KR102321607B1 - Rotating machine fault detecting apparatus and method - Google Patents

Abstract

The present invention relates to a method for detecting a defect of a machine, which increases defect detection accuracy while detecting a defect of a machine by using a vibration signal of the machine. According to one embodiment of the present invention, the method comprises the following steps: receiving a vibration signal generated in a machine by a vibration sensor provided in the machine; converting a time domain vibration signal into a frequency domain vibration signal; calculating statistical characteristic values for the frequency domain vibration signal; using a plurality of deep neural network models, which are pre-trained to determine whether a machine is defective by using the statistical characteristic values of the vibration signal extracted from the vibration sensor and are implemented with different algorithms, to output a classification value for whether the machine is defective; and applying a voting algorithm to a plurality of classification values for whether the machine is defective to finally determine whether the machine is defective. Each of the plurality of deep neural network models is a neural network model pre-trained by using training data including statistical characteristic values of vibration signals generated in the machine and the state of the machine.

Description

Apparatus and method for detecting a fault in a machine

The present invention relates to a machine defect detection method and apparatus capable of improving defect detection accuracy while detecting a machine defect using a machine vibration signal.

With the recent development of industrial technology, as rotary machines become larger and more precise, failures in these facilities can cause enormous economic loss or damage to human life, so fault monitoring of these rotary machines has become very important.

Since the machine including these rotating driving parts is driven under physical friction conditions, it is necessary to perform maintenance at frequent intervals to prevent breakdowns and accidents. As such, the rotating machine requires frequent maintenance, so there is a problem that maintenance cost and time are required.

Due to these problems, condition-based maintenance, which performs maintenance only when there is a possibility of a failure, has recently attracted attention. Because condition-based maintenance performs maintenance only when there is a possibility of a failure, it reduces operating costs and increases profits by reducing maintenance costs and time.

Accordingly, Korean Patent Application Laid-Open No. 2020-0101507 also disclosed a system for diagnosing and predicting a malfunction of a rotating machine using machine learning.

The fields where fault diagnosis can be applied to have great effects can be systems that require high maintenance costs, and gearboxes such as high-speed railways and wind turbines are representative.

Since the gearbox is a basic element used to transmit rotational power in a machine, research on detecting and diagnosing defects through various sensor signals is being actively conducted.

The above-mentioned background art is technical information possessed by the inventor for derivation of the present invention or acquired in the process of derivation of the present invention, and cannot necessarily be said to be a known technique disclosed to the general public prior to the filing of the present invention.

Domestic Patent Publication No. 10-2020-0101507 (2020.08.28)

An object of the present invention is to analyze the vibration characteristics generated in a machine to detect a defect and to prevent a machine failure based on this.

An object of the present invention is to improve defect detection accuracy through machine defect detection using statistical analysis and machine learning.

The problem to be solved by the present invention is not limited to the above-mentioned problems, and other problems and advantages of the present invention not mentioned can be understood by the following description, and more clearly understood by the embodiments of the present invention will be In addition, it will be appreciated that the problems and advantages to be solved by the present invention can be realized by means and combinations thereof indicated in the claims.

A method for detecting a machine defect according to an embodiment of the present invention includes the steps of receiving a vibration signal generated in a machine by a vibration sensor provided in the machine, and converting a time domain vibration signal into a frequency domain vibration signal and calculating statistical characteristic values for the vibration signal in the frequency domain, and using the statistical characteristic value of the vibration signal derived from the vibration sensor to determine whether the machine is defective, a plurality of implementations with different algorithms Outputting a classification value for whether or not a machine is defective by using a deep neural network model, and applying a voting algorithm to a classification value for a plurality of machine defects to finally determine whether a machine is defective, Each of the deep neural network models may be a neural network model trained in advance using training data including statistical characteristic values of vibration signals generated in the machine and the state of the machine.

The apparatus for detecting a machine defect according to an embodiment of the present invention includes a vibration sensor provided in the machine to detect a vibration signal generated in the machine, and a time domain vibration signal output from the vibration sensor is converted into a frequency domain vibration signal A preprocessing unit that outputs the result, a calculation unit that calculates statistical characteristic values for the vibration signal of the frequency domain output from the preprocessor, and a statistical characteristic value of the vibration signal derived from the vibration sensor to determine whether the machine is defective A classification unit that outputs a classification value for whether a machine is defective using a plurality of deep neural network models implemented with different algorithms, trained in advance, and a voting algorithm applied to a classification value for whether a plurality of machines are defective A decision unit that finally determines whether or not there is a defect, and each of the plurality of deep neural network models may be a neural network model trained in advance using training data including statistical characteristic values of vibration signals generated in the machine and the state of the machine.

In addition to this, another method for implementing the present invention, another system, and a computer-readable recording medium storing a computer program for executing the method may be further provided.

Other aspects, features and advantages other than those described above will become apparent from the following drawings, claims, and detailed description of the invention.

Advantageous Effects of Invention According to the present invention, it is possible to diagnose a machine defect due to a change in the state of the machine, so that a failure can be prevented in advance.

In addition, by using statistical analysis and machine learning, the defect detection accuracy of the rotating machine can be improved, and the defect detection cost can be reduced.

Effects of the present invention are not limited to those mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

1 is a block diagram schematically illustrating the configuration of a defect detection apparatus for a machine according to an embodiment of the present invention.
FIG. 2 is an exemplary view illustrating an output direction of a vibration signal output by the vibration sensor in FIG. 1 .
3 is an exemplary view illustrating a vibration signal output from a machine in a normal state and a failure state according to the present embodiment.
4 is a view showing a result of pre-processing the vibration signal output from the machine in a normal state and a failure state according to the present embodiment.
5 is a view showing the results of calculating statistical characteristic values from the preprocessing results for the vibration signal output from the machine in the normal state and the broken state according to the present embodiment.
6 is a diagram illustrating distributions of maximum values and kurtosis values among statistical characteristic values from preprocessing results for vibration signals output from the machine in a steady state and a failure state according to the present embodiment.
7 is a block diagram schematically illustrating the configuration of the classification unit in FIG. 1 .
FIG. 8 is an exemplary diagram showing the accuracy of the deep neural network model obtained in the pre-training stage used when the decision unit in FIG. 1 finally determines whether or not the machine is defective.
9 is a flowchart illustrating a method for detecting a machine defect according to the present embodiment.

