KR102444442B1 - System and method for diagnosing facility fault - Google Patents

Abstract

The present invention relates to an apparatus for diagnosing equipment failure. The equipment failure diagnosis apparatus of the present invention detects equipment failure using sensors that collect data related to operation of equipment, a database that is connected to the sensors through a network and stores sensor data collected from the sensors, and sensor data. A first analysis engine to which a black box method for diagnosis is applied, a second analysis engine to which a white box method for diagnosing equipment failures is applied by linking sensor data with facilities, and a first analysis engine and a second analysis engine, and an interlocking interface for diagnosing equipment failure by combining the first failure diagnosis data output from the first analysis engine and the second failure diagnosis data output from the second analysis engine.

Description

Equipment and method for diagnosing equipment failures

The present invention relates to an apparatus for diagnosing equipment failure, and to an apparatus and method for diagnosing equipment failure managed in a large-scale management system.

With the development of the Internet of Things (IoT), sensors having various types, sizes, and specifications are being developed, and a sensor network technology for using these sensors is also developing. The development of low-power small wireless sensors dramatically improves the number of sensors applicable to target facilities and devices compared to the past.

As the number of sensors that can be installed in each facility increases, additional sensors that have not been installed in the past are also being installed. Such additional sensors include, for example, a temperature and humidity sensor, a vibration sensor, an external barometric pressure sensor, and the like. Although it may be necessary depending on the facility, these sensors of low importance were not installed in the past due to cost, power consumption, communication problems, etc., but now, due to the development of sensor technology along with the development of the Internet of Things, there is no restriction according to the needs of the user. Additional installation is possible.

Considering the aspect of sensor data analysis, in the past when computing power was low, even if the amount of data collected through the installation of many sensors increased, it was not possible to analyze information within a desired time. However, the current amount of data that can be used for big data is not a big problem, and an increase in the amount of data available for analysis may make analysis difficult, but has the advantage of providing more precise analysis.

For this reason, the number of sensors used in a large management system such as a building energy management system or an automatic flight control system increases, and a diagnostic apparatus that more accurately diagnoses a failure of a facility by utilizing the increased sensor data is required.

US6216066B1 (Kai Frank), "system and method for generating alerts through multi-variate data assessment", 2001-04-10. KR101316486B1 (Maeda, ji). "Anomaly detection method and system", 2013-10-08.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an apparatus and method for diagnosing equipment failure that enables accurate equipment failure diagnosis from increased sensors and increased sensor data in a large-scale management system.

Another object of the present invention is to provide an apparatus and method for diagnosing equipment failure based on a dual engine in which a black box analysis engine and a white box analysis engine are simultaneously applied in a large management system.

A facility failure diagnosis apparatus according to the present invention uses sensors for collecting data related to the operation of facilities, a database connected to the sensors through a network, and storing sensor data collected from the sensors, and the sensor data. A first analysis engine to which a black box method for diagnosing a failure of the facility is applied, a second analysis engine to which a white box method for diagnosing a failure of the facility is applied by linking the sensor data with the facility, and the first analysis engine; and an interlocking interface connected to the second analysis engine and diagnosing a failure of the facility by combining the first failure diagnosis data output from the first analysis engine and the second failure diagnosis data output from the second analysis engine do.

In this embodiment, the first analysis engine generates a classification map for error detection by machine learning the sensor data, a learning unit that stores the generated classification map in the database, and a current time from the database and a black box analysis unit for diagnosing a failure by loading the classification map generated within a threshold time as , and determining whether the current sensor data matches the values of the sensor data of the failure state included in the loaded classification map.

In this embodiment, the black box analyzer regenerates the classification map through the learning unit if the classification map generated from the database for less than a threshold time based on the current time does not exist.

In this embodiment, the black box analyzer calculates a distance from the current sensor data to representative values of each group included in the classification map, and the calculated distance does not fall below a threshold value or a predefined error If it is defined as a data group, it is judged as a failure state.

In this embodiment, the black box analyzer outputs the first failure diagnosis data indicating a normal operation or a failure operation according to the diagnosis of the failure.

In this embodiment, the second analysis engine includes a white box analyzer that analyzes the failure using the failure diagnosis logic for the sensor data, and the failure diagnosis logic is configured between data based on information related to the facility. This is the logic that performs linkage analysis.

In this embodiment, the white box analyzer calculates a facility condition index indicating the condition of the facility, uses the facility condition index for the fault diagnosis, and the facility condition index includes the efficiency and load factor of the facility .

In this embodiment, the white box analyzer outputs the second failure diagnosis data including a normal operation or a failure operation according to the diagnosis of the failure and a related description corresponding to the failure operation.

