CN106295072A - A kind of quantitative trend diagnosis method for bearing internal external circle fault - Google Patents

Abstract

The invention discloses a kind of quantitative trend diagnosis method for bearing internal external circle fault, the method is a kind of quantitative trend diagnosis method based on Sparsogram Yu Lempel Ziv.The present invention is directed to, the present Research being difficult to quantitative assessment bearing fault development trend at present with determining abort situation, have studied the failure mechanism of bearing, it is proposed that a kind of quantitative trend diagnosis method based on Sparsogram Yu Lempel Ziv.Application Sparsogram algorithm, to measured signal denoising, extracts characteristic signal.Denoised signal carries out Lempel Ziv complexity indexization process.Found by research, fault degree and Lempel Ziv complexity index value proportion relation, the difference of rolling bearing inside and outside circle abort situation simultaneously, Lempel Ziv complexity is differentiated and is presented the incremental and different trend rules successively decreased, thus failure judgement position and fault size.

Description

A kind of quantitative trend diagnosis method for bearing internal external circle fault

Technical field

The invention belongs to fault diagnosis technology field, relate to a kind of bearing internal external circle fault quantitative trend diagnosis method, special Do not relate to a kind of quantitative trend diagnosis method based on Sparsogram Yu Lempel-Ziv.

Background technology

Along with progress and the industrial expansion of science and technology, process industry plant equipment is day by day to maximization, high speed, systematization And automation direction development, this is more and more higher to the stability requirement of equipment running status.And in order to meet production requirement, close The function of button apparatus structure becomes increasingly complex, and it is changeable to add work under bad environment, can be the most aging in During Process of Long-term Operation, The Potential feasibility broken down increases the most accordingly, and once the critical component of equipment breaks down, it is possible to destroy whole equipment Stable operation, even cause huge economic loss, cause catastrophic casualties.Gear-box is as industry drive system Important component part, its effect is the most crucial.Rolling bearing is as the topmost part in gear-box inside, and internal structure is complicated And often run under high speed, fully loaded transportation condition, rate of breakdown is higher.Therefore rolling bearing is carried out condition monitoring and fault diagnosis To guaranteeing production safety, prevent major accident important in inhibiting.

Bearing is the important component part in heavy mechanical equipment, and its health status is directly connected to the normal of whole equipment Operating, its running status is monitored with diagnose significant.And traditional Fault monitoring and diagnosis method is often Presence or absence for bearing fault is analyzed qualitatively, and the direction gradually complicated to more precise treatment along with equipment is developed, right Bearing fault carry out accurate quantitative analysis could disclose accurately the running status of equipment the law of development of pre-measurement equipment and Residual life, thus know detection and the maintenance of equipment.

Summary of the invention

For the rolling bearing fault signal processing problems of different faults type and different faults degree of injury, method is first Primary signal is carried out WAVELET PACKET DECOMPOSITION, is then compared by sparse value and select optimized analysis frequency band, construct this node signal Calculate its Lempel-Ziv complexity value, draw out Lempel-Ziv complexity value and the fault of dissimilar fault the most respectively The graph of a relation of degree.The flow process of method is as follows:

The first step: same type of rolling bearing primary signal is carried out i layer WAVELET PACKET DECOMPOSITION.

Second step: reconstruct all node coefficient of the wavelet packet of length i-th layer as primary signal, and with each wavelet packet The coefficient of node is real part, is transformed to imaginary part structural analysis signal with the Hilbert of wavelet packet node coefficient.Then, analysis is taken The mould of signal obtains analyzing the envelope signal of signal.Subsequently, calculate the power spectrum of this envelope signal, and calculate i-th layer of wavelet packet The sparse value of node.Select the maximum or second largest node of sparse value as optimized analysis frequency band.

3rd step: optimized analysis frequency band is carried out binary system process, calculates Lempel-Ziv complexity normalized value respectively CnNh and CnNl, more comprehensively obtain Lempel-Ziv aggregative indicator CnN.According to the result of calculation of Lempel-Ziv complexity, paint Make the graph of a relation of Lempel-Ziv complexity value and fault degree of injury.

