KR102566876B1 - System and method for detecting anomaly in air drying equipment for radar system using unsupervised learning method - Google Patents

Abstract

The present invention relates to a technology for detecting anomalies in air drying equipment for a radar system, and more particularly, to obtain temperature and humidity data of the air drying equipment for a radar system using a sensor, and to obtain the temperature in a normal state. and an abnormality detection system and method for air drying equipment for a radar system using an unsupervised learning method for discriminating different abnormal states by learning changes in humidity data.

Description

Anomaly detection system and method of air drying equipment for radar system using unsupervised learning method

The present invention relates to a technology for detecting anomalies in air drying equipment for a radar system, and more particularly, to obtain temperature and humidity data of the air drying equipment for a radar system using a sensor, and to obtain the temperature in a normal state. and an abnormality detection system and method for air drying equipment for a radar system using an unsupervised learning method for discriminating different abnormal states by learning changes in humidity data.

Periodic inspections are conducted to detect component failures in the operation maintenance stage of the weapon system currently in operation. These inspection methods include visual inspection using the user's eyes and signal inspection using test equipment. etc. In particular, signal inspection through test equipment utilizes the BIT (Built In Test) function built into the product itself, and accounts for most of the activities for detecting failures of weapon systems.

However, it has been developed or introduced a long time ago, or in the case of analog equipment, the above BIT function is not designed. For this reason, it relies on a failure exclusion procedure in which a user's visual inspection or a set procedure is performed to estimate the failure of the equipment. The visual inspection and failure exclusion method utilizes a technical manual published during weapon system development, and is maintained and managed under the influence of the operator's experience or propensity during operation.

In weapon systems currently under development or in operation, preventive maintenance, which uses self-failure diagnosis and maintenance equipment such as BIT, to determine whether or not there is a failure, and post-maintenance, which performs maintenance after a failure is identified. These methods have limitations in predicting failures that will occur during mission performance. Due to this, if a failure occurs during the mission, a power vacuum in the military may occur, resulting in fatal results. Therefore, it is necessary to carry out maintenance as quickly as possible, and for this purpose, an appropriate amount of replacement equipment, maintenance tools, repair parts and fixtures should be stored.

However, in the case of expensive equipment for rapid maintenance, it is difficult to secure sufficient replacement equipment, and in the case of repair parts and fixtures, failure response is often limited due to low prediction accuracy. In the end, because the system for managing failure information and repair parts is insufficient, there is room for controversy among stakeholders due to poor utilization or inconsistency in actual situations. Therefore, it is necessary to apply CBM+ technology (condition-based maintenance), a tool that can detect abnormal conditions of weapon system components and predict remaining life. CBM+ is a maintenance strategy that adds RUL (Remaining Useful Life) predictive technology to the CBM (Condition Based Maintenance) concept that identifies changes in equipment status information as signs of failures and defects.

KR 10-2409412 B1, 2022. 06. 10. KR 10-2409411 B1, 2022. 06. 10.

An object of the present invention is to detect abnormalities in air drying equipment for a radar system that can detect abnormal signs of weapon system components in advance using temperature and humidity data before failure (normal state) of weapon system components (air drying equipment). To provide systems and methods.

An abnormality detection system of an air drying equipment for a radar system using an unsupervised learning method according to an embodiment of the present invention is installed in the air drying equipment for a radar system and measures the temperature and humidity of the air drying equipment, respectively, to measure the air drying equipment. a sensor that provides temperature and humidity data; and collecting and storing the temperature and humidity data of the air drying equipment provided by the sensor in a database, and detecting changes in the temperature and humidity data of the air drying equipment in a normal state among the temperature and humidity data of the air drying equipment. It includes a web server that detects anomalies in the air drying equipment by using an anomaly detection model using an unsupervised learning methodology that learns and detects an anomaly when it is different from a normal state.

In addition, the abnormality detection model may include building a data set of an operating section of the air drying equipment; extracting normal data by removing outlier data from the constructed data set; The process of selecting key variables; The process of obtaining training and test data sets; And it can be obtained through a process of selecting a final anomaly detection model.

