KR101129466B1 - Method for condition monitoring of rotating machines via wavelet thresholding function and 4th-order moment - Google Patents

Abstract

PURPOSE: A condition diagnosing method of a rotating machine using a fourth moment and the wavelet thresholding function is provided to detect the initial defect of the rotating machine using a property signal of the condition of the machine. CONSTITUTION: A condition diagnosing method of a rotating machine using a fourth moment and the wavelet thresholding function comprises the following steps: measuring a vibration signal using a sensor on the rotating machine(10); removing signals from the vibration signal irrelevant to the condition of the machine using a property signal extractor; converting the vibration signal into a wavelet coefficient in the time-scale domain(20); calculating a threshold value from the wavelet coefficient(30); selecting the wavelet coefficient for a property signal for each scale using the threshold value and the threshold value function(40); converting the obtained wavelet property coefficient into a property signal for the time domain using a reverse wavelet conversion(50); and testing the condition of the machine using the property signal for the time domain(60).

Description

Method for condition monitoring of rotating machines via wavelet thresholding function and 4th-order moment}

The present invention utilizes a threshold value and a thresholding technique obtained from wavelet transform and fourth-order moment from a vibration signal measured by a sensor, and a characteristic signal associated with a state of a machine. And a signal processing method for diagnosing the state of the rotating machine, especially the presence of an initial defect.

Many methods for diagnosing the state of a rotating machine measure vibration signals through sensors and analyze these signals.

In addition, one of the best ways to diagnose the condition of rotating machinery is to detect early stage defects and develop a maintenance and repair plan. However, the vibration signal (characteristic signal or fault feature) generated by the initial defect of the rotating machine is weak and is easily buried by the vibration signal (or noise component) generated by other rotating machines or the like.

1 is a flowchart of a diagnosis method according to a conventional method.

Vibration signal measuring step 100 for measuring the signal in the vibration signal measuring unit, characteristic signal extraction step 110 for extracting the characteristic signal from the signal in the characteristic signal extraction unit, a state of diagnosing the state of the rotating machine in the state diagnosis unit The diagnosis step 120 is made.

In the vibration signal measuring step 100, the vibration signal of the time domain is measured through a sensor attached to the surface of the rotating machine.

The vibration signal of the time domain measured in the vibration signal measuring step 100 is composed of a sum of a characteristic signal associated with the state of the machine and a signal independent of the state of the machine, as shown in Equation 1, Since the characteristic signal does not appear easily due to noise, it is impossible to diagnose the state of the machine in this state.

One of the methods for extracting such characteristic signals is a characteristic signal extraction method by wavelet transform and thresholding. In this method, as shown in Equation 1, the vibration signal measured by the sensor is a state of the machine, in particular, an initial defect. It is based on the assumption that it is composed of a characteristic signal associated with the noise and noise independent of the state of the machine. That is, only the characteristic signal associated with the state of the machine is extracted from the vibration signal to determine whether the machine is defective.

Here, x (t) is a vibration signal measured by the sensor, s (t) is a characteristic signal associated with the state of the machine, and n (t) is a signal independent of the state of the machine, such as noise.

The left side of Equation 1 is information known as a signal measured by the sensor, and the right side is information unknown.

Universal thresholds used for feature signal extraction are described in “Donoho, D. (1995)“ De-noising by soft-thresholding ”, IEEE Transactions on Information Theory, 41 (3) pp.613-627” ( Equation 2 defined in Ref. 1) is as follows.

은 노이즈 신호 n(t) 의 표준편차, N 은 사용된 진동신호의 데이터 길이를 나타낸다.
Where T _univ is the Universal threshold,

Is the standard deviation of the noise signal n (t), and N is the data length of the vibration signal used.

As the thresholding function, hard thresholding and soft thresholding are typically used, which are the same as Equation 3 and Equation 4, which are shown in FIG. 2 [Ref. 1].

As shown in Fig. 2, the thresholding function causes an extreme change around the threshold value. Such a change causes a problem that noise components are incorrectly judged as characteristic signals or characteristic signals are judged as noise around the threshold when extracting the characteristic signals by applying hard thresholding. In addition, soft thresholding may cause difficulty in analyzing the extracted signal due to a sharp decrease in the output value.

