acquisition unit 101 of the prediction device 10 acquires the composition of the material and the manufacturing conditions.

[0049] In step 12 (S12), the prediction unit 103 of the prediction device 10 inputs the composition of the material and the manufacturing conditions acquired in S11 into a machine learning model that predicts a plurality of characteristics of the material from the composition of the material and the manufacturing conditions, and predicts a plurality of characteristics of the material.

[0050] Note that thereafter, the prediction unit 103 of the prediction device 10 may output the predicted characteristics of the material replicated in S12 (for example, display on the prediction device 10 or the like, transmit data to other devices, etc.).

[0051] FIG. 5 is a flowchart of the prediction process for the composition and manufacturing conditions according to an embodiment of the present invention.

[0052] In step 21 (S21), the characteristic acquisition unit 102 of the prediction device 10 acquires a plurality of characteristics of the material.

[0053] In step 22 (S22), the prediction unit 103 of the prediction device 10 uses a machine learning model that predicts a plurality of characteristics of the material from the composition of the material and the manufacturing conditions for inverse analysis, and predicts the composition of the material and the manufacturing conditions from the plurality of characteristics of the material acquired in S21.

[0054] Note that thereafter, the prediction unit 103 of the prediction device 10 may output the composition of the material and the manufacturing conditions predicted in S22 (for example, display on the prediction device 10 or the like, transmit data to other devices, etc.).

[0055] FIG. 6 is a flowchart of the learning process according to an embodiment of the present invention.

[0056] In step 31 (S31), the learning data acquisition unit 201 of the learning device 20 acquires learning data used when generating a machine learning model (specifically, a machine learning model that predicts a plurality of characteristics of the material from the composition of the material and the manufacturing conditions).

[0057] In step 32 (S32), the learning unit 202 of the learning device 20 performs machine learning using the learning data acquired in S31, and generates a machine learning model for predicting a plurality of characteristics of the material from the composition of the material and the manufacturing conditions.

[0058] <Comparison between SMR and GMR> Hereinafter, the comparison results between the mixed student t-regression model (SMR) and the mixed Gaussian regression model (GMR) will be described with reference to FIGS. 7 and 8.

[0059] FIG. 7 shows the comparison results between the mixed student t-regression model (SMR) and the mixed Gaussian regression model (GMR) according to an embodiment of the present invention. In FIG. 7, the values of the prediction accuracy of SMR and GMR (using two accuracy indicators of RMSE (Root Mean Squared Error) and MAE (Mean Absolute Error)) are shown.

[0060] Specifically, the L value of the white background, the a value of the white background, the b value of the white background, the L value of the black background, the a value of the black background, the b value of the black background, the TT value, and the TP value, which are the characteristics of the material, are shown for each value of the accuracy index when predicted. Note that Data: Train indicates the case of prediction using learning data, Data: Train Y-random indicates the case of prediction using data obtained by randomly combining the input data and output data of the learning data, and Data: Test indicates the case of prediction using test data. * value, the a value of the white background * value, the b value of the white background * value, the L value of the black background * value, the a value of the black background * value, the b value of the black background * value, and the values of the accuracy indicators for each of the TT value and the TP value when predicted. Note that Data: Train indicates the case of prediction using learning data, Data: Train Y-random indicates the case of prediction using data obtained by randomly combining the input data and output data of the learning data, and Data: Test indicates the case of prediction using test data.

[0061] As shown in FIG. 7, in both the case of RMSE and the case of MAE, the prediction accuracy of the mixed student t-regression model (SMR) was higher than that of the mixed Gaussian regression model (GMR).

[0062] Figure 8 shows the comparison results between the mixed Student's t regression model (SMR) and the mixed Gaussian regression model (GMR) according to an embodiment of the present invention. In Figure 8, a scatter diagram of the predicted values (values obtained by the prediction of the prediction device 10, Estimation) and the experimental values (values obtained by the experiment, Experiment) of the SMR and the GMR is shown.