The present invention is directed to field measurement fault vibration signal, use product complexity theory to enter the oscillating sequence of faulty bearings Before row is analyzed, need to decompose and extract maximally effective part in analyzed sequence and use product complexity theory to calculate, to letter Number decomposition and the direct relation that extracts of signal live part the quality of diagnostic result.Before using Sparsogram algorithm to carry out The decomposition of phase signal and the extraction of signal live part.After rolling bearing breaks down, the vibration aggravation of rolling bearing, fault The vibration performance caused is to there is shock pulse in vibration signal.Special in order to preferably extract the impact in bearing vibration signal Levy, the WAVELET PACKET DECOMPOSITION basic function of Sparsogram is optimized and chooses.Extract the preferable 10 kinds of small echos of impact signal effect Base Daubechies 5, Daubechies 7, Daubechies 10, Daubechies 13, Symlets 6, Symlets 7, Symlets 8, Coiflet 4, Biorthogonal 2.6, Biorthogonal 3.9, Biorthogonal 5.5, based on this 10 kinds of wavelet basiss, choose db7 and sym8 both wavelet basiss as based on Sparsogram and Lempel-by error analysis The WAVELET PACKET DECOMPOSITION basic function of the quantitative trend diagnosis of bearing fault of Ziv.

After signal Sparsogram algorithm process, carry out Lempel-Ziv complexity indexization and process.Lempel-Ziv Basic thought is: the complexity of sequence is the biggest, and the periodic component in sequence is the fewest, and sequence is the most irregular, levels off to random manner, The frequency content that sequence comprises is the abundantest, illustrates that the complexity of system is the biggest；The complexity of sequence is the least, and in sequence, the cycle becomes Dividing the most obvious, more tend to periodic state, the frequency content that sequence comprises is less, illustrates that the complexity of system is the lowest.

Accompanying drawing explanation

Fig. 1 is quantitative trend diagnosis method figure based on Sparsogram Yu Lempel-Ziv.

Fig. 2 is actual measurement bearing inner race malfunction test signal.

Fig. 3 is actual measurement bearing outer ring malfunction test signal.

Fig. 4 is bearing inner race Lempel-Ziv complexity variation tendency.

Fig. 5 is bearing outer ring Lempel-Ziv complexity variation tendency.

Detailed description of the invention

Below in conjunction with specification drawings and specific embodiments, the invention will be further described

Fig. 1 is the Sparsogram flow chart with lempel-ziv algorithm of the present invention.Below in conjunction with flow chart to fault Quantitative Diagnosis Method And Principle is described in detail.

(1) same type of rolling bearing primary signal is carried out i layer WAVELET PACKET DECOMPOSITION.

(2) reconstruct all node coefficient of the wavelet packet of length i-th layer as primary signal, and with each wavelet packet node Coefficient be real part, be transformed to imaginary part structural analysis signal with the Hilbert of wavelet packet node coefficient.Then, analysis signal is taken Mould obtain analyze signal envelope signal.Subsequently, calculate the power spectrum of this envelope signal, and calculate i-th layer of wavelet packet node Sparse value.Select the maximum or second largest node of sparse value as optimized analysis frequency band.

Sparsogram is a kind of optimum analysis band selection methods based on wavelet packet.By sparse value in WAVELET PACKET DECOMPOSITION The bottom all nodes in find optimal Decomposition node as optimum analysis frequency band to be further analyzed.

The computing formula of sparse value S:

$S(i,j)=\frac{\left|\right|P(i,j)|{|}_2}{\left|\right|P(i,j)|{|}_1}$

In formula: P (i, j) energy spectrum of i-th layer of jth node analysis signal；(i, j) ‖ 1 is P (i, L1 norm j) to ‖ P；‖ (i, j) ‖ 2 is P (i, L2 norm j) to P；(i j) is the sparse value of i-th layer of jth node to S.

Node corresponding for the value that S is maximum or second largest is carried out signal reconstruction, the optimized analysis frequency band obtained.

