CN103868690B - Rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics - Google Patents
Rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics Download PDFInfo
- Publication number
- CN103868690B CN103868690B CN201410070392.9A CN201410070392A CN103868690B CN 103868690 B CN103868690 B CN 103868690B CN 201410070392 A CN201410070392 A CN 201410070392A CN 103868690 B CN103868690 B CN 103868690B
- Authority
- CN
- China
- Prior art keywords
- feature
- index
- rolling bearing
- threshold value
- vibration signal
- 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
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000005096 rolling process Methods 0.000 title claims abstract description 32
- 238000000605 extraction Methods 0.000 title claims abstract description 10
- 230000008569 process Effects 0.000 claims abstract description 17
- 230000000295 complement effect Effects 0.000 claims abstract description 12
- 230000003412 degenerative effect Effects 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 238000013459 approach Methods 0.000 claims description 6
- 230000004044 response Effects 0.000 claims description 3
- 230000035945 sensitivity Effects 0.000 claims description 2
- 230000006378 damage Effects 0.000 abstract description 8
- 201000004569 Blindness Diseases 0.000 abstract description 2
- 238000010187 selection method Methods 0.000 abstract 1
- 230000008859 change Effects 0.000 description 7
- 230000004927 fusion Effects 0.000 description 7
- 238000001228 spectrum Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 230000007935 neutral effect Effects 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 208000027418 Wounds and injury Diseases 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 208000014674 injury Diseases 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000013139 quantization Methods 0.000 description 2
- 241000212384 Bifora Species 0.000 description 1
- ZAKOWWREFLAJOT-CEFNRUSXSA-N D-alpha-tocopherylacetate Chemical compound CC(=O)OC1=C(C)C(C)=C2O[C@@](CCC[C@H](C)CCC[C@H](C)CCCC(C)C)(C)CCC2=C1C ZAKOWWREFLAJOT-CEFNRUSXSA-N 0.000 description 1
- 241000220317 Rosa Species 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 239000007795 chemical reaction product Substances 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013524 data verification Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000009661 fatigue test Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 239000012535 impurity Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- SYHGEUNFJIGTRX-UHFFFAOYSA-N methylenedioxypyrovalerone Chemical compound C=1C=C2OCOC2=CC=1C(=O)C(CCC)N1CCCC1 SYHGEUNFJIGTRX-UHFFFAOYSA-N 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics. The method comprises the steps of extracting multiple time domain, frequency domain and time-frequency domain characteristics of a vibration signal of a rolling bearing, and intelligently selecting a characteristics subset which is sensitive to the fatigue recession process of the rolling bearing and can supply complementary information by using a supervision characteristic-free selection method based on maximum correlation and minimum redundancy. The blindness of manually selecting sensitive characteristics without priori knowledge is overcome, an alarm policy is set up according to an alarm threshold value automatic establishing method, and the early damage warning of the rolling bearing is realized.
Description
Technical field
The present invention relates to mechanical fault diagnosis field, more particularly, to a kind of Rolling Bearing Status automatic early warning method.
Background technology
Typically directly choose the single feature conduct of vibration signal based on the Rolling Bearing Status assessment technology of vibration signal
State estimation index.However, with further study show that: single feature generally only certain stage to certain defect effectively,
Same feature shows difference under different fault modes, and different characteristic shows also difference under same operating mode, and effective shape
State evaluation index should be able to make full use of much information, is not only able to catch the internal performance change in the different operation phase for the bearing
Change, be easily obtained in actual applications simultaneously.In order to improve single index performance, some scholar's research feature based in recent years
The index construction method of integration technology, for example, hai qiu et al. 2003 is in document " robust performance
degradation assessment methods for enhanced rolling element bearing
Prognostics " in propose a kind of state estimation index construction method based on self organizing neural network, using vibration signal
And accordingly the root-mean-square value of envelope signal, kurtosis value, waveform index train the minimum quantization error (mqe) obtaining as bearing
State estimation index.But, the method only artificially specifies a few feature to be merged it is impossible to from numerous primitive characters
Middle intelligence constructs the character subset that can stably embody bearing decline pattern, does not formulate corresponding automatic alarm strategy simultaneously.
Content of the invention
For overcoming disadvantages mentioned above, the invention provides a kind of Rolling Bearing Status being extracted based on various features and selecting are certainly
Dynamic method for early warning.
