CN101251445B - Method for analysis of fractal characteristic of rotating machinery bump-scrape acoustic emission signal - Google Patents

Abstract

The invention relates to a method for analyzing the fractal property of a rubbing sound emission signal in a rotary machine. The invention comprises the following technical proposal that: an AE signal is extracted from an AE sensor of a friction point; the AE signal which is superimposed with white Gaussian noise and nonstationary noise is sampled, quantized and subframed; data comprising noise in each frame is equally divided into short-time frame data segments with the length of m; a fractal dimension value of each short-time frame data segment is calculated by a fractal dimension algorithm based on waveform; the data with a serial number bang in the middle is selected as the output of a median filtered wave. The method solves the disadvantages that the prior method has very large calculation amount, low precision of Katz dimension and large fluctuation of short-time frame fractal dimension curve. Under the environment of strong noise, with the method, logarithm wavelength dimension has the stronger capacity of distinguishing the noise from the AE signal than the Katz dimension and box dimension; the logarithm waveform dimension of an AE signal frame is closer to the fractal dimension of a pure AE signal than the Katz dimension and the box dimension, thereby effectively detecting and reflecting the occurrence of an AE rubbing event, having high precision and reducing the calculation amount.

Description

The method for analysis of fractal characteristic of rotating machinery bump-scrape acoustic emission signal

Technical field

The present invention relates to the analytical approach of rotating machinery bump-scrape acoustic emission signal, particularly the method for analysis of fractal characteristic of rotating machinery bump-scrape acoustic emission signal.

Background technology

Characteristics such as acoustic emission AE (Acoustic Emission) is highly sensitive with it, Hz-KHz is wide, contain much information, detection of dynamic have shown its superiority in bumping the fault detect that rubs.Bumping rubs produces with stronger AE signal when taking place.By the AE characteristic parameter and change and to judge and bump the generation and the intensity of rubbing, determine to bump the position of rubbing a little.But because the diversity in AE source and the complicacy of noise, the characteristic parameter of AE usually can't reflect the time of day of equipment, and especially under strong noise environment, the feature identification of AE is a difficult problem with extracting always.Utilize fractal theory to crash to rub fault to carry out Theoretical Calculation and experimental study, with its characteristic parameter as quantitative description AE signal.State and early warning that can effective recognition equipment obtain good effect.But in actual applications, the calculated amount of FRACTAL DIMENSION algorithm is to limit the bottleneck of its application.

Before the present invention, fractal dimension method commonly used at present has box dimension, correlation dimension and Katz dimension method etc.For the box dimension, owing to the increase along with system's dimension, the box quantitative indicator that needs increases, and the also corresponding increase of the data length that need be used to calculate makes that the convergence of higher-dimension system is very difficult.And correlation dimension depends critically upon two parameters: minimum delay time and the minimum dimension that embeds.If minimum delay time is too big, reach the size of attractor, then all vectors all are correlated with, and the too little relevant dimension that then calculates is exactly the relevant dimension of noise.For the AE signal, because the minimum delay time of every frame has nothing in common with each other, so each frame all must calculate its minimum delay time and just can calculate correlation dimension accurately, and calculated amount is very big.The Katz dimension does not need calculating limit, and is not strong to parameter dependence yet, and is better than other algorithms on noise resisting ability, but the Katz dimension exists the not high shortcoming of precision.

Summary of the invention

Purpose of the present invention just is to solve the defective of above-mentioned prior art, from the definition of FRACTAL DIMENSION, is purpose to reduce calculated amount and to improve precision, has designed a kind of FRACTAL DIMENSION computing method based on waveform length---wavelength logarithm method.

Technical scheme of the present invention is:

The method for analysis of fractal characteristic of rotating machinery bump-scrape acoustic emission signal the steps include:

At first the noise that contains the AE signal behind the sample quantization being carried out the branch frame handles, the short time frame data segment that again each frame signal is divided into equal length, employing is carried out medium filtering to each frame at last based on the FRACTAL DIMENSION of each section of FRACTAL DIMENSION signature analysis algorithm computation of waveform.

