A rotating machinery fault diagnosis method based on blind source separation denoising and envelope iterative feature enhancement

Abstract

The present invention discloses a method for diagnosing rotating machinery faults based on blind source separation noise reduction and envelope iteration feature enhancement, including: obtaining vibration signals and speed signals; blind source separation noise reduction on the vibration signals; determining the center frequency and bandwidth of the source vibration signals; designing a bandpass filter, and filtering the source vibration signals using the bandpass filter; performing square envelope demodulation iteration on the filtered source signals, and synchronously performing low-pass filtering to obtain feature enhancement signals; performing angle resampling on the feature enhancement signals according to the speed signals to obtain vibration signals in the angle domain; performing Fourier transformation on the vibration signals in the angle domain to obtain order spectra, and analyzing the order spectra to determine the faults of the bearings. The present invention is suitable for the diagnosis of rotating machinery faults of non-stationary vibration signals under variable working conditions, and can effectively improve the stability and reliability of the diagnosis of rotating machinery faults under variable working conditions as a whole.

Description

Technical Field

[0001] The invention belongs to the field of non-stationary vibration signal processing and rotating machinery fault diagnosis, and specifically relates to a noise reduction method based on PowerICA and an order analysis rotating machinery fault diagnosis method with square envelope iterative enhancement characteristics. Background Art

[0002] Rotating machinery is widely used in power units such as engines, fans, and diesel engines, and is prone to failure due to the harsh operating conditions of the power units (low-speed and heavy-load conditions, overload operation, alternating shocks and loads). Studies have shown that failures caused by rotating parts such as bearings and gears account for more than 40% of the total number of mechanical failures. This data shows that rotating machinery is more prone to failure than other parts. At the same time, due to the complex operating conditions of rotating machinery, its vibration signal has the characteristics of non-stationary. Therefore, conducting research on fault diagnosis of non-stationary vibration signals of rotating machinery can effectively improve the limitations of traditional fault diagnosis that relies on manual experience.

[0003] Vibration signals often accompany faults, and the signals contain rich fault characteristics, so vibration analysis methods are widely used in the fault diagnosis of rotating machinery. However, in the process of vibration signal acquisition, environmental noise will inevitably be mixed in the effective vibration signal, causing the characteristics of the acquired vibration signal to include not only fault characteristics but also invalid noise characteristics. Therefore, it is necessary to preprocess the acquired vibration signal to improve the signal-to-noise ratio and highlight the equipment fault characteristics. In addition, the vibration signal of rotating machinery is complex and unstable under the changes of load, speed and other working conditions. At this time, traditional time domain and frequency domain analysis methods cannot extract and analyze signal features. Therefore, it is urgent to develop a rotating machinery fault diagnosis method based on non-stationary signals to achieve noise reduction, feature enhancement, feature extraction and fault diagnosis of non-stationary vibration signals of rotating machinery under variable working conditions. Summary of the invention

[0004] Purpose of the invention: In order to overcome the deficiencies in the prior art, a rotating machinery fault diagnosis method based on blind source separation denoising and envelope iteration feature enhancement is provided, which is suitable for rotating machinery fault diagnosis of non-stationary vibration signals under variable operating conditions. Blind source separation denoising can improve the signal-to-noise ratio; envelope iteration and low-pass filtering can reduce the interference of environmental noise on fault characteristics; order analysis can effectively extract fault orders, and as a whole can effectively improve the stability and reliability of rotating machinery fault diagnosis under variable operating conditions.

[0005] Technical solution: To achieve the above purpose, the present invention provides a rotating machinery fault diagnosis method based on blind source separation noise reduction and envelope iteration feature enhancement, comprising the following steps:

[0006] S1: Obtaining vibration signals and speed signals of rolling bearings under variable speed conditions;

[0007] S2: Perform blind source separation and noise reduction on the vibration signal to obtain the source vibration signal after noise reduction;

[0008] S3: Perform fast spectral kurtosis analysis on the source vibration signal after noise reduction to determine the center frequency and bandwidth of the source vibration signal;

[0009] S4: designing a bandpass filter according to the center frequency and bandwidth of the source vibration signal, and filtering the source vibration signal using the bandpass filter;

[0010] S5: performing square envelope demodulation iteration on the filtered source signal, and simultaneously performing low-pass filtering to obtain a feature enhanced signal;

[0011] S6: resampling the characteristic enhanced signal according to the rotation speed signal to obtain a vibration signal in the angle domain;

[0012] S7: Perform Fourier transform on the vibration signal in the angle domain to obtain an order spectrum, and analyze the order spectrum to determine the bearing fault.