The present invention is achieved through the following technical solutions:
A kind of Rolling Bearing Status automatic early warning method being extracted based on various features and selecting, is comprised the steps:
1) vibration signal of monitored rolling bearing is carried out with various features extraction, constitutive characteristic set, it is preferred that described
Vibration signal is the radial vibration signal at bearing block.
2) using the unsupervised feature selection approach based on maximal correlation minimal redundancy, from characteristic set, intelligent selection goes out
, character subset that complementary information can be provided sensitive to rolling bearing fatigue degenerative process.
3) using self organizing neural network, character subset is merged, build state estimation index.The present invention why
Select self organizing neural network, be because in practical application, the fault type of bearing and fail data cannot obtain, in advance from group
Knit neutral net as a kind of learning algorithm of unsupervised competitive mode it is not necessary to give any target output in advance, also need not know
The type of relationship of road input vector, only just need to carry out Feature Mapping by certain inherent law of input data, be more suitable for engineering
The application of practice.Concretely comprise the following steps: by the use of normal condition vibration signal using the character subset preferably going out as input vector, train
Self organizing neural network, obtains the neuron weight vector of normal condition, then by obtain vibration signal character subset and
The neuron weight vector of all normal conditions of mapping layer is made comparisons, and calculates its Euclidean distance, defines Euclidean distance minimum
Neuron is best match unit (bmu), and using this minimum range as a kind of state estimation index, this minimum range essence is
Input data deviates the distance of normal condition, is defined as minimum quantization error mqe:
mqe=||d-mbmu||
Mqe mqe value in formula；
D vibration signal characteristics subset；
mbmuThe weight vector of bmu.
Mqe value is bigger, represents that the degree of bearing state deviation normal condition is bigger, degree of injury is bigger, therefore passes through to chase after
Track mqe value, can be with quantitative description bearing state.
It is inevitably generated disturbance of data, mqe curve during vibration signals collecting, feature extraction and Feature Fusion
Easily produce a lot of burrs, therefore, using the low frequency of the fine Time-Frequency Localization feature extraction mqe curve of WAVELET PACKET DECOMPOSITION technology
Trend signal is as final state estimation index.
Therefore, application self organizing neural network merges to character subset, can be to different characteristic optimum organization, with single spy
Levy or the feature of a few artificial subjective selection is compared, can truer, accurately and comprehensively reflect that bearing runs different
The internal performance change in stage.
4) automatically set up strategy setting alarm threshold value using alarm threshold value, enter when state estimation index exceedes alarm threshold value
Row early warning.
General threshold value set-up mode is with reference to certain rule, such as iso10816, then comes according to the experience of user
Final given threshold.But, iso10816 simply to set this vibration equipment and can connect according to the size of equipment and mounting means
The degree being subject to, and specify only root-mean-square value for unique monitoring index without referring to the setting offer of the threshold value of other indexs
Lead.2006, ginart et al. was in document " automated feature selection for embeddable
Prognostic and health monitoring (phm) architectures " in, according to the maximum time constant of equipment
Propose a kind of automatic establishing method of more general alarm threshold value with the statistical property of General System threshold value.The method basic
Step is: first, using normal condition data, calculates the amplitude fluctuations size of state estimation index and the ratio of time span, that is,
The unit-step response of first-order system increases slope | m |, calculates mean μ and the standard deviation sigma of state estimation index simultaneously, will report to the police
Threshold value initial setting is μ；Then look up alarm index table and obtain corresponding alarm index, be multiplied by initial setting with alarm index
Alarm threshold value, obtains final alarm threshold value.Accurate shape is obtained based on the character subset preferably going out based on before the present invention
State evaluation index, therefore for improving the effect of the present invention further, the alarm threshold value that available ginart et al. proposes is set up automatically
Strategy setting alarm threshold value, carries out early warning when state estimation index exceedes alarm threshold value.
