KR20210108168A - A diagnosis method of malfunction of pumps that is based on machine-learning - Google Patents

Abstract

The present invention relates to a machine learning-based pump failure diagnosis monitoring method that generates learning result information for fault diagnosis that can determine whether a diagnostic target pump is normal or abnormal and diagnoses the failure of the diagnostic target pump by using the generated learning result information, but accurately diagnoses even a specific failure part in a diagnostic target pump through the precise diagnosis of the diagnostic target pump using various sensors.

Description

A diagnosis method of malfunction of pumps that is based on machine-learning}

The present invention relates to a method for monitoring a pump failure diagnosis based on machine learning, and more particularly, to a pump for diagnosis using the generated learning result information after generating failure diagnosis learning result information that can determine whether the pump is normal or abnormal. It is an invention related to a machine learning-based pump failure diagnosis monitoring method for diagnosing the failure of

In particular, the present invention relates to a machine learning-based pump failure diagnosis monitoring method for accurately diagnosing even a specific faulty part of a diagnosis target pump through precise diagnosis of a diagnosis target pump using various sensors.

In general, most of the water we use every day in real life is tap water supplied from a water supply facility, and groundwater is still used in some limited areas.

Since environmental pollution was not severe before the start of industrialization in Korea, drinking water in its natural state was not a problem.

On the other hand, the pump is a device that works to pull up water from the underground to the ground or to supply water stored in a water tank to the outside.

At this time, the types of pumps are largely divided into a centrifugal pump using rotational force and a positive displacement pump using a piston reciprocating motion, and are subdivided according to use and operation method. Among them, the centrifugal pump is the most widely used.

However, if the centrifugal pump is used for a long period of time, failures such as bearing defects, shaft bending, misalignment, and water leakage may occur. possibility arises.

Therefore, there is a need to maintain the reliability of the centrifugal pump through maintenance and repair by periodically performing a performance test to diagnose a faulty part, and to replace the part when a faulty part is found.

The following are prior art related to this.

1. Republic of Korea Patent Publication No. 10-1247337 Vacuum pump testing device 2. Republic of Korea Patent Publication No. 10-2017-0030823 Nuclear power plant pump reliability evaluation test apparatus and method 3. Republic of Korea Patent Publication No. 10-1745103 High-pressure pump test device

The present invention is to solve the above problems,

The present invention provides a machine learning-based pump failure diagnosis monitoring method for diagnosing a failure of a diagnostic target pump using the generated learning result information after generating failure diagnosis learning result information that can determine whether the pump is normal or abnormal. aim to

Another object of the present invention is to provide a machine learning-based pump failure diagnosis monitoring method for accurately diagnosing even a specific faulty part of the diagnosis target pump through precise diagnosis of the diagnosis target pump using various sensors.

In order to achieve the above object, the present invention's machine learning-based pump failure diagnosis monitoring method,

a learning result information generation step (S1000) of generating learning result information for fault diagnosis that can determine whether the pump is normal or abnormal using the pump demonstration system;

and a failure diagnosis step (S2000) of diagnosing a failure of a target pump for failure diagnosis by using the generated failure diagnosis learning result information.

Since the present invention diagnoses the failure of the diagnostic target pump by using the generated learning result information after generating the failure diagnosis learning result information that can determine whether the pump is normal or abnormal, the failure of the diagnostic target pump is quickly and accurately identified. so that maintenance can be carried out in a timely manner.

In addition, the present invention accurately diagnoses even a specific faulty part of the diagnostic target pump through precise diagnosis of the diagnostic target pump using various sensors. Otherwise, it avoids the cost of replacing the entire pump without knowing the part of the failure.

1 is a flowchart of the present invention;
2 is a flowchart of the learning result information generation step of the present invention;
3 is a schematic diagram of a normal pump demonstration system of the present invention;
4 is a block diagram of a normal pump demo system of the present invention;
5 is a state diagram in which each sensor is installed in a normal pump demonstration system of the present invention;
6 is a schematic diagram of an abnormal pump demonstration system of the present invention;
7 is an abnormal pump demonstration system configuration block diagram of the present invention
8 is a flowchart of the fault diagnosis step of the present invention;

An embodiment of the present invention will be described in detail with reference to the accompanying Figures 1 to 8.