로 변환하고, 여러 스케일에서 기계의 상태와 연관된 특성신호를 추출한다. 웨이블렛 변환은 CWT (Continuous Wavelet Transform), DWT (Discrete Wavelet Transform) 등과 Morlet wavelet, Meyer wavelet 등의 웨이블렛을 모 웨이블렛 (Mother wavelet)로 사용할 수 있다.
In the characteristic signal extraction step 120, wavelet coefficients of the time-scale domain are subjected to wavelet transform of the vibration signal of the time domain.

And extract characteristic signals associated with the state of the machine at different scales. The wavelet transform may use a wavelet such as a continuous wavelet transform (CWT), a discrete wavelet transform (DWT), a morlet wavelet, a meyer wavelet, etc. as a mother wavelet.

The prior art as described above has the following disadvantages in the process of calculating the threshold and in the process of defining the threshold function.

The universal threshold is a value calculated from the standard deviation of the noise. Since there is no information on the noise component, the universal threshold includes a large error in estimating the standard deviation of the noise. As a result, a large amount of errors are included in the step of extracting the characteristic signal, and the reliability of the diagnosis may be degraded when the rotating machine is diagnosed using the extracted characteristic signal. Therefore, it is necessary to reduce the uncertainty (for example, an error generated in estimating the standard deviation of noise) existing in the calculation of the threshold value in order to increase the reliability in the step of extracting the characteristic signal.

Since it is difficult to accurately calculate the threshold value that distinguishes between the characteristic signal and the noise, the following problem occurs when the threshold function is changed extremely around the threshold value.

The noise component may be misjudged as the characteristic signal around the threshold value, or the characteristic signal may be misjudged as noise. This decreases the reliability of the extracted characteristic signal, and as a result, may cause difficulty in diagnosing the initial defect of the rotating machine.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and extracts a characteristic signal or fault feature related to the state of the machine from a vibration signal measured through a sensor to produce a machine component, such as a gear or a bearing. The aim is to diagnose early defects early.

In addition, the present invention is to determine the parameters used to extract the characteristic signal, for example, the threshold value is automatically determined from the measured vibration signal, by defining a threshold function that changes gently around the threshold value An object of the present invention is to provide a characteristic signal extraction method that is easy to cope with a vibration signal having various characteristics.

In addition, an object of the present invention is to provide a method for diagnosing an initial defect of a mechanical component even without prior information on the specifications of the mechanical component, so that the present invention can be easily applied to the diagnosis of the mechanical components of various fields.

The present invention provides a vibration signal measuring step of measuring a vibration signal from the rotating machine to the sensor in the vibration signal measuring unit to solve the above problems; In the characteristic signal extraction unit, 1) the vibration signal in the time domain is transformed into the wavelet coefficient in the time-scale domain in order to remove the signal unrelated to the state of the machine from the vibration signal measured by the sensor and to extract only the characteristic signal related to the state of the machine. 2) a threshold value calculation step of calculating a threshold value from the transformed wavelet coefficients; and 3) a characteristic for each wavelet scale by applying the calculated threshold value and a thresholding function. A characteristic coefficient extraction step of selecting only wavelet coefficients corresponding to the signal; and 4) an inverse wavelet transformation step of converting the extracted wavelet characteristic coefficients in the time-scale domain into a characteristic signal in the time domain through inverse wavelet transformation; Diagnosing the state of the machine from the characteristic signal of the region; And it employs a method of condition monitoring of rotating machinery using the method quaternary moment.

In addition, the present invention may include diagnosing a state of the machine, including calculating a fourth moment from the extracted characteristic signal to determine whether an initial defect exists in a rotating machine, and generating an alarm if a defect exists. It adopts the means of state diagnosis method of rotating machine using wavelet thresholding function and 4th moment.

In addition, the present invention is a method for diagnosing the state of the rotating machine using the wavelet thresholding function and the fourth moment, the step of determining whether the initial defect is present if the fourth moment exceeds 3 Adopt the means of

According to the present invention, since only the fault feature related to the state of the machine is extracted from the vibration signal and used to monitor the state of the machine component, it is possible to diagnose the initial fault sensitive to ambient noise, thereby maintaining the equipment. It allows for a more relaxed response when planning.

In addition, according to the present invention, since advance information such as machine specification is unnecessary and only the vibration signal measured by the sensor is used, it is possible to provide improved reliability in diagnosing the state of the machine. It can greatly improve.

