Information processing device, information processing method, and process execution method

Abstract

An information processing device 10 analyzes a correlation between measurement data in a manufacturing process 20 of a product and quality information of the product, and comprises a control unit 15. The control unit 15 acquires the measurement data in each step in the manufacturing process 20 including a plurality of steps, calculates a partial correlation between factors including each of the measurement data and the quality information to acquire the correlation, connects the factors having the correlation with each other in a causal direction based on a conditional probability by a path, and generates a path diagram visualizing the correlation between the measurement data and the quality information in the manufacturing process 20.

Description

Information Processing Apparatus, Information Processing Method, and Process Execution Method

[0001] The present disclosure relates to an information processing apparatus, an information processing method, and a process execution method.

[0002] Conventionally, in products manufactured through a plurality of processes such as steel products, there are many processes that affect quality information including the quality and presence or absence of defects, so it may not be easy to identify factors related to the quality information. In contrast, various studies have been conducted for factor estimation.

[0003] For example, Patent Document 1 discloses a factor analysis apparatus that can easily grasp the relationship between measurement items even when there is no knowledge of the product being manufactured through a large number of item measurements in the manufacturing process. The apparatus takes as input a plurality of measurement data and their measurement order or process order, selects explanatory variables in the regression analysis from the previous measurement items for each measurement item, and generates a regression equation. The apparatus specifies the measurement item to be the target of factor analysis and hierarchically displays the explanatory variables using a characteristic factor diagram, thereby automatically visualizing the hierarchical relationship between a large number of measurement items.

[0004] Patent Document 2 discloses a method of collecting data measured by manufacturing equipment, constructing a causal relationship model using graphical modeling, deriving the correlation coefficient between each variable for the manufacturing process data, determining the presence or absence of correlation, and continuously updating the causal relationship model for the purpose of constructing a highly accurate and general-purpose causal relationship model using regression analysis.

[0005] Japanese Unexamined Patent Application Publication No. 2011 - 150496, Japanese Unexamined Patent Application Publication No. 2008 - 084039

[0006] In the conventional technology described in Patent Document 1, when performing factor estimation from among a plurality of measurement items in the manufacturing process of a product, variables with high correlation are selected from the explanatory variables in the regression analysis. The regression analysis was performed by specifying the variables used for explanation according to the measurement order or process order.

[0007] When using multiple regression equations to find highly correlated variables, there remained the possibility of spurious correlations between variables. That is, even if there was no correlation between variables, the calculation might make it appear as if there was a correlation due to the influence of other variables or coincidences. In addition, if variables were specified by measurement order or process order, there was a possibility that variables not belonging to an ordinal relationship, such as categorical variables representing product type, might not be considered. As described above, conventional techniques had problems such as the visualized results not being correct correlations or important factor variables not being considered. As a result, the accuracy of correlation analysis decreased, and the accuracy of estimating factors related to quality information decreased.

[0008] In the graphical modeling used in Patent Document 2, it was sometimes difficult to explicitly indicate the direction of causal relationships. In particular, when the direction of causal relationships was unclear between correlated variables, it could lead to misinterpretations. Furthermore, while graphical modeling infers causal relationships based on conditional independence, conditional independence may not hold in real-world data, posing a risk of constructing an incorrect model. Additionally, while graphical modeling can construct complex models, overly complex models can become difficult to interpret.

[0009] This disclosure has been made in view of the above issues and aims to provide an information processing device, an information processing method, and a process implementation method that improve the accuracy of estimating factors related to quality information of products manufactured by a manufacturing process.

[0010] (1) An information processing apparatus according to one embodiment of the present disclosure is an information processing apparatus for analyzing the correlation between measurement data in a product manufacturing process and quality information of the product, comprising a control unit, the control unit acquires the measurement data at each step in the manufacturing process which includes a plurality of steps, calculates partial correlations between factors including the measurement data and the quality information to obtain correlations, connects the factors that are correlated with each other in a causal direction based on conditional probability using paths, and generates a path diagram that visualizes the correlation between the measurement data and the quality information in the manufacturing process.

[0011] (2) In one embodiment of the present disclosure, in (1), the control unit may determine that the factors whose partial correlation coefficient is equal to or greater than a first threshold are correlated with each other.

[0012] (3) In one embodiment of the present disclosure, in (1) or (2), the control unit may determine that the correlation between the factors connected by the path corresponds to the sequence of steps in the manufacturing process, and determine that the correlation is valid.

[0013] (4) In one embodiment of the present disclosure, in any of (1) to (3), the control unit may, before calculating the partial correlation, test the correlation between the factors using a statistical testing method, group the factors that are determined to be significant, calculate the partial correlation between the factors including the measurement data and the quality information for each group of factors, obtain the correlation, and for each group of factors, connect the factors that are correlated with each other in a causal direction based on conditional probability using paths to generate a path diagram that visualizes the correlation between the measurement data and the quality information in the manufacturing process.

[0014] (5) In one embodiment of the present disclosure, in (4), the control unit may test the correlation for all combinations of the factors.

[0015] (6) In one embodiment of the present disclosure, in (4) or (5), the test method may include either Fisher's exact test method or a statistical causal inference method including Bayesian networks and DAG (Directed Acyclic Graph)-DNN (Deep Neural Network).

[0016] (7) In one embodiment of the present disclosure, in any of (1) to (6), the control unit may calculate a first conditional probability that the first factor occurs and the second factor occurs, and a second conditional probability that the second factor occurs and the first factor occurs, for a first factor and a second factor that are correlated with each other, and connect the first factor and the second factor with a path in the causal direction between the first factor and the second factor estimated from the first conditional probability and the second conditional probability.

