Method and system for setting factor variable area

Abstract

To provide a technique capable of setting a range of factor variables which can reduce a defect.SOLUTION: A method of the present disclosure includes the steps of: (a) retrieving from memory a plurality of measured values of a factor variable and a label representing good or bad of a quality corresponding to each of the plurality of measured values; (b) equally dividing a range determined from a maximum value to a minimum value of the plurality of measured values for each factor variable, thereby dividing a factor variable space defined by the factor variable into a plurality of grids; (c) setting a plurality of candidate areas each of which includes one or a plurality of adjacent grids and deriving, for each of the plurality of candidate areas, a good density based on the label associated with the measured value falling in the candidate area; and (d) based on the good density, selecting one of the plurality of candidate regions as a factor variable region.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present disclosure relates to a method and system for setting a factor variable range. [Background technology]

[0002] Patent Document 1 discloses a factor analysis device that creates an analytical model from data on manufacturing conditions, quality characteristics, etc., and identifies explanatory variables that are the cause of defects. If a machine learning model is used as the analytical model, it is expected that the analysis can be performed with high accuracy. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2011-150496 Summary of the Invention [Problem to be solved by the invention]

[0004] However, when a machine learning model is used for factor analysis, although it can perform an accurate analysis, there are problems in that the model is difficult to interpret and it is unclear how to control the identified factor variables to reduce defects. For this reason, there has been a demand for a way to effectively set the range of factor variables that can reduce defects, regardless of whether a machine learning model is used or not. [Means for solving the problem]

[0005] According to a first aspect of the present disclosure, there is provided a method for setting a factor variable region that determines a range of values of one or more factor variables when a state of the manufacturing process is expressed as a good quality for a manufacturing process of a product or the product. The factor variable region defines a range of values of the factor variables when the state of the manufacturing process is expressed as a good quality or bad quality. The method includes the steps of: (a) retrieving from a memory a plurality of actual measurement values of the factor variables and labels indicating good or bad quality corresponding to each of the plurality of actual measurement values; (b) equally dividing a range determined by the maximum and minimum values of the plurality of actual measurement values for each of the factor variables to divide a factor variable space defined by the factor variables into a plurality of grids; (c) setting a plurality of candidate regions each including one or a plurality of adjacent grids, and deriving a good density for each of the plurality of candidate regions based on the labels associated with the actual measurement values that fall in the candidate region; and (d) selecting one of the plurality of candidate regions as the factor variable region based on the good density. The method includes, before step (d), a step of acquiring values at positions that equally divide each of the plurality of grids for each of the factor variables as dummy values for the factor variables, and a step of deriving estimated labels that indicate the estimated pass / fail of the quality by inputting the dummy values into a decision model that has learned the correspondence between the actual measurement values and the labels. Step (c) includes a step of deriving, for each of the plurality of candidate regions, (i) a first good density based on the labels associated with the actual measurement values and (ii) a second good density based on the estimated labels associated with the dummy values, as the good densities that fall within the candidate region. Step (d) selects one of the plurality of candidate regions as the factor variable region using the first good density and the second good density. The step (d) includes a step of calculating, for each of the plurality of candidate areas, a first score that is positively correlated with the first good density for the candidate area and the size of the candidate area; a step of calculating, for each of the plurality of candidate areas, a second score that is positively correlated with the second good density for the candidate area and the size of the candidate area; and a step of selecting one of the plurality of candidate areas as the factor variable area according to the first score and the second score.

