KR102212022B1 - Method of automatically determining condition of hydro turbine in hydroelectric power plant and system for the same - Google Patents

Abstract

Disclosed are a method for automatically determining the health of a flap wheel and a system therefor. Here, the health of the flap wheel can be automatically evaluated by executing a program in a computing device to process data measured by a plurality of sensors for monitoring the health of the flap wheel. The program includes a health automatic determination model unit and a health determination unit. The health automatic determination model unit may normalize measurement data which is read from a data storage by means of preprocessing in accordance with a multiple state classification algorithm to input the same into a plurality of artificial neural network models, wherein each artificial neural network model uses the inputted measurement data to perform analysis for health evaluation as an additional independent process. The health determination unit uses the additional health evaluation result deduced by the analysis of the plurality of artificial neural network models to ultimately decide the determination of the health state of the flap wheel based on the majority rule. Here, the program may further comprise a determination result notification unit which can receive the result of the determination of the health state of the flap wheel to convert the same into a format which can be recognized by a system manager to provide the same to the system manager while issuing a notification to the manager when the deduced ultimate evaluation result is determined to be erroneous.

Description

{Method of automatically determining condition of hydro turbine in hydroelectric power plant and system for the same}

The present invention relates to the field of technology for monitoring the health of a hydroelectric power plant or pumping power plant facility, and more particularly, to a technology capable of automatically determining the soundness of a pumping water wheel of a power plant.

During hydroelectric power generation, the pumping water wheel has a structure in which water is always present and rotates at a constant speed of 300 RPM, so it is difficult for the soundness monitoring sensor of the pumping water wheel to be directly attached to the pumping water wheel. For this reason, the monitoring sensor is attached to the outside of the aberration and makes indirect measurement. However, if the data obtained by indirectly measuring amniotic aberration is simply reviewed, it is difficult to directly identify the characteristics of amniotic aberration.

Measurement data of various sensors can be used to monitor the health of pumping water in a hydroelectric power plant. However, the conventional monitoring method for the health of pumped water in a hydroelectric power plant is used in the form of diagnosing the condition by separately analyzing the values measured from each sensor in a single form. The conventional method of analyzing and using the measurement data of several sensors in a single form makes it difficult to perform an integrated analysis of abnormal conditions. The need to increase reliability by analyzing the data measured from each sensor is raised.

There is a conventionally known pumping power failure diagnosis system, but the system is an alarm system that judges only abnormalities of facilities outside of normal operation. For example, in the past, intensive research has been conducted only on cavitation, which is a state anomaly occurring in amniotic aberration. It was difficult for the system to diagnose the state of the pumping water wheel of the pumping power plant and predict its lifespan, or to diagnose the soundness of the state of the pumping power plant as a whole and predict the lifespan.

The soundness of amniotic aberration can also be evaluated through the vibration pattern analysis of the structure. The integrity evaluation method has a limitation in evaluating only a single characteristic of a structure.

One of the known health evaluation methods is a method of notifying an alarm when single sensing data fluctuates beyond the value set by the system. The system implementing this evaluation method frequently issued false alarms.

There is a need for a real-time automatic health evaluation system using integrated analysis to detect major abnormalities in pumped-out power generation facilities and prevent failures in advance.

(1) Patent registration number: 10-1498527 (2015.02.26), Title of invention: Power plant rotation facility condition diagnosis system by vibration frequency analysis program (2) Patent registration number: 10-1903283 (2018.09.20.), Title of invention: Automatic diagnosis system and automatic diagnosis method for power generation facilities (3) Patent registration number: 10-1307526 (2013.09.05.), Title of invention: Real-time vibration monitoring device

An object of the present invention is to provide a method and system capable of automatically determining the soundness of a pumping water wheel based on static data by utilizing a static monitoring system currently in use without changing the design and facility for monitoring pumping water in a hydropower plant.

An object of the present invention is to provide a method (algorithm) capable of automatically determining the health of a power generation facility by integrated analysis of monitoring data generated from pumped water power generation, and reconfiguring accordingly when applying an additional monitoring sensor and system. .

The problem to be solved by the present invention is not limited to the above-described problems, and may be variously expanded without departing from the spirit and scope of the present invention.

