JP6849543B2 - Defective factor analysis system and defective factor analysis method - Google Patents

Description

The present invention relates to a defective factor analysis system and a defective factor analysis method for analyzing defective factors of a product.

In order to evaluate the manufacturing process of a product, the cause of defective products is analyzed. Patent Document 1 divides a data set consisting of process data for a plurality of process items (temperature, pressure, power input to a heater, flow rate of material gas, etc.) for each of a plurality of products into non-defective product data and defective product data. However, a method of performing principal component analysis on non-defective product data and defective product data and selecting defective product data having the minimum distance from the center point in the principal component space of the defective product is disclosed. In this method, the selected defective product data is displayed together with the non-defective product data in order to be used for the analysis of the defective factor by the user.

Japanese Unexamined Patent Publication No. 2011-13195

In the conventional method as described above, the distribution of the non-defective product data and the defective product data in the main component space overlaps, and the central point of the defective product data in the main component space is included in the distribution range of the non-defective product data. In such a case, there is a problem that the data of the selected defective product does not sufficiently reflect the defective factor, and it becomes difficult to grasp the defective factor.

The present invention has been made in view of such circumstances, and a main object thereof is to provide a defect factor analysis system and a defect factor analysis method capable of solving the above problems.

In order to solve the above-mentioned problems, the defect factor analysis system according to one aspect of the present invention includes a plurality of product-specific objective variable data for objective variables used for determining whether a product is good or defective, and the product. For a data set including the explanatory variable data for each of the plurality of products for each of the plurality of explanatory variables showing the characteristics of the objective variable data, the frequency distribution creating means for creating the frequency distribution of the objective variable data and the frequency distribution creating. A setting means for setting two extraction regions separated from each other in the frequency distribution created by the means, a product in which the objective variable data is included in one of the two extraction regions set by the setting means, and a product. On the other hand, the calculation means for calculating the difference in the feature amount representing the difference in the explanatory variable data with the product containing the objective variable data for each explanatory variable, and the feature amount calculated by the calculation means. Based on the difference, the importance evaluation means for evaluating the degree of relevance of the explanatory variable to the defective factor as the importance and the output unit for outputting the evaluation result by the importance evaluation means are provided.

In this aspect, the setting means may be configured to set the two extraction regions based on a predetermined attribute value of the extraction region.

Further, in the above aspect, the setting means may be configured to set the two extraction regions so that the attribute values of the two extraction regions are the same.

Further, in the above aspect, the setting means is configured to reset the two extraction regions so as to be close to each other after the difference in the feature amount is calculated by the calculation means. When the two extraction regions are reset, the difference in the feature amount of the two reset regions is calculated for each explanatory variable, and the importance evaluation means is configured. The importance may be evaluated based on the difference between the plurality of feature quantities calculated by the calculation means.

Further, in the above aspect, the setting means determines an upper value which is the objective variable data separated from each other in the frequency distribution and a lower value smaller than the upper value, and the upper value and the frequency distribution are described. One of the extraction regions is set between the maximum value of the objective variable data, and the other extraction region is set between the lower value and the minimum value of the objective variable data in the frequency distribution. After the difference in the feature amount is calculated, the two extraction regions may be reset by changing the upper value and the lower value so as to be close to each other.

Further, in the above aspect, the setting means resets the two extraction regions without changing the predetermined attribute values of the two extraction regions after the difference in the feature amount is calculated by the calculation means. It may be configured in.

Further, in the above aspect, the attribute value may be the area width of the extraction region.

Further, in the above aspect, the attribute value may be the area of the extraction region.

Further, in the above aspect, the attribute value may be the number of products in which the objective variable is included in the extraction region.

Further, in the above aspect, the importance evaluation means calculates an approximate expression of the relationship between the difference in the feature amount and the position of the extraction region on the frequency distribution, and based on the calculated approximate expression, said. It may be configured to calculate the importance for each explanatory variable.

Further, in the above aspect, the importance evaluation means calculates the first-order approximation formula and the n-th order approximation formula (n is an integer of 2 or more) of the relationship, and is based on the calculated first-order approximation formula and nth-order approximation formula. Therefore, the importance may be calculated for each of the explanatory variables.

Further, in the above aspect, the defect factor analysis system further includes a determination means for determining a representative value of the objective variable data in the frequency distribution created by the frequency distribution creation means as a division point, and the setting means includes the setting means. The two extraction regions may be set so as to sandwich the division point determined by the determination means.