[0006] According to a second aspect of the present disclosure, there is provided a system for executing a process of determining a factor variable region that determines a quality of a manufacturing process for producing a product or the product, the factor variable region defining a range of values of one or more factor variables when a state under the manufacturing process is expressed as a value of the factor variable. The system includes a memory that stores a plurality of actual measured values of the factor variables and labels that indicate whether the quality is good or bad corresponding to each of the plurality of actual measured values, and one or more processors configured to execute a process of determining the factor variable region. The processor performs the following processes: (a) retrieving the plurality of actual measured values and the labels corresponding to each of the plurality of actual measured values from the memory; (b) dividing a factor variable space defined by the factor variables into a plurality of grids by equally dividing a range determined by the maximum and minimum values of the plurality of actual measured values for each of the factor variables; (c) deriving a good density for each of the plurality of grids based on the labels associated with the actual measured values that fall into the grid; and (d) setting one or a plurality of adjacent grids as the factor variable region based on the good density. Before the process (d), the process executes the following steps: acquiring values at positions that equally divide each of the plurality of grids for each of the factor variables as dummy values for the factor variables; and deriving estimated labels that indicate the estimated quality by inputting the dummy values into a decision model that has learned the correspondence between the actual measurement values and the labels. The process (c) includes a process of deriving, for each of the plurality of candidate regions, (i) a first good density based on the labels associated with the actual measurement values and (ii) a second good density based on the estimated labels associated with the dummy values, as the good densities that fall within the candidate region. The process (d) selects one of the plurality of candidate regions as the factor variable region using the first good density and the second good density. The process (d) includes a process of calculating, for each of the plurality of candidate areas, a first score that is positively correlated with the first good density for the candidate area and the size of the candidate area; a process of calculating, for each of the plurality of candidate areas, a second score that is positively correlated with the second good density for the candidate area and the size of the candidate area; and a process of selecting one of the plurality of candidate areas as the factor variable area according to the first score and the second score. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a conceptual diagram of a factor analysis system according to an embodiment. [Figure 2] 5 is a flowchart showing a procedure for setting a factor variable area in the first embodiment. [Figure 3] FIG. 10 is an explanatory diagram showing an example of dividing a factor variable space into grids. [Figure 4] FIG. 10 is an explanatory diagram showing another example of grid division. [Figure 5] FIG. 10 is an explanatory diagram showing the good density for each grid. [Figure 6] FIG. 10 is an explanatory diagram showing the good density for each candidate region. [Figure 7] FIG. 10 is an explanatory diagram showing density differences for each candidate region. [Figure 8] 10 is a flowchart showing a procedure for setting a factor variable area in the second embodiment. [Figure 9] FIG. 10 is an explanatory diagram showing estimated labels and their second best densities for dummy values of factor variables. [Figure 10] FIG. 10 is an explanatory diagram showing the second good density for each candidate region. [Figure 11] FIG. 10 is an explanatory diagram showing a second density difference for each candidate region. [Figure 12] 10 is a flowchart showing a procedure for setting a factor variable area in the third embodiment. [Figure 13] FIG. 10 is an explanatory diagram showing the first score, the second score, and the integrated score for each grid. [Figure 14] FIG. 10 is an explanatory diagram showing the integrated score for each candidate region. DETAILED DESCRIPTION OF THE INVENTION

[0008] A. First embodiment: 1 is a block diagram showing a factor analysis system according to an embodiment of the present invention, which includes an information processing device 100 and a production line 200.

[0009] The manufacturing line 200 includes a manufacturing processing device 210 that processes the product PD and an inspection device 220 that inspects the processed product PD for quality. The processing conditions PC of the manufacturing processing device 210 and the inspection results IR of the inspection device 220 are supplied to the information processing device 100. The inspection results IR include a label indicating whether the product PD is good or bad. This label indicates whether the manufacturing process of the product PD or the quality of the product PD is good or bad. The processing conditions PC include, for example, environmental data such as atmospheric pressure, set values for processing voltage and pressure, and measurement data during the manufacturing process.

[0010] The information processing device 100 has a processor 110, a memory 120, and an interface circuit 130. An input device 140 and a display unit 150 are also connected to the interface circuit 130, and a manufacturing processing device 210 and an inspection device 220 are also connected to the interface circuit 130. The processor 110 not only has the function of executing the processes described in detail below, but also has the function of displaying on the display unit 150 data obtained by the processes and data generated in the process of the processes.

[0011] The processor 110 functions as a factor variable area setting unit 112 that executes the factor variable area setting unit. The function of the factor variable area setting unit 112 is realized by the processor 110 executing a computer program stored in the memory 120. However, the function of the factor variable area setting unit 112 may also be realized by a hardware circuit. The term "processor" used in this specification also includes such hardware circuits. Furthermore, the processor that executes the processing of the factor variable area setting unit 112 may be a processor included in a remote computer connected to the information processing device 100 via a network. Furthermore, the processing of the factor variable area setting unit 112 may be executed by multiple processors.

[0012] The memory 120 stores manufacturing data MD and a judgment model JM. The manufacturing data MD is a set of processing conditions PC and inspection results IR. The judgment model JM is a machine learning model that finds an estimated label from factor variables. The factor variables are some of the variables of the processing conditions PC, and the estimated label is an estimated value of a label indicating the pass / fail of the inspection result IR. As the machine learning model, various neural networks, support vector machines, random forests, etc. can be used. In this embodiment, it is assumed that the judgment model JM has already been trained. However, the judgment model JM may be omitted.

[0013] In addition, the terms "explanatory variable" and "objective variable" are commonly used in factor analysis. The explanatory variable is a variable that is a cause, and the objective variable is a variable that indicates the result that occurs in response to that cause. In this disclosure, the factor variable that is part of the processing conditions PC corresponds to the "explanatory variable," and the label of the test result IR corresponds to the "objective variable."