A method for automatically determining the soundness of amniotic aberration according to embodiments for realizing one object of the present invention is a method for automatically determining the soundness of amniotic aberration using a computer program executed in a computer device. The method includes the steps of reading measurement data from a monitoring data storage storing data measured by a plurality of sensors for monitoring the integrity of amniotic aberration and inputting it to an automatic soundness determination model unit; Obtaining normalized measurement data by preprocessing the measurement data in the automatic soundness determination model unit; Inputting the normalized measurement data to a plurality of artificial neural network models, respectively, and performing an analysis for health evaluation in each artificial neural network model as a separate independent procedure according to a multi-state classification algorithm; And finally determining, in the soundness determination unit, the soundness state of the amniotic aberration according to the majority voting principle by using separate soundness evaluation results derived by analyzing the plurality of artificial neural network models.

In example embodiments, the measurement data may include temperature data, displacement data, and speed data about a turbine, a generator, and a turbine-generator central axis.

In exemplary embodiments, the method includes, in the automatic health determination model unit, before the pre-processing step, reviewing whether the measurement frequency is constant as time series data for each measurement data classified for each sensor, and if not, transfer It may further include performing linear interpolation based on the data to make the measurement frequency constant for each data.

In example embodiments, the obtaining of the normalized measurement data includes: classifying the input measurement data for each sensor; Removing noise due to temperature, humidity, and vibration changed by outside air in addition to state information of positive aberration; In addition, it may include performing data normalization by using data for a predetermined period prior to the corresponding data.

In example embodiments, the plurality of artificial neural network models used by the multi-state classification algorithm are a multi-layer perceptron model (MLP model), a one-dimensional convolutional neural network model. Networks: 1D-CNN model), and Recurrent Neural Networks-Long Short-Term Memory (RNN-LSTM model).

In example embodiments, the method may further include storing, by the soundness determination unit, a final evaluation result derived from the automatic soundness determination model unit together with monitoring data at the same point in a data storage.

In example embodiments, the method may further include issuing an alarm to the manager when the final evaluation result derived by the health determination unit is determined as an abnormal state.

On the other hand, the system for automatically determining the soundness of amniotic aberration according to embodiments for realizing the above object of the present invention is a plurality of sensors that generate measurement data by measuring elements related to the soundness of amniotic aberration in real time. Monitoring sensor unit including; A data storage for receiving and storing the measurement data generated by the monitoring sensor unit; And it may include a computing device configured to evaluate the soundness of the positive aberration by processing the measurement data through the execution of the program. According to a multi-state classification algorithm, the program normalizes the measurement data read from the data storage through pre-processing and inputs it into a plurality of artificial neural network models, and uses the input measurement data in each artificial neural network model. A soundness automatic determination model unit including a function of performing an analysis for soundness evaluation in an independent procedure; In addition, it may include a soundness determination unit including a function of finally determining the soundness state of amniotic aberration according to a majority voting principle by using separate soundness evaluation results derived by analysis of the plurality of artificial neural network models.

In exemplary embodiments, the program receives the result of the soundness state determination of the positive aberration derived by the soundness determination unit, converts it into a form that can be grasped by the system administrator, and provides the function to the system administrator. It may further include a determination result notification unit including a function of generating an alert to the manager when the final evaluation result is determined as an abnormal state.

In example embodiments, the measurement data may include temperature data, displacement data, and speed data about a turbine, a generator, and a turbine-generator central axis.

In example embodiments, the automatic health determination model unit examines whether the measurement frequency is constant as time series data for each measurement data classified by sensor, and if not, performs linear interpolation based on previous data, A function of making the measurement frequency constant for each data may be further included.

In example embodiments, the automatic soundness determination model unit includes a function of classifying measurement data read from the data storage for each sensor, and noise due to temperature, humidity, and vibration changed by outside air in addition to state information of positive aberration. The ability to remove; And it may further include a function of performing data normalization by using data for a predetermined period prior to the corresponding data.

In example embodiments, the plurality of artificial neural network models used by the multi-state classification algorithm may include an MLP model, a 1D-CNN model, and an RNN-LSTM model.

In example embodiments, the soundness determination unit may further include a function of storing the final evaluation result derived from the soundness automatic determination model unit together with monitoring data at the same time in the data storage.

According to exemplary embodiments of the present invention, monitoring data measured by a static monitoring system installed on an existing pumping water can be utilized for health evaluation. Therefore, since it is not necessary to install a separate additional dynamic monitoring system, the present invention is not particularly difficult to apply to an existing power generation facility monitoring system.