[0017] (8) In one embodiment of the present disclosure, in any of (1) to (7), the control unit may set extraction conditions for the path diagram and extract and visualize only the factors that satisfy the extraction conditions from the path diagram.

[0018] (9) In one embodiment of the present disclosure, in (8), the extraction condition may include the factors of the start and end points in the path diagram, the path length, and a second threshold value for the strength of the partial correlation.

[0019] (10) In one embodiment of the present disclosure, in (9), the factor at the starting point in the path diagram is a factor related to the quality information, and the factor at the ending point may be a factor that has a strong correlation with the quality information, a significant factor of the quality information according to Fisher's exact test, or a factor related to the measurement data that has high variable importance in a machine learning model in which the quality information is the dependent variable and the measurement data is the independent variable.

[0020] (11) In one embodiment of the present disclosure, in any of (1) to (10), the control unit may distinguish and visualize the measurement data of a predetermined first factor in the path diagram for each factor in a group of factors that includes a second factor connected to the first factor and a factor of the same type as the second factor.

[0021] (12) An information processing method according to one embodiment of the present disclosure is an information processing method for analyzing the correlation between measurement data in a product manufacturing process and quality information of the product, comprising: acquiring the measurement data at each step in the manufacturing process which includes a plurality of steps; calculating partial correlations between factors which include the measurement data and the quality information respectively to obtain the correlation; and connecting the factors which are correlated with each other in a causal direction based on conditional probability using paths to generate a path diagram which visualizes the correlation between the measurement data and the quality information in the manufacturing process.

[0022] (13) As one embodiment of the present disclosure, (12) may include: testing the correlation between the factors using a statistical testing method before calculating the partial correlation, and grouping the factors that are determined to be significant; for each group of factors, calculating the partial correlation between the factors including the measurement data and the quality information to obtain the correlation; and for each group of factors, connecting the factors that are correlated with each other in a causal direction based on conditional probability using paths, thereby generating the path diagram that visualizes the correlation between the measurement data and the quality information in the manufacturing process.

[0023] (14) In one embodiment of the present disclosure, in (12) or (13), for a first factor and a second factor that are correlated with each other, a first conditional probability that the first factor occurs and the second factor occurs, and a second conditional probability that the second factor occurs and the first factor occurs, may be calculated, and the first factor and the second factor may be connected by a path in the causal direction between the first factor and the second factor estimated from the first conditional probability and the second conditional probability.

[0024] (15) A process implementation method according to one embodiment of the present disclosure includes: identifying factors that have a strong correlation with the quality information of the final product, among factors including the measurement data and quality information of each step in the manufacturing process which includes a plurality of steps, based on the path diagram generated by the information processing method described in any of (12) to (14); optimizing conditions corresponding to the identified factors in order to improve the quality of the final product; and carrying out the manufacturing process under the optimized conditions.

[0025] According to an information processing apparatus, information processing method, and process implementation method according to one embodiment of this disclosure, the accuracy of estimating factors related to the quality information of products manufactured by a manufacturing process is improved.

[0026] This is a block diagram showing an example of the configuration of an information processing system having an information processing device according to one embodiment of the present disclosure. This is a first flowchart showing an example of an operation performed by the information processing device in Figure 1. This is a second flowchart showing an example of an operation performed by the information processing device in Figure 1. This is a first schematic diagram for explaining an example of an operation performed by the information processing device in Figure 1. This is a second schematic diagram for explaining an example of an operation performed by the information processing device in Figure 1. This is a third schematic diagram for explaining an example of an operation performed by the information processing device in Figure 1. This is a fourth schematic diagram for explaining an example of an operation performed by the information processing device in Figure 1. This is a fifth schematic diagram for explaining an example of an operation performed by the information processing device in Figure 1.

[0027] In the following, one embodiment of this disclosure will be mainly described with reference to the attached drawings.

[0028] Figure 1 is a block diagram showing an example of the configuration of an information processing system 1 having an information processing device 10 according to one embodiment of the present disclosure. An example of the configuration and operation of the information processing system 1 having an information processing device 10 according to one embodiment of the present disclosure will be mainly described with reference to Figure 1.

[0029] The information processing system 1 includes a manufacturing process 20 in addition to the information processing device 10. The manufacturing process 20 is a process for manufacturing products such as steel products, and includes multiple steps. The manufacturing process 20 includes a series of steps, for example, starting from a steelmaking plant, passing through a hot rolling mill and a rolling mill, and finally completing the product at a hot galvanizing plant. In the information processing system 1, the information processing device 10 and the manufacturing process 20 are connected to each other in a way that allows them to communicate with one another.

[0030] The information processing device 10 estimates factors related to the quality information of the final product within a manufacturing process 20 that includes multiple steps, in the information processing system 1. In this disclosure, "quality information" includes, for example, the quality of the final product and the presence or absence of defects. For example, the information processing device 10 analyzes the correlation between measurement data in the product manufacturing process 20 and the product quality information. As an example, the information processing device 10 may analyze the correlation between measurement data in the product manufacturing process 20 and defects in the product.

[0031] In this disclosure, “measurement data” includes, for example, the average value of time-series data of parameters measured when a product is processed at least one process in each factory. Parameters include, for example, component values, temperature, reduction ratio, dew point, furnace temperature, and steel plate steepness. Measurement data is not limited to these, and may also be the time-series data of parameters measured when a product is processed at least one process in each factory, or other statistical indicators different from the average value.

