KR102079359B1 - Process Monitoring Device and Method using RTC method with improved SAX method - Google Patents

Abstract

에 대하여(여기서,

는 기 설정된 알파벳 크기) 상기 시계열 데이터를 심볼화하여 심볼화된 데이터를 생성하는 단계, (c) 상기 심볼화된 데이터와 상기 시계열 데이터간에 MSE(mean square error)를 계산하는 단계, (d) 상기 계산된 MSE가 가장 작을 때의 심볼화된 데이터에 대한 중단점을 최적 중단점으로 결정하는 단계, 및 (e) 상기 심볼화된 데이터 중에서 상기 최적 중단점에서의 심볼화된 데이터에 대하여 RTC(real-time contrast) 기법을 적용하여 상기 공정의 이상여부를 탐지하는 단계를 포함한다.Process monitoring methods are disclosed. The process monitoring method is a process monitoring method performed by a process monitoring device, the method comprising: (a) collecting time series data measured through the process, (b) any break point

For (where,

Symbolizing the time series data to generate symbolized data, (c) calculating a mean square error (MSE) between the symbolized data and the time series data, and (d) Determining a breakpoint for the symbolized data when the calculated MSE is the smallest as the optimum breakpoint, and (e) real symbolic data for the symbolized data at the optimal breakpoint among the symbolized data. -detecting the abnormality of the process by applying a (time contrast) technique.

Description

Process monitoring device and method using RTC method with improved SAX method}

The present invention relates to a process monitoring apparatus and method, and more particularly, to a process monitoring apparatus and method for pre-processing input data of the RTC technique to determine more efficient decision boundary in the RTC technique.

As advanced technologies develop, manufacturing processes become more complex and diverse, and these changes make process monitoring more important. As a conventional defect detection method, the SPC (Statistical Process Control) technique is inefficient or leads to unreliable results. In order to solve the problems of the existing SPC technique, we tried to classify normal and abnormal using machine learning, but it is impossible to reflect new data that reflects the state of the current manufacturing process better. There is a risk of deterioration.

RTC (Real-Time Contrasts) technique is a technique for overcoming the limitations of the supervised learning model described above. When a new data is collected, a new classifier is trained and a statistic is calculated based on the learned result. Conventional RTC charts use random forests as classifiers, which provide easy-to-understand information about the criteria for classifying normal and abnormal conditions and can be causally analyzed through variable importance. have. However, random forest classifiers produce discrete statistics based on the number of decision trees, which can be calculated to the same value even though the manufacturing process has larger anomalies, which can cause performance degradation. .

On the other hand, the decision tree constituting the random forest classifier performs classification with decision boundaries for each variable. Therefore, the performance of the random forest based RTC chart is more affected by the class divided by the decision boundary than the value of each variable. Accordingly, unlike the existing RTC management chart for improving the classifier, it is necessary to improve the class prediction performance of each decision tree through pattern classification of data.

Republic of Korea Patent 1187345 Republic of Korea Patent No. 1562623

The object of the present invention is to improve the class prediction performance of each decision tree by making the data and the pattern of the data more clear through the processing of the input data of the RTC technique, and to more accurately distinguish the data in the moving window. To provide a process monitoring apparatus and method that can be.

에 대하여(여기서,

는 기 설정된 알파벳 크기) 상기 시계열 데이터를 심볼화하여 심볼화된 데이터를 생성하는 단계, (c) 상기 심볼화된 데이터와 상기 시계열 데이터간에 MSE(mean square error)를 계산하는 단계, (d) 상기 계산된 MSE가 가장 작을 때의 심볼화된 데이터에 대한 중단점을 최적 중단점으로 결정하는 단계, 및 (e) 상기 심볼화된 데이터 중에서 상기 최적 중단점에서의 심볼화된 데이터에 대하여 RTC(real-time contrast) 기법을 적용하여 상기 공정의 이상여부를 탐지하는 단계를 포함할 수 있다.Process monitoring method according to an embodiment of the present invention is a process monitoring method performed by the process monitoring device, (a) collecting time series data measured through the process, (b) any break point (break point)

For (where,

Symbolizing the time series data to generate symbolized data, (c) calculating a mean square error (MSE) between the symbolized data and the time series data, and (d) Determining a breakpoint for the symbolized data when the calculated MSE is the smallest as the optimum breakpoint, and (e) real symbolic data for the symbolized data at the optimal breakpoint among the symbolized data. The method may include detecting an abnormality of the process by applying a -time contrast technique.

