CN102778357A - Mechanical failure feature extracting method based on optimal parameter ensemble empirical mode decomposition (EEMD) - Google Patents

Abstract

The invention provides a mechanical failure feature extracting method based on optimal parameter ensemble empirical mode decomposition (EEMD), relating to the technical field of mechanical equipment state monitoring and failure diagnosis. According to the mechanical failure feature extracting method, based on the rule that signal extreme point distribution uniformity is changed by a white noise amplitude coefficient, a white noise amplitude coefficient enabling the best signal extreme point uniformity is used as a k value of EEMD, and an M value (total average number) is obtained through calculation by setting an expected decomposition error, thus the EEMD parameter is selected, the EEMD is carried out on a failure signal, and a failure feature is extracted. The feasibility and the effectiveness of the mechanical failure feature extracting method are verified by the successful extraction of the failure feature of a rolling bearing. The mechanical failure feature extracting method is suitable for the fields of mechanical equipment state monitoring, failure diagnosis and the like.

Description

Mechanical fault feature extracting method based on optimized parameter set empirical mode decomposition

Technical field

The present invention relates to plant equipment status monitoring and fault diagnosis technology field, be specifically related to a kind of mechanical equipment fault feature extracting method.

Background technology

Set empirical mode decomposition (Ensemble empirical mode decomposition; EEMD) the noise assistant analysis is applied to empirical mode decomposition (Empirical Mode decompos it ion; EMD) in; To promote anti-mixing to decompose, effectively suppress pattern intrinsic among the EMD and obscure problem.With respect to EMD, (Intrinsic mode function IMF) more can disclose the physical connotation of original signal, makes the physical essence of each IMF more clear to decompose the natural mode function obtain through EEMD.Lei is applied to EEMD in the fault diagnosis of rotary machine rotor system, has accurately diagnosed the fault of rubbing of bumping of generator amature, and has proved EEMD superiority in the rotating machinery fault feature extraction; Zvokelj combines EEMD and principal component analysis, has improved the reliability of rolling bearing fault diagnosis; An uses EEMD and the loosening fault signature of wind power generating set bearing seat has effectively been extracted in Hilbert transform; Zhou uses EEMD and handles the signal that the gear case embedded type sensor obtains, and has successfully realized gear case state on_line monitoring and fault diagnosis; Lei Yaguo has proposed based on the mechanical failure diagnostic method that improves the Hilbert-Huang conversion; At first utilize EEMD to obtain the IMF that non-mode is obscured; Select the IMF responsive through the susceptibility assessment algorithm again, make the Hilbert-Huang spectrum that obtains to diagnose mechanical fault more accurately fault signature.

More than research has shown that EEMD has important value and clear superiority in the mechanical fault feature extraction, yet, when using EEMD, two important parameters must be set, promptly add the amplitude coefficient k and the population mean number of times M of white noise.If it is improper that these two parameters are provided with, resolution error is increased, cause decomposition result meaningless.When k is too small, possibly be not enough to cause the variation of signal Local Extremum, make to add noise and lost meaning with the local time's span that changes signal, when k is excessive, resolution error is increased, even can fall into oblivion the original signal characteristic and make to decompose and lose meaning.Theoretically, more greatly then resolution error is more little for the M value, and until ignoring, but the increase of M will be lost counting yield, make consuming time being multiplied.The problem of choosing to k and M; Huang suggestion k multiply by a mark by the standard deviation of original signal and defines, and generally get the original signal standard deviation 0.2 times, signal medium-high frequency composition k for a long time suitably reduces; Otherwise then suitably increase k, M can confirm through resolution error is set; Recently definite k of the amplitude standard deviation of the radio-frequency component in the old slightly application signal and the amplitude standard deviation of low-frequency component; Obtain M through the expectation resolution error of setting again; This method has certain using value, but the radio-frequency component and the low-frequency component of the signal that is not easily distinguishable in the computation process.