It is preferred that it is the vibration extracting monitored rolling bearing from different perspectives that the various features in described step 1) are extracted
Signal characteristic；Because the feature species of vibration signal is a lot, the present invention also studied and which feature to carry out extraction to and can ensure spy
Levy the comprehensive of extraction and high efficiency, finally draw, the time domain that the various features of said extracted can at least include vibration signal refers to
The frequency of mark, the time domain index of the envelope signal of vibration signal and frequency-domain index, the second generation wavelet packet band signal of vibration signal
Domain index and energy indexes.Because the vibration signal in the time domain index of vibration signal described above substantially just refers to acquired original
The vibration signal obtaining, and the second generation wavelet packet band signal of the envelope signal of described vibration signal and vibration signal is all
On the basis of the vibration signal that acquired original obtains, process draws, therefore, write for convenience and be easy to distinguish, subsequently shake
The time domain index of dynamic signal is referred to as the time domain index of primary signal, and the time domain index of the envelope signal of vibration signal and frequency domain refer to
Mark is referred to as time domain index and the frequency-domain index of envelope signal, the frequency-domain index of the second generation wavelet packet band signal of vibration signal
It is referred to as frequency-domain index and the energy indexes of second generation wavelet packet band signal with energy indexes.This various features is extracted and be ensure that
Needed for assessment bearing state, characteristic information is comprehensive, breaches existing selection artificial in the case of no priori sensitive
The way of feature, so that information is not missed, is that the accuracy of end product provides requisite precondition.
Further, most sensitive to bearing fatigue degenerative process in order to ensure preferred feature subset, ensure state simultaneously
Assessment ageing, described step 2) in the unsupervised feature selection approach based on maximal correlation minimal redundancy can be according to following
Step is carried out:
First, delete and the incoherent feature of bearing degenerative process；
Then, according to the information comprising in homogenous characteristics is similar, the mode of message complementary sense carries out spy between different category features
Levy classification；Classification number is automatically determined by clustering algorithm, is typically no less than two classes.
Finally, feature based score assessment feature local hold capacity, retains in remaining each category feature and locally preserves energy
1～3 optimum feature of power, is organized paired rolling bearing fatigue degenerative process sensitivity, can be provided the feature of complementary information with this
Subset.
The present invention is that the various features that said extracted is obtained carry out rational remove impurity, screening and reservation progressively,
Have studied the impact to result for the various processing methods, through many experiments and trial, finally give following processing procedure:
Deleted and the incoherent feature of bearing degenerative process based on statistical variance (variance) method.It specifically walks
Suddenly it is: calculate the variance fraction of each feature in initial characteristicses set, variance fraction is more than the spy of given threshold
Levy and be defined as uncorrelated features, deleted.The ultimate principle of variance method is that the inter- object distance of similar sample is less, no
The between class distance of similar sample is bigger, then this feature separability is stronger, specifically can be found in the books that bishop cm1995 writes
《neural networks for pattern recognition》.The feature samples being adapted to state estimation should be with table
Levy the gradual change degenerative process of bearing, the feature samples amplitude of normal condition is typically small and fluctuates less, and sample size is also more, because
And the variance fraction of feature is less.The variance fraction computing formula of r-th feature is as follows:
In formula, vrFor variance fraction, friFor i-th sample of r-th feature, n is the sample number of r-th feature,
μrMeansigma methodss for r-th feature.
Based on Pearson (pearson) correlation coefficient by feature according to the information comprising in homogenous characteristics is similar, inhomogeneity
Between feature, the mode of message complementary sense carries out tagsort.It concretely comprises the following steps: calculates the pearson phase relation between residue character
Matrix number, according to correlation matrix, using clustering algorithm, feature is classified.Pearson correlation coefficient is by karl
A kind of variables measure that pearson proposed in the eighties in 19th century.It is preferably no between character subset feature
Association or weak related, pearson correlation coefficient can effectively reflect the degree of association between feature, and its absolute value is bigger,
Illustrate that between feature, degree of association is higher, redundancy is bigger, belong to homogenous characteristics.Two features f being respectively provided with n sample pointsx=
[fx1,f,...,fxn] and fy=[fy1,fy2,...,fyn] between pearson correlation coefficient be:
In formula, ρxyFor pearson correlation coefficient,WithIt is characterized the meansigma methodss of x and feature y, f respectivelyxiAnd fyiRespectively
It is characterized i-th sample of x and feature y.Carry out the clustering algorithm of tagsort: randomly select a certain feature as in cluster
The heart, calculates the pearson correlation coefficient between this feature and other features, and correlation coefficient absolute value is attributed to this more than threshold value
Class；Be new cluster with the feature less than this threshold value, choose a certain feature as the center of cluster, calculate this feature and residue character it
Between correlation coefficient absolute value, be classified as a class more than threshold value；And so on just according to pearson correlation matrix by feature
Classified.Classification number is automatically determined by this clustering algorithm, is typically no less than two classes.