[0063] Specifically, the L value of the white background, the a value of the white background, the b value of the white background, the L value of the black background, the a value of the black background, the b value of the black background, the TT value, and the TP value, which are the characteristics of the material, are shown in the scatter diagram of the predicted value (Estimation) and the experimental value (Experiment). * value, a value of the white background * value, b value of the white background * value, L value of the black background * value, a value of the black background * value, b value of the black background * value, TT value, TP value, and the scatter diagram of the predicted value (Estimation) and the experimental value (Experiment).

[0064] As shown in Figure 8, the mixed Student's t regression model (SMR) had a higher degree of agreement between the predicted value (Estimation) and the experimental value (Experiment) than the mixed Gaussian regression model (GMR).

[0065] From the above, it is preferable to use the mixed Student's t regression model (SMR) with high prediction accuracy. In the SMR using the t-distribution, not only can it cope with the variation (dispersion) of the data, but also because the tails of the distribution are thick, it is robust against data with many outliers (noise).

[0066] [Verification test] Figure 9 is a flowchart of the verification test process of the mixed Student's t regression model (SMR) according to an embodiment of the present invention. Note that · Explanatory variables: Glass composition, heat treatment conditions · Objective variables: L values, a values, b values, TT values, and TP values of the white background and the black background * value, a * value, b * value, TT value, TP value were used.

[0067] In step 1 (S1), the target color tone (value range) is specified (for example, on the screen as shown in Figure 14).

[0068] In step 2 (S2), by performing inverse principal component analysis on the target color tone (value range) of S1, target values of the material properties (specifically, target colorimetric values (L * value, a * value, b * value), TT value, TP value) are obtained.

[0069] In step 3 (S3), by performing inverse analysis on the target values obtained in S2 using a machine learning model, candidates for the glass composition and heat treatment conditions are obtained.

[0070] In step 4 (S4), by performing forward analysis on the candidates for the glass composition and heat treatment conditions obtained in S3 using a machine learning model, predicted values of the material properties are obtained.

[0071] In step 5 (S5), glass ceramics are prototyped using the candidates for the glass composition and heat treatment conditions obtained in S3, and colorimetric values (L * value, a * value, b * value), TT value, and experimental values of the TP value are obtained.

[0072] FIG. 10 shows a comparison result between target values and experimental values according to an embodiment of the present invention. In FIG. 10, a scatter diagram of the target values (target values of S2 in FIG. 9, Target) and experimental values (experimental values of S5 in FIG. 9, Experiment) of the verification test of the mixed Student's t regression model (SMR) in FIG. 9 is shown.

[0073] Specifically, a scatter diagram of the target values (target values of S2 in FIG. 9, Target) and experimental values (experimental values of S5 in FIG. 9, Experiment) of the colorimetric values (L * value on white background, a * value on white background, b * value on white background, L * value on black background, a * value on black background, b * value on black background, TT value, TP value), which are the material properties, is shown.

[0074] As shown in FIG. 10, in each case, the target value and the experimental value are generally the same.

[0075] FIG. 11 shows the comparison result between the predicted value and the experimental value according to an embodiment of the present invention. In FIG. 11, a scatter diagram of the predicted value (Estimation, the predicted value in S4 of FIG. 9) and the experimental value (Experiment, the experimental value in S5 of FIG. 9) in the verification test of the mixed Student t regression model (SMR) of FIG. 9 is shown.

[0076] Specifically, it shows the scatter diagram of the predicted value (Estimation, the predicted value in S4 of FIG. 9) and the experimental value (Experiment, the experimental value in S5 of FIG. 9) of the L value of the white background, the a value of the white background, the b value of the white background, the L value of the black background, the a value of the black background, the b value of the black background, the TT value, and the TP value, which are the characteristics of the material. * value, the a * value of the white background, the b * value of the white background, the L * value of the black background, the a * value of the black background, the b * value of the black background, the predicted value (Estimation, the predicted value in S4 of FIG. 9) and the experimental value (Experiment, the experimental value in S5 of FIG. 9) of the TT value and the TP value.

[0077] As shown in FIG. 11, in each case, the predicted value and the experimental value are generally the same.