In this embodiment, the interworking interface determines that the first fault diagnosis data indicates a normal operation and the second fault diagnosis data indicates a normal operation as a normal operation, and the first fault diagnosis data indicates a normal operation and the If the second fault diagnosis data indicates a fault operation, a message including a fault alarm and a fault-related description for the fault is output, the first fault diagnosis data indicates a fault operation, and the second fault diagnosis data indicates a normal operation , a message including a failure alarm and a request for confirmation of the failure state is output, and if the first failure diagnosis data indicates a failure operation and the second failure diagnosis data indicates a failure operation, a failure alarm and a failure-related explanation for the failure are displayed. Prints a message including

A facility failure diagnosis method according to the present invention includes the steps of collecting data related to the operation of the facility through sensors, storing the sensor data collected from the sensors, and diagnosing the failure of the facility using the sensor data. The step of diagnosing the failure of the box method, the step of diagnosing the failure of the white box method of diagnosing the failure of the equipment by linking the sensor data with the equipment, and the first failure diagnosis data according to the failure diagnosis of the black box method and outputting the fault diagnosis result of the facility by integrating the second fault diagnosis data according to the white box type fault diagnosis.

In this embodiment, the step of diagnosing the failure of the black box method includes receiving sensor data from sensors connected to the facility, generating a classification map using the sensor data, and generating the classification map Determining whether a threshold time has been exceeded based on the current time point, if the threshold time is not exceeded, diagnosing a failure using the classification map, and if the threshold time is exceeded, past sensor data and current and generating a new classification map by machine learning using sensor data, and generating the first failure diagnosis data through the diagnosis of the failure using the generated classification map.

In this embodiment, generating the classification map includes setting error data and missing data included in the input sensor data as preset default values, pre-processing data that has been processed for the error data and the missing data and generating the classification map using the pre-processed sensor data.

In this embodiment, the generating of the first failure diagnosis data comprises calculating a distance between the current sensor data and representative values of each group included in the classification map, and the calculated distance is less than a threshold value. If not, or if it is defined as a predefined error data group, it is judged as a failure state.

In this embodiment, the step of diagnosing the failure of the white box method includes receiving sensor data from sensors connected to the facility, calculating a facility condition index using the sensor data, and using the sensor data. and generating the second failure diagnosis data obtained by diagnosing the failure using failure diagnosis logic, wherein the failure diagnosis logic is logic for performing linkage analysis between data based on information related to the equipment.

In this embodiment, generating the second failure diagnosis data includes diagnosing the failure using the equipment condition index, and the equipment condition index includes efficiency and load factor of the equipment.

In this embodiment, the generating of the second failure diagnosis data includes generating the second failure diagnosis data including a related description corresponding to the failure when it is determined that the operation of the equipment is a failure. .

In this embodiment, in the step of outputting the fault diagnosis result of the equipment, if the first fault diagnosis data indicates a normal operation and the second fault diagnosis data indicates a normal operation, it is determined as a normal operation, and a message indicating that the normal operation is outputting, when the first failure diagnosis data indicates a normal operation and the second failure diagnosis data indicates a failure operation, outputting a message including a failure alarm and a failure-related description of the failure, the first failure diagnosis outputting a message including a failure alarm and a request for confirmation of a failure state when the data indicates a failure operation and the second failure diagnosis data indicates a normal operation, and the first failure diagnosis data indicates a failure operation, 2 If the fault diagnosis data indicates a fault operation, outputting a message including a fault alarm and a fault-related description for the fault.

According to the present invention, facility failure diagnosis is performed by simultaneously utilizing diagnostic engines to which different methods (for example, a black box method and a white box method) are applied to sensors attached to facilities constituting a large management system. By doing so, it is possible to more accurately diagnose equipment failures by utilizing the increased sensors and the increased sensor data.

1 is a view exemplarily showing an apparatus for diagnosing equipment failure according to the present invention;
FIG. 2 is a diagram illustrating a detailed configuration of the server unit of FIG. 1 by way of example;
3 is a flowchart exemplarily illustrating a diagnostic operation of the first analysis engine of FIG. 1;
4 is a flowchart exemplarily illustrating a diagnostic operation of the second analysis engine of FIG. 1;
Fig. 5 is a diagram showing the operation algorithm of the interworking interface with the first and second analysis engines of Fig. 1;
6 is a graph exemplarily showing a failure diagnosis result of the first analysis engine of FIG. 1, and
7 is a graph exemplarily illustrating a failure diagnosis result of the second analysis engine of FIG. 1 .

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings. It should be noted that in the following description, only parts necessary for understanding the operation according to the present invention are described, and descriptions of other parts will be omitted so as not to obscure the gist of the present invention.

The present invention provides a facility failure diagnosis apparatus for diagnosing facility failures by using a plurality of analysis engines simultaneously in a large-scale management system. At this time, the plurality of analysis engines include a black box method for diagnosing equipment failures using only sensor data obtained from sensors and a white box method for diagnosing equipment failures by linking sensor data obtained from sensors with equipment. do.

1 is a diagram illustrating an apparatus for diagnosing equipment failure according to an exemplary embodiment of the present invention.

Referring to FIG. 1 , the equipment failure diagnosis apparatus 100 includes a sensor unit 110 and a server unit 120 .

The sensor unit 110 includes n sensors 111-11n. The n sensors 111-11n are, for example, connected or attached to each part of the facility 10 of the large-scale management system to detect motion data according to the operation of the facility 10 .

Each of the n sensors 111-11n generates sensor data through operation detection of the facility 10 and outputs the generated sensor data. Each of the n sensors 111-11n outputs sensor data to the gateway 21 or the like. For example, the n sensors 111-11n include a temperature sensor, a pressure sensor, and a flow sensor. In this case, the gateway 21 is connected to the network 20 or is included in the network 20 to output sensor data to the server unit 120 .

The server unit 120 may detect a failure of the facility 10 by using the sensor data output from the sensor unit 110 . The server unit 120 detects a normal operation or a malfunctioning operation, and outputs the detected operation state information of the facility 10 to the user device 30 for monitoring. In this case, the server unit 120 may check the normal or faulty state of the facility 10 with the user device 30 through the network 20 , for example.

The server unit 120 includes a database 121 , a first analysis engine 122 , a second analysis engine 123 , and an interworking interface 124 .

The database 121 is connected to the network 20 to receive sensor data from the sensors 111-11n of the sensor unit 110 and store the received sensor data. The database 121 is received in real time by the first analysis engine 122 and the second analysis engine 123 and outputs updated sensor data. In addition, the database 121 stores data for fault diagnosis, temporarily stores data output through the first analysis engine 122 or the second analysis engine 123 , and the first analysis engine 122 and Data requested from the second analysis engine 123 is provided to the corresponding first analysis engine 122 and the second analysis engine 123 .

The first analysis engine 122 is an engine to which a black box method is applied. The black box method is a method of diagnosing a failure of the equipment 10 . The first analysis engine 122 diagnoses a failure of the facility 10 using the black box method, and outputs the diagnosis result to the interworking interface 124 .

The second analysis engine 123 is an engine to which a white box method is applied. The white box method is a method of diagnosing a failure of the facility 10 in connection with the facility 10 . The second analysis engine 123 diagnoses a failure of the facility 10 using the white box method, and outputs the diagnosis result to the interworking interface 124 .

The interworking interface 124 receives fault diagnosis results from both the first analysis engine 122 and the second analysis engine 123 . The interworking interface 124 determines a failure diagnosis by using the first analysis engine 122 and the second analysis engine 123 at the same time. In the interworking interface 124 , the fault diagnosis results derived using each method are integrated, and the fault diagnosis results are transmitted to the client device 30 to provide the user with the fault diagnosis results through the network 20 .

FIG. 2 is a diagram exemplarily illustrating a detailed configuration of the server unit of FIG. 1 .

Referring to FIG. 2 , the server unit 120 includes a database 121 , a first analysis engine 122 , a second analysis engine 123 , and an interworking interface 124 .

The database 121 receives sensor data that is real-time operation information of the facility 10 .

The first analysis engine 122 includes a learning unit 1221 and a black box analysis unit 1222 . The learning unit 1221 generates a classification map for error detection through machine learning, and stores the generated classification map in the database 121 .

The black box analyzer 1222 receives only normal sensor data through error and omission processing on the input data. The black box analyzer 1222 performs preprocessing of converting error and omission-processed sensor data into data for analysis. The black box analyzer 1222 stores pre-processed data in the database 121 .

The black box analyzer 1222 loads the classification map generated within the threshold time based on the current time from the database 121 . If it is not a classification map generated within the threshold time, the black box analysis unit 1222 requests the learning unit 1221 to generate a new classification map, and the learning unit 1221 stores it in the database 121. Reload the classification map. The black box analyzer 1222 diagnoses the failure by determining whether the data of the failure state included in the loaded classification map, that is, the values of the sensor data match. The black box analysis unit 1222 outputs fault diagnosis data according to the fault diagnosis result to the interworking interface 124 .