Feature based Laplce (laplacian) score assessment feature local hold capacity, by local hold capacity relatively
The feature of difference is deleted, and finally retains 1～3 optimum feature of local hold capacity in remaining each category feature.It is concrete
Way is: calculates the laplacian score of each feature, deletes the feature that score is more than given threshold, retains remaining each
1～3 minimum feature of score in category feature.Laplacian score is xiaofei he et al. 2005 in document
Propose in " laplacian score for feature selection ", it is based on, and laplacian expands and local keeps
Projection, according to each feature, in its characteristic dimension, the weight relationship and its nearest neighbor point between calculates obtaining of this feature
Point.The laplacian score of feature is lower, then the local hold capacity of this feature is better, more sensitive to bearing degenerative process.With
lrRepresent the laplacian score of r-th feature, friRepresent i-th sample of r-th feature, i=1 ..., m, laplacian
Score calculating process is as follows:
A. construction one has the neighbour figure g of m node, and i-th node corresponds to xiIf, xiIt is xjK nearest Neighbor Points
One of or xjIt is xiOne of k nearest Neighbor Points, then carry out Neighbor Points line between node i and j.
If b. there being connection between node i and j, then give this connection one weightT is constant, no
Then give weight sij=0.The weight matrix s of figure g reflects the partial structurtes of data space.
C. for r-th feature, it is defined as follows: fr=[fr1,fr2,...,frm]t, d=diag (sb), b=[1 ..., 1
]t,l=d-s.Order
D. by the laplacian score of following formula r-th feature of calculating:
Compared with prior art, its advantage is the present invention:
The present invention by based on various features extract and select Rolling Bearing Status automatic early warning method, in Feature Fusion
Multiple time domains of the bearing vibration signal under front extraction normal condition, frequency domain and time and frequency domain characteristics, then using based on
The unsupervised feature selection approach intelligent selection of related greatly minimal redundancy is sensitive to rolling bearing fatigue degenerative process, can provide
The character subset of complementary information, then using the character subset preferably going out as input vector, train self organizing neural network, and then will
The character subset of the vibration signal of whole monitoring period of time is merged, and draws the most rational state estimation index, finally by report
Alert threshold value sets up strategy setting alarm threshold value automatically, carries out early warning when state estimation index exceedes alarm threshold value, so far completes
Automatic, the accurate early warning of rolling bearing earlier damage, the method overcome in the case of no priori artificial choose quick
The blindness of sense feature, the drawbacks of also overcoming this area and be typically with artificial experience specific characteristic and merged, carries significantly
The state estimation obtaining index and alarm threshold value are also set up strategy by the high robustness of state estimation index and reliability automatically
Perfect adaptation, it is achieved that the automatic early-warning of rolling bearing earlier damage, is that security industry produces and equipment high efficiency runs and provides
Greatly ensure.
Brief description
For ease of explanation, the present invention is described in detail by following specific embodiments and accompanying drawing.
Fig. 1 is the flow chart of the present invention；
Fig. 2 is the variance score chart of each feature of the embodiment of the present invention；
Fig. 3 is the laplacian shot chart of each category feature of the embodiment of the present invention；
Fig. 4 is that the alarm threshold value of the embodiment of the present invention sets up result automatically.
Specific embodiment
Flow process as shown in Figure 1, the present invention is with the rolling in shipborne satellite communication antenna bearingt shaft transmission accelerated fatigue test
Effectiveness of the invention is verified as a example dynamic bearing vibration signal.Shipborne satellite communication antenna bearingt shaft transmission mainly by motor, OK
Star decelerator and gear-box are constituted, and comprise 4 rolling bearings in gear-box, are located at the upper and lower side of input shaft and output shaft respectively.
In experimentation, the rotating speed of motor is 1500rpm, using pcb333b32 vibration acceleration sensor and econ avant one
Formula data collection and analysis instrument gathers the radial vibration acceleration signal at each bearing block of gear-box, and sample frequency is 20khz, adopts
Number of samples is 20480, vibration data of every 20 minutes records, after experiment carries out 600 hours, gearbox input shaft upper end bearing
Observable minor injury in inner ring, after experiment carries out 720 hours, this bearing failure.For verifying effectiveness of the invention, for
The life-cycle vibration signal of gearbox input shaft upper end rolling bearing carries out analysis below.