In addition, the process of constructing a data set in the operating section of the air drying equipment may include separating operation and non-operation time points of the air drying equipment using a gradient of internal humidity among the temperature and humidity data collected through the sensor; generating a derived variable of the internal humidity gradient and designating a gradient reference value for operation start/end for each air drying equipment; and constructing a data set by merging in chronological order only data at operation time separated based on the slope of the internal humidity.

In addition, in the process of separating the operation and non-operation time of the air drying equipment, the operation and non-operation time of the air drying equipment may be separated by determining based on the change rate per minute of the internal humidity.

In addition, in the process of generating the slope derived variable of the internal humidity and designating the slope reference value for starting/stopping operation of each air drying equipment, the slope condition of the internal humidity graph of the air drying equipment is applied to start/stop operation of each air drying equipment. You can specify the slope reference value of

In addition, the process of constructing a data set by combining only the data at the time of operation of the air drying equipment separated based on the slope of the internal humidity in chronological order extracts only the data at the time of operation of the air drying equipment, combines them in chronological order, and then Data sets can be built from time periods.

delete

In addition, in the process of selecting the main variables, the main variables to be included in the final anomaly detection model selection experiment were reviewed based on the failure mechanism relevance, data reliability, and correlation between variables, and two variables, internal temperature and internal humidity, were selected. can do.

In addition, in the process of securing the learning and test data sets, the learning data set in the normal section of the air drying equipment and the normal section and abnormal section of the air drying equipment for performance verification are used to build the abnormality detection model. You can build a test data set that includes

In addition, in the process of selecting the final anomaly detection model, it may be selected according to a local outlier factor (LOF)-based failure diagnosis model performance criterion.

In addition, a method for detecting anomaly in an air drying equipment for a radar system using an unsupervised learning method according to another embodiment of the present invention uses temperature and humidity data of the air drying equipment measured from a sensor installed in the air drying equipment for a radar system. collection process; obtaining an abnormality detection model to be used for abnormality detection of the air drying equipment based on temperature and humidity data of the air drying equipment; A process of selecting the most suitable final anomaly detection model from the obtained anomaly detection model; and defining an anomaly detection alert criterion to be applied to the selected final anomaly detection model.

In addition, the process of securing an anomaly detection model to be used for detecting anomaly of the air drying equipment based on the temperature and humidity data of the air drying equipment may include constructing a data set of an operating section of the air drying equipment; extracting normal data by removing outlier data from the constructed data set; The process of selecting key variables; The process of obtaining training and test data sets; And it can be obtained through a process of selecting a final anomaly detection model.

In addition, the process of constructing a data set in the operating section of the air drying equipment may include separating operation and non-operation time points of the air drying equipment using a gradient of internal humidity among the temperature and humidity data collected through the sensor; generating a derived variable of the internal humidity gradient and designating a gradient reference value for operation start/end for each air drying equipment; and constructing a data set by merging in chronological order only data at operation time separated based on the slope of the internal humidity.

In addition, in the process of separating the operation and non-operation time of the air drying equipment, the operation and non-operation time of the air drying equipment may be separated by determining based on the change rate per minute of the internal humidity.

In addition, in the process of generating the slope derived variable of the internal humidity and designating the slope reference value for starting/stopping operation of each air drying equipment, the slope condition of the internal humidity graph of the air drying equipment is applied to start/stop operation of each air drying equipment. You can specify the slope reference value of

In addition, the process of constructing a data set by combining only the data at the time of operation of the air drying equipment separated based on the slope of the internal humidity in chronological order extracts only the data at the time of operation of the air drying equipment, combines them in chronological order, and then Data sets can be built from time periods.

delete

In addition, in the process of selecting the main variables, the main variables to be included in the final anomaly detection model selection experiment were reviewed based on the failure mechanism relevance, data reliability, and correlation between variables, and two variables, internal temperature and internal humidity, were selected. can do.