에 대하여(여기서,

는 기 설정된 알파벳 크기) 상기 시계열 데이터를 심볼화하여 심볼화된 데이터를 생성하고, 상기 심볼화된 데이터와 상기 시계열 데이터간에 MSE(mean square error)를 계산하여 상기 계산된 MSE가 가장 작을 때의 심볼화된 데이터에 대한 중단점을 최적 중단점으로 결정하는 데이터 가공부, 및 상기 심볼화된 데이터 중에서 상기 최적 중단점에서의 심볼화된 데이터에 대하여 RTC(real-time contrast) 기법을 적용하여 상기 공정의 이상여부를 탐지하는 이상 탐지부를 포함할 수 있다.Process monitoring device according to an embodiment of the present invention collects the time series data measured through the process, any break point (break point)

For (where,

Symbolizes the time series data to generate symbolized data, calculates a mean square error (MSE) between the symbolized data and the time series data, and calculates a symbol when the calculated MSE is smallest. The process by applying a real-time contrast (RTC) technique to the data processing unit for determining the breakpoint for the normalized data as the optimum breakpoint, and the symbolized data at the optimal breakpoint among the symbolized data It may include an abnormality detection unit for detecting the abnormality of.

Process monitoring apparatus and method according to an embodiment of the present invention can improve the class prediction performance of each decision tree through the processing of the input data, it is possible to more accurately distinguish the data in the mobile window.

In addition, it is possible to accurately analyze the cause of the defect and to detect abnormal conditions more quickly.

1 is a block diagram of a process monitoring apparatus according to an embodiment of the present invention.
2 illustrates a parallel coordinate system of time series data and symbolized data according to dimension numbers.
3 shows the MSE calculated between time series data and symbolized data.
4 is a flow chart of a process monitoring method according to an embodiment of the present invention.
5 is a flowchart illustrating a method of detecting an abnormality according to an embodiment of the present invention.
6A illustrates a conventional classification technique and monitoring statistics of data classified according to an embodiment of the present invention.
6B illustrates a value of original data and a value of data converted according to an embodiment of the present invention.
6C illustrates variable importance of data calculated according to an embodiment of the present invention.
7 illustrates a classification result of original data and converted data according to an embodiment of the present invention.

Specific structural or functional descriptions of the embodiments according to the inventive concept disclosed herein are merely illustrated for the purpose of describing the embodiments according to the inventive concept, and the embodiments according to the inventive concept. These may be embodied in various forms and are not limited to the embodiments described herein.

Embodiments according to the inventive concept may be variously modified and have various forms, so embodiments are illustrated in the drawings and described in detail herein. However, this is not intended to limit the embodiments in accordance with the concept of the present invention to specific embodiments, and includes modifications, equivalents, or substitutes included in the spirit and scope of the present invention.

Terms such as first or second may be used to describe various components, but the components should not be limited by the terms. The terms are only for the purpose of distinguishing one component from another component, for example, without departing from the scope of the rights according to the inventive concept, the first component may be called a second component, Similarly, the second component may also be referred to as the first component.

When a component is said to be "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may be present in the middle. Should be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it should be understood that there is no other component in between. Expressions describing relationships between components, such as "between" and "immediately between" or "directly neighboring", should be interpreted as well.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. As used herein, the terms "comprise" or "have" are intended to designate that the stated feature, number, step, operation, component, part, or combination thereof is present, but one or more other features or numbers, It should be understood that it does not exclude in advance the possibility of the presence or addition of steps, actions, components, parts or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art and, unless expressly defined herein, are not construed in ideal or excessively formal meanings. Do not. Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings.

There are various data representation techniques to reduce the size and noise of process data. Such data representation techniques include, for example, Discrete Fourier Transformation (DFT), Discrete Wavelet Transform (DWT), Single Value Decomposition (SVD), Pageswise Aggregate Approximation (PAA), and Symbolic Aggregate Approximation (SAX).

