CN109993105A - A kind of improved adaptive sparse sampling Fault Classification - Google Patents
A kind of improved adaptive sparse sampling Fault Classification Download PDFInfo
- Publication number
- CN109993105A CN109993105A CN201910247145.4A CN201910247145A CN109993105A CN 109993105 A CN109993105 A CN 109993105A CN 201910247145 A CN201910247145 A CN 201910247145A CN 109993105 A CN109993105 A CN 109993105A
- Authority
- CN
- China
- Prior art keywords
- signal
- sparse
- data
- matrix
- wavelet
- 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.)
- Pending
Links
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
- G01M13/04—Bearings
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- 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
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- 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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/513—Sparse representations
Abstract
A kind of improved adaptive sparse sampling Fault Classification belongs to fault diagnosis technology field.The present invention improves the sparse classification method of tradition, firstly, carrying out feature enhancing processing to signal using Wavelet Modulus Maxima and kurtosis method, under the premise of guaranteeing signal sparsity, replaces redundant dictionary using unit matrix；Secondly, carrying out dimensionality reduction to data using gaussian random calculation matrix, the redundancy in signal is reduced, retains effective and a small amount of data；Then, sparse coefficient is solved using degree of rarefication Adaptive matching tracking (SAMP) algorithm, compressed signal is reconstructed；Finally, the judgment basis using cross-correlation coefficient as failure generic, proposes a kind of improved adaptive sparse sampling Fault Classification with this.Experiments verify that the present invention effectively reduces the redundancy in signal, the influence for avoiding time shift deviation from judging fault category, while operation complexity is reduced, improve calculating speed and reconstruction accuracy.
Description
Technical field
The present invention relates to a kind of rotating machinery fault classification method, in particular to a kind of improved adaptive sparse sampling event
Hinder classification method, belongs to fault diagnosis technology field.
Background technique
Rotary machinery fault diagnosis is significant to equipment safety operation, when mechanical equipment breaks down, vibration letter
It number will mutate, therefore the method for diagnosing faults based on vibration signal is widely applied one kind in current fault diagnosis field
Method.It can realize the compression sampling of signal under conditions of not limited by Shannon's sampling theorem using compressive sensing theory, subtract
Redundancy in few signal, is effectively reduced the pressure of data storage and transport, effectively realizes mechanical equipment with a small amount of data
The diagnosis of failure.
Sparse representation theory is meant that by combining less linearly, to represent complete original letter
Number.Sparse representation method is used for bearing Modulation recognition, can effectively identify bearing fault type.Traditional sparse representation method is
By sample to be tested signal decomposition in redundant dictionary, by optimal reconfiguration algorithm obtain sparse coefficient, by sparse coefficient with it is superfluous
The product of remaining dictionary obtains signal reconstruction value, the redundant error of measured value and reconstruction value is finally calculated, to judge letter to be measured
Number generic.And the above method has the disadvantage that: (1) in traditional rarefaction representation classification method redundant dictionary building process
It is complex, it when analyzing a large amount of measured signals, will lead to that calculation amount is larger, and operation time is longer, be extremely unfavorable for magnanimity
Monitoring data are identified；(2) since the degree of rarefication of data is unknown, traditional sparse coefficient derivation algorithm such as orthogonal matching pursuit
(OMP) algorithm or canonical orthogonal matching pursuit (ROMP) algorithm need to estimate by multiple reconstruction result when solving sparse coefficient
Degree of rarefication, operation efficiency are lower；(3) traditional sparse representation method is difficult to solve mass data to ask to fault diagnosis bring pressure
Topic contains bulk redundancy information in monitoring data, calculates overlong time when the amount of data is large according to Shannon's sampling theorem；(4)
Mechanical fault diagnosis field is missed frequently with the method for diagnosing faults based on vibration signal, traditional sparse representation method with redundancy at present
Judgment basis of the difference as failure generic, for vibration signal, it may appear that time shift offset issue, so as to cause failure
Recognition accuracy substantially reduces.
According to the above problem, the present invention improves the sparse classification method of tradition, firstly, utilizing sparse representation method
Feature enhancing processing is carried out to signal, under the premise of guaranteeing signal sparsity, is replaced in original method using unit matrix
Redundant dictionary substantially reduces operation efficiency；Secondly, carrying out dimensionality reduction to data using compression sensing method, reduce superfluous in signal
Remaining information retains effective and a small amount of data；Then, sparse system is solved using degree of rarefication Adaptive matching tracking (SAMP) algorithm
Number, it is not necessary that under the premise of estimating degree of rarefication, signal is reconstructed；Finally, using cross-correlation coefficient as the affiliated class of failure
Other judgment basis solves vibration signal time shift problems with this.Experiments verify that the present invention effectively reduces the letter of the redundancy in signal
Breath, the influence for avoiding time shift deviation from judging fault category, while operation complexity is reduced, improve calculating speed and reconstruct essence
Degree.