Further, in the above aspect, the setting means determines the two extraction regions when both of the two extraction regions do not overlap with the division point after the difference in the feature amount is calculated by the calculation means. Each is configured to be reset to approach the division point, and the importance evaluation means evaluates the importance when either of the two extraction regions overlaps with the division point. It may be configured as follows.

Further, in the defect factor analysis method of another aspect of the present invention, the objective variable data for each of a plurality of products for the objective variable used for determining whether the product is a good product or a defective product, and a plurality of data showing the characteristics of the product separately. For a data set containing the explanatory variable data for each of the plurality of products for each of the explanatory variables of the above, a step of creating a frequency distribution of the objective variable data and two extraction regions separated from each other in the created frequency distribution. And the difference between the explanatory variable data between the product in which the objective variable data is contained in one of the two set extraction regions and the product in which the objective variable data is contained in the other. A step of calculating the difference in the represented feature amount for each explanatory variable, a step of evaluating the degree of relevance of the explanatory variable to the defective factor as the importance based on the calculated difference in the feature amount, and an evaluation result are output. Has steps to do.

According to the defective factor analysis system and the defective factor analysis method according to the present invention, it is possible to present information that well represents the characteristics of defective factors even when the distributions of the non-defective product data and the defective product data overlap.

The schematic diagram which shows the structure of the defect factor analysis system which concerns on Embodiment 1. FIG. The block diagram which shows the structure of the defect factor analysis system which concerns on Embodiment 1. FIG. The flowchart which shows the procedure of the defect factor analysis processing by the defect factor analysis system which concerns on Embodiment 1. The figure which shows an example of a data set. A graph showing an example of a frequency distribution. The graph of the frequency distribution figure for demonstrating the setting of the extraction area by the defect factor analysis system which concerns on Embodiment 1. FIG. A flowchart showing the procedure of the importance evaluation process. A graph exemplifying the relationship between the difference in features and the position of the extraction area. The flowchart which shows the procedure of the defect factor analysis processing by the defect factor analysis system which concerns on Embodiment 2. The graph of the frequency distribution map for demonstrating the setting of the extraction area by the defect factor analysis system which concerns on Embodiment 2. FIG. The graph which shows the relationship between the difference of the feature amount calculated in an Example, and the position of an extraction area.

Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. It should be noted that each of the embodiments shown below exemplifies a method and an apparatus for embodying the technical idea of the present invention, and the technical idea of the present invention is not limited to the following. Absent. The technical idea of the present invention can be modified in various ways within the technical scope described in the claims.

(Embodiment 1)
In the present embodiment, the defect factor analysis system uses the objective variable data for a data set including each product data (objective variable data) for the objective variable and each product data (explanatory variable data) for the explanatory variables. Create a frequency distribution, set two separate extraction regions in the frequency distribution, and between a product that contains objective variable data in one of these extraction regions and a product that contains objective variable data in the other. The difference in the explanatory variable data in is calculated as the difference in the feature amount. The defect factor analysis system resets the extraction areas so that they are close to each other, calculates the difference in the feature amount each time, and selects the explanatory variable representing the characteristic of the defect factor as the important variable based on the difference in each feature amount. Display this.

<Configuration of defect factor analysis system>
FIG. 1 is a schematic diagram showing a configuration of a defect factor analysis system according to the present embodiment. The defective factor analysis system 10 is for analyzing the defective factors of the product. The defect factor analysis system 10 is given as input a data set including target variable data which is data for each product about the target variable and explanatory variable data which is data for each product about the explanatory variable. The target variable is a variable used to determine whether the product 20 is a non-defective product or a defective product, and the explanatory variable is a variable indicating the characteristics of the product. For example, when the rolled steel sheet is the product 20 to be analyzed, test values such as hardness can be used as target variables, and temperature, pressure, material components, and the like in the manufacturing equipment (rolling equipment) 21 can be used as explanatory variables. In the example of FIG. 1, a plurality of sensors 22 such as a temperature sensor and a pressure sensor are installed in the manufacturing facility 21, and explanatory variable data of each product 20 is collected by these plurality of sensors. Further, a test such as a hardness test is performed on the product 20 manufactured by the manufacturing facility 21 by the test device 23, and the test value is collected as objective variable data.