Various features are extracted.This method is unsupervised approaches, only needs normal service data without fault and mistake
Effect data may be selected by out, character subset that can provide complementary information sensitive to rolling bearing fatigue degenerative process, thus
More convenient engineer applied, because the bearing fault fail data installed in practical engineering application on a particular device is difficult to obtain,
And bearing damage is possible to bringing on a disaster property consequence.Carry out feature extraction and the time span of the normal service data of bearing selecting
It is typically chosen as the 1/4～1/3 of bearing service life conservative estimation value.To extract the normal shape of gearbox input shaft upper end rolling bearing
To carry out method validation as a example the feature of 0～150 hour vibration signal of state, to extract 154 features altogether, including the 15 of primary signal
Individual time domain index, 15 time domain indexes of envelope signal and 12 frequency-domain index, 96 of 8 band signals of second generation wavelet packet
Frequency-domain index and 16 energy indexes.Time domain index, frequency-domain index, the expression way of energy indexes are shown in Table 1, wherein, xlFor letter
Number sequence, ylFor frequency spectrum after fast Fourier transform for the signal time domain sequences, x (i) is second generation wavelet packet band signal, n
For sampling number, ppmmFor failure-frequency maximum amplitude and passband signal average ratio, apmmFor failure-frequency maximum amplitude and fault
Frequency neighborhood average ratio.
Table 1 vibration performance index expression formula
Feature selection.A) delete and the incoherent feature of bearing degenerative process.The variance calculating 154 features divides
Number, given threshold is 1.5 times of all feature variance score average, and result is as shown in Fig. 2 in figure tag number and feature
Corresponding relation is: 1 15 is primary signal time domain index；16 30 is envelope signal time domain index, and 31 42 is envelope signal
Frequency-domain index；43 138 is the frequency-domain index of 8 band signals of second generation wavelet packet, 12 indexs of each frequency band, band number
By arranging from small to large, 139 146 is 8 frequency band energy indexs of second generation wavelet packet, and 147 154 is second generation wavelet packet 8
Individual frequency band energy proportion index, so-called frequency band energy proportion index, it is shown in Table 1.Delete variance fraction and be more than given threshold
Feature, according to Fig. 2, deletes 37 features altogether, is envelope spectrum a respectivelypmm, power spectrum ppmmAnd apmm, 8 frequencies of second generation wavelet packet
The envelope spectrum p of band signalpmmAnd apmm, power spectrum ppmmAnd apmm, and second generation wavelet packet the 5th frequency band energy proportion and the 8th frequency
Band energy proportion.
B) carry out tagsort.Remaining 117 features are classified according to pearson correlation coefficient, correlation coefficient
The span of absolute value be [0,1], if by degree of association between feature be more than 0.7 be classified as a class it is ensured that homogenous characteristics it
Between height correlation, the information comprising is similar, and weak related or separate between heterogeneous characteristics, message complementary sense.Still by spy
Levy a degree of association be more than 0.7 be classified as a class, according to clustering algorithm, 117 features have been divided into 6 big class, and classification results are shown in Table
2.This provides benefit for feature selection below: it is desirable that the feature in classification 1～6 is in the character subset carrying out Feature Fusion
In can include because they are belonging to different classes of, message complementary sense, but again must not retain too many spy in each classification
Levy, because the information that in classification, feature comprises too redundancy, not only these redundancies can not bring any benefit to state estimation,
The amount of calculation of Feature Fusion, impact state estimation cost and real-time also can be increased.
Table 2 tagsort result
C) feature based score assessment feature local hold capacity, is deleted deleting the poor feature of local hold capacity
Remove, finally retain 1～3 optimum feature of local hold capacity in remaining each category feature.Calculate each feature
Laplacian score, threshold value is set as the average of all laplacian scores, retains the spy that laplacian score is less than threshold value
Levy, delete remaining feature, result is as shown in figure 3, in figure tag number and feature corresponding relation are: 1 48 is the first category feature,
49 97 is the second category feature, and 98 104 is the 3rd category feature, and 105 114 is the 4th category feature, and 115,116 is that the 5th class is special
Levy, 117 is the 6th category feature, and every category feature is arranged from small to large by sequence number in table 2.The present embodiment is chosen from 6 category features altogether
8 feature composition characteristic subsets below: the envelope spectrum ppmm in the first category feature, second generation wavelet packet the 7th band signal envelope
Spectrum failure-frequency energy, second generation wavelet packet the 1st frequency band energy, the minima in the 3rd category feature and the bag in the second category feature
Network signal maximum, the envelope signal peak index in the 4th category feature, second generation wavelet packet the 2nd frequency band in the 5th category feature
Energy proportion, second generation wavelet packet the 6th frequency band energy proportion in the 6th category feature.
