CN111898447B - Xin Jihe modal decomposition-based wind turbine generator fault feature extraction method - Google Patents
Xin Jihe modal decomposition-based wind turbine generator fault feature extraction method Download PDFInfo
- Publication number
- CN111898447B CN111898447B CN202010612706.9A CN202010612706A CN111898447B CN 111898447 B CN111898447 B CN 111898447B CN 202010612706 A CN202010612706 A CN 202010612706A CN 111898447 B CN111898447 B CN 111898447B
- Authority
- CN
- China
- Prior art keywords
- fault
- xin
- matrix
- jihe
- components
- 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
Links
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 64
- 238000000605 extraction Methods 0.000 title claims abstract description 26
- 239000011159 matrix material Substances 0.000 claims abstract description 61
- 238000000034 method Methods 0.000 claims abstract description 60
- 238000012549 training Methods 0.000 claims abstract description 33
- 239000013598 vector Substances 0.000 claims abstract description 19
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 238000013145 classification model Methods 0.000 claims abstract description 14
- 238000010276 construction Methods 0.000 claims abstract description 9
- 238000007635 classification algorithm Methods 0.000 claims abstract description 6
- 230000009466 transformation Effects 0.000 claims description 10
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000012935 Averaging Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000003066 decision tree Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000003745 diagnosis Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 9
- 238000004088 simulation Methods 0.000 description 8
- 238000005215 recombination Methods 0.000 description 6
- 230000006798 recombination Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 4
- 238000005096 rolling process Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 125000002347 octyl group Chemical group [H]C([*])([H])C([H])([H])C([H])([H])C([H])([H])C([H])([H])C([H])([H])C([H])([H])C([H])([H])[H] 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000008521 reorganization Effects 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Acoustics & Sound (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a method for extracting fault characteristics of a wind turbine generator based on Xin Jihe modal decomposition, which specifically comprises the following steps: collecting an original bearing fault vibration signal to be diagnosed; decomposing the original bearing fault vibration signal by using a pungent geometric mode decomposition method to obtain an initial component; determining an effective component by calculating cosine similarity between the construction components; constructing a cosine similarity matrix to obtain a final Xin Jihe component; selecting the first two octave geometric components with the most information, constructing a two-dimensional feature vector, and carrying out feature extraction to obtain feature data; and training related parameters in the AdaBoost classification algorithm, and conveying the test set to the trained classification model to complete fault classification. According to the method for extracting the wind turbine generator fault characteristics based on Xin Jihe modal decomposition, xin Jihe decomposition and AdaBoost classification are combined, so that the judgment of bearing fault types and fault degrees is realized, and the judgment precision is high.
Description
Technical Field
The invention belongs to the technical field of fault diagnosis, and relates to a wind turbine generator fault feature extraction method based on Xin Jihe modal decomposition.
Background
In an online monitoring system of the wind turbine generator, vibration signals are collected through a sensor, are processed and analyzed, and are combined with a corresponding intelligent classification algorithm to finish fault discrimination of a bearing.
The common signal decomposition methods mainly comprise Fourier transformation, wavelet transformation, empirical mode decomposition, singular spectrum decomposition, corresponding improvement methods thereof and the like, and the common signal decomposition methods respectively comprise: chinese patent (CN 110866519A) is a rolling bearing fault diagnosis method based on Fourier decomposition and multi-scale arrangement entropy deviation average value, chinese patent (EMD and approximate entropy-based rolling bearing emission signal feature extraction method (CN 109000926A) is a bearing fault feature extraction method based on singular value decomposition and self-encoder (CN 110646203A) is a Chinese patent.
Xin Jihe modal decomposition is a time sequence decomposition method based on the octave space, does not destroy the characteristics of the original time sequence, and has better processing effect on nonlinear signals. In the octave geometry decomposition process, the initial single components are not completely independent, some components have similar periods, frequencies, characteristics and the like, cosine similarity is introduced, and the final octave geometry components are determined by constructing a cosine similarity matrix and recombining the similar components. In order to realize intelligent diagnosis of faults, a fault diagnosis model is established by adopting some intelligent classification algorithms, and in the process, proper fault characteristics are required to be selected as samples for training and testing of the classification model. In order to acquire fault information as much as possible and improve fault accuracy, commonly adopted multidimensional feature vectors sometimes cause redundancy of information, which in turn reduces diagnosis accuracy. On the basis of a Xin Jihe modal decomposition (SGMD-CS) method, xin Jihe entropy is calculated, and a low-dimensional feature vector is constructed and is transmitted to a constructed classification model to finish fault classification, so that higher diagnosis precision can be obtained.
In the aspect of time sequence decomposition, the traditional method for fault diagnosis is carried out in European space for decomposing nonlinear time sequence, so that the problem of modal aliasing of different degrees can occur, while the method for decomposing based on the octyl geometric mode is based on the octyl space, and the method is derived from solving the optimization problem of a nonlinear power system by using an octyl algorithm, is introduced into the field of fault diagnosis, and has better decomposition effect for treating the nonlinear problem; in the aspect of feature extraction, in order to obtain more comprehensive fault information, the problem that information redundancy can occur in a high-dimensional feature vector which is generally constructed reduces the accuracy of fault classification, a low-dimensional feature vector which is constructed based on octave geometric pattern decomposition is used as a fault feature, and weak classifiers are combined by adopting an integrated learning idea to obtain a strong classifier, so that higher accuracy is obtained.
