An adaptive method for extracting rolling bearing fault features under strong noise interference

Through improved signal decomposition method and parameter adaptive optimization, the problem of rolling bearing fault feature extraction under strong noise interference is solved, and accurate fault diagnosis is achieved in a noisy environment.

CN118464448BActive Publication Date: 2024-11-22CSC BEARING
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410586925.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-13
Publication Date
2024-11-22
Estimated Expiration
2044-05-13

AI Technical Summary

Technical Problem

Under strong noise interference, rolling bearing failure characteristics are difficult to accurately and adaptively extract, resulting in misdiagnosis or misdiagnosis, increasing maintenance costs and accident risks.

Method used

Adaptive noise-complete ensemble empirical mode decomposition based on 1.5-dimensional square envelope spectrum kurtitude improvement, eigenmodal decomposition with dual-domain Gini index optimization, and sparse maximum harmonic noise ratio solution guided by square envelope harmonic noise ratio Convolution, adaptively optimize parameters to extract rolling bearing failure characteristics.

Benefits of technology

Effectively handle strong noise interference, improve the accuracy and reliability of fault feature extraction, and accurately extract the fault information of rolling bearings in a noisy environment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118464448B_ABST
    Figure CN118464448B_ABST
Patent Text Reader

Abstract

The invention discloses a method for adaptively extracting rolling bearing fault features under strong noise interference. The method mainly comprises the following steps: inputting a collected rolling bearing vibration signal; decomposing the original signal into a plurality of modal components by using adaptive noise complete set empirical mode decomposition improved based on 1.5-dimensional square envelope spectrum kurtosis; calculating the correlation coefficient between each component and the original signal; reconstructing the modal components with higher correlation; optimizing the input parameters of the characteristic mode decomposition by using a dual-domain Gini index, and performing secondary decomposition on the reconstructed signal; extracting sensitive modal components according to a maximum criterion of a square envelope spectrum peak factor; enhancing the fault features of the sensitive modal components by using a parameter adaptive sparse maximum harmonic noise ratio deconvolution guided by a square envelope harmonic noise ratio, and then performing Hilbert envelope demodulation; and extracting rolling bearing fault feature information from the envelope spectrum. The invention overcomes the problem that the rolling bearing fault features are difficult to be accurately and adaptively extracted under strong noise interference.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of fault diagnosis of rotating machinery parts, and in particular to a method for adaptively extracting rolling bearing fault features under strong noise interference. Background Art

[0002] Rolling bearings are extremely important components in rotating machinery, responsible for supporting rotating shafts and transmitting loads. They are widely used in various mechanical equipment, such as high-end CNC machine tools, automobiles, high-speed trains, wind turbines, and aerospace equipment. Therefore, in order to maintain the efficient and safe operation of rotating machinery, continuous monitoring and accurate fault diagnosis of rolling bearings are indispensable. However, the interference of strong noise on rolling bearing fault diagnosis is a common problem. Strong noise may come from the machinery itself or the external environment. Under strong noise interference, the quality and resolution of the signal will be affected, and the signal in this case is usually more complex. Existing fault diagnosis methods are more likely to misdiagnose or miss diagnosis under strong noise interference. This will not only lead to increased maintenance costs, but also ignore mechanical failures, thereby increasing the risk of accidents. Therefore, providing a new method that can effectively process vibration signals with strong noise interference and accurately extract fault feature information from them has important engineering application value.

[0003] Signal decomposition methods are key technologies for processing and analyzing data signals, especially when processing complex, non-stationary or heavily noise-interfered signals. These methods can help identify and extract useful information from signals, which is crucial for fault diagnosis and feature analysis. Complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is a signal decomposition method that improves empirical mode decomposition (EMD). It decomposes the signal by adding white noise to the original signal, reduces the influence of noise on the decomposition result by using the concept of ensemble average, and enhances the stability and reliability of the decomposition. Eigenmode decomposition (FMD) is a new decomposition theory that decomposes the vibration signal into several independent modes by constructing an adaptive filter, and uses the autocorrelation function to get rid of the prior knowledge of the fault cycle, and has a certain robustness to noise and interference signals. However, both of the above methods have some defects. The white noise amplitude weight NS and the number of noise additions NR of the complete ensemble empirical mode decomposition with adaptive noise, the final number of decomposition modes n of the eigenmode decomposition, the number of filters k set initially and the length of the filter L all affect the final decomposition effect of the two decomposition methods, and these parameters cannot be automatically selected for different vibration signals.

[0004] Deconvolution methods are used to analyze and process vibration signals that become blurred due to the influence of propagation paths and system responses. At the same time, the use of deconvolution techniques can help recover the original fault signal from the signal and improve the accuracy and reliability of fault feature extraction. Sparse Maximum Harmonic-to-Noise Ratio Deconvolution (SMHD) is a new deconvolution method that aims to enhance weak fault features from noise signals. By maximizing the harmonic-to-noise ratio and optimizing signal processing using sparse technology, SMHD not only overcomes the limitations of traditional deconvolution methods, but also shows good fault feature enhancement effects without the need for fault cycles to be provided in advance. However, the processing effect of SMHD depends on the empirical setting of the filter length SL and the initial sparse threshold MU, and the parameter values ​​cannot be adaptively selected for different signals. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] Therefore, the purpose of the present invention is to provide a method for adaptively extracting rolling bearing fault features under strong noise interference, so as to overcome the problem that rolling bearing fault features are difficult to be accurately adaptively extracted under strong noise interference.

[0007] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0008] A method for adaptively extracting rolling bearing fault features under strong noise interference, comprising:

[0009] S10, inputting the rolling bearing vibration signal collected in real time;

[0010] S20, decomposing the collected real-time vibration signal into a set of optimal intrinsic mode components using an adaptive noise complete set empirical mode decomposition improved based on 1.5-dimensional square envelope spectrum kurtosis BSESK;

[0011] S30, calculating the correlation coefficient between each eigenmode component in the group of best eigenmode components and the collected vibration signal;

[0012] S40, the correlation coefficient is greater than the correlation threshold c v The eigenmode components of are used to reconstruct the signal to obtain the reconstructed signal;

[0013] S50, optimizing the input parameters of the eigenmode decomposition using the dual-domain Gini index DDGI, and decomposing the reconstructed signal using the eigenmode decomposition with the optimal input parameters into a set of optimal eigenmode components;

[0014] S60, according to the maximum square envelope spectrum peak factor criterion, extracting the characteristic modal component containing the most bearing fault characteristics from the group of best characteristic modal components as the sensitive modal component;

[0015] S70, using a square envelope harmonic noise ratio (SEHNR) guided parameter adaptive sparse maximum harmonic noise ratio deconvolution to enhance fault characteristics of sensitive modal components, and obtaining an optimal characteristic enhancement component;

[0016] S80, performing Hilbert envelope demodulation on the best feature enhancement component and obtaining an envelope spectrum;

[0017] S90. Extract rolling bearing fault feature information from the envelope spectrum of the best feature enhancement component.