In the case of the positive aberration integrity evaluation method through the vibration pattern analysis of the structure, there is a problem that only a single characteristic of the structure is evaluated, but the present invention utilizes sensing data on the overall behavior of the structure such as temperature, displacement, and velocity. It is also possible to judge abnormal conditions caused by external factors.

In the existing soundness evaluation method, when a single sensing data value fluctuates beyond the value set by the system, it is determined as an abnormal state and an alarm is notified, but there is a problem that false alarms are frequently generated due to the uncomfortable judgment. However, since the present invention comprehensively reviews sensing data acquired from various sensors and makes a comprehensive judgment, the probability of making an erroneous judgment is very low, so that the efficient safety management of the structure can be supported.

1 is a block diagram showing the configuration of an apparatus for monitoring the health of pumping water aberration of a hydroelectric power plant according to an exemplary embodiment of the present invention.
2 is a flow chart showing an execution process of a method for automatically determining the soundness of pumping water in a hydropower plant according to an exemplary embodiment of the present invention.
3 schematically shows a method for evaluating the soundness of amniotic aberration using an MLP model according to an exemplary embodiment of the present invention.
4 schematically shows a method for evaluating health using a 1D-CNN model according to an exemplary embodiment of the present invention.
5 schematically shows a method for evaluating health using an RNN-LSTM model according to an exemplary embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the accompanying drawings. The same reference numerals are used for the same elements in the drawings, and duplicate descriptions for the same elements are omitted.

With respect to the embodiments of the present invention disclosed in the text, specific structural or functional descriptions have been exemplified only for the purpose of describing the embodiments of the present invention. The embodiments of the present invention may be implemented in various forms and should not be construed as being limited to the embodiments described in the text. That is, the present invention can be modified in various ways and can have various forms, and specific embodiments will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to a specific form of disclosure, it is to be understood as including all changes, equivalents, or substitutes included in the spirit and scope of the present invention.

The block diagram of FIG. 1 shows the configuration of an apparatus 100 for monitoring the health of pumping water aberration of a hydropower plant according to an exemplary embodiment of the present invention.

Referring to FIG. 1, the amniotic aberration health monitoring device 100 may include a monitoring sensor unit 20, a monitoring data storage 40, and a computing device 60.

The monitoring sensor unit 20 may include a plurality of sensors. A number of these sensors are currently installed for health monitoring of major components constituting generators and water turbines in hydroelectric power plants or pumping power plants. The monitoring sensors may measure the operation or state of various monitoring elements of the monitoring targets, that is, the pumped water generating aberration 10 in real time, or periodically or when necessary. Data measured by the sensors may be provided to the monitoring data storage 40.

The monitoring data storage 40 may receive and store various measurement data generated by the monitoring sensor unit 20. This monitoring data storage 40 may be managed by the computing device 60 or a separate data management server device.

In an exemplary embodiment, the computing device 60 may include hardware of a general-purpose computer required to produce a desired result by processing data through execution of a program and software to be described later. The software may include an automatic soundness determination model unit 70 implemented based on an automatic determination algorithm for the soundness of pumped water of a hydropower or pumped water power plant. The soundness automatic determination model unit 70 may include a plurality of models to which a multi-state classification algorithm based on monitoring data is applied in order to improve the reliability of the positive aberration state determination. In an exemplary embodiment, the health automatic determination model unit 70 includes, for example, three models of a Multi Layer Perceptron model 72, a 1D CNN-Overall Structure model 74, and a Long-Short Term Memory model 76. can do. Each of these three models can process monitoring data provided from the monitoring data storage 40 by applying the automatic determination algorithm for positive aberration soundness proposed in the present invention, and perform soundness evaluation such as amniotic aberration.

The software installed in the computing device 60 may include a soundness determination unit 80 and a determination result notification unit 90. The soundness determination unit 80 may make a final determination on the state of the positive power generation aberration according to a majority vote using the result of the classification algorithm of the three models 72, 74, 76 of the soundness automatic determination model unit 70. .

The determination result notification unit 90 may receive the state determination result derived by the soundness determination unit 80, convert it into a form that the system administrator can grasp, and provide it to the system administrator. When the determination result is determined to be an abnormal state, the determination result may be stored in the monitoring system 100 for a corresponding point in time, and the determination result may be notified to the administrator in the form of an alarm so that rapid action on the abnormal state may be taken.