FIG. 2 is a block diagram showing a configuration of the defect factor analysis system 10 according to the present embodiment. The defect factor analysis system 10 is realized by the computer 11. The computer 11 includes a main body 110, an input unit 120, and a display unit 130. The main body 110 includes a CPU 111, a ROM 112, a RAM 113, a hard disk 114, an input / output interface 115, and a video output interface 116. The CPU 111, ROM 112, RAM 113, hard disk 114, input / output interface 115, and video output interface 116 are connected to each other by a bus.

When the CPU 111 executes the defect factor analysis program 150, which is a computer program for defect factor analysis, the computer 11 functions as the defect factor analysis system 10.

The ROM 112 records a computer program executed by the CPU 111, data used for the program, and the like. The RAM 113 is used to read out the defect factor analysis program 150 recorded on the hard disk 114. Further, the RAM 113 is used as a work area of the CPU 111 when the CPU 111 executes a computer program.

The hard disk 114 is installed with various computer programs such as an operating system and an application program for the CPU 111 to execute, and data used for executing the computer programs. The defect factor analysis program 150 is also installed on the hard disk 114.

An input unit 120 including a keyboard and a mouse is connected to the input / output interface 115. Further, the sensor 22 and the test device 23 are connected to the input / output interface 115. The data output from the sensor 22 and the test device 23 is given to the CPU 111 via the input / output interface 115.

The video output interface 116 is connected to a display unit 130 composed of an LCD, a CRT, or the like, and outputs a video signal corresponding to the video data given by the CPU 111 to the display unit 130. The display unit 130 displays an image according to the input video signal.

<Operation of defect factor analysis system>
Next, the operation of the defect factor analysis system 10 will be described. FIG. 3 is a flowchart showing a procedure of the defect factor analysis process by the defect factor analysis system 10 according to the present embodiment. While the manufacturing equipment 21 is in operation, the sensor 22 collects explanatory variable data, and the test apparatus 23 collects objective variable data. A data set including these explanatory variable data and objective variable data is input to the defect factor analysis system 10. Alternatively, the operator may collect the explanatory variable data measured by the sensor 22 and the objective variable data measured by the test apparatus 23, and manually input these data sets from the input unit 120 into the defect factor analysis system 10. it can. The CPU 111 of the defect factor analysis system 10 receives the data set input in this way (step S11).

FIG. 4 is a diagram showing an example of a data set. In the example shown in FIG. 4, a product ID is assigned to each product, and the objective variable data and the explanatory variable data of the product are associated with each product ID. As in this example, in the data set, the objective variable data for the same product and each explanatory variable data are associated with each other.

The CPU 111 creates a frequency distribution (histogram) of the objective variable data for the input data set (step S12). FIG. 5 is a graph showing an example of the created frequency distribution. In FIG. 5, the vertical axis represents the frequency and the horizontal axis represents the objective variable. The objective variable is used as one of the items for determining whether the product is non-defective or defective. Therefore, in the frequency distribution, non-defective products and defective products are unevenly distributed depending on the size of the objective variable value (objective variable data). In the example shown in FIG. 5, many defective products are included in the region where the objective variable data is large, and many good products are included in the region where the objective variable data is small. In addition, items other than the objective variable (for example, component values of the product) may be used to determine whether the product is non-defective or defective. In this case, since a good product or a defective product cannot be determined only by the objective variable, the good product and the defective product may be distributed in an overlapping manner in the frequency distribution. In the example shown in FIG. 5, defective products are unevenly distributed in the region where the objective variable data is large, non-defective products are unevenly distributed in the region where the objective variable data is small, and the distribution of non-defective products and defective products partially overlaps. There is. In the frequency distribution shown in FIG. 5, it can be said that the product containing the objective variable data in the maximum region is an extremely defective product, and the product contained in the minimum region is an extremely non-defective product.

See FIG. 3 again. The CPU 111 determines a representative value of the objective variable data in the frequency distribution as a division point (step S13), and sets two extraction regions separated from each other so as to sandwich the division point (step S14). FIG. 6 is a graph of a frequency distribution map for explaining the setting of the extraction region. As shown in FIG. 6, the frequency distribution often shows a normal distribution or a distribution close thereto. That is, many products are distributed in the middle part of the objective variable data, and the number of products decreases as they approach both ends. Representative values of the objective variable data in such a frequency distribution, for example, the mode value, the average value, and the median value are set as the division points μ. Such a representative value is objective variable data in which many products exist, and it can be said that the region of non-defective products and the region of defective products are well separated. Further, in step S14, the CPU 111 _{determines the upper value V U , which is the objective variable data larger than the division point μ, and determines the lower value V L} , which is the objective variable data smaller than the division point μ. Further, the CPU 111 sets _{a region between the upper value V U and the maximum value V max} of the objective variable data in the frequency distribution as an extraction region (hereinafter, referred to as “first extraction region”) 31 in which the objective variable data is larger than the division point μ. _{The area between the lower value VL and the minimum value V min} of the objective variable data in the frequency distribution is set to the extraction area where the objective variable data is smaller than the division point μ (hereinafter referred to as “second extraction area”). Set to 32. More specifically, CPU 111 determines the positive number a that sets _{V U} or more regions of the mu + a as _{V U} in the first extraction region 31, a _{V L} following regions mu-a as _{V L} a 2 Set in the extraction area 32. The first extraction region set in this way is an region to which an extremely defective product belongs, and the second extraction region is an region to which an extremely non-defective product belongs.

