KR102193381B1 - System and method for monitoring health of fan using machine learning - Google Patents

Abstract

The method for evaluating the driving soundness of a fan of the present invention includes the steps of collecting data on the driving state of a fan in a time series for at least 3 months using acceleration data, rotation speed data, and acoustic data, and converting the time series data into frequency data. Step, making the frequency data into an appropriate statistic, extracting only some data as feature data, learning the feature data using one or more machine learning algorithms among RF models, k-NN models, and SVM models, and And evaluating the driving state as normal/abnormal based on the learning data. According to the configuration of the present invention, even if only a minimum number of sensors are used, the driving state can be quickly determined by using machine learning.

Description

System and method for monitoring health of fan using machine learning

The present invention relates to a system and method for intelligently evaluating the driving soundness of a fan, and specifically, to a small fan that is difficult for an administrator to access and can check the driving status based only on data transmitted from a sensor. For this, machine learning is used to quickly and accurately evaluate the operation status without being restricted by the administrator's experience, but the data is collected with only a minimum of sensors, and furthermore, only some of the necessary data are extracted as characteristic data. It relates to a system and method for evaluating the driving soundness of a fan used for running.

In general, fans play an important role in various fields and must be operated at all times, so it is very important to evaluate the operational soundness and take appropriate preventive measures.

Two systems can be applied to the method of evaluating the operating state of the fan. One is an Expert System, and the other is a Case Based Machine Learning Inference System.

The expert system is a system that judges the operating state of a fan using a rule set in advance by a fan expert.

According to such an expert system, since a manager directly observes and judges data transmitted from multiple sensors or data transmitted through an imaging device, consistency and persistence are insufficient depending on the manager's experience and skill level. In particular, there is a concern about an increase in cost and a decrease in efficiency due to the installation of multiple sensors and system operation.

For example, according to the expert system, when it is assumed that the driving state of the fan is determined as two (for example, normal/failure) through expert learning, acceleration data in the x direction, acceleration data in the y direction, rotation speed data, and acoustic data Assuming that a total of four data are used, the expert system has a contradiction that results are different in every case because there are numerous cases. And since infinite conditions must be applied to distinguish the two driving states, it is practically impossible to implement.

In addition, since the applicability to new data is poor, the expert system has a clear limitation in determining the driving status.

KR registered patent 10-1539067

Therefore, the present invention was conceived to solve the problems of the prior art as described above, and the object of the present invention is after collecting minimum physical information (eg, acceleration, rotational speed, sound, etc.) about a fan. It is to provide a system and method for evaluating the operating condition health of a fan using machine running techniques so that the operation condition can be inferred or judged by a given acceleration using machine learning.

Another object of the present invention is to use a minimum number of sensors to collect data, and to intelligently evaluate the collected data, but to determine the abnormal state of the fan operation state by dividing it with mechanical defects and other reasons. To provide health assessment systems and methods.

According to a feature of the present invention for achieving the above object, the system for evaluating the driving soundness of a fan of the present invention includes an acceleration sensor that obtains acceleration data of a fan, an infrared sensor that obtains rotation speed data of the fan, the An acoustic sensor for acquiring sound data of a fan, a time series data collection unit collecting the acceleration data, the rotation speed data, and the sound data for a predetermined period of time or longer, a frequency data conversion unit converting the time series data into frequency data, the frequency And a characteristic data extracting unit for extracting only some statistical values of the data as characteristic data, a data machine learning unit for learning the characteristic data, and a driving state determination unit for determining an operating state of the fan based on the learning data.

According to another feature of the present invention, the method for evaluating the driving soundness of a fan of the present invention includes the steps of collecting data on the driving state of the fan in time series for at least 3 months using acceleration data, rotation speed data, and sound data, the Converting time series data to frequency data using Fourier transform, making the frequency data an appropriate statistic, and extracting only some data as feature data, and extracting the feature data from an RF model, a k-NN model, or an SVM model. Learning using one or more machine learning algorithms, and evaluating the driving state as normal/abnormal based on the learning data.

