A method and device for diagnosing bearing faults

Abstract

An embodiment of the present invention provides a method and device for diagnosing bearing faults, including: obtaining the vibration signal of the bearing as the original signal; constructing a phase space matrix according to the original signal; performing dictionary learning processing on the phase space matrix to obtain the power spectrum estimation of each atom in the dictionary; constructing a sparse frequency response spectrum model according to the power spectrum estimation of each atom, and obtaining the central frequency response function through this model; filtering the original signal according to the central frequency response function to obtain a filtered reconstructed signal; obtaining the Hilbert envelope spectrum of the filtered reconstructed signal, and diagnosing the fault type of the bearing according to this envelope spectrum. The weak bearing fault diagnosis is realized, with identification robustness, and the early detection of bearing faults, fault feature extraction and fault type identification under strong interference conditions are achieved.

Description

Technical Field

[0001] The present invention belongs to the field of signal processing, and particularly relates to a method and device for diagnosing bearing faults. Background Art

[0002] With the continuous development of modern industry, the health management and maintenance of mechanical equipment have become extremely important. As an important part of modern mechanical equipment, rotating machinery such as bearings and gears plays an irreplaceable role. With the increasing complexity and harsh operating environment of modern mechanical equipment, rotating components such as bearings and gears are prone to failures. These failures will result in high maintenance costs and possible casualties. The complex mechanical structure and harsh operating environment of modern equipment cause the signals collected by accelerometers to contain noise and other irrelevant random components. Most methods cannot achieve a good balance between the robustness against noise and generalization. Extracting bearing fault features from these interferences and making accurate diagnoses has become a difficult task. Summary of the Invention

[0003] The purpose of the embodiments of the present invention is a method and device for diagnosing bearing faults. Inspired by the receiving characteristics of the animal retina, the sparse representation theory (SRT) of signals is proposed as a new theory and has quickly received extensive attention in the academic community. Since the essence of the sparse constraint in the sparse representation theory is highly related to fault models such as bearings, but the dictionary learning method has its own disadvantages in terms of resisting noise interference, the vast majority of methods cannot achieve a good balance between the robustness against noise and generalization.

[0004] To achieve the above purpose, the embodiments of the present invention provide a method for diagnosing bearing faults, including: obtaining the vibration signal of the bearing as the original signal; constructing a phase space matrix according to the original signal; performing dictionary learning processing on the phase space matrix to obtain the power spectrum estimation of each atom in the dictionary; constructing a sparse frequency response spectrum model according to the power spectrum estimation of each atom, and obtaining the central frequency response function through this model; filtering the original signal according to the central frequency response function to obtain a filtered reconstruction signal; obtaining the Hilbert envelope spectrum of the filtered reconstruction signal, and diagnosing the fault type of the bearing according to this envelope spectrum.

[0005] Optionally, the vibration signal of the bearing is obtained by a vibration acceleration sensor.

[0006] Optionally, the constructing the phase space matrix according to the original signal includes forming the phase space matrix Y according to the segmentation operator of Matrix-Stride, and this phase space matrix is a two-dimensional matrix:

[0007]

[0008]

[0009] Among them, stride is the sliding step size, and f s is the sampling frequency of the original signal, B f is the theoretical fault characteristic frequency of the bearing, η is the scale factor, y is the original signal, y = [y1, y2, … y N , N is the signal length, and y1, y2, … y N are the sampling points of the original signal.

[0010] Optionally, the dictionary learning process for the phase space matrix includes obtaining a sparse coefficient matrix by performing dictionary learning on the phase space matrix through the K-SVD algorithm:

[0011]

[0012] Among them, Y is the phase space matrix, D is the sparse dictionary, and S is the sparse coefficient matrix, and the sparse coefficient matrix is used to update the dictionary column by column.

[0013] Optionally, obtaining the power spectrum estimate of each atom in the dictionary includes:

[0014]

[0015] Among them, G M,l (ω) is the periodogram, U is the normalization factor, ω is the window function, N is the length of the atom, L is the number of estimated segments, and M = N / L.

[0016] Optionally, constructing a sparse frequency response spectrum model based on the power spectrum estimate of each atom, and obtaining the center frequency response function through this model, includes: normalizing the power spectrum estimate of each atom; sorting all atom power spectra according to the maximum value in the power spectrum estimate of each atom to obtain the atom power spectrum estimate index; constructing a sparse frequency response spectrum model according to the atom power spectrum estimate index; and obtaining the center frequency response function according to the sparse frequency response spectrum model.