Regarding Xin Jihe mode decomposition, since the introduction of the analysis method into the fault diagnosis field, less research about recombination of similar components is adopted, so that iterative ideas are mostly adopted, similar components are selected for recombination from the first component, then the recombined components are stripped, similar components are searched for recombination in the residual components until the normalized mean square error of the residual components and an original signal is smaller than a given threshold value, and the iteration is stopped, so that how to measure the similarity is not explicitly described. Cosine similarity, widely used for text comparison, is introduced into this method, and the component reorganization method is clarified. Regarding the Adaboost method, the fault classification method combining Xin Jihe decomposition and the Adaboost method is not available at present, and is widely used for problems of object detection, text analysis, data mining and the like.
Disclosure of Invention
The invention aims to provide a wind turbine generator fault feature extraction method based on Xin Jihe modal decomposition, which combines an SGMD-CS method with information entropy, constructs low-dimensional octal geometric entropy as a fault feature vector, and realizes the judgment of bearing fault types and fault degrees through AdaBoost classification, so that the judgment precision is high.
The technical scheme adopted by the invention is that the wind turbine generator fault feature extraction method based on Xin Jihe modal decomposition is implemented specifically according to the following steps:
step 1, collecting an original bearing fault vibration signal to be diagnosed;
step 2, decomposing the original bearing fault vibration signal by using a pungent geometric mode decomposition method to obtain an initial component;
step 3, determining effective components of the initial components obtained in the step 2 through calculating cosine similarity among the construction components;
step 4, recombining similar initial components in the effective components by constructing a cosine similarity matrix, so as to obtain a final Xin Jihe component;
step 5, selecting the first two octave geometric components with the highest information, obtaining octave geometric entropy, constructing a two-dimensional feature vector, and carrying out feature extraction to obtain feature data;
and 6, attaching classification labels to the fault data of different categories, training related parameters in an AdaBoost classification algorithm by using a training set, and conveying a testing set to a trained classification model to finish fault classification.
The present invention is also characterized in that,
the step 1 specifically comprises the following steps: selecting an appropriate sampling frequency f s Collecting an original bearing fault vibration signal to be diagnosed to obtain a shape x=x 1 ,x 2 ,…,x n Where n is the length of the signal.
The step 2 is specifically as follows:
step 2.1, based on the Takens embedding theorem, x=x through a time sequence delay topology equivalent method 1 ,x 2 ,…,x n Is reconstructed into a track matrix X comprising multi-dimensional signals,
where d is the embedding dimension, τ is the delay time, m=n- (d-1) τ;
step 2.2, constructing a Hamiltonian matrix M according to the track matrix X obtained in the step 2.1,A＝X T X；
step 2.3, calculating the eigenvalue λ of matrix A i (i=1, 2, …, d) eigenvector Q corresponding to the eigenvalue of matrix a i (i＝1，2，…，d)；
Step 2.4, calculating a transformation coefficient matrix S i ，By formula Z i ＝Q i S i (i=1, 2, …, d) to obtain the corresponding reconstruction matrix +.>Then the trajectory matrix x=x 1 +X 2 +…+X d ；
Step 2.5X is transformed by diagonal averaging i Conversion to a time series Y of length n i ＝y 1 ，y 2 ，…y n (i=1, 2, …, d), thereby yielding d time sequences of length n, i.e. d initial components, which sum to a one-dimensional time sequence x=x 1 ，x 2 ，…，x n 。
τ=1 in step 2.1.
The step 2.3 is specifically as follows:
presence of a Householder matrixPerforming Householder transformation on the matrix A to obtain a matrix Q, wherein H is orthogonal Xin Juzhen;
so thatLet b=qaq T Then->Wherein the matrix B is an upper Hessenberg matrix, and the diagonal element is the eigenvalue lambda of B i (i=1, 2, …, d), the eigenvalue of a and the eigenvalue of B are equal, the eigenvalue λ of a i (i=1, 2, …, d), column vector Q in matrix Q i (i=1, 2, …, d) is a eigenvector corresponding to the eigenvalue of matrix a.
X is transformed by diagonal averaging in step 2.5 i Conversion to a time series Y of length n i ＝y 1 ，y 2 ，…y n The method comprises the following steps:
for matrix X i Element x in (a) ij (i is more than or equal to 1 and less than or equal to m, j is more than or equal to 1 and less than or equal to d), and d is the following * ＝min(m，d)，m * =max (m, d), n=m+ (d-1) τ, then the corresponding time sequence Y i Element y of (a) k (k=1, 2, …, n) is calculated as follows:
wherein if m < d, letOtherwise->
The step 3 is specifically as follows:
the first f (f=1, 2, …, d) initial components Y i ＝y 1 ，y 2 ，…y n Adding to obtain d structural componentsAnd calculating cosine similarity of two adjacent construction components: /> And sequentially sorting, finding out the boundary point k, so that the cosine similarity after the boundary point k is not changed, and taking the first k initial components as effective components.