See FIG. 3 again. The CPU 111 extracts the product whose objective variable data is contained in the first extraction area 31 and the product whose objective variable data is contained in the second extraction area 32 from the data set, and extracts the first and second extraction areas 31. , The difference in the average value of the explanatory variable data of the products of 32 products is calculated for each explanatory variable as the difference in the feature amount (step S15). For example, the difference between the average value of the temperature of the product containing the objective variable data in the first extraction region 31 and the average value of the temperature of the product containing the objective variable data in the second extraction region 32 is the difference in the feature amount. It is calculated. Similarly, the difference in features is calculated for other explanatory variables such as pressure and material components.

The CPU 111 determines whether or not any of the first or second extraction regions 31 and 32 overlaps the division point μ (step S16). When both the first or second extraction areas 31 and 32 do not overlap the division point μ (NO in step S16), the CPU 111 returns the process to step S14 and returns the processing to the first and second extraction areas 31 and 32, respectively. Are reset so as to be close to each other, that is, close to the division point μ (step S14). More specifically, the CPU 111 updates a so as to make it smaller by a predetermined value, resets the region above V _U to the first extraction region 31 with _{μ + a as a new upper value V U, and sets μ −a. The region below VL} is reset to the second extraction region 32 as a new lower value _VL. The CPU 111 calculates the difference in the feature amount for the reset first and second extraction areas 31 and 32 (step S15), and either of the first or second extraction areas 31 and 32 overlaps the division point μ. Whether or not it is determined (step S16). In this way, the CPU 111 repeatedly executes the processes of steps S14 to S16, and calculates a plurality of differences in the feature amount for each explanatory variable.

Next, the CPU 111 executes an importance evaluation process for evaluating the importance of the defect factor for each explanatory variable (step S17). Hereinafter, the importance evaluation process will be described in detail. FIG. 7 is a flowchart showing the procedure of the importance evaluation process. The importance is an index showing how the difference in the feature amount changes according to the change in the extraction region, and shows the degree of relevance of the explanatory variable to the defective factor.

In the importance evaluation process, the CPU 111 first creates a relationship between the difference in the feature amount and the position of the extraction region for each explanatory variable (step S171). FIG. 8 is a graph illustrating the relationship between the difference in the feature amount and the position of the extraction region. In FIG. 8, the vertical axis shows the difference in the feature amount, and the horizontal axis shows the position of the extraction region. The positions of the extraction regions are the positions of the first and second extraction regions 31 and 32, and on the horizontal axis in FIG. 8, the intersection with the vertical axis is the frequency of the first and second extraction regions 31 and 32. It indicates that it is located at both ends of the distribution (that is, it corresponds to the initial value of a), and the first and second extraction regions 31 and 32 approach the division point μ (that is, a is) toward the right. To decrease) is shown. In the process of step S171, specifically, the CPU 111 creates a data set of the position of the extraction area and the difference in the feature amount corresponding thereto for each explanatory variable, or the difference in the feature amount in the order of the position of the extraction area. By creating a data set in which the above are arranged, the relationship between the difference in features and the position of the extraction area is created.

In FIG. 8, the solid line graph shows the relationship between the difference in the feature amount for one explanatory variable and the position of the extraction region, and the broken line graph shows the difference in the feature amount for another explanatory variable. The relationship with the position of the extraction area is shown. In the solid line graph, when the first and second extraction regions 31 and 32 are located at both ends of the frequency distribution, the difference in the feature amount is large, and as the first and second extraction regions 31 and 32 approach the division point μ, The difference in features is decreasing. In other words, in this graph, there is a large difference in the feature amount for this explanatory variable between one end side of the frequency distribution containing extremely non-defective products and the other end side of the frequency distribution containing extremely defective products, and generally only non-defective products are included. It is shown that the difference in the feature amount is small in the vicinity of the dividing point μ, which includes defective products that are not good or are close to non-defective products. Therefore, it can be said that this explanatory variable is strongly related to the defective factor. On the other hand, in the broken line graph, the difference in the feature amount changes regardless of the change in the extraction region. This indicates that this explanatory variable is less relevant to the bad factors.