The machine learning-based pump failure diagnosis monitoring method (hereinafter the method) of the present invention generates failure diagnosis learning result information that can determine whether the pump 20 is normal or abnormal, and then diagnoses the failure using the generated learning result information. Since it is a machine learning-based pump 20 failure diagnosis monitoring method for diagnosing the failure of the target pump 20, the maintenance of the pump 20 is performed in a timely manner by quickly and accurately identifying the failure of the target pump 20. It is an invention that can prevent partial maintenance because it is an invention to do so.

Therefore, the method of the present invention, as shown in Fig. 1, largely includes a learning result information generation step (S1000) and a failure diagnosis step (S2000).

Specifically, the machine learning-based pump failure diagnosis monitoring method of the present invention comprises:

a learning result information generation step (S1000) of generating learning result information for fault diagnosis that can determine whether the pump is normal or abnormal using the pump demonstration system;

and a failure diagnosis step (S2000) of diagnosing a failure of a target pump for failure diagnosis by using the generated failure diagnosis learning result information.

Referring to FIG. 1 , the learning result information generation step ( S1000 ) is a step of generating learning result information for fault diagnosis that can determine whether the pump 20 is normal or abnormal using the pump demonstration system 10 . The failure diagnosis learning result information is used to diagnose the failure of the failure diagnosis target pump in the failure diagnosis step (S2000).

The learning result information generation step (S1000) includes a first step (S100), a second step (S200), and a third step (S300) in detail.

Specifically, the learning result information generation step (S1000) is,

After building a normal pump demonstration system, a first step (S100) of acquiring normal raw data for each sensor of a normal pump;

After building the abnormal pump demonstration system, the second step (S200) of acquiring abnormal raw data for each sensor of the abnormal pump for each failure part;

A third step (S300) of generating learning result information for diagnosing pump failure that can determine whether the pump is normal or abnormal using the acquired normal raw data for each sensor and the abnormal raw data for each sensor acquired for each failed part characterized.

Referring to FIG. 2 , the first step (S100) is a step of acquiring normal raw data for each sensor of a normal pump 20 after building a normal pump demonstration system. Step 1-1 (S110) and the first step 1-2 steps (S120) are included.

Specifically, the first step (S100) is,

Step 1-1 (S110) of installing sensors at each position of the normal pump demonstration system;

After the sensor is installed, the pump demo system is operated to obtain normal raw data for each sensor ( S120 ).

Step 1-1 ( S110 ) is a step of constructing a normal pump demonstration system 10 to obtain normal raw data of the normal pump 20 .

Here, the normal pump demonstration system 10 measures the vibration generated during normal pump 20 operation, the flow rate passing through the pipe, the pressure difference (differential pressure) between the front and rear ends of the normal pump 20, and the current of the motor. It is a system that allows to check the performance and operating state of the normal pump 20 .

Therefore, the normal pump demonstration system 10, as shown in FIGS. 3 and 4, includes a normal pump 20, an acceleration sensor 100, a flow sensor 200, a pressure sensor 300, a current sensor 400, and a water tank ( 500 ), a data collection unit 600 , and a pipe 700 .

That is, the normal pump demonstration system 10 connects the normal pump 20 to the water tank 500 using a pipe 700 , and an acceleration sensor 100 and a flow rate sensor 200 at each corresponding position of the pipe and the normal pump. , the pressure sensor 300 and the current sensor 400 are installed.

Specifically, the acceleration sensor 100 is installed in the impeller case of the pump to measure the vibration of the pump, and the flow sensor 200 is installed in the upper pipe sufficiently spaced apart from the pump outlet to measure the operating flow rate, and the The pressure sensor 300 is installed in the piping near the pump inlet and outlet for measuring the differential pressure, and the current sensor 400 is installed in the hotline of the electric wire to measure the power consumed by the pump.

The installation positions of the sensors will be described in more detail.