Summary of the invention

In view of this; The present invention is directed to two important parameter k and the problem of choosing of M among the EEMD; Analyze the white noise amplitude coefficient and changed the rule that the original signal extreme point distributes; And, confirm k through the distributing homogeneity index of extreme point behind the calculating adding white noise to the influence that EEMD decomposes precision, efficient and resolution error, the expectation resolution error that EEMD is set calculates M.Thereby, the EEMD method of optimized parameter has been proposed, so that EEMD is used for the mechanical movement status monitoring or carries out fault diagnosis.

The mechanical fault feature extracting method based on optimized parameter set empirical mode decomposition that provides of the present invention comprises the steps:

1) input signal x (t); The product std_max of the standard deviation of the maximum value sequence consecutive point amplitude difference of signal and spacing behind the white noise of the different amplitude coefficients of calculating adding; And the product std_min of the standard deviation of minimal value sequence consecutive point amplitude difference and spacing; With this index as the large and small value point distributing homogeneity of evaluation pickup electrode; The large and small value of the more little then pickup electrode of std_max and std_min distributes even more, obtains to make std_max to obtain the white noise amplitude coefficient k_max of minimum value, obtains to make std_min obtain the white noise amplitude coefficient k_min of minimum value;

2) ask the parameter k of the mean value of k_max and k_min,, accomplish the EEMD parameter optimization again according to the expectation resolution error e calculated population average time M that is provided with as EEMD;

3) fault-signal is carried out EEMD and decompose, obtain a series of natural mode function IMF components, accomplish fault signature and extract.

Further, said step 1) specifically comprises the steps:

11) obtain the standard deviation of original signal x (t), add the white noise n' of N different amplitude coefficients respectively _i(t), the white noise amplitude coefficient is shown below:

$k\left(i\right)=\frac{i}{100}\mathrm{\σ}(i=1~N),$

Wherein, k (i) expression white noise amplitude coefficient, i representes the i time and adds white noise;

12) the last extreme point sequence extr_max (x1 of signal behind the each adding of the calculating white noise; Y1), and the following extreme point sequence extr_min of signal behind the each adding of the calculating white noise (x2, y2); Wherein, (x1, y1), (x2, y2) are horizontal stroke, the ordinate of corresponding extreme point;

13) calculate extr_max (x1, y1) the long-pending std_max of the amplitude difference standard deviation of consecutive point and separation criteria difference, calculate extr_min (x2, y2) the amplitude difference standard deviation of consecutive point and separation criteria difference amass std_min, as shown in the formula:

$\mathrm{std}\_\max =\sqrt{\frac{1}{N1-1}\underset{i=N1}{\overset{N1-1}{\mathrm{\&Sigma;}}}\left(\right|{x1}_{i+1}-{x1}_i|-u_1)}\&times;\sqrt{\frac{1}{N1-1}\underset{i=N1}{\overset{N1-1}{\mathrm{\&Sigma;}}}\left(\right|{y1}_{i+1}-{y1}_i|-{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="HDA00002016766300011.png"\; state="\{\{state\}\}">u</figure-callout>}}_1^{\&prime;})}$

$\mathrm{std}\_\min =\sqrt{\frac{1}{N2-1}\underset{i=N2}{\overset{N2-1}{\mathrm{\&Sigma;}}}\left(\right|{x2}_{i+1}-{x2}_i|-u_2)}\&times;\sqrt{\frac{1}{N2-1}\underset{i=N2}{\overset{N2-1}{\mathrm{\&Sigma;}}}\left(\right|{y2}_{i+1}-{y2}_i\left|{-\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="HDA00002016766300011.png"\; state="\{\{state\}\}">u</figure-callout>}}_2^{\&prime;}\right)}$

In the formula, N1 be extr_max (x1, counting y1), N2 be extr_min (x 2, y2) count u ₁, u ₁' be extr_max (x1, the y1) mean value of the mean value of consecutive point amplitude difference and spacing, u ₂, u ' ₂Be extr_min (x2, the y2) mean value of the mean value of consecutive point amplitude difference and spacing, x1 _I+1Expression maximum point horizontal ordinate, x1 _iExpression expression maximum point horizontal ordinate, y1 _I+1Expression maximum point ordinate, y1 _iExpression maximum point ordinate, x2 _I+1Expression minimum point horizontal ordinate, x2 _iExpression minimum point horizontal ordinate, y2 _I+1Expression minimum point ordinate, y2 _iExpression minimum point ordinate.