[0013] Furthermore, the vibration signal and the speed signal are obtained in step S1 as follows:

[0014] Using acceleration sensor to collect vibration signal of rolling bearing 1 (t), x 2 (t), x 3 (t); a photoelectric speed sensor is used to synchronously pick up the rolling bearing speed signal n(t).

[0015] Furthermore, the process of blind source separation and noise reduction in step S2 is as follows:

[0016] A1: Normalize and preprocess the collected vibration signal:

[0017] The data is decentralized by row, and the specific calculation formula is:

[0018]

[0019] A2: Data whitening to decorrelate:

[0020] Find the covariance matrix C of the signal:

[0021]

[0022] Perform eigendecomposition on a signal:

[0023] CV=VΛ (3)

[0024] Where V is the matrix composed of columns of eigenvectors of the mixed signal; Λ is the diagonal matrix composed of eigenvalues;

[0025] Finally, the whitened data X of the mixed signal w for:

[0026]

[0027] A3: Unmixing matrix estimation based on power iteration:

[0028] The general power iteration process can be expressed as:

[0029]

[0030] In the formula For all conventional ICA nonlinearities, the nonlinearity is positive definite. However, this power iteration method is sensitive to finite sample errors. Therefore, the unmixing matrix is ​​estimated by two parallel iterative algorithms. The performance of the two methods is evaluated by non-Gaussianity, and the unmixing matrix with the best performance is finally selected.

[0031] Since conventional nonlinear ICA is positive definite, that is, γ(w)>0, the local maximum value w can be obtained by the following formula: k1 :

[0032]

[0033] To find the local minimum w k2 It is necessary to introduce a constant c, which is specifically expressed as:

[0034]

[0035] A4: Source signal estimation:

[0036] Estimate the source signal based on the mixed signal and the unmixing matrix:

[0037] S(t)=WX(t) (8).

[0038] Furthermore, in step S3, the center frequency and bandwidth of the source vibration signal are determined according to the calculation formula of the fast spectral kurtosis, and the formula is as follows:

[0039]

[0040] Furthermore, the design method of the bandpass filter in step S4 is:

[0041] According to the center frequency and bandwidth of the source vibration signal, the upper and lower cutoff frequencies of the bandpass filter are designed as [f c -B w / 2,f c +B w / 2], the source vibration signal is separated by a bandpass filter, and the separated source vibration signal is filtered.

[0042] Furthermore, the implementation process of the square envelope demodulation iteration and low-pass filtering in step S5 is:

[0043] B1: Calculate the square envelope of the signal:

[0044] The square envelope is shown in the following formula, where the envelope signal is recorded as y(t):

[0045]

[0046] B2: Low-pass filtering to eliminate trend terms:

[0047] The cutoff frequency of the low-pass filter is the minimum rotation frequency. Assuming the trend term of the envelope signal is, then

[0048]

[0049] B3: Similarity calculation of adjacent envelope signals:

[0050] Let H k [x(t)] = y(t), then the cross-correlation coefficient of adjacent envelope signals is:

[0051]

[0052] In the formula, <> is the inner product operator, and H 0 [x(t)] = x(t), the larger the cross-correlation coefficient, the more similar the two adjacent envelope signals are;

[0053] B4: Get the vibration signal related to the fault:

[0054] Compare the signal similarity μ(t) with the preset iteration threshold λ. If μ(t)≤λ, set x(t)=y(t), k=k+1, and go to step B3; otherwise, stop the iteration and output the envelope signal z(t):

[0055]

[0056] B5: Low-pass filtering further enhances the fault characteristics:

[0057] The iterative envelope signal is low-pass filtered again, and the filter cutoff frequency is 4 times the fault frequency.

