JP2022173108A - Information processing apparatus and program - Google Patents

Abstract

To provide a technology for assisting a user who is inexperienced in statistical processing, etc., to receive useful information on production conditions.SOLUTION: A raw data acquisition unit 80 acquires production condition information and specification information transmitted from a user device 2. A learning result generation unit 86 executes processing on statistical machine learning on raw data, and generates a prediction model formed by reflecting statistical characteristics in the raw data, and various parameters, as leaning results. A contribution ratio calculation unit 106 calculates a contribution ratio on specification variables, for each of production condition variables to be learned. A contribution variable specifying unit 108 specifies a section based on the contribution ratio calculated by the contribution ratio calculation unit 106, regarding each of production condition variables, such as individually low contribution variables or totally low contribution variables for each of the specification variables. A response curve presentation unit 122 draws a response curve to be presented.SELECTED DRAWING: Figure 3

Description

The present invention relates to an information processing apparatus and program.

2. Description of the Related Art Conventionally, in the manufacturing industry, etc., it is not easy to determine the characteristics of a product suitable for the purpose and the optimum manufacturing process conditions of the product. Therefore, in order to determine the optimum manufacturing conditions, a method that requires a great deal of cost and time, such as creating many prototypes and making prototypes, has been adopted.
For example, in Patent Document 1, the characteristics of the intermediate product and the characteristics of the metal product at each step of a plurality of manufacturing processes for manufacturing the metal product are calculated using a group of formulas and output. Methods have been proposed for determining suitable manufacturing conditions for manufacturing metal products.

JP 2020-184196 A

However, according to the conventional technology including the technology described in the above-mentioned Patent Document 1, the predicted value indicating the characteristics of the metal product is simply calculated based on the values output for each of the plurality of manufacturing processes. It is difficult to handle unless you are good at statistical processing using . Furthermore, it is insufficient to determine manufacturing conditions corresponding to the wide variety of situations in the manufacturing industry.
Moreover, it is useful for users engaged in the manufacturing industry not only to know the optimum manufacturing conditions, but also to know the range of allowable manufacturing conditions in consideration of other external factors. The prior art including the above-mentioned Patent Literature 1 cannot meet such demands of users.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a technique for assisting the provision of useful information to the manufacturing industry, including users who are unfamiliar with statistical processing and the like.

In order to achieve the above object, an information processing device according to one aspect of the present invention includes:
An information processing device used for arithmetic processing related to the manufacture of a predetermined article,
Primary data acquisition means for acquiring, as primary data, information on manufacturing conditions of the article and information on the quality or properties of the article to be manufactured;
learning result generation means for executing a learning process having statistical properties on the primary data and generating a predetermined algorithm reflecting the statistical properties included in the primary data;
response curve generating means for generating a diagram or table that allows visual confirmation of the relationship between the variables related to the manufacturing conditions and the variables related to the quality or properties of the article, based on the results of the learning process;
Prepare.

A program of one aspect of the present invention is also provided as a program corresponding to the information processing apparatus of one aspect of the present invention.

ADVANTAGE OF THE INVENTION According to this invention, the technique which assists provision of information useful to manufacturing industry etc. including the user who is unfamiliar with statistical processing etc. can be provided.

It is a figure showing composition of an information processing system concerning one embodiment of the present invention. 2 is a block diagram showing an example of a hardware configuration of a server in the information processing system of FIG. 1; FIG. FIG. 3 is a functional block diagram showing an example of a functional configuration related to learning processing among the functional configurations of the server of FIG. 2 and the user device of FIG. 1; FIG. 4 is a diagram showing an example of an upload screen displayed on the user device of FIG. 3; FIG. 4 is a diagram showing an example of a learning setting screen displayed on the user device of FIG. 3; FIG. 4 is a diagram showing an example of a learning result display screen displayed on the user device of FIG. 3; FIG. FIG. 7 is a diagram showing an example of a learning result display screen displayed on the user device of FIG. 3, and showing an example different from the example of FIG. 6; 4 is a flowchart for explaining the flow of learning processing executed by the server of FIG. 3; FIG. 3 is a functional block diagram showing an example of a functional configuration related to a recipe generation process among the functional configurations of the server of FIG. 2 and the user device of FIG. 1; It is a figure which shows an example of the recipe list display screen displayed on the user device of FIG. 10. It is a figure which shows an example of the recipe list display screen displayed on the user device of FIG. 3, and is a figure which shows an example different from the example of FIG. It is a figure which shows an example of the recipe detailed display screen displayed on the user device of FIG. FIG. 12 is a diagram showing an example of a recipe list display screen displayed on the user device of FIG. 3, and is a diagram showing an example different from the examples of FIGS. 10 and 11. FIG. FIG. 4 is a flowchart for explaining the flow of recipe generation processing executed by the server of FIG. 3; FIG. FIG. 4 is a diagram showing an example of an image regarding learning progress displayed on the user device of FIG. 3 ; FIG. FIG. 4 is a diagram showing an example of an image related to a learning result displayed on the user device of FIG. 3; FIG.

<Description of overview>
An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing the configuration of an information processing system according to one embodiment of the present invention.

Before describing FIG. 1, an outline of a learning service, which is the first service to which the information processing system according to the embodiment of the present invention is applied, will be described.
Manufacturing conditions can be expressed by one or more variables (hereinafter referred to as "manufacturing condition variables") that represent the characteristics of the product or the conditions of the manufacturing process, and are uniquely determined or specified by the value of the manufacturing condition variable be done.
In general, product specifications change according to manufacturing conditions, and variables representing these specifications are hereinafter referred to as "specification variables." Here, the specification of a product is a product characteristic focused on in order to determine whether or not the performance of the product and the purpose to be achieved by the product can be achieved.
The learning service uses statistical methods such as machine learning to learn the relationship between manufacturing condition variables and specification variables, and to generate learning results (so-called learned models) that can estimate unknown manufacturing conditions. A set of services to generate.
That is, in the learning service, sample data (hereinafter referred to as "raw data") related to manufacturing conditions and specifications provided by the user is subjected to learning processing by methods such as statistical machine learning, and as a result, The learning result thus obtained is used as a prediction model for estimating product specifications that can be produced under unknown manufacturing conditions.
In addition, the learning service does not simply generate learning results from raw data, and even users who do not have knowledge of statistics can easily interpret the learning results. subsumed.
Therefore, the learning service can be easily used by users who are generally assumed to be in the manufacturing industry and who specialize in physical properties without knowledge of statistics. Furthermore, the user can use the learning result more freely by using his/her specialized knowledge such as physical properties.

As shown in FIG. 1, the system includes a server 1 and user devices 2 . The server 1 and the user device 2 are interconnected via a predetermined network N such as the Internet.
Note that the network N is not an essential component, and may use, for example, NFC (Near Field Communication), Bluetooth (registered trademark), LAN (Local Area Network), or the like.