[0017] Optionally, the normalizing the power spectrum estimate of each atom includes:

[0018]

[0019] Among them, is the power spectrum estimate of each atom.

[0020] Optionally, sorting all atom power spectra according to the maximum value in the power spectrum estimate of each atom to obtain the atom power spectrum estimate index θ:

[0021]

[0022] Among them, K is the number of atoms in the dictionary, is the power spectrum estimate of each atom.

[0023] Optionally, constructing the sparse frequency response spectrum model G according to the atom power spectrum estimation index includes:

[0024]

[0025] Among them, θ K is the atom power spectrum estimation index, k = 1, 2... k,

[0026] is the power spectrum estimate of each atom.

[0027] Optionally, obtaining the central frequency response function according to the sparse frequency response spectrum model G includes:

[0028]

[0029] Among them, is the power spectrum of atom a in G j the power spectrum,

[0030] is the central frequency response function,

[0031] class c 1 and class c n are respectively the start and end indices of class in G,

[0032] n is the number of atom power spectra belonging to class.

[0033] Optionally, filtering the original signal according to the central frequency response function to obtain a filtered reconstruction signal includes: performing threshold denoising on the central frequency response function; calculating the frequency-domain amplitude spectrum of the original signal; calculating the inner product of the central frequency response function curve and the amplitude spectrum function curve; performing inverse Fourier transform on the inner product result to obtain the enhanced signal after filtered reconstruction.

[0034] Optionally, performing threshold denoising on the central frequency response function includes: is the central frequency response function,

[0035]

[0036] Among them, μ is the weighting factor.

[0037] Optionally, calculating the frequency-domain amplitude spectrum of the original signal includes:

[0038]

[0039] The inner product Ψ(ω) of the calculated center frequency response function curve and amplitude spectrum function curve includes:

[0040]

[0041] The inverse Fourier transform of the inner product result to obtain the enhanced signal after filtering and reconstruction includes: where y(t) is the original signal and A(ω) is the amplitude spectrum function.

[0042] Optionally, diagnosing the fault type of the bearing according to the envelope spectrum includes: obtaining the fault characteristic frequency of the bearing according to the envelope spectrum for diagnosing the fault type of the bearing; the fault type of the bearing includes at least one of outer ring fault, inner ring fault, rolling element fault and cage fault.

[0043] Correspondingly, an embodiment of the present invention further provides a diagnostic device for weak faults of a bearing, which is characterized by including: a signal acquisition module for acquiring the vibration signal of the bearing as the original signal; a processing module for processing the original signal to obtain the Hilbert envelope spectrum, including: constructing a phase space matrix according to the original signal; performing dictionary learning processing on the phase space matrix to obtain the power spectrum estimation of each atom in the dictionary; constructing a sparse frequency response spectrum model according to the power spectrum estimation of each atom, and obtaining the center frequency response function through the model; filtering the original signal according to the center frequency response function to obtain a filtered and reconstructed signal; obtaining the Hilbert envelope spectrum of the filtered and reconstructed signal; a diagnosis module for diagnosing the fault type of the bearing through the envelope spectrum.

[0044] Optionally, the performing dictionary learning processing on the phase space matrix includes performing dictionary learning processing on the phase space matrix through the K-SVD algorithm to obtain a sparse coefficient matrix:

[0045]

[0046] where Y is the phase space matrix, D is the sparse dictionary, and S is the sparse coefficient matrix, and the sparse coefficient matrix is used to update the dictionary column by column.

[0047] Optionally, obtaining the power spectrum estimation of each atom in the dictionary includes:

[0048]

[0049] where G M,l(ω) is the periodogram, U is the normalization factor, ω is the window function, N is the length of the atom, L is the estimated number of segments, and M = N / L.

[0050] Optionally, constructing a sparse frequency response spectrum model based on the power spectrum estimation of each atom, and obtaining the central frequency response function through this model, includes: performing normalization processing on the power spectrum estimation of each atom; sorting all atom power spectra according to the maximum value in the power spectrum estimation of each atom to obtain the atom power spectrum estimation index; constructing a sparse frequency response spectrum model according to the atom power spectrum estimation index; and obtaining the central frequency response function according to the sparse frequency response spectrum model.