Further, said step 2) specifically comprise the steps:

14) calculating makes std_max and std_min reach the white noise amplitude coefficient k_max of minimum value and the mean value k of k_min, again according to computes M:

$M={\left(\frac{k}{e}\right)}^2,$

And whether the size of judging M less than 20, if M=20 is got in M＜20 o'clock.

Further, said step 3) specifically comprises the steps:

15) in original vibration signal x (t), adding M average in such a way respectively is 0, and amplitude coefficient is the white Gaussian noise n of k _i(t):

x _i?(t)＝x(t)＋n _i?(t)，

In the formula, i=1～M, i represent the i time and add white noise, x _i(t) for adding white Gaussian noise signal afterwards;

16) to x _i(t) gather empirical mode decomposition respectively, obtain K natural mode function component and a remainder r _i(t):

$x_i\left(t\right)=\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{c}_{\mathrm{ij}}\left(t\right)+r_i\left(t\right)$

In the formula, c _Ij(t) after expression adds white Gaussian noise the i time, decompose resulting j natural mode function component, j=1～K;

17) with above-mentioned steps 16) corresponding natural mode function component carries out the population mean computing, obtains gathering the natural mode function component c after the empirical mode decomposition _j(t) and remainder r (t):

$c_j\left(t\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{c}_{\mathrm{ij}}\left(t\right),$

$r\left(t\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i\left(t\right),$

In the formula, c _j(t) expression is gathered resulting j natural mode function component after the empirical mode decomposition to original signal;

18) natural mode function component and a remainder of K the ascending orderly arrangement of characteristic time scale of acquisition:

$x\left(t\right)=\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{c}_j\left(t\right)+r\left(t\right),$

Wherein, x (t) representes original signal, c _j(t) j IMF component of expression, r (t) representes remainder.

Further, the fault signature in the said step 3) is that the amplitude spectrum of the IMF component after decomposing through EEMD is composed with Hilbert and accomplished the fault signature extraction.

Further, said error e generally gets 0.01.

Further, said white noise n' _i(t) amplitude coefficient N generally gets 50.

The present invention changes signal extreme point distributing homogeneity rule and to the EEMD resolution error and decompose the rule of precision from the white noise of different amplitude coefficients; Searching makes the k value of the most uniform white noise amplitude coefficient of signal extreme point distribution as EEMD; Come calculated population average time M through the expectation resolution error that is provided with, thereby accomplishing the EEMD Parameter Optimization.Through the parameter after the optimization fault-signal is carried out EEMD and decompose, obtain a series of IMF component, observe the IMF component that characterizes fault signature, or ask for corresponding amplitude spectrum and Hilbert spectrum extraction fault signature.When the present invention carries out the mechanical fault feature extraction for using EEMD, parameter choose a kind of new method that provides, this method is from the characteristics of signal own, the adaptive EEMD parameter that is fit to signal to be analyzed of choosing.

Other advantages of the present invention, target; To in instructions subsequently, set forth to a certain extent with characteristic; And to a certain extent,, perhaps can from practice of the present invention, obtain instruction based on being conspicuous to those skilled in the art to investigating of hereinafter.Target of the present invention and other advantages can be passed through instructions, claims, and the structure that is particularly pointed out in the accompanying drawing realizes and obtains.