As described above, according to the configuration of the present invention, the following effects can be expected.

First, since the operation status of the fan can be determined using a machine learning system, contradictory situations in which the diagnosis result is changed according to the skill level or subjective judgment of the site manager are fundamentally eliminated, and the help of experienced experts. Even without it, it is possible to know in real time whether the fan is operating normally or abnormally, so the effect of increasing the convenience of consumers is expected.

Second, since the operation state of the fan can be known using only the acceleration sensor, the infrared sensor, and the acoustic sensor, it is possible to omit the use of a large number of expensive sensors or imaging devices, and an economical effect is expected.

Third, since it shows normal/abnormal diagnosis results, it is possible to quickly respond according to diagnosis, and even in the case of abnormality, it is possible to determine an abnormal state due to environmental factors other than mechanical failure of the fan. In spite of being in a normal state, it is expected to increase productivity by preventing unreasonable confusion that is judged as abnormal and processed.

In particular, the manager does not need to be an expert with a high degree of knowledge, and a person with knowledge enough to understand and cope with the diagnosis result is sufficient.

1 is a block diagram showing the configuration of a system for evaluating the operational soundness of a fan according to the present invention.
2 is a table of statistical values used as characteristic data of the acceleration frequency in the x direction of the fan according to the present invention.
3 is a table of statistical values used as characteristic data of the y-direction acceleration frequency of the fan according to the present invention.
4 is a table of statistical values used as characteristic data of the acoustic frequency of the fan according to the present invention.
5 is an experimental example showing the performance and test data reliability of the k-NN model when evaluating the operating state of the fan according to the present invention as two normal/abnormal.
6 is an experimental example showing the performance and test data reliability of the SVM model when evaluating the operating state of the fan according to the present invention as normal/abnormal two.
7 is an experimental example showing the performance of the RF model and the reliability of test data when evaluating the operating state of the fan according to the present invention as normal/abnormal two.
8 is an experimental example showing the performance of the k-NN model and reliability of test data when evaluating the operating state of the fan according to the present invention as three normal/mechanical defects/other factors.
9 is an experimental example showing the performance and test data reliability of the SVM model when evaluating the operating state of the fan according to the present invention as three normal / mechanical defect / other factors.
10 is an experimental example showing the performance of the RF model and the reliability of test data when evaluating the operating state of the fan according to the present invention as three normal/mechanical defects/other factors.
11 is a flow chart showing a method for evaluating the operational integrity of a fan according to the present invention.
12 is a graph of acceleration data in the x direction of the fan for each normal/mechanical defect/other factors measured by the acceleration sensor according to the present invention.
13 is a graph of acceleration data in the y direction of a fan for each normal/mechanical defect/other factors measured by the acceleration sensor according to the present invention.
14 is a graph of sound data of a fan for each normal/mechanical defect/other factors measured by the acoustic sensor according to the present invention.
15 is a graph of the frequency conversion of FIG. 12;
16 is a graph of frequency conversion of FIG. 13;
17 is a graph of frequency conversion of FIG. 14;

Advantages and features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but will be implemented in various forms different from each other, and only these embodiments make the disclosure of the present invention complete, and common knowledge in the technical field to which the present invention pertains. It is provided to completely inform the scope of the invention to the possessor, and the invention is only defined by the scope of the claims. In the drawings, the sizes and relative sizes of layers and regions may be exaggerated for clarity of description. The same reference numerals refer to the same components throughout the specification.

Embodiments described in the present specification will be described with reference to a plan view and a cross-sectional view, which are ideal schematic diagrams of the present invention. Accordingly, the shape of the exemplary diagram may be modified by manufacturing technology and/or tolerance. Accordingly, embodiments of the present invention are not limited to the specific form shown, but also include a change in form generated according to the manufacturing process. Accordingly, the regions illustrated in the drawings have schematic properties, and the shapes of the regions illustrated in the drawings are intended to illustrate a specific shape of the region of the device, and are not intended to limit the scope of the invention.

