A margin factor weighted reconstruction method applied to rotating and static collision fault identification

Abstract

The application designs a margin factor weighted reconstruction method applied to rotating static rub-impact fault identification, and belongs to the technical field of fault diagnosis; firstly, the vibration signal on a rotor tester is acquired to perform wavelet transform, and the margin factor of each detail and approximate signal after wavelet transform is calculated; secondly, the weight of the margin factor of each component signal in the total sum of the margin factors of all component signals is taken as a weight coefficient, the corresponding component signal is subjected to feature enhancement, and the component signal after feature enhancement is acquired; finally, according to the component signal after feature enhancement, the signal is weighted and reconstructed, and based on the frequency spectrum of the autocorrelation function of the signal after weighted reconstruction, the rub-impact fault is subjected to feature extraction and fault identification; the application can effectively suppress noise, further highlight the low-frequency component in the signal, and realize effective enhancement of the rub-impact fault feature and accurate identification of the fault.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fault diagnosis, and particularly relates to a margin factor weighted reconstruction method applied to rotor-stator rub-impact fault identification. BACKGROUND

[0002] With the rapid development of industry, higher requirements are put forward for large rotating machinery including aero-engines, and high thrust-to-weight ratio and high work efficiency become the overall goal of design and manufacture of large rotating machinery. In the process of high-speed operation of the equipment, as the gap between the rotor and the stator gradually decreases, the possibility of friction failure between the rotor and the stator greatly increases. Rub-impact failure greatly affects the safe operation of the rotating system, and even directly affects the safe operation of the rotor system and the entire aero-engine, causing major safety accidents. Therefore, it is of great significance to efficiently and accurately detect rub-impact failure.

[0003] In order to effectively identify the rub-impact failure, various signal analysis methods are applied to feature extraction of rub-impact failure. The vibration signal of the rotor system when rub-impact failure occurs usually shows non-stationary and nonlinear characteristics, and various signal decomposition methods such as wavelet transform, empirical mode decomposition and variational mode decomposition are applied to rub-impact failure diagnosis. Among them, empirical mode decomposition has defects such as mode mixing and end effect; compared with empirical mode decomposition, variational mode decomposition is better in decomposing complex signals. However, the biggest limitation is the existence of boundary effect and burst signal processing problem; the effect of burst signal processing is poor. The key to rub-impact failure diagnosis is to accurately extract the fault characteristic frequency generated by the impact signal.

[0004] In addition, how to realize fault feature analysis according to the component signals obtained after decomposition is also a problem often studied by scholars. The commonly used method is to use signal evaluation index to screen part of the component signals for analysis. The commonly used signal evaluation index is kurtosis. Although kurtosis can well measure the impact characteristics of the vibration signal, this index does not always work well, especially in the case of high noise.

[0005] Moreover, when the rubbing fault occurs, the fault characteristics contained in the vibration signal are very weak, and the noise interference and other factors will greatly increase the difficulty of rubbing fault identification. At present, the methods that can be used to solve this problem are noise reduction processing to reduce the interference of noise signals, or effective enhancement of fault characteristic signals. As a simple nonlinear signal processing method, singular value decomposition has unique advantages in noise reduction and feature extraction, and is widely used in fault diagnosis. However, when the noise is large or the effective order of the determined singular value is inaccurate, the noise reduction effect is usually not ideal, and the extraction of signal fault characteristic information is also difficult. If the singular value order is too low, the fault signal will be lost due to over-truncation; if the order is too high, the noise in the signal will be left due to under-truncation. SUMMARY

[0006] In order to overcome the shortcomings of the prior art, the present application provides a margin factor weighted reconstruction method applied to rotating static rubbing fault identification. First, the margin factor of each detail and approximate signal after wavelet transform is calculated. Second, the weight of each component signal margin factor in the total sum of all component signal margin factors is taken as the weight coefficient, the corresponding component signal is enhanced in feature, and the component signal after feature enhancement is obtained. Finally, the signal is weighted and reconstructed according to the component signal after feature enhancement, and the rubbing fault is extracted and identified based on the frequency spectrum of the autocorrelation function of the weighted and reconstructed signal. Compared with the prior art, the method of weighted reconstruction based on margin factor has the beneficial effect of effectively suppressing noise while more accurately enhancing the low-frequency characteristic information in the rubbing fault signal, which is more conducive to real-time monitoring and fault identification of the equipment operating state.

