Bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition

Through the multi-dimensional synchronous optimal decomposition method, the correlation kurtitude and time-frequency domain correlation coefficients are used to solve the problem of fault feature extraction in multi-channel signal processing, and efficient diagnosis of bearing failures is achieved, reducing the alarm leakage rate and improving Decomposition efficiency.

CN119064011APending Publication Date: 2024-12-03BEIHANG UNIV
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411173506.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The prior art is difficult to effectively process multi-channel signals in bearing fault diagnosis, resulting in insufficient fault information to cope with noise and random interference, which may lead to missed reports. The traditional decomposition method fails to make full use of multi-channel information, limiting fault feature extraction and diagnosis. Application.

Method used

Adoptional filtering and extraction of multi-channel signals is achieved by using multi-dimensional synchronous optimal decomposition method, using correlation kurtiness as the decomposition target, quickly eliminate redundant modes by using envelope spectrum kurtiness and time-frequency domain correlation coefficients to extract multiple faults feature.

Benefits of technology

It realizes targeted decomposition of periodic fault shock components in complex signals, reduces the alarm leakage rate caused by randomness of a single channel, makes full use of multi-channel information, and improves the decomposition efficiency and intelligent identification of fault characteristics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119064011A_ABST
    Figure CN119064011A_ABST
Patent Text Reader

Abstract

The invention provides a bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition, and relates to the technical field of fault diagnosis, and the method comprises the steps: carrying out the preprocessing of vibration signals collected by a plurality of sensors; then obtaining customized division of each channel signal analysis frequency band by means of a sequence statistics filter, and initializing a mode decomposition filter by adopting a Hanning window filter bank; a filter is used for filtering, the related kurtosis based on a specific input period set is used as a decomposition target constraint solution, and the filter is iteratively updated in sequence by solving a maximum generalized eigenvalue problem; after iteration of a set number of times, a specified number of dominant decomposition modes are reserved according to a related kurtosis value; calculating correlation coefficients between every two residual decomposition modes, selecting two decomposition modes of which the time domain correlation coefficients and the frequency domain correlation coefficients are greater than 0.8, and abandoning mode components of which the envelope spectrum kurtosis values are smaller; and finally, constructing an envelope spectrum based on the residual modes and carrying out fault diagnosis. According to the method, fault signal periodicity and impact strength evaluation index correlation kurtosis are taken as decomposition targets, adaptive filtering and extraction of a multi-channel signal decomposition mode are realized, multi-channel information is fully utilized, and multi-fault features are effectively identified.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition. Background Art

[0002] Rolling bearings are core components of rotating machinery. Malfunctions in these bearings often lead to performance degradation and even failure. Therefore, rolling bearing fault diagnosis is crucial for ensuring the safe operation of mechanical equipment. However, test signals are not only subject to interference from non-faulty components and the environment, but are also severely attenuated by lengthy transmission paths. Therefore, the key to bearing fault diagnosis lies in extracting the periodic impulses associated with the fault from complex test signals.

[0003] Deconvolution methods are widely used because they can effectively resist the influence of transmission paths. The core of deconvolution is to design an adaptive filter and achieve accurate matching of the optimal filter with potential fault information through iterative or non-iterative methods. In 1978, Wiggins first proposed minimum entropy deconvolution (MED) to recover the characteristics of seismic pulse signals from complex signals. Subsequently, McDonald et al. proposed maximum correlated kurtosis deconvolution (MCKD) in 2012, which takes into account the periodicity of the signal. In 2018, Buzzoni et al. proposed maximum second-order cyclostationarity blind deconvolution (CYCBD), which takes into account the cyclostationary properties of rotating machinery faults. Although the above methods are simple and effective in most cases, they mainly rely on single-channel information for fault feature extraction. The fault information is not sufficient to cope with noise and random interference, which may lead to missed detections.

[0004] In contrast, decomposition methods have more advantages in multi-component separation and extraction. Methods such as empirical mode decomposition (EMD), empirical wavelet transform (EWT), and variational mode decomposition (VMD) have been proposed one after another, but these traditional decomposition methods are not fully applicable to bearing fault diagnosis and do not consider typical mechanical fault characteristics. In order to overcome the limitations of traditional decomposition methods, Miao et al. proposed fault feature-based mode decomposition (FMD), which uses correlation kurtosis as the objective function and has high robustness. However, the number of output modes of FMD still needs to be manually set.

