CN104634571B - A kind of Fault Diagnosis of Roller Bearings based on LCD MF - Google Patents
A kind of Fault Diagnosis of Roller Bearings based on LCD MF Download PDFInfo
- Publication number
- CN104634571B CN104634571B CN201510065072.9A CN201510065072A CN104634571B CN 104634571 B CN104634571 B CN 104634571B CN 201510065072 A CN201510065072 A CN 201510065072A CN 104634571 B CN104634571 B CN 104634571B
- Authority
- CN
- China
- Prior art keywords
- fault
- signal
- fractal
- analysis
- scale component
- 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.)
- Expired - Fee Related
Links
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The present invention is a kind of Fault Diagnosis of Roller Bearings based on LCD MF, some intrinsic scale components (ISC) are obtained using local feature Scale Decomposition, choose useful intrinsic scale component and calculate Teager energy operators (TEO) respectively, so as to obtain instantaneous amplitude and instantaneous frequency corresponding to each intrinsic scale component.Then go trend fluction analysis (MFDFA) to analyze the instantaneous amplitude of each intrinsic scale component using multi-fractal, extract multi-fractal features of its generalized Hurst index as intrinsic scale component.Afterwards, dimensionality reduction is carried out using principal component analysis (PCA), using the result of principal component analysis as fault feature vector.By gathering the rolling bearing operational vibration signal under variable working condition in real time, the vibration signal of collection is obtained into corresponding fault feature vector M by the processing of above step, fault feature vector recognized to realize the fault detect of rolling bearing and fault location.
Description
Technical field
The invention belongs to the fault diagnosis technology field of rolling bearing, more particularly to a kind of rolling bearing based on LCD-MF
Method for diagnosing faults.
Background technology
The effect of rolling bearing is support rotating shaft and parts on shaft, and the normal operation position of holding shaft and rotation are smart
Degree, is characterized in that working service convenience, reliable operation, starting performance are good, bearing capacity is higher under medium speed.Rolling bearing
It is the key components and parts commonly used in plant equipment, whether its working condition is normally directly connected to the normal operation of whole production line
State.The failure of rolling bearing can frequently result in being greatly lowered for productivity, and huge property loss is resulted even in when serious.
Run to ensure rolling bearing in the state of good, it is necessary to carry out monitoring and fault diagnosis to rolling bearing.Thus
The fault detection and diagnosis technology of rolling bearing is studied, is tieed up for avoiding major accident, reducing manpower and materials loss and change
Repairing constitution etc. has important theoretical research value and practical application meaning.
In fault diagnosis field, a most important link is feature extraction, and the feature of extraction can roll to characterize
The running status of bearing.But the presence of the non-stationary property of bearing vibration signal and external interference factor adds vibration letter
The difficulty of number feature extraction.Therefore, research emphasis has been placed on feature extracting method by many scholars.For existing at present normal
With diagnostic method, such as time-domain analysis, frequency-domain analysis and energy spectrometer, although they all have extensive academic engineer applied,
But when objective system is time-varying system, its diagnosis effect is unsatisfactory.Therefore, in recent years time frequency analysis by increasingly
In more analyses for applying to time varying signal, such as WIGNER-VILLEDistribution, Wavelet transformation etc..WIGNER-VILLEWhen distribution is bilinearity
The basis of frequency analysis, very high time frequency resolution can be obtained.But work as WIGNER-VILLEIt is distributed for when handling multicomponent data processing
Inevitably it can be influenceed by cross-interference terms.Before although wavelet analysis has deep Fundamentals of Mathematics and is widely applied
Scape, but need to shift to an earlier date using wavelet analysis and wavelet basis is chosen according to the characteristics of signal.In condition monitoring and fault diagnosis height certainly
Today of dynamicization, this shortcoming of wavelet transformation strongly limit it and further develop.Therefore, when occurring many adaptive
Frequency analysis method.Wherein, most widely used is empirical mode decomposition (EMD).Structure and parameter without any basic function
Set in advance, any sophisticated signal can be decomposed into the line of several basic friction angle components (IMF) by empirical mode decomposition in theory
Property superposition.Instantaneous amplitude and instantaneous frequency can be obtained by carrying out Hilbert transform to each basic friction angle component, therefore, this
Method is especially suitable for analysis time varying signal.
In order to overcome some limitations of empirical mode decomposition, such as envelope error and modal overlap, some for EMD occur are spread out
Generation method, such as integrated empirical mode decomposition (EEMD), local feature Scale Decomposition (LMD) and local integrated empirical mode decomposition
(ELMD).Recently there is a kind of new signal decomposition method, i.e. local feature Scale Decomposition (LCD).Pass through the weight to baseline
Original signal can be decomposed into several intrinsic scale components (ISC) by new definition, local feature Scale Decomposition.In addition, retaining
While the advantages of empirical mode decomposition, local feature Scale Decomposition can effectively reduce reactive component and modal overlap.
Fractals is another branch of feature extracting method, is commonly used to disclose the inherent fractal property of signal.By
In the kinetic mechanism that fractal property can reflect under nonlinear system different conditions, therefore, fractal property can be used as feature
Extracted.Common unifractal method has certain limitation, and the entirety point shape that things is only focused on such as unifractal is special
Property, it but have ignored Local Fractal characteristic；Single component in one sophisticated signal has different fractal properties, only uses unifractal
Analysis sophisticated signal is gone substantially to lack enough robustness.Recently, increasing scholar uses multifractal Analysis from more chis
The fractal property of the angle analysis things of degree.In numerous multifractal Analysis methods, multi-fractal removes trend fluction analysis
(MF-DFA) it is a kind of method of new proposition, this method can effectively show the fractal property of things.
