Apparatus and method for detecting outliers using feature vector clustering

Abstract

The present invention relates to an apparatus and method for detecting outliers and, more particularly, to an apparatus and method for detecting outliers using feature vector clustering. According to one embodiment of the present invention, outliers can be detected even from data lacking defect samples, and the time required for detection can be reduced through rapid outlier detection. Furthermore, the present invention enables predictive maintenance, thereby reducing maintenance costs.

Description

Device and method for detecting outliers using feature vector clustering

[0001] The present invention relates to an outlier detection device and method, and more specifically, to an outlier detection device and method using feature vector clustering.

[0002]

[0003] Predictive maintenance is necessary to reduce downtime in production manufacturing lines.

[0004] Predictive maintenance is an equipment maintenance methodology that collects large amounts of data from sensors and uses it to predict when equipment will stop working in order to minimize downtime caused by defects.

[0005] However, since actual production data is collected in real-time during the actual process, it is difficult to find labeled defect data, which limits the collection of data regarding outliers. Additionally, because the actual process utilizes multiple pieces of equipment and sensors, it is challenging to process large volumes of data per second within a short period.

[0006] To solve these problems, it is necessary to provide an anomaly detection device and method using feature vector clustering that rapidly checks large amounts of sensor data to detect anomalies and utilizes them for predictive maintenance.

[0007] The background technology of the present invention is disclosed in Korean Published Patent No. 10-2023-0067360.

[0008]

[0009] The present invention provides an anomaly detection device and method using feature vector clustering, which rapidly checks a large amount of sensor data to detect anomalies and utilizes them for predictive maintenance.

[0010]

[0011] According to one aspect of the present invention, an outlier detection device is provided.

[0012] An outlier detection device according to one embodiment of the present invention may include a data collection unit for collecting data, a data preprocessing unit for preprocessing the data, a feature extraction unit for extracting a feature vector from the data, and an outlier detection unit for detecting outliers using the feature vector.

[0013] According to another aspect of the present invention, a method for detecting outliers is provided.

[0014] An outlier detection method according to one embodiment of the present invention may include the steps of collecting data, preprocessing the data, extracting a feature vector from the data, and detecting outliers using the feature vector.

[0015]

[0016] According to one embodiment of the present invention, the present invention can detect outliers even in data that is free of defect data.

[0017] According to one embodiment of the present invention, the present invention can reduce the time required by rapidly detecting anomalies and enables predictive maintenance, thereby reducing maintenance costs.

[0018] The effects of the present invention are not limited to the effects described above, and should be understood to include all effects that can be inferred from the composition of the invention described in the description or claims of the present invention.

[0019]

[0020] FIGS. 1 and FIGS. 2 are drawings briefly illustrating an outlier detection device according to one embodiment of the present invention.

[0021] FIG. 3 is a flowchart briefly illustrating an outlier detection method according to one embodiment of the present invention.

[0022] FIG. 4 is a flowchart illustrating in detail a method for detecting outliers according to an embodiment of the present invention.

[0023] FIG. 5 is a drawing showing original data according to an embodiment of the present invention.

[0024] FIG. 6 is a diagram showing a time domain feature vector according to an embodiment of the present invention.

[0025] FIG. 7 is a diagram showing a frequency domain feature vector according to an embodiment of the present invention.

[0026] FIG. 8 is a diagram showing the center of optimal data through the silhouette method and the elbow method according to one embodiment of the present invention.

[0027] FIGS. 9 to 15 are drawings showing the results of applying the K-means method and PCA according to an embodiment of the present invention.

[0028] FIGS. 16 to 22 are graphs visualizing a data source of extreme outliers according to an embodiment of the present invention.

[0029]

[0030] The present invention is susceptible to various modifications and may have various embodiments; therefore, specific embodiments are illustrated in the drawings and described in detail. However, this is not intended to limit the present invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. In describing the present invention, detailed descriptions of related prior art are omitted if it is determined that such detailed descriptions may unnecessarily obscure the essence of the invention. Furthermore, singular expressions used in this specification and claims should generally be interpreted as meaning "one or more" unless otherwise stated.

[0031] Throughout the specification, when it is stated that a part is "connected (connected, in contact, combined)" with another part, this includes not only cases where they are "directly connected," but also cases where they are "indirectly connected" with other members interposed between them. Furthermore, when it is stated that a part "includes" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but rather allows for the inclusion of additional components.

