CN103018044A - Bearing combined failure diagnosis method based on improved impact dictionary matching pursuit - Google Patents

Abstract

The invention discloses a bearing combined failure diagnosis method based on improved impact dictionary matching pursuit. According to the method, a bearing vibration signal is iterated and decomposed successively to be a linear combination based on an i-item atom of an improved impact model dictionary; an improved model is established according to the type of an analyzed bearing, and can very accurately indicate out an impact signal of a fault bearing in the running process; a position parameter u in the improved model is used as a preferentially changing parameter according to the cycle characteristic of a bearing failure signal, an atom base with relatively low redundancy is established by using a method of gradually changing the parameters, and the search speed of an optimal atom is greatly improved; a analyzed signal is cut in the signal decomposing process, the optimal atom is searched from atoms in the atom base, and a corresponding impact signal component can be obtained by impact component reconfiguration; and then the failure characteristic frequency of a bearing is obtained by time-frequency transform, and thus the failure diagnosis of the bearing is achieved.

Description

A kind of bearing combined failure diagnostic method that impacts the dictionary pattern matching tracing algorithm that improves

Technical field

The present invention relates to a kind of bearing combined failure diagnostic method, particularly a kind of based on improving the bearing combined failure diagnostic method that impacts the dictionary pattern matching tracing algorithm.

Background technology

Bearing is the important composition parts of rotation class machinery, and detection and the fault diagnosis tool of its operating condition had very important significance.Its fault vibration signal is the non-stationary signal of a quasi-representative, than stationary signal, its distribution parameter or the regularity of distribution change in time, and what contact in the engineering reality often also is non-stationary signal, are extremely important so the research of this type of signal is used for engineering.

The failure mode of bearing has multiple, divide from abort situation and mainly to be divided into rolling body damage, outer ring damage, inner ring damage, and the combined failure form of several damages, such as outer ring and rolling body damage, inner ring and rolling body damage, the damage of inner ring, outer ring and rolling body etc.The key of bearing failure diagnosis is how to extract fault signature from fault-signal.But because the running environment of the actual centre bearer of engineering is abominable, its vibration signal is very complicated, contains much noise and labile factor, is a kind of typical non-stationary signal, and when particularly early defect appearred in bearing, the signal fault feature was very faint.And shared large percentage occurs in combined failure in engineering reality, and when the various faults feature was superimposed, various trouble units influence each other, and were interfering with each other, made fault signature complicated, is difficult to diagnosis.Therefore how adopting effective analysis tool and algorithm, the initial failure of bearing and combined failure are analyzed and diagnosed, extract fault signature and realize fault progression status monitoring and diagnosis, is a large difficult point of bearing being carried out fault detection and diagnosis.

In the analysis of bearing fault signal, the citation form of the expression of signal is comprised that time domain expresses and frequency domain presentation.Yet for the non-stationary signal of complexity, the characteristic information that simple time-domain representation or frequency domain representation all can not complete portrayal signal be rich in.Therefore time-frequency is expressed and is arisen at the historic moment, but general Time-Frequency Analysis Method is because its single expression to this sophisticated signal of bearing combined failure of decomposing basis function lacks adaptivity.

For realize to signal more flexibly, more succinct and adaptive expression, on the basis of wavelet analysis, Mallat and Zhang have summed up forefathers' achievement in research, matching pursuit algorithm (Marching pursuit based on the former word bank of time-frequency was proposed in 1993, MP), be the strategy that a kind of rarefaction that progressively is similar to ask signal is expressed.This algorithm is selected one group of primitive function from former word bank be the linear expansion that atom calculates signal, and by finding the solution the rectangular projection of signal on each atom signal is carried out Continuous Approximation.

Yet more for the bearing vibration signal ground unrest composition with combined failure, data volume is large, and fault characteristic signals is complicated.Be used for that dictionary model that the matching pursuit algorithm of bearing failure diagnosis uses substantially is more single can not set up one to one relation with parameter and the running status of analyzed bearing, analytical effect also has the space of further improving.Aspect the atom selection, the redundant degree of the dictionary of failing fundamentally to reduce, analysis speed is slower.

Summary of the invention

In order to solve the above-mentioned technical matters of matching pursuit algorithm in the diagnosis of bearing combined failure, the invention provides a kind of bearing combined failure diagnostic method that impacts the dictionary pattern matching tracing algorithm that improves.