See FIG. 7 again. Next, the CPU 111 calculates a first-order approximation formula and an nth-order approximation formula (n is an integer of 2 or more) regarding the relationship between the difference in the created features and the position of the extraction region (step S172), and is important according to the following formula. The degree is calculated (step S173).
(Importance) = λ × (absolute value of slope of first-order approximation formula)-(total error between nth-order approximation formula and data)
However, λ in the above equation is a constant. The larger the change in the feature amount with respect to the position of the extraction region, the larger the absolute value of the slope of the first-order approximation formula. Therefore, the absolute value of the slope of the first-order approximation formula represents the strength of the relationship with the defect factor of the explanatory variable. Further, as shown in the graph shown by the solid line in FIG. 8, in the explanatory variable in which the difference in the feature amount changes smoothly according to the position of the extraction region, the relationship between the difference in the feature amount and the position of the extraction region is determined by the nth-order approximation formula. Can be approximated well, so that the error between the nth-order approximation formula and the data becomes small. On the other hand, as shown in the graph shown by the broken line in FIG. 8, it is difficult to approximate by the n-th order approximation formula with the explanatory variable in which the difference in the feature amount changes regardless of the position of the extraction region. The error with the data becomes large. Therefore, the error between the n-th order approximation formula and the data becomes large for the explanatory variables that change regardless of the position of the extraction region. From the above, the above-mentioned importance becomes larger as the explanatory variable is more strongly related to the defect factor, and becomes smaller as the explanatory variable changes regardless of the position of the extraction region.

Next, the CPU 111 compares the importance calculated as described above, selects an explanatory variable whose importance is equal to or higher than a predetermined reference value as an important variable, and stores it in the hard disk 114 (step S174). This completes the importance evaluation process.

See FIG. 3 again. The CPU 111 causes the display unit 130 to display the evaluation result of the importance evaluation process (step S18). On this display screen, the important variables selected in step S174 are displayed. In addition, when there are a plurality of important variables, the important variables are displayed by arranging them in the order of importance or ranking the importance. As a result, the operator can identify which explanatory variable is the important variable and which of the important variables is of high importance by referring to the display screen of the display unit 130. When the evaluation result of the importance is output, the CPU 111 ends the defect factor analysis process.

(Embodiment 2)
In the present embodiment, the defect factor analysis system 10 repeatedly sets the widths of the first and second extraction regions so as to be close to each other while keeping the widths constant.

<Configuration of defect factor analysis system>
Since the configuration of the defect factor analysis system according to the present embodiment is the same as the configuration of the defect factor analysis system 10 according to the first embodiment, the same components are designated by the same reference numerals and the description thereof will be omitted.

<Operation of defect factor analysis system>
The operation of the defect factor analysis system 10 according to the present embodiment will be described. FIG. 9 is a flowchart showing a procedure of the defect factor analysis process by the defect factor analysis system 10 according to the present embodiment. Since the processes of steps S11 to S13 are the same as those described in the first embodiment, the description thereof will be omitted.

When the division point is determined, the CPU 111 sets two extraction areas having a preset area width so as to sandwich the division point (step S24). FIG. 10 is a graph of a frequency distribution diagram for explaining the setting of the extraction region. _{In step S24, the CPU 111 determines the upper value V U} , which is the objective variable data larger than _{the division point μ, and determines the lower value V L} , which is the objective variable data smaller than the division point μ. Further, the CPU 111 sets an _{area in which the objective variable data is larger by a predetermined area width d from the upper value V U} to the first extraction area 231 in which the objective variable data is larger than the division point μ, and the area width d from the _{lower value VL.} The region where the objective variable data is small is set in the second extraction region 232 where the objective variable data is smaller than the division point μ. More specifically, the CPU 111 determines a certain positive number a, sets a region of μ + a or more and μ + a + d or less in the first extraction region 231 and sets a region of μ−a or less and μ−ad or more in the second extraction region 232. Set to.