In addition, in the process of securing the learning and test data sets, the learning data set in the normal section of the air drying equipment and the normal section and abnormal section of the air drying equipment for performance verification are used to build the abnormality detection model. You can build a test data set that includes

In addition, in the process of selecting the final anomaly detection model, it may be selected according to a local outlier factor (LOF)-based failure diagnosis model performance criterion.

Also, in the process of defining the criteria for anomaly detection alert to be applied to the selected final anomaly detection model, an optimal threshold value for anomaly detection alert and a criterion for a continuous occurrence time of the anomaly detection alert may be respectively set.

According to an embodiment of the present invention, internal temperature and humidity data of the air drying equipment are obtained using a sensor, and unsupervised learning (unsupervised If it is different from normal, it is detected as an anomaly using the learning) method, so it is possible to detect equipment anomalies with high accuracy even in situations where unidentified failure types or failure data collection are impossible.

In addition, according to an embodiment of the present invention, by detecting anomalies in advance using temperature and humidity data before failure with respect to weapon system components (air drying equipment for radar systems), the weapon system components for which abnormal signs are detected are quickly Can be provided for maintenance.

1 is a view showing an anomaly detection system of air drying equipment for a radar system using an unsupervised learning method according to an embodiment of the present invention.
2 is a view showing a state in which sensors are installed in the target equipment shown in FIG. 1;
3 is a diagram illustrating an anomaly detection model applied to an embodiment of the present invention;
4 is a diagram illustrating an LOF-based failure diagnosis model applied to an embodiment of the present invention.
5 is a diagram illustrating a process of defining anomaly detection alarm criteria of a final anomaly detection model applied to an embodiment of the present invention;
6 is a flowchart illustrating a method for detecting anomaly in air drying equipment for a radar system using an unsupervised learning method according to an embodiment of the present invention.
7 is a flowchart illustrating a process of securing the anomaly detection model shown in FIG. 6;
FIG. 8 is a view showing a process of constructing a data set of an operating section of the air drying equipment shown in FIG. 7;
9 is a diagram illustrating a process of extracting normal data by removing outlier data from the data set shown in FIG. 7;
10 is a diagram showing a process of selecting the main variables shown in FIG. 7;
11 is a diagram illustrating a process of securing training and test data sets shown in FIG. 7;
12 is a diagram illustrating a process of selecting a final anomaly detection model shown in FIG. 6;

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention will not be limited to the embodiments disclosed below, but will be implemented in various different forms, only the embodiments of the present invention will make the disclosure of the present invention complete, and will make the scope of the invention clear to those skilled in the art. It is provided to fully inform you. Like reference numerals designate like elements in the drawings.

In the method of determining the failure type for the weapon system component, for example, the method of learning the temperature and humidity patterns before the failure of the equipment can identify the failure type for each equipment. However, the present invention provides a method for determining an abnormal situation when it is different from normal by learning a normal operation. This method can detect equipment abnormalities with high accuracy even in situations where it is impossible to collect failure data or unidentified failure types.

1 is a diagram schematically shown to explain an anomaly detection system of air drying equipment for a radar system using an unsupervised learning method according to an embodiment of the present invention.

Referring to FIG. 1, an abnormality detection system of air drying equipment for a radar system according to an embodiment of the present invention includes a sensor 10 installed in a target equipment 1 for data measurement and collection to collect temperature and humidity data, It receives the temperature and humidity data of the target equipment (1) measured and collected through the sensor (10) and learns changes in the temperature and humidity data in the normal state through the anomaly detection model to determine the abnormal state of the target equipment (1). It includes a web server (20).

FIG. 2 is a diagram showing a state in which a sensor is installed in the target equipment shown in FIG. 1. (a) shows a state in which the sensor is not installed in the target equipment, and (b) shows a state in which the sensor is installed.

