KR102104026B1 - Real-time Solenoid Valve Monitoring System and Method Using Artificial Neural Network - Google Patents

Abstract

Disclosed is a real-time solenoid-operated valve monitoring system which analyzes the integrity of a solenoid-operated valve in real time. The real-time solenoid-operated valve monitoring system uses an artificial neural network to monitor whether a solenoid-operated valve is abnormal, and comprises: a sensor connected to the solenoid-operated valve to measure data sensed from the solenoid-operated valve; and a diagnosis unit connected to the sensor to acquire, process, and analyze data measured by the sensor to monitor and diagnose whether the solenoid-operated valve is abnormal. The diagnosis unit includes: a data acquisition and transmission unit to acquire data measured by the sensor from the sensor connected to the solenoid-operated valve, and transmit the acquired data to a data processing unit; a data processing unit to receive data from the data acquisition and transmission unit to preprocess the data to be suitable for processing to extract core data; and a data analysis unit to use a pre-learned artificial neural network to analyze core data extracted by the data processing unit to monitor and diagnose the solenoid-operated valve in real time. The artificial neural network receives learning data as input data from valves for learning in a normal state and an abnormal state for a prescribed period to be learned to analyze whether the solenoid-operated valve is abnormal.

Description

Real-time Solenoid Valve Monitoring System and Method Using Artificial Neural Network

The present invention relates to a system and method for real-time monitoring of data from a sensor installed in a solenoid valve and analyzing the health of the solenoid valve in real time using an artificial neural network.

Solenoid Operated Valve (SOV) is an electromagnetic valve. When electricity is applied, the coil inside the valve is excited to generate electromagnetic force and the flange rises to open the valve. When electricity is cut off, the valve is automatically closed by the flange weight. It is a device that controls the flow of air, water, oil, etc. by doing the opposite. That is, SOV can convert electrical energy into mechanical force and use it as a driving force. This SOV is one of the automatic control valves required to safely and efficiently operate power generation facilities or other facilities. SOV is small, lightweight, and has a fast response characteristic. Recently, it has been actively researched and developed in the field of electric, magnetic, and fluid mechanical technologies, and is applied to all industries including home appliances, semiconductors, and factory automation, as well as for fluid control of power plants and general plants. .

Recently, as the automatic control facility of a power plant is installed as a more advanced automation facility, the number of various SOVs used for blocking and supplying various fluids is rapidly increasing. For example, SOV used in a nuclear power plant may be a safety system related device that automatically controls the opening and closing of the fluid flow in the piping system or the flow rate. When SOV does not normally open or close due to aging, burnout, and various errors, it affects the related systems, causing power plant output detection, power plant shutdown, etc. do. Looking at various failure cases caused by the unstable operation of SOV, it can be seen that the need for health monitoring and diagnosis is very important. In the case of power plants, the current diagnosis and management methods have a time limit that allows inspection only during maintenance periods, and the trend analysis of measured values is insufficient or very little activity. Therefore, there is a need for an efficient system for monitoring and managing SOV in real time.

The present invention has been devised to solve the above problems, and an object thereof is to provide an apparatus and method for real-time analysis of the health of SOV by monitoring SOV in real time. It is also an object to provide real-time information on health, such as whether the current state of SOV is broken and the type of failure that may occur after a certain period of time, by using an analysis module using an artificial neural network. In particular, it is an object of the present invention to provide an apparatus and method for real-time analysis of the health of an SOV by monitoring the driving current of the SOV in real time.

A real-time monitoring system for a solenoid valve using an artificial neural network according to the present invention for achieving the above object includes: a sensor connected to the solenoid valve and measuring data sensed from the solenoid valve; Includes a diagnostic unit connected to the sensor to monitor and diagnose the presence or absence of the solenoid valve by acquiring, processing, and analyzing the data measured by the sensor, wherein the diagnostic unit includes the sensor from the sensor connected to the solenoid valve A data acquisition and transmission unit that acquires the measured data and transfers the acquired data to the data processing unit; A data processing unit that receives data from the data acquisition and transmission unit and preprocesses it to be suitable for analyzing the data to extract core data; And a data analysis unit for real-time monitoring and diagnosis of a solenoid valve by analyzing the core data extracted from the data processing unit using a pre-trained artificial neural network, and the artificial neural network learning from a learning valve for normal and abnormal states. It was learned to analyze the abnormality of the solenoid valve by receiving the data as input data for a certain time.

