CN104897403A - Self-adaption fault diagnosis method based on permutation entropy (PE) and manifold-based dynamic time warping (MDTW) - Google Patents

Abstract

The present invention discloses a self-adaption fault diagnosis method based on permutation entropy (PE) and manifold-based dynamic time warping (MDTW), enabling a bearing fault diagnosis process to be systematic and raising handleability and real-time performance of the diagnosis method. Firstly, a nonlinear and nonstationary bearing vibration signal is decomposed into a plurality of single-package components by applying an adaptive time-frequency analysis method; the adaptive time-frequency analysis method may be selected from empirical mode decomposition, local mean decomposition and local characteristic-scale decomposition methods; and then, extracting the PE of each single-package component as a fault signature. The PE can reflect complexity of the signal and has high robustness and rapidity. The MDTW method is provided by the present invention so as to rapidly and accurately measure distance test data and training data, thereby determining the current fault state and realizing bearing fault diagnosis, and the method has excellent practical engineering application values.

Description

A kind of adaptive failure diagnostic method improving dynamic time warping based on arrangement entropy and stream shape

Technical field

The present invention relates to the technical field of bearing variable working condition fault diagnosis, be specifically related to a kind of based on arrangement entropy (permutation entropy, PE), the adaptive failure diagnostic method that shape improves dynamic time warping (manifold-based dynamic time warping, MDTW) is flowed.

Background technology

Bearing is widely used in rotating machinery, and its health status directly affects the normal operation of whole rotating machinery, and then affects whole system.In recent years, bearing failure diagnosis has become the focus of research, especially based on the fault diagnosis of vibration signal, has had more effective method at present.The process of bearing failure diagnosis mainly comprises fault signature extraction and malfunction determines two aspects.The inventive method is intended to make bearing failure diagnosis process more adding system, efficient, easy to operate, and ensures good real-time characteristic.

For how extracting effective fault signature, the key of problem is the bearing vibration signal how processing nonlinear and nonstationary.Traditional time domain or frequency-domain analysis method are inapplicable in this case.In recent years, researcher proposes some Time-Frequency Analysis Method, adaptive Time Frequency Analysis method wherein obtains a large amount of concerns, typical method has: empirical mode decomposition (empirical mode decomposition, EMD), local mean value decomposes (local mean decomposition, LMD) and local characteristic dimension decompose (local characteristic-scale decomposition, LCD).EMD is put forward by people such as N.E.Huang for 1998, then in mechanical fault detection and diagnosis, obtains a large amount of application.Subsequently, 2005, Jonathan S.Smith proposed LMD.Compared with EMD, LMD maintains better local feature, avoids deficient envelope and crosses envelope problem, and provides more reasonably physical interpretation for single component component.LCD is put forward on the basis of EMD by scholars such as Cheng Junsheng for 2012, and owing to decreasing invalid component and avoiding mode confounding issues, the effect of LCD is better than EMD.Although, it was better than EMD that LMD and LCD was proved effect, but the effect of LMD and LCD did not contrast, and EMD has achieved good effect since birth in a lot of troubleshooting issue, so the inventive method does not specify adaptive Time Frequency Analysis method.In practical implementation, the vibration signal of these three kinds of methods to bearing can be utilized respectively to process, and comparing result determines final plan.After decomposition original vibration signal obtains single component component, extract fault signature further based on single component component.

In recent years, because entropy can identify nonlinear parameter, the method based on entropy is widely used in fault detection and diagnosis, as approximate entropy, Sample Entropy, fuzzy entropy and multi-scale entropy.But approximate entropy is overly dependent upon data length; Sample Entropy based on unit-step function discontinuous at boundary position, there will be step phenomenon; Fuzzy entropy, based on the concept of membership function, is difficult to determine exactly; The proposition of multi-scale entropy, based on Sample Entropy, only calculates Sample Entropy from multiple yardstick.In order to the complicacy of analytic signal, Bandit and Pompe proposes the concept of arrangement entropy.Due to arrangement, entropy has simply, computing velocity is fast, robustness good, nonlinear transformation is had to the advantage of unchangeability, in a lot of field, obtains application.Subsequently, multiple dimensioned arrangement entropy is born, calculated permutations entropy from multiple different yardstick, but cannot disclose the intrinsic scale feature of signal.And adaptive Time Frequency Analysis method can the local feature of reflected signal, the arrangement entropy based on single component component can provide failure message more accurately.Therefore, the inventive method calculates the fault signature of arrangement entropy as bearing of single component component.