As shown in FIG. 2, the sensor 10 is installed inside the target equipment 1 to measure temperature and humidity data of the target equipment 1 over time. The web server 20 collects changes in temperature and humidity data in a steady state using the temperature and humidity data provided from the sensor 10 . At this time, the target equipment (1) is an air drying equipment, and the air drying equipment plays a role of controlling the humidity of the upper system by operating two drying boxes in the equipment crosswise every 2 hours.

The web server 20 can be driven by a drive terminal 30 such as a desktop or laptop, and is interlocked with the database 21 to obtain temperature and humidity data measured by the target equipment 1, temperature and changes in humidity data are registered, updated, and managed.

The temperature and humidity data measured by the sensor 10 are directly provided to the web server 20 through a communication connection between the sensor 10 and the web server 20, or provided to the web server 20 through a separate interface. It can be. In addition, it may be provided through various known communications.

Such a web server 20 may be installed with software that performs an anomaly detection model in order to detect an abnormal operation of the target equipment 1 . Software for performing the anomaly detection model may be registered and stored in the database 21 linked with the web server 20 .

Anomaly detection for detecting abnormal (abnormal) operation of the target equipment 1 can be performed using an unsupervised learning methodology that detects abnormalities by learning the normal state of the target equipment 1. can Unlike classification, which is a supervised learning technique, the unsupervised learning methodology does not require data labels and is suitable for outlier detection in situations where it is difficult to obtain a failure history.

3 is a diagram schematically illustrating an anomaly detection model applied to an embodiment of the present invention.

As shown in FIG. 3 , in the anomaly detection model applied to the embodiment of the present invention, a normal section is learned by an algorithm and detected as an anomaly when it is different from the information of the learned normal section. For example, in the abnormality detection model, when the operation of the air drying equipment, which is the target equipment 1, is in an operating (ON) state, normal and abnormal (faulty) states of the air drying equipment may be detected.

4 is a diagram schematically illustrating an LOF-based failure diagnosis model applied to an embodiment of the present invention.

Referring to FIG. 4, the LOF (Local Outlier Factor)-based failure diagnosis model is an anomaly detection model that classifies normal and failure data based on density. Considering the density of the normal region, the local outlier factor (LOF)-based failure diagnosis model is relative to the observed value near the high-density region. Normal and faulty data are distinguished by using the fact that the outlier detection criteria for observations in the vicinity of the low area are different.

5 is a diagram schematically illustrating a process of defining an anomaly detection alarm criterion of a final anomaly detection model applied to an embodiment of the present invention, and an optimal threshold value for an anomaly detection alarm and anomaly detection It shows the process of setting the frequency of continuous occurrence of alarms.

Referring to FIG. 5 , an optimal threshold value capable of properly distinguishing between a normal state and an abnormal state of the target equipment 1 is set. For example, the optimal threshold value may be set using a bootstrap technique. The bootstrap technique is a technique that calculates the average value of the statistic to be obtained after creating several sample groups through extraction with replacement from the original sample, and can prevent overfitting or a small data set.

The optimal threshold setting method for the target equipment 1, for example, as shown in FIG. 5, measures the temperature and humidity data from the target equipment 1 over time through the sensor 10, and uses the measured data After calculating the anomaly scores of the steady state, the average value of the anomaly scores of the top 99% positions for each batch is set as the optimal threshold value.

Then, a criterion for a continuous occurrence time for an anomaly detection alarm is set. In order to prevent a false alarm, a continuous occurrence time criterion for determining how many times a value equal to or higher than the optimal threshold value occurs in succession is set as a failure. For example, if a value above the optimal threshold is detected more than N (where N is a natural number) consecutively, a warning is issued.

As shown in FIG. 5, when there is no continuous outlier standard (continuous occurrence time standard) for anomaly detection alarm, an error alarm may be generated even in a normal section, while setting the continuous outlier standard for anomaly detection alarm to '10' In this case, it can be confirmed that no error alarm is generated in the normal section. Therefore, it is possible to adjust the false alarm rate in the normal state and the failure state detection rate by adjusting the continuous outlier criteria for the anomaly detection alarm.

