Bearing fault feature extraction method based on parameter optimization SMHD-FMD

Abstract

The invention provides a bearing fault feature extraction method based on parameter optimization SMHD-FMD, and the method comprises the following steps: S1, collecting a vibration signal of a rolling bearing, carrying out the pre-parameter optimization of the SMHD through employing the Myceland algorithm, and carrying out the noise reduction preprocessing of the SMHD through employing the optimal parameter SMHD; s2, performing PO-FMD on the denoised signal to obtain an IMF component; s3, selecting an optimal component according to the kurtosis of each component; and S4, carrying out envelope demodulation analysis on the optimal component, and extracting a characteristic frequency to realize fault diagnosis. According to the method, an optimal parameter SMHD is used for carrying out noise reduction preprocessing on an original signal to highlight pulse characteristics, so that characteristic modal decomposition is carried out on the original signal to obtain a modal component, then kurtosis is used for selecting an optimal component, finally, fault characteristic frequency is extracted by calculating an envelope spectrum of the optimal component, and fault diagnosis of the rolling bearing is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a bearing fault feature extraction method based on parameter optimized SMHD-FMD. Background Art

[0002] Rolling bearings have been widely used in rotating machinery in many fields such as electric power, petrochemical, aerospace, etc. due to their advantages such as strong load-bearing capacity, easy maintenance, and good starting performance. Therefore, conducting research on rolling bearing fault diagnosis is of great practical significance for improving the performance of mechanical equipment, reducing energy consumption, and extending service life.

[0003] Due to the complexity of the production environment, the impact signal of bearing fault is easily submerged by noise, so it is very necessary to pre-process the collected original signal to reduce noise. Blind source deconvolution can reduce noise interference and enhance the signal impact characteristics. GeoffL et al. proposed multi-point optimization minimum entropy deconvolution, using defined weights and impact position target vectors to restore the submerged impact characteristics in the signal. The quality of the restoration effect mainly depends on the selection of the prior parameter fault cycle. In order to overcome this defect, Miao et al. took a different approach and measured the fault cycle by the harmonic noise ratio, and then proposed sparse maximum harmonics-to-noise-ratio deconvolution (SMHD). However, the quality of SMHD noise reduction is subject to the selection of the pre-parameter filter length. Due to the influence of the working environment and shaft vibration, the bearing vibration signal often has non-stationary and nonlinear characteristics, which makes it difficult to extract fault features from it. In order to solve this problem, the signal decomposition algorithm is proposed. The decomposition algorithms that have been widely studied at present include EMD, EEMD, VMD, EWT, etc., but they all have certain limitations in practical applications. EMD has serious modal aliasing and endpoint effects. Although EEMD effectively suppresses modal aliasing and endpoint effects by adding adaptive noise during the decomposition process, it produces defects such as residual noise and large signal reconstruction errors. VMD is a non-recursive decomposition algorithm developed based on Wiener filtering, but the quality of the decomposition effect is limited by the pre-selected parameters of the mode number and the penalty factor, and there is a problem of sudden signal mutation in the smoothing stage. EWT uses scale space to divide the signal spectrum, which has the problem of over-segmentation of the resonance frequency band, thereby generating pseudo peaks. Feature mode decomposition (FMD) is a new signal decomposition method that takes into account the periodicity and impact of mechanical signals. It uses the deconvolution correlation kurtosis as the objective function to achieve adaptive decomposition of multi-component signals. However, like VMD, achieving a good decomposition effect requires the reasonable selection of pre-parameters. The Puma algorithm is a newly proposed meta-heuristic optimization algorithm with strong global optimization ability and fast convergence speed.

[0004] In summary, how to solve the above problems and propose a new bearing fault feature extraction method based on parameter optimization SMHD-FMD has become a technical problem that needs to be solved urgently by people in this field. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose a bearing fault feature extraction method based on parameter optimization SMHD-FMD. Compared with the existing decomposition method, the experimental signal analysis results of this method are clearer and more effective.

