KR102247945B1 - Method of predicting processing fault considering process factor - Google Patents

Abstract

It predicts process defects in real time using a standard algorithm using ICT, qualitatively assigns relative importance to process factors, and presents a process defect prediction method that considers process factors that take into account the center position and distribution of the measured values. do. The method is to predict process defects, in a process defect prediction platform that uses IoT and big data, extracts independent variables from a plurality of process factors, sets dependent variables, and determines the relative importance of process factors from the independent variables. By applying the qualitatively assigned AHP technique, the main process factor is extracted, the correlation between the major process factors is verified, and the process factor whose correlation is verified is applied to the machine learning regression analysis algorithm to calculate the process defect prediction model equation. The process defects derived from the predictive model equation are analyzed by applying the ANOVA technique in consideration of the center position and distribution of the measured values to analyze the minimum significance level of the process defects, and the process management specification of the process defects after the analysis of the significance level is determined. Operate.

Description

Method of predicting processing fault considering process factor}

The present invention relates to a process defect prediction method, and more particularly, to a process defect prediction method that infers the cause of the defect by analyzing the correlation between the process factor affecting the process and the process defect rate in real time and predicts the trend in real time. About.

Recently, ICT (Information and Communications Technologies) is the core of the 4th Industrial Revolution, and its potential for development is infinite as it enables not only the connection between humans and humans, but also the connection between humans and objects, and between objects and objects. Wireless communication, which is one of the core of the above ICT, is making it possible to connect and communicate anytime, anywhere beyond the limits of time and space. In recent years, by utilizing ICT, efforts of process management to efficiently solve various problems occurring in the process and to achieve improvement in productivity and quality are continuing. Conventional process management has been conducted in such a way that when the trend of process defects rises, an abnormal state of a subsequent process is checked to estimate a factor of increasing process defects. Since such process management is a follow-up measure, measures are taken after a process defect occurs, resulting in inefficient and economical loss of work and facility operation.

Meanwhile, Korean Patent Publication No. 2013-0125527 proposes a method of optimizing digital manufacturing using simulation and process data. However, the above patent is a workaround because the correlation between the process factor and the process defect cannot be confirmed, and only follow-up measures are taken. Accordingly, for process management, a process management platform capable of predicting process defects in real time using a standard algorithm using ICT is required. In addition, the conventional process management does not qualitatively assign the relative importance to the process factors, and the consideration of the center position and distribution of the measured values is insufficient.

The problem to be solved by the present invention is to predict process defects in real time using a standard algorithm using ICT, qualitatively assign relative importance to process factors, and consider the center position and distribution of the measured values. It is to provide a method for predicting process defects taking into account.

The process defect prediction method considering process factors to solve the problems of the present invention is a process defect prediction platform that utilizes IoT and big data to predict process defects. First, independent variables are extracted from a plurality of process factors and dependent Set the variable. Then, the main process factors are extracted by applying the AHP technique that qualitatively assigns the relative importance to the process factors from the independent variable. The correlation between the major process factors is verified. The process factor for which the correlation is verified is applied to a machine learning regression analysis algorithm to derive a process defect prediction model equation. The minimum significance level of the process defect is analyzed by applying the ANOVA technique in consideration of the center position and scatter of the measured values for the process defect derived from the predictive model equation. After the analysis of the significance level is completed, the process management specification of the process defect is determined and operated.

In the method of the present invention, the plurality of process factors may be process factors in an independent input process, a production process, and a completion process, respectively. The correlation analysis code of the correlation verification is

# pip install seaborn <<- in virtualenv !!

import seaborn as sns

fig, ax = plt.subplots(figsize=(10,10))

It may be sns.heatmap(df.corr(), annot=True, linewidth=.5, ax=ax).

In a preferred method of the present invention, before deriving the process defect prediction model equation, the step of verifying the error and optimizing the weight of the main process factor may be further included. The verification of the error may be performed by analyzing the amount of deviation between the predicted value of the regression analysis and the measured value. In the ANOVA technique, if the significance difference index P-value is less than or equal to the significance level, it is determined as an influencing factor of the process and the specification for this is adjusted to a good level. As a result of the process defect determination by the ANOVA method, if the process defect is NG, it is converted to 1 of the ONE-HOT code, and if the process defect is OK, it can be converted to 1 of the ONE-HOT code.

