WO2024070169A1 - 試作条件提案システム、試作条件提案方法 - Google Patents
試作条件提案システム、試作条件提案方法 Download PDFInfo
- Publication number
- WO2024070169A1 WO2024070169A1 PCT/JP2023/027145 JP2023027145W WO2024070169A1 WO 2024070169 A1 WO2024070169 A1 WO 2024070169A1 JP 2023027145 W JP2023027145 W JP 2023027145W WO 2024070169 A1 WO2024070169 A1 WO 2024070169A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- regression model
- prototype
- measurement data
- weighting
- characteristic measurement
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N99/00—Subject matter not provided for in other groups of this subclass
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Definitions
- the present invention relates to a system and method for proposing material prototyping conditions to material developers.
- Patent Document 1 discloses a system that estimates the production conditions for a material with optimal physical properties and structure from a data set that includes the production conditions for each of multiple substances used as samples, and material information that represents the physical properties and structure of each substance.
- the characteristics of the material to be developed are first measured through various evaluation tests, and data representing the measurement results (hereinafter referred to as “measured characteristic data”).
- the obtained measured characteristic data is then input into a computer to construct various trained machine learning models, and the conditions for prototyping the material (hereinafter referred to as “prototype conditions”) are estimated using the trained machine learning models.
- the accuracy with which the computer estimates the prototyping conditions (hereinafter referred to as “estimation accuracy”) is determined by the accuracy of the trained machine learning model used in the process of estimating the prototyping conditions.
- the accuracy of this trained machine learning model is largely determined by the number of measured characteristic data (hereinafter also referred to as "number of samples”) that were trained when the machine learning model was constructed, the degree of distribution variability, etc.
- the characteristic measurement data (hereinafter also referred to as "learning data”) input to a computer to construct a machine learning model generally has the characteristic that it has an extremely large number of explanatory variables, but on the other hand, the number of samples is often small. If the amount of training data is small or if the distribution of the training data is biased, the computer cannot construct a highly accurate machine learning model. Therefore, in such cases, even if the computer is made to execute a process to estimate prototype conditions using the constructed machine learning model, there is a risk that sufficient estimation accuracy cannot be obtained.
- the present invention aims to provide a technology that can accurately suggest optimal prototyping conditions for materials even when the amount of training data used to build a machine learning model is small or when the distribution of the training data is biased.
- the prototype condition proposal system proposes prototype conditions for a material to a material developer, and includes a regression model construction processing unit and a prototype condition proposal processing unit.
- the regression model construction processing unit executes a regression model construction process for property measurement data representing actual measurement results of the material's properties.
- the prototype condition proposal processing unit searches for optimal prototype conditions for the material using the constructed regression model, and executes a prototype condition proposal process based on the search results.
- the regression model construction process includes a process of calculating a weighting criterion that is a criterion for weighting the property measurement data, and a process of weighting the property measurement data based on the calculated weighting criterion.
- a prototype condition proposing method is a method of proposing prototype conditions for a material to a material developer using a computer.
- This prototype condition proposing method causes the computer to execute a regression model construction process and a prototype condition proposing process.
- the regression model construction process represents a process of constructing a regression model for property measurement data representing actual measurement results of the properties of the material.
- the prototype condition proposing process represents a process of searching for optimal prototype conditions for the material using the constructed regression model, and proposing prototype conditions for the material based on the search results.
- the regression model construction process includes a process of calculating a weighting criterion that is a criterion for weighting the property measurement data, and a process of weighting the property measurement data based on the calculated weighting criterion.
- the present invention even when the amount of training data used to construct a machine learning model is small or when the distribution of the training data is biased, it is possible to accurately propose optimal prototyping conditions for materials.
- FIG. 1 is a schematic diagram showing an overview of a prototype condition proposal system according to an embodiment of the present invention
- FIG. 1 is a diagram showing functional blocks of a prototype conditions proposal system according to an embodiment of the present invention.
- 1 is a flowchart showing the overall processing flow of a prototype conditions proposal system according to an embodiment of the present invention.
- 11 is a flowchart showing details of characteristic measurement data pre-processing.
- 11 is a flowchart showing details of a regression model construction process.
- FIG. 11 is a diagram showing a comparison of estimation accuracy before and after weighting.
- 11 is a flowchart showing details of a prototype condition proposal process.
- FIG. 1 is a schematic diagram showing an overview of a prototype condition proposal system according to one embodiment of the present invention.
- the prototype condition proposal system 1 shown in FIG. 1 optimizes various prototype conditions to be considered when prototyping a material, such as the composition of the material and the firing conditions, and proposes the results to a material developer, who is the user of this system.
- the optimization of the prototype conditions is performed using various machine learning algorithms based on actual characteristic measurement data representing the actual measurement results of the characteristics of the material.
- GPR Gaussian Process Regression
- SVR Support Vector Regression
- Logistic Regression Logistic Regression
- LASSO Regression Least Absolute Shrinkage and Selection Operator Regression
- the prototype condition proposal system 1 when a user of the prototype condition proposal system 1 prototypes a material under the prototype conditions proposed by the prototype condition proposal system 1 and actually measures the properties of the resulting material, new property measurement data is generated that represents the evaluation value of the properties of the resulting product.
- the prototype condition proposal system 1 learns this property measurement data as new learning data, the prototype condition proposal system 1 proposes more optimized prototype conditions to the user. Each time the user repeats this cycle, the prototype condition proposal system 1 of this embodiment can propose prototype conditions with more favorable predicted property values.
- the prototype condition proposal system may include a function for prototyping a material and a function for actually measuring the properties of the material to be prototyped, or may be configured integrally with these functions.