Here, the SAX technique is a data representation technique for symbolizing time series data into a string, and is a technique in which a string-based symbolization algorithm is combined with a PAA technique for representing data using an average of sequences.

를 길이 M의 시퀀스

으로 나타낸다.PAA technique is a sequence of length n using Equation 1 below

Sequence of length M

Represented by

Here, n and M satisfy n≥M as a natural number.

That is, in order to reduce the n-dimensional time series to M-dimension, the PAA technique divides the time series data into frames having the same size and represents the data as an average value of data belonging to each frame.

The SAX technique symbolizes each segment of data converted through the PAA technique as described above as a string. Here, the string may include, for example, an alphabet or a number.

The existing SAX technique aggregates the time base, and loss of time information due to aggregation of the time base may be fatal in detecting an error in real time.

On the other hand, the process monitoring apparatus and method according to an embodiment of the present invention is an adaptive decision boundary SAX (ADB-SAX), which is a SAX technique improved from the existing SAX technique to improve the performance of the RTC control chart. ) The input data is processed using the technique, and the process is monitored by applying the RTC technique to the processed input data.

1 is a block diagram of a process monitoring apparatus according to an embodiment of the present invention.

Referring to FIG. 1, the process monitoring apparatus 10 may include a data processor 100 and an abnormality detector 200.

The data processing unit 100 collects time series data measured through a process and processes the collected time series data before applying the RTC technique. More specifically, the data processing unit 100 symbolizes the time series data to generate symbolized data, calculates a mean square error (MSE) between the symbolized data and the time series data, and calculates the smallest MSE. The breakpoint for the symbolized data is determined as the optimal breakpoint.

To this end, the data processing unit 100 includes a symbolization unit 110 and an optimum breakpoint determiner 130.

에 대하여 수집된 시계열 데이터를 심볼화하여 심볼화된 데이터를 생성한다. 여기서, 중단점은 시계열 데이터를 동일한 크기의 영역으로 나누는 경계점이다.The symbolization unit 110 may have any break point.

Symbolized data is generated by symbolizing the collected time series data. Here, the breakpoint is a boundary point that divides time series data into regions of the same size.

A breakpoint according to an embodiment may be a set of points that divide a Gaussian curve of N (0, 1) into regions of the same size.

와 같이 나타낼 수 있다. 여기서,

는 기 설정된 알파벳 크기(alphabet size)이다. 기 설정된 알파벳 크기는 후술할 알파벳 크기 결정부(150)에 의해 결정된 값일 수 있다.The breakpoint according to another embodiment is a set of points defined by the quartiles q _main and q _others ,

It can be expressed as here,

Is a preset alphabet size. The preset alphabet size may be a value determined by the alphabet size determiner 150 to be described later.

은 데이터의 분포에 따라 정렬되는데, 평균에 가까운 중단점일수록 분류에 상대적으로 효과가 없다고 볼 수 있다. 이에 따라, 상기 임의의 중단점은 평균으로부터 보다 멀리 위치한 중단점들의 집합인

와 평균으로부터 보다 가까이 위치한 중단점들의 집합인

으로 분류될 수 있다. 이때, q_main과 q_others는 각각

과

에 대한 분위수(quantile)를 의미한다.

과

에 속해있는 중단점들은 임의로 설정될 수 있는 q_main과 q_others에 따라 정해질 수 있다.Any breakpoint

Are sorted according to the distribution of the data. The closer to the mean the breakpoint, the less effective the classification. Accordingly, the random breakpoint is a set of breakpoints located farther from the mean.

And a set of breakpoints located closer to the mean

Can be classified as. Where q _main and q _others are respectively

and

The quantile for.

and

Breakpoints belonging to can be determined by q _main and q _others , which can be set arbitrarily.

내지

이다. 이때, 임의로 설정된 q_main과 q_others에 대하여

와

는 대칭점으로서 각각 q_main에 1 표준 편차를 더한 값과 1-q_main에 1 표준 편차를 뺀 값이며,

와

는 대칭점으로서 각각 q_main과 1-q_main이고,

와

는 대칭점으로서 각각 q_others와 _1-q_others이다. For example, if the preset alphabet size is 7, the number of breakpoints is 6, so the set of breakpoints is

To

to be. At this time, for the randomly set q _main and q _others

Wow

Are symmetry points, where q _main plus 1 standard deviation and 1-q _main minus 1 standard deviation.