Advantages and features of the present invention, and methods for achieving them will become apparent with reference to the detailed description in conjunction with the accompanying drawings. However, it should be understood that the present invention is not limited to the embodiments presented below, but may be implemented in a variety of different forms, and includes all transformations, equivalents, and substitutes included in the spirit and scope of the present invention. . The embodiments presented below are provided to complete the disclosure of the present invention, and to completely inform those of ordinary skill in the art to which the present invention pertains to the scope of the invention. In describing the present invention, if it is determined that a detailed description of a related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In the present application, terms such as “comprise” or “have” are intended to designate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, but one or more other features It should be understood that this does not preclude the existence or addition of numbers, steps, operations, components, parts, or combinations thereof. Terms such as first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings, and in the description with reference to the accompanying drawings, the same or corresponding components are given the same reference numerals, and overlapping descriptions thereof are omitted. decide to do

1 is a block diagram schematically illustrating a configuration of a defect detection apparatus for a rotating machine according to an embodiment of the present invention. Referring to FIG. 1 , the machine defect detecting apparatus may include a rotating machine 100 and a defect detecting apparatus 200 .

The rotating machine 100 may include a gearbox used to transmit rotational power to a mechanical system. The gearbox may include a structure in which various shaft groups and gear groups (not shown) are disposed in order to transmit power required to rotate the drive shaft. Such a gearbox can be used in a wide variety of industries. And by using the gears of the gear group, a small rotational force can be converted into a large rotational force, or a low rotational force can be converted into a high-speed rotation. In the gearbox, a housing (not shown) of a certain shape forms the exterior. The housing may be formed to have a predetermined size and shape according to various mechanical devices or vehicles in which the gearbox is used. In addition, various gears (not shown), a driving shaft (not shown), a rotating shaft (not shown), etc. may be disposed inside the housing to transmit a driving motor (not shown) and driving force for generating driving force.

On the other hand, the gearbox generates various vibrations in the gearbox due to unstable movement in the process of generating and transmitting the driving force using the driving motor, and these vibrations may be transmitted to the drive shaft and the housing forming the gearbox. In addition, when the shaft vibration of the drive shaft is transmitted to the housing, in addition to the vibration of the housing itself, vibration of other components connected to the housing may be caused, and noise due to the vibration may also be induced.

In the past, experts analyzed the change in vibration amount and vibration characteristics of the gearbox to find the cause of the defect and set up appropriate countermeasures. However, due to the lack of experience of experts and changes in diversified facility characteristics, it is difficult to make an accurate diagnosis, so the method of setting the size of vibration and notifying the manager of an abnormal condition only when vibration exceeding the size occurs has been used. However, this method can only determine the degree of failure of the rotating machine 100, and since the cause or progress of the failure cannot be known and thus cannot be actively dealt with, the judgment is again entrusted to an expert. Therefore, when a defect occurs in the rotating machine 100, it is necessary to identify the cause of the defect in advance at the stage before the failure and predict the progress to prevent accidents in advance. Accurate diagnosis is required.

The defect detection apparatus 200 may detect whether the rotating machine 100 is defective by receiving the vibration signal generated from the rotating machine 100 as an input and performing statistical analysis and machine learning. Here, the defect may include a case in which the rotating machine 100 is operating normally and a case in which the rotating machine 100 is malfunctioning. In this embodiment, the defect detection apparatus 200 may include a vibration sensor 210 , a pre-processing unit 220 , a calculation unit 230 , a coupling unit 240 , a classifying unit 250 , and a determining unit 260 . can

In this embodiment, "unit" may be a hardware component such as a processor or circuit, and/or a software component executed by a hardware component such as a processor.

The vibration sensor 210 may be provided in the rotating machine 100 to detect and output a vibration signal generated in the rotating machine 100 . In this embodiment, the vibration sensor 210 may include a first vibration sensor 211 to a fourth vibration sensor 214 .

FIG. 2 is an exemplary view illustrating an output direction of a vibration signal output by the vibration sensor in FIG. 1 . Referring to FIG. 2 , the first vibration sensor 211 may output a first vibration signal g_x vibration signal by detecting vibration in the horizontal (x) direction generated in the rotating machine 100 . The second vibration sensor 212 may output a second vibration signal (g_y vibration signal) by detecting vibration in the vertical (y) direction generated in the rotating machine 100 . The third vibration sensor 213 may output a third vibration signal (g_z vibration signal) by detecting vibration in the axis (z) direction generated in the rotating machine 100 . The fourth vibration sensor 214 detects the vibration in the diagonal (T) direction, which is a combination of the horizontal (x) direction, the vertical (y) direction, and the axial (z) direction, generated in the rotating machine 100 to generate a fourth vibration signal ( g_T vibration signal) can be output.

3 is an exemplary view illustrating a vibration signal output from a machine in a normal state and a failure state according to the present embodiment. Referring to FIG. 3, FIG. 3A shows a first vibration signal (g_x vibration signal) generated in the rotating machine 100 in a normal state, and FIG. 3B is a rotating machine 100 in a broken state. The first vibration signal (g_x vibration signal) generated in . In particular, FIG. 3b shows a first vibration signal generated in the rotating machine 100 when a tooth is broken.

Returning to FIG. 1 , the preprocessor 220 may convert the time domain vibration signal output from the vibration sensor 210 into a frequency domain vibration signal. The frequency domain vibration signal contains more unique information that can be used to distinguish between normal and malfunction of the rotating machine 100 . In this embodiment, the preprocessor 220 converts the first to fourth vibration signals in the time domain output from the first vibration sensors 211 to the fourth vibration sensors 214 by fast Fourier transform (FFT). ) to output the first to fourth vibration signals in the frequency domain. The Fourier transform of a time function is a complex-valued function of frequency, the magnitude (absolute value) indicates the amount of frequency present in the original function, and the argument may be the offset of the fundamental sinusoid at that frequency.

In this embodiment, the preprocessor 220 may include a first preprocessor 221 to a fourth preprocessor 224 . The first preprocessor 221 may convert the first vibration signal g_x vibration signal in the time domain output from the first vibration sensor 211 into a first vibration signal g_x FFT signal in the frequency domain. The second preprocessor 222 may convert a second vibration signal (g_y vibration signal) of the time domain output from the second vibration sensor 212 into a second vibration signal (g_y FFT signal) of the frequency domain. The third preprocessor 223 may convert the third vibration signal g_z vibration signal of the time domain output from the third vibration sensor 213 into a third vibration signal g_z FFT signal of the frequency domain. The fourth preprocessor 224 may convert the fourth vibration signal g_T vibration signal of the time domain output from the fourth vibration sensor 214 into a fourth vibration signal g_T FFT signal of the frequency domain.