The step 4 is specifically as follows:
a cosine similarity matrix CSM is constructed, selecting Y as high similarity ij Initial component Y > 0.95 i And Y j The final Xin Jihe component SGC is added to combine the initial components with the same period, frequency and characteristics to determine the final octave geometry component.
The step 5 is specifically as follows:
selecting the first two components SGC from Xin Jihe components obtained in the step 4 1 And SGC 2 For SGC respectively 1 And SGC 2 According to the method of step 2.1-2.3, corresponding eigenvalues are obtained, and uncertainty p of entropy in different directions is calculated for the eigenvalues obtained by calculation i ，By the formula->Obtaining Xin Jihe entropy, for each time sequence, selecting the first two octave geometric components obtained by decomposition, wherein each component can obtain a corresponding octave geometric entropy to obtain a two-dimensional feature vector SymEn= [ e ] 1 ，e 2 ]Taking the data as characteristic data, the characteristic extraction of fault signals is further completed.
The step 6 is specifically as follows:
step 6.1, taking binary classification as an example, intercepting data with length of n as a fault sample for vibration data of two different faults to form a time sequence of x=x 1 ，x 2 ，…，x n S samples of two faults are selected, wherein m samples are training samples, the rest samples are test samples, and the steps 2-5 are executed on each fault sample to obtain a corresponding group of two-dimensional feature vectors SymEn= [ e ] 1 ，e 2 ]In addition, two different fault states are marked with the label y= { -1,1} to distinguish;
step 6.2, training set number m (SymEn 1 ，y 1 )，…，(SymEn m ，y m ) Delivering to AdaBoost classification model to train model parameters, wherein y i E, Y, specifically, initializing weights of training setFor t=1, …, T is the number of weak classifiers, using a weight distribution D t Training set learning of (1) to obtain a weak classifier h t (x) The method comprises the steps of carrying out a first treatment on the surface of the Calculate h t (x) Classification error rate on training set +.>And h t (x) Coefficient of-> α t > 0, according to alpha t Updating weight distribution of the training set:
wherein the normalization factor
Finally, a strong classifier is obtained
Wherein the weak classifier selects a decision tree;
step 6.3, s-m test sets are transmitted to the Adaboost classification model trained in step 6.2, and the fault type H predicted by the model is observed final (x) And class label y i And (5) whether the classification accuracy is consistent or not is calculated.
The beneficial effects of the invention are as follows:
(1) According to the invention, cosine similarity is introduced into an octave geometric decomposition method to obtain an SGMD-CS method, so that noise reduction and component recombination of signals are completed, the analysis error is reduced, trend items of original signals can be stripped, and the method is suitable for analyzing nonlinear time sequences;
(2) According to the invention, the SGMD-CS method is combined with the information entropy, the low-dimensional octal geometric entropy is constructed as a fault feature vector, the fault type and the fault degree of the bearing are judged by AdaBoost classification, and the Xin Jihe entropy is selected to be higher in diagnosis precision than the approximate entropy, the sample entropy and the fuzzy entropy.
Drawings
FIG. 1 is a flow chart of a wind turbine generator system fault feature extraction method based on Xin Jihe modal decomposition;
FIG. 2 is a flow chart of an SGMD-CS method in the wind turbine generator system fault feature extraction method based on Xin Jihe modal decomposition;
FIG. 3 is a waveform diagram of a simulation signal time domain in an embodiment of a wind turbine generator fault feature extraction method based on Xin Jihe modal decomposition;
FIG. 4 is a waveform diagram of a simulation signal frequency domain in an embodiment of a wind turbine generator fault feature extraction method based on Xin Jihe modal decomposition;
FIG. 5 is a time domain waveform diagram of SGMD-CS decomposition results in an embodiment of a wind turbine generator system fault feature extraction method based on Xin Jihe modal decomposition;
FIG. 6 is a waveform diagram of a frequency domain of an SGMD-CS decomposition result in an embodiment of a wind turbine generator system fault feature extraction method based on Xin Jihe modal decomposition;
FIG. 7 is a time domain waveform diagram of LMD decomposition results in an embodiment of a wind turbine generator fault feature extraction method based on Xin Jihe modal decomposition;
FIG. 8 is a frequency domain waveform diagram of an LMD decomposition result in an embodiment of a wind turbine generator fault feature extraction method based on Xin Jihe modal decomposition;
FIG. 9 is a time domain waveform diagram of EMD decomposition results in an embodiment of a wind turbine generator fault feature extraction method based on Xin Jihe modal decomposition;
fig. 10 is a frequency domain waveform diagram of an EMD decomposition result in an embodiment of a wind turbine generator fault feature extraction method based on Xin Jihe modal decomposition.