Summary of the invention
It is an object of the invention to propose a kind of improved adaptive sparse sampling Fault Classification, to solve to rotate
Mechanical breakdown classification problem.
To achieve the above object, the technical solution adopted by the present invention is a kind of improved adaptive sparse sampling failure modes
Redundant dictionary in traditional rarefaction representation disaggregated model is changed to unit matrix first by method, this method, carries out noise reduction to signal,
Extract the fault signature in signal；Then, dimensionality reduction is carried out to data using compression sensing method, selects random measurement matrix to survey
Magnitude and unit matrix are compressed；Then, sparse coefficient is solved using SAMP algorithm, signal is reconstructed；Finally, adopting
Maximum cross correlation measure is used to replace redundant error as fault verification criterion.This method effectively reduces the redundancy in signal,
The influence for avoiding time shift deviation from judging fault category, while without estimating signal degree of rarefication, computation complexity is reduced, is improved
Calculating speed and reconstruction accuracy.
S1 sparse sampling optimization method:
Sparse sampling optimization is carried out to test sample and training sample signal using Wavelet Modulus Maxima, signal is carried out
Rarefaction representation feature enhancing processing.In wavelet transformed domain, by the mutation of the extreme point detectable signal of wavelet conversion coefficient mould
Point, i.e. singular point, and the size of the changing rule of small echo module maximum and Signal Singularity corresponds, it being capable of table by singular point
The fault signature of signal is shown, therefore, can realize that signal de-noising and fault signature extract using wavelet transformation.
If signal is f (t), j indicates the scale of wavelet decomposition.Signal f (t) is small to obtain through j layers of Dyadic Wavelet Transform
Wave modulus maximum sequence wjF (t), with the increase of decomposition scale (i.e. the number of plies) j, Wavelet Transform Modulus Maximum value sequence also will be gradually dilute
It dredges.If data are excessively sparse, i.e., the neutral element for including in data is excessive, then can not retain the effective information in original signal, lead
Cause the loss of fault signature.
To filter out optimal wavelet modulus maximum sequence, judged using kurtosis, the mathematic(al) representation of kurtosis K is as follows:
Wherein E (x- μ)4For (x- μ)4Mathematic expectaion, x is the Wavelet Transform Modulus Maximum value sequence of input, and μ is mean value, and σ is side
Difference.Kurtosis value is bigger, then the fault message that signal includes is more, judges the size of the kurtosis value of several groups of sequences, chooses kurtosis K most
Big Wavelet Transform Modulus Maximum value sequence is as optimal sequence.Thereby, it is possible to effectively realize denoising to signal and sparse.
Wavelet Modulus Maxima transformation, including following several steps are carried out to signal:
1.1 pairs of collected signals carry out wavelet decomposition, extract the high frequency obtained in decomposition result by high-pass filter
Wavelet Component；
1.2 calculate each layer Wavelet Modulus Maxima；
1.3 calculate the kurtosis of each layer Wavelet Transform Modulus Maximum value sequence amplitude, and optimal Wavelet Transform Modulus Maximum is determined by maximum kurtosis
It is worth component.
The compressed sensing based signal dimensionality reduction of S2:
Dimension-reduction treatment is carried out to optimal Wavelet Modulus Maxima component using gaussian random calculation matrix, is reduced in data
Redundancy.If the optimal Wavelet Modulus Maxima component that length is N is x ∈ RN×1, wherein R is set of real numbers.Construct M row N column
Random measurement matrix Φ ∈ RM×N, wherein M is the line number of calculation matrix and the length of desired obtained data after dimensionality reduction.
It is 0 that each element in matrix, which obeys mean value, and variance isGaussian Profile.Dimensionality reduction is carried out to sample signal:
Y=Φ x ∈ RM×1
Wherein x is testing data (i.e. through Wavelet Modulus Maxima treated data), and y is compressed data.
S3 adaptive optimization method:
The unit matrix B of N × N is initially set up instead of the redundant dictionary in original method, by unit matrix B and random survey
Moment matrix Φ ∈ RM×NEstablish the sensing matrix A of M × N:
A=Φ B
Testing data x is subjected to rarefaction representation at unit matrix B:
X=Biα, i=0,1,2 ... k
Wherein BiIndicate that i-th of element in unit matrix B, α indicate sparse system of the testing data x at unit matrix B
Number, k are the degree of rarefication of α, i.e. the number of nonzero element in α.Therefore compressed data y may be expressed as:
Y=Aiα=Φ Biα∈RM×1
Using SAMP algorithm solve testing data x sparse coefficient α, by input step-length b and determine constant c, gradually into
The selection of row screening, iteration step length can determine that the general value of constant c is 0.1 by being chosen between 0-N every 100.With tradition
Sparse coefficient derivation algorithm such as OMP or ROMP scheduling algorithm is compared, and without estimating signal degree of rarefication, realizes that degree of rarefication adaptively weighs
Structure.