The technical scheme that the present invention solves the problems of the technologies described above comprises to be set up with the dictionary model of bearing designation parameter correlation, sets up dictionary, bearing vibration signal is improved impacted dictionary pattern matching and follow the trail of decomposition, component reconstruct, time-frequency conversion and obtain the steps such as fault signature according to bearing fault characteristics.

Wherein set up with the improvement impact dictionary model method of bearing designation parameter correlation as described below:

Bearing roller linear velocity s:

s=πdf _r

Pulse width p _x:

$p_x=\frac{d_x}{s}$

The pulse x (t) that can obtain thus the defective generation can be expressed as:

$x\left(t\right)=\left\{\begin{array}{cc}1& u<t<u+p_x\\ 0& \end{array}\right.$

The impact that is produced by defective namely improves to be impacted the dictionary model and can be expressed as:

φ′ _imp(p,u,f,dx,d,fr)=conv(x(t),φ _imp(p,u,f))

φ wherein _Imp(p, u, f) for its expression formula of decaying exponential function is:

φ _imp(p,u,f)=e ^-p(t-u)sin2πf(t-u)

Be φ ' _Imp(p, u, f, dx, d, fr) can be expressed as φ _ImpThe convolution of (p, u, f) and x (t), (conv is the convolution algorithm symbol).

The impact that is produced by defective namely improves to be impacted the dictionary model and can be expressed as:

φ′ _imp(p,u,f,dx,d,fr)=con(x(t),φ _imp(p,u,f))

φ wherein _Imp(p, u, f) for its expression formula of decaying exponential function is:

φ _imp(p,u,f)=e ^-p(t-u)sin2πf(t-u)

Be φ ' _Imp(p, u, f, dx, d, fr) can be expressed as φ _ImpThe convolution of (p, u, f) and x (t).

Wherein d is that the bearing path can be determined f according to bearing designation _rFrequently can record by the sensor of special measurement rotating speed for turning.d _x(unit: mm), wherein p is the damping vibration attenuation characteristic of shock response, and u is the initial time that the shock response event occurs, and f is corresponding to the damped natural frequency of system for fault diameter.

Setting up improvement impact dictionary according to bearing fault characteristics comprises the steps:

(1) set up the prediction atom: being the signal that to analyze a length be n, at first utilizing above-mentioned improvement to impact the dictionary model and set up length and be the prediction atom of n, is that the bearing path can be determined f according to bearing designation with d wherein _rFrequently can record by the sensor of special measurement rotating speed for turning.P in the fixing prediction atomic expression, f, d _x, change the u value.(wherein the f number of winning the confidence frequency spectrum is composed the corresponding frequency f value in peak.P is that the attenuation coefficient of atom is relevant with the model of measured bearing, can obtain the result with apparatus measures by tabling look-up, but in the prediction atom, only need obtain large probable value (because match tracing is a kind of method of iterative analysis, therefore do not need very accurately value, as long as last iteration result reaches requirement).d _xBe the diameter of bearing fault, can directly measure exact value, but in the bearing operational process, not directly measure, only need the value between 0-1 of prediction to get final product.）

(2) select the u value: making the initial position u value of atom is 0, take 1 as unit change n time, to be atom begin every movement by initial position is that the fault-signal of n is done inner product one time with length once, and relatively n inner product result chooses so that the x of inner product value maximum position u value.（x＜1%n）

(3) select the f value: the u value band of choosing is updated to improve one by one impacts in the dictionary model, fixedly p value and d _xValue (its value is described in (1)) changes the f value, obtains respectively on the value of x u so that the y of inner product value maximum f value.（y＜1%n）

(4) select the p value: the u value under will determining and corresponding f value are brought into to improve and are impacted in the dictionary model, fixedly d _xValue (its value is described in (1)) is tried to achieve z so that the p value of inner product value value maximum.（z＜1%n）

(5) select d _xValue: at last with the p that decides, f, the substitution of u value improves impacts the dictionary model, tries to achieve a so that the d of inner product value maximum _xValue.（a＜1%n）

(6) set up dictionary: with p, f, u, d _xValue bring into and improve to impact the dictionary model, namely having set up atomic length is n, atomic quantity is that the former word bank of xyza is dictionary.Each atom is wherein carried out normalized.

Impact dictionary D by the above-mentioned steps improvement that can to set up a size be xyza.