[0014] 2 is a flowchart showing the procedure for setting a factor variable area in the first embodiment. In step S110, the factor variable area setting unit 112 reads the actual measured values of the factor variables and the pass / fail labels from the memory 120. The actual measured values of the factor variables are actual data that are part of the processing condition PC. The pass / fail labels are associated with the actual measured values of the factor variables. In step S120, the factor variable area setting unit 112 divides the factor variable space into grids.

[0015] FIG. 3 is an explanatory diagram showing an example of dividing a factor variable space into grids. The horizontal axis X and vertical axis Y in FIG. 3 represent factor variables. However, the number of factor variables can be set to any number equal to or greater than 1. In this disclosure, the space represented by factor variables is referred to as the "factor variable space." The upper diagram in FIG. 3 shows the distribution of pass / fail labels. White circles represent pass labels, and crosses represent fail labels.

[0016] Points representing multiple pass / fail labels exist between the maximum value Xmax and the minimum value Xmin of the actual measurement values of factor variable X, and between the maximum value Ymax and the minimum value Ymin of the actual measurement values of factor variable Y. The lower diagram in Figure 3 shows an example in which the factor variable space containing all pass / fail labels is divided into nine grids G(i,j). Here, i and j indicate the row and column positions of the grid, where i = 1 to 3 and j = 1 to 3. These grids G(i,j) are set by equally dividing the range determined by the maximum value Xmax to the minimum value Xmin of the actual measurement values of factor variable X, and by equally dividing the range determined by the maximum value Ymax to the minimum value Ymin of the actual measurement values of factor variable Y. However, the outermost X positions of grid G(i,j) are set to the positions obtained by adding a margin δX to each of the maximum value Xmax and the minimum value Xmin of factor variable X. This margin δX may be zero or may be set to a small non-zero value. The same applies to the margin δY of factor variable Y. Generally, when the number of factor variables, m, is an integer equal to or greater than 1, G(i,j) becomes an m-dimensional solid. As shown in the example of Figure 3, when m=2, the grid shape is rectangular, but a rectangle is also a type of m-dimensional solid. Since the ranges of the factor variables X and Y are divided equally, the size of each grid G(i,j) is the same.

[0017] 3, each of the factor variables X and Y is divided into three, but the number of divisions can be set to any value equal to or greater than 2. Also, in FIG. 3, the number of grid divisions is the same for the two factor variables X and Y, but the number of grid divisions may be changed for each factor variable using the contribution or importance of each of the factor variables X and Y calculated using the determination model JM.

[0018] FIG. 4 is an explanatory diagram showing an example of grid division according to the importance I(X) and I(Y) of factor variables X and Y. In the upper diagram of FIG. 4, the importance I(X) of factor variable X is higher than the importance I(Y) of factor variable Y, so the number of divisions of the factor variable X grid is set to a larger value than the number of divisions of the factor variable Y grid. In the upper diagram of FIG. 4, the difference between the importance I(X) of factor variable X and the importance I(Y) of factor variable Y is even larger, so the difference between the number of divisions of the factor variable X grid and the number of divisions of the factor variable Y grid is set even larger. In this way, by increasing the number of divisions on the axis of a factor variable with high importance and decreasing the number of divisions on the axis of a factor variable with low importance, it is possible to emphasize the boundaries of more meaningful factor variables and neglect the boundaries of less meaningful factor variables. This makes it possible to efficiently set factor variable regions while reducing the amount of calculation. Note that, for example, permutation importance, an index value indicating the usefulness of the features of a machine learning model, can be used as the contribution and importance of factor variables. Furthermore, when the determination model JM is configured as a random forest, it is also possible to use the Gini coefficient as the contribution or importance of the factor variables. In the following explanation, the example of grid division shown in Figure 3 is used.

[0019] 2, the factor variable region setting unit 112 derives a good density for each grid. The good density Dg is calculated by the following equation. Dg=Ng(k) / ΣNg (E1) Here, k is the ordinal number of the grid, Ng(k) is the number of good labels contained in the k-th grid, and ΣNg is the total number of good labels in all grids.

[0020] In the above equation (E1), since ΣNg is a constant value, the number Ng(k) of good labels contained in the grid may be used as the good density. In general, the "good density" of a grid can be defined as a variable indicating a value proportional to the number Ng(k) of good labels contained in the grid.

[0021] 5 is an explanatory diagram showing the good density for each grid. In this example, the good density Dg of the two grids G(2,2) and G(3,2) is 0.286, which is the maximum.

[0022] In step S140 of Fig. 2, the factor variable region setting unit 112 sets multiple candidate regions and derives a good density for each candidate region. A candidate region is set as a region of one grid or a region including multiple adjacent grids. Each grid G(i,j) shown in Fig. 5 is also a candidate region, but since the good density for each grid G(i,j) has already been calculated in step S130, in step S140, the good density is calculated for a candidate region including multiple adjacent grids.