4 is a view showing a result of pre-processing the vibration signal output from the machine in a normal state and a failure state according to the present embodiment. Referring to FIG. 4, FIG. 4a is a first vibration signal (g_x FFT signal) in the frequency domain by preprocessing a first vibration signal (g_x vibration signal) in the time domain generated in the rotating machine 100 in a normal state. An example of the output is shown, and FIG. 4b is a first vibration signal (g_x FFT signal) in the frequency domain by preprocessing the first vibration signal (g_x vibration signal) in the time domain generated in the rotating machine 100 in a broken state. ) is shown as an example.

Returning to FIG. 1 , the calculator 230 may calculate a statistical characteristic value for the vibration signal of the frequency domain output from the preprocessor 220 . The frequency domain vibration signal output from the preprocessor 220 can be directly input to the deep neural network model to be described later, but since it requires a complex structure and a long training time when working with high-dimensional input data, it is input to the deep neural network model. Statistical characteristic values can be calculated to reduce the data.

On the other hand, the preprocessor 220, the first vibration sensor 211 to the fourth vibration sensor 214 may be configured to evaluate the first vibration signal to the fourth vibration signal received from the selected collection.

Among the received vibration signals, there may be vibration signals that are not suitable for evaluating the state of the machine due to noise or other external environment changes, internal processing errors, and the like. Therefore, it may be necessary to evaluate the suitability of the vibration signal and select and collect only the appropriate signal.

For example, in the evaluation of the vibration signals, a combination of the first vibration signal to the third vibration signal collected from the first vibration sensor 211 to the third vibration sensor 213 is obtained from the fourth vibration sensor 214 . The first to fourth vibration signals are compared with the collected fourth vibration signal to evaluate whether the first to fourth vibration signals are abnormal, and such selective collection is performed when the first to fourth vibration signals are normal based on the above-described evaluation. can be done

In this embodiment, the calculator 230 calculates a mean value, a median value, a minimum value, a maximum value, a kurtosis value, and a skewness of the vibration signal in the frequency domain. A statistical characteristic value including at least one of a value, a standard deviation value, and a peak-to-peak value may be calculated.

Here, the kurtosis is an index indicating the degree of sharpness of the probability distribution, and the kurtosis may increase as the distribution of signal values gathers near a specific value to form a sharp shape. In addition, skewness is an indicator indicating the asymmetry of the probability distribution, and as the signal bias (the degree to which the distribution of signal values shifts to one side with respect to the average of the signal) increases, the skewness may also increase. The positive amplitude (peak-to-peak) is an index indicating the overall width of the signal and may indicate the difference between the smallest value and the largest value in the signal.

In this embodiment, the calculator 230 may include a first calculator 231 to a fourth calculator 234 . The first calculator 231 may calculate the above-described eight statistical characteristic values for the first vibration signal g_x FFT signal of the frequency domain output from the first preprocessor 221 . The second calculator 232 may calculate the above-described eight statistical characteristic values for the second vibration signal g_y FFT signal of the frequency domain output from the second preprocessor 222 . The third calculator 233 may calculate the above-described eight statistical characteristic values for the third vibration signal g_z FFT signal of the frequency domain output from the third preprocessor 223 . The fourth calculation unit 234 may calculate the above-described eight statistical characteristic values for the fourth vibration signal g_T FFT signal of the frequency domain output from the fourth preprocessor 224 .

The combiner 240 may combine the calculation results of the statistical characteristic values output from the first calculator 231 to the fourth calculator 234 and transmit it to the classifier 250 . In the above-described example, since the first calculator 231 to the fourth calculator 234 calculate 8 statistical characteristic values, the combiner 240 transmits a total of 32 statistical characteristic values to the classification unit 250 . can

Classification unit 250 uses a plurality of deep neural network models implemented with different algorithms, previously trained to determine whether the machine is defective by using the statistical characteristic value of the vibration signal derived from the vibration sensor, to determine whether the machine is defective. Classification values can be output. In this embodiment, the deep neural network model may be a neural network model trained in advance using training data including statistical characteristic values of vibration signals generated in the machine and the state of the machine (either normal or malfunctioning). In this way, the classification unit 250 does not use the vibration signal of the time domain, but uses the statistical characteristic value calculated from the vibration signal of the frequency domain to output the classification value of whether the rotating machine 100 is defective. can be lowered

As an optional embodiment, the deep neural network model includes a first vibration signal (g_x FFT signal) in the frequency domain, a second vibration signal (g_y FFT signal) in the frequency domain, and a third vibration signal (g_z FFT signal) in the frequency domain, It may be configured to determine whether the rotating machine 100 has a failure based on a maximum value and a kurtosis value among statistical characteristic values for the fourth vibration signal (g_T FFT signal) in the frequency domain.

5 is a view showing the results of calculating statistical characteristic values from the preprocessing results for the vibration signal output from the machine in the normal state and the broken state according to the present embodiment. Referring to FIG. 5 , FIG. 5A is a first vibration signal in the frequency domain obtained by preprocessing the first vibration signal g_x vibration signal in the time domain generated in the rotating machine 100 in a normal state and a broken state. The distribution of statistical characteristics calculated from (g_x FFT signal) is shown. 5B is a second vibration signal (g_y FFT signal) in the frequency domain obtained by preprocessing the second vibration signal (g_y vibration signal) in the time domain generated in the rotating machine 100 in a normal state and a broken state. The distribution of the calculated statistical characteristics is shown. Figure 5c is from the first vibration signal (g_z FFT signal) in the frequency domain pre-processed the third vibration signal (g_z vibration signal) in the time domain generated in the rotating machine 100 in a normal state and a broken state The distribution of the calculated statistical characteristics is shown. FIG. 5d shows the fourth vibration signal (g_T FFT signal) in the frequency domain obtained by pre-processing the fourth vibration signal (g_T vibration signal) in the time domain generated in the rotating machine 100 in a normal state and a broken state. The distribution of the calculated statistical characteristics is shown.

Among the distributions of statistical characteristics calculated from the first vibration signal (g_x FFT signal) in the frequency domain in FIG. 5A , the difference between the maximum value of the normal state and the broken state is shown in FIGS. 5B to 5D . A normal state calculated from the second vibration signal of the frequency domain (g_y FFT signal), the third vibration signal of the frequency domain (g_z FFT signal) and the fourth vibration signal of the frequency domain (g_T FFT signal) shown in Fig. It can be seen that the difference between the maximum values of the broken state is larger than the difference.

In addition, in the distribution of statistical characteristics calculated from the first vibration signal (g_x FFT signal) in the frequency domain in FIG. 5A, the difference between the kurtosis values of the normal state and the broken state is shown in FIGS. 5B to The normal calculated from the second vibration signal (g_y FFT signal) of the frequency domain, the third vibration signal (g_z FFT signal) of the frequency domain, and the fourth vibration signal (g_T FFT signal) of the frequency domain shown in FIG. 5D It can be seen that the difference between the kurtosis values of the state and the broken state is larger than the difference.