[0006] The bearing fault feature extraction method based on parameter optimization SMHD-FMD of the present invention comprises the following steps:

[0007] S1. Collect the vibration signal of the rolling bearing, use the Puma algorithm to optimize the SMHD pre-parameters, and then use the optimal parameter SMHD to perform noise reduction pre-processing;

[0008] S2. Perform PO-FMD on the denoised signal to obtain the IMF component;

[0009] S3. Select the optimal component according to the kurtosis of each component;

[0010] S4. Analyze the envelope demodulation of the optimal component and extract the characteristic frequency to realize fault diagnosis.

[0011] Preferably, when performing pre-parameter optimization in S1, the Puma algorithm generates 30 individual values ​​that meet the value range

[2300] in each iteration. The process of generating values ​​is divided into:

[0012] The exploration phase is as follows:

[0013]

[0014] if CostX new <CostX i ,U=U+P; (9)

[0015] Where ub and lb are the upper and lower limits of the pre-parameter selection, the value range of the SMHD filter length L is

[2300] , the value range of the FMD mode number N and the filter length L is

[27] ,

[10100] , R Dim 、rand i is a randomly generated number between 0 and 1, X a～f,G is the individual solution in the whole group; j rand is a randomly generated integer within the parameter optimization dimension; U is the parameter set in the optimization process, and its value is a random number between 0 and 1; N pop is the population size of mountain lions, set to 30;

[0016] The development phase is as follows:

[0017]

[0018] R = 2 rand 11 -1; (13)

[0019]

[0020] F2=w×(v) 2 ·cos((2×rand12 )·w); (15)

[0021] w=randn4; v=randn5; (16)

[0022] Where: sol total is the sum of all solutions, is a randomly selected solution from the entire population, X i is the solution of the current iteration, β takes the value of 0 or 1, α and L are static parameters, Puma male is the optimal solution in the group; Iter is the current number of iterations, and MaxIter is the total number of iterations for parameter optimization, which is set to 100.

[0023] Preferably, in S1, when the optimal parameter SMHD is used for noise reduction preprocessing, fuzzy entropy is used as the objective function to judge the quality of the signal noise reduction effect. The smaller the fuzzy entropy value, the more it indicates that the signal has a regular impact component, the stronger the sparsity of the signal, and the richer the fault information contained. The noise reduction processing expression is:

[0024]

[0025] Where: x(t) is the fault signal, T0 is the fault signal period;

[0026] The update formula of the filter coefficients during the iteration process is:

[0027] f k =A -1 b; (2)

[0028] Where: f k is the filter coefficient obtained by K-th iteration calculation, A is the autocorrelation matrix of the fault signal, and b is the inverse filter input signal of K-th iteration.

[0029] Preferably, in S2, the characteristic mode decomposition takes the correlation kurtosis as the objective function, and the vibration signal is quickly decomposed by iteratively updating the filter bank; the expression of the objective function is shown in formula (3);

[0030]

[0031] Where: M is the shift order, N is the signal length, L is the filter length, T s is the input cycle, U k is the component signal numbered K, f k is the FLR filter numbered k;

[0032]

[0033] The decomposition pattern of eigenmode decomposition can be expressed by formula (4), where the superscript H is the matrix conjugate transpose, R1 is the weighted correlation matrix, and R2 is the correlation matrix;

[0034] R XX f k =R XX f k λ (5)

[0035] The filter coefficients are iteratively updated according to formula (5) and continuously approach the target value. In each iteration, the correlation coefficient of two adjacent component signals is calculated, and the two components with the largest correlation coefficients are screened out. Then, the correlation kurtosis of the two components is calculated, and the component signal with the larger kurtosis value is retained. The iterative process is repeated, and it is determined whether the number of component signals meets the set value. If so, the final component is output.

[0036] Preferably, in S3, when calculating the kurtosis of each component signal of FMD and EEMD, the component signal with the largest kurtosis value is taken as the optimal component.