[0007] A margin factor weighted reconstruction method applied to rotating static rubbing fault identification, specifically comprising the following steps:

[0008] Step 1: Use an acceleration sensor and a data acquisition card to obtain an original discrete vibration signal x(n), where n=1, 2, 3…, N, N is the original length of the discrete signal, and perform discrete wavelet transform on the original discrete vibration signal x(n) to obtain component signals after decomposition of the original signal, wherein the component signals include each detail signal and approximate signal.

[0009] Then x(n) is represented by a set of discrete wavelet bases as follows:

[0010]

[0011] Wherein: j is the order of scale, j0 is an arbitrary starting scale, and j≥j0, is the approximate signal, is the detail signal, is the scale function, is a wavelet function, ω is a shift of position, a wavelet base after discretization is obtained by a scale function and a wavelet function, and an approximate coefficient and a detail coefficient are obtained according to an inner product:

[0012]

[0013]

[0014] A detail signal obtained after a discrete wavelet transform is represented by D1, D2,..., D i , and an approximate signal is represented by A i ; wherein i is a number of wavelet decomposition layers;

[0015] Step 2: Calculate a margin factor value of each detail signal and the approximate signal obtained in step 1, wherein the margin factor of the detail signal D1, D2,..., D i is represented by Mf1, Mf2,..., Mf i , and the margin factor of the approximate signal A i is represented by Mf i+1 ;

[0016]

[0017] Step 3: Based on step 2, calculate a proportion of the margin factor of each component signal in a total sum of margin factors of all component signals, and take the proportion as a weight coefficient Wf k ;

[0018] A specific calculation method is as follows:

[0019]

[0020] Wherein i is a number of discrete wavelet transform decomposition layers;

[0021] Step 4: Perform feature enhancement on each component signal according to the weight coefficient Wf k calculated in step 3; the enhanced detail signal is represented by WD1, WD2,..., WD i , and the enhanced approximate signal is represented by WA i ; then: WD i = Wf i ×D i , WA i = Wf i+1 ×A i , and a reconstructed signal Y is: Y = WD1 + WD2 +..., + WD i + WA i ;

[0022] According to a frequency spectrum of an autocorrelation function of the reconstructed signal Y, a feature extraction and fault recognition of a rotating-to-stationary rubbing fault are performed; specifically, the feature extraction and fault recognition are performed as follows:

[0023] The autocorrelation noise reduction processing is performed on the reconstructed signal Y to obtain an autocorrelation noise reduced signal, and Fourier transform is performed on the autocorrelation noise reduced signal to obtain a spectrum of the noise reduced signal; the prominent characteristic frequency is extracted according to the spectrum, and whether the prominent frequency contains a characteristic frequency of the group of data which can reflect the rub-impact fault and is obtained by theoretical calculation is calculated, including an integer or fractional multiple frequency of a rotation frequency, the rotation frequency = rotation speed / 60, and the characteristic extraction and fault recognition of the rotation static rub-impact fault are performed.

[0024] The present application has the beneficial technical effects:

[0025] The present application considers that the margin factor is a ratio of a signal peak value to a square root amplitude, is sensitive to impact fault characteristics, and is commonly used to detect the wear condition of a mechanical device; a method based on margin factor weighted reconstruction is provided to perform characteristic enhancement on fault characteristic information, and is applied to rotation static rub-impact fault recognition; the method can further highlight low frequency components in the signal while effectively suppressing noise, and can effectively enhance the rub-impact fault characteristics and accurately recognize the fault.

[0026] 1. The noise components in the fault signal can be effectively suppressed.

[0027] 2. The characteristic enhancement of the rub-impact fault can be realized, and the low frequency components in the signal can be further highlighted.

[0028] 3. When the running state of an aero-engine is monitored in real time according to a vibration signal, the low frequency band characteristics of the signal are mainly identified and judged; because the amplification of the acceleration sensor to high frequency signals is much greater than that to low frequency signals, the low frequency component information in the rub-impact fault is more difficult to extract, and the low frequency fault component of the rub-impact can be effectively extracted by the present application.

[0029] 4. The vibration signals collected under different states are verified by experiments, and the accurate extraction of the fault characteristic information can be effectively realized. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The present application has the beneficial technical effects:

[0031] Figure 2 The present application has the beneficial technical effects:

[0032] Figure 3 The present application has the beneficial technical effects: DETAILED DESCRIPTION

[0033] The present application has the beneficial technical effects:

[0034] A margin factor weighted reconstruction method applied to rotating and static collision fault identification, as shown in the accompanying Figure 1 , specifically comprising the following steps:

[0035] Step 1: Use an acceleration sensor and a data acquisition card to obtain an original discrete vibration signal x(n), where n=1, 2, 3…, N, N is the original length of the discrete signal, and perform discrete wavelet transform on the original discrete vibration signal x(n) to obtain a component signal after decomposition of the original signal, wherein the component signal includes each detail signal and an approximate signal.

[0036] 1, 2, 3…, N, N is the original length of the discrete signal, and perform discrete wavelet transform on the original discrete vibration signal x(n) to obtain a component signal after decomposition of the original signal, wherein the component signal includes each detail signal and an approximate signal.

[0037] Then x(n) is represented by a set of discrete wavelet bases as follows:

[0038]

[0039] Wherein j is the order of scale, j0 is an arbitrary starting scale, and j≥j0, is the approximate signal, is the detail signal, is the scale function, is the wavelet function, ω is the offset of the position, and the discrete wavelet base is obtained from the scale function and the wavelet function, and the approximate coefficient and the detail coefficient are obtained according to the inner product:

[0040]

[0041]

[0042] The detail signals obtained after discrete wavelet transform are represented by D1, D2…D i , and the approximate signal is represented by A i ; wherein i is the number of wavelet decomposition.

[0043] Step 2: Calculate the margin factor values of each detail signal and approximate signal obtained in step 1, wherein the margin factors of the detail signals D1, D2…D i are represented by Mf1, Mf2…Mf i , and the margin factor of the approximate signal A i is represented by Mf i+1 .

[0044]

[0045] Step 3: Based on step 2, calculate the proportion of the margin factor of each component signal in the total sum of the margin factors of all component signals, and take it as the weight coefficient Wf k .

[0046] The specific calculation method is as follows:

[0047]

[0048] Where i is the number of discrete wavelet transform decomposition layers;

[0049] Step 4: The weight coefficient Wf calculated in step 3 k , and enhance the characteristics of each component signal; the enhanced detail signals are WD1, WD2…WD i Indicates that the enhanced approximate signal is WA i Indicates that: WD i =Wf i ×D i ,WA i =Wf i+1 ×A i , the reconstructed signal Y is: Y=WD1+WD2+…+WD i +WA i ;

[0050] Based on the spectrum of the autocorrelation function of the reconstructed signal Y, feature extraction and fault identification are performed on the rotor-stationary friction fault; specifically:

[0051] The reconstructed signal Y is subjected to autocorrelation noise reduction processing to obtain the signal after autocorrelation noise reduction, and the signal is subjected to Fourier transform to obtain the spectrum of the noise reduction signal; the prominent characteristic frequencies are extracted based on the spectrum, and the prominent frequencies are calculated to determine whether they contain the characteristic frequencies that can reflect the rubbing fault obtained by theoretical calculation of the set of data, including integer or fractional multiples of the rotational frequency, where the rotational frequency = rotational speed / 60, to perform feature extraction and fault identification on the rotation-static rubbing fault.

[0052] The present invention is further illustrated below by two cases in which the present method is applied.

[0053] Example 1: Randomly select the rubbing fault data collected on the rotor tester with the rubbing position being horizontal and the sensor installation position being horizontal for analysis. The corresponding rotation speed of this state is 1535r / min and the rotation frequency is 25.58Hz. The results are as follows Figure 2 shown. Figure 2 (a) is the time domain diagram of the original signal. Figure 2 (b) Figure 2 (a) Spectrum. Figure 2 (c) shows the detail and approximate signals obtained after discrete wavelet transform of the original signal. The wavelet function used in this case is Sym5, with a decomposition level of 5. In the following example analysis, A5 represents the approximate signal, and D1, D2, D3, D4, and D5 represent the detail signals of layers 1-5, respectively. Figure 2(d) are the values of the margin factors corresponding to each component signal, specifically: A5→1.7795, D5→1.5754, D4→0.4211, D3→0.1451, D2→0.0416, D1→0.0488. Figure 2 (e) are the weight coefficients obtained according to each margin factor, specifically: Wf6=0.4436, Wf5=0.3927, Wf4=0.1050, Wf3=0.0362, Wf2=0.0104, Wf1=0.0122.

[0054] Figure 2 (f) are the signals of the component signals after feature enhancement based on the obtained weight coefficients. Specifically: WA5 is the enhanced approximate signal, WD5, WD4, WD3, WD2, WD1 correspond to the enhanced detail signals of D5, D4, D3, D2, D1 in turn.

[0055] Figure 2 (g), Figure 2 (h) are the reconstructed signals obtained according to the component signals after feature enhancement and the autocorrelation functions of the reconstructed signals. Figure 2 (i1), Figure 2 (i2) are Figure 2 (h) are the corresponding spectra and local amplification of the spectra, which have been normalized.

[0056] Analysis Figure 2 (b), in the spectrum of the original signal, the high frequency component is prominent, but no obvious frequency component related to the rubbing fault is found in the low frequency band.

[0057] Analysis Figure 3 (i1), (i2), it can be found that the method proposed in this paper has the following characteristics:

[0058] 1. The noise component is greatly reduced;

[0059] 2. The high frequency signal is relatively weakened, and the low frequency component is more prominent, which is conducive to real-time monitoring and fault identification of the device state;

[0060] 3. In the spectrum of the signal, the frequency components of 25.6Hz, 51.2Hz (51.2 / 2=25.6Hz), 123.3Hz (123.5 / 5=24.7Hz), These frequencies correspond to 1x, 2x, 5x, According to these frequency components of the multiple and fractional multiple of the rotation frequency, it can be judged that the device has a rubbing fault.

[0061] Example 2: To verify the sensitivity of the present invention to different rubbing positions, data from rubbing positions different from those in Case 1 were selected for analysis. Typical data collected on the rotor tester with the rubbing position being horizontal left (case 1 was horizontal right) were randomly selected for analysis, while the sensor installation position was still horizontal. The corresponding rotation speed in this state was 1539 r / min and the rotation frequency was 25.65 Hz. The results are shown in Figure 2. Figure 3 shown. Figure 3 (a) is the time domain diagram of the original signal, Figure 3 (b) Figure 3 (a) Spectrum. Figure 3 (c) is the detail signal and approximate signal obtained after the original signal is transformed by discrete wavelet transform. Figure 3 (d) is the value of the margin factor corresponding to each component signal, specifically: A5→0.9898, D5→0.4186, D4→0.1858, D3→0.0616, D2→0.0294, D1→0.0809. Figure 3 (e) is the weight coefficient obtained based on each margin factor, specifically: Wf6 = 0.5605, Wf5 = 0.2370, Wf4 = 0.1052, Wf3 = 0.0349, Wf2 = 0.0167, Wf1 = 0.0458.

[0062] Figure 3 (f) shows the component signal after feature enhancement based on the obtained weight coefficients. Specifically, WA5 is the enhanced approximate signal, and WD5, WD4, WD3, WD2, and WD1 correspond to the enhanced detail signals of D5, D4, D3, D2, and D1, respectively.

[0063] Figure 3 (g), Figure 3 (h) is the reconstructed signal and its autocorrelation function obtained based on the component signal after feature enhancement. Figure 3 (i1), Figure 3 (i2) is the spectrum of the autocorrelation function and a local amplification of the spectrum, and the spectrum is normalized.

[0064] analyze ​ (i1), (i2), it can be found that when the rubbing positions are different, the present invention has the following characteristics:

[0065] 1. Noise components are effectively suppressed;

[0066] 2. The high-frequency signal is relatively weakened, and the low-frequency component is further highlighted, which is conducive to real-time monitoring and judgment of faults;

[0067] 3. In the spectrum of the signal, frequency components of 25.6 Hz, 51.2 Hz (51.2 / 2 = 25.6 Hz), 74.9 Hz (74.9 / 3 = 25 Hz), 124.5 Hz (124.5 / 5 = 24.9 Hz) can be found. These frequency components correspond to 1x, 2x, 4x, 5x of the rotating frequency respectively. According to these frequency components, it can be judged that the device has occurred a rub-impact failure.

[0068] From the above analysis, it is found that the present application can still accurately extract the characteristic frequency of the rub-impact failure when the rub-impact position is different. That is, the proposed feature enhancement method based on margin factor weighted reconstruction is not sensitive to the rub-impact position.

Claims

1. A margin factor weighted reconstruction method applied to rolling contact failure identification, characterized in that, Specifically comprising the following steps: Step 1: using acceleration sensor and data acquisition card to obtain original discrete vibration signal x(n), wherein n=1, 2, 3…, N, N is the original length of the discrete signal, and the original discrete signal x(n) is subjected to discrete wavelet transform to obtain component signals after decomposition of the original signal, wherein the component signals include each detail signal and an approximate signal; Step 2: Calculate the margin factor values of each detail signal and the approximation signal obtained in step 1, wherein the margin factor of the detail signals D1, D2...D i is represented by Mf1, Mf2...Mf i , respectively, and the margin factor of the approximation signal A i is represented by Mf i+1 ; Step 3: Based on Step 2, calculate the proportion of the margin factor of each component signal in the total sum of all component signal margin factors, and take it as the weight coefficient Wf k ; Step 4: the weight coefficient Wf is calculated according to step 3 k The characteristic of each component signal is enhanced; the enhanced detail signals are represented as WD1, WD2, …, WD i The enhanced approximate signal is represented as WA i The enhanced detail signals are represented as WD1, WD2, …, WD i = Wf i × D i , WA i = Wf i+1 × A i The reconstructed signal Y is: Y = WD1 + WD2 + … + WD i + WA i The characteristic of each component signal is enhanced; the enhanced detail signals are represented as WD1, WD2, …, WD i The enhanced approximate signal is represented as WA i The enhanced detail signals are represented as WD1, WD2, …, WD i = Wf i × D i , WA i = Wf i+1 × A i The reconstructed signal Y is: Y = WD1 + WD2 + … + WD i + WA i The characteristic of each component signal is enhanced; the enhanced detail signals are represented as WD1, WD2, …, WD i The enhanced approximate signal is represented as WA i The enhanced detail signals are represented as WD1, WD2, …, WD i = Wf i × D i , WA < 2. The method of claim 1, wherein the method is applied to a rub fault identification, and the method is characterized in that, Step 1: the original discrete signal x(n) is represented by a set of discrete wavelet bases as follows: where j is the order of the scale, j0 is an arbitrary starting scale, and j ≥ j0, is the approximation signal, is the detail signal, is the scaling function, is the wavelet function, and ω is the shift in position. The discrete wavelet basis is obtained from the scaling function and the wavelet function, and the approximation and detail coefficients are obtained from the inner product: The detail signals obtained after the discrete wavelet transform are represented by D1, D2, …, D i The approximation signals are represented by A i ; wherein i is the number of wavelet decomposition layers.

3. The method of claim 1, wherein the method is applied to a rub fault identification, and the method is characterized by, Step 2: the margin factor values of each detail signal and the approximate signal are specifically as follows:

4. The method of claim 1, wherein the method is applied to a rub fault identification, and the method is characterized by, Step 3: the specific calculation method is as follows: Wherein, i is the number of decomposition layers of the discrete wavelet transform.

5. The method of claim 4, wherein the method is applied to a rub fault identification, and Step 4: according to the frequency spectrum of the autocorrelation function of the reconstructed signal Y, the characteristic extraction and fault identification of the rotating-static rubbing fault are carried out; specifically as follows: The autocorrelation noise reduction processing is carried out on the reconstructed signal Y to obtain the autocorrelation noise reduction signal, and the Fourier transform is carried out on the autocorrelation noise reduction signal to obtain the frequency spectrum of the noise reduction signal; according to the frequency spectrum, the prominent characteristic frequency is extracted, and whether the prominent frequency contains the characteristic frequency which can reflect the rubbing fault and is obtained by theoretical calculation of the data, including the integer or fractional frequency of the rotating frequency, the rotating frequency = rotating speed / 60, the characteristic extraction and fault identification of the rotating-static rubbing fault are carried out.

Images