CN104236905A - Bearing fault diagnosis method - Google Patents

Abstract

The invention discloses a bearing fault diagnosis method and belongs to the field of mechanical faults. The method comprises the steps of denoising an acquired vibration signal with the flexible morphological filtering method to increase the signal to noise ratio, conducting LMD on the denoised vibration signal to obtain PF components, conducting spectral analysis on each PF component to obtain a spectrogram, and extracting fault character frequency from the obtained power spectrogram. The method has the advantages that the vibration signal is denoised with the flexible morphological filtering method so that noise in the vibration signal can be eliminated, fault frequency subjected to frequency spectrum correction is closer to actual fault character frequency than uncorrected fault frequency, and the accuracy of bearing fault diagnosis is improved.

Description

A kind of Method for Bearing Fault Diagnosis

Technical field

The invention belongs to mechanical fault field, relate to the Method for Bearing Fault Diagnosis that a kind of flexible shape filtering method and LMD combine.

Background technology

Rolling bearing is most widely used a kind of mechanical part in various rotating machinery, its running status directly affects the performance of equipment, when its local existing defects, pulse shock can be produced in operation process, the pulse signal frequency that different positions produces is different, if can effectively extract shock pulse signal, the position of existing defects just can be diagnosed.Conventional fault signature extracting method mainly contains fast Fourier and becomes (Fast Fourier transform, FFT), wavelet transformation and empirical mode decomposition (Empirical mode decomposition, EMD) etc.2005, Jonathan.Smith proposes a kind of new adaptive Time Frequency Analysis method, be referred to as local average and decompose ((Local mean decomposition, be called for short LMD), and be successfully applied in the time frequency analysis of brain electricity (EEG) signal, the method can effectively process non-linear, non-stationary signal, achieves good effect in recent years in fault diagnosis.

FFT cannot take into account the overall picture of signal in time domain and frequency domain and Localization Problems simultaneously.The segmentation of wavelet transformation to time frequency plane is mechanical lattice, and wavelet basis is different, and decomposition result is different, the more difficult selection of wavelet basis; Signal decomposition can be multiple IMF (Intrinsic mode function) component by EMD method, the time-frequency distributions that Hilbert conversion can obtain signal is done to all IMF components, but in theory also there are some problems, as the mode in EMD method is obscured, owes envelope, crossed the problem such as envelope, end effect, be all among research.LMD method is a kind of effective Time-Frequency Analysis Method, but be often mingled with a large amount of noises in actual signal, these noises also participate in LMD and decompose, and cause primary fault characteristic information and noise aliasing and not easily extract, moreover, noise component makes LMD Decomposition order increase, produce pseudo-component, algorithm also may be caused not restrain, increase the weight of boundary effect, LMD can be made time serious to decompose and to lose actual physical significance, thus the Accurate Diagnosis of impact to fault.

Summary of the invention

Object of the present invention:

To the method adopting flexible shape filtering and LMD to combine, solve noise problem.

Technical scheme of the present invention:

First adopt flexible shape filtering method to carry out noise reduction to the vibration signal collected, improve signal to noise ratio (S/N ratio), reduce the interference of noise.Flexible shape filtering both can extract the edge contour of signal and the shape facility of signal effectively, had robustness again simultaneously; Secondly adopt LMD method to decompose the vibration signal after noise reduction, each PF component obtained contains original signal Partial Feature information, and its complexity is relatively more simply too much than original signal.So just the analysis to the various characteristic information of original signal, be converted into and divide quantitative analysis to each PF, easilier original signal is familiar with and holds.Contained by PF component, the composition of signal is comparatively simple, wherein comprised fault characteristic information just not easily flood by other information, that from PF component, therefore extracts that failure message just becomes is relatively simple.Then power spectrumanalysis is carried out to PF component, extract fault signature.

Concrete steps are as follows:

A kind of Method for Bearing Fault Diagnosis, is characterized in that, comprise the steps:

The first, adopt flexible shape filtering method to carry out noise reduction to the vibration signal collected, improve signal to noise ratio (S/N ratio);

The second, LMD decomposition is carried out to the vibration signal after noise reduction, obtains PF component;

3rd, spectrum analysis is carried out to each PF component, obtains spectrogram;

4th, from obtaining power spectrum chart, extract fault characteristic frequency.