Referring to Figure 1 of FIG. 5 , the acceleration sensor 100 is a sensor for measuring vibration generated when the pump 20 operates. This is the outside of the impeller case of the pump 20 that is generated greatly, and a total of three acceleration sensors 100 are attached to the outside of the impeller case in the X, Y, and Z axis directions.

At this time, since the uniaxial stud mount method is applied to the acceleration sensor 100 to obtain a high frequency response, when the pump 20 operates, the data measured by the plurality of acceleration sensors 100 is transferred to the data collection unit 600 . ), and the data collection unit 600 easily collects vibration data for each axis of the pump 20 .

5, the flow sensor 200 is a sensor for measuring the flow rate passing through the pipe, and the installation position of the flow sensor 200 is that the flow rate is within the pipe after the pump 20 operates. It is an upper pipe located at a predetermined distance from the outlet of the pump 20 so as to be sufficiently expressed.

Therefore, when the pump 20 is operated, the flow sensor 200 transmits data measuring the flow rate passing through the pipe to the data collection unit 600 .

Referring to the third figure of FIG. 5 , the pressure sensor 300 is a sensor for measuring the pressure generated in the pipe while the pump 20 operates, and in more detail, the front end inlet of the pump 20 . The pressure sensor 300 is installed at the inlet side of the pump 20 to measure the difference between the suction pressure generated at the inlet and the discharge pressure generated at the discharge port at the rear end of the pump 20 (hereinafter, 'differential pressure'). It is configured to include a first pressure sensor 310 to be installed, and a second pressure sensor 320 to be installed in the discharge port side pipe.

At this time, as shown in the drawing, the first pressure sensor 310 and the second pressure sensor 320 are installed in the pipe at an angle of 45 degrees or less from the vertical line from the floor for accurate measurement.

5, the current sensor 400 is a sensor that measures the current of the motor generated when the pump 20 operates, and the current sensor 400 is the pump 20 (specifically, It is installed on one side of the external power hotline (110v, 220V, 380V, 440V power line) connected to the motor).

Referring to Figure 5 of FIG. 5 , the data acquisition unit 600 is connected to each sensor, and when the normal pump 20 operates, a type of DAQ (Data Acquisition) that collects and stores various measurement data from each sensor it is a device

The first and second steps (S120) operate the normal pump 20 and the normal pump demonstration system 10 in which the sensors 100, 200, 300, and 400 are installed to operate the normal pump 20. This is the step of acquiring normal raw data for each sensor.

In this case, the normal raw data acquisition in step 1-2 ( S120 ) is repeatedly performed a set number of times to increase reliability by reducing a measurement error of the acquired data.

The second step (S200) is a process of acquiring the abnormal raw data for each sensor of the abnormal pump for each faulty part after building the abnormal pump demonstration system 11 .

The abnormal construction of the pump demonstration system 11 means that a specific part of the normal pump 20 installed in the above-described normal pump demonstration system 10 is broken.

For example, an abnormal pump demonstration system 11 is constructed by replacing a part installed in a specific part of the normal pump 20 with an abnormal part.

The second step (S200) includes a step 2-1 (S210) and a step 2-2 (S220), as shown in FIG. 2 .

Specifically, the second step (S200) is,

Step 2-1 (S210) of artificially malfunctioning a specific part of the pump constituting the normal pump demonstration system to build an abnormal pump demonstration system;

Including the second-2 step (S220) of acquiring the abnormal raw data for each sensor by operating the built-up abnormal pump demonstration system,

Step 2-1 (S210) and step 2-2 (S220) are repeatedly performed by changing a specific part of the pump to be broken, and step 2-2 (S220) is repeated as many times as set for each failure part. characterized by being

The step 2-1 (S210) artificially malfunctions a specific part of the normal pump 20 installed in the normal pump demonstration system built through the first step (S100) to make the abnormal pump 21, as shown in FIG. As a step of building the abnormal pump demonstration system 11, the part of the normal pump 20 that is artificially broken may be a part where failure frequently occurs, such as a mechanical seal, a bearing, an impeller, a shaft, and the like.