[0032] The information processing device 10 includes general-purpose electronic devices such as PCs (Personal Computers). The information processing device 10 is not limited to these, and may be one or multiple server devices that can communicate with each other, or other electronic devices dedicated to the information processing system 1. In addition, the information processing device 10 may also include general-purpose electronic devices other than PCs, such as tablet PCs, smartphones, and wearable devices such as smartwatches. The information processing device 10 has a communication unit 11, a storage unit 12, an input unit 13, an output unit 14, and a control unit 15.

[0033] The communication unit 11 includes one or more communication interfaces connected to a network. These communication interfaces may support, for example, mobile communication standards such as 4G (4th Generation) and 5G (5th Generation), wired LAN (Local Area Network) standards, or wireless LAN standards, but are not limited to these and may support any communication standard. For example, the communication interface may also support short-range wireless communication standards. In one embodiment, the information processing device 10 is communicably connected to the manufacturing process 20 via the communication unit 11. Various types of information are transmitted and received between the manufacturing process 20 and the communication unit 11.

[0034] The storage unit 12 includes storage modules such as an HDD (Hard Disk Drive), SSD (Solid State Drive), EEPROM (Electrically Erasable Programmable Read-Only Memory), ROM (Read-Only Memory), and RAM (Random Access Memory). The storage unit 12 stores information necessary to realize the operation of the information processing device 10. The storage unit 12 stores information obtained through the operation of the information processing device 10. For example, the storage unit 12 stores system programs, application programs, and various data obtained by any means such as communication.

[0035] The storage unit 12 may function as a main memory module, an auxiliary memory module, or a cache memory. The storage unit 12 is not limited to one built into the information processing device 10, and may also include an external storage module connected by a digital input / output port such as USB (Universal Serial Bus).

[0036] The input unit 13 includes one or more input interfaces that detect user input and acquire input information based on user operations. These input interfaces include physical keys, capacitive keys, a touchscreen integrated with the display of the output unit 14, an imaging module such as a camera, and a microphone that accepts voice input.

[0037] The output unit 14 includes one or more output interfaces that output information to notify the user. These output interfaces include a display that outputs information as an image, a speaker that outputs information as sound, and a vibrator that outputs information as vibration. The display includes LCD (Liquid Crystal Display) and organic EL (Electro Luminescence) displays.

[0038] The control unit 15 includes one or more processors. In this disclosure, "processor" is a general-purpose processor or a dedicated processor specialized for a specific process, but is not limited to these. The control unit 15 includes, for example, a CPU (Central Processing Unit). The control unit 15 is communicably connected to each component constituting the information processing device 10 and controls the operation of the entire information processing device 10.

[0039] Figure 2 is a first flowchart showing an example of the operation performed by the information processing device 10 in Figure 1. The first flowchart in Figure 2 shows the basic processing flow of the information processing method performed using the information processing device 10.

[0040] In step S101, the control unit 15 of the information processing device 10 acquires measurement data from each step in the manufacturing process 20, which includes multiple steps. Based on the measurement data acquired in step S101, the control unit 15 generates a database used for correlation analysis. In this disclosure, the "database" is, for example, a database in which measurement data from when the relevant product was processed at each step is associated with quality information of the final product, using a manufacturing number assigned to each product, for example, a coil number in the case of steel products, as the key.

[0041] The information processing apparatus 10, for example, acquires time-series data of parameters in each process of the manufacturing process 20 via the communication unit 11, calculates data of a representative single point such as an average value for each product, and acquires it as measurement data. The information processing apparatus 10 acquires quality information of the final product manufactured by the manufacturing process 20 by receiving an input operation using the input unit 13 by a quality control person or the like. The information processing apparatus 10 labels the quality information for each manufacturing number, constructs a database, and stores it in the storage unit 12.

[0042] In step S102, the control unit 15 of the information processing apparatus 10 tests the correlation between factors using a statistical test method before calculating the partial correlation in step S103 described later, and groups the factors determined to be significant. In the present disclosure, the "factor" includes, for example, each of the measurement data and quality information in each process in the manufacturing process 20. In the present disclosure, the "test method" includes, for example, either Fisher's exact test method or a statistical causal inference method including a Bayesian network and a DAG (Directed Acyclic Graph)-DNN (Deep Neural Network).

[0043] The control unit 15 tests the correlation for all combinations of factors. That is, the control unit 15 does not need to execute filtering processing for preliminarily examining the presence or absence of correlation with quality information for the measurement data that is the target of the processing after step S102, and actually, measurement data that has no correlation may also be included as a factor in the target of the processing. The filtering processing includes, for example, processing for selecting only factors whose p-value in the result of Fisher's exact test is less than or equal to a threshold value and are determined to be significant for quality information.

[0044] In step S103, the control unit 15 of the information processing apparatus 10 calculates the partial correlation between factors including each of the measurement data and quality information to obtain a correlation relationship. For example, the control unit 15 calculates the partial correlation excluding spurious correlation for each factor for each group of factors grouped in step S102.

[0045] In step S104, the control unit 15 of the information processing apparatus 10 determines whether or not the partial correlation coefficient based on the partial correlation calculated in step S103 is greater than or equal to the first threshold value. The first threshold value may be appropriately determined by receiving an input operation using the input unit 13 by a quality control person or the like. When the control unit 15 determines that the partial correlation coefficient is greater than or equal to the first threshold value, it executes the process of step S105. That is, the control unit 15 determines that the factors for which the partial correlation coefficient is greater than or equal to the first threshold value have a correlation relationship with each other, and executes the process of step S105. When the control unit 15 determines that the partial correlation coefficient is less than the first threshold value, it executes the process of step S106 without executing the process of step S105.

[0046] In step S105, when the control unit 15 of the information processing apparatus 10 determines in step S104 that the partial correlation coefficient is greater than or equal to the first threshold value, it connects the factors having a correlation relationship with each other by a path.