[0078] The glass composition and heat treatment conditions candidates obtained in S3 are actually prototyped, and the experimental values of the characteristics of the obtained material and the data consisting of the glass composition and heat treatment conditions candidates obtained in S3 are added to the learning data acquisition unit 201 as learning data and can be used for the learning of the machine learning model. Thereby, the performance of the machine learning model can be improved.

[0079] <Visualization of Material Characteristics> Hereinafter, the visualization of the material characteristics will be described. Note that the prediction device 10 can visualize at least one of the characteristics of the already prototyped material and the characteristics of the material predicted by the prediction device 10 (that is, both the characteristics of the already prototyped material and the characteristics of the material predicted by the prediction device 10 may be used, or only one of them may be used).

[0080] FIG. 12 is an example of a screen visualizing the characteristics of the material according to an embodiment of the present invention.

[0081] The prediction device 10 estimates a number of properties of a material (in the example of FIG. 12, the L * Value, a on white background * Value, b on white background * Value, L on black background * Value, a on black background * Value, b on black background * A space (i.e., a space representing multiple characteristics of the material) is generated with the axis being the result of reducing the dimension of the matrix (PC1 and PC2 in the example of FIG. 12) of the matrix (PC value, TT value, and TP value) by principal component analysis. Then, the prediction device 10 arranges each material (materials A to D in the example of FIG. 12) in the space representing the multiple characteristics of the material. That is, points representing each material (materials A to D in the example of FIG. 12) are plotted in the space.

[0082] L in Figure 12 * The arrow of _w is L on a white background * The axis of the value (i.e., the L * (indicating the direction of increasing values), a * _w arrow is a on white background * The axis of values ​​(i.e. a * (indicating the direction of increasing values), b * _w arrow is b on white background * The axis of the value (i.e., the white background b * (indicating the direction in which the value increases) and L * The arrow in _b is an L on a black background * The axis of the value (i.e., L on a black background) * (indicating the direction of increasing values), a * The arrow of _b is a on a black background * The axis of the value (i.e., a * (indicating the direction of increasing values), b * The _b arrow is a b on a black background * The axis of the value (i.e., b on a black background) * The TT arrow is the axis of the TT value (i.e., showing the direction in which the TT value increases), and the TP arrow is the axis of the TP value (i.e., showing the direction in which the TP value increases).

[0083] FIG. 13 is an example of a screen that visualizes the properties of a material according to one embodiment of the present invention.

[0084] As in FIG. 12, the prediction device 10 predicts a number of properties of the material (in the example of FIG. 13, the L * Value, a on white background * Value, b on white background * Value, L on black background * Value, a on black background * Value, b on black background * A space (i.e., a space representing multiple characteristics of the material) is generated with the axis being the result of reducing the dimension of the saturation temperature (T1 value, TT value, TP value) by principal component analysis (PC1 and PC2 in the example of FIG. 13). Then, the prediction device 10 arranges each material (in the example of FIG. 13, material a (heat treatment 760°C), material a (heat treatment 860°C), material b (heat treatment 760°C), material b (heat treatment 860°C)) in the space representing multiple characteristics of the material. In other words, points representing each material (in the example of FIG. 13, material a (heat treatment 760°C), material a (heat treatment 860°C), material b (heat treatment 760°C), material b (heat treatment 860°C)) are plotted in the space.

[0085] As in Figure 12, L * The arrow of _w is L on a white background * The axis of the value (i.e., the L * (indicating the direction of increasing values), a * _w arrow is a on white background * The axis of values ​​(i.e. a * (indicating the direction of increasing values), b * _w arrow is b on white background * The axis of the value (i.e., the white background b * (indicating the direction in which the value increases) and L * The arrow in _b is an L on a black background * The axis of the value (i.e., L on a black background) * (indicating the direction of increasing values), a * The arrow of _b is a on a black background * The axis of the value (i.e., a * (indicating the direction of increasing values), b * The _b arrow is a b on a black background * The axis of the value (i.e., b on a black background) *The arrow of TT indicates the direction in which the value increases (i.e., it indicates the axis of the TT value, that is, the direction in which the TT value increases), and the arrow of TP is the axis of the TP value (that is, it indicates the direction in which the TP value increases).