- the prototype conditions proposal system 1 of this embodiment is realized by a single general-purpose computer device, as shown in FIG. 1.
- the prototype conditions proposal system 1 is assumed to be realized by a single general-purpose computer device that includes one or more processor devices, one or more storage devices, one or more input/output devices, and wired or wireless communication lines (none of which are shown) connecting them.
- This computer device is installed as a terminal inside a laboratory, for example, and is connected to various other terminals installed inside and outside the laboratory, various terminals owned by each user such as laptop PCs, tablets, smartphones, etc. (hereinafter referred to as "user terminals"), and other devices such as server devices, via a communication network such as the Internet 400 or a dedicated line.
- the computer device and the Internet 400 are connected by wire via well-known communication equipment (not shown), but may also be connected wirelessly.
- FIG. 2 is a diagram showing the functional blocks of the prototype conditions proposal system according to one embodiment of the present invention. Note that each block described below shows a functional block, not a hardware-based configuration.
- the prototype conditions proposal system 1 of this embodiment is configured to include a control unit 11, a memory unit 12, a user interface unit 13, and a communication unit 14.
- the control unit 11 executes various data processing operations based on user operation input detected by the user interface unit 13, data acquired by the communication unit 14, and programs and data stored in the memory unit 12.
- the control unit 11 also functions as an interface between the user interface unit 13, the communication unit 14, and the memory unit 12.
- the control unit 11 has the functional blocks of a characteristic measurement data preprocessing unit 111, a regression model construction processing unit 112, and a prototype condition proposal processing unit 113.
- the control unit 11 is configured using processor devices (hereinafter simply referred to as "processors") such as a CPU (Central Processing Unit) and various co-processors, and can realize these functional blocks by executing a predetermined program.
- processors such as a CPU (Central Processing Unit) and various co-processors
- the control unit 11 may also be configured using a logic circuit such as an FPGA (Field Programmable Gate Array).
- the control unit 11 may also be configured by combining a processor and a logic circuit.
- the program executed by the control unit 11 may be installed from a program source.
- the program source may be, for example, a program distribution computer or a computer-readable recording medium.
- the program executed by the control unit 11 may also be composed of device drivers, an operating system, various application programs located at higher layers, and libraries that provide common functions to these programs. Furthermore, two or more programs may be realized as one program, and one program may be realized as two or more programs.
- the characteristic measurement data preprocessing unit 111 performs preprocessing on the characteristic measurement data in the so-called raw data state immediately after recording. This processing performed by the characteristic measurement data preprocessing unit 111 is called characteristic measurement data preprocessing.
- the regression model construction processing unit 112 executes a process of constructing a regression model for the preprocessed characteristic measurement data. This process executed by the regression model construction processing unit 112 is referred to as the regression model construction process.
- the prototype condition proposal processing unit 113 searches for optimal prototype conditions for the material using the regression model constructed by the regression model construction processing unit 112 for the preprocessed characteristic measurement data, and performs processing to propose prototype conditions for the material to the user based on the search results. This processing performed by the prototype condition proposal processing unit 113 is referred to as the prototype condition proposal processing.
- the storage unit 12 is configured using a storage device such as a RAM or flash memory, and stores programs that supply various processing commands to the control unit 11, and data representing various information used in the processing executed by the control unit 11. For example, characteristic measurement data that has been preprocessed by the characteristic measurement data preprocessing unit 11 (hereinafter referred to as "preprocessed data") and data representing the regression model constructed by the regression model construction processing unit 112 are stored in the storage unit 12.
- preprocessed data characteristic measurement data that has been preprocessed by the characteristic measurement data preprocessing unit 11
- data representing the regression model constructed by the regression model construction processing unit 112 are stored in the storage unit 12.
- the control unit 11 can realize the functional blocks of the characteristic measurement data preprocessing unit 111, the regression model construction processing unit 112, and the prototype condition proposal processing unit 113 described above by reading and writing this information in the storage unit 12.
- the user interface unit 13 accepts input operations from the user, and is responsible for processing related to the user interface, such as image display and audio output.
- the user interface unit 13 has the functional blocks of an input unit 131 and an output unit 132.
- the input unit 131 detects various operations from the user.
- the input unit 131 is configured using, for example, a keyboard, a pointing device, a touch panel, etc.
- the output unit 132 executes screen display, audio output, etc. for the user.
- the output unit 132 is configured using, for example, an LCD display, a touch screen, etc.
- the communication unit 14 is responsible for processing communications with user terminals owned by each user and other devices such as server devices via the Internet 400.
- the communication unit 14 is configured using, for example, a NIC (Network Interface Card) or an HBA (Host Bus Adapter).
- the functions of the prototype conditions proposal system 1 have been described as being realized integrally by a single computer device. However, these functions may also be realized by multiple computers or server devices connected to each other. Furthermore, the prototype conditions proposal system 1 may be configured to include a general-purpose computer device such as a laptop PC and a web browser installed thereon, or may be configured to include a web server and various portable devices.
- each function is just an example, and multiple functions may be combined into one function, or one function may be divided into multiple functions.
- FIG. 3 is a flowchart showing the overall processing flow of the prototype conditions proposal system according to one embodiment of the present invention. Note that in the following description, the processing may be described with the aforementioned functions or programs as the subject, but the processing described with the function or program as the subject may also be processing performed by a processor or a device having that processor.
- step S310 the control unit 11 executes characteristic measurement data preprocessing using the characteristic measurement data preprocessing unit 111.
- the characteristic measurement data is preprocessed and becomes preprocessed data, making it possible to execute each subsequent process normally. Details of the characteristic measurement data preprocessing performed in step S310 will be explained later with reference to the flowchart in FIG. 4.