According to the automatic determination system 100 for the health of the pumping water vehicle in a hydroelectric power plant according to this embodiment, real-time static monitoring data measured during the operation of the pumping water vehicle 10 of the hydroelectric power plant is collected and stored in a database, and the stored information is By processing, the state of the amniotic aberration 10 can be automatically determined. And by automatically generating an alarm when an abnormal condition occurs in the pumping water wheel 10, it is possible to minimize excessive load operation of the pumping water wheel 10 and clearly determine the abnormal state of the structure.

The flowchart of Fig. 2 shows an execution process of a method for automatically determining the soundness of pumped water aberration of a hydropower plant according to an exemplary embodiment for performing such processing.

Referring to FIG. 2, a procedure for processing and storing and managing sensing data in a method for automatically determining the soundness of pumped water in a hydropower plant is shown. The process described below may be automatically performed when monitoring data generated by the monitoring sensor unit 20 is input from the monitoring data storage 40 to the automatic health determination model unit 70.

First, when the automatic soundness determination model unit 70 receives measurement data generated by the sensors of the monitoring sensor unit 20 and stored in the monitoring data storage 40, it checks the measurement data and classifies it by sensor ( S10). In an exemplary embodiment, the data measured by the pumping water wheel 10 may utilize a total of 59 data about the turbine, the generator, and the turbine-generator central axis as summarized in Table 1 below. For example, the measurement data type may include 44 temperature data, 9 displacement data, and 6 speed data. Of course, the number of data and measurement items measured by the monitoring sensor unit 20 may be more or less than this. Unlike a conventional amniotic aberration monitoring system, it may be characterized in that it is possible to evaluate the health of the amniotic aberration without measurement data about power. Since power data measurement is not required, there is an advantage in that the type of sensor can be reduced and the amount of data to be processed is also reduced.

turbine generator Turbine-Generator central axis Temperature data 24 20 Displacement data 4 3 3 Speed data 3 (bearing) 3 (bearing)

After the measurement data is classified, it is examined whether the measurement frequency is constant as time series data for each data (S12). If not constant, linear interpolation may be performed based on previous data to make the measurement frequency constant for each data (S14).

The interpolated data at the same measurement frequency may be pre-processed in order to remove changes due to temperature, humidity, and vibration changed by outside air in addition to the state information of the positive aberration 10 (S16). In the pre-processing step, data is first collected to remove noise. In addition, data can be normalized using data for 10 days prior to the data.

Data normalized through pre-processing may be input to an artificial neural network model and data processing for analysis may be performed (S18). In an exemplary embodiment, a multi-state classification algorithm based on monitoring data may be applied to improve the reliability of state determination of the positive aberration 10. As an example, the artificial neural network used in the multi-state classification algorithm is a multi-layer perceptron model (72), a one-dimensional convolutional neural network model (1D-CNN model). )(74), and Recurrent Neural Networks-Long Short-Term Memory (RNN-LSTM model)(76), a total of three models can be included.

The three health evaluation models 72, 74, and 76 can each independently analyze input data to derive separate health evaluation results. The multi-state classification algorithm may finally determine a state determination regarding the soundness of the positive aberration according to the majority voting principle by using the soundness evaluation results derived from each of the three models (S20). That is, if two or more of the three health evaluation models 72, 74, and 76 evaluate that an abnormality has occurred, it can be determined that the positive power generation aberration 10 has an abnormality.

3 schematically shows a method for evaluating the soundness of amniotic aberration using the MLP model 72.

Referring to FIG. 3, the MLP model 72 is a model developed by stacking hidden layers using several fully connected layers to understand general characteristics of data. The MLP model 72 is composed of a total of 4 layers. In the three layers, the data dimension is reduced through a fully connected neural network, batch normalization is applied, and the characteristic of the data is characterized by using a characteristic determination algorithm, Rectified Linear Unit (ReLU). To grasp. The MLP model 72 is a method of determining a state by sequentially reducing the dimension of data through this process. Through the last one fully connected neural network, the final features of the data are output with the probability of predicting each state. The soundness can be evaluated by indicating the highest probability from the corresponding final output value. There is a problem that the spatial structure disappears due to the nature of the fully connected layer, but this can be solved by stacking several layers.

4 schematically shows a method for evaluating health using the 1D-CNN model 74.