Description of drawings

Fig. 1 is an algorithm flow chart of the present invention;

Fig. 2 is an original signal;

Fig. 3 is the Fourier spectrum of original signal;

Fig. 4 is first rank IMF component-IMF1 after EEMD decomposes;

Fig. 5 is the amplitude spectrum of IMF1.

Embodiment

Below will combine accompanying drawing, the preferred embodiments of the present invention will be carried out detailed description; Should be appreciated that preferred embodiment has been merely explanation the present invention, rather than in order to limit protection scope of the present invention.

Below will carry out detailed description to the preferred embodiments of the present invention.

Referring to Fig. 1, Fig. 1 is algorithm flow chart of the present invention, and is as shown in the figure, and the mechanical fault feature extracting method based on optimized parameter set empirical mode decomposition provided by the invention may further comprise the steps:

1) obtains the standard deviation of original signal x (t), add the white noise n' of N (N generally gets 50, and N can suitably increase if needed) different amplitude coefficients respectively _i(t), the white noise amplitude coefficient is shown below:

wherein; K (i) expression white noise amplitude coefficient, i representes the i time and adds white noise;

2) calculate each upper and lower extreme point sequence extr_max that adds signal behind the white noise (x1, y1) and extr_min (x2, y2), x1, y1 (x2, y2) they are horizontal stroke, the ordinate of corresponding extreme point;

3) calculate extr_max (x1, y1) with extr_min (x2, y2) the amplitude difference standard deviation of consecutive point and separation criteria difference amass std_max and std_min, as shown in the formula:

$\mathrm{std}\_\max =\sqrt{\frac{1}{N1-1}\underset{i=N1}{\overset{N1-1}{\mathrm{\&Sigma;}}}\left(\right|{x1}_{i+1}-{x1}_i|-u_1)}\&times;\sqrt{\frac{1}{N1-1}\underset{i=N1}{\overset{N1-1}{\mathrm{\&Sigma;}}}\left(\right|{y1}_{i+1}-{y1}_i|-{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="HDA00002016766300011.png"\; state="\{\{state\}\}">u</figure-callout>}}_1^{\&prime;})},$

$\mathrm{std}\_\min =\sqrt{\frac{1}{N2-1}\underset{i=N2}{\overset{N2-1}{\mathrm{\&Sigma;}}}\left(\right|{x2}_{i+1}-{x2}_i|-u_2)}\&times;\sqrt{\frac{1}{N2-1}\underset{i=N2}{\overset{N2-1}{\mathrm{\&Sigma;}}}\left(\right|{y2}_{i+1}-{y2}_i|-{\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="HDA00002016766300011.png"\; state="\{\{state\}\}">u</figure-callout>}}_2^{\&prime;})};$

In the formula, N1, N2 be extr_max (x1, y1) and extr_min (x 2, y2) count u ₁, u ₁', u ₂, u ' ₂(x 1, y1) and extr_min (x2, y2) mean value of the mean value of consecutive point amplitude difference and spacing to be respectively extr_max; X1 _I+1Expression maximum point horizontal ordinate, x1 _iExpression expression maximum point horizontal ordinate, y1 _I+1Expression maximum point ordinate, y1 _iExpression maximum point ordinate, x2 _I+1Expression minimum point horizontal ordinate, x2 _iExpression minimum point horizontal ordinate, y2 _I+1Expression minimum point ordinate, y2 _iExpression minimum point ordinate.

4) calculate and to make std_max and std_min reach the white noise amplitude coefficient k_max of minimum value and the mean value k of k_min, be provided with and expect resolution error e=0.01,,, get M=20 like M＜20 o'clock again according to computes M;

$M={\left(\frac{k}{e}\right)}^2;$

5) in original vibration signal x (t), adding M average respectively is 0, and amplitude coefficient is the white Gaussian noise n of k _i(t), promptly

x _i(t)＝x(t)＋n _i?(t)；

In the formula, i=1～M, x _i(t) for adding white Gaussian noise signal afterwards;

6) to x _i(t) gather empirical mode decomposition respectively, obtain K natural mode function component and a remainder r _i(t):

$x_i\left(t\right)=\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{c}_{\mathrm{ij}}\left(t\right)+r_i\left(t\right);$

In the formula, c _Ij(t) after expression adds white Gaussian noise the i time, decompose resulting j natural mode function component, j=1～K; x _i(t) signal behind the i time adding of expression white noise, r _i(t) expression adds white noise the i time after EMD decomposes the remainder that obtains.