[0086] In FIG. 13, each material (i.e., material a (heat treatment at 760° C.), material a (heat treatment at 860° C.), material b (heat treatment at 760° C.), material b (heat treatment at 860° C.)) is arranged at a position corresponding to the number of times the heat treatment is performed. For example, before the heat treatment, after the first heat treatment (in the example of FIG. 13, the heat treatment at 760° C. or the heat treatment at 860° C.), after the second heat treatment, and after the third heat treatment, points representing each material are arranged at the positions of the characteristics. In this way, it is possible to visualize the characteristics of the material processed according to the manufacturing conditions (e.g., heat treatment conditions (e.g., heating rate, holding temperature, holding time, slow cooling rate, slow cooling time)).

[0087] FIG. 14 is a diagram for explaining the specification of characteristics on a screen visualizing the characteristics of a material according to an embodiment of the present invention. The user 30 can specify characteristics on a screen visualizing the characteristics of a material as shown in FIG. 14 displayed on the prediction device 10 or the like. The prediction device 10 acquires the specification of an area on a screen like FIG. 14 and identifies a plurality of characteristics of the material corresponding to the area. Specifically, the prediction device 10 calculates the L * value of the white background, the a * value of the white background, the b * value of the white background, the L * value of the black background, the a * value of the black background, the b * value of the black background, the TT value, and the TP value by performing inverse principal component analysis (principal component analysis of a plurality of characteristics of the material) on PC1 and PC2 of the specified area.

[0088] <Effect> · In one embodiment of the present invention, the user 30 can easily know a plurality of characteristics of a material manufactured according to the composition and manufacturing conditions specified by the user 30. · In one embodiment of the present invention, the user 30 can easily know the composition and manufacturing conditions for realizing a plurality of characteristics of the material specified by the user 30. Further, the user 30 can easily specify a plurality of desired characteristics on the screen where the characteristics of the material are visualized.

[0089] <Hardware Configuration> FIG. 15 is a hardware configuration diagram of the prediction device 10 and the learning device 20 according to one embodiment of the present invention. The prediction device 10 and the learning device 20 include a CPU (Central Processing Unit) 1001, a ROM (Read Only Memory) 1002, and a RAM (Random Access Memory) 1003. The CPU 1001, the ROM 1002, and the RAM 1003 form a so-called computer. Further, the prediction device 10 and the learning device 20 may include an auxiliary storage device 1004, a display device 1005, an operation device 1006, an I / F (Interface) device 1007, and a drive device 1008. Each hardware of the prediction device 10 and the learning device 20 is interconnected via a bus B.

[0090] The CPU 1001 is an arithmetic device that executes various programs installed in the auxiliary storage device 1004. By executing the program by the CPU 1001, each process described in this specification is performed.

[0091] The ROM 1002 is a non-volatile memory. The ROM 1002 functions as a main memory device that stores various programs, data, etc. necessary for the CPU 1001 to execute various programs installed in the auxiliary storage device 1004. Specifically, the ROM 1002 functions as a main memory device that stores boot programs such as BIOS (Basic Input / Output System) and EFI (Extensible Firmware Interface).

[0092] The RAM 1003 is a volatile memory such as DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory). The RAM 1003 functions as a main memory device that provides a working area where various programs installed in the auxiliary storage device 1004 are expanded when executed by the CPU 1001.

[0093] The auxiliary storage device 1004 is an auxiliary storage device that stores various programs and information used when the various programs are executed.

[0094] The display device 1005 is a display device that displays the internal states of the prediction device 10, the learning device 20, etc.

[0095] The operation device 1006 is an input device through which an operator of the prediction device 10 and the learning device 20 inputs various instructions to the prediction device 10 and the learning device 20.

[0096] The I / F device 1007 is a communication device that connects to a network and communicates with other devices.

[0097] The drive device 1008 is a device for setting the storage medium 1009. The storage medium 1009 here includes media that optically, electrically, or magnetically record information, such as CD (Compact Disk)-ROM, flexible disks, magneto-optical disks, etc. Further, the storage medium 1009 may include semiconductor memories that electrically record information, such as ROM, flash memory, etc.