[0018] As a preferred solution of the method for adaptively extracting rolling bearing fault features under strong noise interference described in the present invention, in step S20, the 1.5-dimensional square envelope spectrum kurtosis BSESK is a frequency domain index for evaluating the sparsity of bearing vibration signals constructed for adaptive noise complete set empirical mode decomposition. The larger the BSESK value, the better the sparsity of the signal, that is, the more periodic fault impact features are contained in the signal. The calculation formula of BSESK is as follows:

[0019]

[0020] Wherein, s is the vibration time series signal, s(t) is the signal amplitude at time t, τ is the time delay, E(·) represents the mathematical expectation operation, j is the imaginary unit, f is the frequency, N is the number of time periods of the 1.5-dimensional square envelope spectrum, i is a positive integer and i=1, 2,···,N.

[0021] As a preferred solution of the method for adaptively extracting rolling bearing fault features under strong noise interference described in the present invention, in step S20, the specific steps of decomposing the collected real-time vibration signal into a set of optimal intrinsic mode components and optimal IMFs by using the adaptive noise complete set empirical mode decomposition CEEMDAN improved based on 1.5-dimensional square envelope spectrum kurtosis BSESK are as follows:

[0022] Step S201, obtain all parameter combinations of two input parameters that affect the decomposition performance of the adaptive noise complete set empirical mode decomposition CEEMDAN, the two input parameters are the white noise amplitude weight NS and the noise addition times NR, the parameter combination is expressed as [NS(u), NR(v)], and their values ​​are as follows:

[0023]

[0024] Where N* is a set of positive integers;

[0025] Step S202: Input a set of parameter combinations [NS(u), NR(v)] into CEEMDAN, and decompose the collected real-time vibration signal into a set of intrinsic mode components IMFs, expressed as:

[0026] [IMFs]=CEEMDAN(x,NS(u),NR(v),MaxIter c )

[0027] Among them, CEEDMAN(·) is the adaptive noise complete set empirical mode decomposition function operation in MATLAB, x represents the real-time vibration signal collected, NS(u) is the white noise amplitude weight in the current parameter combination, NR(v) is the number of noise additions in the current parameter combination, and MaxIter c is the number of iterations of CEEMDAN, MaxIter c Set to infinity, IMFs is a set of intrinsic mode components decomposed by CEEMDAN after applying the current parameter combination;

[0028] Step S203, calculating the 1.5-dimensional square envelope spectrum kurtosis BSESK value of each intrinsic mode component IMF in the current group of IMFs, and recording the maximum BSESK value in the current group as the local maximum BSESK;

[0029] Step S204: Repeat steps S202 and S203 until all local maximum BSESK values ​​corresponding to all parameter combinations of the two input parameters are calculated, and these BSESK values ​​are compared. The largest BSESK value is selected as the global maximum BSESK value, and the corresponding parameter combination is the global optimal parameter combination of CEEMDAN. The global optimal parameter combination of CEEMDAN is expressed as [NS(u) best NR(v) best ];

[0030] Step S205: Using [NS(u) best NR(v) best ]'s CEEMDAN decomposes the collected real-time vibration signal into a set of optimal intrinsic mode components, namely, optimal IMFs.

[0031] As a preferred solution of the method for adaptively extracting rolling bearing fault features under strong noise interference described in the present invention, in step S30, the correlation coefficient C between each intrinsic mode component in the group of best intrinsic mode components and the collected vibration signal is calculated, and the calculation formula is as follows:

[0032]

[0033] Wherein, x represents the collected real-time vibration signal, y(r) is the rth intrinsic mode component in this group of optimal intrinsic mode components, C(r) is the correlation coefficient between the rth intrinsic mode component in this group of optimal intrinsic mode components and the collected real-time vibration signal, M is the length of the signal, m is the mth signal point, and m=1, 2,···,M.

[0034] As a preferred solution of the method for adaptively extracting rolling bearing fault features under strong noise interference described in the present invention, in step S50, the dual-domain Gini index DDGI is a comprehensive sparsity index that considers both the time domain and the frequency domain of the vibration signal and is constructed for optimizing the input parameters of the characteristic mode decomposition. The larger the DDGI value, the better the sparsity of the signal, that is, the more fault cycle impact feature information contained in the signal. The calculation formula of DDGI is as follows:

[0035]

[0036] Wherein, s is the vibration time series signal, M is the length of the signal, m=1,2,···,M, s(m)=(s(1),s(2),s(3),···,s(M)), the elements in s(m) are arranged from small to large, ‖·‖1 is the 1-norm operation, FFT(·) is the fast Fourier transform in MATLAB, abs(·) is the absolute value operation in MATLAB, FS represents the square envelope spectrum of the vibration time series signal s, Q is the length of the signal FS, q=1,2,···,Q, FS(q)=(FS(1),FS(2),FS(3),···,FS(Q)), the elements in FS(q) are arranged from small to large, α s is the sensitivity coefficient of the square envelope Gini index, β s is the sensitivity coefficient of the square envelope spectrum Gini index, α s ∈(0,1),β s ∈(0,1), and α s +β s =1.