[0037] The advantages and technical effects of the present invention are:

[0038] The bearing fault feature extraction method based on parameter optimized SMHD-FMD of the present invention uses the optimal parameter SMHD to perform denoising preprocessing on the original signal to highlight the pulse characteristics, so as to perform characteristic mode decomposition on it to obtain modal components, and then uses the kurtosis to select the optimal component. Finally, the fault characteristic frequency is extracted by calculating the envelope spectrum of the optimal component to realize the fault diagnosis of the rolling bearing. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a rolling bearing fault feature extraction flow chart of the present invention;

[0040] Figure 2 It is a schematic diagram of the rolling bearing test bench in the present invention;

[0041] Figure 3 It is the time domain waveform diagram of the 5280min fault signal in the present invention;

[0042] Figure 4 is the envelope spectrum of the 5280min fault signal in the present invention;

[0043] Figure 5 It is a comparison curve diagram of 4 optimization methods in the present invention;

[0044] Figure 6 It is the envelope spectrum of the fault signal of 5280min after deconvolution in the present invention;

[0045] Figure 7 It is a diagram of the iterative process of optimizing the FMD pre-parameters in the present invention;

[0046] Figure 8 This is a schematic diagram of the FMD decomposition results in the present invention;

[0047] Fig. 9 This is a schematic diagram of the EEMD decomposition results in the present invention;

[0048] Fig.10 is a kurtosis curve diagram of component signals in the present invention;

[0049] Fig.11 is the envelope spectrum of IMF1 in the present invention;

[0050] Fig.12 is the envelope spectrum of IMF3 in the present invention;

[0051] Fig.13 This is a fault wear diagram of the inner ring of the bearing in the present invention;

[0052] Fig.14 The time domain waveform and envelope spectrum of the fault signal in the present invention, wherein (a) the time domain waveform of the fault signal; (b) the envelope spectrum of the fault signal;

[0053] Fig.15 is the envelope spectrum of the fault signal after deconvolution in the present invention;

[0054] Fig.16 is the FMD decomposition result of the deconvolution signal in the present invention;

[0055] Fig.17 It is the envelope spectrum of the optimal component in the present invention. DETAILED DESCRIPTION

[0056] The embodiments of the invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the invention, and cannot be understood as limiting the invention.

[0057] The present invention proposes a bearing fault feature extraction method based on parameter optimization SMHD-FMD

[0058] Here are the steps:

[0059] (1) The vibration signal of the rolling bearing is collected, the SMHD pre-parameters are optimized, and then the optimal parameter SMHD is used to perform noise reduction pre-processing.

[0060] (2) Perform PO-FMD on the denoised signal to obtain the IMF component.

[0061] (3) Select the optimal component based on the kurtosis of each component.

[0062] (4) Perform envelope demodulation analysis on the optimal component and extract characteristic frequencies to realize fault diagnosis.

[0063] The verification of rolling bearing fault diagnosis effect is divided into experimental signal verification and engineering signal verification, which are explained one by one:

[0064] The original vibration signal of the experimental signal verification part comes from the full life cycle acceleration test of rolling bearings at the NSFI / UCR Intelligent Maintenance System Center. Figure 2 As shown. The test bench consists of an AC motor, a spindle, bearing components, a loading device, a lubrication system, a test device, etc. The parameters of the rolling bearing are shown in Table 1. The test uses an acceleration sensor to collect signals, with a sampling frequency of 20kHz. The outer ring fault frequency calculated according to the theoretical empirical formula is 230Hz. The second group of tests lasted for 164h, and a total of 984 data files were collected. The sampling interval was 10min, and the number of sampling points was 20480 points.

[0065] Table 1 Structural parameters of rolling bearings

[0066]

[0067] The existing technology uses the Gini coefficient as an indicator to evaluate the health status of rolling bearings, and obtains the second group of experiments. The initial moment of the bearing failure is 5280 minutes. The bearing failure signal at this moment is used as a sample signal to verify the method. Figure 3 , Figure 4 , which are the time domain waveform and envelope spectrum of the 5280min fault signal respectively.