[0098] Note that various programs installed in the auxiliary storage device 1004 are installed, for example, when the distributed storage medium 1009 is set in the drive device 1008 and various programs recorded on the storage medium 1009 are read by the drive device 1008. Alternatively, various programs installed in the auxiliary storage device 1004 may be installed by being downloaded from a network via the I / F device 1007.

[0099] As described above, the embodiments of the present invention have been described in detail. However, the present invention is not limited to the specific embodiments described above, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims.

Explanation of Reference Numerals

[0100] 1 Prediction system 10 Predictor 20 Learning device 30 User 101 Composition and manufacturing condition acquisition unit 102 Characteristic acquisition unit 103 Prediction unit 104 Visualization unit 105 Machine learning model storage unit 201 Learning data acquisition unit 202 Learning unit 203 Machine learning model storage unit 1001 CPU 1002 ROM 1003 RAM 1004 Auxiliary storage device 1005 Display device 1006 Operating device 1007 I / F device 1008 Drive device 1009 Storage medium

Claims

1. A method for a prediction system to execute, comprising: obtaining a composition of a material and manufacturing conditions; inputting the obtained composition of the material and the manufacturing conditions into a machine learning model that predicts a plurality of characteristics of the material from the composition of the material and the manufacturing conditions, and predicting the plurality of characteristics of the material; wherein the machine learning model is based on a mixture of continuous probability distributions.

2. A method for a prediction system to execute, comprising: obtaining a plurality of characteristics of a material; using, in inverse analysis, a machine learning model that predicts a plurality of characteristics of the material from the composition of the material and the manufacturing conditions, and predicting the composition of the material and the manufacturing conditions from the obtained plurality of characteristics of the material; wherein the machine learning model is based on a mixture of continuous probability distributions.

3. The method according to claim 2, further comprising visualizing a plurality of characteristics of each material by arranging each material in a space representing the plurality of characteristics of the material.

4. The method according to claim 3, wherein the space is a space having the number of dimensions after dimensionality reduction of the plurality of characteristics of the material.

5. The method according to claim 3, wherein obtaining the plurality of characteristics of the material is obtaining a designation of a region in the space and specifying the plurality of characteristics of the material corresponding to the region.

6. The method according to claim 1 or 2, wherein the machine learning model is a mixture Student t regression model.

7. The method according to claim 1 or 2, wherein the characteristics are characteristics related to the color and transparency of the material.

8. The characteristics related to color and transparency include at least one of values in the CIE Lab color system, total light transmittance, parallel light component, diffused (scattered) light component, Translucency Parameter (TP value), haze, and Contrast Ratio (CR value), according to the method of claim 1 or 2.

9. The characteristics regarding color and transparency are L ＊ value, a ＊ value, b ＊ value, total light transmittance, parallel light component, diffused (scattered) light component, Translucency Parameter (TP value), haze, Contrast Ratio (CR value), the method according to claim 1 or 2.

10. The method according to claim 1 or 2, wherein the material is a glass ceramic.

11. The method according to claim 1 or 2, wherein the manufacturing conditions are heat treatment conditions.

12. The method according to claim 1 or 2, wherein the material is a dental material.

13. an acquisition unit that acquires a composition of a material and manufacturing conditions; a prediction unit that inputs the obtained composition of the material and the manufacturing conditions into a machine learning model that predicts a plurality of characteristics of the material from the composition of the material and the manufacturing conditions, and predicts the plurality of characteristics of the material. The prediction system in which the machine learning model is based on a mixture of continuous probability distributions.

14. An acquisition unit that acquires a plurality of characteristics of a material; A prediction unit that uses a machine learning model for predicting a plurality of characteristics of the material from the composition and manufacturing conditions of the material by inverse analysis, and predicts the composition and manufacturing conditions of the material from the plurality of characteristics of the acquired material; and The prediction system in which the machine learning model is based on a mixture of continuous probability distributions.

Images