[0047] In step S105, the control unit 15 of the information processing apparatus 10 calculates and estimates the conditional probability of the order, that is, the causal direction, of the factors connected by the path. Specifically, for the first factor and the second factor having a correlation relationship with each other, the control unit 15 calculates the first conditional probability that the second factor occurs when the first factor occurs and the second conditional probability that the first factor occurs when the second factor occurs, and connects the first factor and the second factor by a path in the causal direction between the first factor and the second factor estimated from the first conditional probability and the second conditional probability.

[0048] In step S106, the control unit 15 of the information processing apparatus 10 determines whether or not the partial correlation has been calculated for all factors. When the control unit 15 determines that the partial correlation has been calculated for all factors, it executes the process of step S107. When the control unit 15 determines that the partial correlation has not been calculated for all factors, it repeats the process from step S103 for the factors for which the partial correlation has not been calculated.

[0049] In step S107, the control unit 15 of the information processing device 10 generates a path diagram if it determines that partial correlations have been calculated for all factors in step S106. For example, the control unit 15 generates a path diagram that visualizes the correlation between measurement data and quality information in the manufacturing process 20, depending on whether or not there are path connections between the factors obtained through the processing cycle from step S103 to step S106.

[0050] Figure 3 is a second flowchart showing an example of the operation performed by the information processing device 10 in Figure 1. The second flowchart in Figure 3 shows the post-processing flow for the process in step S107 of Figure 2, which is part of the information processing method performed using the information processing device 10. When the number of factors to be analyzed is very large, the path diagram generated by the information processing device 10 contains a very large number of factors and causal relationships, making it difficult to identify factors related to the final product information (defects, etc.). The second flowchart shows a process to make it easier to identify important factors from the path diagram of all factors and to improve the efficiency of identifying factors related to product defects.

[0051] In step S201, the control unit 15 of the information processing device 10 sets extraction conditions for the path diagram generated in step S107 of Figure 2. In this disclosure, the "extraction conditions" include, for example, the start and end point factors, the path length, and a second threshold for the strength of the partial correlation in the path diagram. The second threshold may be determined as appropriate by receiving input operations using the input unit 13 by a quality control person or the like. The second threshold may be the same as or different from the first threshold.

[0052] Here, in the path diagram, the starting factor is a factor related to the quality information of the final product within a manufacturing process that includes multiple steps. The ending factor may be a factor that has a strong correlation with the factor corresponding to the quality information of the final product. The ending factor may be a factor that is expected to be related to the quality information of the final product. The ending factor may be a factor that has a very strong correlation with the quality information of the final product, or a factor that is identified by utilizing the knowledge of experts such as quality control personnel. The ending factor may be a factor that is estimated to have a significant relationship with the quality information of the final product (e.g., presence or absence of product defects) by Fisher's exact test. The ending factor may be a factor related to measurement data with high variable importance in a machine learning model where the quality information of the final product is the dependent variable and measurement data of a manufacturing process including multiple steps is the independent variable.

[0053] In step S202, the control unit 15 of the information processing device 10 extracts and visualizes only the factors that satisfy the extraction conditions set in step S201 from the path diagram generated in step S107 in Figure 2.

[0054] In step S203, the control unit 15 of the information processing device 10 determines whether the correlation between the factors connected by paths corresponds to the order of processes in the manufacturing process 20. If the control unit 15 determines that the correlation corresponds to the order of processes, it executes the process in step S204. If the control unit 15 determines that the correlation does not correspond to the order of processes, it executes the process in step S205.

[0055] In step S204, if the control unit 15 of the information processing device 10 determines in step S203 that the correlation corresponds to the order of the processes, it determines that the correlation is valid.

[0056] In step S205, if the control unit 15 of the information processing device 10 determines in step S203 that the correlation does not correspond to the order of the processes, it determines that the correlation is not valid.

[0057] In step S206, the control unit 15 of the information processing device 10 determines whether the determination process in step S203 has been executed for all paths. If the control unit 15 determines that all paths have been processed, it executes the process in step S207. If the control unit 15 determines that all paths have not been processed, it repeats the process from step S203 for the paths for which the determination process has not been executed.

[0058] In step S207, if the control unit 15 of the information processing device 10 determines that processing has been completed for all paths in step S206, it visualizes the measurement data of a predetermined first factor in the path diagram, distinguishing it by factor, including the second factor connected to the first factor and the factor group containing factors of the same type as the second factor. For example, the control unit 15 executes the processing in step S207, treating the factors that were determined to have a valid correlation in step S204 as the first factor and the second factor, respectively.

[0059] Figure 4 is a first schematic diagram illustrating an example of the operation performed by the information processing device 10 in Figure 1. Figure 4 shows an example of a database generated by the control unit 15 of the information processing device 10 in step S101 of Figure 2.

[0060] In the database, for example, measurement data for each process from process a to process n is associated with each manufacturing number 1, 2, ... assigned to each product. In addition, the database associates the final product quality information with each manufacturing number 1, 2, ... assigned to each product. For example, the "Quality Information" item may have "OK" and "NG" as parameters, depending on whether the final product is good or bad. For example, the "Quality Information" item may have "NG" and "OK" as parameters, depending on whether the final product has defects or not.

[0061] Figure 5 is a second schematic diagram illustrating an example of the operation performed by the information processing device 10 in Figure 1. Figure 5 conceptually shows the sequence of processes from step S102 to step S107 in Figure 2. Referring to Figure 5, we will mainly explain an example of the procedure for estimating the factors related to defects in the final product, which is performed by the control unit 15 of the information processing device 10 using the measurement data acquired in step S101 of Figure 2.

