CN109975013B - IVMD-SE-based wind turbine generator gearbox fault feature extraction method - Google Patents

IVMD-SE-based wind turbine generator gearbox fault feature extraction method Download PDF

Info

Publication number
CN109975013B
CN109975013B CN201910360493.2A CN201910360493A CN109975013B CN 109975013 B CN109975013 B CN 109975013B CN 201910360493 A CN201910360493 A CN 201910360493A CN 109975013 B CN109975013 B CN 109975013B
Authority
CN
China
Prior art keywords
imf
vmd
wind turbine
gearbox
turbine generator
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910360493.2A
Other languages
Chinese (zh)
Other versions
CN109975013A (en
Inventor
姜佳辉
包永强
陈文祥
邵琪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Institute of Technology
Original Assignee
Nanjing Institute of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing Institute of Technology filed Critical Nanjing Institute of Technology
Priority to CN201910360493.2A priority Critical patent/CN109975013B/en
Publication of CN109975013A publication Critical patent/CN109975013A/en
Application granted granted Critical
Publication of CN109975013B publication Critical patent/CN109975013B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/02—Gearings; Transmission mechanisms
    • G01M13/021—Gearings
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/02—Gearings; Transmission mechanisms
    • G01M13/028—Acoustic or vibration analysis

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a wind turbine generator gearbox fault feature extraction method based on IVMD-SE, belonging to the technical field of wind power, and the method comprises the steps of respectively acquiring original vibration signals of a gear of a wind turbine generator gearbox under three working conditions of normal, abrasion and tooth breakage by using a vibration acceleration sensor; optimizing a punishment parameter alpha and the decomposition layer number K in the VMD parameter by adopting a PSO algorithm based on integer programming; decomposing the gear vibration signal under each working condition by adopting an improved variation modal method to obtain each modal component of the vibration signal; selecting IMF components closely related to the original signal by using a correlation coefficient method; extracting singular entropies of the modal components as fault features of the fan gearbox; and inputting the features into the multi-classification SVM, and verifying the feature extraction effect. According to the method, the VMD and the singular entropy are combined to extract the gear fault characteristics of the gearbox of the wind turbine generator under the noise interference, the signal characteristics are enhanced, and the characteristic extraction effect is more obvious.