From FIG. 5 , the maximum value and the kurtosis value become the criteria for determining whether the rotating machine 100 is defective is due to the analysis result obtained by training the deep neural network with the above-described training data through an experiment. That is, the first vibration signal (g_x FFT signal) of the frequency domain, the second vibration signal (g_y FFT signal) of the frequency domain, the third vibration signal (g_z FFT signal) of the frequency domain, and the fourth vibration of the frequency domain As a result of analyzing the trained learning model after training the deep neural network with statistical characteristic values for the g_T FFT signal, it was found that good criteria for judging whether the rotating machine 100 is defective are the maximum value and the kurtosis value.

6 is a diagram illustrating distributions of maximum values and kurtosis values among statistical characteristic values from preprocessing results for vibration signals output from the machine in a steady state and a failure state according to the present embodiment. Referring to FIG. 6 , FIG. 6A is a first vibration signal in the frequency domain obtained by preprocessing the first vibration signal g_x vibration signal in the time domain generated in the rotating machine 100 in a normal state and a broken state. The distribution of the maximum value and the kurtosis value among the statistical characteristic values calculated from (g_x FFT signal) is shown. 6B is a second vibration signal (g_y FFT signal) in the frequency domain obtained by preprocessing the second vibration signal (g_y vibration signal) in the time domain generated in the rotating machine 100 in a normal state and a broken state. The distribution of the maximum value and the kurtosis value among the calculated statistical characteristic values is shown. Figure 6c shows the third vibration signal (g_z FFT signal) in the frequency domain obtained by preprocessing the third vibration signal (g_z vibration signal) in the time domain generated in the rotating machine 100 in the normal state and the broken state. The distribution of the maximum value and the kurtosis value among the calculated statistical characteristic values is shown. Figure 6d is from the fourth vibration signal (g_T FFT signal) in the frequency domain pre-processed the fourth vibration signal (g_T vibration signal) in the time domain generated in the rotating machine 100 in a normal state and a broken state The distribution of the maximum value and the kurtosis value among the calculated statistical characteristic values is shown.

In FIG. 6A , the distribution of the maximum (max) value and the kurtosis value of the normal state and the broken state calculated from the first vibration signal (g_x FFT signal) of the frequency domain is identified in FIGS. 6B to The normal calculated from the second vibration signal (g_y FFT signal) of the frequency domain, the third vibration signal (g_z FFT signal) of the frequency domain, and the fourth vibration signal (g_T FFT signal) of the frequency domain shown in FIG. 6D It can be seen that it is clearer than the identification of the distribution of the max and kurtosis values of the state and the broken state.

The fact that the first vibration signal (g_y FFT signal) in the frequency domain from FIG. 6 serves as a criterion for determining whether or not the machine is defective is not limited thereto. signal) to the fourth vibration signal (g_T FFT signal) may be a reference.

In this embodiment, the defect detection apparatus 200 may execute an artificial intelligence (AI) algorithm and/or a machine learning algorithm in a 5G communication environment to detect a defect of the rotating machine 100 . have.

Here, artificial intelligence (AI) is a field of computer science and information technology that studies how computers can do the thinking, learning, and self-development that can be done with human intelligence. It could mean making it possible to imitate intelligent behavior.

Also, AI does not exist by itself, but has many direct and indirect connections with other fields of computer science. In particular, in modern times, attempts are being made to introduce artificial intelligence elements in various fields of information technology and use them to solve problems in that field.

Machine learning is a branch of artificial intelligence, which can include fields of study that give computers the ability to learn without explicit programming. Specifically, machine learning can be said to be a technology that studies and builds a system and an algorithm for learning based on empirical data, making predictions, and improving its own performance. Algorithms in machine learning can take the approach of building specific models to make predictions or decisions based on input data, rather than executing strictly set static program instructions.

As a machine learning method of such an artificial neural network, both unsupervised learning and supervised learning may be used.

In addition, deep learning technology, which is a type of machine learning, can learn by going down to a deep level in multiple stages based on data. Deep learning can represent a set of machine learning algorithms that extract core data from a plurality of data as the level increases.

The deep learning structure may include an artificial neural network (ANN). For example, the deep learning structure is composed of a deep neural network (DNN) such as a convolutional neural network (CNN), a recurrent neural network (RNN), and a deep belief network (DBN). can be The deep learning structure according to the present embodiment may use various well-known structures. For example, the deep learning structure according to the present embodiment may include CNN, RNN, DBN, and the like. RNN is widely used in natural language processing, etc., and is an effective structure for processing time-series data that changes with the passage of time. DBN may include a deep learning structure composed of multiple layers of restricted boltzman machine (RBM), a deep learning technique. By repeating RBM learning, when a certain number of layers is reached, a DBN having the corresponding number of layers can be configured. CNNs can include models that simulate human brain functions based on the assumption that when a person recognizes an object, it extracts the basic features of an object, then performs complex calculations in the brain and recognizes an object based on the result. .

7 is a block diagram schematically illustrating the configuration of the classification unit in FIG. 1 . In the following description, descriptions of parts overlapping with those of FIGS. 1 to 5 will be omitted. Referring to FIG. 7 , the classification unit 250 may include a first deep neural network model 251 , a second deep neural network model 252 , and a third deep neural network model 253 .

In this embodiment, the first deep neural network model 251 to the third deep neural network model 253 may be implemented with different algorithms. In an embodiment, the first deep neural network model 251 may be implemented as a feed forward neural network (FFNN) algorithm, and the second deep neural network model 252 may be implemented as a support vector machine (SVM) algorithm, and the first deep neural network model 251 may be implemented as a support vector machine (SVM) algorithm. 3 The deep neural network model 253 may be implemented as a logistic regression (LR) algorithm. In addition, in this embodiment, although the number of deep neural network models is limited to three, it is not limited thereto, and various embodiments such as three or more or less than three may be applied.

The calculation result of the combined statistical characteristic value output from the combining unit 240 may be input to each of the first deep neural network model 251 to the third deep neural network model 253 of the classifying unit 250 . Each of the first deep neural network model 251 to the third deep neural network model 253 may output a classification value for whether the machine is defective. In one embodiment, each of the first deep neural network model 251 to the third deep neural network model 253 is a classification value indicating the normality of the machine 100 as a classification value for whether the machine is defective or a normal classification value or machine 100 ), the failure classification value can be output as the classification value indicating the failure.