The second analysis engine 123 includes a white box analysis unit 1231 . The white box analyzer 1231 receives only normal sensor data through error and omission processing on the input data. The white box analyzer 1231 calculates a facility condition index indicating the condition of the facility 10 from the sensor data processed for errors and omissions. The facility condition index includes, for example, the efficiency and load factor of the facility 10 , and the selection and calculation method of the facility condition index is determined based on the knowledge of the facility 10 . The white box analyzer 1231 stores the calculated facility state index in the database 121 .

The white box analyzer 1231 loads a failure diagnosis logic for performing linkage analysis between data based on information related to the facility 10 from the database 121 . The white box analyzer 1231 diagnoses a failure based on the failure diagnosis logic loaded with operation information of the equipment 10 according to the equipment state index and sensor data. The white box analyzer 1231 outputs fault diagnosis data according to the fault diagnosis result to the interworking interface 124 .

The interworking interface 124 receives fault diagnosis data from each of the first analysis engine 122 and the second analysis engine 123 , and diagnoses a malfunction for a normal operation or a malfunctioning operation of the facility 10 based on the received information. print the result When the interworking interface 124 receives a failure diagnosis result indicating that both the first analysis engine 122 and the second analysis engine 123 are operating normally, the device 10 outputs a message in which it operates normally.

When an error occurs in at least one of the first analysis engine 122 and the second analysis engine 123 , the interworking interface 124 outputs a failure diagnosis message indicating a failure operation. At this time, information related to a failure acquired according to a failure of one of the first analysis engine 122 and the second analysis engine 123 or information related to a request for setting change for the failure is output together.

As such, the interworking interface 124 integrates the failure diagnosis results output from the plurality of analysis engines including the first analysis engine 122 and the second analysis engine 123 and provides them to the client device 30 .

FIG. 3 is a flowchart exemplarily illustrating a diagnostic operation of the first analysis engine of FIG. 1 .

Referring to FIG. 3 , the first analysis engine 122 receives data, ie, sensor data, in real time from the sensors 111-11n through the database 121 . Here, the sensor data includes real-time operation information (or real-time operation information) of the facility 10 (step S111). The sensor data received from the sensors 111-11n are transmitted in the form of a single vector to have a multi-dimensional form. At this time, the database 121 provides the sensor data to the first analysis engine 122 in real time and simultaneously stores the sensor data therein.

The first analysis engine 122 processes errors and missing data included in the received sensor data (step S113). For example, the first analysis engine 122 determines that sensor data exceeding or less than a predetermined threshold value for omission and error processing is an error, and sets the sensor data in which the error occurs as a preset default value. The first analysis engine 122 also sets a preset default value for data that is missing due to an error occurring in the transmission process. The operation of processing the sensor data generated by the error in the first analysis engine 122 is shown in Equation 1 below.

Here, i is a sensor information index (information for sensor classification), and xi is a sensor input value, that is, sensor data. MAX is the maximum allowed, MIN is the minimum allowed, and DEFAULT is the default. Through Equation 1, the first analysis engine 122 outputs the sensor data input from the sensor as it is if it exists within the preset allowable range, and if it does not exist within the allowable range, removes it and outputs it as a default value. The operation of processing the missing sensor data in the first analysis engine 122 is shown in Equation 2 below.

Here, i is a sensor information index, and xi is a sensor input value, that is, sensor data. DEFAULT is the default. Through Equation 2, the first analysis engine 122 outputs the input sensor data as it is if it exists, and if it does not exist, removes it and sets it as a default value and outputs it.

The first analysis engine 122 pre-processes the data for which error and omission processing has been completed (step S115). Pre-processing refers to conversion processing of inputted sensor data (operation information data) into an easy-to-analyze form. For example, the first analysis engine 122 normalizes the sensor data within the range of the maximum and minimum values so as not to affect the analysis due to the difference in measurement units, or based on the average and standard deviation of the sensor data. Preprocessing can be performed through normalization. In addition, the first analysis engine 122 can perform various tasks such as giving numerical meaning to discontinuous sensor data in the form of grades or redefining variables by using statistical techniques such as Principle Component Analysis. do.

For example, the method of normalizing within the range of the maximum value and the minimum value in the first analysis engine 122 may be expressed as Equation 3 below.

Here, i is a sensor information index, and yi is a normalized sensor input value, that is, normalized sensor data. xi is the sensor input value, xi,max is the maximum value for normalization, and xi,min is the minimum value for normalization.

The first analysis engine 122 stores the preprocessed sensor data in the database 121 (step S117). In addition, since the classification map does not exist in the database 121 during the initial operation, the classification map is generated by machine learning using the preprocessed sensor data. The generated classification map is stored in the database 121 .