The system for evaluating the operational soundness of a fan according to the present invention allows an intelligent system to self-evaluate the operational soundness of a fan used for cooling and ventilation purposes in electronic devices or various mechanical systems to take quick action in an emergency. Define the machine learning method to enable.

The object for evaluating the driving soundness of the present invention relates to a fan. A fan is used to convey gas by rotational motion of a rotating blade for ventilation or cooling, and includes all of a centrifugal fan, a four-flow fan, an axial fan, and the like.

In particular, the present invention has a greater practical benefit in evaluating the operational soundness of a mini fan. For example, since small fans are mounted inside the device or in a duct that is difficult to access, it is difficult for the manager to check it directly unless the device is opened, and the number of installations is infinite depending on the purpose, so it is operated by the efforts of the manager one by one. It is difficult to check the condition.

Here, there are various types of small fans, such as a cooling fan for a computer, a ventilation fan for a microwave oven, and a blower fan for a vehicle, and may be independently manufactured and attached, or may be installed integrally in the device.

The present invention excludes expensive sensors or imaging devices for driving soundness evaluation and uses only a minimum number of sensors. If multiple sensors or imaging devices have to be installed for each fan, the fan management cost will double.

Hereinafter, a preferred embodiment of a system for evaluating the operational soundness of a fan according to the present invention having the configuration as described above will be described in detail with reference to the accompanying drawings.

Referring to FIG. 1, the fan driving soundness evaluation system 100 according to the present invention acquires the acceleration sensor 110 that acquires the fan acceleration data and transmits it to a time series data collection unit to be described later, and the fan rotation speed data. Infrared sensor 120 transmitting to the time series data collection unit, the acoustic sensor 130 acquiring the sound data of the fan and transmitting it to the time series data collecting unit, acceleration data, rotational speed data, and acoustic data are collected in time series for at least a predetermined period A time series data collection unit 140 that converts time series data into frequency data, a frequency data conversion unit 150 that converts time series data into frequency data, a feature data extraction unit 160 that extracts only some statistical values of frequency data as feature data, and learns feature data It includes a machine learning unit 170 to be used, and a driving state determination unit 180 that determines a normal or abnormal driving state based on the learning data.

Among them, the time series data collection unit 140, the frequency data conversion unit 150, the characteristic data extraction unit 160, the data machine learning unit 170, and the driving state determination unit 180 are provided in one processor and collectively It can be processed separately or separately installed in multiple devices to perform each function. In particular, the processor may support the above-described functions through a separate application (APP).

In addition, a communication unit 192 and an output unit 194 may be further included. The communication unit 192 performs communication between the sensors 110, 120, and 130 and the processor or between the processor and an external device, and includes both wireless communication and wired communication. For example, the driving state can be transmitted to an external device through wired or wireless communication. The output unit 194 includes a display or speaker that informs the manager or the management system of the driving soundness evaluation result by a method such as audio-visual or the like.

In this way, the processor controls the communication unit 192 and the output unit 194, etc., calculates and processes various data obtained from the sensor through the communication unit 192, and evaluates it through the communication unit 192 or the output unit 194. Results can be provided to an administrator or external device.

As the acceleration sensor 110, a 3-axis acceleration sensor may be used. For example, acceleration can be detected in the x and y directions in the horizontal plane and in the y directions in the vertical plane. As will be described later, the data used may include only the x-direction acceleration and the y-direction acceleration in the horizontal plane. The acceleration sensor 110 may be used regardless of a piezoelectric type, a capacitive type, or a strain gauge type.

The infrared sensor 120 is not necessarily limited to infrared light, and may include an optical sensor including a light emitting element and a light receiving element. For example, infrared rays are radiated to a fan using a light emitting element, infrared rays reflected through the rotating blades and infrared rays blocked by the rotating blades are sensed by a light receiving element, and the number of rotations of the fan is calculated per a certain time. Can be detected.