Wow

Are symmetry points, q _main and 1-q _main ,

Wow

Are symmetry points, q _others and _1- q _others, respectively.

For each time series data area divided by any breakpoint as described above, the symbolizing unit 110 calculates the averaged value of each time series data area from the collected time series data as shown in Equation 2 below. Create

는 시계열 데이터의 표준 편차를 의미한다.Here, E (area _i ) means an average value for the i th region of the time series data divided by the breakpoint,

Is the standard deviation of the time series data.

On the other hand, unlike the conventional SAX technique of counting the time axis prior to symbolization as described above, the symbolization unit 110 may symbolize only the data value without counting the time axis. That is, the symbolization unit 110 may omit the step of representing the data as the average of the sequence by the PAA technique (n = M in Equation 1), and minimize the loss of time information by symbolizing only the data value. have.

As described above, the time series data is converted into categorical data according to the symbolization unit 110.

Referring to FIG. 2, which shows a parallel coordinate system of time series data and symbolized data according to the number of dimensions, the x-axis is the number of variables, and the y-axis is a scale value for the maximum and minimum values in each variable. Through the symbolization process, it can be seen that the values of each variable change into categorical data according to the alphabet size, and the pattern of each data does not change significantly.

In addition, the change in categorical data affects the determination of the decision boundary of the random forest. The decision tree constituting the random forest compares the values of each variable to determine the decision boundary. The greater the distance between the values, the higher the probability of occurrence of the decision boundary. If there is a decision boundary between healthy data, the monitoring statistics will rise. In contrast, when the data is categorized according to an embodiment of the present invention to minimize the interval of each value, the statistics within the normal categorical region may be lowered.

The optimum breakpoint determiner 130 determines an optimal breakpoint for any breakpoint. More specifically, the optimum breakpoint determiner 130 calculates an MSE between data symbolized by the symbolizer 110 and time series data for an arbitrary breakpoint, and symbolizes when the calculated MSE is smallest. The breakpoint for the generated data is determined as the optimum breakpoint.

Here, MSE is an average squared error indicating information loss due to symbolization of time series data. That is, since the loss of information due to symbolization is minimal when the minimum value of each MSE calculated between arbitrary breakpoint combinations and time series data is represented, the optimal breakpoint determiner 130 may determine that the MSE is the minimum. The breakpoint in time is determined as the optimum breakpoint.

Referring to FIG. 3, which shows the MSE calculated between time series data and symbolized data, the x-axis represents the index for the combination of breakpoints, and the y-axis is the MSE value at each index. For example, if q _main changes from 81 to 99 in 1st percentile intervals, and q _others changes from 60 to 80 in 1st percentile intervals, the combination of breakpoints changes and the MSE value changes accordingly. The point at which the index is approximately 230 (q _main = 92, q _others = 68), the MSE value represents the minimum, so the breakpoint at this time can be determined as the optimal breakpoint.

Here, the data processing unit 100 may further include an alphabet size determining unit 150.

The alphabet size determiner 150 determines an alphabet size for determining the number of breakpoints. In the SAX technique, the alphabet size is a parameter that determines the number of alphabets to be symbolized. If the time axis is not aggregated in the PAA technique (n = M in Equation 1 above), the size of the category representing the data at each time point is determined. Determine the number This alphabet size is preferably set smaller when the data is a simple symmetric distribution, and is set larger as the skewness has a larger skewness.

On the other hand, the skewness in the statistical approach is related to the distribution of mean and mode. For example, for a left-skewed distribution, the mean is located to the left of the mode, and for a right-skewed distribution, the mean is to the right of the mode. In other words, the difference between the mode and the mean may be a useful index for determining the alphabet size.

Accordingly, the alphabet size determining unit 150 according to an embodiment may determine the alphabet size as shown in Equation 3 below.