6 is a schematic flowchart to explain a method for detecting an anomaly in air drying equipment for a radar system using an unsupervised learning method according to an embodiment of the present invention.

Referring to FIG. 6 , a method for detecting anomaly in air drying equipment for a radar system using an unsupervised learning method according to an embodiment of the present invention detects temperature and The process of collecting humidity data (S1), the process of securing an anomaly detection model to be used for detecting anomalies in air drying equipment (S2), and the process of selecting the most appropriate final anomaly detection model from the obtained anomaly detection models (S3) and a process of defining criteria for an anomaly detection alert to be applied to the selected final anomaly detection model (S4).

In the process of collecting temperature and humidity data of the air drying equipment (S1), as shown in FIGS. 1 and 2, the temperature and humidity data are collected through the sensor 10 installed in the air drying equipment.

The process of securing an anomaly detection model (S2) can be secured through the same process as shown in FIG.

7 is a flowchart schematically illustrating a process of securing the anomaly detection model shown in FIG. 6 .

Referring to FIG. 7, the process of securing an anomaly detection model according to the present invention (S2) includes the process of constructing a data set of the operating section of the air drying equipment (S21), and removing outlier data from the built data set to obtain normal data It includes a process of extracting (S22), a process of selecting main variables (S23), and a process of securing training and test data sets (S24).

FIG. 8 is a diagram schematically illustrating a process of constructing a data set of an operating section of the air drying equipment shown in FIG. 7, and the dryer described in FIG. 8 represents the air drying equipment.

As shown in FIG. 8, the process of building a data set in the operating section of the air drying equipment (S21) is the operation and operation of the air drying equipment using the slope of the internal humidity among the temperature and humidity data collected through the data collection process (S1) The process of separating the non-operation (ON/OFF) time, the process of creating a derived variable of the internal humidity gradient and specifying the slope reference value for the start/end of operation for each air drying equipment, and the operation separated based on the slope of the internal humidity It includes the process of building a data set by merging only the data at the point in time in chronological order.

The internal humidity gradient mentioned in the process of separating the operation and non-operation time of the air drying equipment corresponds to the change rate per minute of the internal humidity. Therefore, as shown in FIG. 8, in the process of separating the operation and non-operation time of the air drying equipment, the operation and non-operation time of the air drying equipment can be separated based on the change rate per minute of the internal humidity.

In the process of generating the slope variable of the internal humidity and designating the slope reference value for starting/stopping the operation of each air drying equipment, the slope condition of the internal humidity graph of the air drying equipment is applied to designate the slope reference value for starting/stopping the operation of each air drying equipment. can At this time, the slope condition of the internal humidity graph can be represented by [Equation 1] below.

[Equation 1]

Starting point of air dryer operation:

At the end of operation of the air drying equipment:

here, indicates

The process of constructing a data set by combining only the data at the time of operation of the air drying equipment separated based on the slope of the internal humidity in chronological order extracts only the data at the time of operation of the air drying equipment and combines them in chronological order, and then data from the entire period build three

9 is a diagram schematically illustrating a process of extracting normal data by removing outlier data from the data set shown in FIG. 7, and the dryer represents air drying equipment.

As shown in FIG. 9, the process of extracting normal data by removing outlier data from the data set constructed in the same way as in FIG. 8 (22) is a process for improving the quality of the normal data set, and is a technique for learning the normal data set Anomaly detection is used. Since the quality of a normal data set is proportional to the quality of an anomaly detection model, the quality of an anomaly detection model can be improved by removing outlier data in the data set.

In the process of extracting normal data by removing outlier data from the built data set (22), for example, outlier data (hunting) due to replacement of the drying container and abnormality of the sensor 10 at the time of operation of the air drying equipment is removed.

10 is a diagram schematically showing a process of selecting the main variables shown in FIG. 7, and the air dryer described in FIG. 10 represents air drying equipment.