Feature Fusion.Using normal condition data, preferred 8 features are formed input vector, train from group after normalization
Knit neutral net, network initial parameter select be: topological structure is hexagon, and learning rate is 0.9, distance function be Euclidean away from
From output layer dimension is 10 × 10, and frequency of training is 500 times.Can see, until this step, using be all normal shape
State data, this is that feature selection is designed to unsupervised reason by the present invention, is also to select self organizing neural network to carry out feature
The reason fusion: do not need fault and fail data just can carry out feature selection and Fusion training, more convenient engineer applied.Existing
Effectiveness using life-cycle data verification preceding method.8 characteristic indexs of life-cycle vibration data are input to after training
Neutral net, and constantly calculate corresponding mqe value, extract mqe trend signal as state estimation index.
Alarm threshold value is set up automatically.Mqe trend signal of change using normal condition data goes out mean μ=0.0767, standard
Difference σ=0.0056, the unit-step response of first-order system increase slope | m |=1.1012e-5, search alarm index table and obtain accordingly
Alarm index be 3, so using ginart et al. propose alarm threshold value automatically set up strategy alarm threshold value is set to
0.2301, result such as Fig. 4.Visible threshold set up using be also normal condition data it is not necessary to fault data.
Can be seen by Fig. 4, according to data display, state estimation index mqe value characterizes the decline track of bearing, from opening
Begin to run to 604 hours, mqe value changes are slow and amplitude is less, and bearing is in normal condition；604 hours, state estimation index
Mqe value reaches 0.2403, exceedes alarm threshold value, and subsequent mqe value rises comparatively fast, illustrates that bearing begins to deviate from normal operating condition,
Occur in that initial damage；640 hours, mqe value declined, and big ups and downs, illustrates that bearing is in peeling Growth period, with stripping
Sharp, the continuous fluctuation of mqe value constantly sliding is arrived by sharp to being polished again in the border that falls；705 hours, mqe value exponentially rose, bearing
It is in expiration date.Experimental result is coincide with actual observation situation.
Therefore, can be seen by this embodiment, state estimation index can catch rolling bearing and run each stage
Performance change, the present invention can identify the boundary of normal condition and initial damage state exactly.
The above, the only specific embodiment of the present invention, but protection scope of the present invention is not limited thereto, and any
The change or replacement expected without creative work, all should be included within the scope of the present invention.Therefore, the present invention
Protection domain should be defined by the protection domain that claims are limited.
Claims (2)
1. a kind of based on various features extract and select Rolling Bearing Status automatic early warning method it is characterised in that include as
Lower step:
1) various features extraction, constitutive characteristic set are carried out to the vibration signal of monitored rolling bearing, described vibration signal is
Radial vibration signal at bearing block；
2) selected to the axis of rolling from characteristic set using the unsupervised feature selection approach based on maximal correlation minimal redundancy
Hold the character subset that tired degenerative process is sensitive, can provide complementary information；
3) using self organizing neural network, character subset is merged, build state estimation index；
4) automatically set up strategy setting alarm threshold value using alarm threshold value, carry out when state estimation index exceedes alarm threshold value pre-
Alert；
Wherein, described step 2) in comprised the following steps based on the unsupervised feature selection approach of maximal correlation minimal redundancy:
Delete and the incoherent feature of bearing degenerative process；
According to the information comprising in homogenous characteristics is similar, between different category features, the mode of message complementary sense carries out tagsort, class
Other number is automatically determined by clustering algorithm, no less than two classes；
Feature based score assessment feature local hold capacity, retains 1 of hold capacity optimum in local in remaining each category feature
～3 features, are organized paired rolling bearing fatigue degenerative process sensitivity, can be provided the character subset of complementary information with this；
Described step 4) described in alarm threshold value automatically set up the step of strategy to comprise: first, using normal condition data, count
Calculate the amplitude fluctuations size of state estimation index and the ratio of time span, i.e. the unit-step response of first-order system increases slope | m
|, calculate mean μ and the standard deviation sigma of state estimation index simultaneously, alarm threshold value initial setting is μ；Then look up alarm index
Table obtains corresponding alarm index, is multiplied by the alarm threshold value of initial setting with alarm index, obtains final alarm threshold value.