<Hardware configuration>
FIG. 2 is a block diagram showing an example of a hardware configuration of a server in the information processing system of FIG. 1. As shown in FIG.
The server 1 is composed of a personal computer or the like. As shown in FIG. 2, the server 1 includes a control unit 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a bus 14, an input/output interface 15, an output unit 16, an input A unit 17 , a storage unit 18 , a communication unit 19 and a drive 20 are provided.

The control unit 11 includes a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), and a microcomputer including a semiconductor memory. Various processes are executed according to the program.
The RAM 13 also appropriately stores information necessary for the control unit 11 to execute various processes.

Control unit 11 , ROM 12 and RAM 13 are interconnected via bus 14 . An input/output interface 15 is also connected to this bus 14 . An output unit 16 , an input unit 17 , a storage unit 18 , a communication unit 19 and a drive 20 are connected to the input/output interface 15 .

The output unit 16 includes various liquid crystal displays, speakers, and the like, and outputs various information as images and sounds.

The input unit 17 is composed of a keyboard, a mouse, etc., and inputs various kinds of information.

The storage unit 18 is configured by an HDD (Hard Disk Drive), an SSD (Solid State Drive), or the like, and stores various data. In this embodiment, for example, various information including various programs and various databases are stored.

The communication unit 19 controls communication with another device (for example, the user device 2) via a network N including the Internet.

A drive 20 is provided as required. A removable medium 31 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is mounted in the drive 20 as appropriate. A program read from the removable medium 31 by the drive 20 is installed in the storage section 28 as necessary. The removable medium 31 can also store various data stored in the storage unit 18 in the same manner as the storage unit 18 .

Note that the hardware configuration of the user device 2 can be basically the same as the hardware configuration of the server 1, so description thereof will be omitted here.

Next, among the functional configurations of the server 1, the functional configuration during execution of various processing related to the learning service (hereinafter referred to as “learning processing”) will be described.
FIG. 3 is a functional block diagram showing an example of a functional configuration related to learning processing among the functional configurations of the server of FIG. 2 and the user device of FIG.

During the learning process, only the learning processing section 60 of the server 1 functions, and the recipe generation processing section 140 does not function. As shown in FIG. 3, the learning processing unit 60 of the server 1 includes a raw data acquisition unit 80, a data content confirmation unit 82, a learning setting acquisition unit 84, a learning result generation unit 86, and a learning result presentation unit 88. and are provided.
In one area of the storage unit 18 of the server 1, a learning result DB 500 and a recipe generation result DB 600 are provided. The learning result DB 500 stores various learning results generated as a result of learning processing (prediction models generated as a result of learning, various parameters, etc.).

The raw data acquisition unit 80 of the server 1 acquires information on the manufacturing conditions of the article (hereinafter referred to as "manufacturing condition information") and information on the specifications of the article to be manufactured (hereinafter referred to as "spec information"). Get it as data.
That is, the raw data acquisition unit 80 acquires the manufacturing condition information and the specification information input via the user device 2 and transmitted from the user device 2 as raw data via the communication unit 19 .

The data content confirmation unit 82 confirms whether the raw data acquired by the raw data acquisition unit 80 has conditions suitable for information (input information) used for learning, which will be described later.
Specifically, the data content confirmation unit 82 determines whether or not the content and amount of information included in each of the manufacturing condition information and the specification information are appropriate for learning, which will be described later. Note that the data content confirmation unit 82 can make the above-described determination by any method, but in the present embodiment, the above-described determination can be performed based on, for example, a standard using an orthogonal array in the experimental design method. It should be noted that the data content confirmation unit 82 does not need to confirm the data content when acquiring the raw data, and appropriately confirms the data content in parallel with the learning process.

The learning setting acquisition unit 84 acquires setting information for learning performed on raw data. The learning setting information is, for example, specification of various explanatory variables and objective variables, specification items related to various learning environments, and the like, as will be described later in conjunction with the example of FIG.

The learning result generation unit 86 performs processing related to statistical machine learning on the raw data, and generates a prediction model and various parameters reflecting the statistical properties of the raw data as learning results. The learning result generation unit 86 stores the generated learning result, various information regarding the contribution rate calculated by the contribution rate calculation unit 106, and various information regarding classification of the contribution rate in the learning result DB 500. FIG.
Here, the method of statistical machine learning (for example, deep learning, clustering, etc.) used by the learning result generation unit 86 is arbitrary, but in the present embodiment, for example, a neural network, a regression model by sparse modeling, etc. are combined. to generate learning results.

Further, the learning result generation unit 86 is provided with a data preprocessing unit 100 , a prediction model generation unit 102 , a prediction accuracy calculation unit 104 , a contribution ratio calculation unit 106 , and a contribution variable identification unit 108 .

The data preprocessing unit 100 performs processing on variables set as learning targets so that prediction models with higher accuracy can be expected to be generated.

The predictive model generation unit 102 executes one or more statistical machine learning methods and employs a trained model with the best accuracy as a predictive model. At this time, depending on the method of statistical machine learning, a relational expression between manufacturing condition variables and specification variables can be obtained.

The prediction accuracy calculator 104 calculates the value of the accuracy index of the prediction model (coefficient of determination, correlation coefficient, root-mean-square error, etc. between measured values and predicted values given by raw data).

The contribution rate calculation unit 106 calculates the contribution rate to the specification variable for each of the manufacturing condition variables that are the targets of learning.

As a result, in the case where there is no prior knowledge of which manufacturing condition variables have little relation to specification determination, knowing this can be useful knowledge. Data that can be manufacturing condition variables without prejudice are added to the raw data, and in the process of learning processing, the contribution rate of each manufacturing condition variable is calculated in each specification variable, and at the same time in all specifications Unrelated manufacturing condition variables can be known.

Here, with respect to the contribution rate calculated by the contribution rate calculation unit 106, a manufacturing condition variable that contributes less than a predetermined reference value to one spec variable is called an individual low contribution variable.
Also, a manufacturing condition that contributes less than a predetermined reference value to all specification variables is called a completely low-contribution variable. That is, the concept of contribution rate is classified into individual low-contribution variables and completely low-contribution variables in terms of the contribution rate to "what".

Therefore, the contribution variable identification unit 108 identifies categories based on the contribution ratio calculated by the contribution ratio calculation unit 106 for each manufacturing condition variable, such as individual low contribution variables and complete low contribution variables for each specification variable. .
Based on the categories identified in this way, by excluding variables with low contributions and generating a forecast model, it is possible to generate a forecast model that is more accurate, versatile, and easy to interpret. Be expected.

Note that the categories such as individual low contribution variables and complete low contribution variables specified by the contribution variable specifying unit 108 may be presented to the user or the like as part of the learning results by the learning result presentation unit 88 described later. The user can determine the manufacturing conditions and future data collection policies while confirming the presented categories.

In addition, the learning result generation unit 86 stores the generated learning result, information on the contribution rate calculated by the contribution rate calculation unit 106, and information on the contribution rate category specified by the contribution variable specifying unit 108 in the learning result DB 500. Store.

The learning result presentation unit 88 presents various types of information about the learning results to the user via the user device 2 . The learning result presentation unit 88 is provided with a prediction model accuracy presentation unit 120 and a response curve presentation unit 122 .

The prediction model accuracy presentation unit 120 uses the learning result generated by the learning result generation unit 86, the information on the accuracy evaluation calculated by the prediction accuracy calculation unit 104, the contribution rate calculated by the contribution rate calculation unit 106, and other contributing variable identification. Various types of information regarding the categories of the contribution rates specified by the unit 108 are presented to the user via the user device 2 .

The response curve presentation unit 122 draws and presents the response curve presented as part of the learning result by the learning result presentation unit 88 .
The prediction model can be regarded as a function by using manufacturing condition variables as input values and outputting prediction values (prediction model values). At this time, if there are n (n is an arbitrary natural number equal to or greater than 1) manufacturing condition variables, the response characteristic of the predicted value to the manufacturing conditions is expressed as an n+1-dimensional hyperplane.
However, since it is difficult to recognize a plane that exceeds three dimensions in visualization, it is common to reduce the dimensions and draw. Specifically, by using one of the manufacturing condition variables as a variable and setting predetermined fixed values to the other variables, it is possible to draw a graph as a function of one variable. By arbitrarily changing this fixed value, the user can confirm the behavior of graphs of various cross sections of the prediction model as a multivariable function, and use it to determine manufacturing conditions and consider future development policies. can.

Here, the response characteristic drawn by the response curve presentation unit 122 can be drawn as a three-dimensional graph (two-variable function), but is preferably drawn as a two-dimensional graph (one-variable function). A graph (response surface) showing the relationship between variables is generally drawn not as a two-dimensional (single-variable function) but as a three-dimensional (two-variable function) graph in which two variables are selected. The user confirms such changes in the graph while operating a mouse or the like.
On the other hand, the response curve presentation unit 122 selects only one variable, draws a two-dimensional graph (one-variable function), changes the values of the other variables with a slider bar or the like, and presents the user with a change in the curve. can be confirmed. This allows the user to more easily and accurately grasp the characteristics, contribution rate, etc. of each variable.
Note that the response curve presentation unit 122 may draw the above-described response curve after normalizing the value of each variable.

Next, with reference to FIGS. 4 and 5, the user's specific operations related to the user device 2 when using the study service will be described.
4 is a diagram showing an example of an upload screen displayed on the user device of FIG. 3. FIG.
5 is a diagram showing an example of a learning setting screen displayed on the user device of FIG. 3. FIG.

In the example of FIG. 4, the message "select file" and the image of the file "0000.csv" are displayed in the center of the screen. The user operates the user device 2, moves a file in any format to a part of the display area (eg, drag and drop), and presses (eg, clicks, etc.) the display area labeled "file upload". Thus, raw data used for learning can be uploaded to the server 1 .
Similarly, in the example of FIG. 5, specific variables included in the raw data uploaded via the screen of FIG. , "Height", "Angle", "Attenuation rate", "Defective product rate", "Cost", "Durability", "Energy conversion efficiency", and "Touch feeling" are displayed.
In addition, in the example of FIG. 5, for each variable, the user selects "use for learning", "variable type (explanatory variable or objective variable)", "numerical value/category", and "allowable error". , "value range", "remarks", etc., for use in the learning process can be set or described. In the example of FIG. 5, "No.", "Injection rate", "Expansion rate", "Temperature", "Width", "Height", "Angle", and "Attenuation rate" are explanatory variables, and "Defect rate , "Cost", "Durability", "Energy Conversion Efficiency", and "Touch Feel" are set as objective variables.
Here, the objective variable and the explanatory variable are terms used in regression analysis, etc. The variable that you want to explain as a variable value is called the objective variable, and each variable that is set to explain the objective variable is called the objective variable. called an explanatory variable. Therefore, in this embodiment, typically, each of the manufacturing condition variables described above corresponds to explanatory variables, and the spec variables of the manufactured product correspond to objective variables.
While confirming such a screen, the user can arbitrarily set the conditions for each variable and generate the learning result, which makes it easier to obtain the learning result that meets his or her desires.

Next, the details of the learning result display screen generated after the processing such as the learning setting described above will be described with reference to FIGS. 6 and 7. FIG.
6 and 7 are diagrams showing an example of a learning result display screen displayed on the user device of FIG. 3. FIG.

In the example of FIG. 6, the approximation formula related to the prediction model as the learning result, the contribution rate of each manufacturing condition variable, the classification related to the contribution rate of the manufacturing condition variable, the information related to the accuracy evaluation of the learning result of the learning process, etc. are displayed. there is
Specifically, in the example of FIG. 6, the formula of Formula 1 below is displayed as an approximation formula related to the prediction model, and "injection rate", "expansion rate", "temperature", "width", "height", " The contribution ratios of the parameters "angle" and "attenuation rate" are "16.9%", "13.3%", "2.03%", "2.01%", "1.50%", and "1 .20%" and "0.40%" are displayed.

In the graph relating to the accuracy evaluation of the learning result, the value of the coefficient of determination is displayed on the vertical axis, and the progress (step) of the learning progress is displayed on the horizontal axis. By checking such a graph, the user can easily grasp the variation in accuracy and the presence or absence of generalization performance in the learning process, and can also use it to judge whether the learning result is appropriate or not.

Next, in the example of FIG. 7, a forecast scatter diagram and a histogram of forecast error distribution are shown for each of the learning data and the verification data. The predicted actual scatter diagram is a scatter diagram in which the horizontal axis indicates predicted values based on learning results, and the vertical axis indicates actual measured values. Therefore, the closer each point is to the diagonal line, the higher the accuracy, and the farther from the diagonal line, the lower the accuracy. The graph of the expected scatterplot shows the corresponding accuracy index. Also, the prediction error distribution histogram shows the distribution of the error between the measured values and the predicted values. Also, the histogram of the prediction error distribution is displayed in association with a statistical index representing the magnitude of the prediction error distribution. Thus, the user can visually confirm the accuracy of the obtained model by confirming the histogram of the prediction error distribution in FIG.
In addition, in the example of FIG. 7, for example, the concept of allowable error can be reflected on the horizontal axis. This makes it easier for the user to visually confirm the accuracy of the model.

Here, in statistical machine learning such as neural networks, if the degree of dependence on learning data, which is the data used for learning, is too high, the statistical properties of the learning data will be excessively reflected in the learning results. So-called over-learning becomes a problem.
The verification data shown on the right side of FIG. 7 is an example of test data set aside at a predetermined ratio from the raw data in order to cope with such overlearning. Such test data is not directly used for learning, but used for verification of learning results. The engine (algorithm) generated as a result of statistical machine learning does not overly fit the training data, and a model with good overall results for both training data and validation data can be used for general purposes. It is considered to be a predictive model that is easy to
The user can confirm that the generated learning result is actually a general-purpose model by confirming the graph or the like shown in FIG. The server 1 may adjust the length of the horizontal axis and the like so that the narrowness of the distribution of the histogram can be compared with the target width, the allowable error, and the like, which will be described later.
Also, if the user determines that there is a problem with the results shown in the graph, he or she can adopt a different model from among the generated learning results, change the parameter settings, or change the raw data to be uploaded. etc., a more versatile and highly accurate learning result can be generated.
Each graph shown in FIG. 7 is a scatter diagram showing the distribution of predicted values and actual values in learning data (expected actual scatter diagram), and a scatter diagram showing the distribution of predicted values and actual values in verification data (expected actual scatter diagram). , a diagram showing the distribution of prediction errors in learning data (prediction error distribution diagram), and a diagram showing the distribution of prediction errors in verification data (prediction error distribution diagram). However, each graph shown in FIG. 6 and FIG. 7 is an example, and is an example of a diagram or a table that can contribute to the user confirming the contents and results of learning.

Next, the flow of learning processing executed by the server having the functional configuration of FIG. 3 will be described.
FIG. 8 is a flow chart explaining the flow of learning processing executed by the server of FIG.

In step S<b>1 , the raw data acquisition unit 80 acquires manufacturing condition information regarding the manufacturing conditions of the article and target specification information regarding the properties of the article to be manufactured, that is, raw data transmitted from the user device 2 .

In step S2, the data content confirmation unit 82 confirms whether the raw data acquired by the raw data acquisition unit 80 has conditions suitable for information (input information) used for learning, which will be described later.

In step S3, the learning setting acquisition unit 84 acquires setting information for learning performed on the raw data. At that time, the consistency of the learning setting information is confirmed.

In step S4, the data preprocessing unit 100 performs processing on the variable set as the target of learning so that a prediction model with higher accuracy can be expected to be generated.

In step S5, the learning result generation unit 86 executes processing related to statistical machine learning on the raw data, and generates a prediction model and various parameters reflecting the statistical properties of the raw data as learning results. .

In step S<b>6 , the contribution calculation unit 106 calculates the contribution to the spec variables for each of the learning target manufacturing condition variables.

In step S7, the contribution variable identification unit 108 classifies each manufacturing condition variable, such as an individual low contribution variable or a completely low contribution variable for each specification variable, based on the contribution rate calculated by the contribution rate calculation unit 106. Identify. Note that the user who has confirmed the identified categories and the like may generate learning results by excluding low-contribution variables and the like.

In step S8, the learning result presenting unit 88 presents the learning result generated by the learning result generating unit 86, the contribution rate calculated by the contribution rate calculating unit 106, and various information related to the division specified by the contributing variable specifying unit 108. Send to user device 2 .

In step S9, the response curve presentation unit 122 draws and presents a response curve. Specifically, in step S<b>9 , the response curve presenting unit 122 draws a response curve and transmits to the user device 2 various kinds of information necessary for presenting the response curve. With this, the learning process ends.

Next, an overview of the recipe generation service, which is the second service to which the information processing system according to the embodiment of the present invention is applied, will be described.
The recipe generation service uses the learning results generated by using the above-mentioned learning service etc. to create specific manufacturing conditions for manufacturing products based on the quality and characteristics of the product desired by the user. A set of services that presents candidates to the user.
In the recipe generation service, for the target specifications of the product set by the user, based on the learning results stored in the learning result DB 500, manufacturing condition candidates (hereinafter referred to as , called a “recipe”) to the user. Then, the user can select the optimum manufacturing conditions for himself/herself from the presented recipes and attempt manufacturing.

9 is a functional block diagram showing an example of a functional configuration related to a recipe generation process among the functional configurations of the server of FIG. 2 and the user device of FIG. 1. FIG. With reference to FIG. 9, the functional configuration during execution of various processes related to the recipe generation service (hereinafter referred to as "recipe generation processing") will be described.

During recipe generation processing, only the recipe generation processing unit 140 of the server 1 functions, and the learning processing unit 60 does not function. As shown in FIG. 9 , the recipe generation processing unit 140 of the server 1 is provided with a desired condition acquisition unit 160 , a recipe generation unit 162 , and a recipe generation result presentation unit 164 .

Here, the desired condition acquisition unit 160 of the server 1 is provided with a target specification value acquisition unit 180 , a target width acquisition unit 182 , and a constraint condition acquisition unit 184 .
The desired condition acquisition unit 160 confirms whether the desired conditions such as target specification value information, target width information, and constraint information, which will be described later, are consistent.

The target spec value acquisition unit 180 acquires information on the target specs of the product for which the user wishes to generate a recipe (hereinafter referred to as "target spec value information at recipe generation").

The target width acquisition unit 182 acquires information (hereinafter referred to as “target width information”) on the range (width) of the spec value that the user wants to target.

The constraint condition acquisition unit 184 acquires information (hereinafter referred to as “constraint condition information”) regarding conditions to be satisfied by the manufacturing condition variables and the specification variables in the recipe.
Note that the target specification value information at the time of recipe generation, the target width information, and the constraint information are input by a user or the like and transmitted from the user device 2 to the server 1 .

The recipe generation unit 162 generates a recipe, which is a combination of variable values that define manufacturing conditions for an article that suggests the possibility of satisfying the desired condition, based on the results of statistical processing of the primary information.
That is, the recipe generation unit 162 obtains target spec value information at recipe generation acquired by the target spec value acquisition unit 180, target width information acquired by the target width acquisition unit 182, constraint conditions acquired by the constraint condition acquisition unit 184, and Based on the information and learning results stored in the learning result DB 500, a recipe that can satisfy the target specifications desired by the user is generated.
Note that the recipe generation unit 162 can generate a recipe by any method, but in the present embodiment, for example, a genetic algorithm is used to generate the recipe.

Also, the recipe generation unit 162 is provided with a learning result management unit 200 , a recipe search unit 202 , and a clustering unit 204 .

The learning result management unit 200 acquires and manages learning results (prediction models generated as a result of learning, various parameters, etc.) generated as a result of the above-described learning process stored in the learning result DB 500 . Various types of information managed here are used, for example, to calculate the scores of recipe candidates in the recipe generation process.

The recipe search unit 202 searches the recipe candidate space based on the recipe generation target spec information, target width information, constraint information, learning results, and the like, and generates a recipe (recipe group).

The clustering unit 204 performs clustering processing on the results searched by the recipe search unit 202, and classifies the obtained recipe groups.
Specifically, the clustering unit 204 performs, for example, non-hierarchical clustering, selects the cluster with the highest score as a representative of each cluster, and classifies it as one type of recipe.
Thereby, the recipe generation unit 162 can efficiently exclude recipes that are excessively similar and present various recipe options to the user. For clustering by the clustering unit 204, an input allowable error or aggregate width may be used.

The recipe generation result presenting unit 164 generates output information to be presented to the user based on the recipe generated by the recipe generating unit 162 and the like.

Here, the recipe generation result presentation unit 164 is provided with a recipe list generation unit 220, a radar chart generation unit 222 that displays scores by spec variable, a response curve generation unit 224, and a recipe difference graph generation unit 226. there is

The recipe list generation unit 220 generates a table for listing recipe values, corresponding predicted values, and the like. For each recipe, a predicted value and a score based on the learning result are obtained, and can be used as an index for the order of arrangement.

Each recipe can be scored for each target spec value set for each spec variable. At this time, it is possible to perform scoring considering not only the degree of proximity to the target spec, but also the target width and the like.
The spec variable score radar chart generation unit 222 presents the score for each spec as a radar chart. For example, an example of a radar chart will be described later with reference to FIG.

The response curve generator 224 draws and presents a response curve. The prediction model can be regarded as a function by using manufacturing condition variables as input values and outputting prediction values (prediction model values). Specifically, by using one of the manufacturing condition variables as a variable and setting predetermined fixed values to the other variables, it is possible to draw a graph as a function of one variable. By arbitrarily changing this fixed value, the user can confirm the behavior of graphs of various cross sections of the prediction model as a multivariable function, and use it to determine manufacturing conditions and consider future development policies. can. For example, an example of a response curve is described below with reference to FIG.

The recipe difference graph generation unit 226 generates information about the difference graph (hereinafter referred to as “difference graph information”). A concrete difference graph will be described with reference to FIG. 13 .

Next, operations and the like when the user confirms the generated recipe will be described with reference to FIGS. 10 to 13. FIG.
First, FIGS. 10 and 11 are diagrams showing an example of a recipe list display screen displayed on the user device of FIG.

In the example of FIG. 10, the generated recipe "number (recipe No.)", "graph", "average score", "manufacturing condition variables (injection rate, expansion rate, temperature, width, height, angle, attenuation Recipes R1 to R5 are shown as specific examples of the "value of rate".
Note that, in the example of FIG. 10, similarly to the example shown in FIG. Attenuation rate" is the average value set of manufacturing condition variables when "defective product rate", "cost", "durable strength", "energy conversion efficiency", and "touch feeling" are set as objective variables. They are displayed in descending order of score.
Here, the "average score" indicates the average value of the scores for each of the target specs calculated by an arbitrary method set in recipe generation. It is considered that there is a high possibility that a product with specifications close to the target specifications can be manufactured. For example, in the example of FIG. 10, the average score of recipe R1 is calculated as "87.6", and compared with other recipes, it is a promising recipe that is highly likely to produce a product close to the target specifications desired by the user. I can guess the recipe.

In addition to the average score, the order of the recipes can also be rearranged according to other indicators for evaluating the recipes, such as individual scores for the specifications of interest.

Furthermore, "graph" will be described with reference to FIG. The graph shown in FIG. 11 can visualize the score for each of the target specification variables of "defective product rate", "cost", "durability strength", "energy conversion efficiency", and "touch feeling", using recipe R1 as an example. It is graphed as For example, the user can transition to a screen showing the details of the graph shown in FIG. 11 by operating (clicking, etc.) the display of the graph of each recipe in the "recipe list" shown in FIG.

Next, FIG. 12 is a diagram showing an example of a recipe details display screen displayed on the user device of FIG. An example of a user's operation on the recipe details display screen will be briefly described with reference to FIGS.
The response curve of FIG. 12A shows how the target spec (endurance strength) changes when the injection speed (explanatory variable) is changed, focusing on the injection speed parameter of the recipe. .
Here, looking at the response curve in FIG. 12A, there is a range T1 as a range of injection rates that can take a value of 120 set as the target spec (endurance strength). The predicted value of the endurance strength at the injection speed included in the range T1 is included in the range of the target specifications and target width defined by the user. In other words, if the user can control the injection rate within the range T1, it is possible to manufacture a product that satisfies the target specifications. However, as described above, during the manufacturing process of the product, the injection rate (explanatory variable) may change due to unexpected considerations as well as changes in other variables. Therefore, the wider the injection rate range that can satisfy the target specifications, the better.

In such a case, the user can use the slider bar shown in FIG. 12(B). As shown in FIG. 12B, the user can change the content of the response curve by freely adjusting any parameter (in the example of FIG. 12, the expansion rate). In the example of FIG. 12B, the user can adjust the response curve to be displayed by arbitrarily adjusting the slider bar SB within the range of 1.0 to -1.0.

A response curve generated as a result is, for example, the response curve shown in (C) of FIG. The response curve of FIG. 12(C) includes a range T2 as the injection rate range that satisfies the target specifications. This range T2 is wider than the previous range T1. Therefore, by referring to the recipe shown in FIG. 12C and setting the value of the injection rate within the range T2 and near the center, even if the injection rate changes somewhat, It is considered highly likely that the user will be able to manufacture a product within the range of the target width. In this way, the user takes into consideration changes in the response curve so that the user can efficiently manufacture a product that satisfies the target specifications even with slight changes in the explanatory variables in the manufacturing process of the product. You can arbitrarily decide which recipe to use.

Here, the response curve generated by the response curve generator 224 shown in FIG. A brief explanation of the differences in the response curves”).
This system provides a recipe generation function as a means for discovering a combination of variables that satisfies the user's desired target specifications. The recipe generation function of this system freely searches for data in all ranges and predicts recipes. Since the search range is wide, processing tends to take a long time.
By using the post-recipe-generation response curve and the learning-response curve depending on the situation, the user can find combinations of variables more accurately and efficiently than when using the recipe generation function alone. there is a possibility.
For example, when the user wants to not only discover the desired combination of variables, but also check whether the target specifications can be met even if each variable changes slightly (robustness). By using the post-recipe generation response curve, it is possible to check the robustness of each generated recipe. This may allow the user to find a more robust recipe among multiple candidates.
On the other hand, if the user has an idea of the manufacturing conditions that can satisfy the target specifications to some extent in advance (for example, if there is a recipe that has been used before), or if the user wants to prioritize time over confirmation of robustness, etc. If so, users can make their own predictions by using learning result response curves instead of executing recipe generation that takes a long time to process. This may allow the user to find the desired recipe more quickly than using the recipe generation function. In this way, the user can selectively use the learning result response curve and recipe generation according to his/her purpose and situation.

Next, the dissimilarity graph in this embodiment will be described with reference to FIG. 13 .
13 is a diagram showing an example of a recipe list display screen displayed on the user device of FIG. 3, and is a diagram showing an example different from the examples of FIGS. 10 and 11. FIG.
In the example of FIG. 13, the values of manufacturing condition variables (injection rate, expansion rate, temperature, . . . ) are represented as line graphs for each of recipes R1 to R5. Values of variables other than numerical variables, such as categorical variables, can also be drawn in the same graph if the order is defined. The user can confirm the diversity of a plurality of recipes by confirming the presented graph.
By confirming the graph presented in this way, the user can fully consider which recipe is truly suitable for his/her purpose. Note that the reason why the normalized values are used in the graph shown in FIG. 13 is to facilitate comparison of the diversity of recipes.

Next, the flow of recipe generation processing executed by the server having the functional configuration of FIG. 9 will be described.
FIG. 14 is a flow chart illustrating the flow of recipe generation processing executed by the server of FIG.

In step S21, the target specification value acquisition unit 180 acquires target specification value information at recipe generation related to the target specifications of the product for which the user desires recipe generation.
That is, the target specification value acquisition unit 180 acquires the target specification value information at recipe generation transmitted from the user device 2 via the communication unit 19 .

In step S22, the target width acquisition unit 182 acquires target width information regarding the range (width) of the spec value that the user wants to target.
That is, the target width acquisition unit 182 acquires target width information transmitted from the user device 2 via the communication unit 19 .

In step S<b>23 , the constraint acquisition unit 184 acquires constraint information regarding the conditions that the manufacturing condition variables and the specification variables satisfy in the recipe.
That is, the constraint acquisition unit 184 acquires the constraint information transmitted from the user device 2 via the communication unit 19 .

In step S24, the desired condition obtaining unit 160 confirms whether the desired conditions such as the target specification value information at the time of recipe generation, the target width information, and the constraint information are consistent.

In step S<b>25 , the recipe generation unit 162 acquires the learning result or the like generated as a result of the learning process described above and stored in the learning result DB 500 .

In step S26, the recipe search unit 202 searches the recipe candidate space based on the target spec information at recipe generation, target width information, constraint information, learning results, etc., and generates a recipe (recipe group).

In step S27, the clustering unit 204 performs clustering-related processing on the result obtained in step S26, and classifies the obtained recipe group.

In step S28, the recipe generation result presenting unit 164 generates output information to be presented to the user based on the recipe generated by the recipe generating unit 162 and the like. In step S<b>28 , the generated output information is transmitted to the user device 2 . With the above, the recipe generation processing ends.

Although one embodiment of the present invention has been described above, the present invention is not limited to the above-described embodiment, and modifications, improvements, etc. within the scope of achieving the object of the present invention are included in the present invention. be.

Further, examples of images displayed to the user for confirmation of learning results are shown with reference to FIGS. 15 and 16. FIG.
FIG. 15 is a diagram showing the degree of completion of learning and the process of progress. Specifically, in the example of FIG. 15, the horizontal axis indicates the number of times of learning, and the vertical axis indicates the accuracy of the generated trained model (the lower the value, the higher the accuracy of the trained model). . Therefore, in the example of FIG. 15, repeating the learning gradually improves the accuracy of the trained model, indicating that the learning is progressing normally.
Note that, for example, the various images showing the learning results shown in FIGS. 6 and 15 are examples. For example, this system may modify or replace all or part of the elements of various images shown in FIGS.

On the other hand, in the example of FIG. 16, the accuracy (five levels) of the generated trained model is displayed as an image diagram simulating the human evolution process. In the example of Fig. 16, along with an image diagram showing humans (final) in the process of human evolution, a message saying "The level 5 trained model has extremely high accuracy. We can also expect prediction results. Let's make a prediction." is displayed. That is, FIG. 16 shows an example of an image diagram shown when the accuracy of the generated trained model is extremely high. Note that in the example of FIG. 16, although the index for evaluating the trained model is arbitrary, the accuracy of the trained model may be evaluated based on, for example, the coefficient of determination and the predicted value coverage. The predicted value coverage rate is a value that indicates how much the error of the predicted value with respect to the measured value is within the allowable error range.

Here, the recipe generation in this system is supplemented.
Recipe generation in this system is effective in the first place when the user does not know the manufacturing conditions for each variable to satisfy the target specifications intended by the user. On the other hand, this system makes predictions based on statistical calculations multiple times, scores based on the distance from the target specifications, and presents recipes with high scores as recipes. .
However, depending on the accuracy of learning, etc., there is a risk that the result of the statistically calculated score will deviate from the desired product to the extent that it cannot be used at the manufacturing site. Therefore, the present system can take into account the target width, which includes the user's experience and subjectivity, in scoring. This allows the system to generate a recipe that facilitates the production of a user-acceptable product.
Moreover, when there are a plurality of target variables, it is conceivable that the target specifications cannot be sufficiently satisfied with only one recipe. Therefore, this system assumes specifications that a plurality of recipes that are as diverse as possible are presented, and that the user can select an easy-to-use recipe from among the presented recipes. Specifically, this system can employ a function of presenting as many recipes as possible to the user by combining genetic algorithms, non-hierarchical clustering, and the like.
In addition, even if such measures alone are not sufficient, it is possible to consider tolerances, aggregate widths, etc. so that the manufacturing conditions do not appear to be the same from the user's point of view, and the user can make a more visual judgment. To make this possible, the differences for each recipe can also be displayed graphically (such as a recipe difference graph). This allows the user to more visually confirm a wider variety of recipes.

Further, for example, in the recipe generation process in the above-described embodiment, the recipe is generated using the learning result generated during the learning process, but the invention is not particularly limited. The learning results used in the recipe generation process need not be those generated as the result of the learning process described above. Prediction models, relational expressions, standards, etc. that include the statistical properties of

Further, for example, although the description is omitted in the above-described embodiment (especially the embodiment of FIG. 5), the user sets the range for each objective variable or explanatory variable used for calculation when setting parameters, for example. good too. As a result, the user can study only within the scope desired by the user, thereby saving waiting time and memory consumption for unnecessary arithmetic processing.

Also, for example, in the above-described embodiment, the recipe difference graph generation unit 226 has been described as standardizing the respective parameters of the manufacturing conditions and the target specifications, but it is not particularly limited. That is, the recipe difference graph generation unit 226 may use any statistical method such as standardization or normalization to make it easier for the user to check.

Further, for example, in the above-described embodiment, the constraint information was explained as information about conditions to be satisfied by the manufacturing condition variable and the specification variable, but it is not particularly limited.
For example, the constraint information may be information about the recipe search range, other settings, and various other information about the conditions that the recipe should satisfy. Further, for example, the constraint information may include target width information, allowable error information, and the like.

Also, for example, although the description is omitted in the above-described embodiments, the quality and properties of the article are sufficient if they are characteristics that can be noticed when determining whether the performance and purpose of the article are realized, and are not particularly limited. not.

Also, for example, in the above-described embodiment (especially the embodiment of FIG. 12), the response curve generation unit 224 (or the response curve presentation unit 122) draws a response curve (two-dimensional), but it is not particularly limited.
The response curve generating unit 224 (or the response curve presenting unit 122) may draw a response surface (three-dimensional) according to the content of the learning result or the user's desire. For example, in the case of a response surface, two variables are selected from among n (n is any natural number equal to or greater than 1) manufacturing condition variables and drawn. -1)/2 patterns are conceivable, but in the case of a response curve, n graphs should be drawn. Note that the section of each graph may be changed, for example, by changing the fixed values shown below.
That is, the response surface and response curve are expressed by the following equations, for example.
Response surface: y=f(xa, xb, other xi fixed values)
Response curve: y=f(xa, other xi fixed values)

Further, for example, in the above-described embodiment (especially the embodiment of FIG. 12), the slider bar is used to adjust each variable, but the present invention is not limited to this. For example, the present system may adopt methods such as direct input of values, input by a pull-down method, etc., and adjust each variable.

Further, for example, the desired condition acquisition unit 160 has been described as acquiring each desired condition and the like from the user who wishes to generate a recipe, but is not particularly limited.
That is, the desired condition acquisition unit 160 may acquire desired conditions from a third party different from the user, for example. Furthermore, the desired condition acquisition unit 160 may automatically acquire each desired condition by implementing a program or the like without requiring intervention by a user or the like.

Here, although only a part of the description has been given in the above-described embodiment, each of the terms "allowable error", "target width", and "aggregate width" will be described in detail.
(1) Tolerance Tolerance in this system is the criterion for accuracy evaluation based on the difference between the predicted value of the model and the measured value. Therefore, the permissible error in this system does not simply have a statistical meaning, but may also be influenced by, for example, prior knowledge and subjectivity of the user. For example, if a given variable is rated as important by a manufacturing-savvy user, the tolerance is determined to reflect such assumptions independently of statistical variability. .
In other words, the allowable error of this system is the range (width) regarding whether or not the error between the actual measurement value and the predicted value of the spec variable can be regarded as the same spec in consideration of the prior knowledge and subjectivity of the user in this service. means Note that since this criterion takes into consideration the prior knowledge and subjectivity of the user, the criterion may change due to updates of various information, passage of time, and the like.
Specifically, the tolerances of the system can be used, for example, on the horizontal axis in the embodiment of FIG. Whether the prediction model is good or bad can be easily evaluated by the extent to which each peak of the histogram falls within the range of the allowable error.
In addition, the permissible error of this system can be used as a condition for judging the end of learning. For example, if the difference between the predicted value by the generated model and the measured value is smaller than the allowable error, learning may be terminated.
(2) Target Width (of Spec Variables) The target width means the range (width) of spec values that the user wants to target as a target, taking into consideration the user's prior knowledge, subjectivity, and the like. For example, the target width can be used for recipe evaluation and score calculation. It can be considered that the recipe is as good as it achieves the target spec value, but if it is within the target range, it can be evaluated that all of them have the same performance.
In addition, when actually manufacturing a product, it is not uncommon for the value of each variable to have an error due to some factor. Therefore, rather than simply setting the target specs to the best pre-determined values, the system generates (if possible) multiple recipes that fall within the user's acceptable range. It is designed to make it easy to select recipes that easily produce products within the target range, even if errors occur in the variables.
In other words, the target width can be used to confirm robustness together with the response surface (response curve). Having robustness means that the model exhibits high estimation performance even when outliers and the like are included (for example, see FIG. 12).
(3) Aggregate width (manufacturing condition variables) It means the range (width) regarding whether or not it can be aggregated as one recipe. For example, the aggregate width can be used for clustering or the like when generating recipes. As a result, the user can receive more effective recipe proposals in the process of recipe generation.
Note that the aggregate width is acquired by, for example, the following method.
(a) Obtained as a default fixed value.
(b) Taken as an arbitrary reference value entered by the user.
(c) Acquired as a default value using the contribution rate of the learning result.

Further, in the above-described embodiment, the contribution ratio calculation unit 106 has been described as calculating the contribution ratio to the spec variables, but the present invention is not limited to this. The contribution calculation unit 106 calculates a value that serves as an index for a wide range of interpretations, such as contribution and correlation.

Also, in the above-described embodiment, the contributing variable identification unit 108 generates a prediction model by excluding variables with low contributions, but the present invention is not limited to this. The contribution variable identification unit 108 may generate a prediction model by excluding low-contribution variables, or may generate a prediction model without excluding low-contribution variables.

Also, in the above-described embodiment, the user uploads the raw data, but the upload is not limited to this. For example, the raw data may be uploaded by another third party, or may be automatically obtained by the server 1 .

Further, in the above-described embodiment, an approximate expression related to the prediction model is generated and displayed as part of the learning result, but the present invention is not limited to this. That is, the generation and display of approximate expressions are arbitrary.

Also, in the above-described embodiment (especially the embodiment shown in FIG. 7), an example of the actual scatter diagram has been described, but the present invention is not limited to this. For example, the graph showing the relationship between predicted values and measured values is not limited to a scatter diagram, and may be any type of graph. Also, for example, the accuracy index corresponding to the graph is not limited to the coefficient of determination or the correlation coefficient, and any index may be adopted.
Similarly, the prediction error in the graph of the prediction error distribution may be any statistical index such as mean square error, root mean square error, mean absolute error, and the like.

Moreover, in the above-described embodiments, the preprocessing of the raw data (for example, step S4 in FIG. 8) is optional and not necessarily performed.

Also, in the above-described embodiment, the clustering unit 204 has been described as simply classifying recipes by clustering, but the present invention is not limited to this. For example, the clustering unit 204 can design various variations generally used in clustering, such as a grouping method, a distance calculation method, and whether or not there is a hierarchy. Furthermore, the clustering unit 204 is not limited to clustering classification, and may classify recipes using a similar statistical method that does not apply to clustering.

Also, for example, the series of processes described above can be executed by hardware or by software.
That is, the functional configurations of FIGS. 3 and 9 are merely examples and are not limiting. It is sufficient for the information processing system to have a function capable of executing the series of processes described above as a whole, and there is no particular limitation on what kind of functional block is used to realize this function. Also, the locations of the functional blocks are not limited to the examples in FIG. 3 or 9, and may be arbitrary.
Furthermore, one functional block may be composed of hardware alone, software alone, or a combination thereof.

When a series of processes is to be executed by software, a program constituting the software is installed in a computer or the like from a network or a recording medium.
The computer or the like may be a computer built into dedicated hardware. The computer or the like may be a computer capable of executing various functions by installing various programs, for example, a server, a general-purpose smart phone, or a personal computer.

A recording medium containing such a program not only consists of a removable medium (not shown) that is distributed separately from the device main body in order to provide the program to the user, but is also preinstalled in the device main body and delivered to the user. It may be composed of a provided recording medium or the like.

In this specification, the steps of writing a program recorded on a recording medium are not necessarily processed chronologically according to the order, but may be executed in parallel or individually. It also includes the processing to be executed.
That is, some of the steps in FIGS. 8 and 14 may be changed or omitted as appropriate.

Also, for example, in this specification, the term "system" means an overall device composed of a plurality of devices, a plurality of means, or the like.

In other words, the information processing apparatus to which the present invention is applied only needs to have the following configuration, and can take various forms.
That is, the information processing apparatus to which the present invention is applied is
An information processing device used for arithmetic processing related to the manufacture of a predetermined article,
Primary data acquisition means (for example, raw data acquisition unit 80) for acquiring, as primary data, information regarding manufacturing conditions of the article and information regarding the quality or properties of the article to be manufactured;
Learning result generation means (
For example, a learning result generator 86),
Based on the results of the learning process, response curve generating means (e.g., response curve a presentation unit 122);
It is sufficient to have

Further, the information processing apparatus may further include contribution rate calculation means for calculating the degree of contribution to the quality or properties of the article for each of the variables related to the manufacturing conditions.

Further, the response curve generating means selects one of the variables related to the manufacturing conditions and the variables related to the quality or property of the article, draws a two-dimensional graph, and draws a two-dimensional graph. The chart or the table can be generated in which values can be changed and changes in the curve observed.

The information processing apparatus may further include data content confirmation means for determining whether or not the content of the obtained primary data is necessary and sufficient for the learning process by the learning result generation means.

Further, the information processing apparatus may further include contribution variable identification means for determining whether or not each of the variables related to the manufacturing conditions contributes more than a predetermined value to each or all of the quality or the property. can.

Further, the information processing device can express a relationship between a plurality of variables including at least one of the variable related to the manufacturing condition and the variable related to the quality or property of the article based on the result of the learning process. A formula generating means for generating a formula may further be provided.

Further, the information processing apparatus may further include learning setting acquisition means for acquiring information regarding settings of the learning process.

Also, the information processing apparatus may further comprise tolerance information acquiring means for acquiring tolerance information regarding a difference between two different values in each of the variables relating to the quality or property of the article.

Further, the information processing apparatus can further include learning result confirmation means for generating information that can contribute to confirmation of the result of the learning process generated by the learning result generation means.

Further, the learning result confirming means can display a diagram or table relating to a learning curve showing the process of the learning process based on the generated result of the learning process.

Further, the learning result confirming means can display a diagram or a table regarding the coefficient of determination of the learning process based on the generated result of the learning process.

Further, the learning result confirming means can display a diagram or a table relating to prediction and actual measurement relating to the learning process based on the generated result of the learning process.

Further, the learning result confirming means can display a diagram or table relating to the distribution of the prediction error of the learning process based on the generated result of the learning process.

Further, the information processing apparatus may further include learning result evaluation means for evaluating the result of the learning process in consideration of the difference between two different values in each of the variables relating to the quality or the property. .

Also, the learning result evaluation means can evaluate the result of the learning process including an allowable error.

Further, the learning result evaluation means can evaluate the result of the learning process using an index including a coefficient of determination or a predicted value coverage rate.

The learning result evaluation means can display the evaluation of the result of the learning process using a predetermined index as a diagram or table in a predetermined format.

<Server>
1... Server 11... Control unit (learning result processing)
60... Learning processing unit 80... Raw data acquisition unit 82... Data content confirmation unit 84... Learning setting acquisition unit 86... Learning result generation unit 100... Data preprocessing unit 102... Prediction model generation unit 104 Prediction accuracy calculation unit 106 Contribution rate calculation unit 108 Contribution variable identification unit 88 Learning result presentation unit 120 Prediction model accuracy presentation unit 122 Response curve presentation section (recipe generation processing)
140 Recipe generation processing unit 160 Desired condition acquisition unit 180 Target specification value acquisition unit 182 Aim width acquisition unit 184 Constraint condition acquisition unit 162 Recipe generation unit 200. Learning result management unit 202 Recipe search unit 204 Clustering unit 164 Recipe generation result presentation unit 220 Recipe list generation unit 222 Radar chart generation unit 224 Response Curve generation unit 226 Recipe difference graph generation unit (DB)
500 Learning result DB
600...Recipe creation result DB
<User device>
2 User device

Claims (18)

An information processing device used for arithmetic processing related to the manufacture of a predetermined article,
Primary data acquisition means for acquiring, as primary data, information on manufacturing conditions of the article and information on the quality or properties of the article to be manufactured;
learning result generation means for executing a learning process having statistical properties on the primary data and generating a predetermined algorithm reflecting the statistical properties included in the primary data;
response curve generating means for generating a diagram or table that allows visual confirmation of the relationship between the variables related to the manufacturing conditions and the variables related to the quality or properties of the article, based on the results of the learning process;
Information processing device. Further comprising contribution rate calculation means for calculating the degree of contribution to the quality or properties of the article for each of the variables related to the manufacturing conditions,
The information processing device according to claim 1 . The response curve generating means selects one of the variables related to the manufacturing conditions and the variables related to the quality or property of the article, draws a two-dimensional graph, and calculates the values of the other variables. generating said diagram or said table that can be modified to observe curve changes;
The information processing apparatus according to claim 1 or 2. further comprising data content confirmation means for determining whether or not the content of the acquired primary data is necessary and sufficient for the learning process by the learning result generation means;
The information processing apparatus according to claim 1 or 2. Contribution variable identification means for determining whether each of the variables related to the manufacturing conditions contributes more than a predetermined value to each or all of the quality or the property,
The information processing apparatus according to claim 1 or 2, comprising: a formula generating means for generating a formula capable of expressing a relationship between a plurality of variables including at least one of the variables relating to the manufacturing conditions and the variables relating to the quality or properties of the article, based on the results of the learning process; prepare further,
The information processing apparatus according to claim 1 or 2. Further comprising learning setting acquisition means for acquiring information about the setting of the learning process,
The information processing apparatus according to claim 1 or 2. further comprising tolerance information obtaining means for obtaining tolerance information regarding a difference between two different values in each of said variables relating to the quality or property of said article;
The information processing apparatus according to claim 2. Further comprising learning result confirmation means for generating information that can contribute to confirmation of the result of the learning process generated by the learning result generation means,
The information processing apparatus according to claim 1 or 2. The learning result confirmation means displays a diagram or table regarding a learning curve showing the process of the learning process based on the generated result of the learning process.
The information processing apparatus according to claim 9 . The learning result confirmation means displays a diagram or table regarding the coefficient of determination of the learning process based on the generated result of the learning process.
The information processing apparatus according to claim 9 . The learning result confirmation means displays a diagram or table regarding predictions and actual measurements related to the learning process based on the generated results of the learning process.
The information processing apparatus according to claim 9 . The learning result confirmation means displays a diagram or table regarding the distribution of the prediction error of the learning process based on the generated result of the learning process.
The information processing apparatus according to claim 9 . Further comprising learning result evaluation means for performing an evaluation on the result of the learning process, taking into account the difference between two different values in each of the quality or the property variables,
The information processing apparatus according to claim 1 or 2. The learning result evaluation means evaluates the result of the learning process including an allowable error.
The information processing apparatus according to claim 14. The learning result evaluation means evaluates the result of the learning process using an index including a coefficient of determination or a predicted value coverage rate.
The information processing apparatus according to claim 14. The learning result evaluation means displays the evaluation of the result of the learning process using a predetermined index as a diagram or table in a predetermined format.
The information processing apparatus according to claim 14. In the computer used for arithmetic processing related to the manufacture of the prescribed article,
a primary data acquisition step of acquiring, as primary data, information regarding manufacturing conditions of the article and information regarding the quality or properties of the article to be manufactured;
a learning result generation step of executing a learning process having statistical properties on the primary data and generating a predetermined algorithm that reflects the statistical properties contained in the primary data;
a response curve generating step of generating a diagram or table that allows visual confirmation of the relationship between the variables related to the manufacturing conditions and the variables related to the quality or properties of the article, based on the results of the learning process;
A program that executes a process including

Images