Detailed Description
The invention will be described in detail below with reference to the drawings and the detailed description.
The invention aims to provide a method for extracting fault characteristics of a wind turbine generator based on Xin Jihe modal decomposition, which solves the problems in the prior art.
The technical scheme adopted by the invention is that the method for extracting the fault characteristics of the wind turbine generator based on Xin Jihe modal decomposition is implemented as shown in a figure 1, and specifically comprises the following steps:
step 1, collecting an original bearing fault vibration signal to be diagnosed; the method comprises the following steps: selecting an appropriate sampling frequency f s Collecting an original bearing fault vibration signal to be diagnosed to obtain a shape x=x 1 ，x 2 ，…，x n Where n is the length of the signal;
step 2, decomposing the original bearing fault vibration signal by using a pungent geometric mode decomposition method to obtain an initial component; the flow is shown in fig. 2, and specifically comprises the following steps:
step 2.1, based on the Takens embedding theorem, x=x through a time sequence delay topology equivalent method 1 ，x 2 ，…，x n Is reconstructed into a track matrix X comprising multi-dimensional signals,
where d is the embedding dimension, τ is the delay time, m=n- (d-1) τ, τ=1, and the d value is determined by calculating the Power Spectral Density (PSD) of the time series if the maximum peak frequency f estimated by the PSD method max Normalized to be below the threshold value 10 -3 The d value is set to n/3, otherwise d=1.2× (f s /f max )；
Step 2.2, constructing a Hamiltonian matrix M according to the track matrix X obtained in the step 2.1,A＝X T x is a group; the method comprises the following steps:
presence of a Householder matrixPerforming Householder transformation on the matrix A to obtain a matrix Q, wherein H is orthogonal Xin Juzhen;
so thatLet b=qaq T Then->Wherein the matrix B is an upper Hessenberg matrix, and the diagonal element is the eigenvalue lambda of B i (i=1, 2, …, d), the eigenvalue of a and the eigenvalue of B are equal, the eigenvalue λ of a i (i=1, 2, …, d), column vector Q in matrix Q i (i=1, 2, …, d) is a eigenvector corresponding to the eigenvalue of matrix a;
step 2.3, calculating the eigenvalue λ of matrix A i (i=1, 2, …, d) eigenvector Q corresponding to the eigenvalue of matrix a i (i＝1，2，…，d)；
Step 2.4, calculating a transformation coefficient matrix S i ，By formula Z i ＝Q i S i (i=1, 2, …, d) to obtain the corresponding reconstruction matrix +.>Then the trajectory matrix x=x 1 +X 2 +…+X d ；
Step 2.5X is transformed by diagonal averaging i Conversion to a time series Y of length n i ＝y 1 ，y 2 ，…y n (i=1, 2, …, d), thereby yielding d time sequences of length n, i.e. d initial components, which sum to a one-dimensional time sequence x=x 1 ，x 2 ，…，x n The method comprises the steps of carrying out a first treatment on the surface of the Wherein X is transformed by diagonal averaging i Conversion to a time series Y of length n i ＝y 1 ，y 2 ，…y n The method comprises the following steps:
for matrix X i Element x in (a) ij (i is more than or equal to 1 and less than or equal to m, j is more than or equal to 1 and less than or equal to d), and d is the following * ＝min(m，d)，m * =max (m, d), n=m+ (d-1) τ, then the corresponding time sequence Y i Element y of (a) k (k=1, 2, …, n) is calculated as follows:
wherein if m < d, letOtherwise->
And 3, determining an effective component of the initial component obtained in the step 2 by calculating cosine similarity among the construction components, wherein the effective component is specifically as follows: the first f (f=1, 2, …, d) initial components Y i ＝y 1 ，y 2 ，…y n Adding to obtain d structural componentsAnd calculating cosine similarity of two adjacent construction components: />And sequentially sorting, finding out the boundary points k, so that the cosine similarity after the boundary points k is not changed, and taking the first k initial components as effective components.