[0051] Optionally, filtering the original signal according to the central frequency response function to obtain a filtered reconstructed signal, includes: performing threshold denoising on the central frequency response function; calculating the frequency domain amplitude spectrum of the original signal; calculating the inner product of the central frequency response function curve and the amplitude spectrum function curve; and performing inverse Fourier transform on the inner product result to obtain the enhanced signal after filtered reconstruction.

[0052] Optionally, diagnosing the fault type of the bearing according to the envelope spectrum, includes: obtaining the fault characteristic frequency of the bearing according to the envelope spectrum for diagnosing the fault type of the bearing; the fault type of the bearing includes at least one of outer ring fault, inner ring fault, rolling element fault, and cage fault.

[0053] Through the above technical solutions, the bearing weak fault diagnosis method of the signal sparse structure frequency analysis model using dictionary learning in the present invention enhances the weak bearing fault diagnosis and identification robustness of the envelope spectrum method, and realizes early detection of bearing faults, fault feature extraction, and fault type identification under strong interference conditions.

[0054] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation part. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification. Together with the following specific implementation, they are used to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0056] Figure 1 and Figure 2 is a schematic flow chart of the bearing fault diagnosis method of the present invention;

[0057] Figure 3 is the time domain waveform of the original bearing vibration signal in an embodiment of the present invention;

[0058] Figure 4 is a sparse frequency structure diagram in an embodiment of the present invention;

[0059] Figure 5 It is the graph of the changing trend of the maximum value of the sparse frequency structure diagram of an embodiment of the present invention;

[0060] Figure 6 It is the differential diagram of the maximum value frequency change of an embodiment of the present invention;

[0061] Figure 7 It is the curve graph of the central frequency response function of an embodiment of the present invention;

[0062] Figure 8 It is the time-domain waveform diagram of the filtered signal of an embodiment of the present invention;

[0063] Figure 9 It is the envelope spectrum diagram of the filtered signal of an embodiment of the present invention. Specific Embodiment

[0064] The following will describe in detail the specific embodiments of the embodiments of the present invention with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention.

[0065] Figure 1 It is the flow schematic diagram of the bearing fault diagnosis method of the present invention.

[0066] Step S101 is to obtain a vibration signal. The vibration signal is the vibration signal of the bearing to be measured. According to a preferred embodiment, the vibration signal of the bearing is obtained through a vibration acceleration sensor. Obtaining the vibration signal of the bearing is the original signal.

[0067] Step S102 is to construct a phase space matrix. The phase space matrix is based on the prior knowledge of the fault frequency, and the theoretical fault frequency is known in advance. Preferably, the present invention uses an algorithm based on K-SVD to perform dictionary learning processing on the vibration signal. K-SVD is a classic dictionary training algorithm. According to the principle of minimum error, the error term is decomposed by SVD, and the decomposition term that makes the error minimum is selected as the updated dictionary atom and the corresponding atom coefficient. Through continuous iteration, an optimized solution is obtained. The general application object of the K-SVD method is two-dimensional signals. Therefore, in order to process one-dimensional vibration signals and balance the accuracy and computational efficiency of the algorithm, the present invention proposes a segmentation operator called Matrix-Stride (MS) to form a two-dimensional phase space matrix. Preferably, according to the original signal, a phase space matrix is constructed, including forming a phase space matrix Y according to the segmentation operator of Matrix-Stride. This phase space matrix is a two-dimensional matrix:

[0068]

[0069]

[0070] Among them, stride is the sliding step size, and f s is the sampling frequency of the original signal, B f is the theoretical fault characteristic frequency of the bearing, η is the scale factor, y is the original signal, and y = [y1, y2, … y N , N is the signal length, and y1, y2, … y N are the sampling points of the original signal. is the floor operation (Floor).

[0071] Step S103 is to perform dictionary learning processing to obtain the power spectrum estimation of each atom. By performing dictionary learning processing on the phase space matrix, the power spectrum estimation of each atom in the dictionary is obtained. The performing dictionary learning processing on the phase space matrix includes obtaining a sparse coefficient matrix by performing dictionary learning processing on the phase space matrix through the K-SVD algorithm:

[0072]

[0073] Among them, Y is the phase space matrix, D is the sparse dictionary, S is the sparse coefficient matrix, and the sparse coefficient matrix is used to update the dictionary column by column. The K-SVD dictionary learning algorithm is an efficient dictionary learning algorithm. It uses the orthogonal matching pursuit algorithm (OMP) to implement an efficient sparse coding process, and then synchronously updates the dictionary and the sparse coefficients through singular value decomposition (SVD) during the dictionary learning process. The solution process of K-SVD can be represented by equations (1) and (2):

[0074]

[0075]

[0076] The Welch power spectrum estimation of each atom a k in the calculated dictionary A The sparse representation of the signal is closely related to the signal filtering operation in the frequency domain. Let the sparse representation signal related to the rolling bearing raceway fault be y sr , then y sr can be expressed as a combination of the atoms and a series of sparse coefficients in equation (3). The corresponding frequency domain signal can be obtained through the Fourier transform in equation (4).