See FIG. 9 again. The CPU 111 extracts the product whose objective variable data is contained in the first extraction area 31 and the product whose objective variable data is contained in the second extraction area 32 from the data set, and extracts the first and second extraction areas 31. , The difference in the average value of the explanatory variable data of the products of 32 products is calculated for each explanatory variable as the difference in the feature amount (step S15). Next, the CPU 111 determines whether or not any of the first or second extraction regions 231,232 overlaps the division point μ (step S16). When both the first or second extraction areas 231 and 232 do not overlap the division point μ (NO in step S16), the CPU 111 returns the process to step S24 and returns the processing to the first and second extraction areas 31 and 32, respectively. Are reset so as to be close to each other, that is, close to the division point μ (step S24). More specifically, the CPU 111 updates a so as to make it smaller by a predetermined value, sets μ + a as a new upper value V _U , and sets a region in which the objective variable data is larger than the _{upper value V U by the} same region width d as the previous time. Is reset to the first extraction region 231, and the region where the objective variable data is smaller by the region width d from _{VL is reset to the second extraction region 232 with μ−a as the new lower value VL.} The CPU 111 calculates the difference in the feature amount for the reset first and second extraction areas 31 and 32 (step S15), and either of the first or second extraction areas 31 and 32 overlaps the division point μ. Whether or not it is determined (step S16). In this way, the CPU 111 repeatedly executes the processes of steps S24, S15, and S16, and calculates a plurality of differences in the feature amount for each explanatory variable.

Since the processes of steps S17 to S18 are the same as those described in the first embodiment, the description thereof will be omitted.

With the above configuration, the region widths of the two extraction regions are fixed and the extraction regions are reset so as to approach each other. Therefore, the extraction area is not expanded every time the setting is made, but a division point in which only good products are included or close to good products are included from both sides of the frequency distribution in which extremely defective products and extremely non-defective products are included. The extraction area is sequentially moved to the vicinity of μ with the same area width. Therefore, the difference in the feature amount more strongly reflects the defect factor as compared with the first embodiment, and the importance calculated by using the difference in the feature amount is the degree of relevance of the explanatory variable to the defect factor. Is more prominently represented.

<Example>
The defect factor analysis method according to the first embodiment was applied to a certain product by using a certain test value as an objective variable and five elements such as a measured value as explanatory variables (variables 1 to 5). FIG. 11 is a graph showing the relationship between the difference in the feature amount calculated in this embodiment and the position of the extraction region. In FIG. 11, the vertical axis shows the difference in the feature amount, and the horizontal axis shows the position of the extraction region (specifically, the value of a). In this example, the division point μ was 1055.202. As shown in FIG. 11, the explanatory variable having the largest difference in features is variable 4. In the variable 4, the difference in the feature amount is larger as the extraction regions are farther from each other, and the difference in the feature amount is smaller as the extraction regions are closer to each other (closer to the division point μ). Further, in the variable 4, the change in the difference in the feature amount is generally smooth. The difference in the feature amount of the variable 3 is the second largest after the variable 4, and the difference in the feature amount decreases as the position of the extraction region approaches the division point μ, although the difference is a little. In variable 1, the change in the feature amount is smooth, but the change amount is small. In the variable 2, the difference in the feature amount hardly changes over the entire position of the extraction region. In the variable 5, the difference in the feature amount slightly increases according to the change in the position of the extraction region, and the variation is large.

The following table shows the results of calculating the importance of each of the above variables 1 to 5 with λ = 1.5 and n = 2. In this table, the explanatory variables are arranged in descending order for importance.

The farther the extraction regions are from each other, the larger the difference in the features, and the closer they are to each other, the smaller the difference in the features. That is, the variable 4 having a large absolute value of the slope of the linear approximation formula has the highest importance. Further, the variable 5 having a large variation in the difference in the feature amount has a low importance because the error between the quadratic approximation formula and the data is large. The explanatory variable with the least importance is variable 2, which is considered to be because the absolute value of the slope of the linear approximation formula is small.