[0005] With the development of sensor technology, multi-channel information input provides a solution for fault feature information mining. Methods such as EMD, EWT, and VMD have been extended to multivariate versions, but have not gotten rid of the shortcomings of the original decomposition methods. Considering the excellent characteristics of FMD, Yan et al. attempted to extend FMD to multivariate feature pattern decomposition. By performing FMD on a single channel, multiple components were obtained as input to the subsequent convolutional neural network to learn discriminant features. The above multivariate decomposition methods produce numerous decomposition patterns, requiring further manual observation to identify faults. In addition, a large number of strategies for multi-channel signal processing based on data fusion have been studied, which have shown better performance but rely to a certain extent on a large amount of labeled data. These shortcomings limit the application and promotion of traditional decomposition methods in bearing multi-channel signal processing and multi-dimensional fault feature extraction and diagnosis. Summary of the Invention

[0006] The purpose of the present invention is to provide a bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition, taking the fault signal periodicity and impact strength evaluation index correlation kurtosis as the decomposition target, realizing adaptive filtering and extraction of multi-channel signal decomposition mode, thereby making full use of multi-channel information and effectively identifying multiple fault characteristics.

[0007] A bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition, comprising:

[0008] S1, obtain the multi-channel vibration signal X of the bearing to be tested raw =[x1 x2 … x O ], where each channel signal is x o =[x o (1) x o (2) … x o (N)] T,o∈O, O is the number of channels, N is the length of the signal, and the superscript T is the transpose; the vibration signal is preprocessed, including high-frequency sampling, truncation and de-averaging, to obtain a preprocessed signal;

[0009] S2, set the filter length to L, the sliding window length to W, the number of pre-iterations to I, the number of target fault categories to Q, the total number of patterns to K, and the theoretical fault cycle to T Q ;

[0010] S3, calculate the O channel signals to obtain O upper envelope sets, where W is the sliding window length; take the minimum value point of each group of upper envelope sets as the spectrum segmentation set to obtain the initial filter cutoff frequency set of each channel; construct O groups of channel filter banks based on the Hanning window, and the number of filters in each group is K o ,o∈O.

[0011] The calculation formula of the envelope set on each channel is as follows:

[0012]

[0013] Where: UpEnv o (n) represents the oth channel signal x o The calculated parameter value of the nth length point in the upper envelope set, Y o Represents x o The Fourier spectrum of Y o =[Y o (1) Y o (2) … Y o (N)] T , where Y o (n) represents Y o The parameter value at the nth length point, fft is the Fourier function;

[0014] The calculation formula of the cutoff frequency set of each channel is as follows:

[0015]

[0016] Where: [f lower ,f upper ] o Indicates the lower cutoff frequency and upper cutoff frequency of the oth group of initial filters, FreBound o represents the minimum value point set of the envelope set on the oth channel signal, B o is the number of minimum points of the envelope set on the o-th channel signal, FreBound o (b) represents the parameter value of the bth length point position of the minimum value point set of the envelope set on the oth channel signal;

[0017] S4, let i = 1, j = 1, i∈I, j∈Q, I is the number of pre-iterations, Q is the number of fault categories;

[0018] S5, judge i, if i≤I, execute S6, if i>I, execute S8;

[0019] S6, calculating the correlation kurtosis based on Q theoretical fault cycles, filtering the signals of each channel, and selecting, for each filter, a decomposition pattern corresponding to the maximum eigenvalue in the generalized Rayleigh entropy equation as the final decomposition pattern;

[0020] The relevant kurtosis calculation formula is as follows:

[0021]

[0022] Where: u o,k =[u o,k (1)u o,k (2)…u o,k (N)] T represents the kth decomposition mode of the oth channel, o∈O,k∈K o ,u o,k (n) represents the parameter value of the nth length point position in the kth decomposition pattern of the oth channel, M is the shift order, f o,k is the vector form of the kth filter of the oth channel, f o,k =[f o,k (1)f o,k (2)…f o,k (L)] T , f o,k (l) represents the parameter value of the lth length point position in the kth filter of the oth channel, T j is the theoretical failure period corresponding to the jth type of fault, j∈Q;

[0023]

[0024] Where: R XWX and R XX are weighted correlation matrices, x o is the oth channel signal, and the superscript H is the conjugate transpose;

[0026] The generalized Rayleigh entropy equation is as follows:

[0027] f o,k R XWX =λf o,k R XX ;

[0028] Where: λ is the eigenvalue;

[0029] S7, adjust the coefficients of all the filters, set i=i+1 and return to S5;

[0030] S8, judge j, if j≤Q, execute S9, if j>Q, execute S11;

[0031] S9, for each channel K o The final decomposition patterns are calculated, and the relevant kurtosis of the final decomposition patterns is obtained. o Calculate the number of retained P, and select the final decomposition mode corresponding to the first P maximum values ​​of the relevant kurtosis as the optimal decomposition mode;

[0032] The calculation formula for the reserved quantity is as follows:

[0033] P=floor(ΣK o / Q·O);

[0034] Where: floor is the rounding down operation, K o is the number of filters for each channel, Q is the number of fault categories, and O is the total number of channels;

[0035] S10, set j = j + 1 and return to S8;

[0036] S11, calculating the time domain correlation coefficient and the frequency domain correlation coefficient between each of the K optimal decomposition patterns, and selecting all optimal decomposition pattern combinations where both coefficients are greater than 0.8 as screening decomposition pattern combinations;

[0037] The calculation formula of the time domain correlation coefficient is as follows:

[0038]

[0039] Where: and Respectively represent the pth optimal decomposition mode u p The average value and the qth optimal decomposition mode u q The average value of

[0040]

[0041] Where: S represents the square envelope spectrum of the optimal decomposition mode u, and Respectively represent the square envelope spectrum S of the pth optimal decomposition mode p The average value and the square envelope spectrum S of the qth optimal decomposition mode q The average value of

[0042] S12, calculating the two optimal decomposition patterns in the screened decomposition pattern combination to obtain a first envelope spectrum kurtosis value and a second envelope spectrum kurtosis value, discarding the decomposition pattern with the smaller envelope spectrum kurtosis value, and obtaining K output decomposition patterns;

[0043] The calculation formula of the envelope spectrum kurtosis is as follows:

[0044]

[0045] S13, judge K, if K≤Q, execute S14, if K>Q, execute S4;

[0046] S14, obtaining K output decomposition patterns, constructing an envelope spectrum based on the K output decomposition patterns, and performing fault diagnosis on the bearing to be tested based on the envelope spectrum.

[0047] The effects of the present invention are as follows:

[0048] The bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition of the present invention can process multi-channel signals simultaneously, and takes the correlation kurtosis of periodic and impact intensity measurement indicators as the decomposition target. It can specifically decompose the periodic fault impact components in complex signals and effectively reduce the missed alarm rate caused by the randomness of a single channel.

[0049] Compared with traditional decomposition methods, the bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition in the present invention is guided by multi-fault cycle characteristics, fully utilizes multi-channel and multi-fault characteristics, is not restricted by conditions such as the shape and bandwidth of traditional filters, and can extract discrete fault information more comprehensively and completely, and is particularly suitable for multi-fault scenarios.

[0050] The bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition of the present invention uses the envelope spectrum kurtosis index and the time-frequency domain correlation coefficient to quickly eliminate the number of redundant modes, improve the decomposition efficiency, and realize the intelligent identification of multi-channel signal multi-fault characteristics, which is suitable for actual industrial needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of a bearing test bench according to an example of the present invention;

[0052] Figure 2 This is a flow chart of the bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition of the present invention;

[0053] Figure 3 This is a schematic diagram of the time domain waveform and original envelope spectrum of the dual-channel vibration signal intercepted by the example of the present invention;

[0054] Figure 4 Schematic diagram of the optimal decomposition mode and its envelope spectrum obtained by the method of the present invention;

[0055] Figure 5 The diagram of the optimal decomposition mode and its envelope spectrum obtained by the FMD method;

[0056] Figure 6 Schematic diagram of the optimal decomposition mode and its envelope spectrum obtained by the existing MVMD method. DETAILED DESCRIPTION

[0057] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0058] Figure 2 This is a flow chart of the bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition of the present invention, as shown in FIG. Figure 2 As shown, the present invention provides a bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition, which includes:

[0059] Step S1, obtaining the multi-channel vibration signal X of the bearing to be tested raw =[x1 x2 … x O ], where each channel signal is x o =[x o (1) x o (2) … x o (N)] T ,o∈O, O is the number of channels, N is the length of the signal, and the superscript T is the transpose; the vibration signal is preprocessed, including high-frequency sampling, truncation and de-averaging, to obtain a preprocessed signal;