[0032] The terms used herein are merely for describing specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as “comprising” or “having” are intended to indicate the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0033] The present invention will be described below with reference to the attached drawings. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification have been given similar reference numerals.

[0034]

[0035] FIGS. 1 and FIGS. 2 are drawings briefly illustrating an outlier detection device according to one embodiment of the present invention.

[0036] Referring to FIG. 1, the outlier detection device (10) collects data from a sensor attached to a process line (50), preprocesses the data, extracts a feature vector, and determines the outlier.

[0037] For example, the process line (50) may include a DES line that performs at least one of a developing process, an etching process, a stripping process, and a drying process in the PLC manufacturing process.

[0038] The anomaly detection device (10) collects data in real time from the sensor unit (110). The sensor unit (110) may include, for example, a vibration sensor, a current sensor, a temperature sensor, etc.

[0039] The sensor can be attached to the developing process, etching process, stripping process, drying process, etc.

[0040] For example, the sensor can be attached to a total of 13 pieces of equipment, including 2 developing pumps in the developing process, 4 etching pumps, 1 etching circulation pump, 1 HCl pump, 1 NaClO3 pump in the etching process, 1 stripping pump in the stripping process, 1 Air SQ motor, 1 Air Cut motor, and 1 Hot Air motor in the drying process.

[0041] Since each piece of equipment has an x-axis, a y-axis, and a z-axis, three sensors are attached per piece of equipment, and a total of 39 sensors can be attached.

[0042] In the developing process, the developing pump plays the role of spraying the developer solution.

[0043] In the etching process, the etching pump plays the role of spraying the etching solution, the etching circulation pump plays the role of circulating the etching solution within the etching solution tank, the HCl pump plays the role of spraying HCl (hydrogen chloride) required for the etching process, and the NaClO3 pump plays the role of spraying NaClO₃ (sodium chlorate) required for the etching process.

[0044] In the peeling process, the peel pump plays the role of spraying the peeling solution.

[0045] In the drying process, the Air SQ motor, Air Cut motor, and Hot Air motor spray air for drying.

[0046] The sensor collects 4,096 samples per operation and can collect at a sampling rate of 3,200 Hz.

[0047] The sensor is 0g ~ 2g (g = 9.81 It can be used to measure acceleration within the range.

[0048]

[0049] Referring to FIG. 2, the outlier detection device (10) includes a data collection unit (210), a data preprocessing unit (220), a feature extraction unit (230), an outlier detection unit (240), and a result analysis unit (250).

[0050] The data collection unit (210) collects data from the sensor unit (110).

[0051] For example, the data collection unit (210) collects vibration data through the sensor unit (110) of the process line (50) and transmits the collected data to an edge computer using the subgiga protocol.

[0052] The data preprocessing unit (220) preprocesses the data collected by the data collection unit (210).

[0053] For example, edge computers can preprocess collected data.

[0054] The feature extraction unit (230) extracts feature vectors in the time domain and frequency domain from the data preprocessed by the data preprocessing unit (220).

[0055] For example, the time domain extracts 15 features and the frequency domain extracts 7 features.

[0056] Features are extracted in the frequency domain by applying the Fast Fourier Transform (FFT).

[0057] The anomaly detection unit (240) analyzes the extracted feature vector using a clustering method and detects anomalies based on the clustering results.

[0058] The clustering method involves an outlier detection device (10) selecting an ideal k value using the silhouette method and the elbow method to find the number of suitable data centers, and then clustering each feature vector using the K-means method with the selected ideal k value.

[0059] The outlier detection device (10) determines a feature vector that exceeds a certain value at the center of the data as an outlier.

[0060] That is, the outlier detection device (10) calculates the average distance of a data point from the cluster center and determines a data value exceeding the average distance + threshold × standard deviation as an outlier.

[0061] The result analysis unit (250) visualizes the clustering results and detects outliers by applying PCA (Principal Component Analysis).

[0062]

[0063] FIG. 3 is a flowchart briefly illustrating an outlier detection method according to one embodiment of the present invention.

[0064] In step S310, the outlier detection device (10) collects data.

[0065] Here, the collected data is vibration data collected from a sensor attached to the process line (50).