Description

IVMD-SE-based wind turbine generator gearbox fault feature extraction method
Technical Field
The invention belongs to the technical field of wind power, and particularly relates to a wind turbine generator gearbox fault feature extraction method based on IVMD-SE.
Background
In recent years, with the continuous development of wind power generation technology, the installed capacity of a fan is larger and larger, and the probability of failure of the fan is also improved year by year. The gearbox is a key component in a transmission system of the wind turbine generator, and once the gearbox breaks down, the whole generator is directly stopped, and serious power generation loss is caused. Meanwhile, the gear box is also the component with the highest failure rate in the unit. Therefore, the method has great practical application significance for carrying out timely fault diagnosis on the wind turbine gearbox and guaranteeing safe and reliable operation of the wind turbine and safety of property and related operators.
The wind turbine generator gearbox is complex in operation condition, vibration signals of the wind turbine generator gearbox often contain a large amount of noise, and the wind turbine generator gearbox presents nonlinear non-stationary multi-component characteristics. At present, time-frequency analysis methods are mostly adopted for processing the complex signals. Some common time frequency methods such as wavelet analysis, short-time fourier transform, EMD decomposition, etc. have disadvantages such as low decomposition precision, modal aliasing, poor noise immunity, etc.
The Variable Mode Decomposition (VMD) is a non-recursive and variable self-adaptive Mode Decomposition method, and has obvious superiority compared with the traditional time-frequency method, so that the method is applied to the fault feature extraction of the wind turbine generator gearbox. However, the VMD method needs to preset a penalty parameter α and a decomposition layer number K, and the selection of these two parameters determines whether the fault information can be accurately extracted. The prior art does not form a theoretical basis of a system for selecting the parameters, so that the application of a VMD method in the field of mechanical fault diagnosis is limited to a certain extent.
The modal component (IMF) decomposed by the VMD contains rich information, and a key problem for analyzing the IMF component is how to reliably and effectively extract the characteristic information contained in the IMF component. On one hand, after the gearbox vibration signal is decomposed by the VMD, a certain amount of noise and irrelevant information are still mixed in each IMF, and on the other hand, because the fault information contained in each IMF is massive and irregular, the fault identification purpose is difficult to achieve by directly analyzing the fault information. Due to the two reasons, the effective extraction of the fault characteristic information is difficult.
In the published patents and documents, many scholars have studied extensively and achieved certain results on how to effectively extract fault features in VMD modal components, but these methods also have disadvantages such as severe noise interference, large computation amount, and low reliability.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a wind turbine generator gearbox fault feature extraction method based on IVMD-SE, signal features are enhanced, and a feature extraction effect is more obvious.
The technical scheme is as follows: in order to achieve the purpose, the invention adopts the following technical scheme:
the IVMD-SE-based wind turbine generator gearbox fault feature extraction method comprises the following steps:
step 1, respectively acquiring original vibration signals of a gear of a gearbox of a wind turbine generator under three working conditions of normal, abrasion and tooth breakage by using a vibration acceleration sensor;
step 2, improving the VMD method by adopting a PSO algorithm based on integer programming, and specifically optimizing a punishment parameter alpha and a decomposition layer number K in VMD parameters;
step 3, decomposing the gear vibration signals under the three working conditions of normal, abrasion and broken teeth in the step 1 by adopting the improved variation modal method in the step 2 respectively to obtain each modal component of the vibration signals;
step 4, selecting the modal component closely related to the original signal by using a correlation coefficient method;
step 5, extracting singular entropies of the optimal modal components as fault features of the fan gearbox;
and 6, inputting the features into the multi-classification SVM, and verifying the feature extraction effect.
Further, in the step 2, as both α and K in the VMD algorithm are positive integers, a PSO algorithm of integer programming is adopted; the method comprises the following specific steps:
2.1) initializing the ith particle x according to equation (1)iGroup position x ofi(0)：
Figure GDA0002596153190000021
In the formula, xL,xURepresenting the lower and upper bounds of x, respectively. [ x ] of]Represents the largest integer not exceeding x, rand is the random number between (0,1) generated, i ═ 1,2,. M; thus, M whole-value initial particles are randomly generated in the search space;