[0068] from Figure 3 , Figure 4 It can be seen that the periodic pulse of the bearing is submerged by the signal, there are many interference lines in the envelope spectrum, the characteristic frequency of the outer ring fault and its multiples cannot be observed, and thus the fault diagnosis cannot be made. Therefore, it is necessary to perform noise reduction processing to highlight the impact component in the signal. Sparse Maximum Harmonic Noise Ratio Deconvolution (SMHD) is a noise reduction algorithm developed based on minimum entropy deconvolution to restore the original pulse of the vibration signal. However, the quality of SMHD noise reduction is subject to the selection of the pre-parameter filter length (the value range of the SMHD filter length L is

[2300] ).

[0069] It is time-consuming and laborious to try out the SMHD filter length parameters one by one by enumeration method, so the Puma algorithm is used to optimize the parameters. The Puma algorithm generates 30 individual values ​​that meet the value range

[2300] in each iteration. The process of generating values ​​is divided into the exploration stage (the main steps are formula (6) to formula (9))

[0070]

[0071]

[0072] if CostX new <CostX i ,U=U+P (9)

[0073] and development phase (the main steps are formula (11) to formula (16)).

[0074]

[0075] R = 2 rand 11 -1 (13)

[0076]

[0077] F2=w×(v) 2 ·cos((2×rand 12 )·w)(15)

[0078] w=randn4;v=randn5 (16)

[0079] The 30 numbers generated in each iteration are input into SMHD one by one to perform noise reduction on the 5280-min fault vibration signal. The main steps of the noise reduction process of SHMD are formula (1) to formula (2). The quality of the signal noise reduction effect requires an objective function to describe, that is, a standard to judge. Fuzzy entropy is used as the objective function. The smaller the fuzzy entropy value, the more it indicates that the signal has a regular impact component, the stronger the sparsity of the signal, and the richer the fault information contained.

[0080] The Puma algorithm uses formula (10) to iterate the objective function value, that is, in each iterative calculation, only the filter length L value of the SHMD calculation with the smallest fuzzy entropy value is taken and retained.

[0081] X a,G =X new ,if X i,new <X a,G (10)

[0082] After 100 iterations, the global optimal solution is obtained, that is, the final filter length L value. (Each iteration is 30 to take 1, 100 iterations is 100 times 30 to take 1, and finally there are 100 individual values, and then the individual value that minimizes the objective function value is taken from these 100 individual values).

[0083] The Puma algorithm is used to adaptively optimize the pre-parameter filter length L of SMHD, and the value of L is determined to be 14. In order to reflect the superiority of the Puma algorithm, it is compared with other algorithms. The Puma algorithm (PO), honey badger algorithm (HBA), genetic algorithm (GA), and particle swarm algorithm (PSO) are compared in the convergence speed and the best fitness value of the four different optimization algorithms in the process of SMHD parameter optimization. Figure 5 It can be seen that the Puma algorithm is better than other algorithms in terms of convergence speed and fitness value.

[0084] Figure 6 The envelope spectrum of the 5280-min fault signal processed by the optimal parameter SMHD is shown in the figure. It can be seen from the figure that the impact component is prominent, and the purpose of noise reduction has been achieved. Although the fault characteristic frequency can be extracted, the characteristic frequency is surrounded by the passband noise, and the frequency multiple cannot be extracted. In order to prevent misdiagnosis, it is necessary to further process it to improve the signal-to-noise ratio of the signal.

[0085] Due to the influence of the working environment and the vibration of the shaft system, the bearing vibration signal is often non-stationary and nonlinear, which makes it difficult to extract fault features from it. In order to solve this problem, the signal decomposition algorithm is proposed. Therefore, the deconvolution signal is subjected to eigenmode decomposition (FMD). The main calculation steps of eigenmode decomposition are formula (3) to formula (5).

[0086]

[0087] R XX f k =R XX f k λ (5)

[0088] FMD can realize adaptive decomposition of multi-component signals, but to achieve a good decomposition effect, it is necessary to reasonably select the number of pre-parameter modes N and the filter length L. The numerical range of the number of modes N and the filter length L is

[27] ,

[10100] . Since there are too many permutations and combinations of the two values, and the two pre-parameters are coupled and affect each other, it is not advisable to use the enumeration method for selection. Therefore, the Puma algorithm is used to optimize its parameters. Each iteration of the Puma algorithm generates 30 individual arrays that meet the value range

[27] ,

[10100] . The process of generating values ​​is divided into the exploration stage (the main steps are formula (6) to formula (9)) and the development stage (the main steps are formula (11) to formula (16)).