[0023] FIG. 6 is an explanatory diagram showing an example of candidate area setting and the good density for each candidate area. Here, the good densities of the nine grids G(i,j) shown in FIG. 5 and the other four candidate areas CA1 to CA4 are illustrated. The first candidate area CA1 is composed of two grids G(2,2) and G(3,2) adjacent to each other on the top and bottom. The second candidate area CA2 is composed of three grids G(1,2), G(2,2), and G(3,2) adjacent to each other on the top and bottom. The third candidate area CA3 is composed of two grids G(1,2) and G(2,2) adjacent to each other on the top and bottom. The fourth candidate area CA4 is composed of two grids G(2,1) and G(2,2) adjacent to each other on the left and right. Each of these four candidate areas CA1 to CA4 is set to include the central grid G(2,2), which has the highest good density Dg for each grid G(i,j). The reason for this is that the candidate area including the central grid G(2,2) with the highest density Dg is likely to be selected as the final factor variable space. However, other candidate areas may also be set.

[0024] The good density Dg for each candidate region is given by the following equation, which is similar to the above equation (E1). Dg=Ng(n) / ΣNg (E2) Here, n is the ordinal number of the candidate region, Ng(n) is the number of good labels contained in the nth candidate region, and ΣNg is the total number of good labels in all grids. Note that each grid also constitutes a candidate region, so equation (E2) can be considered a generalization of equation (E1) above. In the example of Figure 6, the second candidate region CA2 has the highest good density at 0.714.

[0025] In the above equation (E2), since ΣNg is a constant value, the number Ng(n) of good labels contained in the candidate region may be used as the good density. In general, the "good density" of a candidate region can be defined as a variable indicating a value proportional to the number Ng(n) of good labels contained in the candidate region.

[0026] 2, the factor variable region setting unit 112 derives the density difference for each candidate region. The density difference ΔD is given by the following equation. ΔD=Dg-Db (E3a) Dg = Ng(n) / ΣNg (E3b) Db=Nb(n) / ΣNb (E3c) Here, Dg is the good density, Db is the bad density, n is the ordinal number of the candidate region, Ng(n) is the number of good labels contained in the nth candidate region, ΣNg is the total number of good labels in all grids, Nb(n) is the number of bad labels contained in the nth candidate region, and ΣNb is the total number of bad labels in all grids. The above formula (E3b) is the same as the above formula (E2).

[0027] 7 is an explanatory diagram showing the density difference for each candidate area. In this example, the density difference ΔD between the first candidate area CA1 and the second candidate area CA2 is both 0.500, the maximum.

[0028] In step S160 of Fig. 2, the factor variable region setting unit 112 selects a factor variable region from the candidate regions using the good density. This selection can be performed using the good density for each candidate region shown in Fig. 6 or the density difference for each candidate region shown in Fig. 7.

[0029] 6 is used, the second candidate area CA2 has the highest good density among all the candidate areas, and therefore is selected as the factor variable area. This factor variable area is used as the range of processing conditions in the manufacturing processing device 210 when processing the product PD in the manufacturing line 200 shown in FIG. 1. As a result, it is possible to increase the likelihood that the processed product PD will be determined to be good by the inspection device 220.

[0030] When the density difference shown in Figure 7 is used, the density difference between the first candidate area CA1 and the second candidate area CA2 is the largest among all the candidate areas. However, since the second candidate area CA2 includes grid G(1,2) where the density difference is zero, it is preferable to select the first candidate area CA1, which does not include this grid G(1,2), as the factor variable area. Note that this method of determining a factor variable area based on the density difference ΔD can also be considered a type of method of determining a factor variable area based on good density.

[0031] As a method for selecting a factor variable region, the method using density difference shown in Fig. 7 is preferable to the method using good density shown in Fig. 6. The reason for this is that in the method using density difference shown in Fig. 7, the first candidate region CA1 selected as the factor variable region is a region that does not include grid G(1,2) with many defective labels shown in Fig. 3. Therefore, the method using density difference shown in Fig. 7 can select a factor variable region with a lower rate of occurrence of defects.

[0032] The candidate area CA1 to be the factor variable area is, as shown in FIG. 7, the upper limit X of the factor variable X. U and lower limit X L and the upper limit Y of the factor variable Y U and the lower limit Y LTherefore, since the ranges of the individual factor variables X and Y can be controlled independently as the ranges of the processing conditions in the manufacturing processing apparatus 210, there is an advantage in that the processing conditions can be easily controlled. It is preferable that the other candidate regions are also set to have shapes that are respectively defined by the ranges of the upper and lower limits of the individual factor variables.