[0060] Step S2: Set the filter length to L, the sliding window length to W, the number of pre-iterations to I, the number of target fault categories to Q, the total number of patterns to K, and the theoretical fault cycle to T. Q ;

[0061] Step S3, calculate the O channel signals to obtain O upper envelope sets, where W is the sliding window length; take the minimum value point of each group of upper envelope sets as the spectrum segmentation set to obtain the initial filter cutoff frequency set of each channel; construct O groups of channel filter banks based on the Hanning window, and the number of filters in each group is K o ,o∈O;

[0062] The calculation formula of the envelope set on each channel is as follows:

[0063]

[0064] Where: UpEnv o (n) represents the oth channel signal x o The calculated parameter value of the nth length point in the upper envelope set, Y o Represents xo The Fourier spectrum of Y o =[Y o (1) Y o (2) … Y o (N)] T , where Y o (n) represents Y o The parameter value at the nth length point, fft is the Fourier function;

[0065] The calculation formula of the cutoff frequency set of each channel is as follows:

[0066]

[0067] Where: [f lower ,f upper ] o Indicates the lower cutoff frequency and upper cutoff frequency of the oth group of initial filters, FreBound o represents the minimum value point set of the envelope set on the oth channel signal, B o is the number of minimum points of the envelope set on the o-th channel signal, FreBound o (b) represents the parameter value of the bth length point position of the minimum value point set of the envelope set on the oth channel signal;

[0068] Step S4, let i=1, j=1, i∈I, j∈Q, I is the number of pre-iterations, Q is the number of fault categories;

[0069] Step S5, judge i, if i≤I, execute S6, if i>I, execute S8;

[0070] Step S6, calculating the correlation kurtosis based on Q theoretical fault cycles, filtering the signals of each channel, and selecting, for each filter, the decomposition mode corresponding to the maximum eigenvalue in the generalized Rayleigh entropy equation as the final decomposition mode;

[0071] The relevant kurtosis calculation formula is as follows:

[0072]

[0073] Where: u o,k =[u o,k (1) u o,k (2) … u o,k (N)] T represents the kth decomposition mode of the oth channel, o∈O,k∈K o ,u o,k (n) represents the parameter value of the nth length point position in the kth decomposition pattern of the oth channel, M is the shift order, f o,k is the vector form of the kth filter of the oth channel, f o,k =[f o,k (1) f o,k (2) … f o,k (L)] T , f o,k (l) represents the parameter value of the lth length point position in the kth filter of the oth channel, T j is the theoretical failure period corresponding to the jth type of fault, j∈Q;

[0074]

[0075] Where: R XWX and R XX are weighted correlation matrices, x o is the oth channel signal, and the superscript H is the conjugate transpose;

[0077] The generalized Rayleigh entropy equation is as follows:

[0078] f o,k R XWX =λf o,k R XX ;

[0079] Where: λ is the eigenvalue;

[0080] Step S7, adjusting the coefficients of all the filters, setting i=i+1 and returning to S5;

[0081] Step S8, judge j, if j≤Q, execute S9, if j>Q, execute S11;

[0082] Step S9: for each channel, o The final decomposition patterns are calculated, and the relevant kurtosis of the final decomposition patterns is obtained. o Calculate the number of retained P, and select the final decomposition mode corresponding to the first P maximum values ​​of the relevant kurtosis as the optimal decomposition mode;

[0083] The calculation formula for the reserved quantity is as follows:

[0084] P=floor(ΣK o / Q·O);

[0085] Where: floor is the rounding down operation, K o is the number of filters for each channel, Q is the number of fault categories, and O is the total number of channels;

[0086] Step S10, set j=j+1 and return to S8;

[0087] Step S11, calculating the time domain correlation coefficient and the frequency domain correlation coefficient between each of the K optimal decomposition patterns, and selecting all optimal decomposition pattern combinations where both the time domain correlation coefficient and the frequency domain correlation coefficient are greater than 0.8 as screening decomposition pattern combinations;

[0088] The calculation formula of the time domain correlation coefficient is as follows:

[0089]

[0090] Where: and Respectively represent the pth optimal decomposition mode u p The average value and the qth optimal decomposition mode u q The average value of

[0091]