- the control unit 11 proceeds to step S320.
- step S320 the control unit 11 executes a regression model construction process using the regression model construction processing unit 112. As a result, a regression model is constructed for the preprocessed data. Details of the regression model construction process performed in step S320 will be described later with reference to the flowchart in FIG. 5. When the regression model construction process is completed, the control unit 11 proceeds to step S330.
- step S330 the control unit 11 executes a regression model evaluation process.
- the generalization performance which is an index representing the predictive accuracy of a regression model, is evaluated for each of the multiple regression models constructed as a result of each process up to step S320.
- This evaluation is performed, for example, by performing cross validation with other regression models.
- the evaluation results are visualized by graphs such as scatter plots and box plots. This allows the user to receive suggestions for prototyping conditions based on a regression model with good generalization performance.
- the control unit 11 proceeds to step S340.
- step S340 the control unit 11 executes a prototype condition proposal process by the prototype condition proposal processing unit 113.
- the user of the prototype condition proposal system 1 can appropriately modify the prototype conditions of the material proposed by the prototype condition proposal system 1 to make them more preferable.
- the control unit 11 obtains a predicted value of the material properties when prototyped under the prototype conditions modified by the user by applying it to the selected regression model, and presents it to the user. In other words, the user can interactively modify the prototype conditions while checking the predicted value.
- the prototype condition proposal system 1 is a system that can incorporate the knowledge of the user, who is the developer of the material, into the prototype conditions of the material proposed to the user. Details of the prototype condition proposal process performed in step S340 will be described later with reference to the flowchart in FIG. 7.
- the control unit 11 temporarily ends the process shown in the flowchart in FIG. 3.
- the prototype condition proposal system 1 of this embodiment executes each process of steps S310 to S340 in FIG. 3 to propose favorable prototype conditions to the user. That is, in step S310, the prototype condition proposal system 1 of this embodiment automatically performs the necessary preprocessing on the characteristic measurement data in the raw data state. Therefore, each process of steps S320 to S340 can be executed without manually performing complicated preprocessing on the characteristic measurement data. Furthermore, in the prototype condition proposal system 1 of this embodiment, the user can select a regression model with good generalization performance. Therefore, the prototype condition proposal system 1 can propose prototype conditions with favorable predicted characteristic values to the user.
- a user of the prototype conditions proposal system 1 can prototype a material under the prototype conditions proposed by the prototype conditions proposal system 1, measure the properties of the resulting product, and have the prototype conditions proposal system 1 learn the data representing the actual measurement results as new property measurement data, and then have the prototype conditions proposal system 1 execute the processes of steps S310 to S340 in FIG. 3 again.
- the prototype conditions proposal system 1 can propose more optimized prototype conditions to the user.
- the prototype conditions proposal system 1 of this embodiment can propose prototype conditions with better predicted property values each time it repeats the processes of steps S310 to S340 in FIG. 3 for the same prototype object.
- Figure 4 is a flowchart showing the details of preprocessing of characteristic measurement data.
- step S410 the control unit 11 causes the characteristic measurement data preprocessing unit 111 to accept input of characteristic measurement data from the user via the input unit 131 or the communication unit 14.
- the characteristic measurement data input to the prototype condition proposal system 1 may be, for example, categorical data, continuous data, or discrete data.
- the specific data format of the characteristic measurement data input to the prototype condition proposal system 1 may be appropriately determined.
- step S420 the control unit 11 sets the type of variables for the characteristic measurement data input from the user in step S410 through the characteristic measurement data preprocessing unit 111.
- an explanatory variable is a variable that is the basis for determining a predicted value of a characteristic.
- the explanatory variables are the composition of the materials that make up the prototype conditions and the firing conditions.
- the target variable is a variable that represents the characteristic value of the prototype material that is the subject of prediction.
- the explanatory variables may be set as defaults, and a setting operation may be received from the user who wishes to change the target variable.
- step S430 the control unit 11 determines whether or not there is an abnormal value in the characteristic measurement data for which one of the explanatory variables and the objective variable was set in step S420, by using the characteristic measurement data preprocessing unit 111.
- the process of determining whether or not there is an abnormal value is performed, for example, by first displaying the characteristic measurement data in a histogram and determining whether or not there is an outlier that falls outside the range of the average value ⁇ 2 ⁇ , and then determining whether or not there was a data input error or a malfunction of the evaluation test machine when generating the characteristic measurement data for which it is determined that there is an outlier.
- the control unit 11 deletes the abnormal value and sets it as a missing value, and proceeds to step S440. Furthermore, if the type of abnormal value contained in the characteristic measurement data is one of the objective variables when the explanatory variables completely overlap and there are different levels of the characteristics, the control unit 11 deletes the sample related to the abnormal value and proceeds to step S440. This is because, when all explanatory variables are duplicated and levels with different characteristics are treated as abnormal values, the level itself must be deleted, unlike when one explanatory variable is an abnormal value due to an input error, etc. On the other hand, if it is determined that the characteristic measurement data does not contain any abnormal values, the control unit 11 proceeds directly to step S440.
- the prototype condition proposal system 1 of this embodiment can omit the process in step S430.
- step S440 the control unit 11 determines whether or not there is a missing value in the characteristic measurement data by the characteristic measurement data preprocessing unit 111. This is because the characteristic measurement data may already contain missing values. If it is determined that the preprocessed data contains missing values, the control unit 11 complements the missing values and proceeds to step S450.
- the missing value complementation process is performed, for example, by using the average value, median value, minimum value, maximum value, etc. of the characteristic measurement data excluding abnormal values as a value to complement the missing value.