[0033] As described above, in the first embodiment, a factor variable region is determined based on good density in a candidate region including one grid or multiple adjacent grids, so that a factor variable region with good quality can be easily found.

[0034] B. Second embodiment: Fig. 8 is a flowchart showing the procedure for setting the factor variable region in the second embodiment. Note that the device configuration is the same as that shown in Fig. 1, so a description thereof will be omitted here. Steps S110 to S130 in Fig. 8 are the same as steps S110 to S130 in Fig. 2 in the first embodiment, and in the second embodiment, the processing from step S140 onwards in Fig. 2 is replaced by steps S210 to S250. When the processing up to step S130 has been performed, the good density for each grid has been calculated, as shown in Fig. 5.

[0035] In step S210, the factor variable region setting unit 112 sets division positions within each grid and determines dummy values for the factor variables at each division position. In step S220, the factor variable region setting unit 112 derives estimated labels corresponding to the dummy values using the determination model JM.

[0036] The upper part of Figure 9 shows an example of setting division positions within each grid and an example of estimated labels according to the dummy values of factor variables X and Y at each division position. In this example, 2 x 2 division positions are set uniformly distributed within each grid G(i, j). The dummy values of factor variables X and Y at each division position are automatically determined by these division positions. Estimated labels according to the dummy values of factor variables X and Y at each division position can be found by inputting the dummy values into the judgment model JM.

[0037] In step S230 of FIG. 8, the factor variable region setting unit 112 sets candidate regions and derives a first good density and a second good density for each candidate region. Similar to the first embodiment, a candidate region is set as a region of one grid or a region including multiple adjacent grids. The "first good density" is a good density based on labels associated with the actual measured values of the factor variables X and Y, and is calculated using the above formula (E2) described in the first embodiment. Therefore, the first good density is the same as the good density shown in FIG. 6 described in the first embodiment. The "second good density" is a good density based on estimated labels associated with dummy values. The value of the second good density for each grid G(i, j) is shown at the bottom of FIG. 9.

[0038] Fig. 10 is an explanatory diagram showing the second good density for each candidate region. In the example of Fig. 10, the second good density of the second candidate region CA2 is 0.471, which is the highest among all the candidate regions.

[0039] In step S240 of FIG. 8, the factor variable region setting unit 112 derives a first density difference and a second density difference for each candidate region. The "first density difference" and "second density difference" are the density difference ΔD given by equation (E3a) described in the first embodiment. In the second embodiment, the density difference ΔD calculated using the actual measured values of the factor variables X and Y and the pass / fail labels is called the "first density difference," and the density difference ΔD calculated using the dummy values of the factor variables X and Y and the estimated labels is called the "second density difference." The first density difference is the density difference shown in FIG. 7 described in the first embodiment.

[0040] 11 is an explanatory diagram showing the second density difference for each candidate area. In this example, the second density difference ΔD between the first candidate area CA1 and the fourth candidate area CA4 is 0.359, which is the largest.

[0041] In step S250 of Fig. 8, the factor variable region setting unit 112 selects a factor variable region from the candidate region using the first good density and the second good density. This selection can be performed using the first good density and the second good density for each candidate region shown in Fig. 6 and Fig. 10, or the second density difference and the second density difference for each candidate region shown in Fig. 7 and Fig. 11.

[0042] As a method for selecting one candidate region as a factor variable region using the first good density shown in FIG. 6 and the second good density shown in FIG. 10, one of the following methods can be used.

[0043] <Method M11> Only one of the first and second good densities is used, and the candidate region having the maximum value is selected as the factor variable region. For example, if only the first good density is used, this method M11 is the same as the method used in step S150 of the first embodiment described above. Alternatively, only the second good density may be used to select a factor variable region. In the example of Figure 10, in this case, the second candidate region CA2 is selected as the factor variable region.

[0044] <Method M12> Both the first and second good densities are used to select candidate regions as factor variable regions. For example, the first and second good densities may be added together to calculate an integrated good density, and the candidate area with the largest integrated good density may be selected as the factor variable area. This addition may be a simple addition or a weighted addition. Alternatively, the first and second good densities may be multiplied together to calculate the integrated good density, and the candidate area with the largest integrated good density may be selected as the factor variable area.

[0045] The method of selecting one candidate region as a factor variable region using the first density difference shown in FIG. 7 and the second density difference shown in FIG. 11 may also be similar to the above-described method M11, in that only one of the first density difference and the second density difference is used to select the candidate region with the largest value as the factor variable region. Alternatively, similar to the above-described method M12, both the first density difference and the second density difference may be used to select a factor variable region. For example, when determining a factor variable region using only the second density difference ΔD, either the first candidate region CA1 or the fourth candidate region CA4 is selected as the factor variable region. Referring to the lower diagram of FIG. 3, it can be seen that these candidate regions CA1 and CA4 are regions with many good labels corresponding to the actual measured values of the factor variables X and Y and almost no bad labels. Therefore, these candidate regions CA1 and CA4 are particularly preferable regions as control ranges of the factor variables that will result in good inspection results. The method of determining a factor variable region based on the second density difference ΔD can also be considered a type of method of determining a factor variable region based on the second good density.