The sensor includes a voltage sensor that senses the voltage of the solenoid valve, a current sensor that senses the current of the solenoid valve, a temperature sensor that senses the temperature of the solenoid valve, a thermal image sensor that senses the heat of the solenoid valve, and an operation sound of the solenoid valve. It may be configured to include any one of the acoustic sensor for sensing.

The data acquisition and transmission unit is a power supply for supplying power to the sensor; A signal adjusting unit having a filter that amplifies a signal of data from the sensor or removes noise of the signal; And a data acquisition unit that acquires data measured by the sensor.

The data processing unit receives and stores the data acquired by the data acquisition and transmission unit, and a data storage unit for preprocessing the acquired data; It may include; a data extraction unit for extracting the core data for determining the presence or absence of the solenoid valve and the type of failure from the data pre-processed in the data storage unit.

The data analysis unit analyzes the extracted core data using a pre-trained artificial neural network, an artificial intelligence analysis unit that monitors and diagnoses the solenoid valve in real time; And a classical analysis unit for real-time monitoring and diagnosis of the solenoid valve by analyzing the extracted core data without using an artificial neural network.

The data analysis unit may further include a data output unit that outputs an analysis result by the AI analysis unit and the classic analysis unit in real time and generates an alarm when an abnormal waveform is detected.

When the sensor is a current sensor that senses the current of the solenoid valve, the pre-processing of the data processing unit differentiates the current data sensed from the sensor with respect to time, and the current data with respect to time is 0. The time interval variable can be obtained, and the average value of the current values for each time interval can be obtained.

The core data may be a time interval variable obtained by data preprocessing of the data processing unit or an average value of current values for a time interval.

The real-time monitoring method of a solenoid valve that monitors the abnormality of a solenoid valve using the artificial neural network according to the present invention for achieving the above object is to input learning data from a learning valve in a normal state and an abnormal state as input data for a predetermined time. , Learning the artificial neural network to analyze whether the solenoid valve is abnormal; Acquiring data sensed from the solenoid valve by a sensor connected to the solenoid valve; Pre-processing the acquired data; Extracting core data for determining whether a solenoid valve has a failure and a failure type from the pre-processed data; And inputting and analyzing the core data into the artificial neural network; and outputting the analysis result in real time.

In the step of learning the artificial neural network, the artificial neural network may be trained using the driving current data of the learning valve in the normal state and the abnormal state as the learning data.

Data sensed from the solenoid valve may be driving current data of the solenoid valve.

The pre-processing of the acquired data may be performed by differentiating the driving current data of the solenoid valve with respect to time, obtaining a time interval variable in which the driving current data with respect to time is 0, and each time interval It may be to obtain the average value of the current value for.

The core data may be a time interval variable obtained by pre-processing the acquired data or an average value of current values for a time interval.

After the step of inputting and analyzing the core data into the artificial neural network, the method may further include diagnosing a failure state by performing pattern comparison with the normal state and the state of each failure mode without using the artificial neural network. have.

The driving current data of the learning valve may be expressed as a voltage value proportional to the current as time series data.

The driving current data of the solenoid valve may be expressed as a voltage value proportional to the current as time series data.

The solenoid valve real-time monitoring system and the solenoid valve real-time monitoring method using an artificial neural network according to the present invention enable real-time failure monitoring of SOV using a pre-trained artificial neural network, and also analyzes SOV failure type and aging trend It has the effect of enabling analysis.

1 shows a conceptual diagram of an SOV real-time monitoring system according to the present invention.
2 is a view showing an SOV real-time monitoring system according to an embodiment of the present invention.
3 is a picture of the current sensor used in the SOV real-time monitoring system according to an embodiment of the present invention.
4 is a block diagram of a diagnostic unit of an SOV real-time monitoring system according to an embodiment of the present invention.
5 is a conceptual diagram for explaining an artificial neural network.
6 is a view showing the operation of the brain neural network and the activation function of the artificial neural network.
7 is a photograph showing a pneumatic valve test system.
8 is a photograph showing a valve test system for hydraulic pressure.
9 is a flow chart for a real-time SOV monitoring method according to an embodiment of the present invention.
FIG. 10 shows an example of a current signal waveform when operating an SOV acquired using a current sensor in a normal state of a DC driven SOV in an SOV real-time monitoring system according to an embodiment of the present invention.
11 is a graph showing a current signal modeled in a steady state of the DC driving SOV of FIG. 10 according to an embodiment of the present invention.
12 is a flowchart illustrating a method for real-time monitoring of SOV using an artificial neural network when operating a DC-driven SOV according to an embodiment of the present invention.
13 is a flowchart illustrating a method for real-time monitoring of SOV using a classic analysis method when operating a DC-driven SOV according to an embodiment of the present invention.
14 is a graph showing a current signal according to a failure type of a DC driven SOV.