2. the Rolling Bearing Status automatic early warning method being extracted based on various features and selecting according to claim 1, its
Be characterised by, described step 1) in the vibration signal to monitored rolling bearing carry out various features extract be from different perspectives
Extract the vibration signal characteristics of monitored rolling bearing, described various features at least include the time domain index of vibration signal, vibration
The time domain index of the envelope signal of signal and frequency-domain index, the frequency-domain index of second generation wavelet packet band signal of vibration signal and
Energy indexes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410070392.9A CN103868690B (en) | 2014-02-28 | 2014-02-28 | Rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410070392.9A CN103868690B (en) | 2014-02-28 | 2014-02-28 | Rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103868690A CN103868690A (en) | 2014-06-18 |
| CN103868690B true CN103868690B (en) | 2017-02-01 |
Family
ID=50907463
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410070392.9A Expired - Fee Related CN103868690B (en) | 2014-02-28 | 2014-02-28 | Rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103868690B (en) |
Families Citing this family (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104198184A (en) * | 2014-08-11 | 2014-12-10 | 中国人民解放军空军工程大学 | Bearing fault diagnostic method based on second generation wavelet transform and BP neural network |
| CN105620290B (en) * | 2015-12-23 | 2018-01-16 | 郑州日产汽车有限公司 | Driving motor of pure electric automobile failure real time early warning method based on power spectrumanalysis |
| CN105975749B (en) * | 2016-04-28 | 2019-03-12 | 上海交通大学 | A method and system for bearing health assessment and prediction |
| JP2018004473A (en) * | 2016-07-04 | 2018-01-11 | ファナック株式会社 | Mechanical learning device for learning estimated life of bearing, life estimation device, and mechanical learning method |
| CN106323635B (en) * | 2016-08-04 | 2018-10-19 | 中国航空综合技术研究所 | A kind of rolling bearing fault on-line checking and state evaluating method |
| DE102016117190A1 (en) * | 2016-09-13 | 2018-03-15 | fos4X GmbH | A method and apparatus for monitoring a condition of at least one wind turbine and computer program product |
| SG11201902898WA (en) * | 2016-10-14 | 2019-05-30 | Univ Osaka Prefect Public Corp | Swallowing diagnosis apparatus and program |
| CN108663995B (en) * | 2017-03-30 | 2020-11-06 | 郑州大学 | Method and device for detecting abnormal trend of industrial process variable |
| CN107831024B (en) * | 2017-04-11 | 2020-06-16 | 上海发电设备成套设计研究院 | Diagnosis method of fan vibration fault state based on eigenvalues of multi-point vibration signal |
| CN107655690A (en) * | 2017-10-30 | 2018-02-02 | 王才旺 | Motor bearings method for diagnosing faults based on artificial neural network |
| CN108985898B (en) * | 2018-07-12 | 2022-02-15 | 广东工业大学 | Site scoring method and device and computer readable storage medium |
| CN108833596B (en) * | 2018-08-05 | 2019-06-21 | 北京吉宝通科技发展有限公司 | A kind of circumference vibration alarming system based on Internet of Things |
| CN109446877B (en) * | 2018-09-01 | 2022-01-07 | 哈尔滨工程大学 | Radar radiation source signal modulation identification method combined with multi-dimensional feature migration fusion |
| CN109597315B (en) * | 2018-10-31 | 2020-08-18 | 华中科技大学 | A kind of mechanical equipment health degradation state identification method, equipment and system |
| CN109708907B (en) * | 2018-11-30 | 2020-10-16 | 浙江胄天科技股份有限公司 | Equipment fault feature extraction method based on envelope information |
| CN109632355B (en) * | 2018-12-20 | 2021-03-16 | 广州航天海特系统工程有限公司 | Fault prediction method and system based on electromechanical device state data drift |
| CN110208377A (en) * | 2019-06-19 | 2019-09-06 | 南京邮电大学 | A kind of more characteristic parameters damage degree assessment method based on Lamb wave |
| CN110186684A (en) * | 2019-06-25 | 2019-08-30 | 东北大学 | A kind of aero-engine mechanical oscillation fault-signal feature extracting method |
| CN110262460B (en) * | 2019-07-01 | 2020-07-03 | 浪潮集团有限公司 | Concrete piston fault prediction method for extracting features by combining clustering idea |
| CN110412467B (en) * | 2019-07-30 | 2021-07-23 | 重庆邮电大学 | A Lithium Battery Fault Data Screening Method Constrained by Normalized Mutual Information Criterion |
| CN110674725B (en) * | 2019-09-20 | 2022-06-03 | 电子科技大学 | Device Signal Type Recognition Method Based on Multidimensional Feature Vector Combination of Detection Signals |