[0046] As described above, in the second embodiment, a factor variable region is selected using the first good density corresponding to the actual measured value of the factor variable and the second good density corresponding to the dummy value of the factor variable, so that the factor variable region can be set more accurately.

[0047] C. Third embodiment: FIG. 12 is a flowchart showing the procedure for setting factor variable regions in the third embodiment. Note that the device configuration is the same as that shown in FIG. 1, so a description thereof will be omitted here. Steps S110 to S130 in FIG. 12 are the same as steps S110 to S130 in FIG. 2 in the first embodiment, and steps S210 to S230 in FIG. 12 are the same as steps S210 to S230 in FIG. 8 in the second embodiment. In the third embodiment, the processing from step S240 onwards in FIG. 8 is replaced by steps S310 to S320. When the processing up to step S230 is performed, the first good density shown in FIG. 6 and the second good density shown in FIG. 10 are obtained for each candidate region.

[0048] In step S310, the factor variable region setting unit 112 calculates a first score and a second score for the candidate region. The first score is calculated using the first good density, and the second score is calculated using the second good density. These scores Sc are calculated, for example, using the following equations. Sc={(Dg-Db) / (Dg+Db)}×Dg (E4) Here, Dg is the good density given by the above formula (E3b), and Db is the bad density given by the above formula (E3c). Both the first score and the second score are calculated by the above formula (E4), but when distinguishing between the two, they are called the first score Sc_1 and the second score Sc_2 by adding "_1" and "_2" to the end.

[0049] The value in the curly brackets on the right side of (E4) above can be considered to be an index representing the ratio of the number of good labels to bad labels in the candidate region. Also, Dg, the last element on the right side of (E4) above, can be considered to be an index representing the size of the candidate region. In this way, the score Sc is a value that has a positive correlation with the ratio of the number of good labels to bad labels in the candidate region and the size of the candidate region. Alternatively, the score Sc can be considered to be a value that has a positive correlation with the density of goodness for the candidate region and the size of the candidate region.

[0050] The first score Sc_1 is a score obtained by calculating the good density Dg and the defective density Db in the above formula (E4) based on the good / fail labels associated with the actual measured values of the factor variables X and Y. The second score Sc_2 is a score obtained by calculating the good density Dg and the defective density Db in the above formula (E4) based on the estimated labels associated with the dummy values of the factor variables X and Y.

[0051] Figure 13 shows the first score and second score calculated for each grid. For the first score, the score value for the central grid G(2,2) is 0.286, which is the highest. For the second score, the score value for the central grid G(2,2) is also 0.235, which is the highest. For candidate areas that include multiple grids, the first score and second score are calculated in a similar manner, but are not shown in the figures.

[0052] 12, the factor variable region setting unit 112 selects a factor variable region from the candidate regions using the first score and the second score. As a method for selecting one candidate region as a factor variable region using the first score and the second score, any of the following methods can be used.

[0053] <Method M21> Only one of the first score and the second score is used, and the candidate region having the maximum value is selected as the factor variable region.

[0054] <Method M22> Both the first and second scores are used to select candidate regions as factor variable regions. For example, the first score and the second score may be added together to calculate an integrated score, and the candidate region with the largest integrated score may be selected as the factor variable region. This addition may be a simple addition or a weighted addition. Alternatively, the first score and the second score may be multiplied together to calculate an integrated score, and the candidate region with the largest integrated score may be selected as the factor variable region.

[0055] In the third embodiment, of the above-mentioned method M22, a method of calculating an integrated score by multiplying the first score and the second score is adopted. The integrated score St is given by the following equation. St = Sc_1 × Sc_2 (E5) where Sc_1 is the first score and Sc_2 is the second score.

[0056] The integrated score St for each grid is shown in the lower part of Fig. 13. The integrated score St also has the largest value of 0.067 in the central grid G(2,2).

[0057] FIG. 14 is an explanatory diagram showing the total score St for each candidate area. Here, among the nine grids G(i,j) and the other four candidate areas CA1 to CA4, the first candidate area CA1 has the largest total score St at 0.142. Therefore, in step S320, this candidate area CA1 is selected as the factor variable area. Referring to the lower diagram in FIG. 3, it can be seen that this candidate area CA1 is an area with many good labels corresponding to the actual measured values of the factor variables X and Y and almost no bad labels. Therefore, this candidate area CA1 is a particularly preferable area as a control range of the factor variables that will produce good inspection results.