The content of the invention
The invention aims to solve conventional Fault Diagnosis of Roller Bearings to lack enough robustness, lack
The adaptability that changes to operating mode and to the classifying quality of different faults degree it is poor etc. the defects of.Therefore, propose a kind of
The Fault Diagnosis of Roller Bearings that trend fluction analysis is combined is gone based on local feature Scale Decomposition and multi-fractal.
The present invention is a kind of Fault Diagnosis of Roller Bearings based on LCD-MF, combine local feature Scale Decomposition and
Multi-fractal Theory carries out fault diagnosis to rolling bearing, using multi-fractal Theory to each intrinsic scale component extraction broad sense
Hurst indexes simultaneously obtain fault feature vector by PCA dimensionality reductions, comprise the following steps：
Step 1: rolling bearing is obtained in normal, inner ring failure, outer ring failure and rolls monomer failure totally four kinds of states
Under time-domain signal.If obtaining N group time-domain signals under every kind of state, every group of time-domain signal includes n sampled data.
Step 2: carrying out local feature Scale Decomposition to every group of time-domain signal, decomposition obtains some intrinsic scale components
(ISC), these intrinsic scale components contain the local feature signal of the different time scales of pending time-domain signal.
Step 3: k intrinsic scale component before choosing, Teager energy is calculated to each intrinsic scale component of selection respectively
Operator (TEO) is measured, so as to obtain instantaneous amplitude and instantaneous frequency corresponding to each intrinsic scale component.
Step 4: trend fluction analysis is gone to divide the instantaneous amplitude of each intrinsic scale component using multi-fractal
Analysis, extracts multi-fractal features of its generalized Hurst index as intrinsic scale component.
Step 5: dimensionality reduction is carried out using pivot analysis (PCA), using the result of pivot analysis as fault feature vector.
Step 6: gathering rolling bearing operational vibration signal in real time, the vibration signal of collection is passed through into step 2 to step
Five processing obtains corresponding fault feature vector, fault feature vector is recognized to realize the fault detect of rolling bearing
And fault location.
Advantages of the present invention is with good effect：
(1) the advantages of making full use of local feature Scale Decomposition to be directed to non-stationary signal, is decomposed into sophisticated signal limited
Individual intrinsic scale component, each intrinsic scale component contain the local feature signal of the different time scales of original signal；In this base
While calculating instantaneous amplitude using Teager energy operators on plinth, the work of demodulation is also played for modulated signal
With being advantageous to the accurate analysis further to fault location.
(2) the intrinsic scale component of high band correspond to M frequency race (depending on M big ISC number for neglecting selection), and
Other ISC components are noise, and therefore, the inventive method takes full advantage of local feature Scale Decomposition method in separated M frequency
Reach the purpose for removing noise while rate race.
(3) multi-fractal is made full use of to go trend fluction analysis to show ability to the good of signal fractal property, from more chis
Spend the main intrinsic scale component of angle analysis extraction；Compared to unifractal analysis method, trend ripple is gone based on multi-fractal
The multifractal Analysis method of dynamic analysis more comprehensively more universal can show signal fractal property, the characteristic vector of extraction
Bearing current state can be preferably characterized, improves the judgement precision of Method for Bearing Fault Diagnosis.
(4) compared with conventional Method for Bearing Fault Diagnosis, the inventive method is directed to same failure mould under different operating modes
The eigenvectors matrix that the unlike signal of formula is obtained has higher registration, shows this method to the event of the same race under variable working condition
It is higher to hinder pattern-recognition degree；The mutual difference of eigenvectors matrix that different faults mode signal is obtained is larger, shows to exchange work
The mutual discrimination of different faults pattern under condition is higher；In addition, the feature obtained by fault mode different faults degree of the same race
The mutual difference of vector matrix is larger, does not occur aliasing, shows that the inventive method can effectively realize fault mode of the same race
Under different faults degree identification.
(5) in view of the desirable features extraction effect of the inventive method, can combine corresponding neutral net can not also combine
Neutral net is directly judged according to characteristic vector, so as to realize that the fault diagnosis of rolling bearing positions；Realize fault diagnosis
Afterwards, the division of fault severity level can also be further carried out according to characteristic vector.Feature extraction can pass through sensor completely
Be connected realization monitoring and real-time diagnosis in real time with computer, and the knowledge excessively professional without related personnel's study only need to be to phase
Closing knowledge has certain understanding to carry out fault diagnosis and failure severity differentiation, and reducing will to the specialty of Operations Analyst personnel
Ask.
Brief description of the drawings
Fig. 1 is the Fault Diagnosis of Roller Bearings entirety flow chart of steps of the present invention；
Fig. 2 is local feature Scale Decomposition decomposition process figure；
Fig. 3 is inner ring fault-signal time-domain diagram in the embodiment of the present invention；
Fig. 4 is normal signal LCD decomposition result figures in the embodiment of the present invention；
Fig. 5 is inner ring fault-signal LCD decomposition result figures in the embodiment of the present invention；
Fig. 6 is outer ring fault-signal LCD decomposition result figures in the embodiment of the present invention；
Fig. 7 is that monomer fault-signal LCD decomposition result figures are rolled in the embodiment of the present invention；
Fig. 8 is inner ring fault-signal ISC in the embodiment of the present invention1Instantaneous amplitude spectrogram；
Fig. 9 is that generalized Hurst index is illustrated under the different faults pattern that fault diameter is 7mils in the embodiment of the present invention
Figure, wherein figure (a) is ISC1Generalized Hurst index schematic diagram, figure (b) be ISC2Generalized Hurst index schematic diagram, scheme (c)
For ISC2Generalized Hurst index schematic diagram；
Figure 10 is that generalized Hurst index is illustrated under the different faults pattern that fault diameter is 14mils in the embodiment of the present invention
Figure, wherein figure (a) is ISC1Generalized Hurst index schematic diagram, figure (b) be ISC2Generalized Hurst index schematic diagram, scheme (c)
For ISC2Generalized Hurst index schematic diagram；
Figure 11 is that generalized Hurst index is illustrated under the different faults pattern that fault diameter is 21mils in the embodiment of the present invention
Figure, wherein figure (a) is ISC1Generalized Hurst index schematic diagram, figure (b) be ISC2Generalized Hurst index schematic diagram, scheme (c)
For ISC2Generalized Hurst index schematic diagram；
Figure 12 is scatter diagram corresponding to data set 1 in table 1；
Figure 13 is scatter diagram corresponding to data set 2 in table 1；
Figure 14 is scatter diagram corresponding to data set 3 in table 1；
The characteristic vector scatter diagram of different faults degree when Figure 15 is inner ring failure in the embodiment of the present invention；
The characteristic vector scatter diagram of different faults degree when Figure 16 is outer ring failure in the embodiment of the present invention；
The characteristic vector scatter diagram of different faults degree when Figure 17 is rolling element failure in the embodiment of the present invention.