(1) extraction of AE signal and pre-service:

(1-1) extract the AE signal near AE sensor of friction point or the 2nd AE sensor that is positioned at the shaft coupling other end by bumping the acoustic emission experiment platform that rubs;

(1-2) bumping stack white Gaussian noise and nonstationary noise on the AE signal that rubs continuously;

(1-3) sampling quantizes, and divides frame;

(1-4) each frame is contained the short time frame data segment that the data five equilibrium of noise is grown into m;

(2) calculate the fractal dimension value of each frame:

(2-1) adopting the fractal dimension algorithm based on waveform is logarithm wavelength dimension algorithm, calculates the fractal dimension value of each short time frame data segment:

$D_i=a-\frac{\ln L\left(\mathrm{\δ}\right)}{\ln {\mathrm{\δ}}_0}-\frac{{\mathrm{b\δ}}_0}{L\left(\mathrm{\δ}\right)\ln L\left(\mathrm{\δ}\right)}$

Wherein $a=1+\frac{{\mathrm{<figure-callout\; id="1"\; label="ln\; k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">ln</figure-callout>}k}_1}{{\ln \mathrm{\&delta;}}_0},$ $b=r(\frac{\mathrm{\&delta;}}{{\mathrm{\&delta;}}_0}-1),$ δ is the yardstick of box covering curve, δ ₀Be minimum sampling interval, δ is minimum sampling interval δ ₀Integral multiple, i.e. δ=k δ ₀, L (δ) is corresponding length of curve.

Be D (2-2) with every section _I-v..., D _I-1, D _i, D _I+1..., D _I+vArrange by numerical values recited, get of the output of the data of middle sequence number as medium filtering:

$Y_i=\mathrm{Med}\{D_{i-v},...,D_{i-1},D_i,D_{i+1},...,D_{i+v}\},v=\frac{m-1}{2}$

In the method for analysis of fractal characteristic of described rotating machinery bump-scrape acoustic emission signal, described sample frequency is 5MHz, and the A/D precision during quantification is 12, and the frame length of every frame is 512 sampled points when dividing frame, and 50% frame is overlapping, and m is necessary for odd number.

In the method for analysis of fractal characteristic of described rotating machinery bump-scrape acoustic emission signal, the AE signal that an AE sensor that more approaches true sound-source signal is extracted is analyzed usually.

Advantage of the present invention and effect are:

1. under strong noise environment, logarithm wavelength dimension has the ability of stronger differentiation noise and AE signal than Katz peacekeeping box dimension, and the logarithm wavelength of AE signal frame dimension more approaches the branch dimension of pure AE signal than Katz peacekeeping box dimension, thereby can detect effectively and reflect the generation of bumping the AE incident of rubbing.

2. on the stability of computation complexity, degree of accuracy, logarithm wavelength dimension also is better than Katz peacekeeping box dimension, thereby has solved the fractal theory big problem of calculation of fault amount of rubbing that crashes that adopts effectively.

3. utilize median filtering technology, alleviated the short time frame that causes by noise and divided the dimension curve fluctuation bigger problem.

Other advantages of the present invention and effect will continue to describe below.

Description of drawings

Fig. 1---bump the acoustic emission experiment platform synoptic diagram that rubs.

Fig. 2---bump the acoustic emission signal waveform figure when rubbing continuously.

Fig. 3---when δ hour, AE squiggle figure in the box.

The branch dimension average comparison diagram of Fig. 4---logarithm wavelength dimension, Katz peacekeeping box dimension.

The computation complexity comparison diagram of Fig. 5---logarithm wavelength dimension, Katz peacekeeping box dimension.

Three kinds of frames that bump the AE signal that rubs under Fig. 6---the white noise environment (0dB) continuously divide the dimension change curve.

Fig. 7---three kinds of frames that bump the AE signal that rubs under the coloured noise environment (0dB) continuously divide the dimension change curve.

Embodiment

Below in conjunction with drawings and Examples, technical solutions according to the invention are further elaborated.