[0089] The 30 arrays generated in each iteration are input into FMD one by one to decompose the deconvolution signal. The quality of the signal noise reduction effect requires an objective function to describe, that is, a standard to judge. Fuzzy entropy is used as the objective function. The smaller the fuzzy entropy value, the more it indicates that the signal has regular impact components, the stronger the sparsity of the signal, and the richer the fault information it contains.

[0090] The Puma algorithm uses formula (10) to iterate the objective function value, that is, in each iterative calculation, only the modal number N and filter length L of the FMD calculation with the smallest fuzzy entropy value are taken and retained.

[0091] After 100 iterations, the global optimal solution is obtained, that is, the final number of modes N and filter length L. (Each iteration is 30 to take 1 array, 100 iterations is 100 times 30 to take 1 array, and finally there are 100 individual arrays, and then the individual array that minimizes the objective function value is taken from these 100 individual arrays).

[0092] The Cougar algorithm is used to optimize the number of pre-parameter modes N and the filter length L, N = 2, L = 51. Figure 7 It is an iterative process diagram.

[0093] Figure 8 The figure is the decomposition result diagram of the deconvolution signal by FMD. In order to show the superiority of FMD, it is compared with EEMD. Fig. 9 This is the decomposition result diagram of the deconvolution signal after EEMD decomposes it. It can be seen from the figure that EEMD has serious modal aliasing phenomenon.

[0094] Calculate the kurtosis of each component signal of FMD and EEMD, and take the component signal with the largest kurtosis value as the optimal component, such as Fig.10 As shown, the optimal component of FMD is IMF1, and the optimal component of EEMD is IMF3.

[0095] Envelope demodulation analysis is performed on the optimal component, and the results are as follows: Fig.11 , Fig.12 As shown in the figure, compared with the envelope spectrum of the EE MD optimal component IMF3, the envelope spectrum of the FMD optimal component IMF1 has fewer interference spectral lines and more useful information can be extracted. The characteristic frequency of the bearing outer ring fault and its multiples can be extracted to achieve fault diagnosis.

[0096] The original vibration signal of the engineering signal verification part comes from the Yangming 1.5MW wind turbine No. 30 in the Guishan Wind Power Plant in Zhuhai, Guangdong. The fault type is the wear of the inner ring of the front bearing of the generator (such as Fig.13The sampling frequency is 25.6KHz, the sampling time is 2.56s, the fault characteristic frequency is 108Hz, and the bearing model is 6330M / C3 NSK.

[0097] Fig.14 It is the time domain waveform and envelope spectrum of the engineering fault signal. From the envelope spectrum, it can be seen that there are many interference spectral lines and the fault characteristic frequency is difficult to extract, so it needs to be denoised.

[0098] The fault signal is deconvoluted using SHMD (filter length L = 188) with pre-parameter adaptive selection. Fig.15 is the envelope spectrum of the fault signal after deconvolution. It can be seen from the figure that compared with Fig.14 (b) The noise is effectively suppressed, and the fault characteristic frequency and its frequency multiple components can be extracted. However, there are still interference spectral lines with large amplitudes in the passband, which affect the fault diagnosis. Therefore, they need to be further processed.

[0099] The deconvolved signal is subjected to characteristic mode decomposition (after optimization by the Puma algorithm, the number of modes N = 2 and the filter length L = 49) to obtain two modal components, such as Fig.16 Calculate the kurtosis values ​​of the two modal components (IMF1 = 20.38, IMF2 = 14.53), take IMF1 as the optimal component, and perform envelope demodulation. The result is as follows Fig.17 As shown in the figure, we can clearly see the characteristic frequency of the inner ring fault of the rolling bearing and its frequency multiple components, and the spectrum is relatively clear.

[0100] Finally, all the unexplained parts of the invention adopt mature products and mature technical means in the existing technology.