Embodiment
Below in conjunction with drawings and examples, the present invention is described in further detail.
Fault Diagnosis of Roller Bearings proposed by the present invention, trend is gone based on local feature Scale Decomposition and multi-fractal
Fluction analysis is combined, and make use of what local feature Scale Decomposition had a clear superiority on processing non-stationary and nonlinear data
Characteristic, and multi-fractal go good analysis characteristic of the trend fluction analysis to the fractal property of signal.Test result indicates that this
Inventive method can effectively realize rolling bearing fault diagnosis, the characteristic vector pair based on the inventive method extraction when operating mode changes
Operating mode change is insensitive, can preferably complete the classification of the rolling bearing fault pattern under variable working condition, and classification results precision is high,
Fault diagnosis can effectively be carried out；On the basis of fault diagnosis is realized, the characteristic vector based on the inventive method extraction can be effective
The different faults degree under fault mode of the same race is distinguished, can preferably complete the differentiation of rolling bearing fault severity.
The overall steps flow chart of Fault Diagnosis of Roller Bearings under the variable working condition of the present invention is as shown in figure 1, specific step
It is rapid as follows：
Step 1: obtain four kinds of states of rolling bearing under time-domain signal, described four kinds of states be respectively normal condition,
Inner ring malfunction, outer ring malfunction and rolling monomer malfunction.
Under rolling bearing running status, with sample frequency set in advance and sampling time, to normal, inner ring failure,
Outer ring failure, the rolling bearing rolled under monomer four kinds of states of failure respectively gather N group vibration signals, and gathered under malfunction
Vibration signal includes the signal under different faults degree.N group vibration signals under every kind of state are exactly the time domain to be obtained letter
Number, if every group of vibration signal has n sampled point.
Step 2: carrying out local feature Scale Decomposition to every group of time-domain signal, decomposition obtains some intrinsic scale components
(ISC), these intrinsic scale components contain the local feature signal of the different time scales of pending time-domain signal.Such as figure
Shown in 2, carry out local feature Scale Decomposition and specifically include：
Step 2.1：Read pending original time domain signal and be assigned to pending sequence x (t), setting counter p's
Initial value is 1, and pending sequence x (t) is assigned into residual signal sequence r (t).
Step 2.2：Determine all M Local Extremum (t of pending sequence x (t)k,xk) (k=1 ..., M), including pole
Big value point and minimum point, two maximum point (minimum point) (tk-1,xk-1) and (tk+1,xk+1) connected with straight line, in
Between minimum point (maximum point) (tk,xk), calculate its corresponding A on this linek(k=1 ..., M)：
Step 2.3：Calculate AkL corresponding to (k=1 ..., M)k(k=1 ..., M)：
Lk=aAk+(1-a)xk a∈(0,1) (2)
Wherein a=0.5, calculate A1And AMWhen, it is necessary to using image method to Local Extremum (tk,xk) (k=1 ..., M) enter
Row continuation, obtains (t0,x0) and (tM+1,xM+1) carry out A1And AMCalculating.
Step 2.4：By all Lk(k=1 ..., M) is connected with SPL SL (t), and it is office to define SL (t) here
The baseline that portion's characteristic dimension decomposes.
Step 2.5：Make and baseline SL (t) is subtracted in sequence x (t), obtain signal difference sequences h (t)=x (t)-SL (t).Inspection
Survey two conditions whether h (t) meets intrinsic scale component needs:In whole time range, Local modulus maxima as evidence, office
Portion's minimum point is negative；For appointing extreme point, xkAnd AkRatio must be consistent：
If meeting two above condition, h (t) is exactly an intrinsic scale component ISC of original time domain signal, first
Obtain for ISC1.Conversely, then using h (t) as new pending sequence x (t)：X (t)=h (t), then go to step 2.2 and hold
OK, until h (t) is an intrinsic scale component, it is denoted as ISCp(t)：
ISCp(t)=h (t) (4)
Step 2.6：Residual signal sequence r (t) is updated, an intrinsic scale component is decomposited in current original series
ISCp(t) afterwards, it is necessary to subtract ISC from current original seriesp(t), the residual signal sequence r (t) of renewal is：
R (t)=r (t)-ISCp(t) (5)
Step 2.7：Judge whether residual signal sequence r (t) is a monotonic function, if so, then terminating local feature chi
Spend decomposable process；Otherwise, using residual signal sequence r (t) as pending sequence x (t), and refresh counter p=p+1, then
Go to step 2.2 execution.If finally give P intrinsic scale component ISC1(t),...,ISCp(t)。
Step 3: k intrinsic scale component before choosing, Teager energy is calculated to each intrinsic scale component of selection respectively
Operator (TEO) is measured, so as to obtain instantaneous amplitude and instantaneous frequency corresponding to each intrinsic scale component.