[0077]

[0078]

[0079] The frequency-domain form of the signal can be sparsely represented by the frequency-domain transformation of the atoms in the sparse dictionary. Thus, the local frequency response characteristics of the signal can be obtained by calculating the power spectral estimate of the atoms in the dictionary. To avoid local interference and obtain the large-scale power spectral estimate of the atoms, the Welch power spectral estimation algorithm is selected to calculate the power spectral estimate of each atom in the dictionary. The Welch algorithm is an improvement of the periodogram method. The obtaining of the power spectral estimate of each atom in the dictionary includes:

[0080]

[0081] where G M,l (ω) is the periodogram, U is the normalization factor, ω is the window function, N is the length of the atom, L is the number of segments estimated, and M = N / L.

[0082] Step S104 is to construct a sparse frequency response spectrum model and obtain the central frequency response function. The central frequency response function is the frequency response function, which is the quotient obtained by dividing the cross-power spectrum function by the self-power spectral estimate. The frequency response function is a complex function, which is the description of the dynamic characteristics of the measured system in the frequency domain, that is, the description of the transmission characteristics of the measured system itself for the input signal in the frequency domain. The frequency response function is of special importance for the dynamic characteristic test of the structure. It includes: normalizing the power spectral estimate of each atom; sorting all the atom power spectra according to the maximum value in the power spectral estimate of each atom to obtain the atom power spectral estimate index; constructing a sparse frequency response spectrum model according to the atom power spectral estimate index; and obtaining the central frequency response function with the strongest repetitive characteristics according to the frequency response pattern with repetitive characteristics characterized by the sparse frequency response spectrum model. Since this frequency response function characterizes the component with the strongest repetitive characteristics in the signal, which conforms to a typical characteristic reflected in the vibration signal after the bearing failure, the central frequency response function is the optimal central frequency response function. Determining the atoms related to the fault impact characteristics is the key to the fault diagnosis method based on vibration signals. We propose the SFRSM to analyze the quasi-periodic characteristics in the signal in order to identify the atoms related to bearing failure in the signal. The SFRSM is the sparse frequency response spectrum model, which is used to characterize the fault information. The normalizing the power spectral estimate of each atom includes:

[0083]

[0084] where is the power spectral estimate of each atom.

[0085] The sorting of all the atom power spectra according to the maximum value in the power spectral estimate of each atom to obtain the atom power spectral estimate index θ:

[0086]

[0087] Among them, K is the number of atoms in the dictionary, is the power spectrum estimation of each atom.

[0088] Constructing the sparse frequency response spectrum model G according to the atomic power spectrum estimation index includes:

[0089]

[0090] Among them, θ K is the atomic power spectrum estimation index, k = 1, 2... k, is the power spectrum estimation of each atom.

[0091] Obtaining the central frequency response function according to the sparse frequency response spectrum model G includes:

[0092]

[0093] Among them, is the power spectrum of atom a in G j the power spectrum,

[0094] is the central frequency response function, class c 1 and class c n are respectively the start and end indexes of the atoms belonging to class in G, and n is the number of atomic power spectra belonging to class. is an optimal response function derived from the response spectrum model analysis.

[0095] Step S105 is to filter the original signal according to the central frequency response function to obtain the filtered and reconstructed signal, including: performing threshold denoising on the central frequency response function; calculating the frequency-domain amplitude spectrum of the original signal; calculating the inner product of the central frequency response function curve and the amplitude spectrum function curve; performing inverse Fourier transform on the inner product result to obtain the enhanced signal after filtering and reconstruction.

[0096] Performing threshold denoising on the central frequency response function includes: is the central frequency response function,

[0097]

[0098] Among them, μ is the weighting factor.

[0099] Calculating the frequency-domain amplitude spectrum of the original signal, is the Fourier transform result of the original signal, including:

[0100]

[0101] Calculate the Fourier transform of the original signal by this formula.

