CN104729853A - Rolling bearing performance degradation evaluation device and method - Google Patents

Abstract

The invention provides a rolling bearing performance degradation evaluation device and method. The device comprises an acceleration sensor, a data acquisition module, a feature extraction module, an SVDD evaluation module and a verification module. According to the method, the acceleration sensor is made to be used for acquiring vibration signals of a bearing to be measured and converting the vibration signals to simulation signals; the data acquisition module is made to be used for conducting amplifying, smoothing and other processing on the simulation signals, converting the simulation signals into digital signals and then transmitting the digital signals to a computer; the feature extraction module is used for extracting wavelet packet singular spectrum entropies of the vibration signals to serve as input feature vectors so as to be used for the SVDD evaluation module; The SVDD evaluation module is used for establishing a self-adaption SVDD model and evaluates the performance degradation process of the rolling bearing through the self-adaption SVDD module to obtain a performance degradation index DI; the verification module verifies the correctness of the evaluation result through an Hilbert envelope demodulation method based on EMD. The rolling bearing performance degradation evaluation device and method are applied to performance degradation evaluation of the rolling bearing in the whole life cycle.

Description

A kind of rolling bearing performance degradation assessment device and method

Technical field

A kind of rolling bearing performance degradation assessment device and method, belongs to mechanical product quality reliability assessment and fault diagnosis technology field.

Background technology

Along with the fast development of industrial requirement, plant equipment is constantly improved in complicated, efficient, light-duty etc., also faces harsher working environment simultaneously.Once the critical component of equipment breaks down, just may affect whole production run, cause huge economic loss, even can cause the problems such as casualties.Therefore, plant-maintenance system is just changed to the condition maintenarnce based on state by traditional periodic maintenance or correction maintenance, and as setting up the prerequisite of rational maintenance strategy, equipment performance degradation assessment also starts to receive much concern.

Rolling bearing is as one of key components and parts in rotating machinery, and the quality of its performance state directly affects the operational reliability of whole equipment.In general, rolling bearing in use all can experience from normal to degenerating until the process lost efficacy, and usually will experience a series of different performance degradation state during this.If monitor the degree that bearing performance is degenerated in the process can degenerated at rolling bearing performance, so just can organize targetedly and produce and formulate rational maintenance plan, prevent the generation that unit exception lost efficacy.

At present, in engineering, conventional time domain index monitors the running status of rolling bearing.Stability indicator (as root-mean-square value, root amplitude etc.) in time domain index can increase gradually along with fault progression, but cannot judge the position of initial damage.And susceptibility index (as kurtosis index) is though can identify the position of initial damage, along with the development of fault can present the trend of falling after rising, and do not meet the development trend of bearing fault degree.Therefore, the performance degradation index building novelty is needed to reflect the performance degradation process of rolling bearing all sidedly.

The actual bearing vibration signal measured non-linear, non-stationary signal often, vibration signal can decompose by WAVELET PACKET DECOMPOSITION on different frequency bands, and not only the low frequency part of signal is decomposed, also decompose the HFS of signal, therefore WAVELET PACKET DECOMPOSITION can realize more meticulous the portraying of signal.

Wavelet packet singular spectrum entropy is theoretical based on svd, the matrix of coefficients of vibration signal after wavelet package transforms is decomposed into a series of singular value that can reflect former matrix of coefficients essential characteristic, recycling information entropy statistical property analysis on Uncertainty is carried out to singular value set, therefore wavelet packet singular spectrum entropy can to the complexity of original vibration signal provide one determine measure.

Support Vector data description (Support Vector Data Description, SVDD) be a kind of monodrome sorting technique based on border thought grown up in Statistical Learning Theory and support vector machine basis, only need normal sample to carry out model training, this is that the deficient problem of abnormal data in fault diagnosis provides solution route.In addition, the method has the features such as computing velocity is fast, strong robustness.

Summary of the invention

The object of the invention is, in order to obtain the performance degradation index of rolling bearing, the initial failure moment of Timeliness coverage bearing and inefficacy moment, the generation of prevention major accident, the invention provides a kind of rolling bearing performance degradation assessment device and method.