[0066] In step S320, the outlier detection device (10) preprocesses the collected vibration data.

[0067] In step S330, the outlier detection device (10) extracts a feature vector from the preprocessed data.

[0068] In step S340, the outlier detection device (10) analyzes the feature vector using a clustering method and detects outliers based on the clustering results.

[0069] In step S350, the outlier detection device (10) visualizes the clustering results by applying PCA (Principal Component Analysis).

[0070]

[0071] FIG. 4 is a flowchart illustrating in detail a method for detecting outliers according to one embodiment of the present invention.

[0072] In step S405, the outlier detection device collects data and integrates data collected from the same sensor at the same location.

[0073] For example, if there are 39 sensors, the variable is 39.

[0074] For example, an outlier detection device can collect vibration data of the process line (50).

[0075] In step S410, the outlier detection device extracts time domain features and frequency domain features from the data, respectively, to extract a feature vector.

[0076] There are 15 time domain features and 7 frequency domain features.

[0077] Time domain features may include the mean, median, standard deviation, variance, range, min, max, skewness, kurtosis, root mean square, peak to peak, crest factor, shape factor, impulse factor, margin factor, etc.

[0078] Frequency domain features may include frequency center, mean square frequency, root mean square frequency, root variance frequency, spectral skewness, spectral kurtosis, spectral entropy, etc.

[0079] In step S415, the outlier detection device selects an ideal k value using the silhouette method and the elbow method.

[0080] That is, the outlier detection device (10) finds the centroids of the data using the silhouette method and the elbow method of the feature vectors and finds the optimal number of data centroids.

[0081] In step S420, the outlier detection device clusters feature vectors using the K-means method.

[0082] That is, the outlier detection device (10) clusters each feature vector using the K-means method with a selected ideal k value.

[0083] In step S425, the outlier detection device calculates the distance between each data point from the center of the data and determines a feature vector that exceeds a certain distance value from the center of the data as an outlier.

[0084] The outlier detection device (10) calculates the distance between each data point from the centroid of the data and determines a vector with a distance greater than the sum of the average value and the threshold value as an outlier.

[0085] Here, the threshold is the value obtained by multiplying the standard deviation by the set integer threshold.

[0086] That is, the outlier detection device (10) calculates the average distance of a data point from the cluster center and determines a data value exceeding the average distance + threshold × standard deviation as an outlier.

[0087] In step S430, the outlier detection device applies Principal Component Analysis (PCA) to visualize the clustering results and detect outliers.

[0088] The outlier detection device (10) visualizes the actual data of the outliers based on the axis with high explanatory power through PCA (Principal Component Analysis) and then checks the actual data of the outliers.

[0089] PCA (Principal Component Analysis) is a statistical method that reduces high-dimensional data to a lower dimension while preserving the variability of the data as much as possible.

[0090] PCA has the advantage of finding two axes that are fast and highly explanatory to the data.

[0091] The outlier detection device (10) is more suitable for point anomaly detection in the case of outlier detection using a time domain feature vector, and may be more suitable for contextual anomaly detection in the case of outlier detection using a frequency domain feature vector.

[0092] The time taken for data integration by the outlier detection device (10) is 6.8 seconds.

[0093] The time taken for the outlier detection device (10) to vectorize time domain features is 10 seconds.

[0094] The time taken for the outlier detection device (10) to vectorize frequency domain features is 8.6 seconds.

[0095] The time taken for the outlier detection device (10) to detect an outlier is 8.8 seconds.

[0096] That is, the outlier detection device (10) takes 34.2 seconds to detect outliers in about 6 months' worth of data collected from one sensor, which is less than 35 seconds.

[0097] For example, the outlier detection device (10) takes about 23 minutes when using 39 sensors.

[0098] That is, the outlier detection device (10) analyzes the collected time series data using an outlier detection method and can detect outliers (abnormal data) that deviate from the normal range based on the temporal flow of the data.

[0099] That is, the outlier detection device (10) can detect outliers even in data without defect data, and the time required is shorter than the method of using raw data as is and the method of using deep learning.

[0100]

[0101] FIG. 5 is a drawing showing original data according to one embodiment of the present invention.

[0102] Referring to Fig. 5, the original data of vibration data collected from a vibration sensor attached to a developing motor in the developing process is shown.