As shown in FIG. 10, in the process of selecting the main variables (S23), the main variables to be included in the final anomaly detection model selection experiment are reviewed based on the failure mechanism relevance, data reliability, and correlation between variables, respectively, and the internal temperature and internal humidity of 2 Select species variables.

In relation to the failure mechanism, external temperature and humidity are environmental factors that cannot represent failures inside the air drying equipment, and are excluded when selecting major variables. Regarding the reliability of the above data, as external temperature data of 30°C or higher was observed in March and April, major variables were excluded due to deterioration in data reliability. Regarding the correlation between the variables, external humidity is a factor that lowers the analysis performance due to its low correlation coefficient with other variables, so it is excluded when selecting the main variable. Therefore, in the process of selecting the main edges (S23), two variables of internal temperature and internal humidity are selected as the main variables.

FIG. 11 is a diagram schematically illustrating a process of securing training and test data sets shown in FIG. 7 .

As shown in FIG. 11, in the process of securing a training data set and a test data set (S24), a training data set in a normal interval for building an anomaly detection model is built, and a test including a normal interval and an abnormal interval for performance verification build a data set

The anomaly detection model learns normal intervals and then verifies anomaly and steady state detection. In order to verify the performance of the anomaly detection model, the anomaly section data history is required. For example, a test data set is built using three failure data in 2021. Failure types include defective drying box (moisture absorption is not possible due to defective desiccant, cam motor failure (drying box cannot be replaced due to cam motor failure), and heater OFF (drying agent inside the drying box cannot be dried).

FIG. 12 is a diagram schematically illustrating a process of selecting a final anomaly detection model shown in FIG. 6 .

As shown in FIG. 12 , in the process of selecting the final abnormality detection model (S3), the final abnormality detection model is selected according to the criteria for determining the performance of the LOF (Local Outlier Factor) based failure diagnosis model shown in FIG. 4 . For example, a failure diagnosis model includes a model that detects a normal section as normal and a model that detects an abnormal section as a failure. A model that detects a normal section as normal may derive a low anomaly score, and a model that detects an anomaly section as a failure may derive a higher anomaly score than the normal section.

When the final abnormality detection model is selected, abnormality detection alarm criteria are defined as shown in FIG. 6 (S4). In this process, as shown in FIG. 5 , an optimal threshold value for an anomaly detection alarm and a frequency of continuous generation of the anomaly detection alarm are respectively set.

In the above, although preferred embodiments of the present invention have been described and illustrated using specific terms, such terms are only intended to clearly explain the present invention, and the embodiments and described terms of the present invention are the technical spirit of the following claims. And it is obvious that various changes and changes can be made without departing from the scope. Such modified embodiments should not be individually understood from the spirit and scope of the present invention, and should be said to fall within the scope of the claims of the present invention.

1: Target equipment (air drying equipment)
10: sensor
20: web server
21: database
30: drive terminal

Claims (21)