In this case, when the fan is installed in the peripheral device, each sensor may be directly installed on the fan or may be installed in the peripheral device. Alternatively, when a cover is installed on the fan, the sensor may be installed on the cover. In particular, the acceleration sensor can be installed directly on the fan. On the other hand, the infrared sensor or the acoustic sensor need not be installed directly on the fan, but may be installed around it.

As the acoustic sensor 130, a microphone may be used. The microphone converts mechanical vibrations into electrical signals. However, depending on the conversion method, a carbon type, a crystal type, a coil type, a condenser type, etc. may be included.

The time series data collection unit 140 collects acceleration data from the acceleration sensor 110 in time series for 3 months (preferably 6 months) or more, collects rotation speed data from the infrared sensor 120, and the acoustic sensor ( 130) to collect sound data.

The frequency data conversion unit 150 may convert time series data into frequency data using filtering and Fourier transform.

The feature data extracting unit 160 extracts only feature data to be used in the data machine learning unit 170. For example, a representative component required to determine the operating state of the fan will be the rotational speed of the fan. If the fan rotation speed is higher (or lower) than the specified speed or the rotation speed is not constant and increases or decreases irregularly, it indicates an abnormal operation condition. For example, when the rotating blade is eccentric with respect to the rotating shaft or the rotating blade is damaged and vibrates, in particular, when a vortex is formed due to a loss of air pressure around the rotating blade, an abnormal operation is performed due to a pressure difference.

Therefore, the acceleration frequency in the x direction (see Fig. 2), the acceleration frequency in the y direction (see Fig. 3), the acoustic frequency (see Fig. 4), and the number of revolutions are extracted as characteristic data and combined to drive the abnormal driving state. It is necessary to judge the condition.

The data machine learning unit 170 may learn the extracted feature data using a machine learning algorithm. For example, the machine learning algorithm may include one or more of a nearest neighbor (k-NN) model, a support vector machine (SVM) model, and a random forest (RF) model.

As a result of examining the reliability of the result obtained by inputting the characteristic data into the machine learning algorithm, it can be confirmed that there are differences for each model.

When judging the operation status of a small fan by dividing it into two criteria (normal/abnormal), the result of the judgment shows that there is a difference in the reliability of the test data and the performance of each model that can be expressed as normal/abnormal. 1].

Referring to FIG. 5, it can be seen that the performance of the nearest neighbor (k-NN) model is 88.137%, and the data reliability is 88.9%. Further, referring to FIG. 6, it can be seen that the performance of the support vector machine (SVM) model is 65.676%, and the data reliability is 66.7%. On the other hand, referring to FIG. 7, the performance of the random forest (RF) model is 97.805%, and its data reliability is 94.4%, which is relatively accurate.

The same is true when judging the operating state of a small fan by dividing it into three criteria (normal, mechanical defect, and other factors). As shown in [Table 2], there is a greater difference than the two criteria in the performance of each model and the reliability of test data, which can be expressed as normal, mechanical defects, and other factors.

Referring to FIG. 8, it can be seen that the performance of the nearest neighbor (k-NN) model is 84.294%, and the data reliability is 83.9%. Further, referring to FIG. 9, it can be seen that the performance of the support vector machine (SVM) model is 50.134%, and the data reliability is 50.0%. On the other hand, referring to FIG. 10, the performance of the random forest (RF) model is 97.391%, and the data reliability is 91.1%, which is relatively accurate.

However, in terms of overall performance or reliability, it can be seen that it is more complicated to predict the determination result with three criteria than with two criteria, and performance is degraded.

As described above, the driving state determination unit 180 may determine the driving state of the small fan as normal/abnormal.

In addition, the operation state determination unit 180 includes evaluating the operation state of the small fan as a normal/mechanical defect/other factor. Here, other factors are not mechanical defects, but an abnormal condition is defined due to environmental factors. For example, this may include a case in which pressure is formed around the small fan to hinder the rotation of the fan (eg, when wind blows around the fan).