[0062] The control unit 15 of the information processing device 10 tests the correlation between factors for all combinations of factors. For example, the control unit 15 uses Fisher's exact test to test the significance between factors. Fisher's exact test is a testing method that examines the distribution of each variable for results expressed as binary values ​​such as OK / NG, and if there is a difference in the distribution between the OK data group and the NG data group for that variable, then that variable is considered significant to the result. For example, the control unit 15 uses Fisher's exact test to test the significance between factors for all combinations of factors, such as factor a and factor b, factor b and factor c, etc., for a group of factors including factors a through z.

[0063] The control unit 15 groups the factors that have been determined to be significant using Fisher's exact test. For example, the control unit 15 generates a first group containing factors a and b, and a second group containing factors x, y, and z.

[0064] The control unit 15 calculates partial correlations for each factor in each of the divided groups, eliminating spurious correlations. For example, the control unit 15 calculates a partial correlation between factor a and factor b in the first group. For example, the control unit 15 calculates a partial correlation between factor x and factor y in the second group. For example, the control unit 15 calculates a partial correlation between factor x and factor z in the second group. For example, the control unit 15 calculates a partial correlation between factor y and factor z in the second group. In addition, the control unit 15 also calculates partial correlations between each of factors a, b, x, y, and z and quality information factors such as OK / NG.

[0065] The control unit 15 connects factors with stronger partial correlations via paths. For example, in the first group, the control unit 15 connects factor a and factor b, whose partial correlation coefficients are greater than or equal to a first threshold, via paths. For example, in the second group, the control unit 15 connects factor x and factor y, whose partial correlation coefficients are greater than or equal to a first threshold, via paths. For example, in the second group, the control unit 15 connects factor x and factor z, whose partial correlation coefficients are greater than or equal to a first threshold, via paths.

[0066] In addition, the control unit 15 connects, for example, factor b, whose partial correlation coefficient is greater than or equal to a first threshold, to factor OK / NG via a path. The control unit 15 connects, for example, factor y, whose partial correlation coefficient is greater than or equal to a first threshold, to factor OK / NG via a path. The control unit 15 connects, for example, factor z, whose partial correlation coefficient is greater than or equal to a first threshold, to factor OK / NG via a path.

[0067] The control unit 15 determines the causal relationship between factors connected by paths by calculating conditional probabilities (for example, a Bayesian network, a conditional p-value calculated by applying Fisher's exact test to a 2x2 contingency table based on the conditional event frequencies occurring between each factor). For example, the control unit 15 calculates a first probability that factor a occurs and then factor b occurs, and a second probability that factor b occurs and then factor a occurs. The control unit 15 determines the direction of causality for factors a and b that corresponds to the larger of the first and second probabilities. For example, if the first probability is greater than the second probability, the control unit 15 assumes that a causal relationship exists from factor a to factor b and connects a path from factor a to factor b. The same applies to other factors. By estimating the direction of causality using conditional probabilities, spurious correlations (false correlations) can be eliminated, and a statistically sound causal structure can be estimated.

[0068] The control unit 15 repeats the path connection process as described above to generate a path diagram of the entire factor, as shown in the lower part of Figure 5. The control unit 15 outputs the path diagram obtained through the above process to, for example, the output unit 14, thereby visualizing the factors that lead to the occurrence of defects in the final product.

[0069] Figure 6 is a third schematic diagram illustrating an example of the operation performed by the information processing device 10 in Figure 1. Figure 6 conceptually shows the sequence of processes from step S201 to step S206 in Figure 3. Referring to Figure 6, we will mainly explain an example of post-processing performed by the control unit 15 of the information processing device 10 on the path diagram acquired in step S107 of Figure 2.

[0070] When the control unit 15 of the information processing device 10 generates a path diagram, it is conceivable that a path diagram with a very large number of factors to be analyzed may be obtained. In this case, there will also be a very large number of paths showing correlations between factors, i.e., causal relationships. Therefore, it may not be easy to identify the factors related to defects in the final product using the obtained path diagram as is.

[0071] The control unit 15 may, for example, accept an input operation from the input unit 13 to identify a factor if the factors have been narrowed down to a certain extent based on the knowledge possessed by a quality control officer, and extract only the area around that factor from the overall path diagram. In this disclosure, "knowledge" includes, for example, insights obtained from past literature and field experience. Alternatively, if there is a factor for which a very strong correlation has been obtained, the control unit 15 may extract only the area around that factor from the overall path diagram.

[0072] For example, if the control unit 15 has found that defects are more likely to occur in the final product only in a specific type of product, "steel type A," it may extract only the area around the factor "steel type A" from the overall path diagram and perform the analysis, as shown in Figure 6. Figure 6 shows, as an example, a path diagram in which only the area around the factor "steel type A" has been extracted.

[0073] The set extraction conditions were, for example, a starting point of "Steel type A", an ending point of "Defect (NG)", a path length of "4 or less", and a partial correlation coefficient of "0.5 or higher". The extracted path diagram yielded the analysis result that any of the following factors may be contributing factors to defects in steel type A: "raw steel component value X1", "hot rolling mill X2 equipment temperature", "cold rolling mill X3 equipment reduction ratio", "hot galvanizing mill X4 equipment dew point", and "hot galvanizing mill X5 equipment dew point".

[0074] In addition to the factors extracted in Figure 6, other factors such as "furnace temperature of the hot-dip galvanizing plant Y1 equipment," "alkali temperature of the hot-dip galvanizing plant Y2 equipment," and "steepness of the cold-rolled steel sheet Y3" may also be included.