For example, after detaching the mechanical seal installed in the normal pump 20 , it is worn and damaged, and then installed in the normal pump 20 again to make the abnormal pump 21 .

Here, the abnormal pump demonstration system 11 measures the vibration generated when the abnormal pump 20 operates, the flow rate passing through the pipe, the pressure difference (differential pressure) between the front and rear ends of the abnormal pump 21, and the current of the motor. It is a system that allows to check the performance and operating state of the abnormal pump 21 .

Therefore, the abnormal pump demonstration system 11 is, as shown in FIGS. 6 and 7, an abnormal pump 21, an acceleration sensor 100, a flow sensor 200, a pressure sensor 300, a current sensor 400, a water tank ( 500 ), a data collection unit 600 , and a pipe 700 .

That is, the abnormal pump demonstration system 11 replaces the normal pump 10 constituting the above-described normal pump demonstration system 10 with the abnormal pump 21, so that the acceleration installed at each corresponding position of the pipe and the abnormal pump. The installation positions of the sensor 100 , the flow sensor 200 , the pressure sensor 300 , and the current sensor 400 are the same as in the normal pump demonstration system 10 .

Step 2-2 (S220) is a step of acquiring abnormal raw data for each sensor by operating the abnormal pump 21 installed in the abnormal pump demonstration system 11 built through the step 2-1 (S210).

In particular, the step 2-1 (S210) and the step 2-2 (S220) are repeatedly performed by changing a specific part of the pump to be broken, and the step 2-2 (S220) is performed by the number of times set for each failed part. It is characterized in that it is performed repeatedly.

For example, after constructing an abnormal pump demonstration system 11 including an abnormal pump in which part A is malfunctioning, the abnormal pump in part A is operated to acquire abnormal raw data for each sensor, but repeatedly for a set number of times. . At this time, the abnormal raw data for each sensor that is repeatedly acquired becomes the abnormal raw data for each sensor matched to the faulty part A.

Next, after building the abnormal pump demonstration system 11 including the abnormal pump in which part B is malfunctioning, the abnormal pump in part B is operated to acquire abnormal raw data for each sensor, but repeatedly acquired by a set number of times. At this time, the abnormal raw data for each sensor that is repeatedly acquired becomes the abnormal raw data for each sensor matched to the faulty part B.

This process is performed with different faulty parts such as faulty part C, faulty part D, and faulty part E.

In this case, the reason for repeating step 2-2 ( S220 ) for the number of times set for each faulty part is to increase reliability by reducing measurement error of acquired data.

If step 2-1 (S210) and step 2-2 (S220) are performed as described above, it is possible to acquire abnormal raw data for each sensor for each defective part.

In summary, the purpose of the above-described first step (S100) is to acquire normal raw data of the sensors through a normal pump demonstration system, and the purpose of the above-described second step (S200) is to use an abnormal pump demonstration system for each failure part. Since the abnormal raw data of the sensors is acquired, learning result information for diagnosing pump failure can be generated through the normal raw data and the abnormal raw data for each sensor in a third step (S300), which will be described later.

The third step (S300) is performed using the normal raw data for each sensor acquired through the first step (S100) and the abnormal raw data for each sensor acquired for each faulty part through the second step (S200). As a step of generating learning result information for diagnosing a pump failure that can determine an abnormality, the pump 20 in the failure diagnosis step (S2000) using the learning result information for diagnosing a pump failure generated through the third step (S300). to diagnose the fault.

In particular, the abnormal raw data for each sensor obtained through the second step (S200) is characterized in that the abnormal raw data for each sensor is generated for each defective part, and the characteristic that the abnormal raw data for each sensor is generated for each defective part is characterized in that the second step (S200) Detailed description in the description part (Step 2-1 (S210) and Step 2-2 (S220) are repeatedly performed by changing a specific part of the pump to be broken, and Step 2-2 (S220) is a failure repeated as many times as set for each part), a detailed description thereof will be omitted.