7) above step is corresponding natural mode function component is carried out the population mean computing, obtains gathering the natural mode function component c after the empirical mode decomposition _j(t) and remainder r (t):

$c_j\left(t\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{c}_{\mathrm{ij}}\left(t\right),$

$r\left(t\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i\left(t\right);$

In the formula, c _j(t) expression is gathered resulting j natural mode function component after the empirical mode decomposition to original signal;

8) natural mode function component and a remainder of K the ascending orderly arrangement of characteristic time scale of acquisition:

$x\left(t\right)=\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{c}_j\left(t\right)+r\left(t\right).$

9) the IMF component being carried out Hilbert is transformed to:

$H\left[c_i\left(t\right)\right]={\hat{c}}_i\left(t\right)=\frac{1}{\mathrm{\&pi;}}{\&Integral;}_{-\&infin;}^{\&infin;}\frac{c_i\left(t\right)}{t-\mathrm{\&tau;}}\mathrm{d\&tau;}(i=1~n)$

Wherein, H [c _i(t)] expression Hilber t transformation results, Expression Hilbert transformation results, c _i(t) i IMF component of expression, the t express time, τ representes integration amount,

So obtain the amplitude function:

$a_i=\sqrt{c_i^2\left(t\right)+{\hat{c}}_i^2\left(t\right)}$

And phase function:

Instantaneous frequency is:

Instantaneous frequency that obtains and instantaneous amplitude combination just can be obtained the Hilbert spectrum:

$H(\mathrm{\&omega;},t)=\mathrm{RP}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_i\left(t\right)e^{j\&Integral;f\left(t\right)\mathrm{dt}}$

In the formula, RP representes to get real part.

Through the integration of time being got final product the Hilbert marginal spectrum h (ω) of picked up signal:

$h\left(\mathrm{\&omega;}\right)={\&Integral;}_0^{T}H(\mathrm{\&omega;},t)\mathrm{dt}.$

Below illustrate present embodiment:

The first step: receive fault-signal to be analyzed.Original signal x in the present embodiment (t) is the vibration signal that has the rolling bearing of outer ring fault, and the signal sampling frequency is 12000Hz, and sampling length is 1024 points, and rotating speed is 800 rev/mins.Bearing designation is UN205, and the rolling body diameter is 7.5mm, and the rolling body number is 12, and pitch diameter is 38.7mm, and contact angle is 0 degree.Promptly changeing frequently is 29.4Hz, and the bearing outer ring failure-frequency is 64.5Hz.The waveform of original signal such as Fig. 2, its amplitude spectrum is as shown in Figure 3, observes Fig. 3 and fails to find fault signature.

Second step: obtain the standard deviation of original signal x (t), add the white noise n' of 50 different amplitude coefficients respectively _i(t), the white noise amplitude coefficient is shown below:

$k\left(i\right)=\frac{i}{100}\mathrm{\&sigma;}(i=1~N);$

The 3rd step: calculate each upper and lower extreme point sequence extr_max that adds signal behind the white noise (x1, y1) and extr_min (x2, y2), x1, y1 (x2, y2) they are horizontal stroke, the ordinate of corresponding extreme point;

The 4th step: calculate extr_max (x1, y1) with extr_min (x2, y2) the amplitude difference standard deviation of consecutive point and separation criteria difference amasss std_max and std_min;

The 5th step: calculating makes std_max and std_min reach the white noise amplitude coefficient k_max of minimum value and the mean value k=0.09 σ of k_min, and expectation resolution error e=0.01 is set, and calculates population mean number of times M=41;