The technique described below may be applied to various changes and may have various embodiments, and specific embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the technology described below to specific embodiments, and should be understood to include all modifications, equivalents, or substitutes included in the spirit and scope of the technology described below.

The SOV real-time monitoring system according to the present invention monitors and diagnoses the status of a running SOV in real time.

1 shows a conceptual diagram of an SOV real-time monitoring system 1 according to the present invention.

Referring to FIG. 1, the SOV real-time monitoring system 1 is a valve sensor for detecting voltage, current, and temperature data from SOVs (SOV # 1 to n, 100) and SOVs (SOV # 1 to n), respectively. It is connected to the modules (VSM # 1 ~ n, 200) and the valve sensor modules (VSM # 1 ~ n) to acquire, process, and analyze the data transmitted from the valve sensor modules (VSM # 1 ~ n). It may include a diagnostic unit 300 for monitoring and diagnosing the presence or absence of an abnormality.

At this time, the sensors provided in the valve sensor modules (VSM # 1 ~ n, 200) include a voltage sensor that senses the voltage of the SOV, a current sensor that senses the current, a temperature sensor that senses the temperature, and a heat sensor that senses the heat of the SOV. It may be any one or more of a variety of sensors, such as a thermal image sensor, an acoustic sensor that detects the operating sound of the SOV.

The diagnostic unit 300 may be a set of diagnostic devices for real-time monitoring and diagnosis of SOV, or may be a computer including a module that performs each function in one diagnostic device, but is not limited thereto.

2 is a view showing an SOV real-time monitoring system according to an embodiment of the present invention, and FIG. 3 is a picture of a current sensor used in an SOV real-time monitoring system according to an embodiment of the present invention.

As shown in FIG. 2, the SOV real-time monitoring system 1 according to an embodiment of the present invention is installed on the power line 110 that supplies power to the SOV 100 and the SOV 100, and senses current. It includes a sensor 200 and a diagnosis unit 300 connected to the current sensor 200 for monitoring and diagnosing the presence or absence of an SOV by acquiring, processing, and analyzing data transmitted from the current sensor 200.

In power plants, SOVs of AC 220V, DC 110V, DC 48V, and DC 24V operating power sources are generally used, and the use current is generally less than 1A. Therefore, it is possible to acquire the driving current data of the SOV by using an indirect measuring current sensor suitable for the SOV. Current sensors are largely divided into AC current sensors and DC current sensors. Several types of non-contact current sensors have been developed and used. 3 is a photograph of a hook-type current sensor, (a) is a hinge-type current sensor, and (b) is a sensor using a Hall effect. For example, a current sensor using the Hall effect can measure the current drawn into the SOV.

Therefore, the current sensor 200 may use the various current sensors described above for sensing the current of the power line 110 of the SOV 100, but is not limited thereto.

4 is a block diagram of a diagnostic unit of an SOV real-time monitoring system according to an embodiment of the present invention.

As shown in FIG. 4, the diagnosis unit 300 acquires signal data from a sensor installed in an SOV and transmits it to a subsequent data processing unit 320, the data acquisition and transmission unit 310, and the data acquisition and transmission unit Data processing unit 320 for extracting the core data by pre-processing the data to receive data from 310 and suitable for analyzing the data, and monitoring the SOV by analyzing the core data extracted by the data processing unit 320 and And a data analysis unit 330 for diagnosis.