(Other embodiments)
In the first embodiment described above, the upper value V _{U larger than the dividing point μ and the lower value VL} smaller than the dividing point μ are determined in the frequency distribution, and the upper value V _U and the maximum value V of the objective variable data are defined. _{A configuration} is described in which the first extraction region 31 is set between _{max and the second extraction region 32 is set between the lower value VL} and the minimum value V _min of the objective variable data. In the second embodiment, the upper side is set. The configuration in which the area in which the objective variable data is large by the area width d from the _{value V U} is set as the first extraction area 231 and the area in which the objective variable data is small by the area width d from the _{lower value VL is set as the second extraction area 232 will be described.} It was. However, the frequency distribution is not limited to the above as long as it has a configuration in which two extraction regions separated from each other are set. For example, in the frequency distribution, an upper value V _{u larger than the division point μ and a lower value V l} smaller than the division point μ by b different from a are determined, and the upper value V _u and the maximum value V _{max of the objective variable data are defined.} It is also possible to set between the lower value V _l and the minimum value V _min of the objective variable data as the second extraction area, with the area between the upper value V l as the first extraction area, or the area width d1 from _{the upper value V l.} It is also possible to set the region where the variable data is large as the first extraction region and the region where the objective variable data is small by the region width d2 different from the _{lower value V l to d1 as the second extraction region.} Further, based on the attribute values of the first and second extraction regions (region width of the extraction region, area of the extraction region, number of products containing the objective variable in the extraction region, etc.), the first position is separated from each other in the frequency distribution. And the second extraction area may be set. For example, the first and second extraction regions may be set so that the areas of the first and second extraction regions are the same, and the number of products in which the objective variable is included in each of the first and second extraction regions is The first and second extraction regions may be set so as to be the same.

Further, in the first embodiment described above, the upper value V _U and the lower value V _L are updated so as to be close to each other, and the new upper value V _U and the maximum value V _max of the objective variable data are set between the new upper value V U and the maximum value V max of the objective variable data. 1 The configuration of resetting as the extraction area 31 and resetting between the new lower value _VL and the minimum value _Vmin of the objective variable data as the second extraction area 32 will be described, and in the second embodiment, the area The configuration for resetting the first and second extraction regions 231,232 so as to be close to each other without changing the width d has been described. However, the configuration is not limited to the above as long as the two extraction regions are reset so as to be close to each other. As a configuration for resetting the first and second extraction areas based on attribute values (area of the extraction area, number of products whose objective variable is included in the extraction area, etc.) different from the area widths of the first and second extraction areas. May be good. For example, the areas of the first and second extraction regions can be reset so as to be close to each other without changing the area, or the number of products contained in each of the first and second extraction regions cannot be changed. Can also be reconfigured to be closer to each other. If the number of products containing the objective variable in each of the first and second extraction regions at the same time point is the same, the products containing the objective variable in each of the first and second extraction regions before and after the resetting. The numbers may be different. That is, the number of products in which the objective variable is included in each of the first and second extraction regions set in the first time is P1, and the objective variable is included in each of the first and second extraction regions set in the second time. The number of products to be specified is P2, the number of products in which the objective variable is included in each of the first and second extraction regions set in the third time is P3, ..., The number of products in the first and second extraction regions set in the nth time is The number of products in which the objective variable is included in each may be Pn. Here, P1, P2, P3, ..., Pn are arbitrary natural numbers, and these may be different from each other.

Further, in the above-described first and second embodiments, the configuration in which the difference in the average value of the explanatory variable data of the products included in the first and second extraction regions is calculated as the difference in the feature amount has been described. Not limited. The difference between the representative values (for example, the median value) of the explanatory variable data of the products included in the first and second extraction regions can be used as the difference in the feature amount.

Further, in the above-described first and second embodiments, the configuration in which all the processes of the defect factor analysis program 150 are executed by a single computer 11 has been described, but the present invention is not limited to this and is defective. It is also possible to make a distributed system in which the same processing as that of the factor analysis program 150 is distributed and executed by a plurality of devices (computers).

The defective factor analysis system and the defective factor analysis method of the present invention are useful as a defective factor analysis system and a defective factor analysis method for analyzing defective factors of a product.

10 Defective factor analysis system 11 Computer 110 Main unit 120 Input unit 130 Display unit 111 CPU
112 ROM
113 RAM
114 Hard disk 115 Input / output interface 116 Video output interface 150 Defect cause analysis program 20 Products 21 Manufacturing equipment 22 Sensors 23 Test equipment 31,231 1st extraction area 32,232 2nd extraction area

Claims (13)