According to the process defect prediction method considering process factors of the present invention, by using AHP and ANOVA techniques as a machine running technique to predict process defects, by using a standard algorithm using ICT to predict process defects in real time, and The relative importance of the factors is given qualitatively, and it is predicted by considering the center position and scatter of the measured values.

1 is a flowchart showing a method for predicting process defects in consideration of process factors according to the present invention.
2 is a diagram for explaining verification of a correlation applied to a method for predicting process defects according to the present invention.
3 is a graph for explaining weight analysis applied to the method for predicting process defects according to the present invention.
4 is a graph for explaining the ANOVA technique applied to the process defect prediction method according to the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. The embodiments described below may be modified in various forms, and the scope of the present invention is not limited to the embodiments described below. Embodiments of the present invention are provided to more completely describe the present invention to those of ordinary skill in the art.

An embodiment of the present invention predicts process defects using AHP and ANOVA techniques as a machine running technique, predicts process defects in real time using a standard algorithm using ICT, and determines the relative importance of process factors. It is given as a grade, and a method for predicting process defects in consideration of the center position and distribution of the measured values is presented. To this end, a standard algorithm for predicting process defects will be described in detail, and a method of increasing the accuracy of prediction by adding AHP and ANOVA techniques will be described in detail. The embodiment of the present invention is a process management that predicts process defects in real time by a standard algorithm, and is distinguished from the conventional process management, which is limited to follow-up measures.

1 is a flowchart showing a method for predicting process defects in consideration of process factors according to an embodiment of the present invention. However, the drawings are not expressed in a strict sense, and there may be components not shown in the drawings for convenience of description.

Referring to FIG. 1, in the method for predicting process defects, first, data measured from process factors of each process are transmitted by IoT and stored in a platform (S10). Then, the independent variable is extracted from the process factor (S11), and the dependent variable is set (S12). Table 1 is for explaining the relationship between the process factor and the defective rate applied to the prediction method according to an embodiment of the present invention.

Input process production process … production process Completion process Process factor 1 Process factor 2 … Process factor (n-1) Process factor n Defective rate 1 Defective rate 2 … Defective rate (m-1) Defective rate m Additional factor 1 Additional factor 2 … Additional factor (k-1) Additional factor k

According to Table 1, each of the processes is an independent process such as an input process for manufacturing a product, a plurality of production processes, and a completion process. In order to manufacture a product, a product is produced through a plurality of production processes after an input process and an input process in which raw materials and parts are input, and finally the product is manufactured through a completion process. Each process includes process factors that affect process failure. Such a process factor includes a real-time defect rate predicted in real time by the prediction method according to an embodiment of the present invention, and an additional factor affecting the process defect may be added in the process of predicting the defect rate in real time. Additional factors are factors that affect process defects in addition to the process factors.

Process factors may include, for example, current, voltage, pneumatic pressure, temperature, humidity, static electricity, illuminance, and amount of dust. An additional factor can be, for example, humidity. Process factor 1 in the input process, process factor 2 in the production process,… , A process factor (n-1) and a process factor n may be present in the completed process. The process factors in each of the process factors 1 to n may differ slightly from the process factors presented above, but include at least one process factor. Process factors 1, 2 to (n-1) and n in the input process, production process, and completion process may be different, respectively. For example, the process factor 1 of the input process is current, voltage, pneumatic pressure, temperature, and static electricity, and the process factors 2 to (n-1) of the production process exclude current and voltage, and the amount of dust may be added.

The defective rate is 1 defective rate in the input process, 2 defective rate in the production process,… , Defect rate (m-1) and process factor m may be present in the completed process. Likewise, the additional factor is an additional factor 1 in the input process, an additional factor 2 in the production process,… , Additional factor (k-1) and process factor k may be present in the completion process. Each process may not include an additional factor or may include at least one or more. Additional factors may include, for example, the amount of sleep the worker has, the worker's body temperature, the worker's career, and the worker's vision.

Next, the main process factors are extracted from the process factors using the AHP technique (S13). The analytic hierarchy process (AHP) technique is a multi-criteria decision making model developed by Saaty, and is for systematically analyzing the decision making process. It is a model to derive reasonable results by performing pairwise comparison for all evaluation items and calculating the relative importance of the evaluation items. In particular, AHP interprets qualitative problems in a quantitative way, supports systematic decision-making, organizes complex and ambiguous problems into several layers, and separates partial relationships through pairwise comparison. Assessing the importance of qualities on a numerical scale helps them make more accurate decisions.