The determination unit 260 may finally determine whether a machine is defective by applying a voting algorithm to a classification value for whether a plurality of machines are defective. That is, the determiner 260 may finally determine whether or not the machine is defective by applying a voting algorithm to the classification values output from the first deep neural network model 251 to the third deep neural network model 253 .

In an embodiment of the voting algorithm, an output value determined by the determination unit 260 by applying a major sample voting (MSV) algorithm is as shown in Equation 1 below.

은 정상 투표 결과값을 나타내고,

는 고장 투표 결과값을 나타낼 수 있다. 이로써, 수학식 1은 정상 투표 결과값 및 고장 투표 결과값 중 더 큰 값을 출력한다는 의미를 포함할 수 있다.in Equation 1

represents the normal voting result,

may represent the failure voting result. Accordingly, Equation 1 may include the meaning of outputting a larger value among the normal voting result value and the malfunctioning voting result value.

) 및 고장 투표 결과값(

)를 산출하는 수식은 하기 수학식 2와 같다.Here, the normal voting result (

) and failure voting results (

) is the same as in Equation 2 below.

In Equation 2, A is a first representative value assigned to the classification values of the first deep neural network model 251 to the third deep neural network model 253 when calculating the normal voting result value, and when the classification value is normal, the second A first value (eg, 1) may be allocated as the 1 representative value, and a second value (eg, 0) may be allocated as the first representative value when the classification value is a failure. In addition, B is a second representative value assigned to the classification values of the first deep neural network model 251 to the third deep neural network model 253 when calculating the failure voting result value, and when the classification value is a failure, the second representative value A first value (eg, 1) may be allocated as a value, and a second value (eg, 0) may be allocated as a second representative value when the classification value is normal.

는 가중치를 나타내며, 수학식 3과 같이 산출할 수 있다. 본 실시 예에서 가중치는 훈련 프로세스 이후에 한 번 업데이트 될 수 있다. 반면에, 투표값

및

는 복수개의 심층신경망 모델을 선호하며, 심층신경망 모델의 개수가 많을수록 더 정확한 투표 결과를 획득할 수 있다.In Equation 2,

denotes a weight, and can be calculated as in Equation (3). In this embodiment, the weights may be updated once after the training process. On the other hand, the vote

and

prefers multiple deep neural network models, and as the number of deep neural network models increases, more accurate voting results can be obtained.

In Equation 3, T may be the number of deep neural network models, 3 in this embodiment. In addition, the accuracy may include the accuracy of the deep neural network model obtained in the pre-training stage, as shown in FIG. 8 .

Referring to FIG. 8 , after training the first deep neural network model 251 to the third deep neural network model 253 using training data, the first deep neural network model 251 to the third deep neural network model 251 using test data The accuracy as a result of evaluating the neural network model 253 is shown.

When testing the deep neural network model from FIG. 8 , when the test data among the Full FFT signals output from the preprocessor 220 is input as the input of the first deep neural network model 251 to the third deep neural network model 253 , It can be seen that 100% accuracy is calculated. In addition, when testing the deep neural network model, when the test data among the reduced FFT signals output from the calculator 230 is input as the input of the first deep neural network model 251 to the third deep neural network model 253 , It can be seen that the accuracy may be 100% or less than 100%. Therefore, it is possible to improve the accuracy of voting results by applying a larger weight to a deep neural network model with higher accuracy and a smaller weight to a deep neural network model having lower accuracy.

The determiner 260 may calculate a weight based on the accuracy obtained in the pre-training step for each of a plurality of deep neural network models implemented with different algorithms.

로 산출할 수 있다. 또한, 제2 심층신경망 모델(252)의 정확도는 91.7% 즉 0.917이고, 이를 수학식 3에 대응하면, 제2 심층신경망 모델(252)의 가중치를

로 산출할 수 있다. 또한, 제3 심층신경망 모델(253)의 정확도는 91.7% 즉 0.917이고, 이를 수학식 3에 대응하면, 제3 심층신경망 모델(253)의 가중치를

로 산출할 수 있다.For example, from FIG. 8, the accuracy of the first deep neural network model 251 is 100%, that is, 1, and if it corresponds to Equation 3, the weight of the first deep neural network model 251 is

can be calculated as In addition, the accuracy of the second deep neural network model 252 is 91.7%, that is, 0.917, and if it corresponds to Equation 3, the weight of the second deep neural network model 252 is

can be calculated as In addition, the accuracy of the third deep neural network model 253 is 91.7%, that is, 0.917, and if it corresponds to Equation 3, the weight of the third deep neural network model 253 is

can be calculated as

The determination unit 260 is a first representative value for calculating the normal voting result value in the classification value for whether or not the machine is defective output from the deep neural network model in which the weight is set, and when the classification value is normal, the first value (for example, , 1), and when the classification value is a failure, a second value (eg, 0) may be assigned.

)은 제2값(0)으로 할당되고, 제2 심층신경망 모델(252)의 제1 할당값(

)은 제2값(0)으로 할당되고, 제3 심층신경망 모델(253)의 제1 할당값(

)은 제1값(1)로 할당될 수 있다.For example, from FIG. 8 , the classification value of the first deep neural network model 251 is classified as failure, the classification value of the second deep neural network model 252 is classified as failure, and the third deep neural network model 253 is classified as failure. When the classification value of is classified as normal, the first allocation value of the first deep neural network model 251 (

) is assigned as the second value (0), and the first assigned value (

) is assigned as the second value (0), and the first assigned value (

) may be assigned as the first value (1).

In addition, the determination unit 260 is a second representative value for calculating the abnormal voting result value in the classification value for whether or not the machine is defective output from the deep neural network model in which the weight is set, and the first value ( For example, 1) may be allocated, and when the classification value is normal, a second value (eg, 0) may be allocated.

)은 제1값(1)으로 할당되고, 제2 심층신경망 모델(252)의 제2 할당값(

)은 제1값(1)으로 할당되고, 제3 심층신경망 모델(253)의 제2 할당값(

)은 제2값(0)로 할당될 수 있다.For example, from FIG. 8 , the classification value of the first deep neural network model 251 is classified as failure, the classification value of the second deep neural network model 252 is classified as failure, and the third deep neural network model 253 is classified as failure. When the classification value of is classified as normal, the second allocation value of the first deep neural network model 251 (

) is assigned as the first value (1), and the second assigned value (

) is assigned as the first value (1), and the second assigned value (

) may be assigned as the second value (0).