- the missing value may also be complemented by linear interpolation. Note that, for example, when the characteristic measurement data preprocessing unit 111 complements the missing value in step S440, it may display the complemented value in red to make it easier to identify.
- the characteristic measurement data preprocessing unit 111 may delete the level itself without complementing the missing value. Furthermore, in the prototype condition proposal system 1 of this embodiment, the characteristic measurement data preprocessing unit 111 can delete the explanatory variables themselves when the explanatory variables have a large missing ratio, for example, when more than 50% of the data is missing. On the other hand, if it is determined that the characteristic measurement data does not contain missing values, the control unit 11 proceeds directly to step S450.
- step S450 if the explanatory variables are categorical values rather than continuous values, the control unit 11 causes the characteristic measurement data preprocessing unit 111 to perform encoding processing on the explanatory variables and convert them into numerical data.
- the characteristic measurement data preprocessing unit 111 executes this encoding processing by, for example, referring to records in a table that represents the correspondence between categorical data and numerical data, which is stored in the memory unit 12.
- the control unit 11 completes the processing in step S450, it proceeds to step S460.
- step S460 the control unit 11 determines whether the characteristic measurement data includes redundant explanatory variables through the characteristic measurement data preprocessing unit 111. This determination is based on whether a combination of explanatory variables with a correlation coefficient of a predetermined number or more, for example, 0.8 or more, can be extracted. If it is determined that the characteristic measurement data includes redundant explanatory variables, the control unit 11 deletes one of the redundant explanatory variables and proceeds to step S470. In the prototype condition proposal system 1 of this embodiment, combinations of explanatory variables with a correlation coefficient of 0.8 or more are visualized to the user through the output unit 132, and the user can select the explanatory variable to be deleted through the input unit 131. On the other hand, if it is determined that the characteristic measurement data does not include redundant explanatory variables, the control unit 11 proceeds directly to step S470.
- step S470 the control unit 11 uses the characteristic measurement data preprocessing unit 111 to perform standardization processing on the characteristic measurement data as necessary.
- the control unit 11 completes the processing in step S470, it stores the preprocessed data, which is the characteristic measurement data that has been preprocessed in steps S410 to S470 in FIG. 4, in the storage unit 12, and ends the characteristic measurement data preprocessing shown in the flowchart in FIG. 4.
- the control unit 11 may use the characteristic measurement data preprocessing unit 111 to perform normalization processing on the characteristic measurement data as necessary. In such a case, the control unit 11 may execute each subsequent process without performing the standardization processing in S470.
- Figure 5 is a flowchart showing the details of the regression model construction process.
- step S510 the control unit 11 selects the implementation conditions of cross-validation for evaluating the regression model to be constructed for the preprocessed data by the regression model construction processing unit 112.
- the prototyping condition proposal system 1 of this embodiment evaluates each regression model by K-fold cross validation.
- the prototyping condition proposal system 1 evaluates the regression model by 10-fold cross validation.
- the user can also select the implementation conditions of cross validation.
- the regression model construction processing unit 112 can accept the implementation conditions of cross validation from the user via the input unit 131 or the communication unit 14.
- step S520 the control unit 11 selects, by the regression model construction processing unit 112, a candidate regression model to be used as a predictive model in searching for prototype conditions.
- the regression model construction processing unit 112 selects as a candidate a regression model for which a selection process has been accepted from the user via the input unit 131 or the communication unit 14.
- the user can select as candidates a plurality of regression models from various regression models such as Gaussian process regression and the aforementioned linear regression, regression tree (including the case of the ensemble method), regression by neural network, support vector regression, logistic regression, and LASSO regression.
- the control unit 11 proceeds to step S530.
- step S530 the control unit 11 executes a process of calculating a weighting criterion, which is a criterion for weighting the actual property measurement data, by the regression model construction processing unit 112.
- the weighting criterion is calculated based on the difference between the objective variable included in the actual property measurement data and the target characteristic that represents the target value of the material characteristic, or a statistic that represents the rarity of the explanatory variable included in the actual property measurement data (described in detail later). Note that a specific example of a statistic that represents the rarity of an explanatory variable is the occurrence probability of an explanatory variable that satisfies a predetermined condition.
- step S540 the control unit 11 executes a process of weighting the actual characteristic measurement data based on the weighting criterion calculated in step S530 by the regression model construction processing unit 112.
- the regression model construction processing unit 112 can also directly weight the regression model based on the weighting criterion calculated in step S530.
- the process executed by the regression model construction processing unit 112 in step S540 based on the result of step S530 is specifically one of the following processes: a process of setting a loss function, which is a function that serves as an index of learning; an oversampling process of amplifying rare or important actual characteristic measurement data and adding it as learning data; and an undersampling process of deleting redundant or unimportant actual characteristic measurement data from the learning data (described in detail later).
- a process of setting a loss function which is a function that serves as an index of learning
- an oversampling process of amplifying rare or important actual characteristic measurement data and adding it as learning data and an undersampling process of deleting redundant or unimportant actual characteristic measurement data from the learning data (described in detail later).
- step S550 the control unit 11 causes the regression model construction processing unit 112 to search for and set optimal hyperparameters for each regression model of each method selected as a candidate in step S520.
- the regression model construction processing unit 112 automatically searches for all parameters for each regression model, and automatically sets those that will best improve the generalization performance of the regression model when set as hyperparameters during construction of the regression model.
- the control unit 11 completes the processing in step S550, it proceeds to step S560.
- step S560 the control unit 11 performs a process of creating a regression model for each method, with the optimal hyperparameters set, using the regression model construction processing unit 112.
- the regression model construction processing unit 112 selects the regression model with the highest generalization performance from all the created regression models, and performs a process of determining the final regression model.