In order to analyze the preprocessed data, a black box data mining technique is used. Various methods can be applied as a data mining technique for analyzing sensor data. For example, when the amount of data accumulated in the past is small, a k-nearest neighbors (KNN) algorithm or a local outlier factor (LOF) algorithm may be applied. In this way, even without machine learning, it is possible to immediately check whether a new input value is a failure corresponding to an outlier indicating a different tendency when compared with the past data. In this case, the first analysis engine 122 does not include the learning unit 1221 and includes only the black box analysis unit 1222 . That is, there is no need to perform the operation of the learning unit 1221 to be described below, that is, steps S129 to S135.

On the other hand, when the amount of integrated data to be used for diagnosis is large, the method of diagnosing a failure immediately without machine learning cannot be applied. If a direct data mining technique is applied to a large amount of data, the calculation time may increase beyond the desired level. To this end, a data mining technique to which machine learning is applied can be used. The first analysis engine 122 to which the black box method is applied based on data mining techniques and machine learning analyzes past data and classifies them into several groups. Each group has its own unique characteristics, and the data in the group consists of sensor data suitable for the group's characteristics. For example, equipment operation data can be classified through data mining into three groups: operation start state, operation progress state, and operation end state. has the unique characteristics of

It is possible to use various data mining algorithms to analyze past data and classify them into groups. For example, there is a method of classifying the first analysis engine 122 into groups having a classification criterion set in advance by an administrator or the like. The first analysis engine 122 may use a data mining algorithm such as cluster analysis to group the collected sensor data into similar ones. Sensor data collected using various cluster analysis algorithms, such as K-means algorithm, Expectation-maximization algorithm, DBSCAN algorithm, and OPTICS algorithm, were compared to similar data. They can be classified into groups.

After classifying the information of sensor data into several groups through data classification using data mining and machine learning, the first analysis engine 122 reversely utilizes the characteristic information of each group to determine which sensor data is newly added. A classification map that determines whether or not to be classified into a group is created. In this case, the generated classification map is stored in the database 121 . A data mining technique may be applied to the classification map generation method, and a decision tree or a support vector machine may be utilized.

The first analysis engine 122 loads a classification map (clustering map) from the database 121 (step S119).

The first analysis engine 122 determines whether the difference between the generation time (time _{map _gen} ) of the classification map loaded with the current time (time _now ) is less than a preset threshold time (T) based on the loaded classification map (time _now - time _{map _gen} < T) (step S121). Through this, when the generation time of the classification map is older than the current time by more than a certain time, the first analysis engine 122 reloads and uses the classification map newly created through the regeneration of the classification map, and the generation time of the classification map is not old. The classification map is applied to the analysis as it is.

As a result of the determination in step S121, the first analysis engine 122 generates a classification map if the difference between the generation time (time _{map _gen} ) of the classification map loaded with the current time (time _now ) is equal to or greater than a preset threshold time (T) To load the pre-processed data (step S123).

The first analysis engine 122 performs machine learning for generating a classification map using the pre-processed past data (ie, the past sensor data) and the pre-processed new data (ie, the current sensor data) (step S125). For the specific operation of machine learning using data mining techniques, refer to the above description.

The first analysis engine 122 generates a new classification map through machine learning (step S127).

The first analysis engine 122 stores the generated classification map in the database 121 and proceeds to step S119 (step S129).

As a result of the determination in step S121, the first analysis engine 122 converts the classification map to new data if the difference between the generation time (time _{map _gen} ) of the classification map loaded with the current time (time _now ) is less than a preset threshold time (T) , that is, applied to the preprocessed sensor data (step S131).

The first analysis engine 122 calculates a distance between the new data and the representative value of each group in the classification map (step S133). This diagnoses the failure by comparing whether new data belongs to each group. and can be expressed as Equation 4 below.

Here, k is the index of each group and d is the distance. xk_center is a representative value of group k, and x is newly input sensor data. Here, the distance calculation using the absolute value is exemplary, and various methods such as the Euclidean distance, the Manhattan distance, and the Mahalanobis distance may be applied.

The first analysis engine 122 checks whether the absolute value (d _k ) of the difference between the representative value (xk_center) of the group and the new data (x) is a group that is less than the threshold distance value (R) (d _k < R) do (step S135). Here, when the absolute value of the difference between the representative value of the group and the new data is less than the threshold distance value, the first analysis engine 122 includes the corresponding sensor data in the corresponding group.

As a result of the confirmation of step S135, the first analysis engine 122 determines whether the absolute value of the difference between the representative value of the group and the new data is less than the threshold distance value, whether it is defined as a predefined erroneous data group (step S137) ).

As a result of the determination in step S137, when the first analysis engine 122 is not defined as a predefined error data group, it outputs a message indicating a normal operating state and terminates (step S139).

As a result of the determination in step S137, when the first analysis engine 122 is defined as a predefined error data group, it proceeds to step S141.

On the other hand, as a result of the confirmation of step S135, if the absolute value of the difference between the representative value of the group and the new data is greater than or equal to the threshold distance value, the first analysis engine 122 outputs a message indicating the malfunctioning state and ends (step S141) ).

For reference, referring to the detailed structure of the first analysis engine 122 as a reference, steps S123 to S129 in the flowchart may be performed by the learning unit 1221 , and the remaining steps are performed through the black box analysis unit 1222 . can be

FIG. 4 is a flowchart exemplarily illustrating a diagnostic operation of the second analysis engine of FIG. 1 .

Referring to FIG. 4 , the second analysis engine 123 receives data, ie, sensor data, in real time from the sensors 111-11n through the database 121 . Here, the sensor data includes real-time facility operation information (or operation information) (step S211). The sensor data received from the sensors 111-11n are transmitted in the form of a single vector to have a multi-dimensional form. At this time, the database 121 provides the sensor data to the second analysis engine 123 in real time and simultaneously stores the sensor data therein.

The second analysis engine 123 processes errors and missing data included in the received sensor data (step S213). For example, the second analysis engine 123 determines that sensor data exceeding or less than a predetermined threshold value for omission and error processing is an error, and sets the sensor data in which the error occurs as a preset default value. The second analysis engine 123 also sets a preset default value for data that is missing due to an error occurring in the transmission process. The operation of processing the erroneous sensor data in the second analysis engine 123 refers to Equation 1 above, and the operation of processing the missing sensor data refers to Equation 2 above.

The second analysis engine 123 calculates a facility condition index indicating the condition of the facility (step S215). For example, a facility condition index includes the efficiency or load factor of a facility. The selection of the condition index for each facility and the determination of the calculation method are predetermined based on the knowledge of the facility in a white box manner.

The second analysis engine 123 stores the facility state index in the database 121 (step S217). Here, the equipment condition index may be used for failure analysis.

The second analysis engine 123 loads a failure diagnosis logic for performing linkage analysis between data based on information related to equipment (step S219). The fault diagnosis logic is a linkage analysis logic between data, such as determining whether a specific value deviates from a threshold value or determining whether the product of two values is less than the other value.

After applying the real-time sensor data to the failure diagnosis logic, the second analysis engine 123 determines whether a failure is detected (step S221 ).

As a result of the determination in step S221, if no failure is detected, the second analysis engine 123 outputs a normal operating state and terminates (step S223).

As a result of the determination in step S221, when a failure is detected, the second analysis engine 123 outputs a message indicating the failure operation state and ends. In this case, the message indicating the failure operation state includes a message including a failure-related description of the failure, that is, information on what kind of failure occurred in which sensor.

For reference, referring to the detailed structure of the second analysis engine 123 , the overall operation of the flowchart is performed through the white box analysis unit 1223 .

3 and 4 are illustrated to explain the operation for determining the normal operation and the malfunctioning operation of the sensor, and the above-described operation process may be repeatedly performed while operating the equipment for diagnosing the failure of the equipment. Through this, the equipment failure diagnosis apparatus 100 diagnoses a failure according to the operation of the equipment.

FIG. 5 is a diagram illustrating an operation algorithm of an interworking interface with the first and second analysis engines of FIG. 1 .

Referring to FIG. 5 , the interworking interface 124 receives data or messages related to fault diagnosis from the first analysis engine 122 and the second analysis engine 123 .

The interworking interface 124 integrates the diagnostic results of the first analysis engine 122 to which the black box method is applied and the second analysis engine 123 to which the white box method is applied to perform a final fault diagnosis of the equipment.

As for the interworking interface 124 , four cases according to fault diagnosis of the first analysis engine 122 and the second analysis engine 123 may be considered.

First, the first analysis engine 122 outputs a message indicating a normal operation to the interworking interface 124 , and the second analysis engine 123 outputs a message indicating a normal operation to the interworking interface 124 . In this case, the interworking interface 124 determines the operating state of the facility as a normal operation, and does not output a message indicating the normal operation or perform a separate operation.

Second, the first analysis engine 122 outputs a message indicating a normal operation to the interworking interface 124 , and the second analysis engine 123 outputs a message indicating a malfunctioning operation to the interworking interface 124 . In this case, the interworking interface 124 outputs a message indicating a failure alert and provides a failure-related explanation for the failure. In this case, the failure-related description information is received from the second analysis engine 123 . In addition, the interworking interface 124 may request a setting so that the sensor data in which a failure has occurred is included in a new or existing failure state group in the first analysis engine 122 .

Third, the first analysis engine 122 outputs a message indicating a malfunctioning operation to the interworking interface 124 , and the second analysis engine 123 outputs a message indicating a normal operation to the interworking interface 124 . At this time, the interworking interface 124 outputs a message indicating a failure alert. In addition, the interworking interface 124 may request to confirm whether a failure state is correct and to request addition of new failure diagnosis logic in case of failure.

Fourth, the first analysis engine 122 outputs a message indicating the failure operation to the interworking interface 124 , and the second analysis engine 123 outputs a message indicating the failure operation to the interworking interface 124 . In this case, the interworking interface 124 outputs a message indicating a failure alert and provides a failure-related explanation for the failure. In this case, the failure-related description information is received from the second analysis engine 123 .

In the present invention, fault diagnosis using the output information of the first analysis engine 122 and the second analysis engine 123 using two different methods has been described, but a failure using the output information of three or more analysis engines. The present invention can also be applied to diagnosis.

FIG. 6 is a graph exemplarily illustrating a failure diagnosis result of the first analysis engine of FIG. 1 .

Referring to FIG. 6 , the first analysis engine 122 shows a result of an embodiment in which fault diagnosis is performed on the equipment. New data (ie, current sensor data) is applied to the classification map and shown for each group. For example, it is a result of analyzing data of each of the sensors included in six groups defined as a cluster. The distance calculation result with the shortest group among the distance calculation results with each group is shown on the vertical axis. And, the horizontal axis represents the date and time according to the facility operation.

Through this, in the case where it is determined that the distance is not within the desired specific range through the first analysis engine 122 and is classified as a failure, it is located in the middle between 6:00 and 12:00 on January 16th. is the point of

7 is a graph exemplarily illustrating a failure diagnosis result of the second analysis engine of FIG. 1 .

Referring to FIG. 7 , it is a result of diagnosis through the second analysis engine 123 at a time similar to that of FIG. 6 . The vertical axis represents a specific value quantified in advance for failure determination, and the horizontal axis represents the date and time. The specific value indicated on the vertical axis is a preset value based on an understanding of the operation of the equipment. For example, there is a method of calculating the efficiency of the equipment and checking the failure diagnosis by determining whether the efficiency of the equipment is greater than or equal to a specific value as above. To this end, the sensor data may be displayed on the graph as a numerical value by applying a predetermined specific formula.

The second analysis engine 123 determines the failure of the boiler corresponding to the identification number 1 based on January 16, 2015 at 10:18.

At this time, the detected failure algorithm is that the amount of water used is excessive (very high), and the treatment method is to treat it to have a normal water flow rate. Also, the detected failure algorithm is that the oxygen ratio in the exhaust gas is excessive, and the treatment method is a request for repair of the burner or repair of the sensor.

In the graph, at 10:18, 2015, the numerical value according to the sensor value corresponds to about 1.8, and thus it is determined as a failure state. It can be confirmed as "excessive use of water" as a description of the failure related to the failure.

The apparatus 100 for diagnosing equipment failure of the present invention outputs a single failure diagnosis result by linking and integrating a plurality of analysis engines to which different algorithms for failure diagnosis are applied.

Meanwhile, in the detailed description of the present invention, specific embodiments have been described, but various modifications are possible without departing from the scope of the present invention. Therefore, the scope of the present invention should not be limited to the above-described embodiments, but should be defined by not only the claims described below, but also the claims and equivalents of the present invention.

10: equipment 20: network
21: gateway 110: sensor unit
111-11n: sensors 120: server unit
121: database 122, 123: analysis engines
124: interworking interface 1221: learning unit
1222: black box analysis unit 1222: white box analysis unit
1231: white box analysis unit

Claims (17)