[0102] The calculation of the inner product Ψ(ω) of the center frequency response function curve and the amplitude spectrum function curve includes:

[0103] Obtain the inner product of the center frequency response function and the Fourier transform of the original vibration signal by this formula. Finally, perform a Fourier transform on the inner product result to restore the time-domain waveform of the filtered signal.

[0104] The inverse Fourier transform of the inner product result is used to obtain the enhanced signal after filter reconstruction, including: Where y(t) is the original signal and A(ω) is the amplitude spectrum function.

[0105] Step S106 is to obtain the Hilbert envelope spectrum, diagnose the fault type of the bearing according to the envelope spectrum, and find the Hilbert envelope spectrum of the filtered and reconstructed signal and perform fault diagnosis and fault type identification based on the theoretical fault characteristic frequency. The envelope spectrum is: after performing a hilbert transform on the signal, then taking the extreme values, then taking the envelope of the one-dimensional data obtained after taking the extreme values, and performing an FFT transform on the envelope signal to obtain the data. The envelope spectrum (with frequency on the abscissa and amplitude on the ordinate) is sensitive to faults of impact events. The distribution of the amplitudes of each frequency in the envelope spectrum diagram is different from that of the spectrum diagram. The amplitudes of the fault characteristic frequencies in the spectrum diagram are small, while the amplitudes of the fault characteristic frequencies in the envelope spectrum diagram are very high and are easy to identify. Therefore, compared with spectrum analysis, envelope spectrum analysis eliminates unnecessary frequency interference and can more prominently highlight the fault characteristic frequencies. It is easier to judge the fault types of rolling bearings according to the envelope spectrum diagram. Diagnosing the fault type of the bearing according to the envelope spectrum includes: obtaining the fault characteristic frequency of the bearing according to the envelope spectrum for diagnosing the fault type of the bearing; the fault type of the bearing includes at least one of outer race fault, inner race fault, rolling element fault, and cage fault.

[0106] Figure 2 is Figure 1Schematic diagram of the process of a specific bearing fault diagnosis method, including: dictionary learning, SFRSM model, AFEF filtering, and characteristic frequency analysis. The present invention first introduces a dictionary learning algorithm based on K-SVD (generalized K-means clustering) to capture the internal structure information of the fault signal, and then proposes a sparse frequency response spectrum model (SFRSM) to characterize the fault information. On the basis of the sparse frequency response model, a sparse frequency structure graph (SFSG) is used to extract the fault-related frequency response patterns. Subsequently, the proposed frequency response function editing filtering method (AFEF) is used to filter the original signal to obtain the relevant fault signals masked by noise. Finally, envelope spectrum analysis is performed on the filtered signal to achieve fault feature extraction and fault type identification. To enhance the weak bearing fault diagnosis and identification robustness of the traditional envelope spectrum method and realize the early detection of bearing faults under strong interference conditions.

[0107] The present invention provides a specific implementation case of a bearing fault diagnosis method, including the following steps:

[0108] Step 1: Use a vibration acceleration sensor to collect bearing vibration signals; as Figure 3 shown, the signal sampling rate is 51200Hz / s, and 0.5 seconds of data is taken as the analysis object of the method of the present invention.

[0109] Step 2: Construct the phase space matrix of the original signal;

[0110] Step 2.1: Set the sliding step size stride according to Equation (2), where fs is the signal sampling frequency, which is 51200Hz here, the bearing fault characteristic frequency B f is 270Hz, and the shaft rotation frequency is 50Hz. η is taken as 0.7. It is calculated that d = 270 and stride = 189.

[0111]

[0112]

[0113] Step 2.2: Construct the phase space matrix of the signal according to Equation (1), y is the original signal, N is the signal length, and y1, y2,... y N are the sampling points of the signal.

[0114] Step 3: Use the K-SVD algorithm to perform dictionary learning processing on the phase space matrix obtained in Step 2. Loop Steps 3.1 to 3.3 a total of 50 times until Equation (3) is satisfied, and ε is taken as 0.25.

[0115]

[0116] Step 3.1: Randomly select 200 columns of the phase space matrix Y as the initial dictionary D0.

[0117] Step 3.2: Fix the dictionary D. According to Equation (4), use the orthogonal matching pursuit algorithm (OMP) to perform sparse decomposition on the phase space matrix Y and solve the sparse coefficient matrix S.