The extraction of one .AE signal and pre-service

1.AE the extraction of signal

This test adopts 3 supportings 2 to stride rotor-support-foundation system, as shown in Figure 1.Clutch shaft bearing 2, second bearing 6 and the 3rd bearing 8 all are the hydrodynamic lubrication sliding bearings, and near friction point, the 2nd AE sensor 7 is positioned at the other end of shaft coupling 5 to friction point 4 near motor 1, the one AE sensor 3.By the AE signal waveform that an AE sensor or the 2nd AE sensor extract, extract the AE signal waveform from an AE sensor in this example, can be used for analyzing the decay and the discontinuous medium coupling of the process back signal distortion situation of acoustic signal propagation.Here, we analyze the signal of an AE sensor 3 extractions of more approaching true sound-source signal.Setting sample frequency in the experiment is 5MHz, bumps continuously and rubs the AE signal waveform as shown in Figure 2.

2.AE the pre-service of signal

Bumping stack white Gaussian noise and nonstationary noise (noise source is provided by the Dutch RSRE research centre under the Britain TNO perception association) on the AE signal that rubs continuously.Signals and associated noises is sampled, quantize, divide frame.The frame length of every frame is 512 sampled points when dividing frame, and 50% frame is overlapping.

Consider that The noise may cause the branch dimension curve fluctuation of each frame bigger.Behind minute frame, the data that each frame is contained noise are the five equilibrium short time frame data segment (m is an odd number) of growing into m again, is convenient to like this after calculating each short time data frame fractal dimension value it be carried out medium filtering.

Two. based on the fractal dimension method of waveform

By the FRACTAL DIMENSION definition as can be known:

$D=-\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }\frac{\ln N_{\mathrm{\&delta;}}\left(F\right)}{\ln \mathrm{\&delta;}}$ (formula 1)

N wherein _δ(F) for yardstick be the minimum number of the box covering curve of δ, l _i(δ) for the length of side be the i box inner curve length of δ.

Order: $L\left(\mathrm{\&delta;}\right)=\underset{i=1}{\overset{N_{\mathrm{\&delta;}}\left(F\right)}{\mathrm{\&Sigma;}}}{l}_i\left(\mathrm{\&delta;}\right)$ (formula 2)

Analyze the AE signal waveform, when δ trended towards 0, the box inner curve mainly can be divided into several situations shown in Figure 3.Wherein (a) and (b) are monotonous curve, (c), (d) for comprising the curve of an extreme point, and (e), (f) comprise the curve of a plurality of extreme points.For the convenience of analyzing, when δ trends towards 0, be with the equivalence of box inner curve length:

(formula 3)

k ₁, k ₂Be respectively the ratio that comprises an extreme point curve at the most and comprise equivalent length with the box length of side of a plurality of extremals.

Then: $L_s=\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }L\left(\mathrm{\&delta;}\right)=\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }\underset{i=1}{\overset{N_{\mathrm{\&delta;}}\left(F\right)}{\mathrm{\&Sigma;}}}{l}_i\left(\mathrm{\&delta;}\right)=\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }(N_{\mathrm{\&delta;}}\left(F\right)-M)k_1\mathrm{\&delta;}+\mathrm{<figure-callout\; id="2"\; label="M\; k"\; filenames="S2008100238086E00011.png,S2008100238086E00012.png"\; state="\{\{state\}\}">M</figure-callout>}{k}_{}{}_2\mathrm{\&delta;}$

M is the box number that comprises a plurality of limits, then $N_{\mathrm{\&delta;}}\left(F\right)=\frac{L\left(\mathrm{\&delta;}\right)+M(k_1-k_2)\mathrm{\&delta;}}{{\mathrm{<figure-callout\; id="1"\; label="k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">k</figure-callout>}}_1\mathrm{\&delta;}}$

Then (formula 1) can be rewritten as:

$D=-\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }\frac{\ln [(L\left(\mathrm{\&delta;}\right)+M(k_1-k_2)\mathrm{\&delta;})/{\mathrm{<figure-callout\; id="1"\; label="k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">k</figure-callout>}}_1\mathrm{\&delta;}]}{\ln \mathrm{\&delta;}}=\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }(1+\frac{{\mathrm{<figure-callout\; id="1"\; label="ln\; k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">ln</figure-callout>}k}_1}{\ln \mathrm{\&delta;}}-\frac{\ln [L\left(\mathrm{\&delta;}\right)+M(k_1-k_2)\mathrm{\&delta;}]}{\ln \mathrm{\&delta;}})$

With Taylor series expansion (δ ₀For near 0 very little value):