When the extraction of the main process factors is finished, the correlation is verified (S14). The correlation degree is verified as shown in FIG. 2. According to FIG. 2, the correlation verification chart utilizes a chart of correlation coefficients showing the correlation between process factors. On the right, the correlation value is expressed by indicating the range of the correlation coefficient in the range of 1 to -1 as the color saturation. And X-axis and Y-axis list the items of the dependent variable (Y) and the independent variable (X1~X8), respectively, and express the degree of correlation between each process factor, so that the correlation coefficient of the dependent variable (Y) and the independent variable (X1~X8). Can be seen. Also, the correlation coefficient between the independent variables (X1~X8) can be known. Through this, it is possible to grasp the relationship and the degree of relationship between the normal and negative correlations between process factors.

At this time, the correlation analysis code is as follows.

# pip install seaborn <<- in virtualenv !!

import seaborn as sns

fig, ax = plt.subplots(figsize=(10,10))

sns.heatmap(df.corr(), annot=True, linewidth=.5, ax=ax)

When the correlation is verified, a machine learning regression analysis algorithm is applied (S15), and a process defect prediction model equation is derived through it (S18). Regression analysis is one of the machine learning algorithms based on artificial intelligence. The regression analysis is a field of speculative statistics that analyzes the relationship between two or more process factors, especially the correlation between process factors. In regression analysis, the correlation is inferred by grasping the mathematical linear functional expression of the change in the measured value of a specific process factor and the change in the measured value of another process factor. The estimated functional expression is called a regression formula. Through this regression equation, if the change in the measured value of a specific process factor (referred to as an independent variable) is related to the change in the measured value of another process factor (referred to as a dependent variable), the change in the measured value of a certain process factor is the cause. It is possible to analyze matters such as which process factor actual measured value change is the resultant phenomenon.

This regression analysis differs from the correlation analysis, which examines only the closeness of the relationship between the measured values of process factors, not the causal relationship. Regression analysis is based on the functional expression between two variables that have a causal relationship, namely the independent variable and the dependent variable. Regression analysis is a useful tool for examining the validity of a hypothesis through empirical analysis. When the regression equation is valid, the value of the dependent variable can be estimated or predicted based on the value of the independent variable. Regression analysis is divided into a case with one independent variable and a case with two or more independent variables. One case is called a simple regression analysis, and two or more cases is called a multiple regression analysis.

Specifically, regression analysis is a methodology for analyzing the relationship between a function relationship (regression equation) and a coefficient of determination when trying to predict the effect of some process factors, the input variable (X), on the output value (Y). Through this, information on the level of process management of the input variable (X) to obtain the desired output value by checking what process factor affects the output value (Y) and how much. Regression analysis proceeds in the order of checking the scatterplot of the collected data, drawing a fitted line, performing variance analysis, verifying the regression model, and performing residual analysis.

Regression analysis is a method of formulating and expressing the relationship between one independent variable (X) and one dependent variable (Y). The model of regression analysis is as follows.

y = α + βχ ₁ + ε _i ,

Here, ε _i ~ N(0, σ ² ) is independent and follows a normal distribution, and σ ² is an unknown constant. y = a + bx is a straight line estimated through the collected data, and E(Y1) = α + βχ is an unknown true straight line.

On the other hand, the regression equation, which is a functional relationship, is a prediction equation that enables prediction of an output value (Y) corresponding to the process factor data, which is an input variable (X), and does not necessarily have to be linear unlike the previous one. The regression equation is simple regression with one independent variable and a linear relationship with the dependent variable, curve regression with one independent variable and a curve with a relationship with the dependent variable, and a linear relationship with several independent variables and a dependent variable. There is a linear function, multiple regression. In addition, there are polynomial regression in which there are several independent variables and the relationship with the dependent variable is more than a linear function, and nonlinear regression in which _{the shape of the regression equation does not form a linear relationship with the unknown parameter β i.}

An embodiment of the present invention implements a prediction equation through polynomial regression in the previous regression equation. The polynomial regression is an analysis that considers two or more independent variables to explain the change of the dependent variable. Since the dependent variable is affected by two or more independent variables, if you select several independent variables with a large influence and check the changing characteristics of the dependent variable, you can perform an accurate analysis compared to simple regression. Therefore, set the independent variable X as current, voltage, temperature, air pressure, humidity, static electricity, illuminance, etc., and create a relational expression with high predictive suitability for the type of process defect for which the dependent variable Y is to be known, and the optimal process management specification for the independent variable. Is set.