Referring to FIG. 2, the third step (S300) includes steps 3-1 (S310), 3-2 (S320), 3-3 (S330), and 3-4 (S340). consists of including

Specifically, the third step (S300) is,

Step 3-1 (S310) of extracting characteristic values for each normal sensor from the acquired normal raw data for each sensor;

Step 3-2 (S320) of extracting characteristic values for each abnormal sensor for each faulty part from the abnormal raw data for each sensor acquired for each faulty part (S320);

Step 3-3 of determining the normal characteristic value range for each sensor using the extracted characteristic value for each normal sensor and the characteristic value for each abnormal sensor extracted for each faulty part, and determining the abnormal characteristic value range for each sensor for each faulty part ( S330) and

and a step 3-5 of generating learning result information for fault diagnosis by using the determined characteristic value ranges for each sensor (S340).

The step 3-1 (S310) is a step of extracting a characteristic value for each sensor from the normal raw data for each sensor obtained through the first step (S100).

That is, in the 3-1 step (S310), the pump vibration (specifically, the vibration of the impeller case) measured by the acceleration sensor 100 when the pump 20 in a steady state is operated, data, the flow sensor 200 The characteristic value for each sensor is obtained from the normal raw data for each sensor including the flow data of the pipe measured by This is the extraction step.

At this time, the characteristic values extracted from the normal raw data are Peak-to-Peak, Standard Variation, Mean, and Root Mean Square for each set time interval of the normal raw data. ), crest factor, skewness, kurtosis, and frequency of max amplitude, the present invention provides a model with the highest classification accuracy for each set time interval Standard deviation is used.

Peak-to-Peak, Standard Variation, Mean, Root Mean Square, crest factor, skewness, kurtosis ( kurtosis) and the frequency of max amplitude are used as characteristic values. Because the data volume of normal raw data for each sensor received in real time is huge, it is not practically easy to compare all data.

Therefore, normal raw data for each sensor received in real time is sampled for each set time section, and peak-to-peak values, standard deviation, mean, and square of the sampled data for each time section are sampled. Comparing the root mean square, crest factor, skewness, kurtosis, and frequency of max amplitude as characteristic values, it is possible to shorten the time to compare massive data. can have an effect.

In particular, in the present invention, the standard deviation for each time section set as a characteristic value is used, which is a peak-to-peak value, a mean, a root mean square, and a crest factor. ), skewness, kurtosis, and frequency of max amplitude when comparing data, the comparison accuracy is the highest. The higher the accuracy, the higher the diagnostic accuracy.

In step 3-2 (S320), in the same manner as in step 3-1 (S310), the characteristic value of each abnormal sensor for each faulty part is obtained from the abnormal raw data for each sensor acquired for each faulty part through the second step (S200) in the same manner as in step 3-1 (S310). It is an extraction process.

The characteristic value for each abnormal sensor extracted for each faulty part in step 3-2 (S320) is also a standard deviation, and the specific characteristic thereof has been described in step 3-1 (S310), and detailed description thereof will be omitted.

In step 3-3 (S330), the normal characteristic value range for each sensor is determined using the extracted characteristic value for each normal sensor and the characteristic value for each abnormal sensor extracted for each faulty part, and the abnormal characteristic value range for each sensor is determined for each faulty part. It is a decision-making process

The normal characteristic value range for each sensor determined through step 3-3 (S330) is, for example, the normal characteristic value range of 1 to 5 of the acceleration sensor, 2-7 of the normal characteristic value of the flow sensor, and the normal characteristic of the pressure sensor. It may be a value range of 10 to 20, a normal characteristic value range of a current sensor of 0.5 to 1, and the like.

In addition, the range of abnormal characteristic values for each sensor determined for each faulty part through step 3-3 (S330) is, for example, the range of abnormal characteristic values of the acceleration sensor when a mechanical seal fails, 6-10, mechanical seal ( The abnormal characteristic value range of the flow sensor in case of mechanical seal failure 8~15, the abnormal characteristic value range of the pressure sensor in case of mechanical seal failure 20~30, the abnormal characteristic value range of the current sensor in case of mechanical seal failure 1~2, the abnormal characteristic value range of the acceleration sensor in case of bearing failure 8~13, the abnormal characteristic value range of the flow sensor in case of bearing failure 10~18, the abnormal characteristic value range of the pressure sensor in the case of bearing failure 25~33, in case of bearing failure The abnormal characteristic value range of the current sensor may be 1.5 to 2.7, and the like.