는 시계열 데이터가 정규 분포를 따를 때의 중단점의 최대값, mode와 mean은 각각 이동식 창(moving window)에서의 시계열 데이터의 최빈값과 평균값을 의미한다. 한편, 알파벳 크기의 범위는 데이터의 왜도에 따라 달리 설정될 수 있다.Here, S may be arbitrarily set as a range of alphabetical sizes,

Denotes the maximum value of the breakpoint when the time series data follow a normal distribution, and mode and mean, respectively, the mode and mean value of the time series data in the moving window. On the other hand, the range of the alphabet size may be set differently according to the skewness of the data.

The abnormality detection unit 200 detects an abnormality of the manufacturing process by applying a real-time contrast (RTC) technique to the data processed through the data processing unit 100.

To this end, the abnormality detector 200 includes a learner 210, a classification probability calculator 230, a monitoring statistics calculator 250, and a detector 270.

The learning unit 210 uses the data when the collected data is in the normal state among the data converted through the data processing unit 100 as reference data, and then contrasts the data collected in real time. Learn the classifier. Here, the control data applies a movable window every time a new measurement is made at time t, so that only N _w data of the most recently collected data are included.

)를 생성한다. 여기서, x_i는 변환된 데이터이고, t_j는 j번째 분류기를 의미한다.A classifier is a random forest classifier that includes at least one decision tree that learns the reference data and the control data to generate a decision boundary at every time t, classifying and matching the reference data into class 0. The data is classified as Class 1. Through learning, the classifier can predict

) Here, x _i is transformed data and t _j means j-th classifier.

In this case, even if the number of decision trees in the random forest classifier is large, the problem of overfitting does not occur.

The classification probability calculation unit 230 calculates a classification probability as shown in Equation 4 below by using the predicted value generated by the learning unit 210.

The random forest classifier uses bagging to fully grow each individual decision tree to reduce bias and to reduce correlation between individual trees. At this time, out of bagging (OOB) means data that is not used when learning a decision tree for bagging.

는 지시자 함수로서, 예측치가 실제 클래스 k와 일치하면 1을 반환하고, 일치하지 않으면 0을 반환하는 함수이다.

Is an indicator function that returns 1 if the prediction matches the actual class k and 0 if it does not.

The monitoring statistics calculator 250 calculates the monitoring statistics using the classification probabilities calculated by the classification probability calculator 230. In general, the monitoring statistics for the reference data are much larger than the monitoring statistics for the control data, which makes them more stable and detectable. Accordingly, the monitoring statistics calculator 250 calculates the monitoring statistics for the reference data as shown in Equation 5 below.

Here, N ₀ means the size of the reference data S ₀ .

The detector 270 detects an abnormal state when the monitoring statistics calculated by the monitoring statistics calculator 250 fall outside a preset control limit. Accordingly, the process state changes from in-conrtorl to out-of-control state.

On the other hand, the preset control limit line is a value that can be set differently according to the type or environment of the process, the type of the product, or the abnormality to be detected. It can be set using.

Here, the abnormality detection unit 200 may further include a cause diagnosis unit 290.

The cause diagnosis unit 290 calculates variable importance to diagnose the cause of the abnormality when the process is detected as an abnormal state through the detection unit 270, and diagnoses the cause of the abnormality using the variable importance. . Variable importance is the degree of contribution to detecting anomalies, which can be calculated through impurity scores. Here, the impurity score may be, for example, a Gini index or Shannon entropy.

According to an embodiment, when using the Gini index, the Gini index in the random forest is expressed by Equation 6 below.

Where v is a node in the decision tree, c is the number of classes, and r _i represents the proportion of class i at each node. According to the calculated Gini index, the variable importance is calculated as in Equation 7 below.

는 불순도의 감소도이다. 이때, 불순물의 감소도는 노드 v에서 자식 노드들에 대해 나누어지는 비율을 나타내는 가중치 w_L과 w_R을 반영하는 것으로서, 아래의 수학식 8과 같다.Where ntree is the total number of trees, D _t is the individual classifier,

Is the decrease in impurity. At this time, the degree of reduction of impurities reflects the weights w _L and w _R indicating the ratio divided by the child nodes at the node v, as shown in Equation 8 below.

The cause diagnosis unit 290 calculates the importance of the variable as described above, and may diagnose that the cause of the abnormality is in the variable having the high variable importance.