2.2) calculating the fitness value of each particle; the method adopts the IMF local minimum entropy value as the fitness function of the PSO algorithm; first, an IMF envelope entropy is defined, as shown in equation (2):
Figure GDA0002596153190000022
in the formula, a (j) is an envelope signal obtained after the IMF component is subjected to Hilbert demodulation; p is a radical ofjThe method is in a (j) normalized form, if the IMF component obtained after VMD decomposition has obvious fault characteristics and strong impact, the envelope entropy value is smaller, otherwise, the envelope entropy value is larger; when the ith particle is at position xiAccording to its pairVMD decomposition is carried out on the parameter combination (α, K) to obtain K IMF components, E of all IMF components is calculatedpTaking the local minimum entropy value minEpAs a fitness function; after the calculation is finished, finding out the historical optimal position pbest of each particleiAnd a population optimal position gbest;
2.3) updating the velocity v of the ith particleij(h +1) and position xij(h+1)：
vij(h+1)＝vij(h)+c1r1[pbesti(h)-xij(h)]+c2r2[gbest(h)-xij(h)](3)；
Figure GDA0002596153190000031
In the formula (3), h is the number of iterations, c1、c2Is a positive acceleration constant, r1、r2∈ U (0,1), j representing the particle dimension, j being 1,2, formula (4) c ∈ U (0,1),
Figure GDA0002596153190000032
for formula (4), when vij(h +1) > 0 represents a particle xi(h) Is a good evolutionary direction, and is thus shifted by one unit, xij(h+1)＝xij(h) +1, conversely to xi(h) Is moved by one unit in the j-th negative direction, i.e. xij(h+1)＝xij(h) -1; when v isijWhen (h +1) is 0, x is representedi(h) The positive and negative directions of the jth dimension are the same, and at the moment, the particles move by one unit to the positive direction of the jth dimension and move by one unit to the negative direction of the jth dimension or keep still according to the probability of 1/3;
2.4) calculating the fitness of each particle after the position updating and updating the historical optimal position pbest of each particleiAnd a population optimal position gbest;
2.5) checking the termination condition, if the condition is not met, returning to 2.3); otherwise, stopping iteration and outputting the optimal solution.
Further, in step 3, the method includes the following steps:
3.1) initialization modality
Figure GDA0002596153190000033
Center frequency
Figure GDA0002596153190000034
Lagrange operator [ lambda ]1And n (n ═ 0);
3.2) n ← n +1, and obtaining the kth modal component u of the (n +1) th generation according to the updating strategies of the formulas (5) and (6)k n+1(w) and its corresponding center frequency wk n+1：
Figure GDA0002596153190000041
Figure GDA0002596153190000042
3.3) update λ according to equation (7):
Figure GDA0002596153190000043
3.4) verifying whether the stop condition is satisfied
Figure GDA0002596153190000044
If yes, stopping iteration and outputting a result; otherwise, return to 3.2).
Further, in step 4, the correlation coefficient between the jth IMF component and the original signal (denoted as x (t)) is defined as follows:
Figure GDA0002596153190000045
wherein N is the number of signal points, Rx(m) represents the autocorrelation function of signal x (t):
Figure GDA0002596153190000046
the correlation coefficient is a statistical index for representing the degree of correlation closeness between two random variables, and the range of the correlation coefficient is between [0 and 1 ]; if the value is larger, the correlation degree between the two variables is larger, otherwise, the correlation degree is smaller.
Further, in the invention, the correlation coefficient value of each modal component and the original signal after VMD decomposition is calculated, and the component with better correlation with the original signal is reserved. Screening the components according to the principle that r (j) > eta, wherein the threshold eta is defined as follows:
Figure GDA0002596153190000047
further, in step 5, the remaining p IMF components after modal screening are recorded as c1(t),c2(t),···,cp(t), the step of extracting singular entropy is as follows:
5.1) carrying out phase space reconstruction on the ith component, and embedding the ith component into (N-m +1) x m dimensional phase space to obtain an IMF reconstruction matrix
Figure GDA0002596153190000051
Wherein m is the embedding dimension, and the value is determined by adopting a G-P method;
5.2) to XiPerforming singular value decomposition to obtain
Figure GDA0002596153190000052
Wherein, U(N-m+1)×(N-m+1)、Vm×mRespectively (N-m +1) and m-order identity orthogonal matrices, Λ(N-m+1)×mIs a diagonal matrix with non-negative diagonal elements λ b1,2,. l. The diagonal elements are XiThe singular value of (a); defining the singular entropy of the ith IMF component as:
Figure GDA0002596153190000053
has the advantages that: compared with the prior art, the IVMD-SE-based wind turbine generator gearbox fault feature extraction method has the advantages that the self-adaptive selection of the VMD parameters is realized by adopting the PSO algorithm of integer programming aiming at the characteristics of the parameters in the VMD algorithm, the difficulty of parameter selection in the traditional VMD decomposition method is overcome, and the difficulty of mechanical fault diagnosis by researchers by using the VMD method is reduced; the singular entropy has the characteristics of basic modal characteristics of singular value decomposition mining data and the function of measuring the signal complexity by the information entropy; the VMD algorithm has stronger anti-noise performance, the VMD and the singular entropy are combined to extract the gear fault characteristics of the wind turbine generator gearbox under noise interference, the signal characteristics are strengthened, and the characteristic extraction effect is more obvious.