[0037] As a preferred solution of the method for adaptively extracting rolling bearing fault features under strong noise interference described in the present invention, in step S50, the input parameters of the characteristic mode decomposition are optimized by the dual-domain Gini index DDGI, and the reconstructed signal is decomposed by the characteristic mode decomposition with the optimal input parameters to decompose into a set of optimal characteristic mode components. The specific steps are as follows:

[0038] Step S501, obtain all parameter combinations of three input parameters that affect the decomposition performance of the eigenmode decomposition, the three input parameters are the number of modes n of the final decomposition, the number of filters k initially set and the length L of the filter, and the three eigenmode decomposition input parameter combinations are expressed as [n(p), k(g), L(h)], and their values ​​are as follows:

[0039]

[0040] Where N* is a set of positive integers;

[0041] Step S502: Decompose the reconstructed signal into a set of eigenmode components Modes using eigenmode decomposition with a set of parameter combinations [n(p), k(g), L(h)],

[0042] Step S503, calculating the dual-domain Gini index DDGI value of each characteristic mode component Mode in the current group Modes, and recording the maximum DDGI value in the current group as the local maximum DDGI;

[0043] Step S504, repeat steps S502 and S503 until all local maximum DDGI values ​​corresponding to all parameter combinations of the three input parameters of the eigenmode decomposition are calculated, compare these DDGI values, select the largest DDGI value as the global maximum DDGI value, and the corresponding eigenmode decomposition input parameter combination is the global optimal parameter combination of the eigenmode decomposition. The global optimal parameter combination of the eigenmode decomposition is expressed as [n(p) best , k(g) best 、L(h) best ];

[0044] Step S505: using [n(p) best , k(g) best 、L(h) best ]’s eigenmode decomposition decomposes the reconstructed signal into a set of optimal eigenmode components, namely, optimal Modes.

[0045] As a preferred solution of the method for adaptively extracting rolling bearing fault features under strong noise interference described in the present invention, in step S60, the square envelope spectrum peak factor SE cThe maximum criterion refers to calculating the square envelope spectrum peak factor SE of each component in the best eigenmode component group. c value, and select the largest SE c The modal component corresponding to the value is the sensitive modal component; the square envelope spectrum peak factor SE of a signal c The larger the value, the better the periodicity of the signal, the higher the signal-to-noise ratio, and the square envelope spectrum peak factor SE. c The calculation formula is as follows:

[0046]

[0047] Among them, N bearing Indicates the bearing speed in rpm; f r represents the theoretical rotation frequency of the bearing; f is the abscissa of the square envelope spectrum, i.e., the frequency; f s is the sampling frequency of the signal; s is the vibration time series signal; FFT(·) is the fast Fourier transform in MATLAB, abs(·) is the absolute value operation in MATLAB, FS represents the square envelope spectrum of the vibration time series signal s; max(·) is the maximum value operation in MATLAB, rms(·) is the root mean square value operation in MATLAB.

[0048] As a preferred solution of the method for adaptively extracting rolling bearing fault features under strong noise interference described in the present invention, in step S70, the sensitive modal components are enhanced by using the parameter adaptive sparse maximum harmonic noise ratio deconvolution SMHD guided by the square envelope harmonic noise ratio SEHNR to enhance the fault features, and the specific steps of obtaining the optimal feature enhancement component are as follows:

[0049] Step S701, obtain all parameter combinations of parameters to be adapted in the sparse maximum harmonic-to-noise ratio deconvolution SMHD, the parameters to be adapted are the filter length SL and the initial sparse threshold MU, the parameter combination is expressed as [SL(w), MU(z)], and their values ​​are as follows:

[0050]

[0051] Among them, S m is the sensitive modal component, rms(·) is the root mean square value operation in MATLAB, and N* is a positive integer set;

[0052] Step S702: Input a set of parameter combinations [SL(w), MU(z)] into SMHD, and enhance the fault characteristics of the sensitive modal components to obtain the enhanced characteristic components, which are expressed as:

[0053] [zmode]=SMHD(f s ,S m,SL(w),Iter SMHD ,MU(z),T,Plotmode)

[0054] Where SMHD(·) is the sparse maximum harmonic noise ratio deconvolution function in MATLAB, f s is the sampling frequency of the signal, S m is the sensitive modal component, SL(w) is the filter length in the current parameter combination, Iter SMHD is the maximum number of iterations of SMHD, which is set to 30, MU(z) is the initial sparse threshold in the current parameter combination, T is the signal failure period, Plotmode indicates whether to display the processed signal, and zmode indicates the feature enhancement component after SMHD processing with the current parameter combination;

[0055] Step S703: Calculate the square envelope harmonic noise ratio SEHNR value of the current feature enhancement component, and record it as the local maximum SEHNR. The larger the SEHNR value, the richer the fault feature information in the signal. The calculation formula of SEHNR is as follows:

[0056]

[0057] Among them, s z (t) is the feature enhancement component, t is time, T g is the lag time when the autocorrelation function reaches the local maximum, i.e., the estimated fault period;

[0058] Step S704, repeat steps S702 and S703 until all local maximum SEHNR values ​​corresponding to all parameter combinations of the parameters to be adapted are calculated, compare these SEHNR values, select the largest SEHNR value as the global maximum SEHNR value of SMHD, and the corresponding parameter combination is the global optimal parameter combination of SMHD. The global optimal parameter combination of SMHD is expressed as [SL(w) best MU(z) best ];

[0059] Step S705: Using [SL(w) best MU(z) best ]'s SMHD enhances the fault characteristics of the sensitive modal components and obtains the optimal feature enhancement component, namely the optimal zmode.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] (1) The present invention constructs a frequency domain index for evaluating the sparsity of bearing vibration signals for adaptive noise complete set empirical mode decomposition, namely, the 1.5-dimensional square envelope spectral kurtosis (BSESK), and uses this index to complete the adaptive optimization of the white noise amplitude weight NS and the number of noise additions NR of the adaptive noise complete set empirical mode decomposition, giving full play to the noise reduction preprocessing performance of the adaptive noise complete set empirical mode decomposition.

[0062] (2) The present invention aims at optimizing the input parameters of eigenmode decomposition and constructs a comprehensive sparsity index that considers both the time domain and the frequency domain of the vibration signal, namely the dual-domain Gini index DDGI. The index is used to optimize the final number of decomposition modes n, the number of filters initially set k, and the length L of the filter, thereby giving full play to the adaptive secondary filtering denoising effect of the eigenmode decomposition.