[0118]

[0119] Step 3.3: Dictionary update. Use the obtained sparse coefficient matrix S to update the dictionary column by column.

[0120] Step 3.3.1: Calculate the error matrix using Equation (5).

[0121]

[0122] Step 3.3.2: Calculate using Equations (6) and (7).

[0123]

[0124]

[0125] Step 3.3.3: Perform SVD decomposition on As shown in Equation (8).

[0126]

[0127] Step 3.3.4: Update the dictionary and the coefficient As shown in Equation (9), where column1() is to extract the first column of the matrix.

[0128]

[0129]

[0130] Step 4: Calculate the Welch power spectrum estimate of each atom a in the dictionary A obtained in Step 3. k as shown in Equation (10). Where G M,l (ω) is the l-th segmented modified periodogram, U is the normalization factor, and ω is the window function. Take N = 25600, L = 20, M = 1280.

[0131]

[0132] Step 5: Construct a sparse frequency response spectrum model based on the power spectrum estimates of each atom obtained in Step 4.

[0133] Step 5.1: Normalize the power spectrum curve of each atom as shown in Equation (11).

[0134]

[0135] Step 5.2: Sort all atom power spectra according to the maximum value in the power spectrum estimation of each atom and obtain the sorting index θ. Specifically, it is shown in Equation (12).

[0136]

[0137] where K is the number of atoms in the dictionary, is the power spectrum estimation of each atom.

[0138] Step 5.3: Construct a sparse frequency structure diagram according to the atom power spectrum estimation index θ. Specifically, it is shown in Equation (13). The constructed sparse frequency structure diagram is as Figure 4 shown.

[0139]

[0140] Step 5.4: According to the sparse frequency structure Figure 3 , draw a trend diagram of the maximum value of the structure diagram as Figure 5 shown. The vertical coordinate PF is the frequency of the sparse dictionary response pattern, and the horizontal coordinate is the atom power spectrum estimation index θ. At the same time, draw a Figure 5 differential diagram according to the frequency corresponding to the maximum value of each frequency response function, as Figure 6 shown. The vertical coordinate CPF is the central response frequency, and the horizontal coordinate is the atom power spectrum estimation index θ. Combine Figure 5 , Figure 6 Set the range of a class of repeated frequency response patterns class1 as j = 23, n = 97.

[0141] Step 5.5: Obtain the central frequency response function from Equation (14) according to the range of class1 (j = 23, n = 97) as Figure 7 .

[0142]

[0143] Step 6: Filter and enhance the original signal according to the central frequency response function obtained in Step 5.

[0144] Step 6.1: Perform threshold denoising on the central frequency response function, specifically as shown in Equation (15). Where μ takes 0.3. The denoised frequency response function is as Figure 6 shown.

[0145]

[0146] Step 6.2: Calculate the frequency-domain amplitude spectrum of the original signal y as shown in Equation (16). Where A(ω) is the amplitude spectrum function.

[0147]

[0148] Step 6.3: Calculate the inner product Ψ(ω) between the central frequency response function curve and the signal amplitude spectrum function curve.

[0149] As shown in Equation (17). Where upsample() is the upsampling operation and ⊙ is the inner product operation.

[0150]

[0151] Step 6.4: Perform the inverse Fourier transform on the inner product frequency function Ψ(ω) according to Equation (18) to obtain the enhanced signal after filtering and reconstruction As Figure 7 shown.

[0152]

[0153] Step 7: Obtain the Hilbert envelope spectrum of the filtered and reconstructed signal as Figure 8 and Figure 9 , and the fault characteristic frequency 270Hz of the bearing inner ring and its multiple frequencies and sideband components can be identified from the envelope spectrum. Thus, it can be diagnosed that the bearing has an inner ring fault.

[0154] In order to overcome the defects of the traditional sparse DL method, which has low accuracy in extracting fault bearing signal features under strong background noise interference and poor adaptability in fault diagnosis, the present invention proposes a fault feature extraction and diagnosis method based on a sparse dictionary structure frequency response model. First, the K-SVD sparse dictionary learning method is used to capture the global features of the signal. According to the proposed sparse frequency response spectrum model (SFSRSM), the frequency-domain power spectrum of each atom is calculated to obtain the relevant frequency correspondence and all atoms are rearranged and modeled. Then, the best core quasi-periodic repeated feature frequency response pattern (RFP) is obtained according to the sparse frequency structure diagram. According to the obtained frequency response pattern, the proposed adaptive frequency response editing filter (AFEF) algorithm is used to filter the original signal to ensure the integrity of the filtered signal. Finally, envelope spectrum analysis is performed on the filtered signal to obtain the typical feature spectrum of the bearing fault, and the bearing is diagnosed accordingly through the fault characteristic frequency. The specific implementation process and results of the present invention prove that this method has strong robustness in dealing with actual weak bearing fault signals, can effectively improve the accuracy of bearing weak fault diagnosis, and is applicable to the fault diagnosis of rotating machinery.