After described third step completes, adopt correcting algorithm to correct frequency spectrum, obtain the spectrogram after correcting.

Flexible shape filtering method is adopted to carry out noise reduction to the vibration signal collected in described first step.

Carry out LMD decomposition to the vibration signal after noise reduction in described first step, the concrete grammar obtaining PF component is: for any given signal x (i), and its decomposable process simplified summary is as follows: find out all Local Extremum n of signal x (i) _i, obtain the absolute value that all adjacent Local Extremum mean value and all adjacent Local Extremum are subtracted each other, and respectively divided by 2, obtain m _iand a _i:

$m_i=\frac{n_i+n_{i+1}}{2}---\left(1\right)$

$a_i=\frac{|n_i-n_{i+1}|}{2}---\left(2\right)$

Then by all adjacent m _icouple together with straight line, then use the smoothing process of moving average method, obtain local mean value function m ₁₁(t).Use the same method and obtain envelope estimation function a ₁₁(t).

By local mean value function m ₁₁t () separates from original signal x (t), obtain:

h ₁₁(t)＝x(t)-m ₁₁(t) (3)

Use h again ₁₁t () is divided by envelope estimation function a ₁₁t () is with to h ₁₁t () carries out demodulation, obtain:

s ₁₁(t)＝h ₁₁(t)/a ₁₁(t) (4)

Ideally, s ₁₁t () is a pure FM signal, i.e. its envelope estimation function a ₁₂t () meets a ₁₂t ()=1, if a ₁₂t () ≠ 1, then by s ₁₁t () repeats above-mentioned iterative process as raw data, until s _1nt () is a pure FM signal, namely its envelope estimates letter a _{1 (n+1)}t () meets a _{1 (n+1)}(t)=1.In practical application, under the prerequisite not affecting discomposing effect, a variation Δ can be set, when meeting 1-Δ≤a _1nduring≤1+ Δ, iteration ends.

Finally all envelope estimation functions produced in iterative process are multiplied and obtain envelope signal:

$a_1\left(t\right)=a_{11}\left(t\right)a_{12}\left(t\right)...a_{1n}\left(t\right)=\underset{q=1}{\overset{n}{\mathrm{\Π}}}{a}_{1q}\left(t\right)---\left(5\right)$

By envelope signal a ₁(t) and pure FM signal s _1nt () is multiplied and obtains original signal first PF component:

PF ₁(t)＝a ₁(t)s _1n (6)

It comprises highest frequency component in Setting signal, PF ₁t () is a simple component AM/FM amplitude modulation/frequency modulation signal, its instantaneous amplitude is exactly envelope signal a ₁(t), its instantaneous frequency f ₁t () then can by pure FM signal s _{1n (t)}obtain.

By PF ₁t () separates from Setting signal x (t), obtain a new signal u ₁t (), by u ₁t () repeats above step as raw data, circulation k time, until u _kt () is a monotonic quantity till.

Given like this original signal x (t) is broken down into k PF component and u _k(t) sum, namely

$x\left(t\right)=\underset{p=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{PF}}_p\left(t\right)+u_k\left(t\right)---\left(7\right)$

In formula: u _kt () is discrepance; PF _pt () is envelope signal and pure FM signal product.This illustrates that LMD decomposes rear original signal information and keeps good, does not cause information dropout.

Beneficial effect of the present invention:

Flexible shape filtering method is adopted to carry out noise reduction to vibration signal, eliminate the noise contribution in vibration signal, failure-frequency after Spectrum Correction, than uncorrected failure-frequency, closer to real fault characteristic frequency, improves the accuracy of bearing failure diagnosis.

Accompanying drawing explanation

Fig. 1 is Noise signal graph

Fig. 2 is signal graph after noise reduction

Signal LMD exploded view after Fig. 3 noise reduction

Fig. 4 Noise signal LMD exploded view

Fig. 5 non-noise reduction outer ring fault data signal graph

Fault data time domain beamformer in outer ring after Fig. 6 noise reduction

Outer ring fault data LMD exploded view after Fig. 7 noise reduction

Outer ring fault data PF component spectrogram after Fig. 8 noise reduction

Embodiment

Utilize the method that flexible shape filtering and LMD combine, can the noise contribution of effective erasure signal, reduce noise to the interference of LMD method, fault characteristic frequency can be extracted more accurately.More effective method is provided for successfully carrying out diagnosis to kinematic train rolling bearing fault.