(3) optimized analysis frequency band is carried out binary system process, calculate Lempel-Ziv complexity normalized value CnNh respectively And CnNl, more comprehensively obtain Lempel-Ziv aggregative indicator CnN.According to the result of calculation of Lempel-Ziv complexity, draw out Lempel-Ziv complexity value and the graph of a relation of fault degree of injury.

For signal S (i) (i=1,2 ..., N), first we be translated into binary sequence, makes a=mean (S (i)), if S (i) >=a, then s (i)=1, otherwise s (i)=0.Thus signal S (i) is converted for binary sequence SN= S1, s2, s3 ..., sN}, the Lempel-Ziv complexity value of sequence SN can be obtained by the n times cycle calculations of following CN (r) (r≤N) Arrive:

Initialize Sv, 0={}, Q0={}, CN (0)=0.R=1, makes Qr={Qr-1Sr}, owing to Qr is not belonging to Sv, r- 1, then CN (r)=CN (r-1)+1, Qr={}, r=r+1；

Make Qr={Qr-1Sr}, it is judged that whether Qr belongs to Sv, r-1={Sv, r-2sr-1}, the most then CN (r)=CN (r- 1), r=r+1, repeats step (2)；

If it is not, then CN (r)=CN (r-1)+1, Qr={}, r=r+1, repetition step (2).

It can be seen that along with the value of the increase complexity of the length N value of SN can increase or constant, such complexity can be subject to To the impact of length N of SN, in order to make Lempel-Ziv complexity index relatively independent, Lempel and Ziv proposes normalization Formula:

$0\≤C_{nN}=\frac{C_{N\left(N\right)}}{\underset{N\→\mathrm{\∞}}{lim}{C}_{N\left(N\right)}}=\frac{C_{N\left(N\right)}}{\underset{N\→\mathrm{\∞}}{lim}\frac{N}{(1-\mathrm{\β}){\log }_{k}N}}\≈\frac{C_{N\left(N\right){\log }_{k}N}}{N}$

In formula: k is the number (for binary sequence SN, k=2) of element in SN；CnN is that Lempel-Ziv normalization refers to Mark.The condition that formula is set up is that N is sufficiently large, and the experience value of N is: N >=3600.

Fig. 2 is actual measurement bearing inner race malfunction test signal, for the correctness of verification algorithm；

Fig. 3 is actual measurement bearing outer ring malfunction test signal；

Fig. 4 is bearing inner race Lempel-Ziv complexity variation tendency, for inner ring fault along with the increase of fault size Lempel-Ziv complexity index value diminishes；

Fig. 5 is bearing outer ring Lempel-Ziv complexity variation tendency, for outer ring fault along with the increase of fault size Lempel-Ziv complexity index value becomes big.

Claims (3)

1. the quantitative trend diagnosis method for bearing internal external circle fault, it is characterised in that: comprise the following steps:

Step (1) carries out i layer WAVELET PACKET DECOMPOSITION to same type of rolling bearing primary signal；

Step (2) reconstruct all node coefficient of the wavelet packet of length i-th layer as primary signal, and with each wavelet packet node Coefficient be real part, be transformed to imaginary part structural analysis signal with the Hilbert of wavelet packet node coefficient；Then, analysis signal is taken Mould obtain analyze signal envelope signal；Subsequently, calculate the power spectrum of this envelope signal, calculate i-th layer of wavelet packet node Sparse value；Select the maximum or second largest node of sparse value as optimized analysis frequency band；

(3) optimized analysis frequency band is carried out binary system process, respectively calculate Lempel-Ziv complexity normalized value CnNh and CnNl, more comprehensively obtain Lempel-Ziv aggregative indicator CnN；According to the result of calculation of Lempel-Ziv complexity, draw out Lempel-Ziv complexity value and the graph of a relation of fault degree of injury.