Step 4, recombining similar initial components in the effective components by constructing a cosine similarity matrix, so as to obtain a final Xin Jihe component; the method comprises the following steps:
the cosine similarity matrix CSM is constructed with the first k initial components,selecting Y as high similarity ij Initial component Y > 0.95 i And Y j Adding to obtain a final Xin Jihe component SGC, so that initial components with the same period, frequency and characteristics are combined to determine a final Xin Jihe component;
step 5, selecting the first two octave geometric components with the highest information, obtaining octave geometric entropy, constructing a two-dimensional feature vector, and carrying out feature extraction to obtain feature data; selecting the first two components SGC from Xin Jihe components obtained in the step 4 1 And SGC 2 For SGC respectively 1 And SGC 2 According to the method of step 2.1-2.3, corresponding eigenvalues are obtained, and uncertainty p of entropy in different directions is calculated for the eigenvalues obtained by calculation i ，By the formulaObtaining Xin Jihe entropy, for each time sequence, selecting the first two octave geometric components obtained by decomposition, wherein each component can obtain a corresponding octave geometric entropy to obtain a two-dimensional feature vector SymEn= [ e ] 1 ，e 2 ]Taking the data as characteristic data, and further finishing the characteristic extraction of fault signals;
step 6, pasting classification labels on fault data of different categories, training relevant parameters in an AdaBoost classification algorithm by using a training set, and conveying a test set to a trained classification model to complete fault classification, wherein the method specifically comprises the following steps:
step 6.1, taking binary classification as an example, intercepting data with length of n as a fault sample for vibration data of two different faults to form a time sequence of x=x 1 ，x 2 ，…，x n Selecting s samples of two faults, wherein m samples are training samples, the rest samples are test samples, and executing steps 2-5 on each fault sample to obtain a corresponding group of two-dimensional characteristicsVector symen= [ e ] 1 ，e 2 ]In addition, two different fault states are marked with the label y= { -1,1} to distinguish;
step 6.2, training set number m (SymEn 1 ，y 1 )，…，(SymEn m ，y m ) Delivering to AdaBoost classification model to train model parameters, wherein y i E, Y, specifically, initializing weights of training setFor t=1, …, T is the number of weak classifiers, using a weight distribution D t Training set learning of (1) to obtain a weak classifier h t (x) The method comprises the steps of carrying out a first treatment on the surface of the Calculate h t (x) Classification error rate on training set +.>And h t (x) Coefficient of-> According to alpha t Updating weight distribution of the training set:
wherein the normalization factor
Finally, a strong classifier is obtained
Wherein the weak classifier selects a decision tree;
step 6.3, s-m test sets are transmitted to the Adaboost classification model trained in step 6.2, and the fault type H predicted by the model is observed final (x) And class label y i And (5) whether the classification accuracy is consistent or not is calculated.
In order to verify the superiority of the SGMD-CS method in decomposing signals, a complex simulation signal is constructed as shown below, in which x (t) contains an amplitude-frequency modulated component, a sine component and a cosine component, and the waveforms are shown in fig. 3-4. The SGMD-CS method, the LMD method and the EMD method are respectively used for decomposing, and the decomposition results are shown in figures 5-10:
as can be seen from fig. 5-6, the SGMD-CS method can well separate the trend terms of the original simulation signal, so that the decomposed components are almost equal to those of the original signal, no under decomposition or over decomposition occurs, and the decomposition error is small. The LMD decomposition decomposes the original signal into 4 components, while the EMD decomposes the signal into 8 components, as is evident from fig. 7-8 and fig. 9-10, the individual components appear to have modal aliasing, and either over-decomposition or under-decomposition occurs, with undesirable decomposition effects.
Here, parameters in the SGMD-CS decomposition of the simulation signal will be explained, first, for a given simulation signal, the sampling frequency F is selected s 1000, which is converted into an original time series of length 1000, the frequency f at the maximum peak is estimated by calculating the power spectral density max 29.7852, so as to determine that the d value is 40, that is, the initial components obtained by a series of transformation processes and finally by using an octave geometric decomposition method are 40, the components with similar components are required to be recombined, and in order to reduce the calculation amount, the first f (f=1, 2, …, 40) initial components are added to obtain a new structural component N f Cosine similarity between two adjacent structural components was found as shown in table 1.
TABLE 1 cosine similarity between adjacent construction components
As can be seen from table 1, from k=6, the calculated adjacent construction signal N f Cosine similarity of 1, cos θ 6,7 =1 indicates the sum N of the first 7 components 7 Sum of the first 6 components N 6 The difference of the contained information is very small, k=6 is selected according to the characteristics of experimental data, so that the subsequent recombination of the components only needs to be conducted on the first 6 components, the latter components are regarded as noise, and then the cosine similarity matrix is constructed on the first 6 components to conduct component recombination. The cosine similarity matrix obtained by calculation is shown as follows:
since the constructed cosine similarity matrix characterizes the similarity between any two components, it is a symmetric matrix, where only elements above the diagonal are listed for simplicity. For the experimental data, when the cosine similarity is higher than 0.95, the components can be considered to be similar, and then the components are combined. As is evident from the cosine similarity matrix: the cosine similarity of component 1 and component 2 is 0.9901, which can be considered to be similar in composition, and is recombined as component SGC 1 Similarly, component 3 and component 4 have cosine similarity of 0.9970 and are recombined into component SGC 2 Finally, component 5 and component 6, which are of similarity 0.9713, are recombined into component SGC 3 . The result of the component reorganization is shown in the first three graphs of the time domain waveform diagram of fig. 4, and the fourth graph represents the remaining components, namely the sum of the first 7 th to the fourth 40 th components, and the sum of the four parts is equal to the original simulation signal.
In order to verify that the constructed Xin Jihe entropy is still high in diagnosis precision when being used as a low-dimensional fault feature vector for fault classification, rolling bearing vibration data of an electric engineering laboratory of Kassi university in the United states is taken as an example, driving end bearing fault data with the sampling frequency of 12KHZ is selected for experimental analysis, and the model of the bearing selected for experiment is 6205-2RS JEM SKF, which is a deep groove ball bearing.