Realizing technical scheme of the present invention is, the invention provides a kind of rolling bearing performance degradation assessment device, comprises,

Acceleration transducer, for gathering the vibration signal of bearing to be measured and vibration signal being converted into simulating signal;

Data acquisition module, being converted to digital signal for described simulating signal being carried out amplifying, after the process such as filtering, then described digital signal being sent to computing machine, and being stored as data file;

Characteristic extracting module, for extracting the wavelet packet singular spectrum entropy of vibration signal as input feature vector vector, for the use of assessment models in described SVDD evaluation module;

SVDD evaluation module and authentication module, for setting up self-adaptation SVDD model, and being assessed by the performance degradation process of described self-adaptation SVDD model to rolling bearing, obtaining performance degradation index DI; Authentication module is for verifying the correctness of performance degradation assessment result.

The input end of described acceleration transducer connection data acquisition module; Data acquisition module connects SVDD evaluation module and authentication module by characteristic extracting module; Described data acquisition module comprises NI SCXI accelerometer load module, NI SCXI signal condition cabinet and NI multifunctional data acquisition card, wherein accelerometer load module is encapsulated in signal condition cabinet, data collecting card can adopt the NI data collecting card of usb bus or pci bus, to adapt to different computer requirements.

The invention provides a kind of rolling bearing performance degradation assessment method, comprise data acquisition, feature extraction, Performance Evaluation and the checking to assessment result, by rolling bearing performance degradation assessment device, rolling bearing performance degeneration is assessed.Described method comprises:

After bearing vibration signal is carried out WAVELET PACKET DECOMPOSITION, respectively the WAVELET PACKET DECOMPOSITION coefficient of each node of last one deck is reconstructed, then svd is carried out to the wavelet package reconstruction coefficient obtained, and then ask for the wavelet packet singular spectrum entropy of each node of last one deck;

Using the input feature vector vector of wavelet packet singular spectrum entropy as SVDD evaluation module, set up self-adaptation SVDD model by input feature vector vector and obtain performance degradation index DI;

Adopt and based on the Hilbert envelope demodulation method of EMD, performance degradation assessment result is verified.

Feature extraction in the inventive method comprises the following steps:

(1) WAVELET PACKET DECOMPOSITION, according to the vibration signal waveforms of rolling bearing, selects db5 wavelet basis to carry out 4 layers of WAVELET PACKET DECOMPOSITION as wavelet basis function to the vibration signal collected, obtains the WAVELET PACKET DECOMPOSITION coefficient of each node of last one deck;

(2) coefficient of dissociation is reconstructed, the WAVELET PACKET DECOMPOSITION coefficient of each for last one deck node is reconstructed, obtains reconstruction coefficients;

(3) carry out svd to reconstruction coefficients, carry out svd to the wavelet package reconstruction coefficient of each node of last one deck, then each sample standard deviation can obtain 16 singular value r ₁, r ₂..., r ₁₆, and these singular values are normalized, $g_i=r_i/\underset{i=1}{\overset{16}{\mathrm{\Σ}}}{r}_i;$

(4) calculate wavelet packet singular spectrum entropy, by the definition of information entropy, the wavelet packet singular spectrum entropy of vibration signal can be expressed as: S _i=-g _ilog ₂g _i; Then the vibration signal in each moment all comprises 16 wavelet packet singular spectrum entropy vectors.

Performance Evaluation in the inventive method comprises the following steps:

(1) using the input vector of the feature samples under normal condition as SVDD model training, obtain the suprasphere in envelope normal sample eigen space, and obtain the radius R of this suprasphere;

(2) input new feature samples a, calculate the generalized distance R of this feature samples to SVDD suprasphere center d _a;

(3) R is compared _awith the size of R, if R _a-R≤0, then perform step (4), if R _a-R>0, then show that SVDD model training terminates, and performs step (5);

(4) using feature samples a input feature vector vector as SVDD model training together with the feature samples under normal condition, continue to upgrade SVDD model by step (1)-(3);

(5) assess in new feature sample substitution self-adaptation SVDD model, calculate the generalized distance R of new feature sample to self-adaptation SVDD suprasphere center d _b, and try to achieve performance degradation index DI.

In the inventive method, assessment result is verified, comprises the following steps:

(1) EMD decomposition is carried out to the vibration signal in rolling bearing initial failure moment and moment of losing efficacy, obtain limited intrinsic mode functions IMF;

(2) respectively correlation analysis is carried out to each IMF component and original signal;

(3) choose the first two IMF component high with original signal correlativity to carry out superposing and reconstruct, obtain reconstruction signal;

(4) utilized by reconstruction signal Hilbert to convert and obtain its signal envelope;

(5) fast fourier transform is carried out to signal envelope, obtain the envelope spectrum of reconstruction signal, verify the correctness of assessment result according to the relation between spectral line frequency differentiable in envelope spectrum and fault characteristic frequency.

The invention has the beneficial effects as follows, the present invention adopts wavelet packet singular spectrum entropy as the input feature vector vector of SVDD assessment models, can reflect non-linear, the non-stationary characteristic of bearing vibration signal, and to the complexity of vibration signal provide one determine measure.Invention introduces self-adaptation SVDD model to assess, SVDD suprasphere border is constantly updated along with the increase of sample to be tested, greatly can improve the accuracy of assessment models.The present invention adopts and carries out double verification based on the Hilbert envelope demodulation method of EMD to assessment result, ensure that correctness and the validity of assessment result.

The present invention is suitable for the performance degradation assessment in rolling bearing life cycle management.

Accompanying drawing explanation

Fig. 1 is rolling bearing performance degradation assessment device schematic diagram;

Fig. 2 is characteristic extracting module process flow diagram;

Fig. 3 is self-adaptation SVDD model algorithm process flow diagram;

Fig. 4 is authentication module process flow diagram.

Embodiment

Embodiments of the present invention relate to a kind of rolling bearing performance degradation assessment device, as shown in Figure 1.

The present embodiment is a kind of rolling bearing performance degradation assessment device, comprises acceleration transducer, data acquisition module, characteristic extracting module, SVDD evaluation module and authentication module.

Acceleration transducer is for gathering the vibration signal of bearing to be measured and vibration signal being converted into simulating signal; Be converted to digital signal after the process such as data acquisition module is used for described simulating signal to carry out amplifying, filtering, then described digital signal be sent to computing machine, and be stored as data file; Characteristic extracting module for extracting the wavelet packet singular spectrum entropy of vibration signal as input feature vector vector, for the use of assessment models in described SVDD evaluation module; SVDD evaluation module is used for setting up self-adaptation SVDD model, and is assessed by the performance degradation process of described self-adaptation SVDD model to rolling bearing, obtains performance degradation index DI, can determine the initial failure moment of bearing and lost efficacy the moment by this index; Described authentication module is for verifying the correctness of performance degradation assessment result.

The installation site of the acceleration transducer of the present embodiment device must be fixed, to ensure the comparability of signal.NI SCXI accelerometer load module can select SCXI-1531 accelerometer load module, and degree of will speed up meter load module is encapsulated in SCXI signal condition cabinet.The conditionings such as simulating signal mainly carries out amplifying by accelerometer load module, filtering.The signal that degree of will speed up meter load module exports is by the core bus of signal condition cabinet and cable adaptor input data collecting card, data collecting card can adopt the NI data collecting card of pci bus or usb bus, wherein the data collecting card of pci bus directly inserts in the PCI slot of computer motherboard, the data collecting card of usb bus is connected with the USB interface of computing machine, so just can adapt to different computer requirements.Characteristic extracting module, SVDD evaluation module and authentication module complete all in a computer.Computer application carries MATLAB software and LABVIEW software.By the collection of LABVIEW programming realization bearing vibration signal, the vibration signal of collection is stored as data file.

In the present invention, the embodiment of rolling bearing performance degradation assessment method is realized by device.And characteristic extracting module, SVDD evaluation module and the authentication module in device is realized by MATLAB software, embodiment is as follows:

1, feature extraction

Characteristic extracting module process flow diagram as shown in Figure 2, is implemented according to the following steps:

(1) WAVELET PACKET DECOMPOSITION

According to the vibration signal waveforms of rolling bearing, select db5 wavelet basis to carry out 4 layers of WAVELET PACKET DECOMPOSITION as wavelet basis function to the vibration signal collected, obtain the WAVELET PACKET DECOMPOSITION coefficient of each node of last one deck.Wavelet transformation can be thought to utilize a series of wavelet basis function to approach vibration signal, general utilization Mallat fast algorithm in actual computation, this algorithm is decomposed approximation signal by a series of low-pass filter group of structure and Hi-pass filter group:

$a_{j,k}\left(i\right)=\underset{n}{\mathrm{\Σ}}{a}_{j-1,k}\left(n\right)p_{n-2i}^{*}---\left(1\right)$

$b_{j,k}\left(i\right)=\underset{n}{\mathrm{\Σ}}{a}_{j-1,k}\left(n\right)q_{n-2i}^{*}---\left(2\right)$

Wherein a _{0, k}=x (i), i=0,1,2 ..., N-1, N is that signal sampling is counted, and x (i) is discrete time signal, and j is the WAVELET PACKET DECOMPOSITION number of plies, k=0,1,2 ..., the impulse response that 15, p (n), q (n) are conjugate mirror filter P, Q, a _j,k(i), b _j,ki () is respectively low frequency, high-frequency decomposition coefficient.

(2) coefficient of dissociation is reconstructed

Be reconstructed by the WAVELET PACKET DECOMPOSITION coefficient of each for last one deck node, restructing algorithm is as follows:

$a_{j,k}\left(i\right)=\underset{n}{\mathrm{\Σ}}{a}_{j-1,k}\left(n\right)p_{i-2n}+\underset{n}{\mathrm{\Σ}}{b}_{j+1,k}\left(n\right)q_{i-2n}---\left(3\right)$

(3) svd is carried out to reconstruction coefficients

Carry out svd to the wavelet package reconstruction coefficient of each node of last one deck, then each sample standard deviation can obtain 16 singular value r ₁, r ₂..., r ₁₆, and these singular values are normalized

$g_i=r_i/\underset{i=1}{\overset{16}{\mathrm{\Σ}}}{r}_i---\left(4\right)$

(4) wavelet packet singular spectrum entropy is calculated

By the definition of information entropy, the wavelet packet singular spectrum entropy of vibration signal can be expressed as

S _i＝-g _ilog ₂g _i(5)

Then the vibration signal in each moment all comprises 16 wavelet packet singular spectrum entropy vectors.

2, SVDD assessment

SVDD evaluation module mainly sets up self-adaptation SVDD model by input feature vector vector, and the basic thought of SVDD generates a minimal hyper-sphere exactly, makes it comprise all normal characteristics samples as far as possible, variously constructs SVDD model by following:

$\mathrm{mod}L(R,\mathrm{\ξ},d)=R^2+C\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ξ}}_i---\left(6\right)$

$s.t.\left\{\begin{array}{c}{(x_i-d)}^T(x_i-d)\≤R^2+{\mathrm{\ξ}}_i\\ {\mathrm{\ξ}}_i\≥0,i=\mathrm{1,2},...,n\end{array}\right.---\left(7\right)$

Wherein d is the suprasphere centre of sphere of SVDD model, and R is suprasphere radius, and C is penalty factor, ξ is relaxation factor, and L is objective function, introduces Lagrange multiplier and imports in objective function by constraint condition, inner product adopts kernel function to replace, and can obtain following quadratic programming formula:

$\max L\left(\mathrm{\α}\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_{i}K(x_i,x_i)-\underset{i,j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_i{\mathrm{\α}}_{j}K(x_i,x_j)---\left(8\right)$

$s.t.\left\{\begin{array}{c}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_i=1\\ 0\≤{\mathrm{\α}}_i\≤C\end{array}\right.---\left(9\right)$

Wherein α _ifor Lagrange multiplier, meet 0≤α _ithe sample of≤C condition is support vector, thus obtains centre of sphere d and the radius R of SVDD suprasphere.R can be determined by following formula:

$R^2=K(x_s\·x_s)+\underset{i,j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_i{\mathrm{\α}}_{j}K(x_i\·x_j)-2\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_{i}K(x_i\·x_s)---\left(10\right)$

Wherein, x _sfor support vector, α _ifor Lagrange multiplier, x _ifor target sample.

In the process setting up SVDD model, add adaptive process, as shown in Figure 3, the concrete implementation step of algorithm is as follows for self-adaptation SVDD model algorithm process flow diagram:

(1) the wavelet packet singular spectrum entropy vector under the normal condition recorded by off-line, as input feature vector vector, obtains SVDD suprasphere, and obtains the radius R of this suprasphere;

(2) the feature samples a that records of Input Online, calculates the generalized distance R of this feature samples to SVDD suprasphere center d _a, R _acan be determined by following formula:

${R_a}^2=K(x_a\·x_a)+\underset{i,j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_i{\mathrm{\α}}_{j}K(x_i\·x_j)-2\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_{i}K(x_a\·x_i)---\left(11\right)$

(3) R is compared _awith the size of R, if R _a-R≤0, then show feature samples a belong to normal condition under feature samples, perform step (4), if R _a-R>0, then show that self-adaptation SVDD model training terminates, and performs step (5);

(4) using feature samples a input feature vector vector as SVDD model training together with the feature samples under normal condition, continue by step (1) to (3), upgrade SVDD model, upgrading the radius terminating rear self-adaptation SVDD model is R1;

(5) assess in the new feature sample recorded online substitution self-adaptation SVDD model, calculate the generalized distance R of new feature sample to self-adaptation SVDD suprasphere center d _b, R _bdetermined by formula (11) equally, finally can try to achieve a series of performance degradation index DI value, DI value is determined by following formula:

$\mathrm{DI}=\left\{\begin{array}{c}0,R_b-R1\≤0\\ (R_b-R1)/R1,R_b-R1>0\end{array}\right.---\left(12\right)$

Then the decision criteria in rolling bearing initial failure moment and moment of losing efficacy is: if DI value is all greater than 0 after sometime, and be ascendant trend, then this moment can be judged to be the bearing initial failure moment always; If the slope after when DI value curve rises to sometime between a moment and this moment reaches maximal value, then this moment can be judged to be the bearing failure moment.In addition, the conspicuousness turning point of DI value curve in uphill process can think the turning point of different phase in bearing performance degenerative process.

3, assessment result is verified

Rolling bearing initial failure moment and moment of losing efficacy in order to ensure correctness and the validity of assessment result, carry out double verification by authentication module to assessment result after determining.Authentication module process flow diagram as shown in Figure 4, is implemented according to the following steps:

(1) the vibration signal data file being defined as the initial failure moment is transferred, EMD decomposition is carried out to vibration signal, determine the maximum point that signal is all and minimum point, by cubic spline matching maximum point and minimum point respectively, obtain coenvelope line and the lower envelope line of signal, and ask the averaged curve of coenvelope line and lower envelope line, then adopt screening principle by signal decomposition according to this averaged curve and obtain limited intrinsic mode functions IMF;

(2) respectively correlation analysis is carried out to each IMF component and original signal, and obtain related coefficient;

(3) choose the maximum the first two IMF component of related coefficient and carry out superposition reconstruct, obtain reconstruction signal;

(4) utilized by reconstruction signal Hilbert to convert and obtain its signal envelope;

(5) fast fourier transform is carried out to signal envelope, obtain the envelope spectrum of reconstruction signal, and calculate the fault characteristic frequency of each position of rolling bearing to be measured (comprising inner ring, outer ring, rolling body and retainer), the correctness of assessment result is verified according to the relation between spectral line frequency differentiable in envelope spectrum and fault characteristic frequency.

Checking the bearing failure moment then transfer be defined as lost efficacy the moment vibration signal data file and verify to (5) by step (1).

During the checking initial failure moment, if there is the spectral line and doubly spectrum line thereof that can distinguish " carpet " noise in signal envelope spectrum, and the fault characteristic frequency at this spectral line frequency and a certain position of rolling bearing is very close, then can determine to have occurred initial failure at this moment bearing, namely this moment is the initial failure moment.

During the checking inefficacy moment, if there is very outstanding spectral line and doubly spectrum line thereof in signal envelope spectrum, " carpet " noise now in envelope spectrum is very little, and the fault characteristic frequency at this spectral line frequency and a certain position of rolling bearing is very close, then can determine in this moment bearing failure, namely this moment is the bearing failure moment.

Claims (6)

1. a rolling bearing performance degradation assessment device, comprises,

Acceleration transducer, for gathering the vibration signal of bearing to be measured and vibration signal being converted into simulating signal;

Data acquisition module, being converted to digital signal for described simulating signal being carried out amplifying, after the process such as filtering, then described digital signal being sent to computing machine, and being stored as data file;

Characteristic extracting module, for extracting the wavelet packet singular spectrum entropy of vibration signal as input feature vector vector, for the use of assessment models in described SVDD evaluation module;

SVDD evaluation module and authentication module, for setting up self-adaptation SVDD model, and being assessed by the performance degradation process of described self-adaptation SVDD model to rolling bearing, obtaining performance degradation index DI; Authentication module is for verifying the correctness of performance degradation assessment result.

2. a rolling bearing performance degradation assessment device, is characterized in that, the input end of described acceleration transducer connection data acquisition module; Data acquisition module connects SVDD evaluation module and authentication module by characteristic extracting module; Described data acquisition module comprises NI SCXI accelerometer load module, NI SCXI signal condition cabinet and NI multifunctional data acquisition card, wherein accelerometer load module is encapsulated in signal condition cabinet, data collecting card can adopt the NI data collecting card of usb bus or pci bus, to adapt to different computer requirements.

3. a rolling bearing performance degradation assessment method, is characterized in that, described method comprises: data acquisition, feature extraction, Performance Evaluation and the checking to assessment result;

After bearing vibration signal is carried out WAVELET PACKET DECOMPOSITION, respectively the WAVELET PACKET DECOMPOSITION coefficient of each node of last one deck is reconstructed, then svd is carried out to the wavelet package reconstruction coefficient obtained, and then ask for the wavelet packet singular spectrum entropy of each node of last one deck;

Using the input feature vector vector of wavelet packet singular spectrum entropy as SVDD evaluation module, set up self-adaptation SVDD model by input feature vector vector and obtain performance degradation index DI;

Adopt and based on the Hilbert envelope demodulation method of EMD, performance degradation assessment result is verified.

4. a kind of rolling bearing performance degradation assessment method according to claim 3, it is characterized in that, described feature extraction comprises the following steps:

(1) WAVELET PACKET DECOMPOSITION, according to the vibration signal waveforms of rolling bearing, selects db5 wavelet basis to carry out 4 layers of WAVELET PACKET DECOMPOSITION as wavelet basis function to the vibration signal collected, obtains the WAVELET PACKET DECOMPOSITION coefficient of each node of last one deck;

(2) coefficient of dissociation is reconstructed, the WAVELET PACKET DECOMPOSITION coefficient of each for last one deck node is reconstructed;

(3) carry out svd to reconstruction coefficients, carry out svd to the wavelet package reconstruction coefficient of each node of last one deck, then each sample standard deviation can obtain 16 singular value r ₁, r ₂..., r ₁₆, and these singular values are normalized, $g_i=r_i/\underset{i=1}{\overset{16}{\mathrm{\Σ}}}{r}_i;$

(4) calculate wavelet packet singular spectrum entropy, by the definition of information entropy, the wavelet packet singular spectrum entropy of vibration signal can be expressed as: S _i=-g _ilog ₂g _i; Then the vibration signal in each moment all comprises 16 wavelet packet singular spectrum entropy vectors.

5. a kind of rolling bearing performance degradation assessment method according to claim 3, it is characterized in that, described Performance Evaluation comprises following concrete steps:

(1) using the input vector of the feature samples under normal condition as SVDD model training, obtain the suprasphere in envelope normal sample eigen space, and obtain the radius R of this suprasphere;

(2) input new feature samples a, calculate the generalized distance R of this feature samples to SVDD suprasphere center d _a;

(3) R is compared _awith the size of R, if R _a-R≤0, then perform step (4), if R _a-R>0, then show that SVDD model training terminates, and performs step (5);

(4) using feature samples a input feature vector vector as SVDD model training together with the feature samples under normal condition, continue to upgrade SVDD model by step (1)-(3);

(5) assess in new feature sample substitution self-adaptation SVDD model, calculate the generalized distance R of new feature sample to self-adaptation SVDD suprasphere center d _b, and try to achieve performance degradation index DI.

6. a kind of rolling bearing performance degradation assessment method according to claim 3, is characterized in that, describedly verifies assessment result, comprises the following steps:

(1) EMD decomposition is carried out to the vibration signal in rolling bearing initial failure moment and moment of losing efficacy, obtain limited intrinsic mode functions IMF;

(2) respectively correlation analysis is carried out to each IMF component and original signal;

(3) choose the first two IMF component high with original signal correlativity to carry out superposing and reconstruct, obtain reconstruction signal;

(4) utilized by reconstruction signal Hilbert to convert and obtain its signal envelope;

(5) fast fourier transform is carried out to signal envelope, obtain the envelope spectrum of reconstruction signal, verify the correctness of assessment result according to the relation between spectral line frequency differentiable in envelope spectrum and fault characteristic frequency.