$D=(1+\frac{{\mathrm{<figure-callout\; id="1"\; label="ink"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">ink</figure-callout>}}_1}{\ln {\mathrm{\&delta;}}_0}-\frac{\ln [L\left({\mathrm{\&delta;}}_0\right)+M(k_1-k_2){\mathrm{\&delta;}}_0]}{{\ln \mathrm{\&delta;}}_0})+\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }{(1+\frac{{\mathrm{<figure-callout\; id="1"\; label="ln\; k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">ln</figure-callout>}k}_1}{\ln \mathrm{\&delta;}}-\frac{\ln [L\left(\mathrm{\&delta;}\right)+M(k_1-k_2)\mathrm{\&delta;}]}{\ln \mathrm{\&delta;}})}^{\&prime;}{|}_{\mathrm{\&delta;}={\mathrm{\&delta;}}_0}(\mathrm{\&delta;}-{\mathrm{\&delta;}}_0)$ (formula 4)

Can get (formula 4) the 2nd abbreviation:

$(\frac{{\mathrm{<figure-callout\; id="1"\; label="ln\; k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">ln</figure-callout>}k}_1}{{\mathrm{\&delta;}}_0{\mathrm{<figure-callout\; id="2"\; label="ln"\; filenames="S2008100238086E00011.png,S2008100238086E00012.png"\; state="\{\{state\}\}">ln</figure-callout>}}^2{\mathrm{\&delta;}}_0}+\frac{1}{{\mathrm{\&delta;}}_0\ln {\mathrm{\&delta;}}_0}-\frac{(L^{\&prime;}\left({\mathrm{\&delta;}}_0\right)+M(k_1-k_2))}{[L\left({\mathrm{\&delta;}}_0\right)+M(k_1-k_2){\mathrm{\&delta;}}_0]\ln [L\left({\mathrm{\&delta;}}_0\right)+M(k_1-k_2){\mathrm{\&delta;}}_0]})\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }(\mathrm{\&delta;}-{\mathrm{\&delta;}}_0)$ (formula 5)

In (formula 5) the 3rd, work as δ ₀→ 0, L (δ ₀) variation be tending towards slowly making L ' (δ ₀)=r works as δ ₀→ 0, the box number that comprises a plurality of limits trends towards 0.Then:

$(-\frac{r}{L\left({\mathrm{\&delta;}}_0\right)\ln L\left({\mathrm{\&delta;}}_0\right)})\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }(\mathrm{\&delta;}-{\mathrm{\&delta;}}_0)$ (formula 6)

Can get according to (formula 4), (formula 5) and (formula 6):

$D=(1+\frac{{\mathrm{<figure-callout\; id="1"\; label="ln\; k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">ln</figure-callout>}k}_1}{{\ln \mathrm{\&delta;}}_0}-\frac{\ln [L\left({\mathrm{\&delta;}}_0\right)+M(k_1-k_2){\mathrm{\&delta;}}_0]}{\ln {\mathrm{\&delta;}}_0})+\underset{\mathrm{\&delta;}\&RightArrow;0}{\lim }(\frac{{\mathrm{<figure-callout\; id="1"\; label="ln\; k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">ln</figure-callout>}k}_1}{{\mathrm{<figure-callout\; id="2"\; label="ln"\; filenames="S2008100238086E00011.png,S2008100238086E00012.png"\; state="\{\{state\}\}">ln</figure-callout>}}^2{\mathrm{\&delta;}}_0}+\frac{1}{\ln {\mathrm{\&delta;}}_0}-\frac{r{\mathrm{\&delta;}}_0}{L\left({\mathrm{\&delta;}}_0\right)\ln L\left({\mathrm{\&delta;}}_0\right)})(\frac{\mathrm{\&delta;}}{{\mathrm{\&delta;}}_0}-1)$

Work as δ ₀→ 0 o'clock, in the following formula $\frac{\ln {\mathrm{<figure-callout\; id="1"\; label="k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">k</figure-callout>}}_1}{{\mathrm{<figure-callout\; id="2"\; label="ln"\; filenames="S2008100238086E00011.png,S2008100238086E00012.png"\; state="\{\{state\}\}">ln</figure-callout>}}^2{\mathrm{\&delta;}}_0}$ With $\frac{1}{{\ln \mathrm{\&delta;}}_0}$ All → 0, further abbreviation gets:

$D=a-\frac{\ln [L\left({\mathrm{\&delta;}}_0\right)+\mathrm{Mc}{\mathrm{\&delta;}}_0]}{\ln {\mathrm{\&delta;}}_0}-\frac{{\mathrm{b\&delta;}}_0}{L\left({\mathrm{\&delta;}}_0\right)\ln L\left({\mathrm{\&delta;}}_0\right)}$ (formula 7)

In order further to reduce calculated amount, consider and work as δ ₀≈ δ → 0, the box that comprises a plurality of limits is counted M and is also trended towards 0, can get:

$D=a-\frac{\ln L\left(\mathrm{\&delta;}\right)}{\ln {\mathrm{\&delta;}}_0}-\frac{{\mathrm{b\&delta;}}_0}{L\left(\mathrm{\&delta;}\right)\ln L\left(\mathrm{\&delta;}\right)}$ (formula 8)

Wherein: $a=1+\frac{\mathrm{<figure-callout\; id="1"\; label="ln\; k"\; filenames="S2008100238086E00013.png,S2008100238086E00021.png"\; state="\{\{state\}\}">ln</figure-callout>}{k}_{}{}_1}{\ln {\mathrm{\&delta;}}_0},$ $b=r(\frac{\mathrm{\&delta;}}{{\mathrm{\&delta;}}_0}-1)$

(formula 8) is the fractal dimension algorithm based on waveform, is referred to as logarithm wavelength dimension.δ in the formula ₀Can be the integral multiple of minimum sampling interval δ, make δ ₀=k δ, L (δ ₀) be corresponding length of curve.A, b, the k parameter adopts fractal Blang's curve to determine.

Calculate the fractal dimension value of each short time frame data segment, i.e. D by (formula 8) _I-v..., D _I-1, D _i, D _I+1..., D _I+v, and, get of the output of the data of middle sequence number as medium filtering by the numerical values recited arrangement:

$Y_i=\mathrm{Med}\{D_{i-v},...,D_{i-1,}{D}_i,D_{i+1},...,D_{i+v}\}v=\frac{m-1}{2}$ (formula 9)

Because medium filtering is a kind of nonlinear filtering, therefore the statistical property of undesired signal can weaken the influence of random disturbance and impulse disturbances well, and frequency spectrum not had influence substantially.

Three. performance evaluation

For the performance of logarithm wavelength dimension relatively, it and box peacekeeping Katz dimension are compared, Fig. 4 is a comparative result, the fractal Blang's curve in this comparison procedure is that computing machine passes through 15 iteration generations, totally 2 ¹⁵=32786 points, program is divided frame with this curve, and 512 points of every frame calculate the FRACTAL DIMENSION of every frame, and the fractal dimension value of each frame is averaged obtains net result.As can be seen from Figure 4, the degree of accuracy of logarithm wavelength dimension is the highest, secondly is the box dimension, is the Katz algorithm at last.

Fig. 5 is the computation complexity comparative result of logarithm wavelength dimension, Katz peacekeeping box dimension, m is a median filter length, assumed curve has N point, the step-length of calculating logarithm wavelength peacekeeping Katz dimension is 1, the capsule length of calculation box dimension also is 1, total k point participated in the curve fitting of box dimension, and its box length is respectively 1, l ₁, l ₂..., l _kAs can be seen from Figure 5, the addition number of times of box dimension, multiplication number of times, nonlinear operation time number average will obviously be tieed up greater than Katz peacekeeping logarithm wavelength; And logarithm wavelength dimension all is being in minimum in the three aspect addition, multiplication, comparison, the nonlinear operation.

Fig. 6 and Fig. 7 have provided three kinds of frames that bump the AE signal that rubs under white noise environment (0dB) and the coloured noise environment (0dB) continuously respectively and have divided the curve map of dimension variation, as can be seen from the figure:

(1) in the noisy AE signal, logarithm wavelength method frame divides dimension curve relatively level and smooth, has the many places fractal dimension value will be significantly less than the fractal dimension value of noise in Fig. 6 and Fig. 7, and can judge these several places has the AE signal to produce;

(2) analyzes three kinds of frames and divide the noise segment of dimension curve can find out that the fluctuation of logarithm wavelength method is minimum, and the fluctuation ratio of Katz dimension, box dimension is bigger, can draw logarithm wavelength dimension than Katz tie up, the box dimension is more stable;

(3) the signal segment logarithm wavelength of noisy AE signal dimension obviously is less than noise segment, and box ties up that then difference of them is little, and under the coloured noise environment, box peacekeeping Katz is not for almost distinguishing noise and AE signal.