[0092] Where: S represents the square envelope spectrum of the optimal decomposition mode u, and Respectively represent the square envelope spectrum S of the pth optimal decomposition mode p The average value and the square envelope spectrum S of the qth optimal decomposition mode q The average value of

[0093] Step S12, calculating the two optimal decomposition patterns in the screened decomposition pattern combination to obtain a first envelope spectrum kurtosis value and a second envelope spectrum kurtosis value, discarding the decomposition pattern with the smaller envelope spectrum kurtosis value, and obtaining K output decomposition patterns;

[0094] The calculation formula of the envelope spectrum kurtosis is as follows:

[0095]

[0096] Step S13, judge K, if K≤Q, execute S14, if K>Q, execute S4;

[0097] Step S14: obtaining K output decomposition patterns, constructing an envelope spectrum based on the K output decomposition patterns, and performing fault diagnosis on the bearing to be tested based on the envelope spectrum.

[0098] Specifically, build Figure 1The test bench shown in the figure consists of a motor, a planetary gearbox, a parallel shaft gearbox, a speed sensor, a bearing to be tested, a brake and other components. The motor drives the inner ring of the bearing after two-stage speed increase through the planetary gearbox and the parallel shaft gearbox. The outer ring of the bearing is fixed to the bearing seat, and the acceleration sensor is fixed directly above the bearing seat to measure the vibration signal of the bearing. The rolling bearing is the bearing to be tested. It has a composite fault of a crack in the outer ring and a crack in the inner ring. The sampling frequency is 20 kHz and the duration is 1 second. The dual-channel vibration signal and its corresponding envelope spectrum are shown as follows: Figure 3 As shown by Figure 3 It can be observed that there is a lot of noise in the signal, and the energy is concentrated in the low-frequency band below 100Hz. Only obvious inner race faults can be shown, and no effective information about outer race faults can be found.

[0099] based on Figure 3 The vibration signal shown in FIG, the optimal decomposition modes M1, M2 and M3 and their envelope spectra obtained by the method of the present invention are as follows Figure 4 As shown, Figure 3 In comparison, the method of the present invention reveals the outer race fault characteristics that are completely submerged by noise, and extracts clear sideband information of the inner race fault frequency, further enhancing the inner race fault characteristics.

[0100] based on Figure 3 The vibration signal shown in the figure uses the existing FMD method to obtain the optimal decomposition mode and its envelope spectrum as shown in the figure. Figure 5 As shown in the figure, it can be found that in the optimal decomposition result, FMD is still limited to the prominent bearing inner race fault features, while the bearing outer race fault features are not successfully extracted, and the decomposition pattern is highly repetitive. Therefore, in this case, FMD cannot diagnose the composite fault.

[0101] based on Figure 3 The vibration signal shown in the figure uses the existing MVMD method to obtain the optimal decomposition mode and its envelope spectrum as shown in the figure. Figure 6 As shown in the figure, most MVMD decomposition patterns enhance the inner race fault signature. Only the M6 ​​decomposition pattern of channel 2 extracts a weak outer race fault signature, but its signal-to-noise ratio is significantly lower than that of the decomposition pattern obtained by the method of the present invention. Furthermore, the MVMD method also generates a large number of redundant repetitive patterns and non-fault patterns. Figures 1-6 In, f I Indicates the characteristic frequency of bearing inner ring fault, f O Indicates the characteristic frequency of bearing outer ring fault, f r Indicates frequency conversion.