Determine for malfunction, key is the similarity of measuring exactly between test data and sample data.Dynamic time warping (dynamic time warping, DTW) method is set forth in 1978, is the problem in order to solve speech recognition at first.Then, as a kind of mode-matching technique, DTW obtains application at a lot of other field, as fingerprint authentication, Activity recognition, on-line signature checking, data mining, computer vision and computer animation, process monitoring and fault diagnosis etc.Compared with other method for mode matching, DTW is simple, easy, has good real-time capacity.But in DTW algorithm, similarity measurement, based on Euclidean distance square, cannot ensure the separability between small data sample, more cannot reflect the global coherency of data.

Summary of the invention

The technology of the present invention is dealt with problems: overcome the deficiencies in the prior art, a kind of adaptive failure diagnostic method improving dynamic time warping based on arrangement entropy and stream shape is provided, in order to measure the distance between test data and training data quickly and accurately, thus determine current malfunction, realize bearing failure diagnosis.

The technical solution used in the present invention is: a kind of self-adaptation bearing variable working condition method for diagnosing faults based on PE-MDTW, and step is as follows:

Step (1), application self-adapting Time-Frequency Analysis Method decompose original vibration signal, obtain several single component component of signals;

Step (2), for each single component component of signal, extract its stable arrangement entropy as fault eigenvalue, to reduce the impact of working conditions change on eigenwert;

Step (3), based on the fault feature vector extracted, application MDTW measures the similarity between test data and training data, thus determines the malfunction that current data is corresponding, realizes failure modes.

Further, described step (1) is specially: the original vibration signal x (t) of application self-adapting signal processing method to bearing nonlinear and nonstationary processes, and obtains several single component component of signals.Here adaptive signal processing method can select the one in EMD, LMD or LCD.In order to ensure the real-time of failure diagnostic process, split original vibration signal, only analyze 1024 points through a large amount of repetition test, effect is best at every turn.

Further, described step (2) is specially: to each single component component of signal, extracts its stable arrangement entropy as fault eigenvalue, to reduce the impact of working conditions change on eigenwert.Process is as follows:

(1) set original vibration signal as x (t), through Time-Frequency Analysis Method process, x (t) is broken down into m single component component c _i(t) and residual components v (t), that is, x (t)=c ₁(t)+c ₂(t)+... + c _m(t)+v (t);

(2) to each single component component c _it (), calculating its arrangement entropy is PE _i, and vectorial W=[PE ₁, PE ₂..., PE _m] be exactly a fault feature vector of original vibration signal.

Further, described step (3) is specially: on the basis of the arrangement entropy feature vector extracted, the MDTW method of application real-time, similarity measurement better effects if calculates the distance between test data and each sample data, and then judge the malfunction of current data, thus realize the fault diagnosis of bearing.

Process is as follows:

(1) first, to the original vibration signal under various health status, carry out Time-frequency Decomposition and extract arrangement entropy feature vector, as sample characteristics matrix during follow-up health status classification, if the data of total k kind health status, then this sample characteristics matrix V=[W ₁, W ₂..., W _k], wherein W _iit is the proper vector of i-th kind of health status;

(2) then, for the vibration signal of arbitrary state to be determined, by Time-Frequency Analysis Method decomposed signal, and its arrangement entropy feature vector is extracted;

(3) similarity that MDTW algorithm measures each proper vector in the proper vector of state to be determined and sample characteristics matrix is applied, metric is less, prove that the state of current state to be determined and this label characteristics vector is more close, thus determine the health status of current data.

The present invention's advantage is compared with prior art:

(1) changeable for bearing operating mode complicated condition, existing diagnostic method flow process is complicated, the present situation of poor real, propose the effective ways of a kind of bearing adaptive failure diagnosis, improve the real-time of diagnostic procedure, strengthen the separability between different faults state, and then improve the effect of failure modes.

(2), for the feature of bearing vibration signal nonlinear and nonstationary non-gaussian, original vibration signal is decomposed several single component components by the non-linear Time-Frequency Analysis Method of application self-adapting, obtains the time-frequency distributions that original signal is complete.

(3), extract the arrangement entropy of each single component component as fault signature, improve the stability of fault signature and the rapidity of feature calculation, reduce working condition change to the impact of eigenwert and improve the real-time performance of method.

(4) apply MDTW and measure similarity between test data and sample data, improve the real-time of separability between small data and pattern match, improve the effect of failure modes under bearing variable working condition.

Calculate the arrangement entropy of each single component component in described step (2) as fault eigenvalue, obtain more stable proper vector thus to reduce the impact of working conditions change on proper vector.

Accompanying drawing explanation

Fig. 1 is the effect case diagram of MDTW metric range;

Fig. 2 is the overall flow figure of Method for Bearing Fault Diagnosis;

Fig. 3 is the testing table schematic diagram of bearing data center of Washington Catholic University of America;

Fig. 4 is the broken line graph of data set 1 characteristic results IMF-PE (a) under different operating mode, ISC-PE (b) and PF-PE (c);

Fig. 5 is the broken line graph of data set 2 characteristic results IMF-PE (a) under different operating mode, ISC-PE (b) and PF-PE (c);

Fig. 6 is the broken line graph of data set 3 characteristic results IMF-PE (a) under different operating mode, ISC-PE (b) and PF-PE (c);

Fig. 7 is the PF-PE feature clustering effect scatter diagram of data set 1 under different operating mode;

Fig. 8 is the PF-PE feature clustering effect scatter diagram of data set 2 under different operating mode;

Fig. 9 is the PF-PE feature clustering effect scatter diagram of data set 3 under different operating mode;

Figure 10 is the Clustering Effect scatter diagram of IMF-PE (a), ISC-PE (b) and PF-PE (c) under data set 4 inner ring fault different faults degree;

Figure 11 is the Clustering Effect scatter diagram of IMF-PE (a), ISC-PE (b) and PF-PE (c) under data set 4 outer shroud fault different faults degree;

Figure 12 is the Clustering Effect scatter diagram of IMF-PE (a), ISC-PE (b) and PF-PE (c) under data set 4 rolling body fault different faults degree;

Figure 13 is the similarity measurement result based on DTW, SDTW and MDTW method.

Embodiment

The present invention proposes dynamic time warping method (the manifold-based dynamic time warping that stream shape is improved, MDTW), similarity measurement is carried out based on the line segment length on stream shape, improve different classes of between separability, and then improve the accuracy of bearing failure diagnosis.The present invention proposes based on the self-adaptation Method for Bearing Fault Diagnosis of PE-MDTW, demonstrates the validity that the method diagnoses under bearing variable working condition condition, have good practical engineering application and be worth the analysis result of test figure.

The present invention is further illustrated below in conjunction with accompanying drawing and specific embodiment.

One of the present invention is based on arrangement entropy (permutation entropy, PE), flow shape and improve dynamic time warping (manifold-based dynamic time warping, MDTW) adaptive failure diagnostic method, concrete steps are as follows:

1, adaptive signal processing method

EMD, LMD and LCD belong to adaptive signal processing method, and original vibration signal can be decomposed into the single component component that several have local scale feature by these methods, for the extraction of consequent malfunction feature lays the foundation.EMD method is that original signal is decomposed into a series of orthogonal component, i.e. intrinsic mode function (intrinsic mode function, IMF).Original signal is decomposed into several multiplicative functions (product function, PF) by LMD, and each PF component is the product of an envelope signal and a pure FM signal, can extract the instantaneous frequency with the physics meaning from pure FM signal.On EMD algorithm basis, original signal, by the new baseline of definition and iteration stopping criterion, is decomposed into a series of intrinsic scale component (intrinsic scale component, ISC) by LCD.IMF, PF and ISC component is all single component component, can the local scale feature of reflected signal.Because the ultimate principle of these three kinds of methods has introduction in a lot of documents and materials, just repeat no more here.

2, the key concept of entropy is arranged

Arrangement entropy is at first in order to the complicacy analyzing data proposes, and have simply, computing velocity is fast, robustness good, nonlinear transformation is had to the advantage of unchangeability, obtain application in a lot of field now, its ultimate principle is described below simply:

For One-dimension Time Series x (k), k=1,2 ..., N}, its m in the i moment tie up delay embedding Definition of Vector and are:

$X_i^m=\[x\left(i\right),x(i+\mathrm{\τ}),...,x(i+(m-1)\mathrm{\τ})\]---\left(1\right)$

Wherein, i=1,2 ..., N, m >=2 are Embedded dimensions, and τ is time delay.