a sensor installed in the air drying equipment for a radar system to measure the temperature and humidity of the air drying equipment and provide temperature and humidity data of the air drying equipment; and
The temperature and humidity data of the air drying equipment provided from the sensor are collected and stored in a database, and among the temperature and humidity data of the air drying equipment, changes in the temperature and humidity data in the normal state of the air drying equipment are learned. A web server that detects anomalies in the air drying equipment using an anomaly detection model using an unsupervised learning methodology that detects an anomaly when it is different from a normal state;
The anomaly detection model,
building a data set of an operating section of the air drying equipment;
extracting normal data by removing outlier data from the constructed data set;
The process of selecting key variables;
The process of obtaining training and test data sets; and
process of selecting a final anomaly detection model; obtained through
The process of building a data set in the operating section of the air drying equipment,
Separating operation and non-operation time points of the air drying equipment using a gradient of internal humidity among temperature and humidity data collected through the sensor;
generating a derived variable of the internal humidity gradient and designating a gradient reference value for operation start/end for each air drying equipment; and
constructing a data set by merging, in chronological order, only data at operation time separated based on the slope of the internal humidity;
Anomaly detection system of air drying equipment for radar system using unsupervised learning method including.
delete delete According to claim 1,
In the process of separating the operation and non-operation time of the air drying equipment, air drying equipment for a radar system using an unsupervised learning method that determines and separates the operation and non-operation time of the air drying equipment based on the rate of change per minute of the internal humidity. anomaly detection system.
According to claim 1,
In the process of generating the slope derived variable of the internal humidity and specifying the slope reference value of the operation start/stop of each air drying equipment, the slope condition of the internal humidity graph of the air drying equipment is applied to determine the operation start/end slope of each air drying equipment. An anomaly detection system for air drying equipment for radar systems using an unsupervised learning method that specifies a reference value.
According to claim 1,
The process of constructing a data set by combining only the data at the time of operation of the air drying equipment separated based on the slope of the internal humidity in chronological order extracts only the data at the time of operation of the air drying equipment and combines them in chronological order, and then Anomaly detection system of air drying equipment for radar system using unsupervised learning method to build data set.
delete According to claim 1,
In the process of selecting the main variables, the main variables to be included in the final anomaly detection model selection experiment are reviewed on the basis of failure mechanism relevance, data reliability, and correlation between variables, and two types of variables, internal temperature and internal humidity, are selected. Anomaly detection system of air drying equipment for radar system using degree learning method.
According to claim 1,
In the process of securing the learning and test data sets, a test including a learning data set in the normal section of the air drying equipment and a normal section and an abnormal section of the air drying equipment for performance verification to build the abnormality detection model Anomaly detection system of air drying equipment for radar system using unsupervised learning method to build data set.
a sensor installed in the air drying equipment for a radar system to measure the temperature and humidity of the air drying equipment and provide temperature and humidity data of the air drying equipment; and
The temperature and humidity data of the air drying equipment provided from the sensor are collected and stored in a database, and among the temperature and humidity data of the air drying equipment, changes in the temperature and humidity data in the normal state of the air drying equipment are learned. A web server that detects anomalies in the air drying equipment using an anomaly detection model using an unsupervised learning methodology that detects an anomaly when it is different from a normal state;
The anomaly detection model,
building a data set of an operating section of the air drying equipment;
extracting normal data by removing outlier data from the constructed data set;
The process of selecting key variables;
The process of obtaining training and test data sets; and
The process of selecting the final anomaly detection model; obtained through,
In the process of selecting the final anomaly detection model, an anomaly detection system of air drying equipment for a radar system using an unsupervised learning method selected according to LOF (Local Outlier Factor)-based failure diagnosis model performance criteria.
Claims 1, 4 to 6, and 8 to 10 of any one of the unsupervised learning method using the air drying equipment for the radar system using the abnormality detection system of the air drying equipment for the radar system As an anomaly detection method,
collecting temperature and humidity data of the air drying equipment measured by the sensor;
obtaining an anomaly detection model to be used for detecting anomaly of the air drying equipment based on temperature and humidity data of the air drying equipment through the web server;
selecting the most suitable final anomaly detection model from the obtained anomaly detection models through the web server; and
A process of defining an anomaly detection alert criterion to be applied to a selected final anomaly detection model through the web server;
The process of securing an abnormality detection model to be used for abnormality detection of the air drying equipment based on the temperature and humidity data of the air drying equipment,
building a data set of an operating section of the air drying equipment;
extracting normal data by removing outlier data from the constructed data set;