SAMP specific algorithm the following steps are included:
3.1 input sensing matrix A, compressed data y, iteration step length b, determine constant c, residual error en, empty set Jn, empty set
Hn, pnIndicate candidate collection length, n indicates the number of iterations；
3.2 initializing residual error e0=y, enables pn=b；
3.3 calculate each of sensing matrix A atom AiWith residual error enInner product < Ai×en>, institute is calculated interior
Product is ranked up from high to low, selects preceding pnAtom A in A corresponding to a maximum inner productn, and it is saved into candidate collection Jn；
3.4 calculate y=AnαnLeast square solution:Wherein | |
y-Anαn| | e is sought in expressionny-Anαn0 norm,A is sought in expressionnTransposition,Expression is askedInverse matrix, from
In select the sparse coefficient of pn maximum absolute value, by AnIn corresponding atom be denoted asAnd by atomIt saves to Hn；
3.5 update residual errorIf | | en| |≤c × | | y | |, wherein | | en| | with | | en+1| | respectively
E is sought in expressionnWith en+10 norm, then terminate iteration, utilize set HnReconstruction signal；If residual error is greater than in last iterative process
Residual error, then iteration step length is updated to b+1, and candidate collection length is updated to b × pn, return to 3.3 and be iterated；If residual error is less than
Residual error in last iterative process, then the number of iterations is updated to n+1, returns to 3.3 and is iterated；
S4 is using cross-correlation coefficient as criterion:
Degree of correlation between two signals is described using cross-correlation coefficient, for discrete data, model is as follows:
Y is every group of data in training sample, whereinFor reconstruction value, σ is standard deviation,ForWith y's
Covariance,E (y) withRespectivelyY withExpectation.Test is judged by the size of cross-correlation coefficient
Signal generic, related coefficient is bigger, then measured signal and training sample signal are closer.
Compared with prior art, the invention has the following beneficial effects:
(1) redundant dictionary in traditional rarefaction representation classification method is changed to unit matrix by the present invention, and it is multiple to reduce building process
Miscellaneous degree, when analyzing a large amount of measured signals, calculation amount is smaller, operation time section；(2) present invention by sparse representation method with
Compressed sensing combines, and reduces the redundancy in data, retains effective information, reduces operation time；(3) SAMP algorithm is used,
Without the degree of rarefication of estimating signal, operation efficiency is effectively improved, and guarantees reconstruction accuracy；(4) using cross-correlation coefficient as event
The judgment basis for hindering generic, effectively solves the problems, such as that time shift deviation is brought.
Detailed description of the invention
Fig. 1 is of the invention based on Wavelet Modulus Maxima rarefaction representation flow chart
Fig. 2 is adaptive sparse sampling Fault Classification flow chart of the invention
Fig. 3 is each layer Wavelet Modulus Maxima sequence chart of middle (center) bearing signal of the present invention.
Fig. 4 is the recognition effect figure of the lower four kinds of states bearing signal of single group test sample in the present invention.
Fig. 5 is the recognition result figure of four kinds of state bearing signals of multiple groups test sample in the present invention.
Specific embodiment
(1) acquisition of vibration signal: bearing vibration signal is acquired by rotating machinery fault simulation experiment platform, bearing lacks
It falls into, sample frequency, the speed of mainshaft, signal length can sets itself.In sample frequency, the identical situation of the speed of mainshaft, respectively
It acquires that bearing is normal, vibration signal of inner ring failure, outer ring failure, rolling element failure, constructs four class signal testing samples respectively
With training sample, every group of signal length is N.Signal is normalized, guarantees the unification of signal amplitude magnitude；
(2) feature enhances: carrying out sparse table to training sample signal and test sample signal using Wavelet Modulus Maxima
Show that feature enhancing is handled, and extracts fault signature.T layers of small wavelength-division are carried out to training sample signal and test sample signal respectively
Solution, obtains the wavelet coefficient of each layer signal.Kurtosis is calculated to the Wavelet Modulus Maxima component of each layer of high-frequency wavelet coefficient.Due to
Decomposition result is different each time, thus need to only be calculated according to kurtosis as a result, choosing maximum one layer of kurtosis as optimal small
Wave modulus maximum component realizes the denoising of signal and sparse；
(3) compressed sensing dimensionality reduction: the gaussian random calculation matrix of building M × N, M are institute after dimensionality reduction to test sample signal
Wavelet Modulus Maxima component carry out compression processing, obtain compressed signal y；
(4) sparse coefficient and reconstruction value are solved: in the sparse classification method model of unit matrix substitution tradition of building N × N
Redundant dictionary, be calculated sensing matrix A, input sensing matrix A, data y after compression, step-length b is set, determine constant c,
Reconstruction coefficients α of the modulus maximum sequence of test sample under unit matrix is calculated using SAMP algorithm, and finds out each group of letter
Number reconstruction value yi=Aiαi, i=0,1,2...k, wherein AiIndicate i-th of atom in A, αiIndicate i-th of sparse coefficient；
(5) data classification: the cross-correlation coefficient of four class data in reconstruction value and training sample is calculated, eventually by cross-correlation
The maximum value of coefficient obtains the failure generic of input signal.