[0103] Referring to FIG. 5, step S405 of FIG. 4 illustrates an example of an outlier detection device integrating data collected from the same sensor at the same location.

[0104] The time taken for data integration is 6.8 seconds.

[0105]

[0106] FIG. 6 is a diagram showing a time domain feature vector according to one embodiment of the present invention.

[0107] Referring to Figure 6, the time domain feature vector of vibration data collected from a vibration sensor attached to a developing motor in the developing process is summarized and shown.

[0108] Referring to FIG. 6, an example is shown of an outlier detection device extracting time domain features from data and extracting a feature vector in step S410 of FIG. 4.

[0109] The time taken to vectorize time domain features is 10 seconds.

[0110] Referring to Figure 6, there are 5,505 data points and 15 time domain features extracted from each data point.

[0111] Time domain features may include the mean, median, standard deviation, variance, range, min, max, skewness, kurtosis, root mean square, peak to peak, crest factor, shape factor, impulse factor, margin factor, etc.

[0112] Mean is the average value of the signal, median is the median value of the signal, standard deviation is the variability or dispersion of the signal, variance is the degree to which the data is spread out from the mean, range is the difference between the maximum and minimum values ​​in the entire data, min is the minimum value of the signal, max is the maximum value of the signal, skewness is the asymmetry of the signal distribution, kurtosis is the degree of peaking of the signal distribution, root mean square is the average magnitude of the signal energy, peak to peak is the difference between the maximum and minimum values ​​in a specific interval of the signal, crest factor is the ratio of the signal peak to the RMS, shape factor is the ratio of the RMS to the mean value, impulse factor is the ratio of the peak value to the absolute mean value, and margin factor is the ratio of the peak value to the squared RMS value.

[0113]

[0114] FIG. 7 is a diagram showing a frequency domain feature vector according to one embodiment of the present invention.

[0115] Referring to Fig. 7, the frequency domain feature vector of vibration data collected from a vibration sensor attached to a developing motor in the developing process is summarized and shown.

[0116] Referring to FIG. 7, an example is shown of an outlier detection device extracting frequency domain features from data and extracting a feature vector in step S410 of FIG. 4.

[0117] The time taken to vectorize frequency domain features is 8.6 seconds.

[0118] Referring to Fig. 7, there are 5,505 data points, and 6 frequency domain features extracted from each data point.

[0119] Frequency domain features may include frequency center, mean square frequency, root mean square frequency, root variance frequency, spectral skewness, spectral kurtosis, spectral entropy, etc.

[0120] The frequency center is the center of mass of the spectrum and is the frequency at which signal energy is mainly distributed; the mean square frequency is the average of the squared values ​​of the frequency components and is the energy distribution of the signal; the root mean square frequency is the square root of the mean square frequency and is the central frequency of the signal energy; the root variance frequency is the square root of the variance of the frequency components and is the degree of spreading of the signal bandwidth; spectral skewness is the asymmetry of the spectral distribution and is the degree to which signal energy is skewed toward low or high frequencies; spectral kurtosis is the degree of peaking of the spectral distribution and is the degree to which energy is concentrated in a specific frequency band; and spectral entropy is the disorder or uncertainty of the spectrum and is the complexity and diversity of the signal.

[0121]

[0122] FIG. 8 is a diagram showing the center of optimal data through the silhouette method and the elbow method according to one embodiment of the present invention.

[0123] Referring to FIG. 8, step S415 of FIG. 4 shows that the outlier detection device finds the number of optimal data centroids through the silhouette method and the elbow method.

[0124] Referring to Figure 8 (a), the result using the silhouette method is shown, where the x-axis represents the number of clusters (Number of Clusters (k)) and the y-axis represents the silhouette score.

[0125] The silhouette coefficient can be expressed by the following mathematical formula 1.

[0126] [Mathematical Formula 1]

[0127]

[0128] Here, is the silhouette coefficient, is the density within the cluster and is the average distance to all other points within the same cluster, and is the separation between clusters and is the distance to the nearest cluster among the average distances to all points in other clusters.

[0129] Referring to Figure 8(b), the results using the elbow method are shown, where the x-axis represents the number of clusters (Number of Clusters (k)) and the y-axis represents inertia (sum of squares of distances between the data and the cluster centers).