Obtained all intrinsic scale components to be chosen, the ISC components of high band correspond to several frequency races, and
Other ISC components are noise, it is necessary to cast out to reach noise reduction purpose.Due to preceding several ISC component frequencies highests, only to comprising
The preceding k IMF components of major failure information are extracted, and its Teager is calculated respectively to each intrinsic scale component of selection
Energy operator (TEO)：
ψ [ISC (n)]=ISC2(n)-ISC(n-1)×ISC(n+1) (6)
Wherein, ISC (n) ISC (n-1) and ISC (n+1) are used for the Teager energy operators for calculating each moment, ensure that
Instantaneity and higher temporal resolution.Then, instantaneous amplitude and instantaneous frequency are further calculated：
Wherein, f (n) represents instantaneous frequency, | a (n) | represent instantaneous amplitude.Thus, k intrinsic scale component pair before obtaining
The instantaneous amplitude answered | a1(n)|,...,|ak(n) | composition characteristic vector space w, for further analysis：
So, the characteristic vector space for extracting the instantaneous amplitude composition of the ISC components containing effective fault message reaches first
The purpose of dimensionality reduction, be advantageous to improve fault diagnosis speed and diagnostic accuracy.
Step 4: trend fluction analysis is gone to divide the instantaneous amplitude of each intrinsic scale component using multi-fractal
Analysis, extracts multi-fractal features of its generalized Hurst index as intrinsic scale component.
The fractal property of things is generally existing, and different things has different fractal properties.For a complicated letter
Number, each of which intrinsic scale component also has different fractal properties, analyzes its internal fractal property and is advantageous to preferably
Extract fault signature.Generally existing unifractal only belongs to an example, is that time series is described from a yardstick, so
Single due to yardstick, it is consistent to be likely to appear in sequence different under a certain yardstick its fractal property, causes to obscure.It is and multiple
Divide shape, also referred to as Multi-scale Fractal, can be from more comprehensively, deeply for describing the Local Fractal characteristic of things from different yardsticks
Angle the fractal property of time series is described.Common Multifractal Method has some limitations, such as the time of analysis
Sequence is necessary for stationary time series, otherwise may obtain the result of mistake.Comparatively speaking, multi-fractal goes trend fluctuation point
Analysis will go trend fluction analysis and more adequacies to be combined, and can effectively reduce interference trend, be advantageous to excavate the non-stationary time
More adequacy characteristics in sequence.Multi-fractal goes trend fluction analysis to specifically include：
Step 4.1：For time series xk, its length is N.Then ' support ' Y (i) is：
Step 4.2：Sequence Y (i) is divided into the m not overlapping subsequences with equal length s, wherein m=int (N/s).It is logical
Normal N is not sub-sequence length s integral multiple.In order to make full use of data, after sequence Y (i) is rearranged from back to front still
It is divided into the m not overlapping subsequences with equal length s.So, 2m subsequence is obtained.
Step 4.3：The polynomial trend of each subsequence is fitted using least-squares algorithm, each subsequence
Variance F2(s, v) is represented：
Work as v=1 ..., m
Work as v=m+1 ..., 2m
Wherein yv(i) it is subsequencevPolynomial fitting, polynomial fitting reflect trend removal degree.
Step 4.4：Q rank wave functions are defined as：
For different sub-sequence length s, repeat step 4.2- steps 4.4 are to obtain Fq(s) for q and s function.Pass through
To the fitting of a polynomial of each subsequence, trend present in each subsequence can be removed, so as to be advantageous to local point of identification
Shape feature.
Step 4.5：In order to determine the scaling property of wave function, log (F are analyzed to each qq(s)) between log (s)
Relation, the average value F of wave functionq(s) following power law relation between yardstick s be present：
Fq(s)∝sH(q) (15)
Wherein H (q) is exactly generalized Hurst index, also known as long coefficient correlation, for characterize time in the past sequence pair now and
The influence of time in the future sequence.For rolling bearing, the normal condition being in the past do not ensure that bearing currently still in
Normal condition.But due to the unrepairable of self, the malfunction of rolling bearing can be continued for.It is past based on this point
The time series of normal condition and malfunction is different, its Hurst index to present and future time sequence influence
Difference, therefore Hurst indexes can be used for the fault diagnosis of rolling bearing.In addition, when bearing is in different malfunctions, its
Hurst indexes are also different.For unifractal time series, due to wave function Fq(s) scaling property is for all
Subsequence is all consistent, and Husrt index H (q) (q=2) are uncorrelated to yardstick q.For the time series with multifractal property,
Generalized Hurst index depends on yardstick q, and generalized Hurst index corresponding to different scale q is different.Due to most of unifractals
Obtained under some extreme environments, and multi-fractal is widely present, therefore, multi-fractal can be from wider range of angle
Degree describes the fractal property that each ISC corresponds to instantaneous amplitude.
Characteristic vector space w every a line (instantaneous amplitude corresponding to i.e. each ISC) is entered according to step 4.1- steps 4.5
Row multi-fractal trend fluction analysis, obtains multi-fractal features vector space wM：
Wherein, q=(- m ..., 1,0,1 ..., m).
Step 5: dimensionality reduction is carried out using pivot analysis (PCA), using the result of pivot analysis as fault feature vector.
PCA (Principal Component Analysis), claim principal component analysis, be one for statistical angle
Kind multivariate statistical method.PCA by linear transformation by multiple variables by selecting less significant variable.It can often have
Effect ground obtains most important element and structure from the excessively data message of " abundant ", the noise and redundancy of data is removed, by original
Carry out complicated Data Dimensionality Reduction, disclose the simple structure for being hidden in complex data behind.Due to instantaneous amplitude corresponding to each ISC
Carry out multi-fractal trend fluction analysis, obtained multi-fractal features vector space wMDimension is larger, is directly inputted to nerve
Network may result in the convergence speed for reducing neutral net and the judgement precision for influenceing neutral net.In addition, obtain
In multi-fractal features vector space there is contact, it is necessary to further extract main component in many compositions each other, therefore utilize
PCA analyses carry out dimensionality reduction to obtained multi-fractal features vector space, obtain final characteristic vector space M：
M=[1stPC 2ndPC 3rdPC] (17)
Step 6: gathering rolling bearing operational vibration signal in real time, the vibration signal of collection is passed through into step 2 to step
Five processing obtains corresponding fault feature vector M, fault feature vector is recognized to realize that the failure of rolling bearing is examined
Survey and fault location.
Embodiment：
This example is verified that the external diameter of bearing is 39.04mm using the experimental data of 6205-2RS type deep groove ball bearings,
Thickness is 12mm, pitch diameter 28.5mm, a diameter of 7.94mm of rolling element, and rolling element number is 9, and contact angle is 0 °.Bearing tries
Test and be made up of motor, torque sensor/encoder, dynamometer and the electric apparatus control apparatus that power is 1.5kW, by motor band
Driven input shaft, output shaft band dynamic load.
The sample signal gathered when respectively using rolling bearing normal condition, inner ring failure, outer ring failure, rolling element failure
The Fault Diagnosis of Roller Bearings of trend fluction analysis is gone to enter based on local feature Scale Decomposition and multi-fractal the present invention
Row detection checking, is comprised the following steps that：
Step 1: under rolling bearing running status, rolling bearing is gathered with the sample frequency of setting and sampling time
Normal condition, inner ring failure, outer ring failure and roll the time-domain signal of monomer failure under totally four kinds of states.
Bearing data set is normal, the data of inner ring failure, outer ring failure and rolling element failure, the data set bag of normal condition
Include 95 groups of samples.Such as
Shown in table 1, each data set according to 4 kinds of different operating modes (here operating mode refer to motor load difference, be 0HP, 1HP,
Tetra- gears of 2HP, 3HP, corresponding motor speed is tetra- shelves of 1730r/min, 1750r/min, 1772r/min, 1797r/min
Position) fault diameter (7mils, 14mils and 21mils) different with 3 kinds be divided into 12 Sub Data Sets, and each Sub Data Set includes
23 groups of samples, sample frequency 12kHz.Fig. 3 is the original signal time-domain diagram of one group of inner ring failure of collection, wherein abscissa
The time (unit is s) is represented, ordinate represents that (unit is 10 to amplitude-3m)。
The bearing data set of table 1
(1. 1mil=0.001inch；2. every group of sample includes 5000 data points)
Step 2: carrying out local feature Scale Decomposition to every group of time-domain signal, decomposition obtains some intrinsic scale components
(ISC), these intrinsic scale components contain the local feature signal of the different time scales of pending time-domain signal.Normally
Signal, inner ring fault-signal, outer ring fault-signal and the LCD decomposition results such as Fig. 4~Fig. 7 for rolling monomer fault-signal.Fig. 4
For normal signal LCD decomposition result figures, Fig. 5 is inner ring fault-signal LCD decomposition result figures, and Fig. 6 is outer ring fault-signal LCD points
Result figure is solved, Fig. 7 is rolling monomer fault-signal LCD decomposition result figures.In Fig. 4~Fig. 7, abscissa represents signal acquisition point
Number, ordinate represent that (unit is 10 to amplitude-3m)。
Step 3: k intrinsic scale component before choosing, Teager energy is calculated to each intrinsic scale component of selection respectively
Operator (TEO) is measured, so as to obtain instantaneous amplitude and instantaneous frequency corresponding to each intrinsic scale component.
It is as follows to the theoretical value of Rolling Bearing Fault Character frequency analysis：
By taking r=1750 revs/min of rotating speed as an example,
Outer ring failure-frequency
Inner ring failure-frequency
Rolling element single fault frequency
Wherein, r- bearings rotating speed (rev/min)；N- ball numbers；D- rolling element diameters；D- bearing diameters；∝-rolling element
Contact angle.
Here 3 intrinsic scale components before choosing, Teager energy is calculated to each intrinsic scale component of selection respectively
Operator (TEO), so as to obtain instantaneous amplitude and instantaneous frequency corresponding to each intrinsic scale component.With one group of inner ring fault signature
Analyzed exemplified by signal, after local feature Scale Decomposition, it is calculated to the 1st obtained intrinsic scale component ISC
Teager energy operators, and obtain instantaneous amplitude, its spectrogram is as shown in figure 8, abscissa represents that (unit is frequency in Fig. 8
Hz), ordinate represents that (unit is 10 to frequency amplitude-3m).Inner ring fault characteristic frequency f is clearly visible from figureiAnd its frequency multiplication
Amplitude.Method the having for spectrum analysis of instantaneous amplitude is taken after the change of local feature Scale Decomposition by Teager energy operators
Effect property and advantage are confirmed.
Step 4: trend fluction analysis is gone to divide the instantaneous amplitude of each intrinsic scale component using multi-fractal
Analysis, extracts multi-fractal features of its generalized Hurst index as intrinsic scale component.
Multi-fractal is carried out to the instantaneous amplitude of each intrinsic scale component and removes trend fluction analysis, here yardstick q=(-
2,1,0,1,2), so for each intrinsic scale component ISC of extractionk, its generalized Hurst index isFig. 9, Figure 10 and Figure 11 correspond to respectively fault diameter be 7mils, fault diameter 14mils
And the broad sense Hurst curves under the different faults pattern that fault diameter is 21mils, abscissa represent yardstick q, ordinate table
Show H (q).From Fig. 9, Figure 10 and Figure 11, for ISC1, ISC2, ISC3, normal condition, inner ring failure, outer ring failure and rolling
Broad sense Hurst curves corresponding to kinetoplast failure are different, that is, the rolling bearing of different conditions has different multi-fractals special
Property, the generalized Hurst index in multifractal property is chosen as multi-fractal parameter, can effectively disclose Rolling Bearing Status
Difference.
Step 5: dimensionality reduction is carried out using pivot analysis (PCA), using the result of pivot analysis as fault feature vector.
It is 3 according to the intrinsic scale component number extracted, each intrinsic scale component carries out multi-fractal and removes trend ripple
The yardstick q=(- 2,1,0,1,2) of dynamic analysis, it is known that obtained multi-fractal features vector space wMIn the multi-fractal that includes
Parameter totally 15.For Intelligent rolling bearing failure diagnosis system, feature vector dimension causes fault diagnosis robustness compared with conference
Reduction, or even judgement precision can be reduced.Therefore, high dimensional data is mapped to lower dimensional space using PCA, and extracts main influence
Factor.Here first three principal component, respectively 1stPC, 2ndPC, 3rdPC are mainly extracted, first three principal component composition characteristic to
Quantity space M.
In actual industrial production, motor rotary speed would generally change over time, and its corresponding load also can be different, and this causes
Bearing is usually operated under variable working condition.Because fault characteristic frequency is relevant with bearing rotating speed, when operating mode changes, its fault signature
Frequency can also change therewith.Under the conditions of variable working condition, traditional rolling bearing fault diagnosis based on fault characteristic frequency extraction
Very great fluctuation process can occur for the characteristic vector that method obtains, and this can cause mutually to mix between characteristic vector corresponding to different faults pattern
It is folded, it will to cause the reduction of fault diagnosis precision.Therefore, whether Fault Diagnosis of Roller Bearings is applied to the rolling under variable working condition
Dynamic bearing fault diagnosis becomes particularly significant.Figure 12, Figure 13, Figure 14 are respectively that data set 1 in table 1, data set 2, data set 3 are right
The scatter diagram answered.It is seen that under the conditions of variable working condition, the characteristic vector M obtained based on the inventive method also may be used
To keep good stability, every kind of fault mode is gathered in certain higher dimensional space, is distributed corresponding to different faults pattern
Do not occur aliasing between space.For different fault diameters, such as 7mils, 14mils and 21mils, above-mentioned conclusion still into
It is vertical.Therefore, the characteristic vector based on the inventive method extraction changes insensitive, i.e., rolling bearing proposed by the present invention for operating mode
Method for diagnosing faults, operating mode, which changes caused fluctuation, to be ignored.In addition, between distribution space corresponding to different faults pattern not
The Fault Diagnosis of Roller Bearings proposed by the present invention that appearance aliasing demonstrates proposition can effectively distinguish rolling bearing not
Same fault mode.
It is determined which kind of fault mode rolling bearing is under, it is sometimes desirable to which the fault severity level of rolling bearing is carried out
Assess, therefore, the data set 4 applied the inventive method in table 1 is to verify the inventive method for the different orders of severity
Fault distinguish effect.Here, there are 95 groups of samples under normal condition, fault mode includes inner ring failure, outer ring failure and rolling element
Failure, every kind of fault mode include different fault degree samples (7mils, 14mils, 21mils).When Figure 15 is inner ring failure
The characteristic vector scatter diagram of different faults degree, the characteristic vector scatter diagram of different faults degree when Figure 16 is outer ring failure, figure
17 be respectively rolling element failure when different faults degree characteristic vector scatter diagram.From Figure 11-Figure 17, in failure mould of the same race
Under formula, characteristic vector corresponding to the different faults degree that is obtained based on the inventive method is in different spaces, mutual boundary
Limit is obvious, does not occur aliasing.Therefore, the Fault Diagnosis of Roller Bearings based on the inventive method can be used for recognizing the axis of rolling
The fault severity level held, and precision is higher.
Step 6: gathering rolling bearing operational vibration signal in real time, the vibration signal of collection is passed through into step 2 to step
Five processing obtains corresponding fault feature vector M, fault feature vector is recognized to realize that the failure of rolling bearing is examined
Survey and fault location.
Pass through above Fault Locating Method and the detailed description of result, it is seen that of the invention based on local feature yardstick point
The Fault Diagnosis of Roller Bearings that solution and multi-fractal go trend fluction analysis to be combined has obvious advantage：Operating mode is one
When determining to change in scope, this method need not change parameter, and effectively the Rolling Bearing Fault Character under variable working condition can be carried out
Extraction；On this basis, rolling bearing is in after which kind of fault mode recognizes, can determines whether that rolling bearing is worked as
Preceding fault severity level., can be directly according to feature due to the clear superiority of characteristic vector obtained based on the inventive method
Vector is judged, the intelligent diagnostics of rolling bearing can also be further realized using characteristic vector as the input of neutral net.
Claims (1)
1. a kind of Fault Diagnosis of Roller Bearings based on LCD-MF, combine local feature Scale Decomposition and multi-fractal reason
Fault diagnosis is carried out by rolling bearing, using multi-fractal Theory to each intrinsic scale component extraction generalized Hurst index
And fault feature vector is obtained by pivot analysis PCA dimensionality reductions, it is characterised in that：This method comprises the following steps：
Step 1: rolling bearing is obtained under normal, inner ring failure, outer ring failure and rolling monomer failure totally four kinds of states
Time-domain signal；If obtaining N group time-domain signals under every kind of state, every group of time-domain signal includes n sampled data；Wherein, N, n is just
Integer；
Step 2: carrying out local feature Scale Decomposition to every group of time-domain signal, decomposition obtains some intrinsic scale components, in these
Report the local feature signal that scale component contains the different time scales of pending time-domain signal；
Step 3: k intrinsic scale component before choosing, the calculation of Teager energy is calculated to each intrinsic scale component of selection respectively
Son, so as to obtain instantaneous amplitude and instantaneous frequency corresponding to each intrinsic scale component；
Step 4: going trend fluction analysis to analyze the instantaneous amplitude of each intrinsic scale component using multi-fractal, carry
Take multi-fractal features of its generalized Hurst index as intrinsic scale component；
Step 5: dimensionality reduction is carried out using pivot analysis PCA, using the result of pivot analysis as fault feature vector；
Step 6: gathering rolling bearing operational vibration signal in real time, the vibration signal of collection is arrived into step 5 by step 2
Processing obtains corresponding fault feature vector, fault feature vector is recognized to realize the fault detect of rolling bearing and event
Barrier positioning；
The Fault Diagnosis of Roller Bearings based on LCD-MF makes full use of local feature Scale Decomposition to be directed to non-stationary signal
The advantages of, sophisticated signal is decomposed into limited individual intrinsic scale component, when each intrinsic scale component contains the difference of original signal
Between yardstick local feature signal；On this basis using Teager energy operators calculate instantaneous amplitude while, for by
The signal of modulation also plays the effect of demodulation, is advantageous to the accurate analysis further to fault location；
The Fault Diagnosis of Roller Bearings based on LCD-MF, multi-fractal is made full use of to remove trend fluction analysis to signal point
The good of shape characteristic shows ability, the main intrinsic scale component of extraction from multiple dimensioned angle analysis；Compared to unifractal
Analysis method, the multifractal Analysis method for removing trend fluction analysis based on multi-fractal can more comprehensive more universal exhibition
Existing signal fractal property, the characteristic vector of extraction can preferably characterize bearing current state, improve bearing failure diagnosis side
The judgement precision of method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510065072.9A CN104634571B (en) | 2015-02-06 | 2015-02-06 | A kind of Fault Diagnosis of Roller Bearings based on LCD MF |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510065072.9A CN104634571B (en) | 2015-02-06 | 2015-02-06 | A kind of Fault Diagnosis of Roller Bearings based on LCD MF |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104634571A CN104634571A (en) | 2015-05-20 |
| CN104634571B true CN104634571B (en) | 2017-12-08 |
Family
ID=53213567
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510065072.9A Expired - Fee Related CN104634571B (en) | 2015-02-06 | 2015-02-06 | A kind of Fault Diagnosis of Roller Bearings based on LCD MF |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104634571B (en) |
Families Citing this family (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105973593A (en) * | 2016-04-22 | 2016-09-28 | 北京航空航天大学 | Rolling bearing health evaluation method based on local characteristic scale decomposition-approximate entropy and manifold distance |
| CN105910823A (en) * | 2016-06-21 | 2016-08-31 | 上海电机学院 | Rolling bearing fault diagnosis method |
| CN106096313B (en) * | 2016-06-29 | 2018-12-25 | 潍坊学院 | A kind of envelope Analysis Method based on unusual spectral factorization and spectrum kurtosis |
| CN106289777B (en) * | 2016-08-01 | 2018-09-21 | 北京航空航天大学 | A kind of multi-state rolling bearing performance appraisal procedure based on geometry measurement |
| CN107831012B (en) * | 2017-10-11 | 2019-09-03 | 温州大学 | A kind of Method for Bearing Fault Diagnosis based on Walsh transformation and Teager energy operator |
| CN107917805A (en) * | 2017-10-16 | 2018-04-17 | 铜仁职业技术学院 | Fault Diagnosis of Roller Bearings based on depth belief network and support vector machines |
| CN109697271B (en) * | 2017-10-23 | 2024-02-13 | 新天科技股份有限公司 | Rolling bearing health assessment and performance prediction method based on short-time energy change ratio and nuclear extreme learning machine |
| CN108152037A (en) * | 2017-11-09 | 2018-06-12 | 同济大学 | Method for Bearing Fault Diagnosis based on ITD and improvement shape filtering |
| CN108051189B (en) * | 2017-11-20 | 2019-12-31 | 郑州工程技术学院 | Rotating machinery fault feature extraction method and device |
| CN108287194A (en) * | 2018-01-30 | 2018-07-17 | 青岛理工大学 | Structural damage alarming method based on local feature Scale Decomposition and waveform index |
| CN108416106B (en) * | 2018-02-05 | 2022-02-08 | 江苏方天电力技术有限公司 | Water feeding pump fault detection method based on multi-scale principal component analysis |
| CN108446629A (en) * | 2018-03-19 | 2018-08-24 | 河北工业大学 | Rolling Bearing Fault Character extracting method based on set empirical mode decomposition and modulation double-spectrum analysis |
| CN108830129A (en) * | 2018-03-29 | 2018-11-16 | 南京航空航天大学 | A kind of fault signal of mechanical equipment feature extracting method |
| CN108444698B (en) * | 2018-06-15 | 2019-07-09 | 福州大学 | Epicyclic gearbox Incipient Fault Diagnosis method based on TEO demodulation accidental resonance |