Drawings
FIG. 1 is a flow chart of a wind turbine generator gearbox fault feature extraction method based on IVMD-SE;
FIG. 2 is a schematic view of a wind power transmission system simulation test bed;
FIG. 3 is a time domain diagram of an original vibration signal of each fault condition of a gearbox;
FIG. 4 is a flow chart of PSO algorithm optimizing VMD;
FIG. 5 is a PSO parameter optimization fitness curve;
FIG. 6 is a time domain diagram of various modal components of the VMD decomposition of the gear wear vibration signal under the optimal parameters;
FIG. 7 is a frequency spectrum diagram of various modal components of a VMD decomposition of a gear wear vibration signal under optimal parameters;
FIG. 8 is a diagram showing singular entropy eigenvalues of IMF components in 3 different states of the gearbox;
FIG. 9 is a test set classification effect diagram of a wind turbine generator gearbox fault feature extraction method based on IVMD-SE;
FIG. 10 is a comparison graph of SVM classification accuracy based on different feature extraction methods.
Detailed Description
The invention is further illustrated by the following examples and figures. In the invention, IVMD (improved variable Mode decomposition) -SE (singular entropy) is improved variation modal decomposition-singular entropy.
The IVMD-SE-based wind turbine generator gearbox fault feature extraction method comprises the following steps:
step 1, respectively acquiring original vibration signals of a gear of a gearbox of a wind turbine generator under three working conditions of normal, abrasion and tooth breakage by using a vibration acceleration sensor;
step 2, improving the VMD method by adopting a PSO algorithm based on integer programming, and specifically optimizing a punishment parameter alpha and a decomposition layer number K in VMD parameters;
step 3, decomposing the gear vibration signals under the three working conditions of normal, abrasion and broken teeth in the step 1 by adopting the improved variation modal method in the step 2 respectively to obtain each modal component of the vibration signals;
step 4, selecting the modal component closely related to the original signal by using a correlation coefficient method;
step 5, extracting singular entropies of the optimal modal components as fault features of the fan gearbox;
and 6, inputting the features into the multi-classification SVM, and verifying the feature extraction effect.
In the step 2, the parameters alpha and K are considered to be optimally selected by adopting a particle swarm optimization algorithm. Since α and K in the VMD algorithm are both positive integers, a PSO algorithm with integer programming is used here. The method comprises the following specific steps:
2.1) initializing the ith particle x according to equation (1)iGroup position x ofi(0)：
Figure GDA0002596153190000061
In the formula, xL,xURepresenting the lower and upper bounds of x, respectively. [ x ] of]Denotes the largest integer not exceeding x, rand is the random number between (0,1) generated, i ═ 1, 2. Thus, M whole-value initial particles are randomly generated in the search space;
2.2) calculating the fitness value of each particle. The invention adopts the IMF local minimum entropy value as the fitness function of the PSO algorithm. First, an IMF envelope entropy is defined, as shown in equation (2):
Figure GDA0002596153190000071
in the formula, a (j) is an envelope signal obtained after the IMF component is subjected to Hilbert demodulation; p is a radical ofjAnd in a (j) normalized form, if the IMF component obtained after VMD decomposition has obvious fault characteristics and strong impact, the envelope entropy value is smaller, otherwise, the envelope entropy value is larger. When the ith particle is at position xiThen, VMD decomposition is carried out according to the corresponding parameter combination (α, K) to obtain K IMF components, and E of all IMF components is calculatedpTaking the local minimum entropy value minEpAs a fitness function; after the calculation is finished, finding out the historical optimal position pbest of each particleiAnd a population optimal position gbest;
2.3) updating the velocity v of the ith particleij(h +1) and position xij(h+1)：
vij(h+1)＝vij(h)+c1r1[pbesti(h)-xij(h)]+c2r2[gbest(h)-xij(h)](3)；
Figure GDA0002596153190000072
In the formula (3), h is the number of iterations, c1、c2Is a positive acceleration constant, r1、r2∈ U (0,1), j representing the particle dimension, j being 1,2, formula (4) c ∈ U (0,1),
Figure GDA0002596153190000073
for formula (4), when vij(h +1) > 0 represents a particle xi(h) Is a good evolutionary direction, and is thus shifted by one unit, xij(h+1)＝xij(h) +1, conversely to xi(h) Is moved by one unit in the j-th negative direction, i.e. xij(h+1)＝xij(h) -1; when v isijWhen (h +1) is 0, x is representedi(h) The positive and negative directions of the jth dimension are the same, and the grains are in the timeThe son moves one unit to the positive direction of the jth dimension, moves one unit to the negative direction of the jth dimension or keeps still with the probability of 1/3;