A plurality of product-specific objective variable data for the objective variable used to determine whether the product is good or defective, and the plurality of product-specific explanatory variables for each of the plurality of explanatory variables that separately indicate the characteristics of the product. For a data set including data, a frequency distribution creating means for creating a frequency distribution of the objective variable data, and
A setting means for setting two extraction regions separated from each other in the frequency distribution created by the frequency distribution creating means, and a setting means.
A feature representing the difference between the explanatory variable data between a product containing the objective variable data in one of the two extraction regions set by the setting means and a product containing the objective variable data in the other. A calculation means for calculating the difference in quantity for each of the explanatory variables, and
Based on the difference in the feature amount calculated by the calculation means, the importance evaluation means for evaluating the degree of relevance of the explanatory variable to the defective factor as the importance, and the importance evaluation means.
It is equipped with an output unit that outputs the evaluation result by the importance evaluation means .
The setting means is configured to reset the two extraction regions so that they are close to each other after the difference in the feature amount is calculated by the calculation means.
The calculation means is configured to calculate, for each of the explanatory variables, the difference in the feature amount of the two reset extraction regions when the two extraction regions are reset.
The importance evaluation means is configured to evaluate the importance based on the difference between the plurality of feature quantities calculated by the calculation means.
Defective factor analysis system. The setting means is configured to set the two extraction regions based on a predetermined attribute value of the extraction region.
The defect factor analysis system according to claim 1. The setting means is configured to set the two extraction regions so that the attribute values of the two extraction regions are the same.
The defective factor analysis system according to claim 2. The setting means determines an upper value which is the objective variable data separated from each other in the frequency distribution and a lower value smaller than the upper value, and the maximum value of the objective variable data in the upper value and the frequency distribution. The area between the two is set as one of the extraction areas, and the area between the lower value and the minimum value of the objective variable data in the frequency distribution is set as the other extraction area. After the calculation, the two extraction regions are configured to be reset by changing the upper value and the lower value so as to be close to each other.
The defect factor analysis system according to any one of claims 1 to 3. The setting means is configured to reset the two extraction regions without changing predetermined attribute values of the two extraction regions after the difference in the feature amount is calculated by the calculation means.
The defect factor analysis system according to any one of claims 1 to 4. The attribute value is the area width of the extraction area.
The defect factor analysis system according to claim 2, 3 or 5. The attribute value is the area of the extraction area.
The defect factor analysis system according to claim 2, 3 or 5. The attribute value is the number of products in which the objective variable is included in the extraction region.
The defect factor analysis system according to claim 2, 3 or 5. The importance evaluation means calculates an approximate expression of the relationship between the difference in the feature amount and the position of the extraction region on the frequency distribution, and based on the calculated approximate expression, the importantness is described for each of the explanatory variables. It is configured to calculate the degree,
The defect factor analysis system according to any one of claims 1 to 8. The importance evaluation means calculates the first-order approximation formula and the n-th order approximation formula (n is an integer of 2 or more) of the relationship, and based on the calculated first-order approximation formula and the nth-order approximation formula, for each of the explanatory variables. Is configured to calculate the importance of
The defective factor analysis system according to claim 9. Further provided with a determination means for determining the representative value of the objective variable data in the frequency distribution created by the frequency distribution creation means as a division point.
The setting means is configured to set the two extraction regions so as to sandwich the division point determined by the determination means.
The defect factor analysis system according to any one of claims 1 to 10. After the difference in the feature amount is calculated by the calculation means, the setting means sets each of the two extraction regions as the division point when both of the two extraction regions do not overlap with the division point. It is configured to be reconfigured to be closer
The importance evaluation means is configured to evaluate the importance when any of the two extraction regions overlaps with the division point.
The defect factor analysis system according to claim 11. A plurality of product-specific objective variable data for the objective variable used to determine whether the product is good or defective, and the plurality of product-specific explanatory variables for each of the plurality of explanatory variables that separately indicate the characteristics of the product. For the data set including the data, the step of creating the frequency distribution of the objective variable data and
A step of setting two extraction regions separated from each other in the created frequency distribution, and
The difference in the feature amount representing the difference in the explanatory variable data between the product in which the objective variable data is contained in one of the two set extraction regions and the product in which the objective variable data is contained in the other. , Steps to calculate for each explanatory variable,
Based on the calculated difference in the feature amount, the step of evaluating the degree of relevance of the explanatory variable to the defective factor as the importance, and
Possess and outputting the evaluation result,
In the step of setting, after the difference in the feature amount is calculated in the step of calculating, the two extraction regions are reset so as to be close to each other.
When the two extraction regions are reset in the calculation step, the difference in the feature amounts of the reset two extraction regions is calculated for each of the explanatory variables.
In the evaluation step, the importance is evaluated based on the difference between the plurality of feature quantities calculated in the calculation step.
Defective factor analysis method.

Images