[0063] (3) The present invention aims to solve the problem of adaptive parameters of sparse maximum harmonic noise ratio deconvolution and constructs a time domain index for evaluating the sparsity of bearing vibration signals, namely, the square envelope harmonic noise ratio SEHNR. The index is used to adaptively select the filter length SL and the initial sparse threshold MU of the sparse maximum harmonic noise ratio deconvolution, giving full play to the adaptive fault feature enhancement effect of the sparse maximum harmonic noise ratio deconvolution.

[0064] (4) The method provided by the present invention includes three parts: noise reduction preprocessing, secondary filtering noise reduction and fault feature enhancement. Compared with signal processing methods such as CEEMDAN, FMD and SMHD, it can effectively process vibration signals with strong noise interference, and can adaptively and accurately extract rolling bearing fault feature information therefrom. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail below in combination with the accompanying drawings and detailed embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0066] Figure 1 This is a flow chart of a method for adaptively extracting rolling bearing fault features under strong noise interference according to an embodiment of the present invention;

[0067] Figure 2 This is a time domain diagram of a rolling bearing fault signal under strong noise interference according to an embodiment of the present invention;

[0068] Figure 3 This is an envelope spectrum of a rolling bearing fault signal under strong noise interference according to an embodiment of the present invention;

[0069] Figure 4 A time domain diagram of a group of optimal intrinsic mode components decomposed by CEEMDAN improved based on BSESK according to an embodiment of the present invention;

[0070] Figure 5 A variation trend diagram of the correlation coefficient value between each component in the best eigenmode component group and the collected signal according to an embodiment of the present invention;

[0071] Figure 6 A time domain diagram of a reconstructed signal according to an embodiment of the present invention;

[0072] Figure 7 The reconstructed signal envelope spectrum of an embodiment of the present invention;

[0073] Figure 8 A time domain diagram of a group of optimal eigenmode components of a reconstructed signal decomposed after optimizing the input parameters of eigenmode decomposition by DDGI according to an embodiment of the present invention;

[0074] Fig. 9 is an envelope spectrum diagram of a sensitive modal component according to an embodiment of the present invention;

[0075] Fig.10 A time domain diagram of the best feature enhancement component of an embodiment of the present invention;

[0076] Fig.11 It is the envelope spectrum of the best feature enhancement component according to an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0078] The present invention provides a method for adaptively extracting rolling bearing fault features under strong noise interference, which overcomes the problem that rolling bearing fault features are difficult to accurately and adaptively extract under strong noise interference.

[0079] See also Figure 1 , which shows a flow chart of a method for adaptively extracting rolling bearing fault features under strong noise interference provided by an embodiment of the present invention, comprising the following steps:

[0080] Step S10, inputting the rolling bearing vibration signal collected in real time;

[0081] Step S20, decomposing the collected real-time vibration signal into a set of optimal intrinsic mode components by using an adaptive noise complete set empirical mode decomposition improved based on 1.5-dimensional square envelope spectral kurtosis (BSESK);

[0082] Step S30, calculating the correlation coefficient between each eigenmode component in the group of best eigenmode components and the collected vibration signal;

[0083] Step S40: The correlation coefficient is greater than the correlation threshold c v The eigenmode components of are used to reconstruct the signal to obtain the reconstructed signal;

[0084] Step S50, optimizing the input parameters of the eigenmode decomposition using the dual domain Gini index (DDGI), and decomposing the reconstructed signal using the eigenmode decomposition with the optimal input parameters into a set of optimal eigenmode components;

[0085] Step S60, according to the maximum criterion of the square envelope spectrum peak factor, extracting the characteristic modal component containing the most bearing fault characteristics from the group of optimal characteristic modal components as the sensitive modal component;

[0086] Step S70, using parameter adaptive sparse maximum harmonic noise ratio deconvolution guided by square envelope harmonic noise ratio (SEHNR) to enhance fault characteristics of sensitive modal components to obtain optimal characteristic enhancement components;

[0087] Step S80, performing Hilbert envelope demodulation on the best feature enhancement component and obtaining an envelope spectrum;

[0088] Step S90: extract rolling bearing fault feature information from the envelope spectrum of the best feature enhancement component.

[0089] In one embodiment of the present invention, in step S20, 1.5-dimensional square envelope spectrum kurtosis (BSESK) is a frequency domain index for evaluating the sparsity of bearing vibration signals constructed for adaptive noise complete set empirical mode decomposition. The larger the BSESK value, the better the sparsity of the signal, that is, the more periodic fault impact features contained in the signal. The calculation formula of BSESK is as follows:

[0090]

[0091] Wherein, s is the vibration time series signal, s(t) is the signal amplitude at time t, τ is the time delay, E(·) represents the mathematical expectation operation, j is the imaginary unit, f is the frequency, N is the number of time periods of the 1.5-dimensional square envelope spectrum, i is a positive integer and i=1, 2,···,N.

[0092] In one embodiment of the present invention, in step S20, the specific steps of decomposing the collected real-time vibration signal into a set of optimal intrinsic mode components (optimal IMFs) using the improved adaptive noise complete ensemble empirical mode decomposition (CEEMDAN) based on 1.5-dimensional square envelope spectral kurtosis (BSESK) are as follows:

[0093] Step S201, obtain all parameter combinations of two input parameters that affect the decomposition performance of the adaptive noise complete ensemble empirical mode decomposition (CEEMDAN), the two input parameters are the white noise amplitude weight NS and the noise addition times NR, the parameter combination is expressed as [NS(u), NR(v)], and their values ​​are as follows:

[0094]

[0095] Where N* is a set of positive integers;

[0096] Step S202: Input a set of parameter combinations [NS(u), NR(v)] into CEEMDAN, and decompose the collected real-time vibration signal into a set of intrinsic mode components (IMFs), expressed as:

[0097] [IMFs]=CEEMDAN(x,NS(u),NR(v),MaxIter c )

[0098] Among them, CEEDMAN(·) is the adaptive noise complete set empirical mode decomposition function operation in MATLAB, x represents the real-time vibration signal collected, NS(u) is the white noise amplitude weight in the current parameter combination, NR(v) is the number of noise additions in the current parameter combination, and MaxIter c is the number of iterations of CEEMDAN, MaxIter c Set to infinity, IMFs is a set of intrinsic mode components decomposed by CEEMDAN after applying the current parameter combination;

[0099] Step S203, calculating the 1.5-dimensional square envelope spectrum kurtosis (BSESK) value of each intrinsic mode component IMF in the current group of IMFs, and recording the maximum BSESK value in the current group as the local maximum BSESK;

[0100] Step S204: Repeat steps S202 and S203 until all local maximum BSESK values ​​corresponding to all parameter combinations of the two input parameters are calculated, and these BSESK values ​​are compared. The largest BSESK value is selected as the global maximum BSESK value, and the corresponding parameter combination is the global optimal parameter combination of CEEMDAN. The global optimal parameter combination of CEEMDAN is expressed as [NS(u) best NR(v) best ];

[0101] Step S205: Using [NS(u) best NR(v) best ]'s CEEMDAN decomposes the acquired real-time vibration signal into a set of optimal intrinsic mode components (optimal IMFs).