[0155] For the specific implementation details and effects of a bearing fault diagnosis device provided by an embodiment of the present invention, reference may be made to the foregoing embodiments, and details will not be repeated here.

[0156] The optional implementation manners of the embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above implementation manners. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention.

[0157] In addition, it should be noted that, among the various specific technical features described in the above specific implementation manners, they can be combined in any appropriate manner without conflict. To avoid unnecessary repetition, the embodiments of the present invention will not separately describe various possible combination manners.

[0158] Those skilled in the art can understand that all or part of the steps in implementing the methods of the above embodiments can be completed by instructing relevant hardware through a program, and this program is stored in a storage medium, including several instructions for causing a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disc that can store program codes.

[0159] In addition, any combination can be made among various different implementation manners of the embodiments of the present invention, as long as it does not violate the idea of the embodiments of the present invention, and it should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for diagnosing bearing faults, characterized in that, Including: Obtain the vibration signal of the bearing as the original signal; Construct a phase space matrix according to the original signal; Perform dictionary learning processing on the phase space matrix to obtain the power spectrum estimation of each atom in the dictionary; Construct a sparse frequency response spectrum model according to the power spectrum estimation of each atom, and obtain the central frequency response function through this model; Filter the original signal according to the central frequency response function to obtain a filtered and reconstructed signal; Obtain the Hilbert envelope spectrum of the filtered and reconstructed signal, and diagnose the fault type of the bearing according to this envelope spectrum; Obtaining the power spectral estimate of each atom in the dictionary , including: Among them, is the th periodogram after segmented modification, a k is an atom, is ω complex exponential signal of U is a normalization factor, ω is a window function, N is the length of the atom, L is the estimated number of segments, and M = N / L.

2. The method according to claim 1, wherein: Obtain the vibration signal of the bearing through a vibration acceleration sensor.

3. The method according to claim 1, wherein The constructing a phase space matrix according to the original signal includes forming a phase space matrix Y according to the segmentation operator of Matrix-Stride, and this phase space matrix is a two-dimensional matrix: ， ， where stride is the sliding step size, f s is the sampling frequency of the original signal, B f is the theoretical fault characteristic frequency of the bearing, η is the scaling factor, and N is the length of the atom. y is the original signal, y = [y1, y2, … y N , N is the signal length, y1, y2, … y N are the sampling points of the original signal. d is the value obtained by rounding down the operation.

4. The method according to claim 1, characterized in that The performing dictionary learning processing on the phase space matrix includes performing dictionary learning processing on the phase space matrix through the K-SVD algorithm to obtain a sparse coefficient matrix: where Y is the phase space matrix, D is the sparse dictionary, N is the length of the atom, S is a sparse coefficient matrix, and the sparse coefficient matrix is used to update the dictionary column by column.

5. The method according to claim 1, wherein The constructing a sparse frequency response spectrum model according to the power spectrum estimation of each atom and obtaining the central frequency response function through this model includes: Perform normalization processing on the power spectrum estimation of each atom; Sort all atom power spectra according to the maximum value in the power spectrum estimation of each atom to obtain the atom power spectrum estimation index; Construct a sparse frequency response spectrum model according to the atom power spectrum estimation index; Obtain the central frequency response function according to the sparse frequency response spectrum model.

6. The method according to claim 5, characterized in that, The performing normalization processing on the power spectrum estimation of each atom includes: ， wherein, is the power spectrum estimate for each atom, is the minimum value of, is the maximum value of.

7. The method according to claim 5, characterized in that, Sort all atomic power spectrum estimates according to the maximum value in the power spectrum estimate of each atom to obtain an atomic power spectrum estimate index : ， where K is the number of atoms in the dictionary, is the gamma function, sort() is the function for sorting elements, and max() is the function for taking the maximum value. Power spectrum estimation for each atom.