Bearing vibration signal is impacted dictionary pattern matching tracking decomposition to be comprised the steps:

(1) utilize the acceleration vibration transducer that the vibration signal in the bearing operational process is measured, obtain vibration acceleration signal as signal X to be analyzed (t), sampling length is decided to be 2 integer power (being convenient to analyze), sets sample frequency according to bearing rotating speed and model;

(2) analyzed signal is divided into the m section, the length of each section is n

(3) set up corresponding improvement according to the parameter of bearing to be analyzed and impact the dictionary model

(4) setting up atomic quantity according to the parameter of bearing to be analyzed and signal characteristic is xyza, and atomic length is that dictionary D is impacted in the improvement of n

(5) original signal X (t) is assigned to residual signals, obtains initial residual error r ₀, i.e. r ₀=X (t)

(6) residual signals r _i(i=0,1,2 ..., I-1, I are iterations) and in improving impact dictionary D, respectively seek an optimum matching atom d _i, namely utilize residual signals r _iDo inner product operation with improving each atom that impacts among the dictionary D, maximum inner product is worth corresponding atom d _iBe the optimum matching atom.Obtain simultaneously projection coefficient c _i:

c _i=max<r _i-1,d _i>

(7) total projection of i iteration is before the calculating:

$p_i=\underset{i=0}{\overset{I-1}{\mathrm{\&Sigma;}}}{c}_i{d}_i$

(8) r _I+1=r _i-p _iResidual signals deducts total projection, obtains new residual signals

r _i+1=r _i-p _i

(9) iteration is carried out (6) ~ (8) step, until satisfy stopping criterion for iteration (0＜A＜1), wherein the value of A can determine result of calculation accuracy, the value of A is less, iterations is more, result of calculation and analyzed signal are more approaching.The value of A is larger, and iterations is fewer, and there are gap in result of calculation and analyzed signal.Therefore the value of A can be adjusted as required within the specific limits.

(10) Its Sparse Decomposition finishes, and obtains each rank matching factor c _iWith each rank matched atoms d _i

Restructing algorithm is the inverse process of decomposition algorithm:

Observe for the signal after processing, need to be reconstructed signal.Restructing algorithm is the inverse process of decomposition algorithm, and computing formula is as follows:

$s=\stackrel{I-1}{\mathrm{\&Sigma;}}{c}_i{d}_i$

Technique effect of the present invention is: the bearing vibration signal successive iteration is resolved into linear combination based on the i item atom of the former word bank of improved model.Improved model is set up according to the model parameter of analyzed bearing, can point-devicely reflect the impact signal of fault bearing in operational process.According to the cyclophysis of bearing fault signal, choose the preferential Varying parameters of location parameter u conduct in the improved model, set up the less former word bank of redundance by the method that progressively changes parameter, greatly improved the search speed of optimum atom.In the process of signal decomposition analyzed signal is blocked, the atom that travels through in the former word bank is sought optimum atom, can obtain corresponding impact signal composition by impacting component reconstruct.Then time-frequency conversion obtains the fault diagnosis of the fault characteristic frequency realization bearing of bearing.

Description of drawings

The invention will be further described below in conjunction with the drawings and specific embodiments.

Fig. 1 is that dictionary pattern matching trace flow figure is impacted in improvement of the present invention.

Fig. 2 is that dictionary Establishing process figure is impacted in improvement of the present invention.

Fig. 3 is bearing test system schematic of the present invention.

Fig. 4 is the original time domain waveform of experimental signal and frequency domain among the present invention.

Fig. 5 improve to impact time domain waveform and frequency spectrum after the dictionary pattern matching tracking process among the present invention.

Embodiment

Fig. 1 is the signal decomposition process flow diagram that the dictionary pattern matching tracing algorithm is impacted in improvement of the present invention.Below in conjunction with process flow diagram the signal decomposition method principle of impacting the dictionary pattern matching tracing algorithm based on improvement is elaborated.