[0102] In one embodiment of the present invention, in step S30, the correlation coefficient C between each eigenmode component in the set of best eigenmode components and the collected vibration signal is calculated, and the calculation formula is as follows:

[0103]

[0104] Wherein, x represents the collected real-time vibration signal, y(r) is the rth intrinsic mode component in this group of optimal intrinsic mode components, C(r) is the correlation coefficient between the rth intrinsic mode component in this group of optimal intrinsic mode components and the collected real-time vibration signal, M is the length of the signal, m is the mth signal point, and m=1, 2,···,M.

[0105] In one embodiment of the present invention, in step S50, the dual-domain Gini index (DDGI) is a comprehensive sparsity index that considers both the time domain and the frequency domain of the vibration signal and is constructed for optimizing the input parameters of the characteristic mode decomposition. The larger the DDGI value, the better the sparsity of the signal, that is, the more fault cycle impact characteristic information contained in the signal. The calculation formula of DDGI is as follows:

[0106]

[0107] Wherein, s is the vibration time series signal, M is the length of the signal, m=1,2,···,M, s(m)=(s(1),s(2),s(3),···,s(M)), the elements in s(m) are arranged from small to large, ‖·‖1 is the 1-norm operation, FFT(·) is the fast Fourier transform in MATLAB, abs(·) is the absolute value operation in MATLAB, FS represents the square envelope spectrum of the vibration time series signal s, Q is the length of the signal FS, q=1,2,···,Q, FS(q)=(FS(1),FS(2),FS(3),···,FS(Q)), the elements in FS(q) are arranged from small to large, α s is the sensitivity coefficient of the square envelope Gini index, β s is the sensitivity coefficient of the square envelope spectrum Gini index, α s ∈(0,1),β s ∈(0,1), and α s +β s =1.

[0108] In one embodiment of the present invention, in step S50, the input parameters of the eigenmode decomposition are optimized by the dual domain Gini index (DDGI), and the reconstructed signal is decomposed into a set of optimal eigenmode components by using the eigenmode decomposition with the optimal input parameters. The specific steps of this step are as follows:

[0109] Step S501, obtain all parameter combinations of three input parameters that affect the decomposition performance of the eigenmode decomposition, the three input parameters are the number of modes n of the final decomposition, the number of filters k initially set and the length L of the filter, and the three eigenmode decomposition input parameter combinations are expressed as [n(p), k(g), L(h)], and their values ​​are as follows:

[0110]

[0111] Where N* is a set of positive integers;

[0112] Step S502: Decompose the reconstructed signal into a set of eigenmode components (Modes) using eigenmode decomposition with a set of parameter combinations [n(p), k(g), L(h)],

[0113] Step S503, calculating the dual domain Gini index (DDGI) value of each characteristic modal component Mode in the current group Modes, and recording the maximum DDGI value in the current group as the local maximum DDGI;

[0114] Step S504, repeat steps S502 and S503 until all local maximum DDGI values ​​corresponding to all parameter combinations of the three input parameters of the eigenmode decomposition are calculated, compare these DDGI values, select the largest DDGI value as the global maximum DDGI value, and the corresponding eigenmode decomposition input parameter combination is the global optimal parameter combination of the eigenmode decomposition. The global optimal parameter combination of the eigenmode decomposition is expressed as [n(p) best , k(g) best 、L(h) best ];

[0115] Step S505: using [n(p) best , k(g) best 、L(h) best ]'s eigenmode decomposition decomposes the reconstructed signal into a set of optimal eigenmode components (optimal Modes).

[0116] In one embodiment of the present invention, in step S60, the squared envelope spectrum peak factor (SE c ) The maximum criterion refers to calculating the square envelope spectrum peak factor (SE c ) value and select the largest SE c The modal component corresponding to the value is the sensitive modal component; the square envelope spectrum peak factor (SE c ) is larger, the better the periodicity of the signal, the higher the signal-to-noise ratio, and the square envelope spectrum peak factor (SE c ) is calculated as follows:

[0117]

[0118] Among them, N bearing Indicates the bearing speed in rpm; f r represents the theoretical rotation frequency of the bearing; f is the abscissa of the square envelope spectrum, i.e., the frequency; f s is the sampling frequency of the signal; s is the vibration time series signal; FFT(·) is the fast Fourier transform in MATLAB, abs(·) is the absolute value operation in MATLAB, FS represents the square envelope spectrum of the vibration time series signal s; max(·) is the maximum value operation in MATLAB, rms(·) is the root mean square value operation in MATLAB.

[0119] In one embodiment of the present invention, in step S70, the sensitive modal component is enhanced with parameter-adaptive sparse maximum harmonic noise ratio deconvolution (SMHD) guided by square envelope harmonic noise ratio (SEHNR) to obtain the best feature enhancement component, and the specific steps are as follows:

[0120] Step S701, obtain all parameter combinations of parameters to be adapted in sparse maximum harmonic-to-noise ratio deconvolution (SMHD), the parameters to be adapted are filter length SL and initial sparse threshold MU, the parameter combination is expressed as [SL(w), MU(z)], and their values ​​are as follows:

[0121]

[0122] Among them, S m is the sensitive modal component, rms(·) is the root mean square value operation in MATLAB, and N* is a positive integer set;

[0123] Step S702: Input a set of parameter combinations [SL(w), MU(z)] into SMHD, and enhance the fault characteristics of the sensitive modal components to obtain the enhanced characteristic components, which are expressed as:

[0124] [zmode]=SMHD(f s ,S m ,SL(w),Iter SMHD ,MU(z),T,Plotmode)