The 6th step: signal x (t) is carried out EEMD decompose the parameter k=0.09 σ of EEMD, M=41.First IMF component that obtains is as shown in Figure 4.From Fig. 4, can obviously observe periodic shock, its impulse period T=15.5ms, i.e. frequency of impact f=1/T=64.51Hz, the bearing outer ring failure-frequency that is calculated with preamble matches.Calculate the amplitude spectrum of IMF1 simultaneously, as shown in Figure 5, main frequency be 64.52 with multiple frequences such as 128.91Hz, 199.22Hz, 257.81Hz, coincide with the outer ring fault characteristic frequency that calculates, extracted the housing washer fault characteristic information accurately.

Present embodiment is based on the EEMD mechanical fault feature extracting method of optimized parameter; Through the parameter of itself choosing EEMD right according to signal to be analyzed; Carry out EEMD then and decompose the IMF component that obtains characterizing fault signature; Ask for its amplitude spectrum, extracted the housing washer fault characteristic information accurately.Feasibility of the present invention and validity have been proved absolutely.

The above is merely the preferred embodiments of the present invention, is not limited to the present invention, and obviously, those skilled in the art can carry out various changes and modification and not break away from the spirit and scope of the present invention the present invention.Like this, belong within the scope of claim of the present invention and equivalent technologies thereof if of the present invention these are revised with modification, then the present invention also is intended to comprise these changes and modification interior.

Claims (7)

1. based on the mechanical fault feature extracting method of optimized parameter set empirical mode decomposition, it is characterized in that: comprise the steps:

1) input signal x (t); The product std_max of the standard deviation of the maximum value sequence consecutive point amplitude difference of signal and spacing behind the white noise of the different amplitude coefficients of calculating adding; And the product std_min of the standard deviation of minimal value sequence consecutive point amplitude difference and spacing; Acquisition makes std_max obtain the white noise amplitude coefficient k_max of minimum value, obtains to make std_min obtain the white noise amplitude coefficient k_min of minimum value;

2) ask the parameter k of the mean value of k_max and k_min,, accomplish the EEMD parameter optimization again according to the expectation resolution error e calculated population average time M that is provided with as EEMD;

3) fault-signal is carried out EEMD and decompose, obtain a series of natural mode function IMF components, accomplish fault signature and extract.

2. the mechanical fault feature extracting method based on optimized parameter set empirical mode decomposition according to claim 1, it is characterized in that: said step 1) specifically comprises the steps:

11) obtain the standard deviation of original signal x (t), add the white noise n' of N different amplitude coefficients respectively _i(t), the white noise amplitude coefficient is shown below:

$k\left(i\right)=\frac{i}{100}\mathrm{\&sigma;}(i=1~N),$

Wherein, k (i) expression white noise amplitude coefficient, i representes the i time and adds white noise;

12) the last extreme point sequence extr_max (x1 of signal behind the each adding of the calculating white noise; Y1), and the following extreme point sequence extr_min of signal behind the each adding of the calculating white noise (x2, y2); Wherein, (x1, y1), (x2, y2) are horizontal stroke, the ordinate of corresponding extreme point;

13) calculate extr_max (x1, y1) the long-pending std_max of the amplitude difference standard deviation of consecutive point and separation criteria difference, calculate extr_min (x2, y2) the amplitude difference standard deviation of consecutive point and separation criteria difference amass std_min, as shown in the formula:

$\mathrm{std}\_\max =\sqrt{\frac{1}{N1-1}\underset{i=N1}{\overset{N1-1}{\mathrm{\&Sigma;}}}\left(\right|{x1}_{i+1}-{x1}_i|-u_1)}\&times;\sqrt{\frac{1}{N1-1}\underset{i=N1}{\overset{N1-1}{\mathrm{\&Sigma;}}}\left(\right|{y1}_{i+1}-{y1}_i|-u_1^{\&prime;})}$