Then, if S _mfor factorial m! Symmetrical group, be the combination of all arranging situations, one of them arrangement can be expressed as π _j=(j ₁, j ₂..., j _m).So, just can think one is had to arrange π _j, and if only if π _jat S _min the unique and condition met below:

x(i+(j ₁-1)τ)≤x(i+(j ₂-1)τ)≤…≤x(i+(j _m-1)τ) (2)

Wherein, 1≤j _i≤ m, j _i≠ j _j.

To each arrangement π _j, the frequency of its correspondence can be obtained by formula below:

$p\left({\mathrm{\π}}_j\right)=\frac{Number\left\{j\right|j\≤N-(m-1)\mathrm{\τ},X_i^{m}hastype\mathrm{\π}\}}{N-(m-1)\mathrm{\τ}}---\left(3\right)$

Then according to the principle of Shannon entropy, can by formulae discovery arrangement entropy below.

$H_p\left(D\right)=-\underset{{\mathrm{\π}}_j\∈S_m}{\Σ}p\left({\mathrm{\π}}_j\right)ln\left(p\left({\mathrm{\π}}_j\right)\right)---\left(4\right)$

Thus, the arrangement entropy of arbitrary signal can be calculated.

3, dynamic time warping correlation technique

(1) dynamic time warping method

Dynamic time warping (dynamic time warping, DTW) be by Sakoe and Chiba be speech recognition propose method for mode matching, then have also been obtained extensive application at other field.Based on dynamic programming techniques, DTW, by time series is carried out extending and shortening, calculates the bee-line between two time serieses, and then realizes similarity measurement.The principles illustrated of DTW algorithm is as follows:

For two sequence C=c ₁, c ₂..., c _i..., c _mand Q=q ₁, q ₂..., q _j..., q _n, the distance d (C between them between corresponding element _i, Q _j) can be calculated by a distance function, thus obtain the distance matrix of a n × m.In traditional DTW algorithm, distance function is Euclidean distance square.Then, by making Cumulative Distance minimum, a regular path U=(u can be determined ₁, u ₂..., u _k..., u _l), wherein max (m, n)≤L≤m+n-1.Some local restrictive conditions of this paths demand fulfillment, such as:

A () end points limits: the terminal in this path should correspond to first point and last point of distance matrix, ensures that the sequencing of sequence does not change, that is, u ₁=(c ₁, q ₁), u _l=(c _m, q _n).

B () continuity limits: each time, path can only take a step forward, the continuous print that the process of coupling is necessary, across Point matching, that is, can not work as u _k=(c _i, q _j), u _k+1=(c _i+1, q _j+1), then there is c _i+1-c _i≤ 1, q _j+1-q _j≤ 1.

C () monotonicity limits: matching process carries out along sequence dullness, that is, work as u _k=(c _i, q _j), u _k+1=(c _i+1, q _j+1), then there is c _i≤ c _i+1, q _j≤ q _j+1.

Finally, DTW Cumulative Distance is defined as:

$DTW(C_i,Q_j)=d(C_i,Q_j)+min\left\{\begin{array}{c}DTW(C_i,Q_{j-1})\\ DTW(C_{i-1},Q_j)\\ DTW(C_{i-1},Q_{j-1})\end{array}\right.---\left(5\right)$

In actual applications, calculate all possible path and expend time in very much, nor necessary, the overall situation restriction of therefore applying regular path in the matching process reduces the path of calculating.

As seen from the above analysis, have the step that two main in DTW method, one is the distance in calculating two sequences between corresponding element, and one is in distance matrix, find an optimal path make the Cumulative Distance between two sequences the shortest.Due in traditional DTW method, similarity measurement, based on Euclidean distance square, cannot ensure the separability between small distance data, and distance value can be made again when Euclidean distance is larger larger.In order to address these problems, the present invention proposes the DTW method based on stream shape.

(2) standardization dynamic time warping method

Because in traditional DTW algorithm, distance function is Euclidean distance square, it treats the proper vector of all dimensions coequally but in fact these features are unequal.In order to address this problem, calculating distance before can first establishing criteria formula to sequence C=c ₁, c ₂..., c _i..., c _mand Q=q ₁, q ₂..., q _j..., q _ncarry out standardization.Standardization formula is as follows:

$x_i^{*}=\frac{x_i-m}{s}---\left(6\right)$

Wherein, x _i ^*be the point value after standardization, m is the average of sequential element, and s is the standard deviation of sequence.

In similarity measurement, when test data is simultaneously all smaller with several different classes of distance, we wish to strengthen the may differentiate between these distances, thus obtain better classifying quality, so in SDTW algorithm, distance function elects standardized Euclidean distance as, but not Euclidean distance square.

(3) based on the dynamic time warping method of stream shape

No matter be DTW or SDTW algorithm, its similarity measurement is all based on Euclidean distance.Because Euclidean distance is thought " between 2, straight line is the shortest ", but from the data overall situation, the straight line be directly connected between 2 is not necessarily the shortest, and is likely connection distance the shortest between 2 by a series of path coupled together compared with short line segment.Therefore DTW and SDTW algorithm cannot reflect the consistance of data.And the manifold distance defined by Mikhail Belkin can reflect the global coherency of data, manifold distance is exactly the size of the manifold structure metric range along data.In order to describe " between 2, straight line may not be the shortest " this characteristic, in manifold distance, define 2 x _i, x _jbetween line segment length be:

$L(x_i,x_j)={\mathrm{\ρ}}^{dist(x_i,x_j)}-1---\left(7\right)$

Wherein, dist (x _i, x _j) be the Euclidean distance of point-to-point transmission, ρ > 1 is spring factor.

Be subject to the inspiration that formula (7) defines, the present invention is that DTW defines a kind of new distance function:

$l(x_i,x_j)=1-e^{-\mathrm{\β}\·d(x_i,x_j)}---\left(8\right)$

Wherein, β > 0 is spring factor, d (x _i, x _j) be standardized Euclidean distance.

In order to explain this distance function how to improve traditional DTW algorithm, a case more visual pattern.If d (:)=0:0.01:2, β=1.7, carry out compute distance values with formula (8), then draw d-l graph of a relation picture, as shown in Figure 1.As can be seen from the figure, d1 > dx > d2, as d (x _i, x _j) less time, increase a little d (x a little _i, x _j), the l (x of its correspondence _i, x _j) can increase more, thus widen the distinguishable ability between small distance; And as d (x _i, x _j) larger time, same degree ground increases d (x _i, x _j), find the l (x of its correspondence _i, x _j) be only increased a little, prevent the overinflation of distance.In practice, when each distance value is all smaller, be difficult to the ownership determining current data, the distinguishable power between at this moment wishing small distance is strengthened; And when distance value is larger, at this moment can d (x be passed through _i, x _j) be separated different classes of, there is no need to expand d (x again _i, x _j), like this to classifying quality not too large benefit, cause partial distance on the contrary and expand, reduce the resolving power between small distance.

As seen through the above analysis, the distance function based on the concept of manifold that the present invention proposes, can well strengthen the detachability between small distance, avoids distance simultaneously and expands, improve the ability of similarity measurement.Because the improvement DTW method of the present invention's proposition is based on stream shape thought, therefore called after is based on the DTW of stream shape, i.e. MDTW.

4, based on the Method for Bearing Fault Diagnosis of EMD/LCD/LMD-PE-MDTW

The self-adaptation Method for Bearing Fault Diagnosis overall flow that the present invention proposes as shown in Figure 2.Concrete step is as follows:

(1) first, application self-adapting Time-Frequency Analysis Method EMD, LCD or LMD decompose original vibration signal, obtain a limited number of single component component of signal;

(2) then, to each one-component component, its arrangement entropy is extracted as fault signature, thus fast, the failure message of stably reflected signal;

(3) last, based on the fault feature vector extracted, application MDTW measures the similarity between test data and training data, thus determines the malfunction that current data is corresponding, realizes failure modes.

Application example is as follows:

1, bearing Data Source

In order to verify the validity of put forward the methods of the present invention, the method validation result based on bearing data disclosed in bearing data center of Washington Catholic University of America will be shown below.This bearing designation is 6205-2RS JEM SKF.Bearing test-bed comprises the motor of a 2hp, a torque converter, a dynamometer and relevant control circuit, as shown in Figure 3.Test bearing supports motor shaft.Before test, application spark erosion technique is filled with Single Point of Faliure respectively on the inner ring of bearing, outer shroud and rolling body, often kind of fault mode is filled with three kinds of different fault sizes, is respectively 7mils, 14mils and 21mils (1mil=0.001inches); And three kinds of different engine loads (0-3 horsepower, corresponding rotating speed be 1797,1772,1750 and 1720RPM) under gather vibration data, frequency acquisition is 12kHz.

2, the fault signature based on EMD/LCD/LMD-PE extracts

In the present invention, application self-adapting Time-Frequency Analysis Method and the combination of arrangement entropy, extract the fault characteristic information in bearing vibration signal.

First, a kind of method in application EMD, LCD or LMD decomposes original vibration signal, obtains the single component component with local feature information.In the present invention, in order to obtain better real-time fault diagnosis performance, the original signal of download is divided into some parts, every part only comprises 1024 points for signal decomposition, thus decreases the time of feature extraction.

Then, its arrangement entropy is calculated as fault eigenvalue to decomposing the single component component obtained.

In order to the effect of contrast characteristic's extracting method EMD-PE, LCD-PE and LMD-PE, in the present invention, consider that working conditions change and fault degree change contrast the feature that these three kinds of methods are extracted.

(1) performance of fault diagnosis under different operating mode

In engineer applied, because engine speed is indefinite, the working condition of bearing inevitably changes.And different engine speed can produce the vibration signal of varying strength, and then affect the size of fault signature, thus the effect of effective fault diagnosis, be therefore necessary to assess the performance of fault diagnosis under different operating mode.

As previously stated, often organize original vibration signal and comprise 1024 points, Time-frequency Decomposition is carried out to it, calculate the arrangement entropy of component as fault eigenvalue.In order to the validity of the inventive method under different working condition is described, will carry out compliance test result below, the data set of checking composition is as shown in table 1, and wherein data set 1-3 is for verifying the performance of fault diagnosis of the inventive method under different operating mode.

First, contrast the arrangement entropy feature based on EMD, LMD and LCD respectively, to determine best fault signature, the comparing result under different faults degree as shown in Figure 4, Figure 5 and Figure 6.As can be seen from the figure, fault signature based on LMD-PE maintains good consistance under different operating mode, and under different operating mode, there occurs larger difference based on EMD-PE with based on the fault signature of LCD-PE, visible, for these bearing data, the resistivity of fault signature to working conditions change based on LMD-PE is stronger.Under different faults degree based on the three-dimensional scatter diagram of the fault signature of LMD-PE as shown in Figure 7, Figure 8 and Figure 9, as can be seen from the figure, under different operating mode, the good cluster of feature of same fault is together, there is good separability between the feature of different faults, ensure that the accuracy of variable working condition fault diagnosis.

The details table of characteristic data set verified by table 1

(2) performance of fault diagnosis under different faults degree

In order to the validity of the inventive method under different faults degree is described, will carry out compliance test result below, the data set composition of checking is as shown in the data set 4 in table 1.For inner ring fault, outer shroud fault and rolling body fault, based on the characteristic 3 D scatter diagram under the different faults degree of LMD-PE, EMD-PE and LCD-PE as shown in Figure 10, Figure 11 and Figure 12.As can be seen from the figure, for these bearing data, the fault signature based on LMD-PE is better to the distinction of different faults degree, and could not well distinguish different faults degree based on the fault signature of EMD-PE and LCD-PE.

Based on above analysis, find for these bearing data, compared to EMD-PE and LCD-PE, the ability of the anti-operating mode of the fault signature based on LMD-PE disturbance ability and differentiation different faults degree is stronger, therefore, the fault signature extracted based on LMD-PE launches by follow-up malfunction classification.

3, the malfunction based on MDTW is determined

Based on the fault feature vector that LMD-PE extracts, application MDTW measures the distance between test data and sample data collection, and then judges which sample label is current data belong to, and realizes fault diagnosis.In order to verify the advantage of the inventive method, applying DTW and MDTW simultaneously and carrying out similarity measurement.

From analysis result above, the fault signature based on LMD-PE extraction effectively can resist the change of operating mode, and effectively distinguishes fault in various degree.Therefore, nonserviceable in identifying, sample data collection comprises the different faults degree of often kind of fault, but also no longer distinguishes working condition, and with the data instance under 1797RPM, the details of sample data collection are as shown in table 2.Sample data collection comprises 10 kinds of labels, and often kind of label comprises 5 groups of data; Prepare 9 groups of test datas altogether, correspond respectively to label 2-10.The result of application DTW, SDTW and MDTW measured similarity as shown in figure 13, as can be seen from the figure, the distance calculated based on DTW is difficult to distinguish small distance, as subgraph " inner ring fault (14mils) ", the DTW distance of test data is simultaneously all very close with label 2,3 and 6, is difficult to judge which label is test data input on earth.And small distance can be distinguished preferably based on the distance of SDTW and MDTW, realize malfunction more accurately and determine.Compared to SDTW, the distance that MDTW calculates enhances the separability between small distance by a larger margin, effectively control again the excessive expansion of larger distance simultaneously, as subgraph " inner ring fault (7mils) ", MDTW increases corresponding Euclidean distance at (c) place, and do not continue to increase Euclidean distance according to same ratio at (a) and (b) place, and then distance value is controlled in proper scope, both ensure that the separability between small distance, and turn avoid unnecessary distance and expand.

Known by the result above, the method for measuring similarity MDTW that the present invention proposes obtains effect more better than traditional DTW and SDTW method, and the distance based on MDTW is that follow-up cluster analysis provides better foundation.

The details of table 2 sample data collection

In sum, the adaptive failure diagnostic method based on PE and MDTW that the present invention proposes, achieves good effect in the diagnosis of variable working condition fault diagnosis and different faults degree.The inventive method can combine with arbitrary neural net method or support vector machine, carries out fault diagnosis or the health evaluating of many group bearing data.

Non-elaborated part of the present invention belongs to the known technology of those skilled in the art.

Claims (4)

1. improve an adaptive failure diagnostic method for dynamic time warping based on arrangement entropy and stream shape, it is characterized in that performing step is as follows:

Step (1), application self-adapting Time-Frequency Analysis Method decompose original vibration signal, obtain several single component component of signals;

Step (2), calculate the arrangement entropy of each single component component as fault signature, reduce working conditions change to the impact of eigenwert;

Step (3), based on extract fault feature vector, the similarity between test data and training data is measured in application stream shape improvement dynamic time warping (MDTW), thus determine the malfunction that current data is corresponding, realize failure modes, thus realize fault diagnosis.

2. a kind of adaptive failure diagnostic method improving dynamic time warping based on arrangement entropy and stream shape according to claim 1, is characterized in that: the middle adaptive Time Frequency Analysis method of described step (1) is empirical mode decomposition EMD, local mean value decomposes LMD or local feature Scale Decomposition LCD method.

3. a kind of adaptive failure diagnostic method improving dynamic time warping based on arrangement entropy and stream shape according to claim 1, is characterized in that: the arrangement entropy that described step (2) calculates each single component component is as follows as the process of fault signature:

(1) set original vibration signal as x (t), through Time-Frequency Analysis Method process, x (t) is broken down into m single component component c _i(t) and residual components v (t), that is, x (t)=c ₁(t)+c ₂(t)+... + c _m(t)+v (t);

(2) to each single component component c _it (), calculating its arrangement entropy is PE _i, and vectorial W=[PE ₁, PE ₂..., PE _m] be exactly a fault feature vector of original vibration signal.

4. a kind of adaptive failure diagnostic method improving dynamic time warping based on arrangement entropy and stream shape according to claim 1, it is characterized in that: described step (3) is based on the fault feature vector extracted, the similarity between test data and training data is measured in application stream shape improvement dynamic time warping (MDTW), thus determine the malfunction that current data is corresponding, realize failure modes detailed process as follows:

(1) first, to the original vibration signal under various health status, carry out Time-frequency Decomposition and extract arrangement entropy feature vector, as sample characteristics matrix during follow-up health status classification, if the data of total k kind health status, then this sample characteristics matrix V=[W ₁, W ₂..., W _k], wherein W _iit is the proper vector of i-th kind of health status;

(2) then, for the vibration signal of arbitrary state to be determined, by Time-Frequency Analysis Method decomposed signal, and its arrangement entropy feature vector is extracted;

(3) similarity that MDTW algorithm measures each proper vector in the proper vector of state to be determined and sample characteristics matrix is applied, metric is less, prove that the state of current state to be determined and this label characteristics vector is more close, thus determine the health status of current data.