[0058] Furthermore, the process of acquiring the order spectrum in step S7 is as follows:

[0059] C1: Equal angle marking:

[0060] The velocity curve of the reference axis is drawn through the key phase signal, and it is used as the angle increment reference for obtaining the equal angle sampling time mark;

[0061] C2: Angle domain rearrangement:

[0062] The vibration signals sampled at equal time are rearranged into vibration signals at equal angles by interpolation method;

[0063] C3: Equiangular vibration signal processing:

[0064] Perform fast Fourier transform on the vibration signal in the angle domain to obtain the order spectrum of the vibration signal.

[0065] In the present invention, blind source separation and noise reduction separates the source signal and the noise signal of the collected signal through PowerICA to achieve signal noise reduction; iterative envelope feature enhancement enhances the fault characteristics through envelope demodulation and low-pass filtering; finally, based on order analysis, order and time-order features are extracted from the denoised and feature-enhanced signal to determine and analyze the fault characteristics of the rotating machinery.

[0066] Aiming at the problems of non-stationary characteristics of vibration signals, low signal-to-noise ratio, unclear fault characteristics, etc., the present invention is mainly divided into three parts: PowerICA noise reduction processing, separation source signal envelope iteration, and envelope signal order analysis. Order analysis, envelope demodulation and blind source separation methods provide solutions. Order analysis converts non-stationary time domain signals into stable order domain signals, which are used as the basis for fault diagnosis; the blind source separation algorithm directly distinguishes effective vibration signals from noise signals, while eliminating the process of parameter setting, signal component selection and reconstruction to achieve noise reduction of vibration signals; envelope demodulation demodulates low-frequency signals from high-frequency carrier signals, and combines low-pass filtering to achieve fault feature enhancement. Therefore, the present invention proposes a preprocessing method and feature extraction method for vibration signal enhancement and fault feature enhancement to achieve rotating machinery fault diagnosis of non-stationary vibration signals.

[0067] Beneficial effects: Compared with the prior art, the present invention is suitable for rotating machinery fault diagnosis of non-stationary vibration signals under variable operating conditions, and mainly includes three parts: blind source separation and noise reduction, envelope iteration and low-pass filtering, and order analysis. Blind source separation and noise reduction can improve the signal-to-noise ratio; envelope iteration and low-pass filtering can reduce the interference of environmental noise on fault characteristics; order analysis can effectively extract the fault order, and as a whole can effectively improve the stability and reliability of rotating machinery fault diagnosis under variable operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 is an overall flow chart of the method of the present invention;

[0069] Figure 2 A schematic diagram of a test bench for an example of the present invention;

[0070] Figure 3 The time domain spectrograms of three observation signals are collected for the example of the present invention;

[0071] Figure 4 This is a diagram of the speed signal collected by the photoelectric sensor of the present invention;

[0072] Figure 5 It is the time domain spectrogram of the estimated source signal after PoweICA separation and noise reduction in the example of the present invention;

[0073] Figure 6 It is a fast spectral kurtosis analysis diagram of the source signal after noise reduction in the example of the present invention;

[0074] Figure 7 Specific flow chart for feature enhancement of square envelope iteration and low-pass filtering;

[0075] Figure 8 The time domain spectrogram of the last envelope signal of the source signal separated by the example of the present invention;

[0076] Fig. 9 Specific flow chart for order analysis;

[0077] Fig.10 This is the fault order spectrum of the gear bearing of the present invention;

[0078] Fig.11 This is the fault order spectrum of the gear bearing after wavelet packet denoising and square envelope demodulation. DETAILED DESCRIPTION

[0079] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0080] The present invention provides a rotating machinery fault diagnosis method based on blind source separation noise reduction and envelope iteration feature enhancement, such as Figure 1 As shown, the following steps are included:

[0081] S1: Obtaining vibration signals and speed signals of rolling bearings under variable speed conditions;