,

,

)에 가중치(

,

,

)를 곱하여 합산한 결과로서의 정상 투표 결과값(

)과, 복수개의 심층신경망 모델 각각에 대하여 제2 대표값(

,

,

)에 가중치(

,

,

)를 곱하여 합산한 결과로서의 고장 투표 결과값(

)을 비교하여 더 큰 투표 결과값에 해당하는 분류값을 상기 기계의 결함 여부로 최종 결정할 수 있다.The determination unit 260 is a first representative value for each of the plurality of deep neural network models (

,

,

) to weight(

,

,

) multiplied by the normal voting result (

) and the second representative value (

,

,

) to weight(

,

,

) and the result of the summation of the failure voting result (

), a classification value corresponding to a larger voting result can be finally determined as whether the machine is defective.

,

,

)에 가중치(

,

,

)를 곱하여 합산한 결과로서의 정상 투표 결과값(

)을 산출할 수 있다.For example, from FIG. 8 , the classification value of the first deep neural network model 251 is classified as failure, the classification value of the second deep neural network model 252 is classified as failure, and the third deep neural network model 253 is classified as failure. When the classification value of is classified as normal, the determiner 260 determines the first representative value (

,

,

) to weight(

,

,

) multiplied by the normal voting result (

) can be calculated.

,

,

)에 가중치(

,

,

)를 곱하여 합산한 결과로서의 고장 투표 결과값(

)을 산출할 수 있다.In addition, the determination unit 260 for each of the plurality of deep neural network models, the second representative value (

,

,

) to weight(

,

,

) and the result of the summation of the failure voting result (

) can be calculated.

) 및 고장 투표 결과값(

)의 비교 결과, 고장 투표 결과값이 더 큼에 따라 기계(100)의 결함 여부로 <고장>을 최종 결정할 수 있다.The determination unit 260 is a normal voting result value (

) and failure voting results (

), as the result of the failure vote is larger, <failure> may be finally determined as whether the machine 100 is defective.

9 is a flowchart illustrating a method for detecting a machine defect according to the present embodiment. In the following description, descriptions of parts overlapping with those of FIGS. 1 to 8 will be omitted.

Referring to FIG. 9 , in step S910 , the defect detection apparatus 200 receives a vibration signal generated in the rotating machine 100 from a vibration sensor provided in the rotating machine 100 . In this embodiment, the vibration sensor may include a first vibration sensor to a fourth vibration sensor. The defect detection device 200 receives a first vibration signal (g_x vibration signal) from a first vibration sensor that senses vibration in the horizontal (x) direction generated in the rotating machine 100 , and is generated in the rotating machine 100 . receiving a second vibration signal (g_y vibration signal) from a second vibration sensor that senses vibration in the vertical (y) direction The third vibration signal (g_z vibration signal) is received from the sensor, and the vibration in the horizontal (x) direction, the vertical (y) direction, and the axial (z) direction generated in the rotating machine 100 is combined in the diagonal (T) direction. A fourth vibration signal (g_T vibration signal) may be received from the detected fourth vibration sensor.

Among the vibration signals received in step S910, there may be vibration signals unsuitable for evaluating the state of the machine due to noise or other external environment changes, internal processing errors, and the like.

Therefore, before transferring data to the following step S920, the combination of the first to third vibration signals collected from the first to third vibration sensors is compared with the fourth vibration signal collected from the fourth vibration sensor, and the first It is possible to evaluate whether the to fourth vibration signal is abnormal.

In addition, based on the result of the evaluation step, only the vibration signals when the first to fourth vibration signals are normal may be selectively collected and transmitted as data to be used in the next step.

The reason for the above evaluation and selection collection is that the fourth direction measured by the fourth vibration sensor is a direction that can be derived by a combination of the first to third directions measured by the first to third vibration sensors. am. That is, the first to third vibration sensors are independent of each other by measuring directions perpendicular to each other, but the vibration signal in the fourth direction measured by the fourth vibration sensor is generated by a combination of vibration signals in the first to third directions. This is because it is a direction that can be derived. Accordingly, the relationship (angle) between the first to third directions and the fourth direction is known, and if the first to third vibration signals are given, the fourth vibration signal can be predicted, and if the fourth vibration signal is different from the expected It can be assumed that there is an error in the signal.

If the fourth vibration signal is a vibration signal that cannot be derived by a combination of the first to third vibration signals when the direction relationship is considered, the corresponding vibration signal set may be evaluated as abnormal data.

Therefore, only the vibration signal suitable for judging the failure of the machine can be used in a subsequent step, thereby enabling more accurate judgment.

In step S920, the defect detection apparatus 200 converts the vibration signal of the time domain output from the vibration sensor into a vibration signal of the frequency domain. The defect detection apparatus 200 generates a first vibration signal (g_x FFT signal) in the frequency domain by fast Fourier transforming the first vibration signal (g_x vibration signal) of the time domain received from the first vibration sensor, and the second vibration A second vibration signal (g_y FFT signal) in the frequency domain may be generated by fast Fourier transforming the second vibration signal (g_y vibration signal) of the time domain received from the sensor. The defect detection apparatus 200 may generate a frequency domain third vibration signal (g_z FFT signal) by fast Fourier transforming the time domain third vibration signal (g_z vibration signal) received from the third vibration sensor, A fourth vibration signal (g_T FFT signal) in the frequency domain may be generated by fast Fourier transforming the fourth vibration signal (g_T vibration signal) in the time domain output from the vibration sensor.

In step S930, the defect detection apparatus 200 calculates a statistical characteristic value for the vibration signal in the frequency domain. The defect detection apparatus 200 includes a mean value, a median value, a minimum value, a maximum value, a kurtosis value, a skewness value, and a standard value of the vibration signal in the frequency domain. A statistical characteristic value including at least one of a standard deviation value and a peak-to-peak value may be calculated. The defect detection apparatus 200 calculates a statistical characteristic value with respect to a first vibration signal (g_x FFT signal) of a frequency domain, calculates a statistical characteristic value with respect to a second vibration signal (g_y FFT signal) of a frequency domain, and frequency A statistical characteristic value may be calculated with respect to a third vibration signal (g_z FFT signal) in the domain, and a statistical characteristic value may be calculated with respect to a fourth vibration signal (g_T FFT signal) in the frequency domain.

In step S940, the defect detection apparatus 200 uses a plurality of deep neural network models implemented with different algorithms, previously trained to determine whether the machine is defective by using the statistical characteristic value of the vibration signal derived from the vibration sensor. Outputs the classification value of whether or not the machine is defective. In this embodiment, the deep neural network model may be a neural network model trained in advance using training data including statistical characteristic values of vibration signals generated in the machine and the state of the machine (either normal or malfunctioning).