2. to go a kind of quantitative trend diagnosis method for bearing internal external circle fault described in 1 according to right, it is characterised in that: Described step (1) carries out i layer WAVELET PACKET DECOMPOSITION to same type of rolling bearing primary signal；

Choose db7 and sym8 both wavelet basiss as the quantitative trend of bearing fault based on Sparsogram Yu Lempel-Ziv The WAVELET PACKET DECOMPOSITION basic function of diagnosis.

3. to go a kind of quantitative trend diagnosis method for bearing internal external circle fault described in 1 according to right, it is characterised in that: Step (1) carries out i layer WAVELET PACKET DECOMPOSITION to same type of rolling bearing primary signal；

Step (2) reconstruct all node coefficient of the wavelet packet of length i-th layer as primary signal, and with each wavelet packet node Coefficient be real part, be transformed to imaginary part structural analysis signal with the Hilbert of wavelet packet node coefficient；Then, analysis signal is taken Mould obtain analyze signal envelope signal；Subsequently, calculate the power spectrum of this envelope signal, and calculate i-th layer of wavelet packet node Sparse value；Select the maximum or second largest node of sparse value as optimized analysis frequency band；

Sparsogram is a kind of optimum analysis band selection methods based on wavelet packet；By sparse value in WAVELET PACKET DECOMPOSITION All nodes of bottom find optimal Decomposition node as optimum analysis frequency band to be further analyzed；

The computing formula of sparse value S:

$S\left(i,j\right)=\frac{\left|\right|P\left(i,j\right)|{|}_2}{\left|\right|P\left(i,j\right)|{|}_1}$

In formula: P (i, j) energy spectrum of i-th layer of jth node analysis signal；(i, j) ‖ 1 is P (i, L1 norm j) to ‖ P；‖P(i, J) ‖ 2 is P (i, L2 norm j)；(i j) is the sparse value of i-th layer of jth node to S；

Node corresponding for the value that S is maximum or second largest is carried out signal reconstruction, the optimized analysis frequency band obtained；

(3) optimized analysis frequency band is carried out binary system process, respectively calculate Lempel-Ziv complexity normalized value CnNh and CnNl, more comprehensively obtain Lempel-Ziv aggregative indicator CnN；According to the result of calculation of Lempel-Ziv complexity, draw out Lempel-Ziv complexity value and the graph of a relation of fault degree of injury；

For signal S (i) (i=1,2 ..., N), first we be translated into binary sequence, makes a=mean (S (i)), if S (i) >=a, then s (i)=1, otherwise s (i)=0；Thus signal S (i) is converted for binary sequence SN={s1, s2, S3 ..., sN}, the Lempel-Ziv complexity value of sequence SN can be obtained by the n times cycle calculations of following CN (r) (r≤N):

Initialize Sv, 0={}, Q0={}, CN (0)=0；R=1, makes Qr={Qr-1Sr}, owing to Qr is not belonging to Sv, r-1, then CN (r)=CN (r-1)+1, Qr={}, r=r+1；

Make Qr={Qr-1Sr}, it is judged that whether Qr belongs to Sv, r-1={Sv, r-2sr-1}, the most then CN (r)=CN (r-1), r =r+1, repeats step (2)；

If it is not, then CN (r)=CN (r-1)+1, Qr={}, r=r+1, repetition step (2)；

Seeing that the value of the increase complexity of the length N value along with SN can increase or constant, such complexity can be by the length of SN The impact of degree N, in order to make Lempel-Ziv complexity index relatively independent, Lempel and Ziv proposes normalization formula:

$0\≤C_{nN}=\frac{C_{N\left(N\right)}}{\underset{N\→\mathrm{\∞}}{\lim }{C}_{N\left(N\right)}}=\frac{C_{N\left(N\right)}}{\underset{N\→\mathrm{\∞}}{\lim }\frac{N}{\left(1-\mathrm{\β}\right){\log }_{k}N}}\≈\frac{C_{N\left(N\right){\log }_{k}N}}{N}$

In formula: k is the number (for binary sequence SN, k=2) of element in SN；CnN is Lempel-Ziv normalization index；N Experience value be: N >=3600.