| DE102019219772A1 (en) * | 2019-09-26 | 2021-04-01 | Robert Bosch Gmbh | Sensor system, linear device and method for a sensor system |
| CN111652031A (en) * | 2019-12-02 | 2020-09-11 | 昆明理工大学 | A Fault Diagnosis Method for Rolling Bearings Based on Improved Empirical Wavelet Transform |
| CN111597682B (en) * | 2020-04-14 | 2023-03-31 | 新疆大学 | Method for predicting remaining life of bearing of gearbox of wind turbine |
| CN112326245B (en) * | 2020-10-21 | 2023-03-10 | 中国航空工业集团公司上海航空测控技术研究所 | Rolling bearing fault diagnosis method based on variational Hilbert-Huang transform |
| CN112284785B (en) * | 2020-10-26 | 2022-12-06 | 华能南京金陵发电有限公司 | Coal mill fault state diagnosis method based on unsupervised clustering algorithm |
| CN113255591A (en) * | 2021-06-25 | 2021-08-13 | 四川九通智路科技有限公司 | Bearing fault diagnosis method based on random forest and fusion characteristics |
| CN113569351A (en) * | 2021-07-08 | 2021-10-29 | 北京远舢智能科技有限公司 | Industrial equipment fault diagnosis method, device and system based on PHM technology |
| CN114117734A (en) * | 2021-10-22 | 2022-03-01 | 中国华能集团清洁能源技术研究院有限公司 | Equipment life detection method and device, electronic equipment and storage medium |
| CN114509266B (en) * | 2022-01-05 | 2023-12-01 | 东南大学 | Bearing health monitoring method based on fault feature fusion |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102721545B (en) * | 2012-05-25 | 2015-05-20 | 北京交通大学 | Rolling bearing failure diagnostic method based on multi-characteristic parameter |
| CN102788695B (en) * | 2012-07-18 | 2014-12-10 | 南京航空航天大学 | Identification method of rolling bearing abrasion |
-
2014
- 2014-02-28 CN CN201410070392.9A patent/CN103868690B/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN103868690A (en) | 2014-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103868690B (en) | Rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics | |
| CN106769052B (en) | A kind of mechanical system rolling bearing intelligent failure diagnosis method based on clustering | |
| CN102112933B (en) | Error detection method and system | |
| CN111459700B (en) | Equipment fault diagnosis method, diagnosis device, diagnosis equipment and storage medium | |
| CN108535635B (en) | EEMD and HMM based analog circuit intermittent fault diagnosis method | |
| CN112254964A (en) | Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network | |
| CN109754110B (en) | Early warning method and system for traction converter faults | |
| Martin-del-Campo et al. | Online feature learning for condition monitoring of rotating machinery | |
| Guo et al. | Deep convolution feature learning for health indicator construction of bearings | |
| CN108304348B (en) | A Bearing Remaining Life Prediction Method Based on Binary Wiener Process | |
| Zheng et al. | Refined time-shift multiscale normalised dispersion entropy and its application to fault diagnosis of rolling bearing | |
| Hoseinzadeh et al. | Modifying the Hilbert-Huang transform using the nonlinear entropy-based features for early fault detection of ball bearings | |
| CN109255395A (en) | A kind of life-span prediction method of ball screw assembly, | |
| Xu et al. | A selective fuzzy ARTMAP ensemble and its application to the fault diagnosis of rolling element bearing | |
| CN107782551A (en) | Method for evaluating damage degree of mechanical part | |
| Desmet et al. | Leak detection in compressed air systems using unsupervised anomaly detection techniques | |
| CN109255201A (en) | A kind of ball screw assembly, health evaluating method based on SOM-MQE | |
| Abburi et al. | A closer look at bearing fault classification approaches | |
| CN112634945A (en) | Intelligent water leakage sound identification method based on cloud platform | |
| CN113505639B (en) | Rotary machine multi-parameter health state assessment method based on TPE-XGBoost | |
| Vununu et al. | Automatic fault diagnosis of drills using artificial neural networks | |
| CN106909136A (en) | A kind of method for diagnosing faults based on index regularization kernel linear discriminant analysis | |
| CN108615018A (en) | Object state identification method based on the extraction of time domain histogram feature | |
| Zhou | Enhanced feature extraction for machinery condition monitoring using recurrence plot and quantification measure | |
| Hao et al. | New fusion features convolutional neural network with high generalization ability on rolling bearing fault diagnosis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170201 Termination date: 20190228 |