[0125] Where SMHD(·) is the sparse maximum harmonic noise ratio deconvolution function in MATLAB, f s is the sampling frequency of the signal, S m is the sensitive modal component, SL(w) is the filter length in the current parameter combination, Iter SMHDis the maximum number of iterations of SMHD, which is set to 30, MU(z) is the initial sparse threshold in the current parameter combination, T is the signal failure period, Plotmode indicates whether to display the processed signal, and zmode indicates the feature enhancement component after SMHD processing with the current parameter combination;

[0126] Step S703: Calculate the square envelope harmonic noise ratio (SEHNR) value of the current feature enhancement component, and record it as the local maximum SEHNR. The larger the SEHNR value, the richer the fault feature information in the signal. The calculation formula of SEHNR is as follows:

[0127]

[0128] Among them, s z (t) is the feature enhancement component, t is time, T g is the lag time when the autocorrelation function reaches the local maximum, i.e., the estimated fault period;

[0129] Step S704, repeat steps S702 and S703 until all local maximum SEHNR values ​​corresponding to all parameter combinations of the parameters to be adapted are calculated, compare these SEHNR values, select the largest SEHNR value as the global maximum SEHNR value of SMHD, and the corresponding parameter combination is the global optimal parameter combination of SMHD. The global optimal parameter combination of SMHD is expressed as [SL(w) best MU(z) best ];

[0130] Step S705: Using [SL(w) best MU(z) best ]'s SMHD enhances the fault characteristics of the sensitive modal components and obtains the optimal feature enhancement component (optimal zmode).

[0131] In order to verify the effect of the adaptive extraction method of rolling bearing fault features under strong noise interference of the present invention, the following specific implementation case is provided:

[0132] The public data set of the Bearing Data Center of Western Reserve University is used to verify the effectiveness of the method of the present invention. The experimental platform consists of a 1.5KW (2 horsepower) motor, a torque sensor, a power tester and an electronic controller. The driving end bearing to be tested is SKF6205-2RS deep groove ball bearing, with an inner ring diameter of 25mm, an outer ring diameter of 52mm, a thickness of 15mm, a rolling element diameter of 7.94mm, a pitch diameter of 39.04mm, and 9 rolling elements. The damage to the outer ring of the bearing is a single-point damage caused by electrospark machining, with a diameter of 0.1778mm and a depth of 0.2794mm. The speed of the motor is 1797rpm, the sampling frequency of the vibration signal is 12000Hz, the number of sampling points is 8192 points, and the theoretical fault frequency of the outer ring of the bearing is 107.37Hz.

[0133] The collected weak bearing fault vibration signal contains some noise interference. In order to further simulate the bearing vibration signal under strong noise interference, a Gaussian white noise interference signal with a signal-to-noise ratio (SNR) of -15dB is added to the weak bearing fault signal collected in the experiment. The time domain diagram of the bearing outer ring fault vibration signal is shown in Figure 2. Figure 2 As shown, the envelope spectrum after Hilbert envelope demodulation is as follows Figure 3 As shown. Figure 2 It can be seen from the time domain diagram that the noise signal in the signal is very strong, and the periodic pulse fault signal is submerged in it, and it is impossible to extract the fault feature information through the time domain diagram. Figure 3 It is observed from the envelope spectrum that due to the interference of strong noise, not only the fault characteristic frequency and its harmonic frequency cannot be observed, but also three abnormal interference frequencies with obvious abnormal amplitudes are added, which seriously hinders the extraction of fault characteristic information. It can be seen that it is impossible to extract effective fault features from the rolling bearing vibration signal under strong noise interference only through the time domain diagram and envelope spectrum.

[0134] Therefore, the method provided by the present invention is used. First, the collected real-time vibration signal is decomposed into a set of optimal intrinsic mode components using the adaptive noise complete set empirical mode decomposition improved based on 1.5-dimensional square envelope spectral kurtosis (BSESK). In this step, the parameter white noise amplitude weight NS(u) after adaptive optimization of the adaptive noise complete set empirical mode decomposition is best = 0.2, Noise addition times NR(v) best =500, the best eigenmode component group contains 13 modal components, such as Figure 4Next, the correlation coefficient between each intrinsic mode component in this group of optimal intrinsic mode components and the collected vibration signal is calculated. In this step, the correlation coefficient values ​​of the 13 components and the collected bearing vibration signal are calculated as follows: 0.7587, 0.4592, 0.4152, 0.3357, 0.2535, 0.1869, 0.1391, 0.0945, 0.0624, 0.0455, 0.0341, 0.0270, 0.0329, as shown in FIG. Figure 5 Then, the correlation coefficient is greater than the correlation threshold c v = 0.4, and the first three eigenmode components are reconstructed to obtain the reconstructed signal. The time domain diagram of the reconstructed signal is shown in Figure 6 As shown, the envelope spectrum is Figure 7 As shown, from Figure 6 and Figure 7 It is also impossible to observe the fault characteristic frequency and its harmonic multiples.

[0135] Furthermore, the input parameters of the eigenmode decomposition are optimized by the dual-domain Gini index (DDGI), and the reconstructed signal is decomposed into a set of optimal eigenmode components using the eigenmode decomposition with the optimal input parameters. In this step, the optimal input parameter of the eigenmode decomposition is n(p) best =5, k(g) best =6, L(h) best = 20, the best set of characteristic mode components after decomposition is as follows Figure 8 As shown in FIG. 1 , furthermore, according to the maximum criterion of the square envelope spectrum peak factor, the characteristic mode component containing the most bearing fault characteristics is extracted from this group of optimal characteristic mode components, that is, the third characteristic mode component Mode3 is used as the sensitive mode component; the envelope spectrum of the sensitive mode component is shown in FIG. Fig. 9 As shown in the figure, it can be observed that the fault harmonic frequency 2f o 、3f o 、4f o , but the fault characteristic frequency f cannot be observed o , and there are still 3 abnormal interference frequencies with obvious abnormal amplitudes, which seriously hinder the extraction of fault feature information.