For example, if unexpected external wind blows behind the fan's rotating blade, the rotational speed and rotational speed of the instantaneous rotating blade will increase, and conversely, if the wind blows in the front, the rotational speed and rotational speed will decrease. Even though there is no mechanical failure of the fan, it can be determined as an abnormal operation state.

Therefore, the determination may be somewhat complicated, but as in the embodiment of the present invention, since it can be analyzed that the operating state may be temporarily changed due to environmental factors other than mechanical failure, this is mistaken as a failure and a normal fan You can avoid the error of replacing. For example, when wind blows from the front and rear, a change occurs in the acceleration in the y-axis direction, but there is no change in the acceleration in the x direction, so other acoustic data can be synthesized to determine that it is a temporary malfunction rather than a mechanical failure.

Hereinafter, a method for evaluating the operational soundness of a fan according to the present invention will be described with reference to the drawings.

Referring to FIG. 11, the method for evaluating the driving soundness of a fan according to the present invention is to collect data on the driving state of a fan in time series for at least 3 months using acceleration data, rotation speed data, and sound data. Step (S110), converting the time series data into frequency data (S120), extracting the frequency data as characteristic data to be used in the intelligent system of the present invention (S130) by making the frequency data into an appropriate statistic, the characteristic data into an RF model , k-NN model, and SVM model learning using one or more machine learning algorithms (S140), and evaluating the driving state as normal/abnormal based on the learning data (S150). Alternatively, the operating state can be evaluated as normal/mechanical defect/other factors.

The time series data may include acceleration data (eg, acceleration data in the x direction of FIG. 12 and acceleration data in the y direction of FIG. 13 ), rotation speed data, and sound data (see FIG. 14 ).

The x-direction acceleration data represents a characteristic factor acting in the left-right direction based on the rotation axis of the fan. For example, when there is a change in the left-right rotation capability of the fan due to an abnormality in the motor rotation shaft or a defect in the rotation blade, there is a change in the measured x-direction acceleration data.

The y-direction acceleration data represents a characteristic factor acting in the front-rear direction with respect to the rotation axis of the fan. For example, if there is a change in the forward/backward rotational state, such as vibration in the front-rear direction due to an abnormality in the motor rotation shaft or a defect in the rotating blade, there is a change in the measured y-direction acceleration data.

Vibration and other obstacles in the vertical direction can be grasped through the z-direction acceleration data, but since the change in the vertical direction accompanies a change in the left and right directions, in this embodiment, a characteristic factor is determined based on the x-direction acceleration data. .

The above-described time series data are judged as normal/abnormal, or normal/mechanical defects/other factors by expert learning. Since the judgment itself is very subjective, the judgment result may vary depending on the level of experience of the expert.

Accordingly, according to an exemplary embodiment of the present invention, such time series data is converted into frequency data and used for machine learning, and can be objectively determined by the fan driving soundness evaluation system 100.

For example, the above-described time series data is converted to the x-direction acceleration data of FIG. 12 through frequency conversion, and the y-direction acceleration data of FIG. 13 to the y-direction acceleration frequency data of FIG. , The acoustic data of FIG. 14 may be converted into the acoustic frequency data of FIG. 17 to be machine learned.

Among the characteristic data, the x-direction acceleration frequency data measured and converted by the acceleration sensor 110 is x-mean (average value of data), x-std (standard deviation of data), and x- min (minimum value of data), x-max (maximum value of data), x-median (median value of data), x-1q (1 quartile of data), x-3q (3 quartile of data), x- It may include fmax (frequency when amplitude is at maximum).

Of the characteristic data, the y-direction acceleration frequency data measured and converted by the acceleration sensor 110 is y-mean (average value of data), y-std (standard deviation of data), y- min (minimum value of data), y-max (maximum value of data), y-median (median value of data), y-1q (1 quartile of data), y-3q (third quartile of data), y- may contain fmax (frequency when the amplitude is at maximum).