In conjunction with example, fault diagnosis is carried out to signal of rolling bearing, embodiment is further illustrated:
(1) bearing vibration signal, sampling frequency the acquisition of vibration signal: are acquired by rotating machinery fault simulation experiment platform
Rate is 100kHz, speed of mainshaft 1300r/min, uses normal rolling bearing and inner ring, outer ring, rolling during test
Body is respectively present the rolling bearing of defect, and flaw size is wide 0.7mm and depth 0.05mm.Acquisition bearing is normal, interior respectively
The vibration signal of failure, outer ring failure, rolling element failure is enclosed, every 10K data point is one group.The test specimens of building 10000 × 4
This signal (each one group of every class signal), 10000 × 16 training sample signal (each four groups of every class signal).Normalizing is carried out to signal
Change processing, guarantees the unification of signal amplitude magnitude；
(2) feature enhances: carrying out sparse table to training sample signal and test sample signal using Wavelet Modulus Maxima
Show that feature enhancing is handled, and extract fault signature, as shown in Figure 3.4 are carried out to training sample signal and test sample signal respectively
Layer wavelet decomposition, obtains the wavelet coefficient of each layer signal.The Wavelet Modulus Maxima component of each layer of high-frequency wavelet coefficient is solved, is counted
Kurtosis is calculated, it is need to only being calculated according to kurtosis as a result, choosing the maximum one layer of work of kurtosis since decomposition result is different each time
For optimal Wavelet Modulus Maxima component, the denoising of signal and sparse is realized；
(3) compressed sensing dimensionality reduction: the gaussian random calculation matrix of building 1000 × 10000, to the small of test sample signal
Wave modulus maximum component carries out compression processing, and compressed sensing based basic concept will be believed under the premise of retention fault feature
Number dimension reduces by 10 times, reduces operational data amount, improves fault signal analysis treatment effeciency；
(4) sparse coefficient and reconstruction value are solved: the sparse classification side of unit matrix substitution tradition of building 10000 × 10000
Sensing matrix A is calculated in redundant dictionary in method model, inputs sensing matrix A, data x after compression, step-length b=500 sentence
Permanent several c=0.1 calculate reconstruction coefficients α ' of the modulus maximum sequence of test sample under unit matrix using SAMP algorithm,
And find out reconstruction value yi'=Eiαi'；
(5) data classification: calculating the cross-correlation coefficient of four class data in reconstruction value and training sample, and every class can be calculated
4 cross-correlation are sparse, find out mean value, obtain the failure generic of input signal eventually by the maximum value of cross-correlation coefficient,
As shown in Figure 4.
Fig. 1 is of the invention based on Wavelet Modulus Maxima rarefaction representation flow chart.
Fig. 2 is a kind of improved adaptive sparse sampling Fault Classification flow chart of the present invention.
Fig. 3 is each layer Wavelet Modulus Maxima sequence chart of 1300r/min bearing signal, signal length 10000 in the present invention
Point, Decomposition order is 4 layers, wherein optimal Wavelet Modulus Maxima component is the 3rd layer.
Fig. 4 is the identification of the different classes of lower lower four kinds of states bearing signal of single group test sample of 1300r/min bearing signal
Effect picture, wherein 1,2,3,4 respectively correspond normal bearing, inner ring failure, outer ring failure and four kinds of states of rolling element failure,
The generic of test sample signal is obtained according to the maximum value of the degree of correlation with training sample signal.
Fig. 5 is four kinds of state recognition result figures of multiple groups test sample, wherein speed of mainshaft 1300r/min, to four kinds of shapes
State signal respectively takes 200 groups of data, and circle, cross, triangle and scatterplot in figure respectively indicate rarefaction representation and solve resulting approximation
Value respectively with normal, inner ring failure, outer ring failure, rolling element failure training sample between cross-correlation coefficient.Cross correlation
The maximum corresponding fault category of number is the generic for testing signal.