[0075] In Figure 6, among the factors extracted, we focus on "Dew point of equipment X4 at the molten zinc plating plant" → "Dew point of equipment X5 at the molten zinc plating plant". Of equipment X4 and equipment X5, equipment X4 is located upstream in the product manufacturing process 20. Therefore, in step S204 of Figure 3, the control unit 15 determines that the correlation between "Dew point of equipment X4 at the molten zinc plating plant" and "Dew point of equipment X5 at the molten zinc plating plant" is valid. In other words, the control unit 15 determines that the causal relationship leading to the occurrence of defects, from the upstream equipment factor to the downstream equipment factor, is calculated and is valid because it is in line with reality.

[0076] The control unit 15 may perform the above-described validation automatically by assigning parameters as 1, 2, 3, ... to each factor from the upstream factory to the downstream factory, and then checking whether a reversal of numbers has occurred in the correlation after the analysis results have been output as a path diagram. The control unit 15 may also perform the above-described validation automatically by similarly assigning parameters to each piece of equipment from the upstream equipment to the downstream equipment, even if there are multiple pieces of equipment within a single factory.

[0077] Figure 7A is a fourth schematic diagram illustrating an example of the operation performed by the information processing device 10 in Figure 1. Figure 7B is a fifth schematic diagram illustrating an example of the operation performed by the information processing device 10 in Figure 1. Figures 7A and 7B conceptually show the process performed in step S207 of Figure 3. Referring to Figures 7A and 7B, we will mainly explain the results of checking the correlation between the factors actually extracted and defects in order to evaluate the path analysis results of the path diagram obtained in step S107 of Figure 2.

[0078] As an example, we examined the "raw steel component value X1". The path diagram shown in Figure 6 suggests that, among the factors related to steel type A, the "raw steel component value X1" may be influencing the occurrence of defects in the final product. How these results are interpreted is shown in Figures 7A and 7B.

[0079] Figure 7A shows histograms of component value X1 for all steel grades, which is one of the factors determined to be significant by Fisher's exact test as a factor correlated with the presence or absence of defects in the final product. The graph in Figure 7A shows that defects are likely to occur below the threshold, and almost never occur above the threshold. However, this analysis alone does not provide enough information to interpret the correlation between component value X1 and the presence or absence of defects.

[0080] Therefore, Figure 7B shows the results of a more detailed examination of the results in Figure 7A based on the path analysis results shown in Figure 6. In the graph shown in Figure 7B, the first factor, extracted from the connection between the presence or absence of defects shown in the path diagram in Figure 6 and the steel type information, is color-coded according to the steel type. The first factor corresponds to "raw steel component value X1". The second factor connected to the first factor corresponds to "steel type A". The group of factors containing the same type as the second factor corresponds to "steel types A and B".

[0081] Figure 7B confirms that in steel type A, which is prone to defects, the component value X1 tends to be relatively lower than in steel type B, while in steel type B, which is less prone to defects, the component value X1 tends to be relatively higher than in steel type A. Similarly, it was confirmed that the distribution of the other extracted factors X2 to X5 differs between steel type A and steel type B. Through this procedure, the meaning of factors considered significant in the occurrence of defects in the final product can be interpreted more deeply. In Figure 7B, by comparing steel type A and steel type B, it is easy to understand that steel type A is the factor related to defects in the final product.

[0082] As factors used to distinguish and display results when interpreting them, categorical variables of the same type and mutually exclusive relationship, such as steel type A and steel type B as shown in Figure 7B, are used, but are not limited to this. For example, when the control unit 15 extracts only the vicinity of a continuous quantity such as temperature and analyzes the path diagram, it may divide the entire range of the continuous quantity into predetermined ranges, convert one range into a categorical variable where one range is "0" and the other ranges are "1", and generate categorical variables of the same type and mutually exclusive relationship.

[0083] According to the information processing device 10 of the above embodiment, the accuracy of estimating factors related to the quality information of products manufactured by the manufacturing process 20 is improved. The information processing device 10 calculates partial correlations between factors and obtains correlation relationships. The information processing device 10 connects factors that have correlation relationships with each other in a causal direction based on conditional probability using paths and generates a path diagram.

[0084] As described above, the information processing device 10, when estimating factors related to the final product quality information, eliminates spurious correlations between factors by using partial correlation to remove the influence of other factors, and clearly estimates the causal direction by using conditional probability. This makes it easier to identify the factors that are the true causes. The information processing device 10 makes it possible to identify factors that can truly affect quality information from measurement data for each of the multiple processes included in the product manufacturing process 20. The information processing device 10 enables analysis that eliminates spurious correlations between factors by visualizing the relationship between factors and quality information for the entire process. The information processing device 10 can generate a path diagram of all factors, making it possible to visualize, for example, the factors that lead to the occurrence of product defects. As described above, the information processing device 10 can contribute to stabilizing product quality, reducing product defects, and accelerating these actions.

[0085] The information processing device 10 determines that factors whose partial correlation coefficient is equal to or greater than a first threshold are correlated with each other. This allows the information processing device 10 to eliminate spurious correlations between factors and facilitate the identification of true factors when estimating factors related to the final product quality information.

[0086] The information processing device 10 determines that the correlation between factors connected by paths corresponds to the sequence of processes in the manufacturing process 20, and if it determines that the correlation is valid, it determines that the correlation is valid. This allows the information processing device 10 to verify the validity of the causal relationships connected by paths based on the actual sequence of processes. Quality control personnel and others can easily determine whether the calculated causal relationships leading to the occurrence of defects, from factors in upstream equipment to factors in downstream equipment, are accurate and reflect reality.