First, the different fault categories of the bearing are classified. And marking the faults representing the faults of the inner ring, the faults of the rolling bodies, the faults of the outer ring and the normal state by using labels 1,2,3 and 4 respectively, selecting 50 groups of data from each fault, using 35 groups of data as training sets, conveying the training sets into a constructed AdaBoost classification model for relevant parameter training, and using the rest 15 groups of data as test sets to verify the fault classification accuracy. Because the training set and the test set samples are randomly selected, the test is repeated for 20 times to obtain the average value, and in addition, the sample entropy, the approximate entropy and the fuzzy entropy are selected for comparison. The classification accuracy obtained is shown in table 2.
Table 2 means of accuracy of classification of different types of faults
And secondly, classifying different fault degrees of the bearing under the same fault type. The fault representing diameters of 0.007 inch, 0.014 inch, 0.021 inch and normal state are marked by labels 1,2,3 and 4 respectively, 50 groups of data are selected for each fault degree, 35 groups are used as training sets, the training sets are transmitted to a constructed AdaBoost classification model for relevant parameter training, and the rest 15 groups are used as test sets for verifying fault classification accuracy. The classification accuracy obtained is shown in table 3.
TABLE 3 Classification accuracy averages for different degrees of failure
As can be seen from tables 2 and 3, whether the fault type or the fault degree is judged, the low-dimensional feature vector Xin Jihe entropy is selected to extract the fault information, so that higher judging precision can be achieved, and compared with the sample entropy, the approximate entropy and the fuzzy entropy, the Xin Jihe entropy has the advantages of being outstanding.
Through the simulation experiment, the provided wind turbine generator set fault feature extraction method based on the SGMD-CS and AdaBoost frames has certain effectiveness, has certain significance for judging the bearing faults of the wind turbine generator set, and further provides corresponding basis for overhauling work of overhaulers.
Claims (5)
1. The method for extracting the fault characteristics of the wind turbine generator based on Xin Jihe modal decomposition is characterized by comprising the following steps of:
step 1, collecting an original bearing fault vibration signal to be diagnosed;
step 2, decomposing the original bearing fault vibration signal by using a pungent geometric mode decomposition method to obtain an initial component;
step 3, determining effective components of the initial components obtained in the step 2 through calculating cosine similarity among the construction components;
step 4, recombining similar initial components in the effective components by constructing a cosine similarity matrix, so as to obtain a final Xin Jihe component;
step 5, selecting the first two octave geometric components with the highest information, obtaining octave geometric entropy, constructing a two-dimensional feature vector, and carrying out feature extraction to obtain feature data;
step 6, pasting classification labels on fault data of different categories, training related parameters in an AdaBoost classification algorithm by using a training set, and conveying a testing set to a trained classification model to finish fault classification;
the step 1 specifically comprises the following steps: selecting an appropriate sampling frequency f s Collecting an original bearing fault vibration signal to be diagnosed to obtain a shape x=x 1 ,x 2 ,…,x n Where n is the length of the signal;
the step 2 specifically comprises the following steps:
step 2.1, based on the Takens embedding theorem, x=x through a time sequence delay topology equivalent method 1 ,x 2 ,…,x n Is reconstructed into a track matrix X comprising multi-dimensional signals,
where d is the embedding dimension, τ is the delay time, m=n- (d-1) τ;
step 2.2, constructing a Hamiltonian matrix M according to the track matrix X obtained in the step 2.1,A＝X T X；
step 2.3, calculating the eigenvalue λ of matrix A i (i=1, 2, …, d) eigenvector Q corresponding to the eigenvalue of matrix a i (i＝1,2,…,d)；
Step 2.4, calculating a transformation coefficient matrix S i ，By formula Z i ＝Q i S i (i=1, 2, …, d) to obtain the corresponding reconstruction matrix +.>Then the trajectory matrix x=x 1 +X 2 +…+X d ；
Step 2.5X is transformed by diagonal averaging i Conversion to a time series Y of length n i ＝y 1 ,y 2 ,…y n (i=1, 2, …, d), thereby yielding d time sequences of length n, i.e. d initial components, which sum to a one-dimensional time sequence x=x 1 ,x 2 ,…,x n ；
The step 3 specifically comprises the following steps:
the first f (f=1, 2, …, d) initial components Y i ＝y 1 ,y 2 ,…y n Adding to obtain d structural componentsAnd calculating cosine similarity of two adjacent construction components: /> Sequentially sorting, finding out a boundary point k, so that cosine similarity after the boundary point k is not changed any more, and taking the first k initial components as effective components at the moment;
the step 4 specifically comprises the following steps:
a cosine similarity matrix CSM is constructed, selecting Y as high similarity ij >Initial component Y of 0.95 i And Y j Adding to obtain a final Xin Jihe component SGC, so that initial components with the same period, frequency and characteristics are combined to determine a final Xin Jihe component;
the step 5 specifically comprises the following steps:
selecting the first two components SGC from Xin Jihe components obtained in the step 4 1 And SGC 2 For SGC respectively 1 And SGC 2 According to the method of step 2.1-2.3, corresponding eigenvalues are obtained, and uncertainty p of entropy in different directions is calculated for the eigenvalues obtained by calculation i ，By the formula->Obtaining Xin Jihe entropy, for each time sequence, selecting the first two octave geometric components obtained by decomposition, each component can obtain a corresponding octave geometric entropy, and obtaining two-dimensional characteristicsVector symen= [ e ] 1 ,e 2 ]Taking the data as characteristic data, the characteristic extraction of fault signals is further completed.