- the control unit 11 ends the regression model construction process shown in the flowchart of FIG. 5.
- the weighting criteria are calculated based on the difference between the objective variable included in the measured characteristic data and the target characteristic that represents the target value of the material's characteristics, or based on a statistic that represents the rarity of the explanatory variable included in the measured characteristic data.
- step S530 The specific details of the process executed in step S530 when calculating the weighting criteria based on the difference between the objective variable included in the actual characteristic measurement data and the target characteristic that represents the target value of the material characteristic are shown below as steps S532 to S534.
- step S532 the control unit 11 calculates the difference between the objective variable included in the characteristic measurement data and the target characteristic representing the target value of the material characteristic by the regression model construction processing unit 112. When the processing in step S532 is completed, the control unit 11 proceeds to step S534.
- step S534 the control unit 11 determines the weighting criterion by the regression model construction processing unit 112 using a function of the difference between the objective variable and the target characteristic calculated in step S532.
- this function include the absolute value or square value of the difference between the objective variable and the target characteristic, which represents the distance between the objective variable and the target characteristic.
- step S530 when calculating the weighting criteria based on statistics that indicate the rarity of the explanatory variables contained in the characteristic measurement data, the specific content of the processing executed in step S530 is as shown below as steps S536 to S538.
- step S536 the control unit 11 uses the regression model construction processing unit 112 to calculate the number of occurrences of explanatory variables that take a specific value or fall within a specific range. Specifically, for example, when the explanatory variables relate to the raw material ratios of various raw materials, the control unit 11 calculates the number of occurrences of explanatory variables that take a value greater than 0, in other words, that represent the raw material ratios that are actually used.
- step S536 the control unit 11 proceeds to step S538.
- step S538 the control unit 11 determines the weighting criterion by the regression model construction processing unit 112 using a function of the number of occurrences of the explanatory variable calculated in step S536.
- This function is the probability of occurrence of the explanatory variable in the entire learning data.
- the regression model construction processing unit 112 may also determine the number of occurrences of the explanatory variable as the weighting criterion, instead of using the probability of occurrence of the explanatory variable as the weighting criterion.
- the reason that a criterion based on the explanatory variable is used is to incorporate the material science knowledge of the user, who is a material developer, particularly for important parameters that do not appear in the objective variables.
- the regression model construction processing unit 112 performs weighting on the characteristic measured data or the regression model by executing one of the following processes: a process for setting a loss function; an oversampling process for amplifying rare or important characteristic measured data and adding it as learning data; and an undersampling process for deleting redundant or unimportant characteristic measured data from the learning data. That is, the regression model construction process further includes one of the following processes, which are executed based on the calculated weighting criterion: a process for setting a loss function; an oversampling process for amplifying rare or important characteristic measured data and adding it as learning data; and an undersampling process for deleting redundant or unimportant characteristic measured data from the learning data.
- step S530 when setting the loss function based on the weighting criteria calculated in step S530 are described below as step S542.
- step S542 the control unit 11 determines, through the regression model construction processing unit 112, machine learning based on the weighting criteria calculated in step S530 so as to reduce the prediction error of rare or important measured characteristic data. This process is performed by weighting the measured characteristic data with the loss function during learning.
- step S538 the control unit 11 ends the process shown in step S540 of reflecting the weighting criteria in the learning data or regression model, and proceeds to step S550.
- step S530 the specific content of the process called oversampling or upsampling, which is executed in step S530 when rare or important measured characteristic data is amplified based on the weighting criteria calculated in step S530 and added as learning data, is described below as step S544.
- step S544 the control unit 11 causes the regression model construction processing unit 112 to copy the rare or important characteristic measurement data to be oversampled based on the weighting criteria calculated in step S530, and add it to the learning data.
- the characteristic measurement data is copied until the number of copies reaches a predetermined threshold value or more.
- the oversampling process may be performed by various methods such as SMOTE, ADASYN, Borderline-SMOTE, and Safe-level SMOTE. This adds rare or important characteristic measurement data to the learning data, improving the balance of the learning data.
- Figure 6 shows a comparison of the estimation accuracy before and after the weighting process by oversampling performed in step S544.
- Figure 6 shows a scatter plot of characteristic measurement data and its predicted value for bending strength, which is one of the characteristics of a ceramic composite material made from a number of raw materials.
- FIG. 6 shows that weighting was performed on the actual characteristic data of a specific raw material, for which the raw material was used only a small number of times in the entire actual measurement data, resulting in insufficient estimation accuracy. As a result, the error between the estimated value and the actual measurement value for this composite material was reduced, or in other words, the estimation accuracy for this composite material was improved.
- the control unit 11 ends the process of reflecting the weighting criteria in the learning data or regression model shown in step S540, and proceeds to step S550.
- step S540 the specific content of the process called undersampling or downsampling, which is executed in step S540 when redundant or less important measured characteristic data is deleted from the learning data based on the weighting criteria calculated in step S530, is described below as step S546.
- step S546 the control unit 11 causes the regression model construction processing unit 112 to delete redundant or unimportant measured characteristic data from the learning data. This improves the balance of the learning data because redundant or unimportant measured characteristic data is deleted from the learning data.
- step S546 the control unit 11 ends the processing shown in step S540 for reflecting the weighting criteria in the learning data or regression model, and proceeds to step S550.
- Figure 7 is a flowchart showing the details of the prototype condition proposal process.
- step S710 the control unit 11 causes the prototyping condition proposal processing unit 113 to perform a prototyping condition search process based on the regression model constructed in step S560 of FIG. 5.
- the prototyping condition proposal processing unit 113 executes this process by optimization processing.