2.4) calculating the fitness of each particle after the position updating and updating the historical optimal position pbest of each particleiAnd a population optimal position gbest;
2.5) checking the termination condition, if the condition is not met, returning to 2.3); otherwise, stopping iteration and outputting the optimal solution.
In step 3, the specific steps of performing variation modal decomposition on the signal are as follows:
3.1) initialization modality
Figure GDA0002596153190000081
Center frequency
Figure GDA0002596153190000082
Lagrange operator [ lambda ]1And n (n ═ 0);
3.2) n ← n +1, and obtaining the kth modal component u of the (n +1) th generation according to the updating strategies of the formulas (5) and (6)k n+1(w) and its corresponding center frequency wk n+1：
Figure GDA0002596153190000083
Figure GDA0002596153190000084
3.3) update λ according to equation (7):
Figure GDA0002596153190000085
3.4) verifying whether the stop condition is satisfied
Figure GDA0002596153190000086
If yes, stopping iteration and outputting a result; otherwise, return to 3.2).
In step 4, the correlation coefficient between the jth IMF component and the original signal x (t) is defined as follows:
Figure GDA0002596153190000087
wherein N is the number of signal points, Rx(m) represents the autocorrelation function of signal x (t):
Figure GDA0002596153190000088
the correlation coefficient is a statistical index for representing the degree of closeness of correlation between two random variables, and is in the range of [0,1 ]. If the value is larger, the correlation degree between the two variables is larger, otherwise, the correlation degree is smaller.
In the invention, the correlation coefficient value of each modal component and the original signal after VMD decomposition is calculated, and the component with better correlation with the original signal is reserved. The components are preferably chosen on the basis of r (j) > η, wherein the threshold η is defined as follows:
Figure GDA0002596153190000091
in step 5, the remaining p IMF components after modal screening are recorded as c1(t),c2(t),···,cp(t), the step of extracting singular entropy is as follows:
5.1) carrying out phase space reconstruction on the ith component, and embedding the ith component into (N-m +1) x m dimensional phase space to obtain an IMF reconstruction matrix
Figure GDA0002596153190000092
Wherein m is the embedding dimension, and the value is determined by adopting a G-P method;
5.2) to XiPerforming singular value decomposition to obtain
Figure GDA0002596153190000093
Wherein, U(N-m+1)×(N-m+1)、Vm×mRespectively (N-m +1) and m-order identity orthogonal matrices, Λ(N-m+1)×mIs a diagonal matrix with non-negative diagonal elements λ b1,2,. l. The diagonal elements are XiThe singular value of (a); defining the singular entropy of the ith IMF component as:
Figure GDA0002596153190000094
examples
In the embodiment, a wind power transmission system simulation test bed is adopted for carrying out experiments, and is shown in figure 2. In the experiment, the rotating speed of the motor is controlled by connecting a frequency converter, the input end of the motor is connected with a driving motor and a coupler, and the driving motor is connected with a load motor after being transmitted by a secondary gear box (a parallel gear box and a planetary gear box). The experiment simulates two different fault working conditions of gear abrasion and tooth breakage respectively in a manual processing defect mode. The vibration sensor is arranged in the direction of the load side of the output shaft of the planetary gear box and is used for acquiring vibration signals of the gear box in the working process. The sampling frequency is set to be 10kHz, and the rotating speed of an output shaft is 1500 r/min.
In the embodiment, the model of the speed reducer of the selected primary planetary gearbox is SPM-3, the model of the cylindrical gearbox is ZDY80-3.15-11, and the speed reducer of the selected primary planetary gearbox is connected with the cylindrical gearbox through a coupler; a wind turbine generator gearbox fault feature extraction method based on improved variational modal decomposition and singular entropy is shown in FIG. 1: the method comprises the following steps:
s1, firstly, acquiring original vibration signals of a planetary gearbox gear under three conditions of normal, abrasion and tooth breakage on a wind power transmission system simulation test bed respectively, acquiring 100 groups of data under each working condition, wherein each group of data is 1000 points, and a time domain diagram of the original vibration signals under each working condition is shown in FIG. 3;
s2, before decomposing the vibration signal, the penalty parameter α and the number of decomposition layers K are determined. Taking the gear wear signal as an example, the parameter is optimized by applying the PSO algorithm according to the flow shown in fig. 4, and the finally obtained corresponding PSO fitness curve graph is shown in fig. 5. In the 6 th generation, the fitness (namely the local minimum entropy) reaches the minimum value, and then the optimal parameter combination [ K, alpha ] corresponding to the VMD is obtained as [5,1550 ];