sensors that collect data related to the operation of the facility;
a database connected to the sensors through a network and storing sensor data collected from the sensors;
a first analysis engine to which a black box method for diagnosing a failure of the facility is applied using only the sensor data;
a second analysis engine to which a white box method for diagnosing a failure of the facility by linking the sensor data with the facility is applied; and
It is connected to the first analysis engine and the second analysis engine, and combines the first failure diagnosis data output from the first analysis engine and the second failure diagnosis data output from the second analysis engine to detect the failure of the facility. including an interlocking interface to diagnose;
The interworking interface is
If the first fault diagnosis data indicates a normal operation and the second fault diagnosis data indicates a normal operation, it is determined as a normal operation;
When the first failure diagnosis data indicates a normal operation and the second failure diagnosis data indicates a failure operation, a message including a failure alarm and a failure-related description for the failure is output;
When the first failure diagnosis data indicates a failure operation and the second failure diagnosis data indicates a normal operation, a message including a failure alarm and a request for confirmation of the failure state is output;
When the first failure diagnosis data indicates a failure operation and the second failure diagnosis data indicates a failure operation, a failure alarm and a message including a failure-related explanation for the failure are output. The method of claim 1,
The first analysis engine is
a learning unit for generating a classification map for error detection by machine learning the sensor data, and storing the generated classification map in the database; and
Loading a classification map generated less than a threshold time based on the current time from the database, and diagnosing a failure by determining whether the current sensor data matches the values of the sensor data of the failure state included in the loaded classification map Equipment failure diagnosis device including a black box analysis unit. 3. The method of claim 2,
The black box analysis unit
If the classification map generated from the database for less than a threshold time based on the current time does not exist, the equipment failure diagnosis apparatus for regenerating the classification map through the learning unit. 3. The method of claim 2,
The black box analysis unit
Equipment that calculates the distance between the current sensor data and the representative values of each group included in the classification map, and determines a failure state if the calculated distance does not fall below a threshold value or is defined as a predefined error data group Troubleshooting device. 3. The method of claim 2,
The black box analyzer outputs the first failure diagnosis data indicating a normal operation or a failure operation according to the diagnosis of the failure. The method of claim 1,
The second analysis engine includes a white box analyzer that analyzes the failure using failure diagnosis logic for the sensor data,
The failure diagnosis logic is a logic for performing linkage analysis between data based on information related to the facility. 7. The method of claim 6,
The white box analysis unit calculates a facility condition index indicating the condition of the facility, and uses the facility condition index for the fault diagnosis,
The facility condition index is a facility failure diagnosis device including the efficiency and load factor of the facility. 7. The method of claim 6,
The white box analyzer outputs the second failure diagnosis data including a normal operation or a failure operation according to the diagnosis of the failure and a related description corresponding to the failure operation. delete collecting data related to the operation of the facility through sensors;
storing sensor data collected from the sensors;
performing fault diagnosis in a black box method of diagnosing a fault of the facility using only the sensor data;
performing fault diagnosis in a white box method for diagnosing a failure of the facility by linking the sensor data with the facility; and
and outputting a fault diagnosis result of the facility by integrating the first fault diagnosis data according to the black box type fault diagnosis and the second fault diagnosis data according to the white box type fault diagnosis,
The step of outputting the failure diagnosis result of the equipment is
determining that the first fault diagnosis data indicates a normal operation and the second fault diagnosis data indicates a normal operation, determining that the operation is normal, and outputting a message indicating the normal operation;
outputting a message including a failure alarm and a failure-related explanation for the failure when the first failure diagnosis data indicates a normal operation and the second failure diagnosis data indicates a failure operation;
outputting a message including a failure alarm and a request for confirmation of a failure state when the first failure diagnosis data indicates a failure operation and the second failure diagnosis data indicates a normal operation; and
and outputting a message including a failure alarm and a failure-related explanation for the failure when the first failure diagnosis data indicates a failure operation and the second failure diagnosis data indicates a failure operation. 11. The method of claim 10,
The step of diagnosing the black box method is
receiving sensor data from sensors connected to the facility;
generating a classification map using the sensor data;
determining whether the generation time of the classification map exceeds a threshold time based on the current time;
diagnosing a failure using the classification map when the threshold time is not exceeded; and
When the threshold time is exceeded, generating a new classification map by machine learning using past sensor data and current sensor data, and generating the first failure diagnosis data through diagnosis of the failure using the generated classification map; A method for diagnosing equipment failures, including. 12. The method of claim 11,
The step of generating the classification map is
setting error data and missing data included in the input sensor data as preset default values;
pre-processing the processed data for the error data and the missing data; and
The facility failure diagnosis method further comprising generating the classification map using the preprocessed sensor data. 12. The method of claim 11,
The step of generating the first failure diagnosis data includes:
Equipment that calculates the distance between the current sensor data and the representative values of each group included in the classification map, and determines a failure state if the calculated distance does not fall below a threshold value or is defined as a predefined error data group How to diagnose a breakdown. 11. The method of claim 10,
The step of diagnosing the failure of the white box method is
receiving sensor data from sensors connected to the facility;
calculating a facility condition index using the sensor data; and
and generating the second failure diagnosis data for diagnosing the failure using the failure diagnosis logic using the sensor data,
The failure diagnosis logic is a logic for performing linkage analysis between data based on information related to the equipment. 15. The method of claim 14,
The step of generating the second failure diagnosis data includes:
diagnosing the failure using the equipment condition index,
The facility condition index is a facility failure diagnosis method including the efficiency and load factor of the facility. 15. The method of claim 14,
The step of generating the second failure diagnosis data includes:
and generating the second failure diagnosis data including a related description corresponding to the failure when it is determined that the operation of the equipment is a failure. delete

Images