Here, the process monitoring device 10 may further include a separate database for storing the collected time series data or the results calculated by the data processing unit 100 and the abnormality detection unit 200.

4 is a flow chart of a process monitoring method according to an embodiment of the present invention. Hereinafter, detailed descriptions of the overlapping portions will be omitted.

Referring to FIG. 4, step S310 is a step of collecting time series data measured through a process.

에 대하여 시계열 데이터를 심볼화하여 심볼화된 데이터를 생성한다. S320 단계는 임의의 중단점에 의해 나누어진 각 시계열 데이터 영역에 대하여, 상기 수학식 2에 의해 각 시계열 데이터 영역의 평균값을 계산하여 시계열 데이터를 심볼화한다.S320 step is a random breakpoint

Symbolize time series data with respect to to generate symbolic data. In step S320, for each time series data area divided by an arbitrary breakpoint, the average value of each time series data area is calculated by Equation 2 to symbolize the time series data.

In operation S330, the MSE is calculated between the symbolized data and the collected time series data. Step S330 may be performed repeatedly for all combinations of breakpoints.

Step S340 determines the breakpoint for the symbolized data when the MSE calculated in step S330 is the smallest as the optimum breakpoint.

Step S350 detects an abnormality of the process by applying the RTC technique to the symbolized data at the optimal breakpoint among the symbolized data in step S340.

5 is a flowchart illustrating a method of detecting an abnormality according to an embodiment of the present invention.

Referring to FIG. 5, step S351 learns symbolized data and control data using a random forest classifier, and generates prediction values of a decision tree.

When the predicted value is generated through the step S351, the step S352 calculates the classification probability as shown in Equation 4 using the predicted value.

Step S353 calculates the monitoring statistics as shown in Equation 5 using the classification probability calculated through step S352.

Step S354 detects an abnormal state when the monitoring statistics calculated through step S353 are out of the preset management limit line.

On the other hand, if the abnormal state is detected by the step S354, in order to diagnose the cause of the abnormality to calculate the importance of the variable as shown in equation (7), and additionally perform the step S355 to extract the variable causing the abnormality through the calculated variable importance Can be.

6A illustrates a conventional classification technique and monitoring statistics of data classified according to an embodiment of the present invention, and FIG. 6B illustrates values of original data and values of data converted according to an embodiment of the present invention. FIG. 6C illustrates the variable importance of data calculated according to an embodiment of the present invention.

Referring to FIG. 6A, the processed data is included as a normal category within the normal category, and is divided into an external category outside the normal category, and the converted data has lower monitoring statistics than when not converted from the normal category. This is because no decision boundary can occur between data having the same category value. On the other hand, data outside the normal category is larger than the existing monitoring statistics because it is classified into the outer category. That is, according to the present invention, the difference between the monitoring statistics of the normal category and the monitoring statistics of the abnormal category is widened, thereby increasing the detection performance. You can see that the statistic does not exceed CL.

Meanwhile, referring to FIG. 6B, when data is converted according to an embodiment of the present invention, a time point indicating a phenomenon in which the converted data value is increased from 2.699 to 3.977 as in time point 22 may occur. Since the misclassification of the point is recognized as one misclassification of the size of the movable window, the phenomenon shown in FIG. 6B is not a problem when the RTC technique is employed.

In addition, referring to FIG. 6C, it can be seen that X1 has a similar value to other variables in variable importance indicating cause analysis.

7 illustrates a classification result of original data and converted data according to an embodiment of the present invention.

Referring to Fig. 7, the decision boundaries for the original data (a) and the processed data (b) are displayed together. Each point is labeled with a class (0 or 1) of the class. Unlike the original data, when classifying using the processed data according to the present invention, it can be seen that the class is clearly divided by the decision boundary.

As described above, in the case of the process monitoring apparatus and method according to an embodiment of the present invention, performance of process monitoring may be improved through data processing that minimizes loss of information prior to applying the RTC technique.

The apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the devices and components described in the embodiments may include, for example, processors, controllers, arithmetic logic units (ALUs), digital signal processors, microcomputers, field programmable arrays (FPAs), It may be implemented using one or more general purpose or special purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to the execution of the software. For convenience of explanation, one processing device may be described as being used, but one of ordinary skill in the art will appreciate that the processing device includes a plurality of processing elements and / or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as parallel processors.