| CN109214097B (en) * | 2018-09-14 | 2021-09-10 | 上海工程技术大学 | Method for predicting long-related fault trend of rolling bearing with dimensionless parameters |
| CN109443719B (en) * | 2018-11-01 | 2020-05-19 | 河南理工大学 | Drill bit vibration signal online virtual test method and system thereof |
| CN111256993A (en) * | 2018-11-30 | 2020-06-09 | 中国电力科学研究院有限公司 | Method and system for diagnosing fault type of main bearing of wind turbine generator |
| CN110146293A (en) * | 2019-06-04 | 2019-08-20 | 昆明理工大学 | A kind of Fault Diagnosis of Roller Bearings based on PCA and ELM |
| TWI706149B (en) * | 2019-12-04 | 2020-10-01 | 財團法人資訊工業策進會 | Apparatus and method for generating a motor diagnosis model |
| CN111735583B (en) * | 2020-06-24 | 2022-01-28 | 东北石油大学 | Pipeline working condition identification method based on LCD-EE pipeline sound wave signal characteristic extraction |
| CN112073345B (en) * | 2020-07-28 | 2021-08-31 | 中国科学院信息工程研究所 | Modulation mode identification method and device, electronic equipment and storage medium |
| CN111982489B (en) * | 2020-08-27 | 2022-05-06 | 江苏师范大学 | Weak fault feature extraction method for selectively integrating improved local feature decomposition |
| CN112633121A (en) * | 2020-12-18 | 2021-04-09 | 国网浙江省电力有限公司武义县供电公司 | Radiation source identification method based on Hilbert transform and multi-fractal dimension characteristics |
| CN113029569B (en) * | 2021-03-11 | 2022-04-29 | 北京交通大学 | Train bearing autonomous fault identification method based on cyclic strength index |
| CN115876511A (en) * | 2022-12-29 | 2023-03-31 | 郑州机械研究所有限公司 | Fault determination method and device for rotating machinery and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103033362A (en) * | 2012-12-31 | 2013-04-10 | 湖南大学 | Gear fault diagnosis method based on improving multivariable predictive models |
| CN103048137A (en) * | 2012-12-20 | 2013-04-17 | 北京航空航天大学 | Fault diagnosis method of rolling bearing under variable working conditions |
| CN104155108A (en) * | 2014-07-21 | 2014-11-19 | 天津大学 | Rolling bearing failure diagnosis method base on vibration temporal frequency analysis |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2012042338A (en) * | 2010-08-19 | 2012-03-01 | Ntn Corp | Roller bearing abnormality diagnosis apparatus and gear abnormality diagnosis apparatus |
-
2015
- 2015-02-06 CN CN201510065072.9A patent/CN104634571B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103048137A (en) * | 2012-12-20 | 2013-04-17 | 北京航空航天大学 | Fault diagnosis method of rolling bearing under variable working conditions |
| CN103033362A (en) * | 2012-12-31 | 2013-04-10 | 湖南大学 | Gear fault diagnosis method based on improving multivariable predictive models |
| CN104155108A (en) * | 2014-07-21 | 2014-11-19 | 天津大学 | Rolling bearing failure diagnosis method base on vibration temporal frequency analysis |
Non-Patent Citations (3)
| Title |
|---|
| 基于LCD和排列熵的滚动轴承故障诊断;郑近德等;《振动、测试与诊断》;20141031;第34卷(第5期);第802-806,971页 * |
| 基于LMD 和平滑Teager能量算子解调的电机滚动轴承故障特征提取;王冰等;《机械传动》;20121231;第36卷(第9期);第89-92页 * |
| 基于多重分形去趋势波动分析的齿轮箱故障特征提取方法;林近山等;《振动与冲击》;20131231;第32卷(第2期);第97-101页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104634571A (en) | 2015-05-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104634571B (en) | A kind of Fault Diagnosis of Roller Bearings based on LCD MF | |
| CN110160791B (en) | System and method for diagnosing faults of induction motor bearing based on wavelet-spectral kurtosis | |
| CN105718876B (en) | A kind of appraisal procedure of ball-screw health status | |
| Yongbo et al. | Review of local mean decomposition and its application in fault diagnosis of rotating machinery | |
| CN102269655B (en) | Method for diagnosing bearing fault | |
| CN106644467B (en) | A kind of gear-box non-stationary signal fault signature extracting method | |
| CN111504645B (en) | Rolling bearing fault diagnosis method based on frequency domain multipoint kurtosis | |
| CN109782603A (en) | The detection method and monitoring system of rotating machinery coupling fault | |
| CN103291600B (en) | Fault diagnosis method for hydraulic pump based on EMD-AR (empirical mode decomposition-auto-regressive) and MTS (mahalanobis taguchi system) | |
| CN109932179B (en) | Rolling bearing fault detection method based on DS self-adaptive spectrum reconstruction | |
| CN105181019A (en) | Computer program product for early fault early-warning and analysis of rotation type machine | |
| CN104215456B (en) | Plane clustering and frequency-domain compressed sensing reconstruction based mechanical fault diagnosis method | |
| CN101534305A (en) | Method and system for detecting network flow exception | |
| CN103616187A (en) | Fault diagnosis method based on multi-dimension information fusion | |
| CN110057581A (en) | Rotary machinery fault diagnosis method based on interval type reliability rule-based reasoning | |
| CN105675274A (en) | Time-domain parameter and D-S evidence theory-based rotor running state monitoring method | |
| CN105626502A (en) | Plunger pump health assessment method based on wavelet packet and Laplacian Eigenmap | |
| CN117251812A (en) | High-voltage power line operation fault detection method based on big data analysis | |
| CN103941722A (en) | Method monitoring and diagnosing equipment failure through component characteristic frequency multiplication amplitude trend | |
| Yi et al. | Second-order Synchrosqueezing Modified S Transform for wind turbine fault diagnosis | |
| Du et al. | Intelligent turning tool monitoring with neural network adaptive learning | |
| Xu et al. | Rolling bearing fault feature extraction via improved SSD and a singular-value energy autocorrelation coefficient spectrum | |
| CN112801033A (en) | AlexNet network-based construction disturbance and leakage identification method along long oil and gas pipeline | |
| CN116738314A (en) | Air compressor fault detection method based on neural network | |
| CN111473975A (en) | Intelligent diagnosis method for secondary gearbox fault based on characteristic vector baseline method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20171208 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |