JP7230654B2 - Data analysis method, data analysis apparatus, and computer program - Google Patents

Description

The present invention relates to a data analysis method, a data analysis device, and a computer program.

Raw materials that form the basis of products have variations in the properties of each raw material (for example, particle size). Conventionally, a method of analyzing data containing such variations is known (see Patent Document 1, for example). The technique described in Patent Document 1 uses a graphical lasso to evaluate log data recorded when an exposure apparatus performs predetermined control processing. Non-Patent Document 1 describes an evaluation method for the degree of relevance of each characteristic in data.

JP 2017-167310 A

Written by Kazunori Yamaguchi, written by Motohisa Hirono, supervised by Masahiko Munechika, "Introduction to SEM Causal Analysis JUSE-StatWorks Official Text (Series 6 that is useful for practical use)", Nikkagiren Publishing Co., Ltd., July 2012

The relevance of various data collected in the manufacturing process may depend on the manufacturing line. For this reason, analysis of various data collected in the manufacturing process is performed by dividing the data for each manufacturing line (hereinafter also referred to as "stratification"). However, the number of specimens for raw material lot data that does not depend on production lines is reduced by stratification. Thus, there is a problem that the degree of association evaluated from data of a small number of samples may deviate greatly from the true value. In addition, when the correlation coefficient is used as the degree of association, there are cases where a high correlation coefficient is obtained by chance. There is a problem that the value becomes unsatisfactory and the degree of relevance cannot be evaluated.

The present invention has been made to solve at least part of the above-described problems, and an object thereof is to correctly evaluate the degree of property relevance between products manufactured from the same raw material lot on different production lines.

The present invention has been made to solve at least part of the above problems, and can be implemented as the following modes. A data analysis method, which is data collected in a manufacturing process and includes independent data that does not depend on a specific process to be analyzed during the manufacturing process and dependent data that depends on the specific process. and an evaluation step of evaluating the relevance of the data set using all of the independent data and the dependent data corresponding to the specific process, The evaluation of the relevance is performed on the element corresponding to the independent data in the variance-covariance matrix or the correlation coefficient matrix obtained with the data set divided for each of the specific processes. A data analysis method by inputting a matrix using a variance-covariance matrix or a correlation coefficient matrix obtained without dividing the data set into a Gaussian graphical model. In addition, the present invention can also be implemented as the following modes.

(1) According to one aspect of the present invention, a data analysis method is provided. This data analysis method includes data collected in a manufacturing process, independent data that does not depend on a specific process to be analyzed during the manufacturing process, and dependent data that depends on the specific process. obtaining a data set; and using all of the non-dependent data and the dependent data corresponding to the particular process to assess the relevance of the data set.

According to this configuration, all non-dependent data that does not depend on a specific process and dependent data that corresponds to a specific process (that is, out of the data divided for each specific process, , data corresponding to a specific process). For this reason, compared to the case where the entire data set is divided for each specific process, it is possible to suppress the decrease in the number of samples of independent data, and the problem caused by the decrease in the number of samples (the degree of relevance is low). It is possible to suppress the occurrence of divergence from the true value, and the positive definiteness of the variance-covariance matrix and correlation coefficient matrix cannot be satisfied when applying the Gaussian graphical model), and to correctly evaluate the degree of association. .

(2) In the data analysis method of the above aspect, the specific process may be a manufacturing line that implements the manufacturing process.
According to this configuration, since the specific process is the manufacturing line, it is possible to evaluate the relevance of the data set including the independent data before processing on the manufacturing line and the dependent data after processing.

(3) In the data analysis method of the above aspect, the non-dependent data is data representing various characteristics of each raw material used in the manufacturing process, and the dependent data is acquired in each step in the manufacturing process. The data may represent at least one of various characteristics of the processed product and processing conditions for processing each raw material.
According to this configuration, it is possible to evaluate the relationship between various characteristics of the raw material as independent data, and at least one relationship between the various characteristics of the processed product as dependent data and the processing conditions in the manufacturing process. can be evaluated.

(4) In the data analysis method of the above form, the evaluation of the relevance is the non-dependence of the variance-covariance matrix or the correlation coefficient matrix obtained with the data set divided for each of the specific processes By inputting a matrix using a variance-covariance matrix or a correlation coefficient matrix obtained without dividing the data set for each specific process into the Gaussian graphical model as an element corresponding to the data good too.
According to this configuration, a variance-covariance matrix or a correlation coefficient matrix is used to evaluate the relevance of data sets. Among the variance-covariance matrix or correlation coefficient matrix of the data set of the divided dependent data, the elements corresponding to the independent data are replaced with the variance-covariance matrix or correlation coefficient matrix of the independent data set. This allows the computation of associations to be performed without the errors in Gaussian graphical models caused by chance high correlation coefficients between variables in the independent data, and the association as a network between variables to be evaluated.

It should be noted that the present invention can be implemented in various aspects, such as a data analysis device, a data analysis system, a data analysis method, a control method for a data analysis device, and a computer program for executing these devices and methods. , a server device for distributing the computer program, a non-temporary storage medium storing the computer program, or the like.

1 is a block diagram of a data analysis device as one embodiment of the present invention; FIG. 1 is an example data set including raw material data and product data; It is an example of a correlation coefficient matrix calculated from raw material data. It is a data set corresponding to a manufacturing line. It is a data set corresponding to a manufacturing line. It is part of the correlation coefficient matrix calculated from the data set corresponding to the production line. 7 is a correlation coefficient matrix obtained by substituting a part of the correlation coefficient matrix shown in FIG. 6; It is a flow chart of the data analysis method in this embodiment. 6 is a flow chart of a data analysis method in the second embodiment;

<First Embodiment>
FIG. 1 is a block diagram of a data analysis device 10 as one embodiment of the invention. As shown in FIG. 1, the data analysis device 10 analyzes various characteristics of the raw material powders A and B in the process of processing the raw material powders A and B in production lines L1 and L2, respectively, and manufacturing them into products. Raw material data DTS representing the product and product data DTL1 and DTL2 representing various characteristics of the product are obtained. The data analysis device 10 evaluates the relationship between various characteristics in each acquired data.

The data analysis device 10 includes a CPU (Central Processing Unit) 1 , a ROM (Read Only Memory) 2 , a RAM (Random Access Memory) 3 and a storage section 9 . The CPU 1 functions as an acquisition unit 4 and an evaluation unit 5 by deploying a computer program stored in the ROM 2 in the RAM 3 and executing it. The storage unit 9 stores various data such as raw material data DTS and product data DTL1 and DTL2. The storage unit 9 is configured by a hard disk drive (HDD) or the like.

In the manufacturing process of the present embodiment, as shown in FIG. 1, raw materials obtained by adding a binder to two types of raw material powders A and B are divided into a manufacturing line L1 and a manufacturing line L2. In each of the production lines L1 and L2, the raw material is agitated, compression-molded into a gear shape, sintered at a predetermined temperature for a predetermined time, and shipped as a product. The acquisition unit 4 acquires raw material data DTS and product data DTL1 and DTL2 collected in the manufacturing process. The raw material data DTS is non-dependent data that does not depend on the production lines L1 and L2, which are the analysis targets of this embodiment, before the binder is added. On the other hand, the product data DTL1 and DTL2 are dependent data dependent on the production lines L1 and L2. As a method for obtaining the raw material data DTS and the product data DTL1 and DTL2, the data may be input via an input unit such as a keyboard or mouse (not shown), or the data may be obtained by wired or wireless communication. . Below, the case where the relevance of the collected data is evaluated for each production line L1, L2 will be exemplified. Therefore, production lines L1 and L2 correspond to specific processes.

FIG. 2 is an example of a data set DS including raw material data DTS and product data DTL1 and DTL2. FIG. 2 shows a table of various characteristics associated with six product IDs. For example, for product ID 1, the raw material lot ID is a, the purity of raw material powder A is 6.2, the moisture content of raw material powder A is 1.1, the production line is production line L1, and the rotation speed of the stirring blade during stirring is is 3.2, the weight before compression is 5.3, and the product hardness is 10.5. Note that the numerical values representing these characteristics do not have units because they represent numerical values with respect to normalized reference values. In other embodiments, the data may be the unnormalized measurements themselves.

As shown in FIG. 2, product IDs 1-3 are manufactured by manufacturing line L1, and product IDs 4-6 are manufactured by manufacturing line L2. The purity of the raw material powder A and the moisture content of the raw material powder A shown in FIG. 2 are independent data that do not depend on the production lines L1 and L2. On the other hand, the number of revolutions of the stirring blades, the weight before compression, and the hardness are dependent data depending on the production lines L1 and L2. Each of the product data DTL1 and DTL2 as dependent data is data representing various characteristics in each step in the manufacturing process and processing conditions for processing each raw material. Other non-dependent data in the present embodiment include particle sizes of the raw material powders A and B, and the like. In addition, other dependent data in this embodiment include the torque of the stirring blade, compression load, sintering temperature, and sintering time as processing conditions, and the weight after compression and sintering as characteristics of the processed product. After the weight and the like.

The evaluation unit 5 evaluates the relevance of various characteristics in the data set DS shown in FIG. 2 using all of the raw material data DTS and the product data DTL1 and DTL2. The evaluation unit 5 evaluates the relationships between various characteristics of the raw material powders A and B using all raw material data DTS independent of the production lines L1 and L2. The evaluation unit 5 of this embodiment calculates the correlation coefficient matrix using the raw material data DTS shown in FIG.

FIG. 3 is an example of a correlation coefficient matrix calculated from raw material data DTS. The correlation coefficient between the purity of raw material powder A and the moisture content of raw material powder A shown in FIG. 3 is 0.7. This correlation coefficient matrix is derived from the six raw material lots IDa to IDf in FIG. Next, the evaluation unit 5 divides the data set DS of FIG. 2 into the production line L1 and the production line L2.

FIG. 4 is a data set DS1 corresponding to the production line L1. FIG. 5 is the data set DS2 corresponding to the production line L2. The data set DS1 shown in FIG. 4 is composed of raw material data DTS1 of three raw material lots IDa to IDc processed on the production line L1, and product data DTL1. The data set DS2 shown in FIG. 5 is composed of raw material data DTS2 of three raw material lots IDd to IDf processed on the production line L2, and product data DTL2.

The evaluation unit 5 calculates a correlation coefficient matrix for each of the divided data sets DS1 and DS2 shown in FIGS. 4 and 5. FIG. FIG. 6 shows part of the correlation coefficient matrix calculated from the data set DS1 corresponding to the production line L1. The correlation coefficient matrix shown in FIG. 6 is calculated based on various characteristics of product IDs 1-3. Therefore, the numerical values in the range RG1 surrounded by the thick line in FIG. 6 are calculated from various characteristics of the three raw material lots IDa to IDc. Numerical values in the range RG1 are calculated from independent data that does not depend on the production line L1.

FIG. 7 is a correlation coefficient matrix in which a part of the correlation coefficient matrix shown in FIG. 6 is replaced. After calculating the correlation coefficient matrix shown in FIG. 6, the evaluation unit 5 replaces the numerical values of the range RG1 corresponding to the independent data with the correlation coefficient matrix (FIG. 3) calculated from the raw material data DTS. Therefore, the numerical values in the range RG1 surrounded by the thick line in FIG. 7 are calculated from the various characteristics of all the six raw material lots IDa to IDf. The evaluation unit 5 receives as input the correlation coefficient matrix after the range RG1 replacement shown in FIG. 7, and uses a Gaussian graphical model to evaluate the degree of association between various characteristics.

FIG. 8 is a flow chart of the data analysis method in this embodiment. In the data analysis method, first, the acquisition unit 4 acquires a data set DS including raw material data DTS and product data DTL1 and DTL2 (step S1). The evaluation unit 5 calculates the correlation coefficient matrix shown in FIG. 3 from the raw material data DTS independent of the production lines L1 and L2 (step S2). The evaluation unit 5 divides the data set DS of FIG. 2 into a data set DS1 for the production line L1 and a data set DS2 for the production line L2 (step S3). The evaluation unit 5 calculates correlation coefficient matrices from the data sets DS1 and DS2 of the production line L1 and the production line L2 (step S4). The evaluation unit 5 stores the data for each of the production lines L1 and L2 in the range RG1 (FIG. 6) corresponding to the independent data in the correlation coefficient matrix corresponding to the data sets DS1 and DS2 of the production lines L1 and L2. The correlation coefficient matrix (FIG. 3) obtained without dividing the sets DS1 and DS2 is replaced (step S5). The evaluation unit 5 inputs the correlation coefficient matrix (FIG. 7) after substitution into the Gaussian graphical model to evaluate the relationship between various characteristics (step S6), and ends the data analysis method. Note that step S1 corresponds to the acquisition step, and steps S2 to S6 correspond to the evaluation step.

As described above, in the data analysis device 10 of the present embodiment, the acquisition unit 4 acquires the raw material data DTS and the product data DTL1 and DTL2 collected in the manufacturing process. The evaluation unit 5 evaluates the relevance of the data set DS using all the raw material data DTS before being divided for each of the production lines L1 and L2 and the product data DTL1 and DTL2 for each of the production lines L1 and L2. . That is, instead of the raw material data DTS1 and DTS2 divided by the manufacturing lines L1 and L2, all the raw material data DTS independent of the manufacturing lines L1 and L2 are used as the independent data used to evaluate the relationship. Therefore, compared to the case where all the data sets DS are divided into the data sets DS1 and DS2 for each of the production lines L1 and L2, it is possible to suppress the decrease in the number of samples of the raw material data DTS, which is independent data. As a result, the occurrence of problems caused by a decrease in the number of samples, such as relevance deviating from the true value, can be suppressed, and relevance can be evaluated correctly.

In addition, in the data analysis device 10 of the present embodiment, raw material data DTS, which is non-dependent data, and product data DTL1, DTL2, which are dependent data, are divided on the basis of production lines L1 and L2. Therefore, the relevance of the data set DS including independent data and dependent data before and after processing on the manufacturing line can be evaluated.

In addition, in the data analysis device 10 of the present embodiment, the raw material data DTS, which is non-dependent data, is data representing characteristics of raw materials. The product data DTL1 and DTL2, which are dependent data, include data representing processing conditions in the production lines L1 and L2 and data representing characteristics of products after processing in the production lines L1 and L2. Therefore, the data analysis device 10 of the present embodiment can evaluate the relationship between various characteristics of the raw material, and can evaluate the relationship between the various characteristics of the processed product and the relationship between the processing conditions in the production lines L1 and L2. can be evaluated.

Further, in the data analysis device 10 of the present embodiment, a correlation coefficient matrix is used for evaluating the relevance of the dataset DS. The evaluation unit 5 replaces the numerical values in the range RG1 corresponding to the independent data in the divided data sets DS1 and DS2 with the correlation coefficient matrix calculated from the raw material data DTS of the independent data. This allows association calculations to be performed without errors in Gaussian graphical models due to chance high correlation coefficients between variables in the independent data. In addition, it is possible to prevent the positive definiteness of the correlation coefficient matrix from being unsatisfied when the Gaussian graphical model is applied. Therefore, the data analysis device 10 of the present embodiment can evaluate network relevance between variables.

<Second embodiment>
FIG. 9 is a flow chart of a data analysis method in the second embodiment. A data analysis device that implements the data analysis method of the second embodiment is the same as the data analysis device 10 of the first embodiment. Evaluation of the data analysis method of the first embodiment is different from evaluation using a Gaussian graphical model using a correlation coefficient matrix. In the second embodiment, a data analysis method different from that of the first embodiment will be described, and description of the same configuration as that of the first embodiment will be omitted.

As shown in FIG. 9, in the data analysis method of the second embodiment, first, the acquisition unit 4 acquires the data set DS (FIG. 2) including raw material data DTS and product data DTL1 and DTL2 (step S11 ). The evaluation unit 5 calculates a correlation coefficient from the raw material data DTS independent of the production lines L1 and L2 (step S12). The evaluation unit 5 divides the data set DS into a data set DS1 for the production line L1 and a data set DS2 for the production line L2 (step S13). The evaluation unit 5 determines correlation coefficients between various characteristics in the product data DTL1 and DTL2, correlation coefficients between various characteristics in the product data DTL1 and DTL2, raw material data , and the correlation coefficients between various characteristics are calculated (step S14).

In the data analysis method of the second embodiment, the evaluation unit 5 uses raw material data DTS and product data DTL1 and DTL2 to evaluate the relevance of data sets DS1 and DS2 of product data DTL1 and DTL2, respectively. Therefore, in the same way as in the first embodiment, the data analysis device 10 of the second embodiment determines the relation of various characteristics between the raw material data DTS in each of the data sets DS1 and DS2 divided for each of the production lines L1 and L2. can more accurately assess gender.

<Modification of this embodiment>
The present invention is not limited to the above-described embodiments, and can be implemented in various aspects without departing from the scope of the invention. For example, the following modifications are possible.

In the first and second embodiments, the data analysis device 10 that implements the data analysis method has been described as an example, but the data analysis device 10 can be modified in various ways. For example, the data analysis device 10 includes a microphone that accepts voice input, a monitor that displays the relationship evaluation results as an image, an output device that outputs the relationship evaluation results as a log, and a speaker that outputs the evaluation results as voice. may be

In the above-described first embodiment, the correlation coefficient matrix input to the Gaussian graphical model is used to evaluate the relevance, but the variance-covariance matrix input to the Gaussian graphical model may be used. In this case, the relationship between various characteristics in the raw material data DTS is also represented by the variance-covariance matrix. Moreover, the evaluation unit 5 does not necessarily have to input the calculated correlation coefficient matrix to the Gaussian graphical model, and may calculate the correlation coefficient matrix and terminate the data analysis method (FIG. 8).

In the first embodiment, the relationship between the data set DS including the raw material data DTS independent of the production lines L1 and L2 and the product data DTL1 and DTL2 dependent on the production lines L1 and L2 was evaluated. may be evaluated for relevance of the combined datasets. For example, when a plurality of different raw material data delivered from different material manufacturers are processed on the production line L1, each of the plurality of raw material data is non-dependent data that does not depend on the production line L1. In this case, for example, the evaluation unit 5 may calculate a correlation coefficient matrix representing the relevance of a plurality of independent data corresponding to each raw material lot.

A specific process that separates the independent data and the dependent data may be a process other than the manufacturing lines L1 and L2 that implement the manufacturing process. For example, the process may be divided into steps of stirring and sintering in production lines L1 and L2. For example, the relevance of collected data may be evaluated for each raw material lot. In this case, if the raw material lot IDs are different, the processing is performed assuming that the specific processes are different.

Dependent data and non-dependent data may be data other than data representing various characteristics. For example, raw material data DTS may include the supplier of the raw material, the purchase price, and the purchase timing. Further, the product data DTL1 and DTL2 may be a measured value, an evaluation value, an index representing the difference between the measured value and the evaluation value, a sales price, and the like.

The present aspect has been described above based on the embodiments and modifications, but the above-described embodiments are intended to facilitate understanding of the present aspect, and do not limit the present aspect. This aspect may be modified and modified without departing from the spirit and scope of the claims, and this aspect includes equivalents thereof. Also, if the technical features are not described as essential in this specification, they can be deleted as appropriate.

1 CPU
2 ROM
3 RAM
4... Acquisition unit 5... Evaluation unit 9... Storage unit 10... Data analysis device A, B... Raw material powder DS, DS1, DS2... Data set DTL1, DTL2... Product data DTS, DTS1, DTS2... Raw material data IDa to IDf... Raw material Lot ID1 to ID6 ... Product L1, L2 ... Production line RG1 ... Range

Claims (5)

A data analysis method comprising:
Acquisition for acquiring a data set that is data collected in a manufacturing process and that includes independent data that does not depend on a specific process to be analyzed during the manufacturing process and dependent data that depends on the specific process process and
an evaluation step of evaluating relevance of the data set using all of the non-dependent data and the dependent data corresponding to the specific process;
with
The evaluation of the relevance is performed on the element corresponding to the independent data in the variance-covariance matrix or the correlation coefficient matrix obtained with the data set divided for each of the specific processes. A data analysis method by inputting a matrix using a variance-covariance matrix or a correlation coefficient matrix obtained without dividing the data set into a Gaussian graphical model. The data analysis method according to claim 1,
The data analysis method, wherein the specific process is a manufacturing line that implements the manufacturing process. The data analysis method according to claim 2,
The independent data is data representing various characteristics of each raw material used in the manufacturing process,
The data analysis method, wherein the dependent data is data representing at least one of various characteristics of the processed product obtained in each step of the manufacturing process and processing conditions for processing each raw material. A data analysis device,
Acquisition for acquiring a data set that is data collected in a manufacturing process and that includes independent data that does not depend on a specific process to be analyzed during the manufacturing process and dependent data that depends on the specific process Department and
an evaluation unit that evaluates the relevance of the data set using all of the non-dependent data and the dependent data corresponding to the specific process;
with
The evaluation of the relevance is performed on the element corresponding to the independent data in the variance-covariance matrix or the correlation coefficient matrix obtained with the data set divided for each of the specific processes. A data analysis device , which is implemented by inputting a matrix using a variance-covariance matrix or a correlation coefficient matrix obtained without dividing the data set into a Gaussian graphical model. A computer program,
Acquisition for acquiring a data set that is data collected in a manufacturing process and that includes independent data that does not depend on a specific process to be analyzed during the manufacturing process and dependent data that depends on the specific process function and
an evaluation function that evaluates the relevance of the data set using all of the non-dependent data and the dependent data corresponding to the specific process;
is realized on a computer,
The evaluation of the relevance is performed on the element corresponding to the independent data in the variance-covariance matrix or the correlation coefficient matrix obtained with the data set divided for each of the specific processes. A computer program executed by inputting a matrix using a variance-covariance matrix or a correlation coefficient matrix obtained without dividing the data set into a Gaussian graphical model.

Images