First intend adopting the FMAM Nonlinear Simulation signal containing random noise to carry out flexible shape filtering and LMD decomposition analysis, its form is as follows:

$\left\{\begin{array}{c}{x}_1\left(t\right)=(1+0.4\cos (2\mathrm{\π}\×6t))\sin (2\mathrm{\π}\×100t+\sin (2\mathrm{\π}\×7.5t))\\ x_2\left(t\right)=\sin (2\mathrm{\π}\×20t+\cos (15\mathrm{\π}\×t))\\ x_3\left(t\right)=0.8\sin (15\mathrm{\πt}+\frac{\mathrm{\π}}{5})\\ x=x_1\left(t\right)+x{\left(t\right)}_2+x_3\left(t\right)+0.3\mathrm{randn}\end{array}\right.---\left(1\right)$

T=[0,0.4] in formula, this signal is made up of FMAM signal, FM signal and sinusoidal signal, is mixed with random white noise simultaneously.Sample frequency is 1000Hz, and time domain waveform as shown in Figure 1, can be found out, has occurred many burrs in signal waveform.Adopt flexible shape filtering to carry out noise reduction to signal, obtain the waveform after noise reduction as shown in Figure 2, and do not add compared with the original signal of making an uproar, have localized distortion, but distortion is very little, less on the Essential Analysis impact of signal.Fig. 3 is that after noise reduction, signal LMD decomposes, and wherein comprises 3 PF components and 1 residual components, can more accurately effective constituent be decomposed out.Fig. 4 carries out LMD decomposition to Noise signal, due to the interference of noise, decomposes and has had more 1 high-frequency components PF1, and PF2 and PF3 component partial waveform serious distortion, substantially do not see the rule of signal.As can be seen from Fig. 3 and Fig. 4 contrast, signal carries out LMD again and decomposes its feature and be obviously better than directly decomposing after flexible shape filtering noise reduction, decomposes accuracy and is also greatly improved.

For verifying the effect of this method, select CWRU of U.S. the department of Electrical Engineering and Computer Science bearing test data experiment Analysis herein, 6205-2RS SKF deep groove ball bearing of supporting motor transmission axle head is test bearing, by power be the motor of 1.47kW, electric apparatus control apparatus and dynamometer, moment of torsion code translator/sensor form testing table.Its structural parameters are as table 1.

Table 1 bearing structure parameter table

Parameter	Numerical value
		Steel ball size d/mm	8
Pitch diameter of ball set D/mm	40
		Steel ball quantity N	9
Contact angle α/(°)	0

Implementation step is as follows:

[1] outer ring fault data is first chosen, sample frequency is 12KHz, and sampling number is 1024, engine speed 1797rpm/min, calculating rotating shaft fundamental frequency according to formula (1) is=29.95Hz, calculates outer ring failure-frequency be about=107.82Hz according to formula (2).Time-domain diagram is Fig. 5.

f _r＝n/60 (2)

$f_i=\frac{N}{2}(1+\frac{d}{D}\cos \left(\mathrm{\α}\right))f_r---\left(3\right)$

[2] carrying out to fault data the time-domain diagram that flexible shape filtering obtains data is Fig. 6, contrasts known flexible shape filtering and can well remove noise, and remain the mutagenic components of signal with Fig. 5.

[3] LMD decomposition is carried out to the vibration signal after noise reduction, obtain PF component, as Fig. 7.Each PF component is carried out power spectrumanalysis, and the power spectrum chart obtained is Fig. 8.Can obviously find out from PF1 component, PF2 component and PF3 component frequency spectrum: the spectrogram of PF1 component has obvious peak value at 105.5Hz, be close with bearing outer ring failure theory, all obvious peak value is had at frequency values 210.9Hz, 316.4Hz, 425.6Hz and 530.1Hz place, these frequencies are close with 2,3,4 and 5 frequencys multiplication of bearing outer ring failure theory frequency respectively, also there is obvious peak value at 58.9Hz place, be doubly close with rotating shaft fundamental frequency 2; The power spectrum chart of PF2 component also there is obvious peak value at 105.5Hz place.And there is obvious peak value at 58.9Hz place; On the frequency spectrum of PF3 component, the peak value at 105.5Hz place is also more obvious.Also there is peak value at 58.59Hz place simultaneously.Can draw thus, be that the outer ring fault of characteristic frequency occurs in rolling bearing with 105.5Hz.Therefore, flexible shape filtering and the LMD method of combining can accurately decompose out the trouble unit comprised in signal, describe the validity of the method.