Claims (2)
1. a kind of improved adaptive sparse samples Fault Classification, which comprises the following steps:
S1 sparse sampling optimization method:
Sparse sampling optimization is carried out to test sample and training sample signal using Wavelet Modulus Maxima, signal is carried out sparse
Indicate feature enhancing processing；
If signal is f (t), j indicates the scale of wavelet decomposition；Signal f (t) is to obtain small echo mould through j layers of Dyadic Wavelet Transform
Very big value sequence wjF (t), with decomposition scale, that is, number of plies j increase, Wavelet Transform Modulus Maximum value sequence also will be gradually sparse；
To filter out optimal wavelet modulus maximum sequence, judged using kurtosis, the mathematic(al) representation of kurtosis K is as follows:
Wherein E (x- μ)4For (x- μ)4Mathematic expectaion, x be input Wavelet Transform Modulus Maximum value sequence, μ is mean value, and σ is variance；It is high and steep
Angle value is bigger, then the fault message that signal includes is more, judges the size of the kurtosis value of several groups of sequences, and it is maximum to choose kurtosis K
Wavelet Transform Modulus Maximum value sequence is as optimal sequence；
The compressed sensing based signal dimensionality reduction of S2:
Dimension-reduction treatment is carried out to optimal Wavelet Modulus Maxima component using gaussian random calculation matrix, reduces the redundancy in data
Information；If the optimal Wavelet Modulus Maxima component that length is N is x ∈ RN×1, wherein R is set of real numbers；Construct the random of M row N column
Calculation matrix Φ ∈ RM×N, wherein M is the line number of calculation matrix and the length of desired obtained data after dimensionality reduction；In matrix
Each element to obey mean value be 0, variance isGaussian Profile；Dimensionality reduction is carried out to sample signal:
Y=Φ x ∈ RM×1
Wherein x is testing data i.e. through Wavelet Modulus Maxima treated data, and y is compressed data；
S3 adaptive optimization method:
The unit matrix B of N × N is initially set up instead of the redundant dictionary in original method, by unit matrix B and random measurement matrix
Φ∈RM×NEstablish the sensing matrix A of M × N:
A=Φ B
Testing data x is subjected to rarefaction representation at unit matrix B:
X=Biα, i=0,1,2 ... k
Wherein BiIndicate that i-th of element in unit matrix B, α indicate sparse coefficient of the testing data x at unit matrix B, k is
The degree of rarefication of α, the i.e. number of nonzero element in α；Therefore compressed data y is indicated are as follows:
Y=Aiα=Φ Biα∈RM×1
The sparse coefficient α that testing data x is solved using SAMP algorithm by input step-length b and is determined constant c, is gradually sieved
Choosing；
SAMP specific algorithm the following steps are included:
3.1 input sensing matrix A, compressed data y, iteration step length b, determine constant c, residual error en, empty set Jn, empty set Hn, pn
Indicate candidate collection length, n indicates the number of iterations；
3.2 initialization residual error e0=y, enables pn=b；
3.3 calculate each of sensing matrix A atom AiWith residual error enInner product < Ai×en>, by the calculated inner product of institute by height
It is ranked up to low, selects preceding pnAtom A in A corresponding to a maximum inner productn, and it is saved into candidate collection Jn；
3.4 calculate y=AnαnLeast square solution:Wherein | | y-An
αn| | e is sought in expressionny-Anαn0 norm,A is sought in expressionnTransposition,Expression is askedInverse matrix, therefrom
Select pnThe sparse coefficient of a maximum absolute value, by AnIn corresponding atom be denoted asAnd by atomIt saves to Hn；
3.5 update residual errorIf | | en| |≤c × | | y | |, wherein | | en| | with | | en+1| | it respectively indicates
Seek enWith en+10 norm, then terminate iteration, utilize set HnReconstruction signal；If residual error is greater than residual in last iterative process
Difference, then iteration step length is updated to b+1, and candidate collection length is updated to b × pn, return to 3.3 and be iterated；If residual error is less than upper one
Residual error in secondary iterative process, then the number of iterations is updated to n+1, returns to 3.3 and is iterated；
S4 is using cross-correlation coefficient as criterion:
Degree of correlation between two signals is described using cross-correlation coefficient, for discrete data, model is as follows:
Y is every group of data in training sample, whereinFor reconstruction value, σ is standard deviation,ForWith the association side of y
Difference,E (y) withRespectivelyY withExpectation；Pass through the size judgement test letter of cross-correlation coefficient
Number generic, related coefficient is bigger, then measured signal and training sample signal are closer.
2. a kind of improved adaptive sparse according to claim 1 samples failure modes, it is characterised in that:
(1) acquisition of vibration signal: by rotating machinery fault simulation experiment platform acquire bearing vibration signal, bearing defect,
Sample frequency, the speed of mainshaft, signal length sets itself；In sample frequency, the identical situation of the speed of mainshaft, axis is acquired respectively
Normal, inner ring failure, outer ring failure, rolling element failure vibration signal is held, constructs four class signal testing samples and training respectively
Sample, every group of signal length are N；Signal is normalized, guarantees the unification of signal amplitude magnitude；
(2) feature enhances: it is special to carry out rarefaction representation to training sample signal and test sample signal using Wavelet Modulus Maxima
Enhancing processing is levied, and extracts fault signature；T layers of wavelet decomposition are carried out to training sample signal and test sample signal respectively, are obtained
To the wavelet coefficient of each layer signal；Kurtosis is calculated to the Wavelet Modulus Maxima component of each layer of high-frequency wavelet coefficient；Due to each
Secondary decomposition result is different, therefore only need to be according to kurtosis calculating as a result, choosing maximum one layer of kurtosis as optimal small echo mould
Maximum component realizes the denoising of signal and sparse；
(3) compressed sensing dimensionality reduction: building M × N gaussian random calculation matrix, M for after dimensionality reduction to the small of test sample signal
Wave modulus maximum component carries out compression processing, obtains compressed signal y；
(4) sparse coefficient and reconstruction value are solved: superfluous in the sparse classification method model of unit matrix substitution tradition of building N × N
Sensing matrix A is calculated in remaining dictionary, inputs sensing matrix A, data y after compression, and step-length b is arranged, and determines constant c, uses
SAMP algorithm calculates reconstruction coefficients α of the modulus maximum sequence of test sample under unit matrix, and finds out each group of signal weight
Structure value yi=Aiαi, i=0,1,2...k, wherein AiIndicate i-th of atom in A, αiIndicate i-th of sparse coefficient；
(5) data classification: the cross-correlation coefficient of four class data in reconstruction value and training sample is calculated, eventually by cross-correlation coefficient
Maximum value obtain the failure generic of input signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910247145.4A CN109993105A (en) | 2019-03-29 | 2019-03-29 | A kind of improved adaptive sparse sampling Fault Classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910247145.4A CN109993105A (en) | 2019-03-29 | 2019-03-29 | A kind of improved adaptive sparse sampling Fault Classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109993105A true CN109993105A (en) | 2019-07-09 |
Family
ID=67131022
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910247145.4A Pending CN109993105A (en) | 2019-03-29 | 2019-03-29 | A kind of improved adaptive sparse sampling Fault Classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109993105A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108037422A (en) * | 2017-12-07 | 2018-05-15 | 国网冀北电力有限公司秦皇岛供电公司 | A kind of photovoltaic system direct current arc fault feature extraction and recognition methods and device |
| CN110765965A (en) * | 2019-10-30 | 2020-02-07 | 兰州理工大学 | Quick dictionary learning algorithm for sparse representation of mechanical vibration signals |
| CN110873833A (en) * | 2019-12-10 | 2020-03-10 | 国网河南省电力公司洛阳供电公司 | Self-adaptive fault section positioning method for power distribution network containing distributed power supply |
| CN111062349A (en) * | 2019-12-23 | 2020-04-24 | 燕山大学 | Bearing fault feature enhancement method and system |
| CN111091159A (en) * | 2019-12-27 | 2020-05-01 | 哈尔滨工程大学 | Weak fault feature extraction method for autonomous underwater robot propeller |
| CN111238816A (en) * | 2020-03-06 | 2020-06-05 | 中原工学院 | Rolling bearing composite fault diagnosis method based on sparse classification algorithm |
| CN111751658A (en) * | 2020-06-24 | 2020-10-09 | 国家电网有限公司大数据中心 | Signal processing method and device |
| CN112287889A (en) * | 2020-11-22 | 2021-01-29 | 西北工业大学 | Bearing fault intelligent diagnosis method based on compressed sensing and related vector machine |
| CN112818762A (en) * | 2021-01-15 | 2021-05-18 | 北华大学 | Large-size composite material and rapid nondestructive testing method for sandwich structure thereof |
| CN112906473A (en) * | 2021-01-19 | 2021-06-04 | 杭州安脉盛智能技术有限公司 | Fault diagnosis method for rotating equipment |
| CN112968741A (en) * | 2021-02-01 | 2021-06-15 | 中国民航大学 | Adaptive broadband compressed spectrum sensing algorithm based on least square vector machine |
| CN113225089A (en) * | 2021-04-23 | 2021-08-06 | 西安交通大学 | Industrial robot multi-channel monitoring data compression method and system |
| CN113420837A (en) * | 2020-08-18 | 2021-09-21 | 河南工业大学 | Fault diagnosis method based on multi-source compressed sensing |
| CN114136604A (en) * | 2021-11-30 | 2022-03-04 | 三门核电有限公司 | Rotary equipment fault diagnosis method and system based on improved sparse dictionary |
| WO2022128465A1 (en) * | 2020-12-17 | 2022-06-23 | Commissariat A L'energie Atomique Et Aux Energies Alternatives | Method for reconstructing an image based on compressed acquisition of measurements |
| FR3122744A1 (en) * | 2021-05-06 | 2022-11-11 | Commissariat A L'energie Atomique Et Aux Energies Alternatives | Method and device for reconstructing a signal from measurements subject to uncertainties. |
| CN117113013A (en) * | 2023-07-19 | 2023-11-24 | 石家庄铁道大学 | Bearing vibration missing data repairing method based on structured compressed sensing |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105281780A (en) * | 2015-11-20 | 2016-01-27 | 重庆大学 | Variable step size regularized adaptive compressed sampling matching pursuit method |
| CN106559744A (en) * | 2016-11-24 | 2017-04-05 | 湖北文理学院 | It is a kind of based on sparse identification and the body area network power-economizing method of compressed sensing |
| CN107368809A (en) * | 2017-07-20 | 2017-11-21 | 合肥工业大学 | A kind of bearing fault sorting technique based on rarefaction representation and dictionary learning |
| CN108037422A (en) * | 2017-12-07 | 2018-05-15 | 国网冀北电力有限公司秦皇岛供电公司 | A kind of photovoltaic system direct current arc fault feature extraction and recognition methods and device |
-
2019
- 2019-03-29 CN CN201910247145.4A patent/CN109993105A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105281780A (en) * | 2015-11-20 | 2016-01-27 | 重庆大学 | Variable step size regularized adaptive compressed sampling matching pursuit method |
| CN106559744A (en) * | 2016-11-24 | 2017-04-05 | 湖北文理学院 | It is a kind of based on sparse identification and the body area network power-economizing method of compressed sensing |
| CN107368809A (en) * | 2017-07-20 | 2017-11-21 | 合肥工业大学 | A kind of bearing fault sorting technique based on rarefaction representation and dictionary learning |
| CN108037422A (en) * | 2017-12-07 | 2018-05-15 | 国网冀北电力有限公司秦皇岛供电公司 | A kind of photovoltaic system direct current arc fault feature extraction and recognition methods and device |
Non-Patent Citations (1)
| Title |
|---|
| 谢馨: "基于稀疏采样的数据修复及故障分类研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108037422A (en) * | 2017-12-07 | 2018-05-15 | 国网冀北电力有限公司秦皇岛供电公司 | A kind of photovoltaic system direct current arc fault feature extraction and recognition methods and device |
| CN110765965A (en) * | 2019-10-30 | 2020-02-07 | 兰州理工大学 | Quick dictionary learning algorithm for sparse representation of mechanical vibration signals |
| CN110765965B (en) * | 2019-10-30 | 2023-09-15 | 兰州理工大学 | Quick dictionary learning algorithm for sparse representation of mechanical vibration signals |
| CN110873833A (en) * | 2019-12-10 | 2020-03-10 | 国网河南省电力公司洛阳供电公司 | Self-adaptive fault section positioning method for power distribution network containing distributed power supply |
| CN110873833B (en) * | 2019-12-10 | 2022-04-08 | 国网河南省电力公司洛阳供电公司 | Self-adaptive fault section positioning method for power distribution network containing distributed power supply |
| CN111062349A (en) * | 2019-12-23 | 2020-04-24 | 燕山大学 | Bearing fault feature enhancement method and system |
| CN111062349B (en) * | 2019-12-23 | 2023-04-18 | 燕山大学 | Bearing fault feature enhancement method and system |
| CN111091159A (en) * | 2019-12-27 | 2020-05-01 | 哈尔滨工程大学 | Weak fault feature extraction method for autonomous underwater robot propeller |
| CN111091159B (en) * | 2019-12-27 | 2023-03-17 | 哈尔滨工程大学 | Weak fault feature extraction method for autonomous underwater robot propeller |
| CN111238816A (en) * | 2020-03-06 | 2020-06-05 | 中原工学院 | Rolling bearing composite fault diagnosis method based on sparse classification algorithm |
| CN111238816B (en) * | 2020-03-06 | 2021-10-22 | 中原工学院 | Rolling bearing composite fault diagnosis method based on sparse classification algorithm |
| CN111751658A (en) * | 2020-06-24 | 2020-10-09 | 国家电网有限公司大数据中心 | Signal processing method and device |
| CN113420837A (en) * | 2020-08-18 | 2021-09-21 | 河南工业大学 | Fault diagnosis method based on multi-source compressed sensing |
| CN113420837B (en) * | 2020-08-18 | 2024-01-26 | 河南工业大学 | Fault diagnosis method based on multi-source compressed sensing |
| CN112287889A (en) * | 2020-11-22 | 2021-01-29 | 西北工业大学 | Bearing fault intelligent diagnosis method based on compressed sensing and related vector machine |