[0058] The score Sc may be calculated using a formula or function other than the formula (E4). In either case, however, it is preferable that the score Sc has a positive correlation with both the quality density of the candidate region and the size of the candidate region. In this way, a region with a higher quality density and a larger size can be set as the factor variable region.

[0059] As described above, in the third embodiment, it is possible to easily set an appropriate factor variable region by using the score.

[0060] In the above-described embodiments, the factor variables X and Y are quantitative variables. However, the present disclosure can also be applied to cases where some of the factor variables are qualitative variables. When qualitative factor variables are included, there is no concept of minimum or maximum values as there is for quantitative factor variables, so grid division is not performed for the qualitative factor variables. Furthermore, when determining candidate areas, a different search method may be required for qualitative factor variables than for quantitative factor variables. For example, for qualitative factor variables, the qualitative factor variables may be classified into multiple categories, and when setting candidate areas, some categories may be combined using a logical OR to set candidate areas. Specifically, multiple categories may be combined using a logical OR in descending order of the number of non-defective labels to set candidate areas. This makes it possible to set factor variable areas in a factor variable space having both qualitative and quantitative factor variables.

[0061] Other forms: The present disclosure is not limited to the above-described embodiments and can be realized in various forms without departing from the spirit thereof. For example, the present disclosure can also be realized in the following aspects. The technical features in the above embodiments corresponding to the technical features in each aspect described below can be appropriately replaced or combined to solve some or all of the problems of the present disclosure or to achieve some or all of the effects of the present disclosure. Furthermore, if a technical feature is not described as essential in this specification, it can be appropriately deleted.

[0062] (1) According to a first aspect of the present disclosure, there is provided a method for setting a factor variable region that determines a quality of a manufacturing process for producing a product or a product, the factor variable region defining a range of values of the factor variable when a state of the manufacturing process is expressed as a value of one or more factor variables. The method includes the steps of: (a) retrieving from a memory a plurality of actual measurement values of the factor variables and labels indicating quality levels corresponding to the plurality of actual measurement values; (b) equally dividing a range determined by the maximum and minimum values of the plurality of actual measurement values for each of the factor variables to divide a factor variable space defined by the factor variables into a plurality of grids; (c) setting a plurality of candidate regions each including one or a plurality of adjacent grids, and deriving a quality density for each of the plurality of candidate regions based on the labels associated with the actual measurement values falling in the candidate region; and (d) selecting one of the plurality of candidate regions as the factor variable region based on the quality density. According to this method, a factor variable region is determined based on good density in a candidate region including one grid or a plurality of adjacent grids, so that a factor variable region with good quality can be easily determined.

[0063] (2) The method may include, before step (d), a step of acquiring values at positions that equally divide each of the plurality of grids for each of the factor variables as dummy values of the factor variables, and a step of deriving estimated labels that indicate the estimated pass / fail of the quality by inputting the dummy values into a decision model that has learned the correspondence between the actual measurement values and the labels. Furthermore, step (c) may include a step of deriving, as the good densities for each of the plurality of candidate regions, (i) a first good density based on the labels associated with the actual measurement values and (ii) a second good density based on the estimated labels associated with the dummy values, which fall within the candidate region, and step (d) may select one of the plurality of candidate regions as the factor variable region using the first good density and the second good densities. According to this method, the factor variable region is selected using the first good density corresponding to the actual measured value of the factor variable and the second good density corresponding to the dummy value of the factor variable, so that the factor variable region can be set more accurately.

[0064] (3) In the above method, step (d) may include a step of calculating, for each of the plurality of candidate regions, a score that is positively correlated with the good density for the candidate region and the size of the candidate region, and a step of selecting one of the plurality of candidate regions as the factor variable region according to the score. According to this method, the factor variable range can be easily set using the score.

[0065] (4) In the above method, step (d) may include the steps of: calculating, for each of the plurality of candidate areas, a first score that is positively correlated with the first good density for the candidate area and the size of the candidate area; calculating, for each of the plurality of candidate areas, a second score that is positively correlated with the second good density for the candidate area and the size of the candidate area; and selecting one of the plurality of candidate areas as the factor variable area according to the first score and the second score. According to this method, the factor variable range can be easily set using the score.