Steps 3-4 ( S340 ) are a process of generating learning result information for fault diagnosis by using the determined characteristic value range for each sensor.

The learning result information for fault diagnosis generated through steps 3-4 ( S340 ) includes information on the characteristic value range for each sensor of the normal pump and information on the characteristic value range for each sensor of the abnormal pump for each faulty part.

Therefore, it will be described later by comparing the characteristic value of each sensor of the pump to be diagnosed with the learning result information for fault diagnosis, which includes information on the characteristic value range for each sensor of the normal pump and the information on the characteristic value range for each sensor of the abnormal pump for each failure part. Through the failure diagnosis step (S2000), the normal or abnormality of the pump, which is a diagnosis target, is diagnosed.

Referring to FIG. 1 , the failure diagnosis step ( S2000 ) of the present invention diagnoses the failure of the target pump 20 for failure diagnosis by using the failure diagnosis learning result information generated through the learning result information generation step ( S1000 ). Step 4-1 (S2100), step 4-2 (S2200), and includes a step 4-3 (S2300).

Specifically, the fault diagnosis step (S2000) is as shown in FIG. 8,

Step 4-1 (S2100) of building a fault diagnosis pump system and installing sensors at each location of the built system;

Step 4-2 (S2200) of operating the fault diagnosis pump system to obtain data for fault diagnosis for each sensor;

and a 4-3 step (S2300) of diagnosing the failure of the pump, which is a failure diagnosis target, by using the acquired failure diagnosis data for each sensor and the failure diagnosis learning result information.

In particular, the sensors installed in step 4-1 (S2100) include an acceleration sensor, a flow sensor, a pressure sensor, and a current sensor, and the acceleration sensor is installed in the impeller case to measure the vibration of the pump, and the flow rate The sensor is installed in the upper pipe sufficiently spaced apart from the outlet to measure the operating flow, the pressure sensor is installed at the inlet and the outlet to measure the differential pressure, and the current sensor is a hotline of the electric wire to measure the power consumed by the pump. It is characterized in that it is installed in

The step 4-1 ( S2100 ) is a step of constructing a fault diagnosis pump system and installing sensors at each position of the built system.

That is, the failure diagnosis pump system built through the step 4-1 (S2100) is the same system as the normal pump demonstration system built in the first step (S100) of the learning result information generation step (S1000), only the difference is The point is that the normal pump 20 constituting the normal pump demonstration system built in the first step ( S100 ) is replaced with a diagnosis target pump for which a failure is to be diagnosed.

That is, the failure diagnosis pump system constructed through step 4-1 (S2100) is a failure diagnosis target pump, acceleration sensor 100, flow sensor 200, pressure sensor 300, current sensor 400, water tank ( 500 ), a data collection unit 600 , and a pipe 700 .

Step 4-2 (S2200) is a process of acquiring fault diagnosis data for each sensor by operating the fault diagnosis pump system built through step 4-1 (S2100).

The 4-3 step (S2300) is a failure diagnosis target using the fault diagnosis data for each sensor acquired through the 4-2 step (S2200) and the fault diagnosis learning result information generated in the learning result information generation step (S1000). As a step of diagnosing the failure of the pump, it includes a 4-3-1 step (S2310) and a 4-3-2 step (S2320).

Specifically, the 4-3 step (S2300) is as shown in FIG. 8,

Step 4-3-1 (S2310) of extracting a characteristic value for each sensor from the acquired fault diagnosis data for each sensor;

Comprising the 4-3-2 step (S2320) of comparing the extracted characteristic values for each sensor with the learning result information for fault diagnosis, and determining whether the pump is normal or abnormal for diagnosis of a fault;

The learning result information for fault diagnosis is characterized in that it includes information on the characteristic value range for each sensor of the normal pump and information on the characteristic value range for each sensor of the abnormal pump for each faulty part.

The step 4-3-1 ( S2310 ) is a step of extracting a characteristic value for each sensor from the acquired fault diagnosis data for each sensor. It is characterized in that it is the standard deviation for each set time section of the data for fault diagnosis.

The reason for using the standard deviation as the characteristic value is the same as described in the learning result information generation step (S1000).

The step 4-3-2 ( S2320 ) is a process of comparing the extracted characteristic values for each sensor with the learning result information for fault diagnosis to determine whether the pump to be diagnosed is normal or abnormal.

Since the failure diagnosis learning result information includes information on the characteristic value range for each sensor of a normal pump and information on the characteristic value range for each sensor of the abnormal pump for each failure part, the extracted characteristic value for each sensor is used for failure diagnosis. Comparing with the result information means comparing the extracted characteristic value for each sensor with the characteristic value range for each sensor of a normal pump, and comparing the extracted characteristic value for each sensor with the characteristic value range for each sensor of the abnormal pump for each faulty part. .

For example, the learning result information for fault diagnosis indicates that the normal characteristic value range of the acceleration sensor is in the range of 1 to 5, the abnormal characteristic value range of the acceleration sensor in case of mechanical seal failure is 6 to 10, and the acceleration sensor in the case of bearing failure. If it is assumed that the abnormal characteristic value range of is 8 to 13, the characteristic value of the acceleration sensor extracted from the failure diagnosis data obtained from the failure diagnosis pump system is in the normal characteristic value range of the acceleration sensor 1 to 5, mechanical seal This is compared with the abnormal characteristic value range of 6 to 10 of the acceleration sensor in case of failure, and the abnormal characteristic value range of 8 to 13 of the acceleration sensor in the case of bearing failure.

As a result of comparison, if the characteristic value of the acceleration sensor extracted from the fault diagnosis data falls within the normal characteristic value range of the acceleration sensor, 1-5, it is normal, and when the mechanical seal fails, the acceleration sensor's abnormal characteristic value range is 6-10. If it belongs to mechanical seal failure, and if it belongs to 8~13, which is the abnormal characteristic value range of the acceleration sensor in case of bearing failure, it is judged as bearing failure.

In the above case, it is determined that the mechanical seal and bearing of the pump for fault diagnosis are faulty.

Through this, the present invention allows the user to accurately identify a specific faulty part of the pump to be diagnosed and promptly take action, so as in the prior art, it is not possible to check the pump unnecessarily or indiscriminately to determine whether there is an abnormality or to determine a specific faulty part. It is possible to solve problems such as replacing the entire pump.

Although the technical idea of the present invention has been described together with the accompanying drawings, the preferred embodiment of the present invention is exemplarily described, and the present invention is not limited thereto. In addition, it is a clear fact that anyone with ordinary skill in the art can make various modifications and imitations without departing from the scope of the technical idea of the present invention.

10: Normal Pump Demo System
11: Abnormal Pump Demo System
20: normal pump
21: Abnormal pump
S1000: Learning result information generation step
S2000: Troubleshooting stage

Claims (7)