[0130] The elbow method can find the optimal number of clusters k for clustering by utilizing Inertia (SSE, Sum of Squared Errors), and can be expressed by Equation 2 below.

[0131] [Mathematical Formula 2]

[0132]

[0133] Here, is the number of clusters, Is centroid of the nth cluster is a cluster Data points belonging to , is a data point and cluster center It is the distance between (Euclidean distance).

[0134]

[0135] FIGS. 9 to 15 are drawings showing the results of applying the K-means method and PCA according to an embodiment of the present invention.

[0136] Referring to Figures 9 through 15, the red circles represent outliers and the blue circles represent extreme outliers.

[0137]

[0138] Referring to Figures 9 through 12, each point is the result of PCA after converting the original data into a time domain feature vector.

[0139] For example, the outlier detection device (10) reduces 15 dimensions to 2 dimensions.

[0140]

[0141] Figure 9 is a diagram showing the results of vibration data collected from an x-axis vibration sensor attached to a developing pump motor in the developing process.

[0142] Referring to Figure 9, the centroid of the data is 6 and the threshold multiplier is 7.

[0143]

[0144] Figure 10 is a diagram showing the results of vibration data collected from a z-axis vibration sensor attached to a developing pump motor in the developing process.

[0145] Referring to Fig. 10, the centroid is 2 and the threshold multiplier is 7.

[0146]

[0147] Figure 11 is a diagram showing the results of vibration data collected from an x-axis vibration sensor attached to an Air SQ motor in a drying process.

[0148] Referring to Fig. 11, the centroid is 2 and the threshold multiplier is 7.

[0149]

[0150] Figure 12 is a diagram showing the results of vibration data collected from a y-axis vibration sensor attached to an Air SQ motor in a drying process.

[0151] Referring to Fig. 12, the centroid is 5 and the threshold multiplier is 7.

[0152]

[0153] Referring to Figures 13 to 15, each point is the result of PCA after converting the original data into a frequency domain feature vector.

[0154] For example, the outlier detection device (10) reduces six dimensions to two dimensions.

[0155]

[0156] Figure 13 is a diagram showing the results of vibration data collected from an x-axis vibration sensor attached to an etching pump motor of an etching process.

[0157] Referring to Fig. 13, the centroid is 2 and the threshold multiplier is 7.

[0158]

[0159] Figure 14 is a diagram showing the results of vibration data collected from a z-axis vibration sensor attached to an etching pump motor of an etching process.

[0160] Referring to Fig. 14, the centroid is 2 and the threshold multiplier is 7.

[0161]

[0162] Figure 15 is a diagram showing the results of vibration data collected from an x-axis vibration sensor attached to a stripping pump motor of a stripping process.

[0163] Referring to Fig. 15, the centroid is 2 and the threshold multiplier is 7.

[0164]

[0165] FIGS. 16 to 22 are graphs visualizing a source of extreme outliers according to an embodiment of the present invention.

[0166] Referring to FIGS. 16 to 22, extreme outliers are shown in blue in FIGS. 9 to 15.

[0167] Referring to FIGS. 16 to 22, Acceleration is acceleration and Time is time.

[0168] The time taken to detect outliers is 8.8 seconds.

[0169] That is, the outlier detection device (10) takes 34.2 seconds to detect outliers in about 6 months' worth of data collected from one sensor, which is less than 35 seconds.

[0170] For example, the outlier detection device (10) takes about 23 minutes when using 39 sensors.

[0171] That is, the outlier detection device (10) analyzes the collected time series data using an outlier detection method and can detect outliers (abnormal data) that deviate from the normal range based on the temporal flow of the data.

[0172]

[0173] Figure 16 is a graph visualizing the source of extreme outliers, which are the blue cause, among the results of vibration data collected from the x-axis vibration sensor attached to the developing pump motor of the developing process of Figure 9.

[0174]

[0175] Figure 17 is a graph visualizing the source of extreme outliers, which are the blue cause, among the results of vibration data collected from the z-axis vibration sensor attached to the developing pump motor of the developing process of Figure 10.

[0176]

[0177] Figure 18 is a graph visualizing the source of extreme outliers, represented by the blue cause, among the results of vibration data collected from the x-axis vibration sensor attached to the Air SQ motor of the Dry process of Figure 11.