The software may include a computer program, code, instructions, or a combination of one or more of the above, and may configure the processing device to operate as desired, or process independently or collectively. You can command the device. Software and / or data may be any type of machine, component, physical device, virtual equipment, computer storage medium or device in order to be interpreted by or to provide instructions or data to the processing device. Or may be permanently or temporarily embodied in a signal wave to be transmitted. The software may be distributed over networked computer systems so that they may be stored or executed in a distributed manner. Software and data may be stored on one or more computer readable recording media.

The method according to the embodiment may be embodied in the form of program instructions that can be executed by various computer means and recorded in a computer readable medium. The computer readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the media may be those specially designed and constructed for the purposes of the embodiments, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the embodiments have been described with reference to the accompanying drawings as described above, various modifications and variations are possible to those skilled in the art from the above description. For example, the described techniques may be performed in a different order than the described method, and / or components of the described systems, structures, devices, circuits, etc. may be combined or combined in a different form than the described method, or other components. Or, even if replaced or substituted by equivalents, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are within the scope of the claims that follow.

Claims (10)

A process monitoring method performed by a process monitoring device,
(a) collecting time series data measured through the process;
(b) any break point

For (where,

Symbolizing the time series data to generate symbolized data;
(c) calculating a mean square error (MSE) between the symbolized data and the time series data;
(d) determining a breakpoint for the symbolized data when the calculated MSE is smallest as an optimal breakpoint; And
(e) detecting an abnormality of the process by applying a real-time contrast (RTC) technique to the symbolized data at the optimum breakpoint among the symbolized data.
The method of claim 1,
Step (b) is a process monitoring method for symbolizing the data value of the time series data, without aggregating the time axis.
The method of claim 1,
The step (b) is a process monitoring method for symbolizing the time series data by calculating the average value of each time series data area for each time series data area divided by the arbitrary breakpoint.
The method of claim 3,
The average value of each time series data area is calculated by Equation 1 below.
[Equation 1]

Where E (area _i ) is the mean value for the i th region,

Is the standard deviation of the time series data.
The method of claim 1,
The preset alphabet size is a process monitoring method defined by Equation 2 below.
[Equation 2]

Where S is a range of alphabetic sizes,

Is the maximum value of the breakpoint when the time series data follow the normal distribution, and mode and mean are the modes and mean values of the time series data in the moving window, respectively.
The method of claim 1,
Step (e) is
(e-1) learning the symbolized data and contrast data using a random forest classifier and generating predictions of Decision Trees;
(e-2) calculating a classification probability using the prediction value;
(e-3) calculating monitoring statistics using the classification probability; And
(e-4) detecting the abnormal state when the monitoring statistic is out of a predetermined control limit.
The method of claim 6,
Step (e) is
(e-5) The process monitoring method further comprises the step of diagnosing the cause of the abnormality using variable importance, if detected as abnormal.
Collect time series data measured by the process, and random breakpoints

For (where,

Symbolizes the time series data to generate symbolized data, calculates a mean square error (MSE) between the symbolized data and the time series data, and calculates a symbol when the calculated MSE is smallest. A data processing unit which determines a breakpoint for the normalized data as an optimum breakpoint; And
And an abnormality detection unit detecting an abnormality of the process by applying a real-time contrast (RTC) technique to the symbolized data at the optimum breakpoint among the symbolized data.
The method of claim 8,
The data processing unit symbolizes data values of the time series data without aggregation of a time axis,
For each time series data area divided by the arbitrary breakpoint, the average value of each time series data area is calculated to symbolize the time series data,
The average value of each time series data area is calculated by Equation 1 below.
[Equation 1]

Where E (area _i ) is the mean value for the i th region,

Is the standard deviation of the time series data.
The method of claim 8,
The data processing unit
And a alphabet size determiner configured to determine the preset alphabet size by Equation 2 below.
[Equation 2]

Where S is a range of alphabetic sizes,

Is the maximum value of the breakpoint when the time series data follow the normal distribution, and mode and mean are the modes and mean values of the time series data in the moving window, respectively.

Images