In addition, the present invention can compensate for the misjudgment problem, which is a disadvantage of the conventional hard / soft thresholding function, by adjusting the change rate of the thresholding function around the threshold.

In addition, the present invention uses the fourth order (Kurtosis) to adjust the rate of change of the thresholding function around the threshold, the rate of change can be automatically adjusted according to the characteristics of the data.

In addition, since the present invention does not require user input in determining a parameter of the thresholding function, an error that may occur in the process of the user's intervention may be blocked.

1 is a schematic diagram of a state diagnosis technique of a rotating machine using a conventional wavelet threshold. FIG. 2 is a hard threshold and soft threshold function used in the related art.
3 is a flow chart showing the processing procedure of the state diagnosis technique of the rotating machine using wavelet thresholding and the fourth moment of the present invention.
4 is a thresholding function proposed in the present invention.
5 is a flowchart for diagnosing the state of the rotating machine using the extracted characteristic signal.
6 is an example (simulation) applied to vibration signals in a defect state and a steady state.
7 is an embodiment of an apparatus for realizing the method of the present invention.

Hereinafter, the present invention will be described in detail with reference to the drawings.

3 is a flow chart of the diagnostic method of the present invention, Figure 4 is a thresholding function proposed in the present invention, Figure 5 is a flow chart for the state diagnosis presented in the present invention, Figure 6 is an application example of the present invention Indicates.

The present invention

1) a vibration signal measuring step (10) of measuring the vibration signal from the rotating machine to the sensor in the vibration signal measuring unit;

2) At the characteristic signal extraction unit

A wavelet transform step 20 of converting the vibration signal of the time domain measured by the sensor into a wavelet coefficient of the time-scale domain;

A threshold value calculation step (30) of calculating a threshold value from the transformed wavelet coefficients of the time-scale region;

A feature coefficient extraction step (40) of extracting a wavelet characteristic coefficient from the wavelet coefficients of the time-scale region by applying the calculated threshold value and a thresholding function;

An inverse wavelet transforming step (50) for converting the extracted wavelet characteristic coefficients into characteristic signals in a time domain;

And 3) a step (60) of diagnosing the state of the machine from the characteristic signal in the time domain at the state diagnosis section.

Each step will be described in detail as follows.

Vibration signal measuring step (10)

The vibration signal measuring unit of the vibration signal measuring step 10 includes a sensor attached to the outside of the rotating machine, for example, an accelerometer and a data collection device for receiving the vibration signal measured by the sensor. Measure and collect.

Wavelet Transform Step (20)

The wavelet transform step 20 converts a vibration signal measured in the time domain into a time-scale domain by applying a continuous wavelet transform (CWT), a discrete wavelet transform (DWT), or a similar transform method.

Threshold value calculation step (30)

The threshold value calculating step 30 of calculating a threshold value from the transformed wavelet coefficients utilizes statistical characteristics of the signal from the wavelet coefficients converted from the measured vibration signal in order to minimize uncertainty. Calculate the threshold value.

Utilizing the statistical characteristics of the signal means that the fourth order moment (Kurtosis), which is a statistical value of the wavelet coefficient, is used in determining the threshold value, as shown in Equation 5 below. In this process, it is one of the features of the present invention to eliminate the process of estimating the standard deviation of the error, that is, the noise, which can occur in Equation 2 by using a fourth order moment calculated from wavelet coefficients.

로부터 스레스홀드 값을 구한다. 스레스홀드 값은 수학식 5 에 정의한 바와 같이 구한다.
Wavelet Coefficients from Time-Scale Domains Obtained by Wavelet Transform

The threshold value is obtained from. The threshold value is obtained as defined in equation (5).

는 스케일

에서의 스레스홀드 값이며,

는 스케일

에서의 웨이블렛 계수,

는 웨이블렛 계수

로 부터 구한 4차 모멘트이다.
here,

Scale

The threshold value at

Scale

Wavelet coefficient at,

The wavelet coefficient

This is the fourth order moment from.

The threshold value is defined using the fourth order moment for the following reasons. In general, characteristic signals associated with the initial failure of rotating machines tend to have periodic impulses that have been shocked and disappeared for a very short time. Since the impulse generated by the initial defect is very weak and easily distorted by noise, it is generally difficult to observe the impulse in the vibration signal of the time domain measured by the sensor. This impulse is well reflected in the fourth order (Kurtosis), the transient impulsiveness associated with the initial failure of the machine will show a high fourth order moment. In the case of the machine operating in the normal state, the fourth moment value of about 3.0 is shown, and if a defect occurs in the machine, the impulse appears in the vibration signal and the fourth moment value is larger than 3.0. Therefore, when the threshold value is defined using the characteristics of the fourth order moment, an error generated in estimating the standard deviation of the noise signal used when calculating the universal threshold, as shown in Equation 2, is described. Can be removed.

Characteristic coefficient extraction step (40)

In the characteristic coefficient extraction step 40 of extracting the wavelet characteristic coefficient, the threshold function reflects the statistical characteristics of the vibration signal (wavelet coefficient) measured to correspond to the characteristics of the various vibration signals, and the threshold value. Define to show a gentle change around.

Here, it is defined to show a gentle change as follows.

As shown in FIG. 4, it can be seen that the thresholding function differs according to the fourth-order moment value calculated from the wavelet coefficients. This means that the output value of the thresholding function is different for the same input value. For example, K = 2.0 and K = 6.0 will be described.

In Figure 4, the horizontal axis is the input and the vertical axis is the output of the thresholding function.

The output value when K = 2.0 is smaller than the output value when K = 6.0. Here, the input value smaller than the threshold value (square marker) has a zero output value. When the output value of the threshold function for the larger input value and the input value smaller than the threshold value are compared with respect to different K values, when the value of K = 6.0, the change in the output value is set to K = 2.0 before and after the threshold. Greater than the case.

This change is automatically adjusted according to the calculated K value as defined in Equation 6.

Through this process, the aforementioned misjudgment can be reduced.

The feature coefficient extraction step 40 is performed by the threshold value defined using the fourth order moment and the threshold function shown in Equation 6 and FIG.

와

를 사용했는데, 이는 센서가 측정한 진동신호의 특성을 반영하여, 기계의 상태와 무관한 신호를 제거하고 기계의 상태와 연관된 특성신호를 추출하는데 보다 효율을 높이기 위해서이다. In the function defined in Equation 6 above, the exponent

Wow

This is to reflect the characteristics of the vibration signal measured by the sensor, in order to remove the signal unrelated to the state of the machine and to increase the efficiency of extracting the characteristic signal associated with the state of the machine.

As shown in FIG. 4, this allows the rate of change of the thresholding function to be automatically adjusted around the threshold value according to the fourth-order moment in which the characteristics of the vibration signal are reflected. It can reduce the likelihood.

When the distinction between the impulse and the noise, which is the characteristic signal, is clear, that is, when the fourth-order moment value is large, a sudden change is displayed around the threshold value so that the impulse and noise can be easily distinguished. In other words, as the fourth moment is larger, the slope increases in the threshold value. On the contrary, when the distinction between impulse and noise is unclear, that is, when the fourth-order moment value is small, a gentle change (or large attenuation) appears around the threshold value to remove the influence of the noise.

The result obtained by Equation 6 is obtained by approximating the s' (t) of the characteristic signal corresponding to the time domain through the inverse wavelet transform with wavelet coefficients corresponding to the characteristic signal (the characteristic signal reflecting the initial defect) associated with the state of the machine. You can get it. Here, since the exact characteristic signal s (t) is unknown, it is represented by the approximation value s' (t) of the characteristic signal s (t) due to the defect of the machine.

Inverse wavelet transform step (50)

The inverse wavelet transform step 50 converts the extracted wavelet characteristic coefficients of the time-scale domain into characteristic signals associated with the machine state of the time domain.

Diagnostic step (60)

After that, the status diagnosis unit proceeds to the diagnosis of the state of the machine.

The diagnostic step 60 is described in the flowchart of FIG.

The approximate value s' (t) of the extracted time-domain characteristic signal follows the process as shown in FIG. 5 to diagnose the state of the machine. If an impulse component that is periodically repeated exists in the extracted characteristic signal, it indicates that a defect has occurred in the rotating machine to be measured.

Such defects can be diagnosed according to this value by obtaining the fourth order moment (Kurtosis, K) from the extracted characteristic signal s' (t).

의 4차 모멘트 (Kurtosis)는 다음 수학식 7 과 같이 구할 수 있다.Signal of interest

The fourth moment of (Kurtosis) can be obtained as shown in Equation 7.

 When an initial defect occurs in a component of a rotating machine such as a gear or a bearing, an impulse that is periodically repeated appears in the vibration signal. This impulse is reflected in the fourth order moment, a statistical characteristic value.

In the present invention, when the fourth moment calculated from the extracted characteristic signal has a value greater than 3.0, it is determined that a defect exists in the machine. When the fourth moment obtained from the extracted characteristic signal is 3.0 or less, an impulse component is added to the characteristic signal. It does not exist, and it is determined that there is no initial defect that generates such an impulse.

Therefore, the diagnosis is made as follows according to the value of the fourth order moment K obtained from the extracted signal.

1) If K> 3.0, there is an initial defect on the machine.

2) If K ≤ 3.0, the machine is in normal condition.

According to the result of diagnosis of the state of the machine through the 4th moment, if an initial fault exists, an alarm is triggered so that the maintenance and inspection of the machine component in which the fault exists can proceed.

Figure 7 shows one embodiment of an apparatus for realizing the method of the present invention.

The vibration signal measuring unit includes a sensor unit including a plurality of vibration sensors, and as a characteristic signal extracting unit and a state diagnosis unit, a sensor interface for filtering and sampling a signal, an analog filter, a data sampler, and a digital filter is provided. After calculating and processing the signal in the processing unit, if the result is displayed on the screen in the display unit or if the value of the fourth moment obtained from the extracted signal is abnormal, an alarm will sound in the alarm generating unit.

Example

In order to verify the performance of the present invention, vibration signals in 1) steady state and 2) defective state were generated to extract characteristic signals related to defects. The used signal was generated to reflect vibration characteristics through simulation.

FIG. 6A shows a characteristic signal (impulse) due to a defect excluding noise, and FIG. 6B shows a vibration signal measured through a sensor from a defective rotating machine.

In general, the vibration signal of a rotary machine having an initial defect is a vibration generated by another rotating machine and its components, and the impulse associated with the defect is buried in the surrounding noise, as shown in FIG. 6B.

FIG. 6D is a vibration signal in a steady state, and when compared with the defective vibration signal shown in FIG. 6B, its characteristics are difficult to distinguish easily.

6C and 6E show the characteristic signals extracted by applying the characteristic signal extraction method according to the present invention. In the case of a defect, the extracted characteristic signal clearly shows an impulse periodically repeated (Fig. 6C). Therefore, it can be judged that a defect exists.

In the steady state, the extracted characteristic signal cannot observe the repeated impulse as shown in FIG. 6E. Therefore, the rotating machine to be measured can be determined to be in a steady state without a defect.

10: vibration signal measurement step 20: wavelet conversion step
30: threshold value calculation step 40: feature coefficient extraction step
50: inverse wavelet transform step 60: diagnostic step

Claims (3)

In the vibration signal measuring unit
Vibration signal measuring step of measuring a vibration signal with a sensor in a rotating machine (10);
From the characteristic signal extraction unit
1) A wavelet transform step of removing a signal unrelated to the state of the machine from the vibration signal measured by the sensor, and converting the vibration signal in the time domain into the wavelet coefficient in the time-scale domain to extract only the characteristic signal associated with the state of the machine. 20; and
2) a threshold value calculation step 30 of calculating a threshold value from the transformed wavelet coefficients; and
3) a characteristic coefficient extraction step 40 of selecting only wavelet coefficients corresponding to the characteristic signals for each scale by applying the calculated threshold value and threshold function; and
4) an inverse wavelet transform step 50 of converting the extracted wavelet characteristic coefficients of the time-scale region into a characteristic signal of the time domain through inverse wavelet transformation; Wow
At the status diagnosis part
Diagnosing a state of the machine from the characteristic signal in the time domain (60);
A method for diagnosing a state of a rotating machine using wavelet thresholding functions and fourth moments. The method of claim 1,
Diagnosing the state of the machine (60) is to determine whether there is an initial defect in the rotating machine by calculating the fourth moment (Kurtosis) from the extracted characteristic signal and
Alarming if a fault exists.
A method for diagnosing the state of rotating machinery using wavelet thresholding function and fourth order moment. The method of claim 2,
The step of determining whether the initial defect is present is a step of determining that the defect exists when the fourth moment exceeds 3
A method for diagnosing the state of rotating machinery using wavelet thresholding function and fourth order moment.

Images