Claims (4)

1. a Method for Bearing Fault Diagnosis, is characterized in that, comprises the steps:

The first, adopt flexible shape filtering method to carry out noise reduction to the vibration signal collected, improve signal to noise ratio (S/N ratio);

The second, LMD decomposition is carried out to the vibration signal after noise reduction, obtains PF component;

3rd, spectrum analysis is carried out to each PF component, obtains spectrogram;

4th, from obtaining power spectrum chart, extract fault characteristic frequency.

2. Method for Bearing Fault Diagnosis according to claim 1, is characterized in that, after described third step completes, adopts correcting algorithm to correct frequency spectrum, obtains the spectrogram after correcting.

3. Method for Bearing Fault Diagnosis according to claim 1, is characterized in that, adopts flexible shape filtering method to carry out noise reduction to the vibration signal collected in described first step.

4. Method for Bearing Fault Diagnosis according to claim 1, it is characterized in that, in described first step, LMD decomposition is carried out to the vibration signal after noise reduction, the concrete grammar obtaining PF component is: for any given signal x (i), and its decomposable process is as follows: find out all Local Extremum n of signal x (i) _i, obtain the absolute value that all adjacent Local Extremum mean value and all adjacent Local Extremum are subtracted each other, and respectively divided by 2, obtain m _iand a _i:

$m_i=\frac{n_i+n_{i+1}}{2}---\left(1\right)$

$a_i=\frac{|n_i-n_{i+1}|}{2}---\left(2\right)$

Then by all adjacent m _icouple together with straight line, then use the smoothing process of moving average method, obtain local mean value function m ₁₁(t).Use the same method and obtain envelope estimation function a ₁₁(t).

By local mean value function m ₁₁t () separates from original signal x (t), obtain:

h ₁₁(t)＝x(t)-m ₁₁(t) (3)

Use h again ₁₁t () is divided by envelope estimation function a ₁₁t () is with to h ₁₁t () carries out demodulation, obtain:

s ₁₁(t)＝h ₁₁(t)/a ₁₁(t) (4)

Ideally, s ₁₁t () is a pure FM signal, i.e. its envelope estimation function a ₁₂t () meets a ₁₂t ()=1, if a ₁₂t () ≠ 1, then by s ₁₁t () repeats above-mentioned iterative process as raw data, until s _1nt () is a pure FM signal, namely its envelope estimates letter a _{1 (n+1)}t () meets a _{1 (n+1)}(t)=1.In practical application, under the prerequisite not affecting discomposing effect, set a variation Δ, when meeting 1-Δ≤a _1nduring≤1+ Δ, iteration ends.

Finally all envelope estimation functions produced in iterative process are multiplied and obtain envelope signal:

$a_1\left(t\right)=a_{11}\left(t\right)a_{12}\left(t\right)...a_{1n}\left(t\right)=\underset{q=1}{\overset{n}{\mathrm{\Π}}}{a}_{1q}\left(t\right)---\left(5\right)$

By envelope signal a ₁(t) and pure FM signal s _1nt () is multiplied and obtains original signal first PF component:

PF ₁(t)＝a ₁(t)s _1n (6)

It comprises highest frequency component in Setting signal, PF ₁t () is a simple component AM/FM amplitude modulation/frequency modulation signal, its instantaneous amplitude is exactly envelope signal a ₁(t), its instantaneous frequency f ₁t () then can by pure FM signal s _{1n (t)}obtain.

By PF ₁t () separates from Setting signal x (t), obtain a new signal u ₁t (), by u ₁t () repeats above step as raw data, circulation k time, until u _kt () is a monotonic quantity till.

Given like this original signal x (t) is broken down into k PF component and u _k(t) sum, namely

$x\left(t\right)=\underset{p=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{PF}}_p\left(t\right)+u_k\left(t\right)---\left(7\right)$

In formula: u _kt () is discrepance; PF _pt () is envelope signal and pure FM signal product.