The data acquisition and transmission unit 310 is a signal having a power supply unit 311 for supplying power to the current sensor and the signal processing circuit, a filter that amplifies the signal of the current sensor or removes noise (Noise Rejection Filter), etc. It includes a conditioning unit (Signal Conditioning Module, 312) and a data acquisition unit 313 for receiving data from the current sensor or the signal conditioning unit 312. Here, the data from the current sensor may be amplified by the signal adjustment unit 312, or noise may be removed and transmitted to the data acquisition unit 313, or directly to the data acquisition unit 313 from the current sensor. It might be.

The data processing unit 320 receives and stores the data acquired by the data acquisition and transmission unit 310 and reduces the dimension of the acquired data using a Principal Component Analysis (PCA) algorithm, or differentiates the data with respect to time. And a data storage unit 321 for preprocessing such as obtaining a time interval variable having a derivative value of 0, a history management unit 322 for recording acquired data and preprocessed data, and managing the history thereof, and the data storage unit ( And a data extracting unit 323 that extracts core data for determining the presence or absence and failure type of SOV from the data pre-processed in 321).

The data analysis unit 330 is an artificial intelligence analysis unit 331 that monitors the extracted core data in real time and monitors and diagnoses the state of SOV in real time using an existing artificial neural network (ANN), When the analysis results by the classical analysis unit 332 and the artificial intelligence analysis unit 331 and the classical analysis unit 332 are performed in real-time to monitor and diagnose SOV by performing an analysis that does not use an artificial neural network, and output an abnormal waveform in real time It includes a data output unit 333 for generating an alarm.

5 is a conceptual diagram for explaining an artificial neural network, and FIG. 6 is a diagram showing the operation of the brain neural network and the activation function of the artificial neural network.

As shown in Fig. 5, the artificial neural network is composed of layers composed of individual units (units, circles in the figure, commonly referred to as perceptrons), and each unit is connected to units of the same layer or different layers. have. This is called an 'artificial neural network' because each unit operates in a way that mimics a nerve cell, and the intensity of connections between units changes in a manner similar to that of synapses, which are connections between nerve cells. As shown in FIG. 6, the unit of the next layer of the artificial neural network receives input of the units of the previous layer, similar to the neural network of a real organism. Here, the 'connection strength' corresponds to the 'synaptic efficiency', and the stronger the connection strength, the greater the influence of the input from the previous unit. If the sum of the inputs from the previous floor units exceeds a threshold, the output is output as much as the sum of the input sums. Changing the link strength between these units results in different outputs even when the same input is received. This is the key to learning in the neural network. This is one of the important factors that change when Algogo learns while simulating many large countries. The basic principle of learning an artificial neural network is to adjust the connection strength of a neural network according to an error between an output and a correct answer. Back propagation algorithms are commonly used. That is, if the output result is stronger than the correct answer, the connections that led to the corresponding output are weakened, and if it is weaker than the correct answer, the connections that derive the corresponding output are strengthened. At this time, if the size of the error between the output and the correct answer is large, the strength of the connections is changed a lot, and if the size of the error is small, it is slightly changed.

The artificial intelligence analysis unit 331 using the artificial neural network according to the present invention includes an input layer, a hidden layer, and an output layer. Learning can be done using sets. This can be done using a back propagation algorithm that updates synaptic weights to reduce the error rate or loss between the current output and the target output corresponding to the input. This weight can be optimized by being updated several times according to the learning process according to the inverse propagation algorithm.

To train an artificial neural network, the process of learning using millions of input / output pairs has to be repeated several times. This started to be solved with the advent of a graphics processing unit (GPU) that rapidly processes large amounts of information. Currently, cloud services such as Microsoft's Azure, Amazon machine learning, and Google cloud machine learning can perform artificial neural network learning using cloud computing.

Currently, various types of artificial neural networks have been developed and studied, such as Deep Neural Network (DNN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Restricted Boltzmann Machine (RBM), Deep Belief Network (DBN), etc. Is not limited to a specific artificial neural network.

Data used for learning of an artificial neural network is called learning data, and an artificial neural network is constructed using learning data.

The learning data used in the SOV real-time monitoring system 1 according to the present invention can be collected using various valve test systems. FIG. 7 shows a pneumatic valve test system, and FIG. 8 is a photograph showing a hydraulic pressure valve test system. The learning data according to the present invention may be collected using a pneumatic valve test system or a hydraulic valve test system shown in FIGS. 7 and 8.