| WO2022128465A1 (en) * | 2020-12-17 | 2022-06-23 | Commissariat A L'energie Atomique Et Aux Energies Alternatives | Method for reconstructing an image based on compressed acquisition of measurements |
| FR3118251A1 (en) * | 2020-12-17 | 2022-06-24 | Commissariat A L'energie Atomique Et Aux Energies Alternatives | Method for reconstructing an image from a compressed acquisition of measurements |
| CN112818762A (en) * | 2021-01-15 | 2021-05-18 | 北华大学 | Large-size composite material and rapid nondestructive testing method for sandwich structure thereof |
| CN112818762B (en) * | 2021-01-15 | 2022-06-07 | 北华大学 | Large-size composite material and rapid nondestructive testing method for sandwich structure thereof |
| CN112906473A (en) * | 2021-01-19 | 2021-06-04 | 杭州安脉盛智能技术有限公司 | Fault diagnosis method for rotating equipment |
| CN112968741B (en) * | 2021-02-01 | 2022-05-24 | 中国民航大学 | Adaptive broadband compressed spectrum sensing algorithm based on least square vector machine |
| CN112968741A (en) * | 2021-02-01 | 2021-06-15 | 中国民航大学 | Adaptive broadband compressed spectrum sensing algorithm based on least square vector machine |
| CN113225089A (en) * | 2021-04-23 | 2021-08-06 | 西安交通大学 | Industrial robot multi-channel monitoring data compression method and system |
| FR3122744A1 (en) * | 2021-05-06 | 2022-11-11 | Commissariat A L'energie Atomique Et Aux Energies Alternatives | Method and device for reconstructing a signal from measurements subject to uncertainties. |
| CN114136604B (en) * | 2021-11-30 | 2023-11-28 | 三门核电有限公司 | Rotary equipment fault diagnosis method and system based on improved sparse dictionary |
| CN114136604A (en) * | 2021-11-30 | 2022-03-04 | 三门核电有限公司 | Rotary equipment fault diagnosis method and system based on improved sparse dictionary |
| CN117113013A (en) * | 2023-07-19 | 2023-11-24 | 石家庄铁道大学 | Bearing vibration missing data repairing method based on structured compressed sensing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109993105A (en) | A kind of improved adaptive sparse sampling Fault Classification | |
| CN109460618B (en) | Rolling bearing residual life online prediction method and system | |
| CN103245907B (en) | A kind of analog-circuit fault diagnosis method | |
| CN109858352B (en) | Fault diagnosis method based on compressed sensing and improved multi-scale network | |
| Tang et al. | Sparse classification of rotating machinery faults based on compressive sensing strategy | |
| Oliveira et al. | Ultrasound-based identification of damage in wind turbine blades using novelty detection | |
| CN110132554B (en) | Rotary machine fault diagnosis method based on deep Laplace self-coding | |
| CN112287889A (en) | Bearing fault intelligent diagnosis method based on compressed sensing and related vector machine | |
| CN104215456B (en) | Plane clustering and frequency-domain compressed sensing reconstruction based mechanical fault diagnosis method | |
| CN112668527B (en) | Ultrasonic guided wave semi-supervised imaging detection method | |
| CN111401136B (en) | Plunger pump cavitation degree detection method and device and terminal | |
| CN112257530A (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
| CN101900789A (en) | Tolerance analog circuit fault diagnosing method based on wavelet transform and fractal dimension | |
| CN109444863A (en) | A kind of estimation method of the narrowband ultrasonic echo number based on convolutional neural networks | |
| CN112990082B (en) | Detection and identification method of underwater sound pulse signal | |
| CN111272886A (en) | Pressure vessel damage positioning method based on convolutional neural network | |
| CN115824519B (en) | Comprehensive diagnosis method for valve leakage faults based on multi-sensor information fusion | |
| CN106814141A (en) | A kind of phased array supersonic compression method based on orthogonal matching pursuit | |
| CN111665050B (en) | Rolling bearing fault diagnosis method based on clustering K-SVD algorithm | |
| CN115127806A (en) | Gear box fault diagnosis method and device based on multi-sensor vibration signals | |
| CN112949402A (en) | Fault diagnosis method for planetary gear box under minimum fault sample size | |
| CN114964782A (en) | Rolling bearing fault diagnosis acoustic emission signal processing method and system based on compressed sensing | |
| CN110057918B (en) | Method and system for quantitatively identifying damage of composite material under strong noise background | |
| Zhang et al. | Bearing fault diagnosis base on multi-scale 2D-CNN model | |
| CN112183297B (en) | Ultrasonic phased array signal sparse feature extraction method |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190709 |
|
| WD01 | Invention patent application deemed withdrawn after publication |