(1) utilize the acceleration vibration transducer that the vibration signal in the bearing operational process is measured, obtain vibration acceleration signal as signal x to be analyzed (t), sampling length is decided to be 2 integer power, sets sample frequency according to bearing rotating speed and model;

(2) analyzed signal is divided into the m section, the length of each section is n

(3) set up the individual features construction of function according to the parameter of bearing to be analyzed and improve impact dictionary model,

At first try to achieve bearing roller linear velocity s:

S=πdf _r

Ask subsequently pulse width p _x:

$p_x=\frac{d_x}{s}$

The pulse x (t) that can obtain thus the defective generation can be expressed as:

$x\left(t\right)=\left\{\begin{array}{cc}1& u<t<u+{\mathrm{<figure-callout\; id="0"\; label="p\; x"\; filenames="HDA00002451332300031.png,HDA00002451332300032.png"\; state="\{\{state\}\}">p</figure-callout>}}_x\\ 0& \end{array}\right.$

The impact that is produced by defective namely improves to be impacted the dictionary model and can be expressed as:

φ′ _imp(p,u,f,dx,d,fr)=conv(x(t),φ _imp(p,u,f))

φ wherein _Imp(p, u, f) for its expression formula of decaying exponential function is:

φ _imp(p,u,f)=e ^-p(t-u)sin2πf(t-u)

Be φ ' _Imp(p, u, f, dx, d, fr) can be expressed as φ _ImpThe convolution of (p, u, f) and x (t) (conv is the convolution algorithm symbol).

Wherein d is that the bearing path can be determined f according to bearing designation _rFrequently can record by the sensor of special measurement rotating speed for turning.d _x(unit: mm), wherein p is the damping vibration attenuation characteristic of shock response, and u is the initial time that the shock response event occurs, and f is corresponding to the damped natural frequency of system for fault diameter.

(4) setting up atomic quantity according to the parameter of bearing to be analyzed and signal characteristic is xyza, and atomic length is that dictionary D is impacted in the improvement of n.Fig. 2 improves the process flow diagram that impacts dictionary D for setting up, and (4-1)-(4-6) makes detailed description in conjunction with process flow diagram to improving the process of setting up of impacting dictionary:

(4-1) set up the prediction atom: being the signal that to analyze a length be n, at first utilizing above-mentioned improvement to impact the dictionary model and set up length and be the prediction atom of n, is that the bearing path can be determined f according to bearing designation with d wherein _rFrequently can record by the sensor of special measurement rotating speed for turning.P in the fixing prediction atomic expression, f, d _x, change the u value.(wherein the f number of winning the confidence frequency spectrum is composed the corresponding frequency f value in peak.P is that the attenuation coefficient of atom is relevant with the model of measured bearing, can obtain the result with apparatus measures by tabling look-up, but in the prediction atom, only need obtain large probable value (because match tracing is a kind of method of iterative analysis, therefore do not need very accurately value, as long as last iteration result reaches requirement).d _xBe the diameter of bearing fault, can directly measure exact value, but in the bearing operational process, not directly measure, only need the value between 0-1 of prediction to get final product.）

(4-2) select the u value: the initial position that makes atom is 0, take 1 as unit cyclic shift n time, to be atom begin every movement by initial position is that the fault-signal of n is done inner product one time with length once, n inner product result relatively, choose so that the x of inner product value maximum position u value (x＜＜n).

(4-3) select the f value: the u value band of choosing is updated to improve one by one impacts in the dictionary model, fixedly p value and d _xValue (its value is described in (4-1)) changes the f value, obtains respectively on the value of x u so that the y of inner product value maximum f value.（y＜1%n）

(4-4) select the p value: the u value under will determining and corresponding f value are brought into to improve and are impacted in the dictionary model, fixedly d _xValue (its value is described in (4-1)) is tried to achieve z so that the p value of inner product value value maximum.（z＜1%n）

(4-5) select d _xValue: at last with the p that decides, f, the substitution of u value improves impacts the dictionary model, tries to achieve a so that the d of inner product value maximum _xValue.（a＜1%n）

(4-6) set up dictionary: with p, f, u, d _xValue bring into and improve to impact the dictionary model, namely having set up atomic length is n, atomic quantity is that the former word bank of xyza is dictionary.Each atom is wherein carried out normalized.Impact dictionary D by the above-mentioned steps improvement that can to set up a size be xyza.