[0082] S2: Perform blind source separation and noise reduction on the vibration signal to obtain the source vibration signal after noise reduction;

[0083] S3: Perform fast spectral kurtosis analysis on the source vibration signal after noise reduction to determine the center frequency and bandwidth of the source vibration signal;

[0084] S4: designing a bandpass filter according to the center frequency and bandwidth of the source vibration signal, and filtering the source vibration signal using the bandpass filter;

[0085] S5: performing square envelope demodulation iteration on the filtered source signal, and simultaneously performing low-pass filtering to obtain a feature enhanced signal;

[0086] S6: resampling the characteristic enhanced signal according to the rotation speed signal to obtain a vibration signal in the angle domain;

[0087] S7: Perform Fourier transform on the vibration signal in the angle domain to obtain an order spectrum, and analyze the order spectrum to determine the bearing fault.

[0088] Based on the above scheme, in this embodiment, the above rotating machinery fault diagnosis method is applied as an example, as follows:

[0089] The experimental data of this example comes from the VALENIAN-PT500 rotating machinery test bench. The test bench structure is as follows: Figure 2 As shown. The module of experimental gears 1 to 7 is 25, so the test can be carried out by easily sliding and meshing any gear between the good gear and the faulty gear. The gear ratio of the gearbox is 4.97:1, and it consists of two stages of gears, with the number of teeth of each stage being 58:25 and 60:28 respectively. By sliding the gear along the intermediate shaft, the normal gear on the intermediate shaft can mesh with the gear on the input shaft, and the other gear on the intermediate shaft can also mesh with the gear on the output shaft. The experimental bearing is a conventional model of rolling bearing UC206, and the bearing structure parameters are shown in Table 1.

[0090] Table 1 UC206 rolling bearing structural parameters

[0091]

[0092] In the experiment, the motor speed was adjusted by the frequency converter to make the gearbox run at a variable speed. During the experiment, the fault signals of the gears and bearings were collected by the vibration acceleration sensor, and the speed signals of the input and output shafts were collected by the photoelectric speed sensor. Dewesoft was used to synchronously collect the vibration signal and the speed signal, with a signal sampling frequency of 10000Hz and a sampling time of 3s.

[0093] The simulated fault types used in this embodiment are local crack faults of the inner ring of the rolling bearing and gear wear faults. The faulty bearing is set on the bearing 7, the faulty gear is set on the worn gear 7, the photoelectric speed sensor collects the speed of the gear output shaft, and the vibration acceleration sensor is placed on the bearings 2, 4, and 7. Figure 2 shown.

[0094] The specific diagnostic process is:

[0095] 1) Obtain the vibration signal and speed signal of the rotating machinery under variable speed conditions:

[0096] The three vibration signals x collected in this embodiment 1 (t), x 2 (t), x 3 (t) Figure 3 As shown, the speed signal of the input shaft is a linear rising signal. Figure 4 shown.

[0097] 2) Perform blind source separation on the vibration signal to obtain the source vibration signal after noise reduction:

[0098] First, the three observation data are combined into an observation matrix, namely:

[0099] X=[x 1 (t), x 2 (t), x 3 (t)] (1)

[0100] Next, the observation matrix is ​​normalized and preprocessed according to the formula.

[0101]

[0102] Then, the signal covariance matrix C and eigendecomposition are calculated according to the formula to obtain the whitened data of the observation matrix y = (y 1 ,y 2 ,y 3 ,…,y m );

[0103]

[0104] Finally, according to the iterative method in Table 2, the estimated unmixing matrix W is obtained, and the estimated source signal s(t) is calculated according to the formula. Figure 5 shown.

[0105] Table 2 PowerICA pseudo code

[0106]

[0107] S(t)=WX(t) (4)

[0108] 3) Perform fast spectral kurtosis analysis on the noise reduction source signal s(t) to determine the signal center frequency and bandwidth; and design a bandpass filter for filtering:

[0109] According to the formula, the source signal s(t) is quickly analyzed for its kurtosis spectrum, and its kurtosis is as follows: Figure 6 As shown, it can be observed that the center frequency of the signal is 5468 Hz and the bandwidth is 1562 Hz.