Fig. 6,7 explanations, under the very noisy situation, logarithm wavelength method has more the ability of distinguishing noise and bumping the AE signal that rubs, and the logarithm wavelength of AE signal frame dimension more approaches the branch dimension of pure AE signal than Katz peacekeeping box dimension.

Above result shows that under strong noise environment, logarithm wavelength dimension can reflect the generation of bumping the AE incident of rubbing; Logarithm wavelength dimension has the ability of stronger differentiation noise and AE signal than Katz peacekeeping box dimension; On the stability of computation complexity, degree of accuracy, also be better than Katz peacekeeping box dimension.The good performance of this algorithm is the feature identification of bumping the AE signal that rubs and provides a new approach with analysis, is that the AE signal that rubs reasonable branch dimension calculating and analytical approach are bumped in a kind of processing.

The scope that the present invention asks for protection is not limited only to the description of this embodiment.

Claims (3)

1. the method for analysis of fractal characteristic of rotating machinery bump-scrape acoustic emission signal the steps include:

At first the noise that contains the AE signal behind the sample quantization being carried out the branch frame handles, the short time frame data segment that again each frame signal is divided into equal length, employing is carried out medium filtering to each section at last based on the fractal dimension value of each section of FRACTAL DIMENSION signature analysis algorithm computation of waveform;

(1) extraction of AE signal and pre-service:

(1-1) by bumping the 2nd AE sensor extraction AE signal that the acoustic emission experiment platform that rubs is positioned at an AE sensor of shaft coupling one end near friction point or is positioned at the shaft coupling other end;

(1-2) bumping stack white Gaussian noise and nonstationary noise on the AE signal that rubs continuously;

(1-3) sampling quantizes, and divides frame;

(1-4) each frame is contained the short time frame data segment that the data five equilibrium of noise is grown into m;

(2) calculate the fractal dimension value of each section:

(2-1) adopting the fractal dimension algorithm based on waveform is logarithm wavelength dimension algorithm, calculates the fractal dimension value of each short time frame data segment,

$D_i=a-\frac{\ln L\left(\mathrm{\&delta;}\right)}{\ln {\mathrm{\&delta;}}_0}-\frac{{\mathrm{b\&delta;}}_0}{L\left(\mathrm{\&delta;}\right)\ln L\left(\mathrm{\&delta;}\right)}$

Wherein $a=1+\frac{\ln k_1}{\ln {\mathrm{\&delta;}}_0},$ $b=r(\frac{\mathrm{\&delta;}}{{\mathrm{\&delta;}}_0}-1),$ δ is the yardstick of box covering curve, δ ₀Be minimum sampling interval, δ is minimum sampling interval δ ₀Integral multiple, i.e. δ=k δ ₀, L (δ) is corresponding length of curve; k ₁Be the equivalent length that comprises an extreme point curve at the most ratio with the box length of side; R=L ' (δ ₀);

Be D (2-2) with every section _I-v..., D _I-1, D _i, D _I+1..., D _I+vArrange by numerical values recited, get of the output of the data of middle sequence number as the medium filtering of this section:

$\begin{array}{cc}{Y}_i=\mathrm{Med}\{D_{i-v},...,D_{i-1},D_i,D_{i+1},...,D_{i+v}\}& v=\frac{m-1}{2}\end{array}$

M is an odd number.

2. the method for analysis of fractal characteristic of rotating machinery bump-scrape acoustic emission signal according to claim 1 is characterized by, and the sample frequency during sampling is 5MHz, A/D precision during quantification is 12, the frame length of every frame is 512 sampled points when dividing frame, and 50% frame is overlapping, and m is necessary for odd number.

3. the method for analysis of fractal characteristic of rotating machinery bump-scrape acoustic emission signal according to claim 1 is characterized by, and the AE signal that an AE sensor that more approaches true sound-source signal is extracted is analyzed usually.

Images