(5) original signal x (t) is assigned to residual signals, obtains initial residual error r ₀, i.e. r ₀=X (t)

(6) residual signals r _i(i=0,1,2 ..., I-1, I are iterations) and in dictionary D, respectively seek an optimum matching atom d _i, namely utilize residual signals r _iDo inner product operation with improving each atom that impacts among the dictionary D, maximum inner product is worth corresponding atom d _iBe the optimum matching atom.Obtain simultaneously projection coefficient c _i:

c _i=max<r _i-1,d _i>

(7) total projection of i iteration is before the calculating:

$p_i=\underset{i=0}{\overset{I-1}{\mathrm{\&Sigma;}}}{c}_i{d}_i$

(8 residual signals deduct total projection, obtain new residual signals r _I+1

r _i+1=r _i-p _i

(9) iteration is carried out (6) ~ (8) step, until satisfy stopping criterion for iteration

(0＜A＜1), wherein the value of A can determine result of calculation accuracy, can adjust within the specific limits as required

(10) Its Sparse Decomposition finishes, and obtains each rank matching factor c _iWith each rank matched atoms d _i

(11) restructing algorithm is the inverse process of decomposition algorithm:

Observe for the signal after processing, need to be reconstructed signal.Restructing algorithm is the inverse process of decomposition algorithm, and computing formula is as follows:

$s=\underset{i=0}{\overset{I-1}{\mathrm{\&Sigma;}}}{c}_i{d}_i$

Signal after the reconstruct is carried out time frequency analysis can extract fault signature.

Fig. 3 is the bearing test system.Experimental system is comprised of bearing experiment table, HG3528A data collecting instrument, notebook computer.Wherein experiment table by threephase asynchronous machine 1. by flexible coupling 2. with rotor rotating shaft 4. be housed be connected, 3. axle is normal bearing by two 6307 bearings, 5. is the bearing of different spot corrosion patterns.Motor speed R=1496r/min(turns frequently: 24.933Hz), the large footpath D=80mm of bearing, path d=35mm, the rolling body number is Z=8, contact angle α=0.Calculating the bearing outer ring fault characteristic frequency according to above-mentioned parameter is 76.7282Hz, and the inner ring failure-frequency is 122.738Hz, and the rolling body fault characteristic frequency is 99.38Hz, and sample frequency is 15360Hz, and analyzed signal length is 8192 points.

The inner ring that Fig. 4 measures by Fig. 3 pilot system, outer ring, rolling body all contain bearing vibration signal x (t) time domain waveform of spot corrosion, the impact that is produced by fault among the figure substantially be submerged can not failure judgement situation

Fig. 5 is time domain waveform and the spectrogram of reconstruct after improving the processing of impact dictionary, can find out that the impact composition that is produced by fault in the time domain waveform is extracted.At the fault characteristic frequency 75Hz of spectrogram centre bearer outer ring, inner ring, rolling body, 122Hz, 98Hz high-visible (have error, and error being in allowed band), wherein also there is the frequency multiplication composition in the fault characteristic frequency of inner ring and rolling body.Thereby realize that fault signature extracts and the diagnosis of bearing combined failure.

Claims (5)

1. one kind based on the bearing combined failure diagnostic method that improve to impact the dictionary pattern matching tracing algorithm, may further comprise the steps:

(1) gathers bearing vibration signal as signal to be analyzed;

(2 foundation are impacted the dictionary model with the improvement of bearing designation parameter correlation;

(3) set up improvement according to bearing fault characteristics and impact dictionary;

(4) bearing vibration signal is improved the impact dictionary pattern matching and follow the trail of decomposition, obtain each rank matched atoms and matching factor based on the former word bank of each subcharacter;

(5) component is impacted in reconstruct;

(6) reconstruction signal is carried out time-frequency conversion and obtain fault characteristic frequency.

2. according to claim 1 improvement impact dictionary model in the described step (2) is set up according to the model parameter of bearing based on improving the bearing combined failure diagnostic method that impacts the dictionary pattern matching tracing algorithm, and method for building up is as follows:

Bearing roller linear velocity s:

s=πdf _r

Pulse width p _x:

$p_x=\frac{d_x}{s}$

The pulse x (t) that obtains thus the defective generation is expressed as:

$x\left(t\right)=\left\{\begin{array}{cc}1& u<t<u+p_x\\ 0& \end{array}\right.$

The impact that is produced by defective namely improves to be impacted the dictionary model representation and is:

φ′ _imp(p,u,f,dx,d,fr)=conv(x(t),φ _imp(p,u,f))

φ wherein _Imp(p, u, f) for its expression formula of decaying exponential function is:

φ _imp(p,u,f)=e ^-p(t-u)sin2πf(t-u)

Be φ ' _Imp(p, u, f, dx, d, fr) is expressed as φ _ImpThe convolution of (p, u, f) and x (t), conv is the convolution algorithm symbol;

Wherein d is that the bearing path is determined f according to bearing designation _rFor the sensor that turns frequently by special measurement rotating speed records; d _xBe fault diameter, unit: mm, wherein p is the damping vibration attenuation characteristic of shock response, and u is the initial time that the shock response event occurs, and f is corresponding to the damped natural frequency of system.