The valve used to train the artificial neural network in the valve test system is called a learning valve. The learning valve can be divided into a normal valve and a fault valve that has caused a fault in advance according to the fault type. Artificial neural networks can also be built using different models depending on the type of valve. That is, different models may be provided in consideration of the manufacturer and the model year. Also, in some cases, a single artificial neural network may be constructed using various learning data.

The artificial intelligence analysis unit 331 includes an artificial neural network learned by receiving learning data. For learning the artificial neural network, various learning data acquired from the learning valve are continuously acquired for a certain period of time. At this time, the artificial intelligence analysis unit 331 analyzes the extracted data using the artificial neural network.

At this time, the current signal waveform of the SOV current sensor acquired by the data acquisition and transmission unit 310 is different depending on the SOV operating in DC and the SOV operating in AC. Therefore, the input for training the artificial neural network of the artificial intelligence analysis unit 331 is different to implement the sound diagnostic method of SOV and the diagnostic method.

9 is a flowchart of a method for real-time monitoring of SOV according to an embodiment of the present invention.

First, an artificial neural network is trained and built to monitor SOV in real time (S100).

For example, in the SOV real-time monitoring system 1 including a current sensor, the artificial neural network training and construction method may include steps 1 to 3 below.

Step 1: Acquire SOV drive current data (learning data) continuously for a period of time from the learning valve. The current data may be expressed as a voltage value proportional to the current as time series data. At this time, it also secures the current data of the fault valve that matches the defined fault type.

Step 2: Train an artificial neural network using the learning data. At this time, it can be learned to apply the clustering technique that can distinguish the pattern of the normal valve and the failure valve.

Step 3: The learned artificial neural network is mounted on the artificial intelligence analysis unit 331 and used in the SOV real-time monitoring system 1.

Then, the driving current data of the real-time SOV sensed from the valve sensor module is acquired (S110).

Thereafter, the acquired data is stored, and pre-processing of the data is performed (S120).

Thereafter, core data for determining whether a SOV has a failure and a failure type is extracted from the preprocessed data (S130).

Thereafter, the core data is input to the artificial neural network installed in the artificial intelligence analysis unit for analysis (S140).

At this time, the core data may be additionally input and analyzed in the classical analysis unit (S150). The classical analysis unit is an analysis unit for diagnosing whether or not a failure occurs by performing a pattern comparison with a normal state and a state for each failure mode without using an artificial neural network.

The analysis result is output in real time (S160).

At this time, the SOV real-time monitoring system 1 can display the diagnosis result on the display screen. In addition, the SOV real-time monitoring system 1 can also deliver diagnostic results to other devices. For example, the SOV real-time monitoring system 1 can deliver the diagnosis result to a user terminal such as a smartphone. The SOV real-time monitoring system 1 can deliver the diagnosis results to the management server. The user can check the diagnosis result by accessing the management server through the application on the smartphone.

If the use of SOV is not finished (NO in S170), the SOV real-time monitoring system 1 acquires sensing data from the current sensor again and repeats the above-described process. Meanwhile, the SOV real-time monitoring system 1 can update the algorithm of the artificial neural network. For example, after diagnosing a malfunction using an artificial neural network, the current collected result may be reflected in the artificial neural network again. That is, the current artificial neural network can be further trained using data acquired using the SOV real-time monitoring system 1. Furthermore, the SOV real-time monitoring system 1 can store the detected failure record.

FIG. 10 shows an example of a current signal waveform when operating an SOV acquired using a current sensor in a normal state of a DC driven SOV in the SOV real-time monitoring system 1 according to an embodiment of the present invention.

As shown in FIG. 10, the current signal waveform during SOV operation is the current signal during the starting current section (A) in which the SOV is started, and the current signal during the holding current section (B) in which the current signal waveform of the SOV is kept constant. And a current signal during the return current section C in which the current signal waveform of the SOV decreases. Therefore, in the current signal waveform, the pick-up time / current of the starting current section (A), the holding current section (B), and the return current section (C), and the current value after pick-up change to the minimum Check the dividing time and the drop-out time / current for each section.

In addition, the current signal waveform during operation of the SOV acquired through the field test can be compared with the current signal waveform during operation of the SOV in the normal state to determine whether the change is made, thereby identifying the failure type of the SOV. At this time, whether or not the current signal waveform changes can be easily determined by the user's naked eye, but the current signal must be pre-processed to be automatically monitored and diagnosed by the SOV real-time monitoring system 1.