$\mathrm{std}\_\min =\sqrt{\frac{1}{N2-1}\underset{i=N2}{\overset{N2-1}{\mathrm{\&Sigma;}}}\left(\right|{x2}_{i+1}-{x2}_i|-u_2)}\&times;\sqrt{\frac{1}{N2-1}\underset{i=N2}{\overset{N2-1}{\mathrm{\&Sigma;}}}\left(\right|{y2}_{i+1}-{y2}_i\left|{-u}_2^{\&prime;}\right)}$

In the formula, N1 be extr_max (x1, counting y1), N2 be extr_min (x 2, y2) count u ₁, u ₁' be extr_max (x1, the y1) mean value of the mean value of consecutive point amplitude difference and spacing, u ₂, u ' ₂(x 2, the y2) mean value of the mean value of consecutive point amplitude difference and spacing, x1 for extr_min _I+1Expression maximum point horizontal ordinate, x1 _iExpression expression maximum point horizontal ordinate, y1 _I+1Expression maximum point ordinate, y1 _iExpression maximum point ordinate, x2 _I+1Expression minimum point horizontal ordinate, x2 _iExpression minimum point horizontal ordinate, y2 _I+1Expression minimum point ordinate, y2 _iExpression minimum point ordinate.

3. the mechanical fault feature extracting method based on optimized parameter set empirical mode decomposition according to claim 1 is characterized in that: said step 2) specifically comprise the steps:

14) calculating makes std_max and std_min reach the white noise amplitude coefficient k_max of minimum value and the mean value k of k_min, again according to computes M:

$M={\left(\frac{k}{e}\right)}^2,$

And whether the size of judging M less than 20, if M=20 is got in M＜20 o'clock.

4. the mechanical fault feature extracting method based on optimized parameter set empirical mode decomposition according to claim 1, it is characterized in that: said step 3) specifically comprises the steps:

15) in original vibration signal x (t), adding M average in such a way respectively is 0, and amplitude coefficient is the white Gaussian noise n of k _i(t):

x _i(t)＝x(t)＋n _i?(t)，

In the formula, i=1～M, i represent the i time and add white noise, x _i(t) for adding white Gaussian noise signal afterwards;

16) to x _i(t) gather empirical mode decomposition respectively, obtain K natural mode function component and a remainder r _i(t):

$x_i\left(t\right)=\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{c}_{\mathrm{ij}}\left(t\right)+r_i\left(t\right)$

In the formula, c _Ij(t) after expression adds white Gaussian noise the i time, decompose resulting j natural mode function component, j=1～K;

17) with above-mentioned steps 16) corresponding natural mode function component carries out the population mean computing, obtains gathering the natural mode function component c after the empirical mode decomposition _j(t) and remainder r (t):

$c_j\left(t\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{c}_{\mathrm{ij}}\left(t\right),$

$r\left(t\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i\left(t\right),$

In the formula, c _j(t) expression is gathered resulting j natural mode function component after the empirical mode decomposition to original signal;

18) natural mode function component and a remainder of K the ascending orderly arrangement of characteristic time scale of acquisition:

$x\left(t\right)=\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{c}_j\left(t\right)+r\left(t\right),$

Wherein, x (t) representes original signal, c _j(t) j IMF component of expression, r (t) representes remainder.

5. the mechanical fault feature extracting method based on optimized parameter set empirical mode decomposition according to claim 4 is characterized in that: the fault signature in the said step 3) is to compose through the amplitude spectrum of the IMF component after the EEMD decomposition and Hilbert to accomplish the fault signature extraction.

6. the mechanical fault feature extracting method based on optimized parameter set empirical mode decomposition according to claim 1, it is characterized in that: said error e generally gets 0.01.

7. the mechanical fault feature extracting method based on optimized parameter set empirical mode decomposition according to claim 2 is characterized in that: said white noise n' _i(t) amplitude coefficient N generally gets 50.

Images