[0066] (5) According to a second aspect of the present disclosure, there is provided a system for executing a process of setting a factor variable region that determines a quality of a manufacturing process for producing a product or the quality of the product, the factor variable region defining a range of values of one or more factor variables when a state under the manufacturing process is expressed as a value of the factor variable. The system includes a memory that stores a plurality of actual measured values of the factor variables and labels that indicate whether the quality is good or bad corresponding to each of the plurality of actual measured values, and one or more processors configured to execute a process of determining the factor variable region. The processor performs the following processes: (a) retrieving the plurality of actual measured values and the labels corresponding to each of the plurality of actual measured values from the memory; (b) dividing a factor variable space defined by the factor variables into a plurality of grids by equally dividing a range determined by the maximum and minimum values of the plurality of actual measured values for each of the factor variables; (c) deriving a good density for each of the plurality of grids based on the labels associated with the actual measured values that fall into the grid; and (d) setting one or a plurality of adjacent grids as the factor variable region based on the good density.

[0067] The present disclosure may be realized in various forms other than those described above, such as a factor analysis device, a computer program for realizing the functions of the factor analysis device, or a non-transitory storage medium on which the computer program is recorded. [Explanation of symbols]

[0068] 100...information processing device, 110...processor, 112...factor variable area setting unit, 120...memory, 130...interface circuit, 140...input device, 150...display unit, 200...production line, 210...production processing device, 220...inspection device

Claims

1. 1. A method for setting a factor variable range that determines a manufacturing process for manufacturing a product or a quality of the product, the factor variable range defining a range of values of one or more factor variables when a state under the manufacturing process is expressed as a value of the factor variable, comprising: (a) retrieving from a memory a plurality of actual measurement values of the factor variables and labels representing the quality corresponding to each of the plurality of actual measurement values; (b) dividing a factor variable space defined by the factor variables into a plurality of grids by equally dividing a range determined by the maximum value to the minimum value of the plurality of actual measurement values for each of the factor variables; (c) setting a plurality of candidate regions each including one or a plurality of adjacent grids, and deriving a good density for each of the plurality of candidate regions based on the label associated with the actual measurement value falling in the candidate region; (d) selecting one of the plurality of candidate regions as the factor variable region based on the good density; Including, Before the step (d), acquiring, for each of the factor variables, values at positions that equally divide each of the plurality of grids as dummy values of the factor variables; a step of inputting the dummy values into a judgment model that has learned the correspondence relationship between the actual measurement values and the labels, thereby deriving an estimated label that indicates whether the quality is good or bad; Including, the step (c) includes a step of deriving, for each of the plurality of candidate regions, (i) a first good density based on the label associated with the actual measurement value, and (ii) a second good density based on the estimated label associated with the dummy value, which fall within the candidate region, as the good densities; The step (d) selects one of the plurality of candidate regions as the factor variable region using the first good density and the second good density; The step (d) calculating, for each of the plurality of candidate regions, a first score that is positively correlated with the first good density for the candidate region and with the size of the candidate region; calculating, for each of the plurality of candidate regions, a second score that is positively correlated with the second good density for the candidate region and with the size of the candidate region; selecting one of the plurality of candidate regions as the factor variable region according to the first score and the second score; A method comprising:

2. A system for executing a process of setting a factor variable range that determines a manufacturing process for manufacturing a product or a quality of the product, the factor variable range defining a range of values of one or more factor variables when a state under the manufacturing process is expressed as a value of the factor variable, comprising: a memory that stores a plurality of actual measurement values of the factor variables and labels that indicate whether the quality is good or bad and correspond to each of the plurality of actual measurement values; one or more processors configured to execute a process for determining the factor variable domain; Equipped with The processor: (a) retrieving the plurality of actual measurement values and the labels corresponding to each of the plurality of actual measurement values from the memory; (b) dividing a range determined by the maximum value to the minimum value of the plurality of actual measurement values for each of the factor variables equally into a plurality of grids to divide a factor variable space defined by the factor variables; (c) deriving, for each of the plurality of grids, a good density based on the labels associated with the actual measurements falling into the grid; (d) setting one or a plurality of adjacent grids as the factor variable region based on the good density; Run Before the treatment (d), a process of acquiring values at positions that equally divide each of the plurality of grids for each of the factor variables as dummy values of the factor variables; a process of deriving an estimated label representing the estimated quality by inputting the dummy value into a judgment model that has learned the correspondence relationship between the actual measurement value and the label; Run The process (c) includes a process of deriving, for each of the plurality of candidate regions, (i) a first good density based on the label associated with the actual measurement value, and (ii) a second good density based on the estimated label associated with the dummy value, which fall within the candidate region, as the good densities; The process (d) uses the first good density and the second good density to select one of the plurality of candidate regions as the factor variable region; The process (d) a process of calculating, for each of the plurality of candidate regions, a first score that is positively correlated with the first good density for the candidate region and with the size of the candidate region; a process of calculating, for each of the plurality of candidate regions, a second score that is positively correlated with the second good density for the candidate region and with the size of the candidate region; a process of selecting one of the plurality of candidate regions as the factor variable region according to the first score and the second score; Including, the system.

Images