11 is a graph showing a current signal modeled in a steady state of the DC driving SOV of FIG. 10 according to an embodiment of the present invention. Specifically, FIG. 11 shows current values (indicated by solid lines, NORMAL CURRENT) over time of SOV, and derivatives (indicated by dotted lines, DIFFERETIATED) with respect to time of current.

Here, the time at which the differential value for the time of the current becomes 0 is determined for each section. In other words, the time interval from the first zero value to the second zero value is t1, and the second zero time to the third zero value is t2. Let t be the time interval from the third zero value to the fourth zero value, and the time interval from the fourth zero value to the fifth zero value is t4. Let t5 be the time interval from the value from 0 to the 6th value, and 6 from 0 to the 7th value from 0. The time interval from this value to the eighth value is set to t7, and the value obtained by adding t1 to t7 together is called T_total. Therefore, the graph of the differential value with respect to the time of the current can be divided into eight time periods from t1 to t7 and T_total. Using these t1 to t7 and T_total as time interval variables, the average value of the time interval variables of the SOV and the current values in each time interval is determined and compared with the normal state of the SOV to determine whether the SOV is in a normal state, a fault, or an abnormal state. Can be distinguished. At this time, the time interval variables such as t1 to t7 and T_total and the average value of the current values in each time interval are the key data. Also, the error tolerance range of the average value of the time interval variable and the current value is determined differently according to the monitored SOV, and if the average value of the time interval variable or current value is included in the allowable range, it can be grasped as a normal state. For example, if the allowable error range is 5%, if the tolerance of the time interval variable is within 5% compared to the steady state time interval variable, it can be regarded as a steady state. In addition, if the tolerance of the average value of the current value is within 5% compared to the average value of the current value in the corresponding steady state time period, it can be regarded as a steady state.

In one embodiment of the present invention, eight time interval variables of t1 to t7 and T_total were determined, but the number and type of the time interval variables may vary depending on the type of SOV being monitored and the type of valve sensor coupled to the SOV. have. Also, a kind of energy parameter integrating the current-time graph can be used as data for learning the artificial neural network. Accordingly, the core data can be set differently.

12 is a flowchart illustrating a method for real-time monitoring of SOV using an artificial neural network when operating a DC-driven SOV according to an embodiment of the present invention.

First, in step S200, an instantaneous current value (current data) of SOV is acquired.

In step S210, as a pre-process for the acquired data, the instantaneous current value of the acquired SOV is differentiated. For example, the derivative value can be obtained by taking a derivative over time against an instantaneous current value.

In step S220, time periods are determined based on a time point at which the differential value is zero to extract key data for determining the presence or absence and failure type of SOV, and an average value of current values in each time period is obtained.

The average value of the current values can be obtained by the following equation.

For example, in the graph shown in FIG. 11, the average value of the current value in the time period tn

이다

to be

In step S230, data is analyzed by inputting an average value of a time interval and a current value in each section into a previously learned artificial intelligence analysis unit.

Thereafter, in step S240, the artificial intelligence analysis unit exclusively determines and outputs at least one of the normal mode and the failure mode based on the analyzed data.

13 is a flowchart illustrating a method for real-time monitoring of SOV using a classic analysis method when operating a DC-driven SOV according to an embodiment of the present invention.

First, in step S300, the instantaneous current value of SOV is acquired.

Thereafter, in step S310, the obtained instantaneous current value is differentiated. For example, the derivative value can be obtained by taking a derivative over time against an instantaneous current value.

Thereafter, in step S320, time intervals are determined based on a time point at which the derivative value is zero.

Subsequently, in step S330, the size of the time and current for each section is compared with the normal state and the pattern for each failure mode.

Thereafter, in step S340, the SOV may be determined as at least one of the failure modes from the pattern comparison result. For example, if the magnitude of time and current for each section is within a threshold, a failure mode is output based on this.

The real-time SOV monitoring system according to the present invention monitors and diagnoses whether the SOV is in a normal mode or a failure mode using an artificial intelligence analysis unit, and may additionally use a classical analysis unit. At this time, data analysis results from the artificial intelligence analysis unit and the classical analysis unit may be synthesized to determine whether a real-time failure condition of the SOV occurs.