[0102] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0103] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A bearing fault diagnosis method based on multi-dimensional synchronous optimal decomposition, It is characterized in that It includes: S1, obtain the multi-channel vibration signal X of the bearing to be tested raw =[x 1 x 2 … xo], x O ], where each channel signal is x o =[x o (1) x o (2) … x o (N)] T , o∈O, O is the number of channels, N is the length of the signal, and the superscript T is the transpose; the vibration signal is preprocessed, including high-frequency sampling, truncation and de-averaging, to obtain a preprocessed signal; S2, set the filter length to L, the sliding window length to W, the number of pre-iterations to I, the number of target fault categories to Q, the total number of patterns to K, and the theoretical fault cycle to T Q ; S3, calculate the O channel signals to obtain O upper envelope sets, W is the sliding window length; take the minimum value point of each group of upper envelope sets as the spectrum segmentation set to obtain the cutoff frequency set of each channel initial filter; construct O groups of channel filter groups based on the Hanning window, and the number of filters in each group is K o ,o∈O; The calculation formula of the envelope set on each channel is as follows: Where: UpEnv o (n) represents the oth channel signal x o The calculated parameter value of the nth length point in the upper envelope set, Y o Represents x o The Fourier spectrum of Y o =[Y o (1)Y o (2)…Y o (N)] T , where Y o (n) represents Y o The parameter value at the nth length point, fft is the Fourier function; The calculation formula of the cutoff frequency set of each channel is as follows: Where: [f lower ,f upper ] o Indicates the lower cutoff frequency and upper cutoff frequency of the oth group of initial filters, FreBound o represents the minimum value set of the envelope set on the oth channel signal, B o is the number of minimum points of the envelope set on the o-th channel signal, FreBound o (b) represents the parameter value of the bth length point position of the minimum point set of the envelope set on the oth channel signal; S4, let i=1, j=1, i∈I, j∈Q, I is the number of pre-iterations, Q is the number of fault categories; S5, judge i, if i≤I, execute S6, if i>I, execute S8; S6, calculating the correlation kurtosis based on Q theoretical fault cycles, filtering the signals of each channel, and for each filter, selecting the decomposition mode corresponding to the maximum value of the eigenvalue in the generalized Rayleigh entropy equation as the final decomposition mode; The relevant kurtosis calculation formula is as follows: Where: u o,k =[u o,k (1) u o,k (2) … u o,k (N)] T represents the kth decomposition mode of the oth channel, o∈O,k∈K o ,u o,k (n) represents the parameter value of the nth length point position in the kth decomposition pattern of the oth channel, M is the shift order, f o,k is the vector form of the kth filter of the oth channel, f o,k =[f o,k (1)f o,k (2) … f o,k (L)] T , f o,k (l) represents the parameter value of the lth length point position in the kth filter of the oth channel, T j is the theoretical fault cycle corresponding to the jth type of fault, j∈Q; Where: R XWX and R XX are weighted correlation matrices, x o is the oth channel signal, and the superscript H is the conjugate transpose; The generalized Rayleigh entropy equation is as follows: f o,k R XWX =λf o,k R XX ; Where: λ is the eigenvalue; S7, adjusting the coefficients of all the filters, setting i=i+1 and returning to S5; S8, judge j, if j≤Q, execute S9, if j>Q, execute S11; S9, for each channel K o The final decomposition patterns are calculated for each of the final decomposition patterns, and the K o The relevant kurtosis; Calculate the number of retentions P, and select the final decomposition mode corresponding to the first P maximum values ​​of the relevant kurtosis as the optimal decomposition mode; The calculation formula for the reserved quantity is as follows: P=floor(∑K o / Q·O); In the formula: floor is the rounding down operation, K o is the number of filters for each channel, Q is the number of fault categories, and O is the total number of channels; S10, set j=j+1 and return to S8; S11, calculating the time domain correlation coefficient and the frequency domain correlation coefficient between the K optimal decomposition patterns, and selecting all the optimal decomposition pattern combinations where both coefficients are greater than 0.8 as the screening decomposition pattern combinations; The calculation formula of the time domain correlation coefficient is as follows: Where: and Respectively represent the pth optimal decomposition mode u p The average value and the qth optimal decomposition mode u q The average value of s.t.S=abs(fft(u 2 )) Where: S represents the square envelope spectrum of the optimal decomposition mode u, and Respectively represent the square envelope spectrum S of the pth optimal decomposition mode p The average value and the square envelope spectrum S of the qth optimal decomposition mode q The average value of S12, calculating two optimal decomposition modes in the screening decomposition mode combination to obtain a first envelope spectrum kurtosis value and a second envelope spectrum kurtosis value, discarding the decomposition mode with a smaller envelope spectrum kurtosis value, and obtaining K output decomposition modes; The calculation formula of the envelope spectrum kurtosis is as follows: S13, judge K, if K≤Q, execute S14, if K>Q, execute S4; S14, obtaining K output decomposition modes, constructing an envelope spectrum based on the K output decomposition modes, and performing fault diagnosis on the bearing to be tested based on the envelope spectrum.

Citation Information

Cited By

  • PON optical transmission and monitoring method

    CN119743196A

  • Numerical control machine tool main shaft bearing feature extraction method based on improved FMD

    CN121350567A

  • High-voltage cable partial discharge identification method and system based on TMR array probe

    CN122410242A