2. The method for extracting fault characteristics of a wind turbine generator based on Xin Jihe modal decomposition according to claim 1, wherein τ=1 in step 2.1.
3. The method for extracting fault characteristics of a wind turbine generator based on Xin Jihe modal decomposition according to claim 1, wherein the step 2.3 specifically comprises:
presence of a Householder matrixPerforming Householder transformation on the matrix A to obtain a matrix Q, wherein H is orthogonal Xin Juzhen;
so thatLet b=qaq T Then->Wherein the matrix B is an upper Hessenberg matrix, and the diagonal element is the eigenvalue lambda of B i (i=1, 2, …, d), the eigenvalue of a and the eigenvalue of B are equal, the eigenvalue λ of a i (i=1, 2, …, d), column vector Q in matrix Q i (i=1, 2, …, d) is a eigenvector corresponding to the eigenvalue of matrix a.
4. The method for extracting fault characteristics of a wind turbine generator based on Xin Jihe modal decomposition according to claim 1, wherein in step 2.5, X is determined by diagonal average transformation i Conversion to a time series Y of length n i ＝y 1 ,y 2 ,…y n The method comprises the following steps:
for matrix X i Element x in (a) ij (i is more than or equal to 1 and less than or equal to m, j is more than or equal to 1 and less than or equal to d), and d is the following * ＝min(m,d)，m * ＝max(m,d),n＝m+(d-1) τ, then the corresponding time series Y i Element y of (a) k (k=1, 2, …, n) is calculated as follows:
wherein if m<d, orderOtherwise->
5. The method for extracting fault characteristics of a wind turbine generator based on Xin Jihe modal decomposition according to claim 1, wherein the step 6 is specifically:
step 6.1, taking binary classification as an example, intercepting data with length of n as a fault sample for vibration data of two different faults to form a time sequence of x=x 1 ,x 2 ,…,x n S samples of two faults are selected, wherein m samples are training samples, the rest samples are test samples, and the steps 2-5 are executed on each fault sample to obtain a corresponding group of two-dimensional feature vectors SymEn= [ e ] 1 ,e 2 ]In addition, two different fault states are marked with the label y= { -1,1} to distinguish;
step 6.2, training set number m (SymEn 1 ,y 1 ),…,(SymEn m ,y m ) Delivering to AdaBoost classification model to train model parameters, wherein y i E, Y, specifically, initializing weights of training setFor t=1, …, T is the number of weak classifiers, using a weight distribution D t Training set learning of (1) to obtain a weak classifier h t (x) The method comprises the steps of carrying out a first treatment on the surface of the Calculate h t (x) Classification error rate on training set +.>And h t (x) Coefficient of-> According to alpha t Updating weight distribution of the training set:
wherein the normalization factor
Finally, a strong classifier is obtained
Wherein the weak classifier selects a decision tree;
step 6.3, s-m test sets are transmitted to the Adaboost classification model trained in step 6.2, and the fault type H predicted by the model is observed final (x) And class label y i And (5) whether the classification accuracy is consistent or not is calculated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010612706.9A CN111898447B (en) | 2020-06-30 | 2020-06-30 | Xin Jihe modal decomposition-based wind turbine generator fault feature extraction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010612706.9A CN111898447B (en) | 2020-06-30 | 2020-06-30 | Xin Jihe modal decomposition-based wind turbine generator fault feature extraction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111898447A CN111898447A (en) | 2020-11-06 |
| CN111898447B true CN111898447B (en) | 2024-02-02 |
Family
ID=73207316
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010612706.9A Active CN111898447B (en) | 2020-06-30 | 2020-06-30 | Xin Jihe modal decomposition-based wind turbine generator fault feature extraction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111898447B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113029567A (en) * | 2021-02-26 | 2021-06-25 | 上海电机学院 | Fault signal diagnosis method based on feature fusion equipotential planet map |
| CN113095151B (en) * | 2021-03-18 | 2023-04-18 | 新疆大学 | Rolling bearing unknown fault detection method based on signal decomposition and complex network |
| CN113155462B (en) * | 2021-03-29 | 2023-03-24 | 温州大学 | Bearing fault diagnosis method based on octyl geometric mode decomposition and graph structure enhanced dynamic time warping |
| CN113435255A (en) * | 2021-06-01 | 2021-09-24 | 西北工业大学 | Engine fault diagnosis method based on feature fusion |