[0110]

[0111] The frequency band of the bandpass filter in this embodiment is [3906,7030].

[0112] 4) Specific envelope iterative feature enhancement methods such as Figure 7 As shown, including:

[0113] 4-1) Calculate the signal square envelope:

[0114] The square envelope is as shown in the formula, and the envelope signal is recorded as y(t);

[0115]

[0116] 4-2) Low-pass filtering to eliminate trend items:

[0117] The cutoff frequency of the low-pass filter is the minimum rotation frequency. Let the trend term of the envelope signal be but

[0118]

[0119] 4-3) Calculation of similarity between adjacent envelope signals:

[0120] Let Γ k [x(t)] = y(t), then the cross-correlation coefficient of adjacent envelope signals is:

[0121]

[0122] In the formula, <> is the inner product operator, and Γ 0 [x(t)] = x(t). The larger the cross-correlation coefficient is, the more similar the two adjacent envelope signals are.

[0123] 4-4) Obtaining fault-related vibration signals:

[0124] Compare the signal similarity μ(t) with the preset iteration threshold λ. If μ(t)≤λ, set x(t)=y(t), k=k+1, and go to 4-3); otherwise, stop the iteration and output the envelope signal z(t):

[0125]

[0126] 4-5) Low-pass filtering strengthens the fault characteristics again. The iterative envelope signal is low-pass filtered again, and the cut-off frequency of the filter is 4 times the fault frequency.

[0127] In this embodiment, the enhanced signal s after the iterative envelope enhace (t) Figure 8 shown.

[0128] 5) The basic process of order analysis is as follows Fig. 9 As shown, the characteristic enhancement signal s is based on the bearing speed signal enhace (t) performing angle resampling to obtain a vibration signal in the angle domain;

[0129] 6) Perform Fourier transform on the vibration signal in the angle domain to obtain the order spectrum, and analyze the order spectrum to determine the bearing fault.

[0130] In this embodiment, the fault order spectrum is as follows: Fig.10 shown. Fig.10 The bearing ring fault order, as well as the single fault order, double fault order, and triple fault order can be clearly found in the figure; at the same time, the meshing order of the gear can also be observed, and it is noted that the sideband near the meshing order can be found to be O m ±5O 4 , which means that a local failure occurred in gear 7.

[0131] In order to verify the effect of the proposed denoising and feature enhancement algorithm, wavelet packet denoising and square envelope demodulation are performed on one observation signal, and finally the order analysis is performed. The order spectrum is as follows: Fig.11 As shown in Figure 2, compared with the analysis results after PowerICA separation and envelope iteration feature enhancement, Fig.11 Only the bearing fault characteristic order and its double order and the gear meshing characteristic order are observed, and other detailed features are covered by noise signals, making it impossible to perform effective fault diagnosis. It can be seen that the vibration signal after PowerICA blind source separation and noise reduction and envelope iteration feature enhancement can effectively diagnose the fault characteristics of rotating machinery such as bearings and gears through order analysis.

Claims

1. A rotating machinery fault diagnosis method based on blind source separation denoising and envelope iteration feature enhancement, characterized in that: The steps include: S1: Obtaining vibration signals and speed signals of rolling bearings under variable speed conditions; S2: Perform blind source separation and noise reduction on the vibration signal to obtain the source vibration signal after noise reduction; S3: Perform fast spectral kurtosis analysis on the source vibration signal after noise reduction to determine the center frequency and bandwidth of the source vibration signal; S4: designing a bandpass filter according to the center frequency and bandwidth of the source vibration signal, and filtering the source vibration signal using the bandpass filter; S5: performing square envelope demodulation iteration on the filtered source signal, and simultaneously performing low-pass filtering to obtain a feature enhanced signal; S6: resampling the characteristic enhanced signal according to the rotation speed signal to obtain a vibration signal in the angle domain; S7: Perform Fourier transform on the vibration signal in the angle domain to obtain an order spectrum, and analyze the order spectrum to determine the bearing fault.

2. A rotating machinery fault diagnosis method based on blind source separation denoising and envelope iteration feature enhancement according to claim 1, characterized in that: The vibration signal and the speed signal are obtained in step S1 as follows: An acceleration sensor is used to collect the vibration signals x1(t), x2(t), and x3(t) of the rolling bearing; a photoelectric speed sensor is used to synchronously pick up the rolling bearing speed signal n(t).

3. The method for fault diagnosis of rotating machinery based on blind source separation denoising and envelope iteration feature enhancement according to claim 1 is characterized in that: The process of blind source separation and noise reduction in step S2 is as follows: A1: Normalize and preprocess the collected vibration signal: The data is decentralized by row, and the specific calculation formula is: A2: Data whitening to decorrelate: Find the covariance matrix C of the signal: Perform eigendecomposition on a signal: CV=VΛ (3) Where V is the matrix composed of columns of eigenvectors of the mixed signal; Λ is the diagonal matrix composed of eigenvalues; Finally, the whitened data X of the mixed signal w for: A3: Unmixing matrix estimation based on power iteration: The general power iteration process can be expressed as: In the formula For all conventional ICA nonlinearities, which are positive definite, the unmixing matrix is ​​estimated by two parallel iterative algorithms, and the performance of the two methods is evaluated by non-Gaussianity, and finally the unmixing matrix with the best performance is selected; Since conventional nonlinear ICA is positive definite, that is, γ(w)>0, the local maximum value w can be obtained by the following formula: k1 : To find the local minimum w k2 It is necessary to introduce a constant c, which is specifically expressed as: A4: Source signal estimation: Estimate the source signal based on the mixed signal and the unmixing matrix: S(t)=WX(t) (8).

4. The method for fault diagnosis of rotating machinery based on blind source separation denoising and envelope iteration feature enhancement according to claim 1, characterized in that: In step S3, the center frequency and bandwidth of the source vibration signal are determined according to the calculation formula of the fast spectral kurtosis, and the formula is as follows:

5. The method for fault diagnosis of rotating machinery based on blind source separation denoising and envelope iteration feature enhancement according to claim 1, characterized in that: The design method of the bandpass filter in step S4 is: According to the center frequency and bandwidth of the source vibration signal, the upper and lower cutoff frequencies of the bandpass filter are designed as [f c -B w / 2,f c +B w / 2].

6. The method for fault diagnosis of rotating machinery based on blind source separation denoising and envelope iteration feature enhancement according to claim 1, characterized in that: The implementation process of the square envelope demodulation iteration and low-pass filtering in step S5 is: B1: Calculate the square envelope of the signal: The square envelope is shown in the following formula, where the envelope signal is recorded as y(t): B2: Low-pass filtering to eliminate trend terms: The cutoff frequency of the low-pass filter is the minimum rotation frequency. Assuming the trend term of the envelope signal is, then B3: Similarity calculation of adjacent envelope signals: Let H k [x(t)] = y(t), then the cross-correlation coefficient of adjacent envelope signals is: Where <> is the inner product operator, and H0[x(t)]=x(t); B4: Get the vibration signal related to the fault: Compare the signal similarity μ(t) with the preset iteration threshold λ. If μ(t)≤λ, set x(t)=y(t), k=k+1, and go to step B3; otherwise, stop the iteration and output the envelope signal z(t): B5: Low-pass filtering further enhances the fault characteristics: The iterative envelope signal is low-pass filtered again, and the filter cutoff frequency is 4 times the fault frequency.

7. The method for fault diagnosis of rotating machinery based on blind source separation denoising and envelope iteration feature enhancement according to claim 1, characterized in that: The method for obtaining the vibration signal in the angle domain in step S6 is: C1: Equal angle marking: The velocity curve of the reference axis is drawn through the key phase signal, and it is used as the angle increment reference for obtaining the equal angle sampling time mark; C2: Angle domain rearrangement: The vibration signals sampled at equal times are rearranged into vibration signals of equal angles by interpolation method.

Images