s3, the wear vibration signal is decomposed into 5-layer VMDs using the parameters obtained above. Finally, time domain and frequency spectrum diagrams of each IMF component of the gear wear vibration signal are obtained, and are respectively shown in fig. 6 and 7;
in order to verify the advantages of the VMD method compared with the recursive signal decomposition method, two indexes of orthogonality IO and energy conservation degree IEC are introduced for evaluating the decomposition performance. For any two components, IO is defined as follows:
Figure GDA0002596153190000101
orthogonality indicator as a general criterion, IOi,jThe larger the value of (A) is, the larger the correlation between the components is, the worse the orthogonality is, and on the contrary, the better the orthogonality between the components is. For IEC, it is defined as follows:
Figure GDA0002596153190000102
wherein x (t) represents the original signal, ci(t) denotes the i-th component obtained after signal decomposition, rn(t) represents a residual term. The energy conservation degree represents the contrast degree of the energy of the signal before and after decomposition, and the closer the value is to 1, the less energy leakage in the decomposition process is, the more lossless the signal is.
For the same section of signal, Empirical Mode Decomposition (EMD), Ensemble Empirical Mode Decomposition (EEMD) and Local Mean Decomposition (LMD) are respectively adopted to perform signal decomposition, and orthogonality and energy conservation indexes obtained after decomposition are shown in table 1:
TABLE 1 evaluation of the IO and IEC indexes of the decomposition algorithm of each mode
Performance index algorithm VMD EMD EEMD LMD
IO 1.805e-03 0.0312 0.0295 0.0267
IEC 0.9781 0.8505 0.8920 0.8052
Comparing the results in table 1, it can be seen that the VMD method is used to decompose the vibration signal of the gearbox, which has the best orthogonality and the closest energy conservation degree to 1, indicating that the decomposition effect is the best.
S4: and respectively carrying out VMD decomposition on the correlation coefficient value r (j) of each IMF component and the original signal and a threshold value eta. The correlation coefficient values for the different modes are shown in table 2:
TABLE 2 correlation coefficient of each modal component with the original signal
IMF number 1 2 3 4 5
Correlation coefficient 0.0136 0.0950 0.1605 0.1453 0.1050
Further, the threshold η is calculated to be 0.10. Thereby performing subsequent correlation processing on the IMF2, IMF3, IMF4 and IMF5 as effective modal components;
s4: and performing singular entropy extraction on each optimized IMF component. For each group of vibration data, a p-dimensional singular entropy characteristic (p is 4) vector [ H ] is constructed respectively1H2...Hp]As a final fan gearbox failure signature.
Specifically, fig. 8 shows singular entropy characteristic values of each IMF component in 3 different states of the wind turbine gearbox. Due to space limitation, only the VMD singular entropy characteristic values corresponding to each group of data under each working condition are listed. The characteristic values of all states of the gear box have better classification identification degree, and the type distinction of various fault working conditions can be better realized by the method which can be intuitively obtained from the graph 8.
In order to verify the superiority of the wind turbine generator gearbox fault feature extraction method in terms of gearbox fault diagnosis precision, a support vector machine is adopted to verify the feature extraction effect of the method. Establishing a characteristic sample set: (x)i,yi)，xi∈R3For the sample VMD singular entropy feature input, yi∈ {1,2,3}, i ═ 1,2,3 are sample outputs, which represent gear normal, wear, and break, respectivelyThree working condition types of the teeth. For each of the three conditions 50 sets of data were extracted, 35 of which were used for training and 15 of which were used for testing. Each group of data respectively extracts VMD singular entropy characteristics by using the method of the invention and inputs the VMD singular entropy characteristics into a one-to-one multi-classification SVM classifier for fault identification. Fig. 9 is a diagram illustrating the classification effect of the corresponding test set.
It can be seen that two samples with only wear failure are mistakenly divided into broken teeth, and the overall classification accuracy reaches 95.6%. For the same signal, the following fault feature extraction methods are respectively adopted: 1) wavelet packet energy characteristics; 2) VMD-kurtosis factor; 3) the VMD-energy entropy is compared with the feature extraction method disclosed by the invention for verification. The following features are input into the same SVM classifier respectively, and the classification accuracy for different faults is obtained as shown in fig. 10.
The comparison test results show that when the support vector machine is used for classification and judgment, the VMD-singular entropy value is used as the characteristic, and a better classification effect can be obtained.
The above examples are only preferred embodiments of the present invention, it should be noted that: any modification, equivalent change and modification of the above embodiments according to the technical spirit of the present invention will still fall within the scope of the technical solution of the present invention for a person of ordinary skill in the art.