In a machine learning-based pump failure diagnosis monitoring method for diagnosing a pump abnormality,
a learning result information generation step (S1000) of generating learning result information for fault diagnosis that can determine whether the pump is normal or abnormal using the pump demonstration system;
A machine learning-based pump failure diagnosis monitoring method, comprising: a failure diagnosis step (S2000) of diagnosing a failure of a target pump for failure diagnosis using the generated failure diagnosis learning result information.
The method according to claim 1,
The learning result information generation step (S1000) is,
After building a normal pump demonstration system, a first step (S100) of acquiring normal raw data for each sensor of a normal pump;
After building the abnormal pump demonstration system, the second step (S200) of acquiring abnormal raw data for each sensor of the abnormal pump for each failure part;
A third step (S300) of generating learning result information for diagnosing pump failure that can determine whether the pump is normal or abnormal using the acquired normal raw data for each sensor and abnormal raw data for each sensor acquired for each faulty part A machine learning-based pump failure diagnosis monitoring method.
3. The method according to claim 2,
The first step (S100) is,
Step 1-1 (S110) of installing sensors at each position of the normal pump demonstration system;
After the sensor is installed, including the first and second steps (S120) of operating the pump demonstration system to obtain normal raw data for each sensor,
The sensors installed in step 1-1 (S110) include an acceleration sensor, a flow sensor, a pressure sensor, and a current sensor,
The acquisition of normal raw data for each sensor in step 1-2 (S120) is repeatedly acquired for a set number of times,
The acceleration sensor is installed in the impeller case to measure the vibration of the pump, the flow sensor is installed in the upper pipe sufficiently spaced apart from the outlet to measure the operating flow, and the pressure sensor is installed at the inlet and outlet to measure the differential pressure. installed, and the current sensor is a machine learning-based pump failure diagnosis monitoring method, characterized in that it is installed on a hotline of a wire to measure the power consumed by the pump.
3. The method according to claim 2,
The second step (S200) is,
Step 2-1 (S210) of artificially malfunctioning a specific part of the pump constituting the normal pump demonstration system to build an abnormal pump demonstration system;
Including the second-2 step (S220) of acquiring the abnormal raw data for each sensor by operating the built-up abnormal pump demonstration system,
Step 2-1 (S210) and step 2-2 (S220) are repeatedly performed by changing a specific part of the pump to be broken, and step 2-2 (S220) is repeated as many times as set for each failure part. Machine learning-based pump failure diagnosis monitoring method, characterized in that.
3. The method according to claim 2,
The third step (S300) is,
Step 3-1 (S310) of extracting characteristic values for each normal sensor from the acquired normal raw data for each sensor;
Step 3-2 (S320) of extracting characteristic values for each abnormal sensor for each faulty part from the abnormal raw data for each sensor acquired for each faulty part (S320);
Step 3-3 of determining the normal characteristic value range for each sensor using the extracted characteristic value for each normal sensor and the characteristic value for each abnormal sensor extracted for each faulty part, and determining the abnormal characteristic value range for each sensor for each faulty part ( S330) and
Steps 3-4 (S340) of generating learning result information for fault diagnosis using the determined characteristic value ranges for each sensor,
The characteristic value is a machine learning-based pump failure diagnosis monitoring method, characterized in that the standard deviation for each set time section.
The method according to claim 1,
The fault diagnosis step (S2000) is,
Step 4-1 (S2100) of building a fault diagnosis pump system and installing sensors at each location of the built system;
Step 4-2 (S2200) of operating the fault diagnosis pump system to obtain data for fault diagnosis for each sensor;
Including the 4-3 step (S2300) of diagnosing the failure of the pump, which is a failure diagnosis target, using the acquired failure diagnosis data for each sensor and the failure diagnosis learning result information,
The sensors installed in step 4-1 (S2100) include an acceleration sensor, a flow sensor, a pressure sensor, and a current sensor,
The acceleration sensor is installed in the impeller case to measure the vibration of the pump, the flow sensor is installed in the upper pipe sufficiently spaced apart from the outlet to measure the operating flow, and the pressure sensor is installed at the inlet and outlet to measure the differential pressure. installed, and the current sensor is a machine learning-based pump failure diagnosis monitoring method, characterized in that it is installed on a hotline of a wire to measure the power consumed by the pump.
7. The method of claim 6,
The 4-3 step (S2300) is,
Step 4-3-1 (S2310) of extracting characteristic values for each sensor from the acquired data for fault diagnosis by sensor;
Comprising the 4-3-2 step (S2320) of comparing the extracted characteristic values for each sensor with the learning result information for fault diagnosis, and determining whether the pump is normal or abnormal for diagnosis of a fault;
The characteristic value for each sensor extracted from the data for fault diagnosis for each sensor is the standard deviation for each set time section,
The failure diagnosis learning result information is a machine learning-based pump failure diagnosis monitoring method, characterized in that it includes information on the characteristic value range for each sensor of the normal pump and information on the characteristic value range for each sensor of the abnormal pump for each failure part. .

Images