Meanwhile, before deriving the process defect prediction model equation, the error is verified (S16), the weights of major process factors are optimized (S17), and the correlation is verified (S14), and then applied to the machine learning regression analysis algorithm. . In the error verification (S16), for example, for each process, the amount of deviation between the predicted regression value and the measured value is verified. Here, the amount of deviation is the linear equation Y = a + bx, and the optimal line is a straight line with a correlation coefficient (a and b) generated by minimizing the sum of the squared difference between the actual data and the straight line, and the points of the actual data. It refers to the difference between straight lines. In addition, process management specifications for process defects are set in consideration of tolerances in predicted values of regression analysis. Big data is formed by continuously accumulating measured data based on the set specifications. The weight analysis is determined by analyzing the correlation as shown in FIG. 3.

When the process defect prediction model equation is derived, the minimum significance level of the process defect is analyzed using the ANOVA technique (S19). In this case, the ANOVA technique compares the averages of two or more recruits, and the data is in a continuous form. The ANOVA technique manages the distribution of process factors related to significant differences in data defects as shown in FIG. 4. Specifically, the ANOVA technique determines the process's significant difference criterion as an influence factor of the process through hypothesis testing for the difference in the mean value between process factors, and if the significance level P-value is, for example, a significance level of, for example, 0.05 or less. , To adjust the specifications for this to a good level. Thereafter, when the specification of the process defect improvement management is determined, it is operated based on this (S20).

Table 2 shows examples of regression analysis results for process management specifications according to an embodiment of the present invention.

According to Table 2, process defects are predicted in real time using a regression analysis algorithm of machinerunnin, and the relative importance of process factors is qualitatively assigned with the AHP technique, and the center position and distribution of the measured values are determined by the ANOVA technique. Process defects can be predicted by taking the considered process factors into account. As a result of the process defect determination, if the process is defective (NG), it is converted to 1 in the ONE-HOT code, and if it is not (OK), it is converted to 0 in the ONE-HOT code. This process can be repeated and used as big data.

In the above, the present invention has been described in detail with reference to a preferred embodiment, but the present invention is not limited to the above embodiment, and various modifications are made by those of ordinary skill in the art within the scope of the technical idea of the present invention. It is possible.

Claims (7)

In order to predict process defects, in a process defect prediction platform that utilizes IoT and big data,
Extracting independent variables from a plurality of process factors and setting dependent variables;
Extracting major process factors by applying an AHP technique that qualitatively assigns relative importance to the process factors from the independent variable;
Verifying the degree of correlation between the major process factors;
Deriving a process defect prediction model equation by applying a machine learning regression analysis algorithm to the process factor whose correlation has been verified;
Analyzing the minimum significance level of the process defect by applying the ANOVA technique in which the process defect derived from the prediction model equation is considered in consideration of the center position and the distribution of the measured values; And
Including the step of determining and operating a process management specification of the process defect that has finished the analysis of the significance level,
The correlation analysis code of the correlation verification is
# pip install seaborn <<- in virtualenv !!
import seaborn as sns
fig, ax = plt.subplots(figsize=(10,10))
A method for predicting process defects considering process factors, characterized in that sns.heatmap(df.corr(), annot=True, linewidth=.5, ax=ax). The method of claim 1, wherein the plurality of process factors are process factors in an independent input process, a production process, and a completion process, respectively. delete The method of claim 1, further comprising verifying errors and optimizing weights of major process factors before deriving the process defect prediction model equation. 5. The method of claim 4, wherein the error is verified by analyzing the amount of deviation between the predicted value of regression analysis and the measured value. The method of claim 1, wherein the ANOVA technique determines that a significant difference index P-value is less than or equal to the significance level as a process influence factor and adjusts the specification to a good level. The method of claim 1, wherein if the process defect determination result by the ANOVA method is NG, converts to 1 of the ONE-HOT code, and if not (OK), converts to 0 of the ONE-HOT code. Process defect prediction method considering process factors, characterized in that.

Images