[0136] Furthermore, the parameter-adaptive sparse maximum harmonic noise ratio deconvolution (SMHD) guided by the squared envelope harmonic noise ratio (SEHNR) is used to enhance the fault characteristics of the sensitive modal components and obtain the optimal feature enhancement component; among them, the global optimal parameter combination SL(w) of SMHD is best =297, MU(z) best =1.5×rms(S m )=1. The time domain diagram of the best feature enhancement component is as follows Fig.10As shown in the figure, the fault characteristics of periodic impact can be clearly observed from the time domain diagram. Finally, the envelope spectrum of the best feature enhancement component after Hilbert envelope demodulation is as follows Fig.11 As shown in the figure, it can be analyzed from the envelope spectrum that the strong noise signal is effectively suppressed, and the characteristic frequency f of the outer ring fault of the rolling bearing can be clearly analyzed. o , and the fault harmonic frequency 2f o 、3f o 、4f o , 5f o This verifies the effectiveness of the method of the present invention in adaptively extracting rolling bearing fault feature information under strong noise interference.

[0137] Although the present invention has been described above with reference to the embodiments, various modifications may be made thereto and parts thereof may be replaced by equivalents without departing from the scope of the present invention. In particular, as long as there is no structural conflict, the various features in the embodiments disclosed in the present invention may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for adaptively extracting rolling bearing fault features under strong noise interference, characterized in that: include: S10, inputting the rolling bearing vibration signal collected in real time; S20, decomposing the collected real-time vibration signal into a set of optimal intrinsic mode components using an adaptive noise complete set empirical mode decomposition improved based on 1.5-dimensional square envelope spectrum kurtosis BSESK; S30, calculating the correlation coefficient between each eigenmode component in the group of best eigenmode components and the collected vibration signal; S40, the correlation coefficient is greater than the correlation threshold c v The eigenmode components of are used to reconstruct the signal to obtain the reconstructed signal; S50, optimizing the input parameters of the eigenmode decomposition using the dual-domain Gini index DDGI, and decomposing the reconstructed signal using the eigenmode decomposition with the optimal input parameters into a set of optimal eigenmode components; S60, according to the maximum square envelope spectrum peak factor criterion, extracting the characteristic modal component containing the most bearing fault characteristics from the group of best characteristic modal components as the sensitive modal component; S70, using a square envelope harmonic noise ratio (SEHNR) guided parameter adaptive sparse maximum harmonic noise ratio deconvolution to enhance fault characteristics of sensitive modal components, and obtaining an optimal characteristic enhancement component; S80, performing Hilbert envelope demodulation on the best feature enhancement component and obtaining an envelope spectrum; S90, extracting rolling bearing fault feature information from the envelope spectrum of the best feature enhancement component; In step S20, the specific steps of decomposing the collected real-time vibration signal into a set of optimal intrinsic mode components and optimal IMFs by using the adaptive noise complete set empirical mode decomposition CEEMDAN improved based on 1.5-dimensional square envelope spectrum kurtosis BSESK are as follows: Step S201, obtain all parameter combinations of two input parameters that affect the decomposition performance of the adaptive noise complete set empirical mode decomposition CEEMDAN, the two input parameters are the white noise amplitude weight NS and the noise addition times NR, the parameter combination is expressed as [NS(u), NR(v)], and their values ​​are as follows: Where N* is a set of positive integers; Step S202: Input a set of parameter combinations [NS(u), NR(v)] into CEEMDAN, and decompose the collected real-time vibration signal into a set of intrinsic mode components IMFs, expressed as: [IMFs]=EXAMPLE(x,NS(u),NR(v),MaxIter c ) Among them, CEEDMAN(·) is the adaptive noise complete set empirical mode decomposition function operation in MATLAB, x represents the real-time vibration signal collected, NS(u) is the white noise amplitude weight in the current parameter combination, NR(v) is the number of noise additions in the current parameter combination, and MaxIter c is the number of iterations of CEEMDAN, MaxIter c Set to infinity, IMFs is a set of intrinsic mode components decomposed by CEEMDAN after applying the current parameter combination; Step S203, calculating the 1.5-dimensional square envelope spectrum kurtosis BSESK value of each intrinsic mode component IMF in the current group of IMFs, and recording the maximum BSESK value in the current group as the local maximum BSESK; Step S204: Repeat steps S202 and S203 until all local maximum BSESK values ​​corresponding to all parameter combinations of the two input parameters are calculated, and these BSESK values ​​are compared. The largest BSESK value is selected as the global maximum BSESK value, and the corresponding parameter combination is the global optimal parameter combination of CEEMDAN. The global optimal parameter combination of CEEMDAN is expressed as [NS(u) best NR(v) best ]; Step S205: Using [NS(u) best NR(v) best ]'s CEEMDAN decomposes the collected real-time vibration signal into a set of optimal intrinsic mode components, namely, optimal IMFs.

2. The method for adaptively extracting rolling bearing fault features under strong noise interference according to claim 1 is characterized in that: In step S20, the 1.5-dimensional square envelope spectrum kurtosis BSESK is a frequency domain index for evaluating the sparsity of bearing vibration signals constructed for adaptive noise complete set empirical mode decomposition. The larger the BSESK value, the better the sparsity of the signal, that is, the more periodic fault impact features contained in the signal. The calculation formula of BSESK is as follows: Wherein, s is the vibration time series signal, s(t) is the signal amplitude at time t, τ is the time delay, E(·) represents the mathematical expectation operation, j is the imaginary unit, f is the frequency, N is the number of time periods of the 1.5-dimensional square envelope spectrum, i is a positive integer and i=1, 2,···,N.

3. The method for adaptively extracting rolling bearing fault features under strong noise interference according to claim 1 is characterized in that: In step S30, the correlation coefficient C between each eigenmode component in the set of best eigenmode components and the collected vibration signal is calculated, and the calculation formula is as follows: Wherein, x represents the collected real-time vibration signal, y(r) is the rth intrinsic mode component in this group of optimal intrinsic mode components, C(r) is the correlation coefficient between the rth intrinsic mode component in this group of optimal intrinsic mode components and the collected real-time vibration signal, M is the length of the signal, m is the mth signal point, and m=1, 2,···,M.