The process of selecting key variables;
The process of obtaining training and test data sets; and
The process of selecting the final anomaly detection model; obtained through,
The process of building a data set in the operating section of the air drying equipment,
Separating operation and non-operation time points of the air drying equipment using a gradient of internal humidity among temperature and humidity data collected through the sensor;
generating a derived variable of the internal humidity gradient and designating a gradient reference value for operation start/end for each air drying equipment; and
constructing a data set by merging, in chronological order, only data at operation time separated based on the slope of the internal humidity;
Anomaly detection method of air drying equipment for radar system using unsupervised learning method including.
delete delete According to claim 11,
In the process of separating the operation and non-operation time of the air drying equipment, air drying equipment for a radar system using an unsupervised learning method that determines and separates the operation and non-operation time of the air drying equipment based on the rate of change per minute of the internal humidity. An anomaly detection method in .
According to claim 11,
In the process of generating the slope derived variable of the internal humidity and specifying the slope reference value of the operation start/stop of each air drying equipment, the slope condition of the internal humidity graph of the air drying equipment is applied to determine the operation start/end slope of each air drying equipment. A method for detecting anomalies in air drying equipment for radar systems using an unsupervised learning method specifying reference values.
According to claim 11,
The process of constructing a data set by combining only the data at the time of operation of the air drying equipment separated based on the slope of the internal humidity in chronological order extracts only the data at the time of operation of the air drying equipment and combines them in chronological order, and then Anomaly detection method of air drying equipment for radar system using unsupervised learning method to build data set.
delete According to claim 11,
In the process of selecting the main variables, the main variables to be included in the final anomaly detection model selection experiment are reviewed on the basis of failure mechanism relevance, data reliability, and correlation between variables, and two types of variables, internal temperature and internal humidity, are selected. Anomaly detection method of air drying equipment for radar system using degree learning method.
According to claim 11,
In the process of securing the learning and test data sets, a test including a learning data set in the normal section of the air drying equipment and a normal section and an abnormal section of the air drying equipment for performance verification to build the abnormality detection model Anomaly detection method of air drying equipment for radar system using unsupervised learning method to build data set.
Claims 1, 4 to 6, and 8 to 10 of any one of the unsupervised learning method using the air drying equipment for the radar system using the abnormality detection system of the air drying equipment for the radar system As an anomaly detection method,
collecting temperature and humidity data of the air drying equipment measured by the sensor;
obtaining an anomaly detection model to be used for detecting anomaly of the air drying equipment based on temperature and humidity data of the air drying equipment through the web server;
selecting the most suitable final anomaly detection model from the obtained anomaly detection models through the web server; and
A process of defining an anomaly detection alert criterion to be applied to a selected final anomaly detection model through the web server;
The process of securing an abnormality detection model to be used for abnormality detection of the air drying equipment based on the temperature and humidity data of the air drying equipment,
building a data set of an operating section of the air drying equipment;
extracting normal data by removing outlier data from the constructed data set;
The process of selecting key variables;
The process of obtaining training and test data sets; and
The process of selecting the final anomaly detection model; obtained through,
In the process of selecting the final anomaly detection model, an anomaly detection method of air drying equipment for a radar system using an unsupervised learning method selected according to LOF (Local Outlier Factor)-based failure diagnosis model performance criteria.
Claims 1, 4 to 6, and 8 to 10 of any one of the unsupervised learning method using the air drying equipment for the radar system using the abnormality detection system of the air drying equipment for the radar system As an anomaly detection method,
collecting temperature and humidity data of the air drying equipment measured by the sensor;
obtaining an anomaly detection model to be used for detecting anomaly of the air drying equipment based on temperature and humidity data of the air drying equipment through the web server;
selecting the most suitable final anomaly detection model from the obtained anomaly detection models through the web server; and
A process of defining an anomaly detection alert criterion to be applied to a selected final anomaly detection model through the web server;
The process of securing an abnormality detection model to be used for abnormality detection of the air drying equipment based on the temperature and humidity data of the air drying equipment,
building a data set of an operating section of the air drying equipment;
extracting normal data by removing outlier data from the constructed data set;
The process of selecting key variables;
The process of obtaining training and test data sets; and
The process of selecting the final anomaly detection model; obtained through,
The process of defining the criteria for the anomaly detection alert to be applied to the selected final anomaly detection model is a radar using an unsupervised learning method that sets the optimal threshold for anomaly detection alert and the standard for the continuous occurrence time of the anomaly detection alert, respectively. A method for detecting anomalies in air drying equipment for systems.