Claims (3)

1. IVMD-SE-based wind turbine generator gearbox fault feature extraction method is characterized by comprising the following steps: the method comprises the following steps:
step 1, respectively acquiring original vibration signals of a gear of a gearbox of a wind turbine generator under three working conditions of normal, abrasion and tooth breakage by using a vibration acceleration sensor;
step 2, improving the VMD method by adopting a PSO algorithm based on integer programming, and specifically optimizing a punishment parameter alpha and a decomposition layer number K in VMD parameters;
step 3, decomposing the gear vibration signals under the three working conditions of normal, abrasion and broken teeth in the step 1 respectively by adopting the step 2 to obtain each modal component of the vibration signals;
step 4, selecting the modal component closely related to the original signal by using a correlation coefficient method;
step 5, extracting singular entropies of the optimal modal components as fault features of the fan gearbox;
step 6, inputting the features into a multi-classification SVM, and verifying the feature extraction effect; in the step 2, because both alpha and K in the VMD algorithm are positive integers, a PSO algorithm with integer programming is adopted; the method comprises the following specific steps:
2.1) initializing the ith particle x according to equation (1)iGroup position x ofi(0)：
Figure FDA0002596153180000011
In the formula, xL,xURespectively representing a lower bound and an upper bound of x; [ x ] of]Represents the largest integer not exceeding x, rand is the random number between (0,1) generated, i ═ 1,2,. M; thus, M whole-value initial particles are randomly generated in the search space;
2.2) calculating the fitness value of each particle; adopting an IMF local minimum entropy value as a fitness function of a PSO algorithm; first, an IMF envelope entropy is defined, as shown in equation (2):
Figure FDA0002596153180000012
in the formula, a (j) is an envelope signal obtained after the IMF component is subjected to Hilbert demodulation; p is a radical ofjThe method is in a (j) normalized form, if the IMF component obtained after VMD decomposition has obvious fault characteristics and strong impact, the envelope entropy value is smaller, otherwise, the envelope entropy value is larger; when the ith particle is at position xiThen, VMD decomposition is carried out according to the corresponding parameter combination (α, K) to obtain K IMF components, and E of all IMF components is calculatedpTaking the local minimum entropy value minEpAs a fitness function; after the calculation is finished, finding out the historical optimal position pbest of each particleiAnd a population optimal position gbest;
2.3) updating the velocity v of the ith particleij(h +1) and position xij(h+1)：
vij(h+1)＝vij(h)+c1r1[pbesti(h)-xij(h)]+c2r2[gbest(h)-xij(h)](3)；
Figure FDA0002596153180000021
In the formula (3), h is the number of iterations, c1、c2Is a positive acceleration constant, r1、r2∈ U (0,1), j representing the particle dimension, j being 1,2, formula (4) c ∈ U (0,1),
Figure FDA0002596153180000022
for formula (4), when vij(h +1) > 0 represents a particle xi(h) Is a good evolutionary direction, and is thus shifted by one unit, xij(h+1)＝xij(h) +1, conversely to xi(h) Is moved by one unit in the j-th negative direction, i.e. xij(h+1)＝xij(h) -1; when v isijWhen (h +1) is 0, x is representedi(h) The positive and negative directions of the jth dimension are the same, and at the moment, the particles move by one unit to the positive direction of the jth dimension and move by one unit to the negative direction of the jth dimension or keep still according to the probability of 1/3;
2.4) calculating the fitness of each particle after the position updating and updating the historical optimal position pbest of each particleiAnd a population optimal position gbest;
2.5) checking the termination condition, if the condition is not met, returning to 2.3); otherwise, stopping iteration and outputting an optimal solution; in step 3, the method comprises the following steps:
3.1) initialization modality
Figure FDA0002596153180000023
Center frequency
Figure FDA0002596153180000024
Lagrange operator [ lambda ]1And n (n ═ 0);
3.2) n ← n +1, and obtaining the kth modal component u of the (n +1) th generation according to the updating strategies of the formulas (5) and (6)k n+1(w) and its corresponding center frequency wk n+1：
Figure FDA0002596153180000025
Figure FDA0002596153180000031
3.3) update λ according to equation (7):
Figure FDA0002596153180000032
3.4) verifying whether the stop condition is satisfied
Figure FDA0002596153180000033
If yes, stopping iteration and outputting a result; otherwise, return to 3.2); in step 4, the correlation coefficient between the jth IMF component and the original signal x (t) is defined as follows:
Figure FDA0002596153180000034
wherein N is the number of signal points, Rx(m) represents the autocorrelation function of x (t):
Figure FDA0002596153180000035
the correlation coefficient is a statistical index for representing the degree of correlation closeness between two random variables, and the range of the correlation coefficient is between [0 and 1 ]; if the value is larger, the correlation degree between the two variables is larger, otherwise, the correlation degree is smaller.
2. The IVMD-SE based wind turbine generator gearbox fault feature extraction method of claim 1, wherein: screening components according to the principle that the correlation coefficient r (j) > eta obtained in the step 4 is obtained, wherein the threshold eta is defined as follows:
Figure FDA0002596153180000036
3. the IVMD-SE based wind turbine generator gearbox fault feature extraction method of claim 1, wherein: in the step 5, the residual p IMF components after modal screening are recorded as c respectively1(t),c2(t),···,cp(t), the step of extracting singular entropy is as follows:
5.1) carrying out phase space reconstruction on the ith component, and embedding the ith component into (N-m +1) x m dimensional phase space to obtain an IMF reconstruction matrix
Figure FDA0002596153180000041
Wherein m is the embedding dimension, and the value is determined by adopting a G-P method;
5.2) to XiPerforming singular value decomposition to obtain
Figure FDA0002596153180000042
Wherein, U(N-m+1)×(N-m+1)、Vm×mRespectively (N-m +1) and m-order identity orthogonal matrices, Λ(N-m+1)×mIs a diagonal matrix with non-negative diagonal elements λb1,2,. l; the diagonal elements are XiThe singular value of (a); defining the singular entropy of the ith IMF component as:
Figure FDA0002596153180000043
CN201910360493.2A 2019-04-30 2019-04-30 IVMD-SE-based wind turbine generator gearbox fault feature extraction method Active CN109975013B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910360493.2A CN109975013B (en) 2019-04-30 2019-04-30 IVMD-SE-based wind turbine generator gearbox fault feature extraction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910360493.2A CN109975013B (en) 2019-04-30 2019-04-30 IVMD-SE-based wind turbine generator gearbox fault feature extraction method