- the prototyping condition proposal system 1 of this embodiment is configured to be able to use various optimization processing methods such as mathematical optimization (MO), Bayesian optimization (BO), genetic algorithm (GA), Newton's method (NM), and simplex method (SM).
- MO mathematical optimization
- BO Bayesian optimization
- GA genetic algorithm
- NM Newton's method
- SM simplex method
- the prototyping condition proposal processing unit 113 performs a sensitivity analysis on the tentative prototyping condition to evaluate the importance of each explanatory variable constituting the tentative prototyping condition, and presents the evaluation result together.
- the regression model used is Gaussian process regression, it is also possible to select the prototype conditions that maximize the acquisition function.
- step S720 the control unit 11 accepts from the user a modification to the provisional prototyping conditions proposed to the user in step S710 by the prototyping condition proposal processing unit 113.
- the prototyping condition proposal processing unit 113 accepts an input operation related to the modification of the values of the explanatory variables constituting the provisional prototyping conditions from the user via the input unit 131 or the communication unit 14, it modifies the provisional prototyping conditions according to the modification content.
- the control unit 11 uses the prototyping condition proposal processing unit 113 to use a regression model to obtain a predicted value of the material characteristics when a prototype is produced under the modified provisional prototyping conditions, and presents the calculation result to the user.
- the prototyping condition proposal processing unit 113 also performs a sensitivity analysis on the modified provisional prototyping conditions, as in step S710, to evaluate the importance of the explanatory variables constituting the modified provisional prototyping conditions, and presents the evaluation result together.
- the evaluation result is updated every time the user modifies the provisional prototyping conditions, and the latest evaluation result is always presented to the user.
- the control unit 11 proceeds to step S730.
- step S730 the control unit 11, using the prototyping condition proposal processing unit 113, determines whether the predicted value of the characteristic obtained in step S720 for the revised tentative prototyping conditions is insufficient as the characteristic value of the material to be prototyped.
- the regression model used is Gaussian process regression
- this determination is made by obtaining an acquisition function (Acquisition Function) for each prototyping condition, which represents the expected value of improvement in the material characteristic when prototyped under that prototyping condition, for the revised tentative prototyping conditions, and determining whether the difference between the value of that acquisition function and the maximum value of the acquisition function is within a specified range.
- the acquisition function is calculated based on the predicted value ⁇ of the material characteristic when prototyped under any prototyping condition, and the standard deviation ⁇ representing the variance of that predicted value. If it is determined that the predicted characteristic values are insufficient, the process returns to step S720 to again accept instructions from the user to modify the prototype conditions. If it is determined that the predicted characteristic values are not insufficient, the predicted values of the material characteristics when prototyping under the modified provisional prototype conditions are sufficient, so the provisional prototype conditions are finalized and proposed to the user as finalized prototype conditions. In other words, this determination process is repeated until it is determined that the predicted values of the material characteristics related to the provisional prototype conditions are not insufficient. When the process in step S730 is completed, the control unit 11 ends the prototype condition proposal process shown in the flowchart of FIG. 7.
- the prototype condition proposal system 1 of this embodiment proposes more optimized prototype conditions to the user based on the newly input characteristic measurement data.
- the control unit 11 of the prototype condition proposal system 1 determines whether the newly input characteristic measurement data contains missing values. If it is determined that a missing value is included, in order to complement the missing value, characteristic measurement data preprocessing is performed on the newly input characteristic measurement data by starting processing from step S440 of FIG. 4.
- the control unit 11 determines whether or not the regression model needs to be updated. If it is determined that the regression model needs to be updated, the control unit 11 executes a regression model construction process for the newly input characteristic measurement data, starting from step S510 in FIG. 5, in order to construct a new regression model. On the other hand, if it is determined that the regression model does not need to be updated, the control unit 11 omits the regression model construction process and the regression model evaluation process, and performs a prototype condition proposal process for the newly input characteristic measurement data, starting from step S710 in FIG. 7, using the previously constructed regression model. Note that even if the newly input characteristic measurement data contains missing values, if the regression model does not need to be updated, the control unit 11 similarly omits the regression model construction process and the regression model evaluation process.
- the prototype condition proposal system 1 is a system that proposes prototype conditions for a material to a material developer, and includes a regression model construction processing unit 112 and a prototype condition proposal processing unit 113.
- the regression model construction processing unit 112 executes a regression model construction process for the characteristic measurement data that represents the actual measurement results of the characteristics of the material (step S320).
- the prototype condition proposal processing unit 113 searches for optimal prototype conditions for the material using the constructed regression model, and executes a prototype condition proposal process based on the search results (step S340).
- the regression model construction process (FIG.
- step S530 includes a process of calculating a weighting criterion that is a criterion for weighting the characteristic measurement data (step S530), and a process of weighting the characteristic measurement data based on the calculated weighting criterion (step S540).
- the weighting criteria are calculated based on the difference between the objective variables included in the characteristic measurement data and the target characteristics that represent the target values of the material characteristics (steps S532 to S534). In this way, even in material development where the weighting criteria cannot be determined based on the number of times the objective variable appears because regression problems are often solved, the weighting criteria can be determined based on the objective variables.
- the weighting criteria are calculated based on statistics that indicate the rarity of the explanatory variables contained in the characteristic measurement data (steps S536 to S538). In this way, the weighting criteria can be calculated based on the explanatory variables. As a result, it is possible to incorporate the material science knowledge of the user, who is a material developer, especially for important parameters that do not appear in the objective variables.
- the regression model construction process (FIG. 5) further includes a process (step S542) of setting a loss function based on the calculated weighting criteria. In this way, it is possible to reduce prediction errors for rare or important characteristic measurement data.