[0178]

[0179] Figure 19 is a graph visualizing the source of extreme outliers, represented by the blue cause, among the results of vibration data collected from the y-axis vibration sensor attached to the Air SQ motor of the Dry process of Figure 12.

[0180]

[0181]

[0182] Figure 20 is a graph visualizing the source of extreme outliers, represented by the blue cause, among the results of vibration data collected from the x-axis vibration sensor attached to the etching pump motor of the etching process of Figure 13.

[0183]

[0184] Figure 21 is a graph visualizing the source of extreme outliers, represented by the blue cause, among the results of vibration data collected from the z-axis vibration sensor attached to the etching pump motor of the etching process of Figure 14.

[0185]

[0186] Figure 22 is a graph visualizing the source of extreme outliers, represented by the blue cause, among the results of vibration data collected from the x-axis vibration sensor attached to the stripping pump motor of the stripping process of Figure 15.

[0187]

[0188]

[0189] The apparatus and method according to the embodiments of the present invention described above may be implemented as computer-readable code on a computer-readable medium. The computer-readable recording medium may be, for example, a removable recording medium (CD, DVD, Blu-ray disc, USB storage device, removable hard disk) or a fixed recording medium (ROM, RAM, computer-equipped hard disk). The computer program recorded on the computer-readable recording medium may be transmitted to another computing device via a network such as the Internet and installed on the other computing device, thereby being used on the other computing device.

[0190] Although it has been described above that all components constituting an embodiment of the present invention are combined or operate as a single unit, the present invention is not necessarily limited to such an embodiment. That is, within the scope of the purpose of the present invention, all components may be selectively combined in one or more ways to operate.

[0191] Although operations are depicted in a specific order in the drawings, it should not be understood that the operations must necessarily be executed in the specific order depicted or in a sequential order, or that all depicted operations must be executed to obtain the desired result. In certain situations, multitasking and parallel processing may be advantageous. Furthermore, the separation of various configurations in the embodiments described above should not be understood as necessarily required, and it should be understood that the described program components and systems can generally be integrated together into a single software product or packaged into multiple software products.

[0192] The present invention has been described above with reference to its embodiments. Those skilled in the art will understand that the present invention may be embodied in modified forms without departing from the essential characteristics of the invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the invention is defined by the claims, not by the foregoing description, and all variations within the scope of the claims should be interpreted as being included in the invention.

[0193]

[0194] The modes for carrying out the invention are described together in the best mode for carrying out the invention above.

[0195]

[0196] The present invention relates to an outlier detection device and method, and more specifically, to an outlier detection device and method using feature vector clustering. Since it can be used in various ways during predictive maintenance, it has industrial applicability.

Claims

1. In an outlier detection device, A data collection unit that collects data; A data preprocessing unit that preprocesses the above data; A feature extraction unit that extracts a feature vector from the above data; and An anomaly detection device comprising an anomaly detection unit that detects anomalies using the above feature vector.

2. In Paragraph 1, An outlier detection device further comprising a result analysis unit for visualizing results.

3. In Paragraph 1, The above data collection unit is an outlier detection device that collects vibration data collected from a process line.

4. In Paragraph 3, The above process line is an outlier detection device comprising at least one of a developing process, an etching process, a stripping process, and a drying process.

5. In Paragraph 4, The above etching process is an outlier detection device comprising at least one of an etching pump, an etching circulation pump, an HCl pump, and a NaClO3 pump.

6. In Paragraph 4, The above drying process is an outlier detection device comprising at least one of an Air SQ motor, an Air Cut motor, and a Hot Air motor.

7. In an outlier detection method performed by an outlier detection device, Data collection step; A step of preprocessing the above data; A step of extracting a feature vector from the above data; and An outlier detection method comprising the step of detecting outliers using the above feature vector.

8. In Paragraph 7, The step of collecting the above data Step of collecting vibration data of the process line; and An outlier detection method comprising the step of integrating data collected from the same sensor at the same location.

9. In Paragraph 7, The step of detecting outliers using the above feature vector is Step to find the center of the data; A step of calculating the distance between each data point from the center of the above data; and An outlier detection method comprising the step of determining a feature vector exceeding a certain distance value from the center of the data as an outlier.

10. In Paragraph 7, An outlier detection method that includes an additional step of visualizing results.

Images