In the present invention, various time series signals collected from DC driven SOV current signals are classified into normal and abnormal signals using various clustering algorithms, and the abnormal signals are also classified into various failure types in FIG. 14.

14A shows the current waveform when the SOV flange is stuck (plunger stick), FIG. 14B shows the current waveform when the flange reaction is delayed (plunger drag), and FIG. 14C shows the SOV leak (leak) ) Shows the current waveform, FIG. 14D shows the current waveform when the SOV coil is short, and FIG. 14E shows the current waveform when the SOV spring is deteriorated (weak spring), and FIG. 14F It shows the current waveform when the coil of SOV is aged.

At this time, the AI analysis unit may include an artificial neural network trained to cluster (classify) any of the six abnormal signals in FIG. 14 using the K-means algorithm. In addition, by learning the artificial neural network through a deep learning algorithm on data according to the current signal waveform, it is possible to detect abnormal signals by analyzing signals generated in the future in real time.

In one embodiment of the present invention, SOV is monitored and diagnosed in real time based on current data using a current sensor, but an artificial neural network learned using learning data such as the voltage sensor, temperature sensor, thermal image sensor, and acoustic sensor described above. It is also possible to monitor and diagnose SOV in real time using this installed AI analysis unit.

The inspection of the existing solenoid valve has been performed through SOV excitation coil resistance measurement, insulation resistance measurement, operation test, and leakage check. In order to analyze the soundness of SOV, the field system was separated from the SOV and the soundness was determined using a general measuring instrument. However, as described above, the SOV real-time monitoring system according to the present invention measures the current waveform driving the SOV using a non-contact current sensor, and then uses the measured current waveform to perform diagnostic analysis to separate it from the field system. There is no need for monitoring at all times. With the existing test method, it is possible to check only the soundness of the coil, the operation of the SOV, and whether or not it is leaked, and it is impossible to identify the detailed cause. SOV real-time monitoring system according to an embodiment of the present invention, in addition to the basic problems as described above, the mechanical problems inside the solenoid valve (for example, a flange stick (Plunger Stick), delay (Drag), spring deterioration (Weak Spring) and Diagnosis such as coil aging is possible. The aging of the solenoid valve cannot be judged only by the general coil resistance, insulation resistance, and operation test, and it is difficult to establish a pre-replacement time. In the case of using the apparatus of one embodiment of the present invention, it is possible to pre-check and replace the solenoid valve by notifying the coil gradation, spring spring, etc. through aging trend analysis in advance. In addition, using the previously learned artificial neural network, it informs the failure type of SOV, the presence of failure, and the possibility of future failure. In the case of a nuclear power plant or a thermal power plant, SOV inspection is possible only during the planned preventive maintenance period, so it is impossible to identify the SOV problem until a major problem that affects the system occurs due to the failure of the SOV. However, in the case of the developed product, the status of SOV is monitored in real time and an alarm is alerted when a problem occurs, so that a quick response is possible.

In the above, specific preferred embodiments of the present invention have been described, but the present invention is not limited to the specific embodiments described above, and the technical field to which the present invention pertains without departing from the gist of the present invention claimed in the claims. In addition, various modifications can be carried out by anyone with ordinary knowledge in the art, and such modifications are within the scope of the claims.

1: SOV real-time monitoring system 100: SOV
200: valve sensor module 300: diagnostic unit
310: data acquisition and transmission unit 311: power supply unit
312: signal conditioning unit 313: data acquisition unit
320: data processing unit 321: data storage unit
322: history management unit 323: data extraction unit
330: data analysis unit 331: artificial intelligence analysis unit
332: classic analysis unit 333: data output unit

Claims (16)