- the regression model construction process (FIG. 5) further includes an oversampling process (step S544) that amplifies rare or important measured characteristic data based on the calculated weighting criteria and adds the data as training data. In this way, rare or important measured characteristic data is added to the training data. As a result, the sensitivity of rare or important measured characteristic data can be improved, improving the balance of the training data.
- the regression model construction process (FIG. 5) further includes an undersampling process (step S546) that deletes redundant or unimportant characteristic measurement data from the training data based on the calculated weighting criterion.
- an undersampling process (step S546) that deletes redundant or unimportant characteristic measurement data from the training data based on the calculated weighting criterion.
- redundant or unimportant characteristic measurement data is deleted from the training data.
- the sensitivity of redundant or unimportant characteristic measurement data can be reduced, improving the balance of the training data.
- Prototype condition proposal system 11 Control unit 12: Storage unit 13: User interface unit 14: Communication unit 111: Property measurement data preprocessing unit 112: Feature quantity selection processing unit 113: Regression model construction processing unit 114: Prototype condition proposal processing unit 131: Input unit 132: Output unit 400: Internet
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Evolutionary Biology (AREA)
- Algebra (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
また、本発明による試作条件提案方法は、コンピュータを用いて材料開発者に材料の試作条件を提案する方法である。この試作条件提案方法は、回帰モデル構築処理と、試作条件提案処理とをコンピュータに実行させる。回帰モデル構築処理は、材料の特性の実測結果を表す特性実測データについて回帰モデルを構築する処理を表す。試作条件提案処理は、構築された回帰モデルを用いて材料の最適な試作条件を探索し、探索結果に基づいて当該材料の試作条件を提案する処理を表す。回帰モデル構築処理は、特性実測データへの重みづけの基準である重み基準を算出する処理と、算出された重み基準に基づいて特性実測データに重みづけを行う処理とを含む。
11:制御部
12:記憶部
13:ユーザーインターフェース部
14:通信部
111:特性実測データ前処理部
112:特徴量選択処理部
113:回帰モデル構築処理部
114:試作条件提案処理部
131:入力部
132:出力部
400:インターネット
Claims (7)
- 材料開発者に材料の試作条件を提案する試作条件提案システムであって、
前記材料の特性の実測結果を表す特性実測データについて回帰モデル構築処理を実行する回帰モデル構築処理部と、
前記構築された回帰モデルを用いて前記材料の最適な試作条件を探索し、探索結果に基づいて試作条件提案処理を実行する試作条件提案処理部と
を備え、
前記回帰モデル構築処理は、
前記特性実測データへの重みづけの基準である重み基準を算出する処理と、
前記算出された重み基準に基づいて前記特性実測データに重みづけを行う処理と
を含む、試作条件提案システム。 - 請求項1に記載の試作条件提案システムにおいて、
前記重み基準は、前記特性実測データに含まれる目的変数と、前記材料の特性の目標値を表す目標特性との差に基づいて算出される、試作条件提案システム。 - 請求項1に記載の試作条件提案システムにおいて、
前記重み基準は、前記特性実測データに含まれる説明変数の希少性をあらわす統計量に基づいて算出される、試作条件提案システム。 - 請求項1に記載の試作条件提案システムにおいて、
前記回帰モデル構築処理は、前記算出された重み基準に基づいて損失関数を設定する処理をさらに含む、試作条件提案システム。 - 請求項1に記載の試作条件提案システムにおいて、
前記回帰モデル構築処理は、前記算出された重み基準に基づいて希少または重要性の高い特性実測データを増幅し、学習データとして追加するオーバーサンプリング処理をさらに含む、試作条件提案システム。 - 請求項1に記載の試作条件提案システムにおいて、
前記回帰モデル構築処理は、前記算出された重み基準に基づいて、冗長または重要性の低い特性実測データを学習データから削除するアンダーサンプリング処理をさらに含む、試作条件提案システム。 - コンピュータを用いて材料開発者に材料の試作条件を提案する試作条件提案方法であって、
前記材料の特性の実測結果を表す特性実測データについて回帰モデルを構築する回帰モデル構築処理と、
前記構築された回帰モデルを用いて前記材料の最適な試作条件を探索し、探索結果に基づいて当該材料の試作条件を提案する試作条件提案処理と
をコンピュータに実行させ、
前記回帰モデル構築処理は、
前記特性実測データへの重みづけの基準である重み基準を算出する処理と、
前記算出された重み基準に基づいて前記特性実測データに重みづけを行う処理と
を含む、試作条件提案方法。
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE112023004076.1T DE112023004076T5 (de) | 2022-09-29 | 2023-07-25 | Testproduktionsbedingungsvorschlagssystem und testproduktionsbedingungvorschlagsverfahren |
| JP2024549786A JP7886957B2 (ja) | 2022-09-29 | 2023-07-25 | 試作条件提案システム、試作条件提案方法 |
| US19/051,384 US20250181672A1 (en) | 2022-09-29 | 2025-02-12 | Trial production condition proposal system and trial production condition proposal method |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2022157142 | 2022-09-29 | ||
| JP2022-157142 | 2022-09-29 |
Related Child Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| US19/051,384 Continuation US20250181672A1 (en) | 2022-09-29 | 2025-02-12 | Trial production condition proposal system and trial production condition proposal method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2024070169A1 true WO2024070169A1 (ja) | 2024-04-04 |
Family
ID=90477067
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2023/027145 Ceased WO2024070169A1 (ja) | 2022-09-29 | 2023-07-25 | 試作条件提案システム、試作条件提案方法 |
Country Status (4)
| Country | Link |
|---|---|
| US (1) | US20250181672A1 (ja) |
| JP (1) | JP7886957B2 (ja) |
| DE (1) | DE112023004076T5 (ja) |