8. The method according to claim 5, characterized in that, The constructing a sparse frequency response spectrum model G according to the atom power spectrum estimation index includes: ， Among them, is the atomic power spectrum estimation index, k = 1, 2... k, Power spectrum estimation for each atom.

9. The method according to claim 5, characterized in that, The obtaining the central frequency response function according to the sparse frequency response spectrum model G includes: Among them, is the power spectrum of the atoms in G , is the center frequency response function, and are the starting and ending indices of the class in G respectively, n is the number of atom power spectra belonging to class.

10. The method according to claim 5, characterized in that The filtering the original signal according to the central frequency response function to obtain a filtered and reconstructed signal includes: Perform threshold denoising on the central frequency response function; Calculate the frequency domain amplitude spectrum of the original signal; Calculate the inner product of the central frequency response function curve and the amplitude spectrum function curve; Perform inverse Fourier transform on the inner product result to obtain the enhanced signal after filtering and reconstruction.

11. The method according to claim 10, wherein Performing threshold denoising on the center frequency response function includes: is the center frequency response function, Wherein, μ is a weighting factor.

12. The method according to claim 10, wherein: The calculating the frequency domain amplitude spectrum of the original signal includes: The inner product of the calculated center frequency response function curve and the amplitude spectrum function curve , including: Performing an inverse Fourier transform on the inner product result to obtain the enhanced signal after filtering and reconstruction includes: , where y(t) is the original signal, is the amplitude spectrum function, upsample() is the upsampling operation, is the inner product operation, is the center frequency response function.

13. The method according to claim 1, characterized in that The diagnosing the fault type of the bearing according to this envelope spectrum includes: Obtain the fault characteristic frequency of the bearing according to the envelope spectrum for diagnosing the fault type of this bearing; The fault type of the bearing includes at least one of outer ring fault, inner ring fault, rolling element fault and cage fault.

14. A diagnostic device for weak faults of bearings, characterized in that, Including: A signal acquisition module for obtaining the vibration signal of the bearing as the original signal; A processing module for processing the original signal to obtain the Hilbert envelope spectrum, including: Constructing a phase space matrix according to the original signal; Performing dictionary learning processing on the phase space matrix to obtain the power spectrum estimation of each atom in the dictionary; Construct a sparse frequency response spectrum model based on the power spectrum estimation of each atom, and obtain the central frequency response function through this model; Filter the original signal according to the central frequency response function to obtain a filtered and reconstructed signal; obtain the Hilbert envelope spectrum of this filtered and reconstructed signal; A diagnosis module for diagnosing the fault type of the bearing through the envelope spectrum; Obtaining the power spectrum estimate of each atom in the dictionary , including: Among them, is the th periodogram after segmented modification, a k is an atom, is ω complex exponential signal of U is a normalization factor, ω is a window function, N is the length of the atom, L is the estimated number of segments, and M = N / L.

15. The device according to claim 14, wherein The dictionary learning process for the phase space matrix includes performing dictionary learning on the phase space matrix through the K-SVD algorithm to obtain a sparse coefficient matrix: ， where Y is the phase space matrix and N is the length of the atom, D is a sparse dictionary, S is a sparse coefficient matrix, and the sparse coefficient matrix is used to update the dictionary column by column.

16. The device according to claim 14, wherein, The construction of the sparse frequency response spectrum model based on the power spectrum estimation of each atom and the obtaining of the central frequency response function through this model include: Normalize the power spectrum estimation of each atom; Sort all atom power spectra according to the maximum value in the power spectrum estimation of each atom to obtain the atom power spectrum estimation index; Construct a sparse frequency response spectrum model according to the atom power spectrum estimation index; Obtain the central frequency response function according to the sparse frequency response spectrum model.

17. The device according to claim 14, characterized in that, The filtering of the original signal according to the central frequency response function to obtain a filtered and reconstructed signal includes: Perform threshold denoising on the central frequency response function; Calculate the frequency domain amplitude spectrum of the original signal; Calculate the inner product of the central frequency response function curve and the amplitude spectrum function curve; Perform an inverse Fourier transform on the inner product result to obtain the enhanced signal after filtering and reconstruction.

18. The device according to claim 14, characterized in that, The diagnosis of the fault type of the bearing according to the envelope spectrum includes: Obtain the fault characteristic frequency of the bearing according to the envelope spectrum for diagnosing the fault type of the bearing; The fault types of the bearing include at least one of outer ring fault, inner ring fault, rolling element fault, and cage fault.

Images