The 1D-CNN-based model 74 can identify data characteristics through the spatial structure of the data. A small group is formed in the data, processed first, and characteristics are derived. Referring to FIG. 4, the 1D-CNN model 74 may be composed of a total of 5 layers. In the three CNN modules, data can be subdivided, and data characteristics can be identified through convolution between the subdivided data. Thereafter, the shape is changed so that it is easy to process through data type conversion (Reshape), and then the final feature of the data can be derived as a probability of predicting each state through a fully connected neural network. Health can be evaluated by indicating the highest probability from the corresponding output value.

5 schematically shows a method for evaluating health using the RNN-LSTM model 76.

Referring to FIG. 5, the measurement data generated by the positive aberration 10 is time series data, and has a characteristic of high correlation with previous data. Using this, you can derive characteristics for additional data that are introduced based on previous data characteristics. The RNN-LSTM model 76 may include a total of three layers. Through two LSTM models, features for a time series of data are extracted, and the extracted features affect the data of the next time. With respect to the extracted characteristics, the final characteristics of the data can be derived with the probability of predicting each state through the fully connected neural network. Health can be evaluated by indicating the highest probability from the corresponding output value.

The soundness determination unit 80 may receive the evaluation results analyzed by the three soundness evaluation models 72, 74, and 76 as described above, and determine whether the positive aberration 10 is in an abnormal state regarding the soundness. Specifically, the soundness determination unit 80 may derive a final result on soundness evaluation by integrating the evaluation results of the three soundness evaluation models 72, 74, 76. The characteristics extracted from the three models of the MLP model 72, the 1D-CNN model 74, and the RNN-LSTM model 76 are all different, and due to this, the final output values may all be different for each model. Accordingly, the health diagnosed by the three models 72, 74, and 76 may be different.

The soundness determination unit 80 may determine the final result of soundness evaluation according to the principle of majority vote. If there are two or more models determined as an abnormal state, the state of the positive aberration 10 is determined as an abnormal state at the time point. For example, the MLP model 72 and the 1D-CNN model 74 determine a normal state, but the RNN-LSTM model 76 may determine an abnormal state. In this case, since the two models were determined to be in a normal state, the final evaluation result can be diagnosed as a normal state. The final evaluation result may be stored in the data storage 40. However, when saving the final evaluation result, it is possible to record the details of the evaluation results different from the final evaluation, that is, in the above example, the RNN-LSTM model 76 determines that it is in an abnormal state.

When applying the method of evaluating soundness according to the principle of majority vote using these three evaluation models, it can be seen that the accuracy is increased by about 15% (75% --> 90%) compared to the case of evaluating soundness using a single model. there was.

The soundness determination unit 80 may store the soundness determination result in the monitoring data storage 40, and may issue an alarm to an administrator when it is determined as an abnormal state. That is, when it is determined as an abnormal state, the soundness determination unit 80 may transmit the determination result to the determination result notification unit 90 so that an alert may be issued to the administrator (S22).

In addition, the soundness determination unit 80 may store the result of evaluating the state of the positive aberration 10 in the monitoring data storage 40 together with the monitoring data at the same time point (S24). If only one of the three models is determined to be an abnormal state, the state is specified as a normal state, but when data is saved, the state can be marked and stored as a suspected abnormal state. If it is determined to be in a normal state, no additional alarm is issued.

The positive aberration soundness automatic determination system 100 described above may be implemented as a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the devices and components described in the embodiments include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It can be implemented using one or more general purpose computers or special purpose computers, such as a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of software. The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to behave as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodyed in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

The present invention can be used to monitor the health of a pumping water wheel of a pumping power plant in real time.

Although the embodiments have been described by the limited drawings as described above, those skilled in the art will variously modify and change the present invention within the scope not departing from the spirit and scope of the present invention described in the following claims. You will understand that you can. For example, the described techniques are performed in a different order from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved. Therefore, other implementations, other embodiments, and claims and equivalents fall within the scope of the claims to be described later.