| CN113673452A (en) * | 2021-08-25 | 2021-11-19 | 南京航空航天大学 | Induction chip sensor characteristic signal extraction method based on octyl geometric modal decomposition |
| CN115345210B (en) * | 2022-10-20 | 2023-04-25 | 深圳市信润富联数字科技有限公司 | Bearing signal fault diagnosis method, device, equipment and readable storage medium |
| CN116819378B (en) * | 2023-08-29 | 2023-12-26 | 中国华能集团清洁能源技术研究院有限公司 | Energy storage battery abnormality detection method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106092574A (en) * | 2016-05-30 | 2016-11-09 | 西安工业大学 | The Method for Bearing Fault Diagnosis selected with sensitive features is decomposed based on improving EMD |
| CN109489977A (en) * | 2018-12-28 | 2019-03-19 | 西安工程大学 | Method for Bearing Fault Diagnosis based on KNN-AdaBoost |
| WO2019090879A1 (en) * | 2017-11-09 | 2019-05-16 | 合肥工业大学 | Analog circuit fault diagnosis method based on cross wavelet features |
| WO2019153388A1 (en) * | 2018-02-12 | 2019-08-15 | 大连理工大学 | Power spectral entropy random forest-based aeroengine rolling bearing fault diagnosis method |
-
2020
- 2020-06-30 CN CN202010612706.9A patent/CN111898447B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106092574A (en) * | 2016-05-30 | 2016-11-09 | 西安工业大学 | The Method for Bearing Fault Diagnosis selected with sensitive features is decomposed based on improving EMD |
| WO2019090879A1 (en) * | 2017-11-09 | 2019-05-16 | 合肥工业大学 | Analog circuit fault diagnosis method based on cross wavelet features |
| WO2019153388A1 (en) * | 2018-02-12 | 2019-08-15 | 大连理工大学 | Power spectral entropy random forest-based aeroengine rolling bearing fault diagnosis method |
| CN109489977A (en) * | 2018-12-28 | 2019-03-19 | 西安工程大学 | Method for Bearing Fault Diagnosis based on KNN-AdaBoost |
Non-Patent Citations (3)
| Title |
|---|
| 杨宇 ; 程健 ; 彭晓燕 ; 潘海洋 ; 程军圣 ; .一种基于改进辛几何模态分解的复合故障诊断方法.湖南大学学报(自然科学版).2020,(第02期),全文. * |
| 杨然,周钢,许晓鸣.求解最优控制问题的改进辛几何算法.上海交通大学学报.2000,(第05期),全文. * |
| 郑直 ; 高崇一 ; 宋金超 ; 姜万录 ; .基于SGMD敏感参数和KFCMC的滚动轴承故障诊断方法.机床与液压.2020,(第11期),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111898447A (en) | 2020-11-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111898447B (en) | Xin Jihe modal decomposition-based wind turbine generator fault feature extraction method | |
| CN111272429B (en) | Bearing fault diagnosis method | |
| CN109916628B (en) | Rolling bearing fault diagnosis method based on improved multi-scale amplitude perception permutation entropy | |
| CN111191740B (en) | Fault diagnosis method for rolling bearing | |
| CN109884419B (en) | Smart power grid power quality online fault diagnosis method | |
| CN110503075B (en) | Planetary gearbox fault diagnosis method and system based on PSO-EEMD method | |
| CN114818780A (en) | Vehicle bearing fault diagnosis method based on CEEMDAN and APSO-SVM | |
| CN108846307B (en) | Microseism and blasting event identification method based on waveform image | |
| CN113705396B (en) | Motor fault diagnosis method, system and equipment | |
| Cheng et al. | A bearing fault diagnosis method based on VMD-SVD and Fuzzy clustering | |
| CN109359550A (en) | Language of the Manchus document seal Abstraction and minimizing technology based on depth learning technology | |
| Hussein et al. | Automated classification of tropical plant species data based on machine learning techniques and leaf trait measurements | |
| CN107545228B (en) | Electrified railway operation condition identification method based on wavelet transformation | |
| CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
| CN114067915A (en) | scRNA-seq data dimension reduction method based on deep antithetical variational self-encoder | |
| CN107045624B (en) | Electroencephalogram signal preprocessing and classifying method based on maximum weighted cluster | |
| CN113627539A (en) | Rolling bearing composite fault diagnosis method and system, storage medium and computing equipment | |
| CN110569727B (en) | Transfer learning method combining intra-class distance and inter-class distance for motor imagery classification | |
| CN116429426A (en) | Bearing fault diagnosis method, device and medium for multi-domain feature fusion | |
| CN114530847A (en) | Non-invasive load identification method combining inactive current harmonic waves with XGboost | |
| CN113704688B (en) | Method for recovering missing vibration signals based on variable decibels leaf parallel factor decomposition | |
| Wang et al. | Mill load identification method for ball milling process based on grinding signal | |
| Ju et al. | Object detection for similar appearance objects based on entropy | |
| Abd Halim et al. | Analysis on current flow style for vehicle alternator fault prediction | |
| CN114218991A (en) | Rotor fault diagnosis method based on weighted BP-AdaBoost |
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 |