[0087] Before calculating partial correlations, the information processing device 10 uses statistical testing methods to test the correlations between factors and groups those factors that are determined to be significant. This allows the information processing device 10 to shorten the calculation time by narrowing the scope of the partial correlation calculation process, which follows the statistical testing, to only the factors determined to be significant, thereby enabling more efficient partial correlation calculation. Furthermore, it becomes possible to perform detailed analysis focusing on specific datasets of factors determined to be significant by statistical testing methods, thereby gaining important insights.

[0088] The information processing device 10 determines the causal relationships between factors connected by paths by calculating conditional probabilities. By estimating the causal direction using conditional probabilities, the information processing device 10 eliminates spurious correlations (false correlations) and enables the estimation of a statistically sound causal structure.

[0089] The information processing device 10 tests correlations for all combinations of factors. By analyzing correlations between all factors without pre-specifying factors, the information processing device 10 can visualize how factors that appear unrelated at first glance influence multiple factors and propagate to the result. For example, the information processing device 10 can also add factors that are not directly correlated with defects to the analysis and illustrate how seemingly insignificant factors influence other factors and lead to defects. As a result, the information processing device 10 can reduce the risk that quality control personnel and others may overlook important factors.

[0090] The information processing device 10 sets extraction conditions for the path diagram and extracts and visualizes only the factors that satisfy the extraction conditions from the path diagram. This allows the information processing device 10 to extract and present only the high-priority factor groups required by quality control personnel from a complex overall path diagram, thereby improving visibility. Therefore, the information processing device 10 facilitates the identification of factors related to defects in the final product. The information processing device 10 can also improve usability.

[0091] The information processing device 10 visualizes the measurement data of a predetermined first factor in the path diagram, distinguishing it for each factor in the factor group that includes the second factor connected to the first factor and factors of the same type as the second factor. This allows the information processing device 10 to more accurately interpret, for example, the correlation between the first factor and the presence or absence of defects.

[0092] <Process Implementation Method> Based on the path diagram generated by the information processing device 10, factors that have a strong correlation with the quality information of the final product are identified from among the measurement data and quality information of each process in a manufacturing process that includes multiple processes. The conditions corresponding to the identified factors are optimized, and the manufacturing process is carried out based on those conditions. In this way, the quality of the final product is improved.

[0093] As a specific method for optimizing manufacturing conditions, the optimal conditions (e.g., operating conditions) for improving the quality of the final product can be identified by following the procedure below. First, the control unit 15 of the information processing device 10 extracts measurement data (process data), such as the state of the intermediate product or the state of the equipment, which are associated with factors that have a strong correlation with the quality information of the final product, as identified based on the path diagram. Next, the control unit 15 identifies the optimal conditions for manufacturing a defect-free product, based on the quality information of the final product linked to the extracted measurement data, using, for example, data science or machine learning techniques.

[0094] Specifically, a classification model (e.g., logistic regression, decision tree, random forest, support vector machine) is constructed to predict the presence or absence of defects in a product. Then, using the constructed model, the optimal conditions (process data) for manufacturing a defect-free product are identified using methods such as grid search or Bayesian optimization.

[0095] The output unit 14 outputs the optimal conditions identified by the control unit 15 to the control device of the manufacturing process 20. As a result, the manufacturing process 20 is carried out based on the optimal conditions under the control of the control device, and the product is manufactured. Ultimately, it becomes possible to manufacture products efficiently while improving product quality.

[0096] This disclosure belongs to the technical field relating to a manufacturing process 20 that includes multiple steps, such as the manufacturing process for steel plates, and is widely applicable in industrial fields such as the steel manufacturing industry and the steel plate processing industry.

[0097] While this disclosure has been described based on the drawings and embodiments, it should be noted that those skilled in the art can make various modifications and alterations based on this disclosure. Therefore, it should be noted that these modifications and alterations are within the scope of this disclosure. For example, the functions included in each configuration or step can be rearranged or omitted in a logically consistent manner, and multiple configurations or steps can be combined into one or divided into two.

[0098] For example, it is also possible to configure a general-purpose electronic device such as a smartphone or computer to function as the information processing device 10 according to the above-described embodiment. Specifically, a program describing the processing content that realizes each function of the information processing device 10 according to the embodiment is stored in the memory of the electronic device, and the processor of the electronic device reads and executes the program. Therefore, this disclosure can also be realized as a program that can be executed by a processor.

[0099] Alternatively, the Disclosure may also be implemented as a non-temporary computer-readable medium storing a program executable by one or more processors for causing an information processing device 10 or the like to perform each function according to one embodiment. It should be understood that these are also included within the scope of the Disclosure.

[0100] In the above embodiment, the information processing device 10 was described as determining that factors whose partial correlation coefficients are equal to or greater than a first threshold are correlated with each other and connecting them via paths, but it is not limited to this. The information processing device 10 may connect factors via paths in other ways. For example, the information processing device 10 may connect factors with the highest partial correlation coefficients among a predetermined number of pairs of factors via paths, or it may connect factors with partial correlation coefficients equal to or greater than the average value of the partial correlation coefficients across all factors via paths.

[0101] In the above embodiment, it was explained that the information processing device 10 performs a determination process to determine whether the correlation corresponds to the sequence of steps in the manufacturing process 20 after the extraction process in step S202 of Figure 3, but it is not limited to this. The information processing device 10 may perform a similar determination process on the entire path diagram for which the extraction process has not been performed. Alternatively, the information processing device 10 may not perform such a determination process at any stage.

[0102] In the above embodiment, it was explained that the information processing device 10 tests the correlation between factors using a statistical testing method before calculating partial correlations, and groups the factors that are determined to be significant; however, it is not limited to this. The information processing device 10 does not have to perform such grouping processing.

[0103] In the above embodiment, the information processing device 10 was described as testing for correlations for all combinations of factors, but it is not limited to this. The information processing device 10 may test for correlations for some combinations of factors, or it may not perform the testing process at all.

[0104] In the above embodiment, the information processing device 10 was described as setting extraction conditions for the path diagram and extracting and visualizing only the factors that satisfy the extraction conditions from the path diagram, but it is not limited to this. The information processing device 10 does not have to perform such an extraction process.

[0105] In the above embodiment, the information processing device 10 was described as visualizing the measurement data by distinguishing it by factor after the extraction process in step S202 of Figure 3, but it is not limited to this. The information processing device 10 may perform a similar visualization process on the overall path diagram for which the extraction process has not been performed. Alternatively, the information processing device 10 may not perform such a visualization process at any stage.

[0106] 1. Information Processing System 10. Information Processing Device 11. Communication Unit 12. Storage Unit 13. Input Unit 14. Output Unit 15. Control Unit 20. Manufacturing Process

Claims

1. An information processing device for analyzing the correlation between measurement data in the manufacturing process of a product and quality information of the product, comprising a control unit, the control unit acquires the measurement data at each step of the manufacturing process which includes multiple steps, calculates partial correlations between factors including the measurement data and the quality information to obtain correlations, connects the factors that are correlated with each other in a causal direction based on conditional probability using paths, and generates a path diagram that visualizes the correlation between the measurement data and the quality information in the manufacturing process.

2. An information processing device according to claim 1, wherein the control unit determines that factors whose partial correlation coefficient is equal to or greater than a first threshold are correlated with each other.

3. An information processing apparatus according to claim 1 or 2, wherein the control unit determines that the correlation between the factors connected by the path corresponds to the sequence of steps in the manufacturing process, and determines that the correlation is valid.

4. An information processing apparatus according to any one of claims 1 to 3, wherein the control unit tests the correlation between the factors using a statistical testing method before calculating the partial correlation, groups the factors that are determined to be significant, calculates the partial correlation between the factors including the measurement data and the quality information for each group of factors, obtains the correlation, connects the factors that are correlated with each other in a causal direction based on conditional probability using paths for each group of factors, and generates a path diagram that visualizes the correlation between the measurement data and the quality information in the manufacturing process.

5. An information processing apparatus according to claim 4, wherein the control unit tests the correlation for all combinations of the factors.

6. An information processing device according to claim 4 or 5, wherein the test method includes either Fisher's exact test method or a statistical causal inference method including a Bayesian network and a DAG (Directed Acyclic Graph)-DNN (Deep Neural Network).

7. An information processing device according to any one of claims 1 to 6, wherein the control unit calculates a first conditional probability that the first factor occurs and the second factor occurs, and a second conditional probability that the second factor occurs and the first factor occurs, for a first factor and a second factor that are correlated with each other, and connects the first factor and the second factor with a path in the causal direction of the first factor and the second factor estimated from the first conditional probability and the second conditional probability.

8. An information processing device according to any one of claims 1 to 7, wherein the control unit sets extraction conditions for the path diagram and extracts and visualizes only the factors that satisfy the extraction conditions from the path diagram.

9. An information processing device according to claim 8, wherein the extraction condition includes, in the path diagram, the factors of the start point and the end point, the path length, and a second threshold value for the strength of the partial correlation.

10. An information processing device according to claim 9, wherein, in the path diagram, the factor at the starting point is a factor related to the quality information, and the factor at the ending point is a factor that has a strong correlation with the quality information, a significant factor of the quality information according to Fisher's exact test, or a factor related to the measurement data with high variable importance in a machine learning model in which the quality information is the dependent variable and the measurement data is the independent variable.

11. An information processing device according to any one of claims 1 to 10, wherein the control unit visualizes the measurement data of a predetermined first factor in the path diagram, distinguishing it for each factor in a group of factors that includes a second factor connected to the first factor and a factor of the same type as the second factor.

12. An information processing method for analyzing the correlation between measurement data in a product manufacturing process and quality information of the product, comprising: acquiring the measurement data at each step in the manufacturing process which includes multiple steps; calculating partial correlations between factors including the measurement data and the quality information to obtain the correlation; and connecting the factors that are correlated with each other in a causal direction based on conditional probability using paths to generate a path diagram that visualizes the correlation between the measurement data and the quality information in the manufacturing process.

13. An information processing method according to claim 12, comprising: testing the correlation between the factors using a statistical testing method before calculating the partial correlation, and grouping the factors that are determined to be significant; calculating the partial correlation between the factors, each including the measurement data and the quality information, for each group of factors to obtain the correlation; and connecting the factors that are correlated with each other in a causal direction based on conditional probability using paths for each group of factors to generate a path diagram that visualizes the correlation between the measurement data and the quality information in the manufacturing process.

14. An information processing method according to claim 12 or 13, comprising: calculating a first conditional probability that the first factor occurs and the second factor occurs, and a second conditional probability that the second factor occurs and the first factor occurs, for a first factor and a second factor that are correlated with each other; and connecting the first factor and the second factor with a path in the causal direction of the first factor and the second factor estimated from the first conditional probability and the second conditional probability.

15. A process implementation method comprising: identifying factors among the measurement data and quality information of each step in the manufacturing process, which includes a plurality of steps, that have a strong correlation with the quality information of the final product, based on the path diagram generated by the information processing method according to any one of claims 12 to 14; optimizing the conditions corresponding to the identified factors in order to improve the quality of the final product; and carrying out the manufacturing process under the optimized conditions.

Images