10: pumping power generation aberration 20: monitoring sensor unit
40: data storage 60: computing device
70: soundness automatic determination model unit 80: soundness determination unit
90: determination result notification unit 100: positive aberration soundness automatic determination system

Claims (15)

As a method for automatically determining the soundness of amniotic aberration using a computer program running on a computer device,
Reading measurement data from a monitoring data storage storing data measured by a plurality of sensors for monitoring the integrity of amniotic aberration and inputting it to an automatic soundness determination model unit;
Obtaining normalized measurement data by preprocessing the measurement data in the automatic soundness determination model unit;
Inputting the normalized measurement data to a plurality of artificial neural network models, respectively, and performing an analysis for health evaluation in each artificial neural network model as a separate independent procedure according to a multi-state classification algorithm; And
And finally determining, in the soundness determination unit, the soundness state of amniotic aberration according to a majority voting principle using separate soundness evaluation results derived by analyzing the plurality of artificial neural network models. A method for automatically determining health. The method of claim 1, wherein the measurement data includes temperature data, displacement data, and velocity data about a turbine, a generator, and a turbine-generator central axis.
The method of claim 1, wherein, before the pre-processing step, the automatic health determination model unit reviews whether the measurement frequency is constant as time series data for each measurement data classified for each sensor, and if not, linearly based on previous data. The method for automatically determining the soundness of positive aberration, further comprising the step of making the measurement frequency constant for each data by performing interpolation. The method of claim 1, wherein the obtaining of the normalized measurement data comprises: classifying the input measurement data for each sensor; Removing noise due to temperature, humidity, and vibration changed by outside air in addition to state information of positive aberration; And performing data normalization by using data for a predetermined period prior to the corresponding data. The method of claim 1, wherein the plurality of artificial neural network models used by the multi-state classification algorithm are a multi-layer perceptron model (MLP model) and a one-dimensional convolutional neural network model. 1D-CNN model), and a Recurrent Neural Networks-Long Short-Term Memory (RNN-LSTM model), a method for automatically determining the soundness of positive aberrations. The automatic soundness of positive aberration according to claim 1, further comprising the step of storing, by the soundness determination unit, the final evaluation result derived from the automatic soundness determination model unit together with monitoring data at the same time point. How to judge as. The method according to claim 1, further comprising the step of issuing an alert to an administrator when the final evaluation result derived by the soundness determination unit is determined to be an abnormal state. A monitoring sensor unit including a plurality of sensors for generating measurement data by measuring elements related to the soundness of amniotic aberration in real time;
A data storage receiving and storing the measurement data generated by the monitoring sensor unit; And
And a computing device configured to process the measurement data through execution of a program to evaluate the soundness of the positive aberration,
The program, according to the multi-state classification algorithm,
The measurement data read from the data storage is normalized through preprocessing and input into a plurality of artificial neural network models, and analysis for soundness evaluation is performed in a separate independent procedure using the input measurement data in each artificial neural network model. Integrity automatic determination model unit including a function to; And a soundness determination unit including a function of finally determining the soundness state of amniotic aberration according to a majority voting principle by using separate soundness evaluation results derived by analyzing the plurality of artificial neural network models. System to automatically determine the health of The function of claim 8, wherein the program receives the result of the soundness state determination result of the positive aberration derived by the soundness determination unit, converts it into a form that can be grasped by the system administrator, and provides it to the system administrator. A system for automatically determining the soundness of amniotic aberration, further comprising: a determination result notification unit including a function of generating an alarm to an administrator when the result is determined as an abnormal state. 9. The system of claim 8, wherein the measurement data includes temperature data, displacement data, and velocity data about a turbine, a generator, and a turbine-generator central axis. The method of claim 8, wherein the automatic health determination model unit reviews whether the measurement frequency is constant as time series data for each measurement data classified by sensor, and if not constant, performs linear interpolation based on previous data to A system for automatically determining the soundness of amniotic aberration, further comprising a function of making the measurement frequency constant. The method of claim 8, wherein the soundness automatic determination model unit classifies measurement data read from the data storage for each sensor, and removes noise due to temperature, humidity, and vibration changed by outside air in addition to state information of positive aberration. function; And a function of performing data normalization using data for a predetermined period prior to the corresponding data. A system for automatically determining the soundness of positive aberrations. The method of claim 8, wherein the plurality of artificial neural network models used by the multi-state classification algorithm are a multi-layer perceptron model (MLP model) and a one-dimensional convolutional neural network model. 1D-CNN model), and a Recurrent Neural Networks-Long Short-Term Memory (RNN-LSTM) model. A system for automatically determining the soundness of positive aberrations. The soundness of positive aberration according to claim 8, wherein the soundness determination unit further comprises a function of storing the final evaluation result derived from the soundness automatic determination model unit together with monitoring data at the same time point in the data storage. System for automatic judgment. A computer readable recording medium having a computer program recorded thereon for performing the method for automatically determining the soundness of the positive aberration according to any one of claims 1 to 7.

Images