3. according to claim 1 foundation improves the impact dictionary according to bearing fault characteristics in the described step (3) based on improving the bearing combined failure diagnostic method that impacts the dictionary pattern matching tracing algorithm, and the process of foundation comprises the steps:

(1) set up the prediction atom: being the signal that to analyze a length be n, at first utilizing above-mentioned improvement to impact the dictionary model and set up length and be the prediction atom of n, is that the bearing path is determined f according to bearing designation with d wherein _rFor the sensor that turns frequently by special measurement rotating speed records; P in the fixing prediction atomic expression, f, d _x, change the u value; Wherein the f number of winning the confidence frequency spectrum is composed the corresponding frequency f value in peak; P is that the attenuation coefficient of atom is relevant with the model of measured bearing, d _xBe the diameter of bearing fault, in the bearing operational process, directly do not measure, only need the value between 0-1 of prediction to be;

(2) select the u value: making the initial position u value of atom is 0, take 1 as unit change n time, namely to begin every movement by initial position be that the fault-signal of n is done inner product one time with length once to atom, relatively n inner product result, choose so that the x of inner product value maximum position u value x＜1%n;

(3) select the f value: the u value band of choosing is updated to improve one by one impacts in the dictionary model, fixedly p value and d _xValue changes the f value, obtains respectively on the value of x u so that the y of inner product value maximum f value; Y＜1%n;

(4) select the p value: the u value under will determining and corresponding f value are brought into to improve and are impacted in the dictionary model, fixedly d _xValue (its value is described in (1)) is tried to achieve z so that the p value of inner product value value maximum, z＜1%n;

(5) select d _xValue: at last with the p that decides, f, the substitution of u value improves impacts the dictionary model, tries to achieve a so that the d of inner product value maximum _xValue, a＜1%n;

(6) set up dictionary: with p, f, u, d _xThe value substitution improve to impact the dictionary model, namely having set up atomic length is n, atomic quantity is that the former word bank of xyza is dictionary; Each atom is wherein carried out normalized;

Impact dictionary D by the above-mentioned steps improvement that to set up a size be xyza.

4. according to claim 1 based on improving the bearing combined failure diagnostic method that impacts the dictionary pattern matching tracing algorithm, the decomposition algorithm process comprises the steps: in the described step (4)

(1) utilize the acceleration vibration transducer that the vibration signal in the bearing operational process is measured, obtain vibration acceleration signal as signal X to be analyzed (t), sampling length is decided to be 2 integer power, sets sample frequency according to bearing rotating speed and model;

(2) analyzed signal is divided into the m section, the length of each section is n;

(3) set up corresponding improvement according to the parameter of bearing to be analyzed and impact the dictionary model;

(4) setting up atomic quantity according to the parameter of bearing to be analyzed and signal characteristic is xyza, and atomic length is that dictionary D is impacted in the improvement of n;

(5) original signal X (t) is assigned to residual signals, obtains initial residual error r ₀, i.e. r ₀=X (t);

(6) residual signals r _i(i=0,1,2 ..., I-1, I are iterations) and in improving impact dictionary D, respectively seek an optimum matching atom d _i, namely utilize residual signals r _iDo inner product operation with improving each atom that impacts among the dictionary D, maximum inner product is worth corresponding atom d _iBe the optimum matching atom; Obtain simultaneously projection coefficient c _i:

c _i=max<r _i-1,d _i>

(7) total projection of i iteration is before the calculating:

$p_i=\underset{i=0}{\overset{I-1}{\mathrm{\&Sigma;}}}{c}_i{d}_i$

(8) residual signals deducts total projection, obtains new residual signals r _I+1

r _i+1=r _i-p _i

(9) iteration is carried out (6) ~ (8) step, until satisfy stopping criterion for iteration

, 0＜A＜1;

(10) Its Sparse Decomposition finishes, and obtains each rank matching factor c _iWith each rank matched atoms d _i

5. according to claim 1 impacting the component restructing algorithm in the described step (5) is the inverse process of decomposition algorithm based on improving the bearing combined failure diagnostic method that impacts the dictionary pattern matching tracing algorithm, and computing formula is:

$s=\stackrel{I-1}{\mathrm{\&Sigma;}}{c}_i{d}_i$ 。

Images