4. The method for adaptively extracting rolling bearing fault features under strong noise interference according to claim 1 is characterized in that: In step S50, the dual-domain Gini index DDGI is a comprehensive sparsity index that considers both the time domain and the frequency domain of the vibration signal and is constructed for the optimization of the characteristic mode decomposition input parameters. The larger the DDGI value, the better the sparsity of the signal, that is, the more fault cycle impact characteristic information contained in the signal. The calculation formula of DDGI is as follows: Wherein, s is the vibration time series signal, M is the length of the signal, m=1,2,···,M, s(m)=(s(1),s(2),s(3),···,s(M)), the elements in s(m) are arranged from small to large, ||·||1 is the 1-norm operation, FFT(·) is the fast Fourier transform in MATLAB, abs(·) is the absolute value operation in MATLAB, FS represents the square envelope spectrum of the vibration time series signal s, Q is the length of the signal FS, q=1,2,···,Q, FS(q)=(FS(1),FS(2),FS(3),···,FS(Q)), the elements in FS(q) are arranged from small to large, α s is the sensitivity coefficient of the square envelope Gini index, β s is the sensitivity coefficient of the square envelope spectrum Gini index, α s ∈(0,1),β s ∈(0,1), and α s +β s =1.

5. The method for adaptively extracting rolling bearing fault features under strong noise interference according to claim 1 is characterized in that: In step S50, the input parameters of the eigenmode decomposition are optimized by the dual-domain Gini index DDGI, and the reconstructed signal is decomposed into a set of optimal eigenmode components by using the eigenmode decomposition with the optimal input parameters. The specific steps are as follows: Step S501, obtain all parameter combinations of three input parameters that affect the decomposition performance of the eigenmode decomposition, the three input parameters are the number of modes n of the final decomposition, the number of filters k initially set and the length L of the filter, and the three eigenmode decomposition input parameter combinations are expressed as [n(p), k(g), L(h)], and their values ​​are as follows: Where N* is a set of positive integers; Step S502: Decompose the reconstructed signal into a set of eigenmode components Modes using eigenmode decomposition with a set of parameter combinations [n(p), k(g), L(h)], Step S503, calculating the dual-domain Gini index DDGI value of each characteristic mode component Mode in the current group Modes, and recording the maximum DDGI value in the current group as the local maximum DDGI; Step S504, repeat steps S502 and S503 until all local maximum DDGI values ​​corresponding to all parameter combinations of the three input parameters of the eigenmode decomposition are calculated, compare these DDGI values, select the largest DDGI value as the global maximum DDGI value, and the corresponding eigenmode decomposition input parameter combination is the global optimal parameter combination of the eigenmode decomposition. The global optimal parameter combination of the eigenmode decomposition is expressed as [n(p) best , k(g) best 、L(h) best ]; Step S505: using [n(p) best , k(g) best 、L(h) best ]’s eigenmode decomposition decomposes the reconstructed signal into a set of optimal eigenmode components, namely, optimal Modes.

6. The method for adaptively extracting rolling bearing fault features under strong noise interference according to claim 1 is characterized in that: In step S60, the square envelope spectrum peak factor SE c The maximum criterion refers to calculating the square envelope spectrum peak factor SE of each component in the best eigenmode component group. c value, and select the largest SE c The modal component corresponding to the value is the sensitive modal component; the square envelope spectrum peak factor SE of a signal c The larger the value, the better the periodicity of the signal, the higher the signal-to-noise ratio, and the square envelope spectrum peak factor SE. c The calculation formula is as follows: Among them, N bearing Indicates the bearing speed in rpm; f r represents the theoretical rotation frequency of the bearing; f is the abscissa of the square envelope spectrum, i.e., the frequency; f s is the sampling frequency of the signal; s is the vibration time series signal; FFT(·) is the fast Fourier transform in MATLAB, abs(·) is the absolute value operation in MATLAB, FS represents the square envelope spectrum of the vibration time series signal s; max(·) is the maximum value operation in MATLAB, rms(·) is the root mean square value operation in MATLAB.

7. The method for adaptively extracting rolling bearing fault features under strong noise interference according to claim 1 is characterized in that: In step S70, the sensitive modal components are enhanced with parameter-adaptive sparse maximum harmonic noise ratio deconvolution SMHD guided by the square envelope harmonic noise ratio SEHNR. The specific steps of obtaining the optimal feature enhancement component are as follows: Step S701, obtain all parameter combinations of parameters to be adapted in the sparse maximum harmonic-to-noise ratio deconvolution SMHD, the parameters to be adapted are the filter length SL and the initial sparse threshold MU, the parameter combination is expressed as [SL(w), MU(z)], and their values ​​are as follows: Among them, S m is the sensitive modal component, rms(·) is the root mean square value operation in MATLAB, and N* is a positive integer set; Step S702: Input a set of parameter combinations [SL(w), MU(z)] into SMHD, and enhance the fault characteristics of the sensitive modal components to obtain the enhanced characteristic components, which are expressed as: [zmode]=SMHD(f s ,S m ,SL(w),Iter SMHD ,MU(z),T,Plotmode) Where SMHD(·) is the sparse maximum harmonic noise ratio deconvolution function in MATLAB, f s is the sampling frequency of the signal, S m is the sensitive modal component, SL(w) is the filter length in the current parameter combination, Iter SMHD is the maximum number of iterations of SMHD, which is set to 30, MU(z) is the initial sparse threshold in the current parameter combination, T is the signal failure period, Plotmode indicates whether to display the processed signal, and zmode indicates the feature enhancement component after SMHD processing with the current parameter combination; Step S703: Calculate the square envelope harmonic noise ratio SEHNR value of the current feature enhancement component, and record it as the local maximum SEHNR. The larger the SEHNR value, the richer the fault feature information in the signal. The calculation formula of SEHNR is as follows: Among them, s z (t) is the feature enhancement component, t is time, T g is the lag time when the autocorrelation function reaches the local maximum, i.e., the estimated fault period; Step S704, repeat steps S702 and S703 until all local maximum SEHNR values ​​corresponding to all parameter combinations of the parameters to be adapted are calculated, compare these SEHNR values, select the largest SEHNR value as the global maximum SEHNR value of SMHD, and the corresponding parameter combination is the global optimal parameter combination of SMHD. The global optimal parameter combination of SMHD is expressed as [SL(w) best MU(z) best ]; Step S705: Using [SL(w) best MU(z) best ]'s SMHD enhances the fault characteristics of the sensitive modal components and obtains the optimal feature enhancement component, namely the optimal zmode.