In this embodiment, the defect detection apparatus 200 is a plurality of deep neural network models implemented with different algorithms, and may include a first deep neural network model to a third deep neural network model. The first deep neural network model may be implemented as a feed forward neural network (FFNN) algorithm, the second deep neural network model may be implemented as a support vector machine (SVM) algorithm, and the third deep neural network model may be implemented as a logistic regression (LR) algorithm. It can be implemented as an algorithm. In addition, in this embodiment, although the number of deep neural network models is limited to three, it is not limited thereto, and various embodiments such as three or more or less than three may be applied.

The calculation result of the statistical characteristic value may be input to each of the first deep neural network model 251 to the third deep neural network model 253 . Each of the first deep neural network model 251 to the third deep neural network model 253 may output a classification value for whether the machine is defective. Each of the first deep neural network model to the third deep neural network model is a classification value for whether the machine is defective, and a classification value indicating the normality of the machine 100 is classified as a normal classification value or a classification value indicating a failure of the machine 100. value can be printed.

In step S950 , the defect detection apparatus 200 finally determines whether the machine is defective by applying a voting algorithm to the classification values for whether the plurality of machines are defective.

,

,

)를 산출할 수 있다. 가중치의 산출은 상술한 수학식 3을 참조하여 산출할 수 있다.The defect detection apparatus 200 performs weighting (

,

,

) can be calculated. The weight can be calculated with reference to Equation 3 above.

,

,

)으로, 분류값이 정상인 경우 제1값(예를 들어, 1)을 할당하고, 분류값이 고장인 경우 제2값(예를 들어, 0)을 할당할 수 있다.The defect detection apparatus 200 includes a first representative value (

,

,

), a first value (eg, 1) may be allocated when the classification value is normal, and a second value (eg, 0) may be allocated when the classification value is malfunctioning.

,

,

)으로, 분류값이 고장인 경우 제1값(예를 들어, 1)을 할당하고, 분류값이 정상인 경우 제2값(예를 들어, 0)을 할당하며,The defect detection device 200 includes a second representative value (

,

,

), a first value (eg, 1) is assigned when the classification value is faulty, and a second value (eg, 0) is assigned when the classification value is normal,

,

,

)에 가중치(

,

,

)를 곱하여 합산한 결과로서의 정상 투표 결과값(

)과, 복수개의 심층신경망 모델 각각에 대하여 제2 대표값(

,

,

)에 가중치(

,

,

)를 곱하여 합산한 결과로서의 고장 투표 결과값(

)을 비교하여 더 큰 투표 결과값에 해당하는 분류값을 상기 기계의 결함 여부로 최종 결정할 수 있다.The defect detection apparatus 200 has a first representative value (

,

,

) to weight(

,

,

) multiplied by the normal voting result (

) and the second representative value (

,

,

) to weight(

,

,

) and the result of the summation of the failure voting result (

), a classification value corresponding to a larger voting result can be finally determined as whether the machine is defective.

The embodiment according to the present invention described above may be implemented in the form of a computer program that can be executed through various components on a computer, and such a computer program may be recorded in a computer-readable medium. In this case, the medium includes a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk, and a ROM. , RAM, flash memory, and the like, hardware devices specially configured to store and execute program instructions.

Meanwhile, the computer program may be specially designed and configured for the present invention, or may be known and used by those skilled in the computer software field. Examples of the computer program may include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

In the specification of the present invention (especially in the claims), the use of the term "above" and similar referential terms may be used in both the singular and the plural. In addition, when a range is described in the present invention, each individual value constituting the range is described in the detailed description of the invention as including the invention to which individual values belonging to the range are applied (unless there is a description to the contrary). same as

The steps constituting the method according to the present invention may be performed in an appropriate order, unless there is an explicit order or description to the contrary. The present invention is not necessarily limited to the order in which the steps are described. The use of all examples or exemplary terms (eg, etc.) in the present invention is merely for the purpose of describing the present invention in detail, and the scope of the present invention is limited by the examples or exemplary terms unless defined by the claims. it's not going to be In addition, those skilled in the art will appreciate that various modifications, combinations, and changes may be made in accordance with design conditions and factors within the scope of the appended claims or their equivalents.

Therefore, the spirit of the present invention should not be limited to the above-described embodiments, and the scope of the spirit of the present invention is not limited to the scope of the scope of the present invention. will be said to belong to

100: machine
200: defect detection device
210: vibration sensor
220: preprocessor
230: output unit
240: coupling part
250: classification unit
260: decision part

Claims (17)

A method for detecting a fault in a machine, comprising:
Receiving a vibration signal generated in the machine by a vibration sensor provided in the machine;
converting the time domain vibration signal into a frequency domain vibration signal;
calculating statistical characteristic values for the vibration signal in the frequency domain;
Using a plurality of deep neural network models implemented with different algorithms, trained in advance to determine whether a machine is defective by using the statistical characteristic value of the vibration signal derived from the vibration sensor, the classification value for whether the machine is defective is output. to do; and
Comprising the step of finally determining whether the machine is defective by applying a voting algorithm to the classification value of whether the plurality of machines are defective,
Each of the plurality of deep neural network models,
It is a pre-trained neural network model using training data including statistical characteristic values of vibration signals generated by the machine and the state of the machine.
Receiving the vibration signal comprises:
receiving a vibration signal in the horizontal direction generated in the machine by a first vibration sensor;
receiving a vibration signal in the vertical direction generated in the machine by a second vibration sensor;
receiving a vibration signal in the depth direction generated in the machine by a third vibration sensor; and
Receiving a vibration signal in a diagonal direction based on a horizontal direction, a vertical direction, and a depth direction generated in the machine by a fourth vibration sensor,
The deep neural network model is
configured to determine whether the machine is defective based on the maximum value and kurtosis value among statistical characteristic values for vibration signals in the horizontal direction, vertical direction, depth direction and diagonal direction of the frequency domain,
How to detect a machine's faults. delete The method of claim 1,
between the receiving step and the converting step,
A combination of first to third vibration signals collected from the first to third vibration sensors is compared with a fourth vibration signal collected from the fourth vibration sensor to evaluate whether the first to fourth vibration signals are abnormal to do; and
Further comprising the step of collecting only the vibration signal when the first to fourth vibration signals are normal based on the result of the evaluating step,
A method of detecting defects in a machine. The method of claim 1,
The step of converting the frequency domain into a vibration signal,
converting a time domain horizontal vibration signal output from a first vibration sensor provided in the machine into a frequency domain horizontal vibration signal;
converting the time domain vertical vibration signal output from a second vibration sensor provided in the machine into a frequency domain vertical vibration signal;
converting a depth direction vibration signal output from a third vibration sensor provided in the machine into a depth direction vibration signal of a frequency domain; and
The vibration signal in the diagonal direction based on the horizontal direction, the vertical direction, and the depth direction of the time domain output from the fourth vibration sensor provided in the machine, the diagonal direction based on the horizontal direction, the vertical direction and the depth direction of the frequency domain Including the step of converting to a vibration signal of
A method of detecting defects in a machine. The method of claim 1,
Calculating the statistical characteristic value comprises:
Mean value, median value, min value, max value, kurtosis value, skewness value, standard deviation value of the vibration signal of the frequency domain And, calculating a statistical characteristic value including one or more of the positive amplitude (peak-to-peak) value,
A method of detecting defects in a machine. 5. The method of claim 4,
Calculating the statistical characteristic value comprises:
calculating statistical characteristic values for the horizontal vibration signal of the frequency domain;
calculating statistical characteristic values for the vertical vibration signal in the frequency domain;
calculating statistical characteristic values for the vibration signal in the depth direction of the frequency domain; and
Comprising the step of calculating statistical characteristic values for the vibration signal in the diagonal direction based on the horizontal direction, the vertical direction and the depth direction of the frequency domain,
A method of detecting defects in a machine. delete The method of claim 1,
The step of finally determining whether the machine is defective is,
calculating weights based on the accuracy obtained in the pre-training step for each of the plurality of deep neural network models implemented with the different algorithms;
A first representative value for calculating a normal voting result value is assigned to a classification value for whether or not the machine is defective output from the deep neural network model in which the weight is set, and a first value is assigned when the classification value is normal, and the classification value assigning a second value in case of this failure;
To the classification value for whether the machine is defective, output from the deep neural network model to which the weight is set, as a second representative value for calculating a failure vote result value, a first value is assigned when the classification value is failure, and the classification assigning a second value if the value is normal; and
For each of the plurality of deep neural network models, a normal voting result value as a result of multiplying the first representative value by the weight and summing the result by multiplying the weight by the second representative value for each of the plurality of deep neural network models Comprising the step of comparing the failure voting result value as a final decision as to whether or not the machine is defective as a classification value corresponding to a larger voting result value,
A method of detecting defects in a machine. A computer-readable recording medium storing a computer program for executing the method of any one of claims 1, 3 to 6, and 8 using a computer. A device for detecting a fault in a machine, comprising:
a vibration sensor provided in the machine to detect a vibration signal generated by the machine;
a pre-processing unit for converting a time domain vibration signal output from the vibration sensor into a frequency domain vibration signal;
a calculator for calculating statistical characteristic values for the vibration signal in the frequency domain output from the preprocessor;
Using a plurality of deep neural network models implemented with different algorithms, pre-trained to determine whether a machine is defective by using the statistical characteristic value of the vibration signal derived from the vibration sensor, the classification value for whether the machine is defective is output. classification unit; and
A decision unit for finally determining whether the machine is defective by applying a voting algorithm to the classification value of whether the plurality of machines are defective,
Each of the plurality of deep neural network models,
It is a neural network model trained in advance using training data including statistical characteristic values of vibration signals generated by the machine and the state of the machine.
The vibration sensor is
a first vibration sensor for detecting horizontal vibrations generated in the machine;
a second vibration sensor for detecting vibration in the vertical direction generated by the machine;
a third vibration sensor for detecting vibration in the depth direction generated by the machine; and
A fourth vibration sensor for detecting vibration in a diagonal direction based on a horizontal direction, a vertical direction and a depth direction generated in the machine,
The deep neural network model is
configured to determine whether the machine is defective based on the maximum value and kurtosis value among statistical characteristic values for vibration signals in the horizontal direction, vertical direction, depth direction and diagonal direction of the frequency domain,
Fault detection device of the machine. delete 11. The method of claim 10,
The preprocessor is
It is configured to evaluate and selectively collect first to fourth vibration signals received from the first to fourth vibration sensors,
The evaluation is performed by comparing a combination of first to third vibration signals collected from the first to third vibration sensors with a fourth vibration signal collected from the fourth vibration sensor to determine the abnormality of the first to fourth vibration signals. This is done by evaluating whether
The selective collection is made when the first to fourth vibration signals are normal based on the evaluation,
Fault detection device of the machine. 11. The method of claim 10,
The preprocessor is
a first pre-processing unit for converting a horizontal vibration signal in the time domain output from the first vibration sensor provided in the machine into a horizontal vibration signal in the frequency domain;
a second preprocessing unit for converting a vertical vibration signal in the time domain output from the second vibration sensor provided in the machine into a vertical vibration signal in the frequency domain;
a third pre-processing unit for converting a time domain depth direction vibration signal output from a third vibration sensor provided in the machine into a frequency domain depth direction vibration signal; and
The vibration signal in the diagonal direction based on the horizontal direction, the vertical direction, and the depth direction of the time domain output from the fourth vibration sensor provided in the machine, the horizontal direction, the vertical direction, and the diagonal direction based on the depth direction of the frequency domain Containing a fourth pre-processing unit for converting into a vibration signal,
Fault detection device of the machine. 11. The method of claim 10,
The calculation unit,
Mean value, median value, minimum (min), maximum (max) value, kurtosis value, skewness value, standard deviation value of the vibration signal of the frequency domain and, configured to calculate a statistical characteristic value comprising one or more of a peak-to-peak value,
Fault detection device of the machine. 15. The method of claim 14,
The calculation unit,
a first calculator for calculating statistical characteristic values for the horizontal vibration signal of the frequency domain;
a second calculation unit for calculating statistical characteristic values for the vertical direction vibration signal of the frequency domain;
a third calculation unit for calculating statistical characteristic values with respect to the vibration signal in the depth direction of the frequency domain; and
A fourth calculation unit for calculating statistical characteristic values for the vibration signal in the diagonal direction based on the horizontal direction, the vertical direction, and the depth direction of the frequency domain,
Fault detection device of the machine. delete 11. The method of claim 10,
The determining unit is
For each of the plurality of deep neural network models implemented with the different algorithms, a weight is calculated based on the accuracy obtained in the pre-training step,
A first representative value for calculating a normal voting result value is assigned to a classification value for whether or not the machine is defective output from the deep neural network model in which the weight is set, and a first value is assigned when the classification value is normal, and the classification value In case of this failure, a second value is assigned,
To the classification value for whether the machine is defective, output from the deep neural network model to which the weight is set, as a second representative value for calculating a failure vote result value, a first value is assigned when the classification value is failure, and the classification If the value is normal, a second value is assigned,
For each of the plurality of deep neural network models, a normal voting result value as a result of multiplying the first representative value by the weight and summing the result by multiplying the weight by the second representative value for each of the plurality of deep neural network models configured to finally determine whether the machine is defective or not as a classification value corresponding to a larger voting result value by comparing the failure voting result value as
Fault detection device of the machine.

Images