Solenoid valve real-time monitoring system that monitors the abnormality of the solenoid valve using an artificial neural network,
A sensor connected to the solenoid valve and measuring data sensed from the solenoid valve;
Includes a diagnostic unit connected to the sensor to monitor and diagnose the presence or absence of the solenoid valve by acquiring, processing and analyzing the data measured by the sensor.
The diagnostic unit,
A data acquisition and transmission unit that acquires data measured by the sensor from the sensor connected to the solenoid valve and transmits the acquired data to a data processing unit;
A data processing unit that receives data from the data acquisition and transmission unit and preprocesses it to be suitable for analyzing the data to extract core data; And
Includes a data analysis unit that analyzes the core data extracted from the data processing unit using a pre-trained artificial neural network to monitor and diagnose the solenoid valve in real time.
The artificial neural network is a solenoid valve real-time monitoring system that is trained to analyze the abnormality of a solenoid valve by receiving learning data as input data from a learning valve in a normal state and an abnormal state for a predetermined time. The method of claim 1, wherein the sensor, a voltage sensor for sensing the voltage of the solenoid valve, a current sensor for sensing the current of the solenoid valve, a temperature sensor for sensing the temperature of the solenoid valve, a thermal image sensor for sensing the heat of the solenoid valve , Solenoid valve real-time monitoring system comprising any one of the sound sensor for sensing the operation sound of the solenoid valve. In claim 1,
The data acquisition and transmission unit is a power supply for supplying power to the sensor; A signal adjusting unit having a filter that amplifies a signal of data from the sensor or removes noise of the signal; And a data acquisition unit that acquires data measured by the sensor. In claim 1,
The data processing unit receives and stores the data acquired by the data acquisition and transmission unit, and a data storage unit for preprocessing the acquired data; Solenoid valve real-time monitoring system comprising a; data extraction unit for performing extraction of the key data to determine the presence or absence of failure and the type of failure of the solenoid valve from the data pre-processed in the data storage unit. In claim 1,
The data analysis unit analyzes the extracted core data using a pre-trained artificial neural network, an artificial intelligence analysis unit that monitors and diagnoses the solenoid valve in real time; And a classical analysis unit for real-time monitoring and diagnosis of the solenoid valve by analyzing the extracted core data without using an artificial neural network. In claim 5,
The data analysis unit outputs the analysis results by the AI analysis unit and the classic analysis unit in real time, and further includes a data output unit that generates an alarm when an abnormal waveform is detected, solenoid valve real-time monitoring system. In claim 1,
When the sensor is a current sensor that senses the current of the solenoid valve, the pre-processing of the data processing unit differentiates the current data sensed from the sensor with respect to time, and the current data with respect to time is 0. A real-time monitoring system for solenoid valves that acquires time interval variables and obtains an average value of current values for each time interval. In claim 7,
The core data is a solenoid valve real-time monitoring system that is an average value of current values for time interval variables or time intervals obtained by preprocessing data of the data processing unit. A real-time monitoring method of a solenoid valve that monitors the abnormality of a solenoid valve using an artificial neural network,
Learning the artificial neural network to analyze the abnormality of the solenoid valve by inputting learning data from the steady state and abnormal state learning valves as input data for a predetermined time;
Acquiring data sensed from the solenoid valve by a sensor connected to the solenoid valve;
Pre-processing the acquired data;
Extracting core data for determining whether a solenoid valve has a failure and a failure type from the pre-processed data;
Analyzing the input of the core data into the artificial neural network; and
Solenoid valve real-time monitoring method comprising the step of outputting the analysis results in real time. In claim 9,
In the step of learning the artificial neural network, a solenoid valve real-time monitoring method of learning an artificial neural network by using driving current data of a learning valve in a normal state and an abnormal state as the learning data. In claim 10,
A method for real-time monitoring of a solenoid valve in which data sensed from the solenoid valve is drive current data of the solenoid valve. In claim 11,
The step of pre-processing the acquired data is to differentiate the driving current data of the solenoid valve with respect to time, obtain a time interval variable in which the driving current data is differentiated with respect to time, and each time interval is obtained. Solenoid valve real-time monitoring method to obtain the average of the current values for. In claim 12,
The core data is a solenoid valve real-time monitoring method that is an average value of a current value for a time period variable or a time period obtained by pre-processing the acquired data. In claim 9,
Solenoid further comprising the step of diagnosing the failure by performing the pattern comparison with the normal state and the state of each failure mode without using the artificial neural network after inputting and analyzing the core data into the artificial neural network. Valve real-time monitoring method. In claim 10,
The driving current data of the learning valve is a time-series data, solenoid valve real-time monitoring method represented by a voltage value proportional to the current. In claim 11,
The driving current data of the solenoid valve is a time-series data, a real-time monitoring method of a solenoid valve expressed as a voltage value proportional to current.

Images