Among the characteristic data, the acoustic frequency data measured and converted by the microphone 130 are s-mean (average value of data), s-std (standard deviation of data), s-min (data), as shown in FIG. Minimum value of), s-max (maximum value of data), s-median (median value of data), s-1q (1 quartile of data), s-3q (three quartile of data), s-fmax (amplitude When this is the maximum, the frequency) can be included.

In an embodiment of the present invention, the statistical value of each characteristic data is obtained by using the minimum, maximum, median, 1 and 3 quartiles, and frequencies of the maximum amplitude as characteristic data in addition to the average value and standard deviation of the data. High, in particular, it is easy to distinguish results due to mechanical defects and environmental factors, and data accuracy can be improved.

Finally, the characteristic data includes the data of the rotation speed measured by the infrared sensor 120 and converted.

In this way, even though electronic devices and various mechanical systems cannot operate without a small fan, it is difficult for the manager to directly access depending on the size or mounting location, making it difficult to manage and supervise the operation status. In the present invention, since machine learning can be implemented using only a minimum number of sensors to intelligently monitor the operating soundness of a small fan, it is expected to increase fan management productivity and significantly reduce cost.

As described above, the present invention is to quickly determine the operating state using only some sensors and sound data to prevent in advance the presence or absence of failure of a small fan that performs an air purification function such as cooling or ventilation in various devices or facilities. It can be seen that the configuration using the random forest (RF) model is a technical idea. Within the scope of the basic technical idea of the present invention, many other modifications may be made to those of ordinary skill in the art.

100: fan operation soundness evaluation system
110: acceleration sensor
120: infrared sensor
130: acoustic sensor
140: time series data collection unit
150: frequency data conversion unit
160: characteristic data extraction unit
170: Materials Machine Learning Department
180: driving state determination unit

Claims (15)

An acceleration sensor that acquires acceleration data of the fan;
An infrared sensor that acquires the rotation speed data of the fan;
An acoustic sensor for acquiring acoustic data of the fan;
A time series data collection unit collecting the acceleration data, the rotation speed data, and the sound data for a predetermined period or more;
A frequency data conversion unit converting the time series data into frequency data;
A characteristic data extracting unit for extracting, as characteristic data, frequencies of some statistical values of the frequency data, such as average value, standard deviation, minimum value, maximum value, median value, 1 quartile, 3 quartile, and maximum amplitude;
A data machine learning unit for learning the characteristic data; And
And a driving state determination unit configured to determine the driving state of the fan by dividing it into normal or mechanical defects or environmental factors based on the learning data. The method of claim 1,
The time series data collection unit, the frequency data conversion unit, the characteristic data extraction unit, the data machine learning unit, and the driving state determination unit are operated within one processor,
The processor calculates and processes the acceleration data, the rotation speed data, and the sound data, and controls an output unit to display the driving state, or controls a communication unit to communicate with an external device, and transfer the driving state to the external device. Fan operation soundness evaluation system, characterized in that for transmitting. The method of claim 1,
The characteristic data extracting unit extracts an x-direction acceleration frequency, a y-direction acceleration frequency, an acoustic frequency, and a rotation speed as characteristic data. The method of claim 1,
The system for evaluating the driving soundness of a fan, wherein the data machine learning unit performs machine learning using a random forest (RF) model. delete Collecting data on the driving state of the fan in time series for at least 3 months using acceleration data, rotation speed data, and sound data;
Converting the time series data into frequency data;
Extracting the average, standard deviation, minimum, maximum, median, 1 quartile, 3 quartile, and maximum amplitude frequencies of the frequency data as characteristic data;
Training the characteristic data using one or more machine learning algorithms of an RF model, a k-NN model, or an SVM model; And
And determining the driving state by dividing the driving state into normal or mechanical defects or environmental factors based on the learning data. The method of claim 6,
The time series data includes acceleration data, rotation speed data, and sound data,
The frequency data is an x-direction acceleration frequency data, y-direction acceleration frequency data, acoustic frequency data, and rotational speed data. delete delete delete delete delete delete delete delete

Images