[0101] In the description of this specification, reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in the embodiment or example of the invention.

[0102] Although embodiments of the invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. Bearing fault feature extraction method based on parameter optimization SMHD-FMD, characterized in that: The following steps are involved: S1. Collect the vibration signal of the rolling bearing, use the Puma algorithm to optimize the SMHD pre-parameters, and then use the optimal parameter SMHD to perform noise reduction pre-processing; S2. Perform PO-FMD on the denoised signal to obtain the IMF component; S3. Select the optimal component according to the kurtosis of each component; S4. Analyze the envelope demodulation of the optimal component and extract the characteristic frequency to realize fault diagnosis.

2. The bearing fault feature extraction method based on parameter optimized SMHD-FMD according to claim 1 is characterized in that: When performing pre-parameter optimization in S1, the Puma algorithm generates 30 individual values ​​that meet the value range [2300] in each iteration. The process of generating values ​​is divided into: The exploration phase is as follows: if CostX new <CostX i ,U=U+P; (9) Where ub and lb are the upper and lower limits of the pre-parameter selection, the value range of the SMHD filter length L is [2300], the value range of the FMD mode number N and the filter length L is [27], [10100], R Dim 、rand i is a randomly generated number between 0 and 1, X a～f,G is the individual solution in the whole group; j rand is a randomly generated integer within the parameter optimization dimension; U is the parameter set in the optimization process, and its value is a random number between 0 and 1; N pop is the population size of mountain lions, set to 30; The development phase is as follows: R=2·rand 11 -1; (13) F2=w×(v) 2 ·cos((2×rand 12 )·w); (15) w=randn4; v=randn5; (16) Where: sol total is the sum of all solutions, is a randomly selected solution from the entire population, X i is the solution of the current iteration, β takes the value of 0 or 1, α and L are static parameters, Puma male is the optimal solution in the group; Iter is the current number of iterations, and MaxIter is the total number of iterations for parameter optimization, which is set to 100.

3. The bearing fault feature extraction method based on parameter optimized SMHD-FMD according to claim 1 is characterized in that: In S1, when the optimal parameter SMHD is used for noise reduction preprocessing, fuzzy entropy is used as the objective function to judge the quality of the signal noise reduction effect. The smaller the fuzzy entropy value, the more it indicates that the signal has a regular impact component, the stronger the sparsity of the signal, and the richer the fault information contained. The noise reduction processing expression is: Where: x(t) is the fault signal, T0 is the fault signal period; The update formula of the filter coefficients during the iteration process is: f k =A -1 b; (2) Where: f k is the filter coefficient obtained by K-th iteration calculation, A is the autocorrelation matrix of the fault signal, and b is the inverse filter input signal of K-th iteration.

4. The bearing fault feature extraction method based on parameter optimized SMHD-FMD according to claim 1 is characterized in that: In S2, the characteristic mode decomposition takes the correlation kurtosis as the objective function, and quickly realizes the decomposition of the vibration signal by iteratively updating the filter bank; the expression of the objective function is shown in formula (3); Where: M is the shift order, N is the signal length, L is the filter length, T s is the input cycle, U k is the component signal numbered K, f k is the FLR filter numbered k; The decomposition pattern of eigenmode decomposition can be expressed by formula (4), where the superscript H is the matrix conjugate transpose, R1 is the weighted correlation matrix, and R2 is the correlation matrix; R XX f k =R XX f k λ (5) The filter coefficients are iteratively updated according to formula (5) and continuously approach the target value; each iteration calculates the correlation coefficient of two adjacent component signals, and selects the two components with the largest correlation coefficients, then calculates the correlation kurtosis of the two components, and retains the component signal with the larger kurtosis value; Repeat the iterative process and determine whether the number of component signals meets the set value. If so, output the final component.

5. The bearing fault feature extraction method based on parameter optimized SMHD-FMD according to claim 1 is characterized in that: In S3, when calculating the kurtosis of each component signal of FMD and EEMD, the component signal with the largest kurtosis value is taken as the optimal component.

Images