| WO (1) | WO2024070169A1 (ja) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010092266A (ja) * | 2008-10-08 | 2010-04-22 | Nec Corp | 学習装置、学習方法及びプログラム |
| JP2014063432A (ja) * | 2012-09-24 | 2014-04-10 | Nippon Telegr & Teleph Corp <Ntt> | 欠損値予測装置、欠損値予測方法、欠損値予測プログラム |
| JP2019191769A (ja) * | 2018-04-20 | 2019-10-31 | 富士通株式会社 | データ判別プログラム、データ判別装置およびデータ判別方法 |
| WO2020255299A1 (ja) * | 2019-06-19 | 2020-12-24 | 日本電信電話株式会社 | 異常度推定装置、異常度推定方法、プログラム |
| JP2021111221A (ja) * | 2020-01-14 | 2021-08-02 | パナソニックIpマネジメント株式会社 | 計測装置、システム、計測方法、及びプログラム |
| JP2022092297A (ja) * | 2020-12-10 | 2022-06-22 | 昭和電工マテリアルズ株式会社 | 設計支援装置、設計支援方法及び設計支援プログラム |
-
2023
- 2023-07-25 DE DE112023004076.1T patent/DE112023004076T5/de active Pending
- 2023-07-25 JP JP2024549786A patent/JP7886957B2/ja active Active
- 2023-07-25 WO PCT/JP2023/027145 patent/WO2024070169A1/ja not_active Ceased
-
2025
- 2025-02-12 US US19/051,384 patent/US20250181672A1/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010092266A (ja) * | 2008-10-08 | 2010-04-22 | Nec Corp | 学習装置、学習方法及びプログラム |
| JP2014063432A (ja) * | 2012-09-24 | 2014-04-10 | Nippon Telegr & Teleph Corp <Ntt> | 欠損値予測装置、欠損値予測方法、欠損値予測プログラム |
| JP2019191769A (ja) * | 2018-04-20 | 2019-10-31 | 富士通株式会社 | データ判別プログラム、データ判別装置およびデータ判別方法 |
| WO2020255299A1 (ja) * | 2019-06-19 | 2020-12-24 | 日本電信電話株式会社 | 異常度推定装置、異常度推定方法、プログラム |
| JP2021111221A (ja) * | 2020-01-14 | 2021-08-02 | パナソニックIpマネジメント株式会社 | 計測装置、システム、計測方法、及びプログラム |
| JP2022092297A (ja) * | 2020-12-10 | 2022-06-22 | 昭和電工マテリアルズ株式会社 | 設計支援装置、設計支援方法及び設計支援プログラム |
Also Published As
| Publication number | Publication date |
|---|---|
| DE112023004076T5 (de) | 2025-08-21 |
| JP7886957B2 (ja) | 2026-07-08 |
| US20250181672A1 (en) | 2025-06-05 |
| JPWO2024070169A1 (ja) | 2024-04-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gromski et al. | The influence of scaling metabolomics data on model classification accuracy | |
| US20200372342A1 (en) | Systems and methods for predictive early stopping in neural network training | |
| US20240111931A1 (en) | Trial manufacturing condition proposing system and trial manufacturing condition proposing method | |
| JP7807463B2 (ja) | 産業特定機械学習アプリケーション | |
| JP4627674B2 (ja) | データ処理方法及びプログラム | |
| CN111080117A (zh) | 设备风险标签的构建方法、装置、电子设备及存储介质 | |
| KR102054500B1 (ko) | 설계 도면 제공 방법 | |
| US20210201179A1 (en) | Method and system for designing a prediction model | |
| Niu et al. | Hybrid Gaussian process regression with temporal feature extraction for partially interpretable remaining useful life interval prediction in Aeroengine prognostics | |
| US20220101198A1 (en) | Automated generation of a machine learning model from computational simulation data | |
| CN116862078B (zh) | 一种换电套餐用户逾期的预测方法、系统、装置及介质 | |
| Zhou et al. | SurvMetrics: An R package for Predictive Evaluation Metrics in Survival Analysis. | |
| EP4310736A1 (en) | Method and system of generating causal structure | |
| JPWO2016111240A1 (ja) | 情報処理システム、変化点検出方法、およびプログラム | |
| Yerlikaya-Özkurt et al. | A review and new contribution on conic multivariate adaptive regression splines (CMARS): a powerful tool for predictive data mining | |
| CN117726460A (zh) | 一站式理赔方法、系统、终端及存储介质 | |
| Pevec et al. | Prediction intervals in supervised learning for model evaluation and discrimination | |
| CN120703012A (zh) | 水质检测数据处理方法、装置、计算机设备和存储介质 | |
| Yang et al. | Double robust conditional independence test for novel biomarkers given established risk factors with survival data | |
| CN119129841A (zh) | 一种地区能源经济预测方法、设备及存储介质 | |
| CN119539177A (zh) | 基于主成分回归模型的经济预测方法及系统 | |
| CN119598858A (zh) | 一种主板性能优化方法及优化系统 | |
| CN119596856A (zh) | 一种基于虚拟量测预测数据的生产控制方法、设备、介质及产品 | |
| US20250181672A1 (en) | Trial production condition proposal system and trial production condition proposal method | |
| Han et al. | RUL prediction method based on sequential health index evaluation with multidimensional coupled degradation data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 23871412 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2024549786 Country of ref document: JP |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 112023004076 Country of ref document: DE |
|
| WWP | Wipo information: published in national office |
Ref document number: 112023004076 Country of ref document: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 23871412 Country of ref document: EP Kind code of ref document: A1 |