Publications (2)

Publication Number Publication Date
CN109975013A CN109975013A (en) 2019-07-05
CN109975013B true CN109975013B (en) 2020-09-18

Family

ID=67087320

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910360493.2A Active CN109975013B (en) 2019-04-30 2019-04-30 IVMD-SE-based wind turbine generator gearbox fault feature extraction method

Country Status (1)

Country Link
CN (1) CN109975013B (en)

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111141519A (en) * 2019-12-27 2020-05-12 西安交通大学 Rolling bearing fault quantitative evaluation method
CN111413089A (en) * 2020-04-08 2020-07-14 北华大学 Gear fault diagnosis method based on combination of VMD entropy method and VPMCD
CN111721528B (en) * 2020-05-18 2022-04-05 浙江工业大学 Wind generating set gear box fault early warning method based on CMS system big data
CN112255316A (en) * 2020-09-23 2021-01-22 南昌航空大学 Ultrasonic signal enhancement method for tiny defects of nickel-based high-temperature alloy additive manufacturing component
CN113049684B (en) * 2021-03-23 2023-07-28 大连海洋大学 Ultrasonic lamb wave defect signal identification method based on variational modal decomposition
CN113255780B (en) * 2021-05-28 2024-05-03 润联智能科技股份有限公司 Reduction gearbox fault prediction method and device, computer equipment and storage medium
CN113406525B (en) * 2021-06-15 2023-06-02 金陵科技学院 Lithium battery pack residual life prediction method based on optimization variation modal decomposition
CN113657268B (en) * 2021-08-13 2023-01-31 江苏国电南自海吉科技有限公司 Signal automatic decomposition method applied to wind turbine generator gearbox fault diagnosis
CN114563181B (en) * 2022-01-10 2023-06-27 浙江工业大学之江学院 Rotary machine fault feature extraction method based on improved variation mode extraction
CN115096588B (en) * 2022-06-28 2023-08-18 重庆大学 Machine tool spindle bearing state monitoring device and method
CN116625688A (en) * 2023-05-24 2023-08-22 石家庄铁道大学 Rolling bearing health monitoring method based on multilayer noise reduction and self-encoder
CN117077064B (en) * 2023-10-13 2024-02-06 云南滇能智慧能源有限公司 Fault detection method, device and equipment for wind power equipment and storage medium

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103033359B (en) * 2012-12-19 2015-10-28 西安交通大学 A kind of main transmission in wind generating set method for diagnosing faults of multiple features Multilateral Comprehensive Judge
CN104748961A (en) * 2015-03-30 2015-07-01 中国矿业大学 Gear fault diagnosis method based on SVD decomposition and noise reduction and correlation EEMD entropy features
CN105203645B (en) * 2015-06-08 2018-03-13 中国铁路总公司 A kind of high-speed switch crackle hurt intelligent detecting method based on vibration signal fusion
CN105699080B (en) * 2015-12-18 2019-09-03 华北电力大学（保定） A kind of Wind turbines bearing fault characteristics extracting method based on vibration data
CN108171263B (en) * 2017-12-26 2019-08-30 合肥工业大学 Based on the Fault Diagnosis of Roller Bearings for improving variation mode decomposition and extreme learning machine

Also Published As

Publication number Publication date
CN109975013A (en) 2019-07-05

Similar Documents

Publication Publication Date Title
CN109975013B (en) IVMD-SE-based wind turbine generator gearbox fault feature extraction method
Qin et al. The optimized deep belief networks with improved logistic sigmoid units and their application in fault diagnosis for planetary gearboxes of wind turbines
Soualhi et al. Pattern recognition method of fault diagnostics based on a new health indicator for smart manufacturing
CN103822793B (en) A kind of complex apparatus acoustic fault recognition positioning method
CN102607845B (en) Bearing fault characteristic extracting method for redundantly lifting wavelet transform based on self-adaptive fitting
CN103291600B (en) Fault diagnosis method for hydraulic pump based on EMD-AR (empirical mode decomposition-auto-regressive) and MTS (mahalanobis taguchi system)
CN105275833A (en) CEEMD (Complementary Empirical Mode Decomposition)-STFT (Short-Time Fourier Transform) time-frequency information entropy and multi-SVM (Support Vector Machine) based fault diagnosis method for centrifugal pump
Bai et al. Research on feature selection for rotating machinery based on Supervision Kernel Entropy Component Analysis with Whale Optimization Algorithm
CN106796579A (en) For the method and system of the defect in automatic detection rotary shaft
CN113420691A (en) Mixed domain characteristic bearing fault diagnosis method based on Pearson correlation coefficient
Jiang et al. Fault diagnosis for rolling bearing using a hybrid hierarchical method based on scale-variable dispersion entropy and parametric t-SNE algorithm
CN108760305B (en) Bearing fault detection method, device and equipment
CN111413089A (en) Gear fault diagnosis method based on combination of VMD entropy method and VPMCD
Wang et al. Gear fault diagnosis based on variational modal decomposition and wide+ narrow visual field neural networks
CN104614179A (en) Method for monitoring state of gearbox of wind power generation set
Li et al. Crack fault detection for a gearbox using discrete wavelet transform and an adaptive resonance theory neural network
CN112729834B (en) Bearing fault diagnosis method, device and system
CN103983452A (en) Failure mode recognition method of epicyclic gearbox using mixed domain feature vector and grey correlation analysis
CN104832418A (en) Hydraulic pump fault diagnosis method based on local mean conversion and Softmax
Han et al. An intelligent fault diagnosis method of variable condition gearbox based on improved DBN combined with WPEE and MPE
CN118296470B (en) Intelligent classification method for multi-classifier fusion
Al Tobi et al. Using MLP‐GABP and SVM with wavelet packet transform‐based feature extraction for fault diagnosis of a centrifugal pump
CN109580260A (en) A kind of inferior health diagnostic method of track vehicle door system
Song et al. Intelligent diagnosis method for machinery by sequential auto-reorganization of histogram
Shao et al. Distribution recurrence plots and measures: Effective signal analysis tools for fault diagnosis of wind turbine drivetrain system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant