CN105678270A - Screening method for stress wave signal characteristics of one-dimensional member - Google Patents

Abstract

The invention discloses a screening method for the stress wave signal characteristics of a one-dimensional member. The method comprises the following steps of 1, pre-processing an original stress wave signal to obtain m sub-waves; 2, adopting the quantized information entropy as the characteristics of the stress wave signal of a one-dimensional member; 3, constructing a time window of a fixed width W, and extracting the signal characteristics according to a step length B for each sub-wave signal to a imensional stress wave quantized information entropy array; 4, conducting the dimension reduction treatment on the multi-dimensional stress wave quantized information entropy array to obtain a one-dimensional quantized information entropy mean vector; 5, based on the multiple parameter correlation analysis method in the grey system theory, screening out the characteristics of the stress wave through calculating the relevancy between the to-be-detected state mode of multiple invalid characteristics and the standard and normal state mode. According to the technical scheme of the invention, the above method is better in practical application value. Meanwhile, the quantized information entropy is adopted as the characteristics, so that the method is good in anti-noise performance and does not occupy the excessive amount of time resource.

Description

The screening technique of one-dimensional component stress ripple signal characteristic

Technical field

The invention belongs to signal characteristic identification field, particularly relate to a kind of based on gray system theory and quantitative information entropy, the screening technique of one-dimensional component stress ripple signal characteristic.

Background technology

The present invention is the key component of one-dimensional component Dynamic Non-Destruction Measurement. One-dimensional component Non-Destructive Testing is to apply technology widely, and particularly at building pile foundation, bearing and anchor pole etc., the selection of stress wave signal eigenvalue and extraction are parts most crucial in this technology.

On one-dimensional member integrity detects, most widely used method is stress wave at present, that is: stress wave exciter and signal receiving sensor is disposed in same one end of one-dimensional component, after elastic stress wave, shockwave sensor can receive complete stress wave signal waveform, waveform comprises abundant information, by analyzing signal waveform characteristic quantity, it is possible to obtain the integrity information of one-dimensional component. Stress wave characteristic quantity common at present includes the amplitude of signal, energy (power spectral density), variance etc., and the mode extracting characteristic quantity has extraction characteristic quantity in the whole time-histories of signal, also has first by signal subsection, extracts the characteristic quantity of every segment signal. No matter it is in the whole time-histories of signal, or the characteristic quantity of stage extraction signal, the omission of damage characteristic and the reduction of detection resolution can be caused all unavoidably.

Summary of the invention

It is an object of the invention to provide a kind of high reliability, high-resolution, the screening technique of one-dimensional component stress ripple signal characteristic.

The screening technique of one-dimensional component stress ripple signal characteristic, comprises the following steps,

Step one: primitive stress ripple signal is carried out pretreatment, decomposes with the method correspondence wave signal of WAVELET PACKET DECOMPOSITION, obtains m wavelet;

Step 2: using the quantitative information entropy characteristic quantity as one-dimensional component stress ripple signal;

Step 3: the time window of structure fixed width W, moves extraction signal characteristic value by stepping length B on each wavelet signal, obtains the stress wave quantitative information entropy matrix of multidimensional;The value of stepping length B meets W/8 < B < W/4;

Step 4: the stress wave quantitative information entropy matrix of multidimensional is carried out dimension-reduction treatment, seeks the average of the eigenvalue of all wavelet correspondence same positions, obtain one-dimensional quantitative information entropy mean vector;

Step 5: the multiparameter association analysis method in application gray system theory, the degree of association of state model to be checked Yu standard normal condition pattern by calculating multiple failure characteristics amount, the size of the contrast degree of association judges the performance state of system, completes the screening of stress wave characteristic quantity.

The screening technique of the one-dimensional component stress ripple signal characteristic of the present invention, it is also possible to including:

1, quantitative information entropy is:

$\mathrm{DH}\left(X\right)=\underset{i=0}{\overset{D-1}{\mathrm{\&Sigma;}}}{\mathrm{Num}}_i{p}_i{\log }_2\frac{1}{p_i}$

Wherein, D is the number that amplitude is interval, Num_iRepresent sampled point number in i-th amplitude interval, p_iFor the statistical probability of sampling number in i-th amplitude interval.

2, stress wave quantitative information entropy matrix is:

$H=\left[\begin{array}{cccc}{\mathrm{DH}}_{11}& {\mathrm{DH}}_{12}& \mathrm{...}& {\mathrm{DH}}_{1n}\\ {\mathrm{DH}}_{21}& {\mathrm{DH}}_{22}& \mathrm{...}& {\mathrm{DH}}_{2n}\\ \mathrm{...}& \mathrm{...}& \mathrm{...}& \mathrm{...}\\ {\mathrm{DH}}_{m1}& {\mathrm{DH}}_{m2}& \mathrm{...}& {\mathrm{DH}}_{mn}\end{array}\right]$

Where each row data represent the quantitative information entropy vector of a wavelet, and m wavelet constitutes m × n characteristic quantity matrix;

Each wavelet entropy vector is:

H_m={ DH_m1, DH_m2, DH_m3..., DH_mn(m=1,2 ..., 2^k)

Wherein, stress wave, after wavelet packet k layer decomposes, obtains m=2^kIndividual signal wavelet, time window number of steps is n.

3, quantitative information entropy mean vector is:

$\stackrel{\&OverBar;}{H}=\left\{{\stackrel{\&OverBar;}{DH}}_1,{\stackrel{\&OverBar;}{DH}}_2,{\stackrel{\&OverBar;}{DH}}_3,\mathrm{...},{\stackrel{\&OverBar;}{DH}}_n\right\}$

${\stackrel{\&OverBar;}{DH}}_i=\frac{1}{m}\left({\mathrm{DH}}_{1i}+{\mathrm{DH}}_{2i}+\mathrm{...}+{\mathrm{DH}}_{mi}\right),\left(i=1,2,\mathrm{...},n\right).$

4, in step 5, state Y to be checked_iState model vector be:

y⁽ⁱ⁾={ y₁ ⁽ⁱ⁾, y₂ ⁽ⁱ⁾, y₃ ⁽ⁱ⁾…y_n ⁽ⁱ⁾}^T

Wherein suspect system has the several characteristic parameter of n;

Standard normal condition Y₀State model vector be:

y⁽⁰⁾={ y₁ ⁽⁰ⁱ⁾, y₂ ⁽⁰ⁱ⁾, y₃ ⁽⁰ⁱ⁾…y_n ⁽⁰⁾}^T

State Y to be checked_jFor standard normal condition Y₀Thresholding coefficient of association on jth characteristic parameter is:

${r_j}^{\left(i\right)}=\frac{{\mathrm{k\&sigma;}}_j^0}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

Then state Y to be checked_jFor standard normal condition Y₀Threshold service be:

$r^{\left(i\right)}=\frac{1}{{\mathrm{\&Sigma;}}_{j=1}^n{\mathrm{\&alpha;}}_j}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_j\frac{{\mathrm{k\&sigma;}}_j^0}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

Work as r_jWhen >=0.5, state to be checked belongs to standard normal condition; Work as r_jWhen≤0.5, state to be checked is not belonging to standard normal condition. Beneficial effect

Compared with existing algorithmic technique, the invention has the beneficial effects as follows the dimensionality reduction achieving characteristic quantity, the screening that gray system theory is stress wave signal characteristic quantity provides new thinking, for the later stage, the Non-Destructive Testing of one-dimensional component improves detection resolution, and noiseproof feature is excellent, do not occupy too much time resource.

The present invention is by signal for quantitative information entropy characteristic quantity; Build time window is instrument, at Signal Move dynamic extraction characteristic quantity; By signal characteristic quantity dimensionality reduction, tectonic stress wave characteristic vector. Compare based on stage extraction stress wave signal amplitude structural feature vector with known, utilize the threshold service evaluation methodology as characteristic vector performance reliability of gray system theory, judge relative analysis to two feature vectors, The inventive method achieves the optimization of characteristic quantity and screening, higher performance reliability is still obtained when signal exists many places singular point, improve Non-Destructive Testing resolution, and noiseproof feature is excellent, do not occupy too much time resource.

Accompanying drawing explanation

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

Fig. 2 is that time window moves extraction signal characteristic value.

Fig. 3 is the normalized vector curve of the quantitative information entropy that the present invention proposes.

Fig. 4 is the normalized vector curve as characteristic quantity of the amplitude as a comparison.

Detailed description of the invention

Below in conjunction with accompanying drawing, the present invention is described in further details.

In order to overcome above-mentioned existing methodical deficiency, the invention provides a kind of screening technique based on gray system theory and the one-dimensional component stress ripple signal characteristic of quantitative information entropy.

Technical scheme is as follows:

The stress wave signal being used for detecting one-dimensional member integrity is carried out pretreatment, decomposes by the method for WAVELET PACKET DECOMPOSITION, obtain a number of wavelet;The concept of quantitative information entropy is proposed, by signal for quantitative information entropy eigenvalue; The time window of structure fixed width, as instrument, moves extraction signal characteristic value by fixing stepping length on each wavelet signal, obtains the signal characteristic quantity of multidimensional; Multidimensional characteristic amount is carried out dimension-reduction treatment, seeks the average of the eigenvalue of all wavelet correspondence same positions, obtain one-dimensional signal characteristic quantity; Utilize the multi parameter analysis method of gray system theory, utilize threshold service evaluation to quantify Information Entropy Features vector sum amplitude Characteristics vector, reach the purpose that stress wave characteristic quantity is evaluated and screened.

Comprising the concrete steps that of the method:

(1) primitive stress ripple signal is carried out pretreatment, carry out the decomposition of the suitable number of plies with the method correspondence wave signal of WAVELET PACKET DECOMPOSITION, obtain some wavelets. As shown in Figure 24 wavelets by primitive stress ripple through 2 layers of WAVELET PACKET DECOMPOSITION.

(2) concept of quantization comentropy is lifted up out at the conceptual foundation of comentropy, as the eigenvalue of stress wave signal. If signal sampling amplitude set is X={x₁, x₂..., x_n, the probability distribution of X is expressed as p_i=P (x_i) (i=1,2 ..., n), simultaneouslyTake the maxima and minima of variable X, be designated as x respectively_maxAnd x_min, in conjunction with practical situation, by amplitude by x_minTo x_maxIt is divided into D quantization amplitude interval. Then the value length of every section of amplitude is:

$\mathrm{\&Delta;}d=\frac{x_{\max }-x_{\min }}{D}$

The quantization span of every section of amplitude is:

(x_min+ i Δ d, x_min+ (i+1) Δ d) (i=0,1,2 ..., D-1)

If the sampled point number in i-th section of amplitude interval is Num_i, then the statistical probability in i-th section of amplitude district is defined as:

$P_i=\frac{{\mathrm{Num}}_i}{n}$

Then the quantitative information entropy of signal is defined as:

$DH\left(X\right)=\underset{i=0}{\overset{D-1}{\&Sigma;}}{\mathrm{Num}}_i{p}_i{\log }_2\frac{1}{p_i}$

Wherein DH (X) is the quantitative information entropy of signal, and D is the number that amplitude is interval, Num_iRepresent sampled point number in i-th amplitude interval, p_iFor the statistical probability of sampling number in i-th amplitude interval.

(3) construct the time window of fixed width as instrument, each wavelet signal moves extraction signal characteristic value by fixing stepping length, obtains the signal characteristic quantity of multidimensional. For the time window of fixed width, the setting of call parameter to be paid special attention to. If time window width is W, stepping length is B. The value of time window width W to cover at least one complete singular point, simultaneously as far as possible less than two singular points; The value of stepping length B meets W/8 < B < W/4, to meet when extracting characteristic quantity by stepping, singular point will not be omitted, without repeating to process same singular point, minimizing processes the time, improves the resolution of singular point.

(4) set stress wave after wavelet packet k layer decomposes, obtain m=2^kIndividual signal wavelet, time window number of steps is set to n, extracts characteristic quantity, and namely after quantitative information entropy, each wavelet entropy vector is:

H_m={ DH_m1, DH_m2, DH_m3..., DH_mn(m=1,2 ..., 2^k)

Stress wave quantitative information entropy matrix construction is as follows:

$H=\left[\begin{array}{cccc}{\mathrm{DH}}_{11}& {\mathrm{DH}}_{12}& ...& {\mathrm{DH}}_{1n}\\ {\mathrm{DH}}_{21}& {\mathrm{DH}}_{22}& ...& {\mathrm{DH}}_{2n}\\ ...& ...& ...& ...\\ {\mathrm{DH}}_{m1}& {\mathrm{DH}}_{m2}& ...& {\mathrm{DH}}_{mn}\end{array}\right]$

Where each row data represent the quantitative information entropy vector of a wavelet, and m wavelet constitutes m × n characteristic quantity matrix.

(5) multidimensional characteristic moment matrix is carried out dimension-reduction treatment, seek the average of the eigenvalue of all wavelet correspondence same positions, obtain one-dimensional signal characteristic quantity, one complicated characteristic quantity matrix can not analyze singular point preferably, on the basis of multidimensional characteristic moment matrix, try to achieve the quantitative information entropy average of the time window of corresponding same segment signal:

${\stackrel{\&OverBar;}{DH}}_i=\frac{1}{m}\left({\mathrm{DH}}_{1i}+{\mathrm{DH}}_{2i}+\mathrm{...}+{\mathrm{DH}}_{mi}\right),(i=1,2,\mathrm{...},n)$

Process obtains quantitative information entropy mean vector:

$\stackrel{\&OverBar;}{H}=\left\{{\stackrel{\&OverBar;}{DH}}_1,{\stackrel{\&OverBar;}{DH}}_2,{\stackrel{\&OverBar;}{DH}}_3,\mathrm{...},{\stackrel{\&OverBar;}{DH}}_n\right\}$

(6) the threshold service method of evaluating performance of gray system theory is as follows: assume a suspect system with the several characteristic parameter of n, its state Y to be checked_iState model vector be:

y⁽ⁱ⁾={ y₁ ⁽ⁱ⁾, y₂ ⁽ⁱ⁾, y₃ ⁽ⁱ⁾…y_n ⁽ⁱ⁾}^T

Standard normal condition Y₀State model vector be:

y⁽⁰⁾={ y₁ ⁽⁰ⁱ⁾, y₂ ⁽⁰ⁱ⁾, y₃ ⁽⁰ⁱ⁾…y_n ⁽⁰⁾}^T

If each characteristic parameter y_j ⁽⁰ⁱ⁾All obeying average isVariance isNormal distribution, define state Y to be checked_jFor standard normal condition Y₀Thresholding coefficient of association on jth characteristic parameter is:

${r_j}^{\left(i\right)}=\frac{{\mathrm{k\&sigma;}}_j^0}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

Then state Y to be checked_jFor standard normal condition Y₀Threshold service be:

$r^{\left(i\right)}=\frac{1}{{\mathrm{\&Sigma;}}_{j=1}^n{\mathrm{\&alpha;}}_j}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_j\frac{{\mathrm{k\&sigma;}}_j^0}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

The judgment criterion of state to be checked is: work as r_jWhen >=0.5, state to be checked belongs to standard normal condition;Work as r_jWhen≤0.5, state to be checked is not belonging to standard normal condition.

Present invention relates particularly to the one-dimensional component stress ripple signal characteristic screening technique based on gray system theory and quantitative information entropy, step is as follows: the stress wave signal being used for detecting one-dimensional member integrity is carried out pretreatment by (1), decompose by the method for WAVELET PACKET DECOMPOSITION, obtain a number of wavelet; (2) by signal for quantitative information entropy eigenvalue; (3) construct the time window of fixed width as instrument, each wavelet signal moves extraction signal characteristic value by fixing stepping length, obtains the signal characteristic quantity of multidimensional; (4) multidimensional characteristic amount is carried out dimension-reduction treatment, seek the average of the eigenvalue of all wavelet correspondence same positions, obtain one-dimensional signal characteristic quantity; (5) utilize the multiparameter association analysis method in gray system theory, based on the performance reliability evaluation methodology of threshold service, evaluate the performance quantifying Information Entropy Features amount; (6) testing process terminates. The present invention has good actual application value in the evaluation and screening of stress wave signal feature, and the quantitative information entropy that the present invention proposes is excellent as characteristic quantity noiseproof feature, does not occupy too much time resource.

Screening technique based on gray system theory and the one-dimensional component stress ripple signal characteristic of quantitative information entropy, step is as follows: the stress wave signal being used for detecting one-dimensional member integrity is carried out pretreatment by (1), decompose by the method for WAVELET PACKET DECOMPOSITION, obtain a number of wavelet; (2) by signal for quantitative information entropy eigenvalue, the quantitative information entropy of each wavelet is extracted; (2) construct the time window of fixed width as instrument, each wavelet signal moves extraction signal characteristic value by fixing stepping length, obtains the signal characteristic quantity of multidimensional; (3) multidimensional characteristic amount is carried out dimension-reduction treatment, seek the average of the eigenvalue of all wavelet correspondence same positions, obtain one-dimensional signal characteristic quantity; (4) utilize the multiparameter association analysis method in gray system theory, based on the performance reliability evaluation methodology of threshold service, evaluate the performance quantifying Information Entropy Features amount and skin quality characteristic quantity, reach the purpose of stress wave characteristic quantity screening.

Based on the method for the one-dimensional component stress ripple detection of Singular Point of quantitative information entropy, concretely comprising the following steps of described method:

A, primitive stress ripple signal is carried out pretreatment, do WAVELET PACKET DECOMPOSITION, obtain some wavelet signals of primitive stress ripple signal.

B, lift up out the concept of quantization comentropy on the basis of comentropy, as the characteristic quantity of one-dimensional component stress ripple signal.

C, structure fixed width W time window, the time-histories of signal moves extraction signal characteristic quantity by certain step-length B. The characteristic quantity matrix of structure signal.

D, characteristic quantity matrix to signal do dimension-reduction treatment, analyze the positional information judging Singular Point.

Primitive stress ripple signal is carried out WAVELET PACKET DECOMPOSITION, it is possible to the high and low frequency part of signal is decomposed simultaneously, both will not make the information dropout of signal, also achieve the decomposition that signal is more fine, reach more excellent Time-Frequency Localization effect. Signal is carried out n-layer wavelet packet decomposition, the 2 of different frequency scope can be obtainedⁿIndividual wavelet signal.

Quantitative information entropy is defined as, if signal sampling amplitude set is X={x₁, x₂..., x_n, the probability distribution of X is expressed as p_i=P (x_i) (i=1,2 ..., n), simultaneouslyTake the maxima and minima of variable X, be designated as x respectively_maxAnd x_min, in conjunction with practical situation, by amplitude by x_minTo x_maxIt is divided into D quantization amplitude interval.Then the value length of every section of amplitude is:

$\mathrm{\&Delta;}d=\frac{x_{\max }-x_{\min }}{D}$

The quantization span of every section of amplitude is:

(x_min+ i Δ d, x_min+ (i+1) Δ d) (i=0,1,2 ..., D-1)

If the sampled point number in i-th section of amplitude interval is Num_i, then the statistical probability in i-th section of amplitude district is defined as:

$p_i=\frac{{\mathrm{Num}}_i}{n}$

Then the quantitative information entropy of signal is defined as:

$DH\left(X\right)=\underset{i=0}{\overset{D-1}{\&Sigma;}}{\mathrm{Num}}_i{p}_i{\log }_2\frac{1}{p_i}$

Wherein DH (X) is the quantitative information entropy of signal, and D is the number that amplitude is interval, Num_iRepresent sampled point number in i-th amplitude interval, p_iFor the statistical probability of sampling number in i-th amplitude interval.

For the time window of fixed width, the setting of call parameter to be paid special attention to. If time window width is W, stepping length is B. The value of time window width W to cover at least one complete singular point, simultaneously as far as possible less than two singular points; The value of stepping length B meets W/8 < B < W/4, to meet when extracting characteristic quantity by stepping, singular point will not be omitted, without repeating to process same singular point, minimizing processes the time, improves the resolution of singular point.

When solving the quantitative information entropy matrix of signal, if stress wave is after wavelet packet k layer decomposes, obtain m=2^kIndividual signal wavelet, time window number of steps is set to n, extracts characteristic quantity, and namely after quantitative information entropy, each wavelet entropy vector is:

H_m={ DH_m1, DH_m2, DH_m3..., DH_mn(m=1,2 ..., 2^k)

Stress wave quantitative information entropy matrix construction is as follows:

$H=\left[\begin{array}{cccc}{\mathrm{DH}}_{11}& {\mathrm{DH}}_{12}& ...& {\mathrm{DH}}_{1n}\\ {\mathrm{DH}}_{21}& {\mathrm{DH}}_{22}& ...& {\mathrm{DH}}_{2n}\\ ...& ...& ...& ...\\ {\mathrm{DH}}_{m1}& {\mathrm{DH}}_{m2}& ...& {\mathrm{DH}}_{mn}\end{array}\right]$

Where each row data represent the quantitative information entropy vector of a wavelet, and m wavelet constitutes m × n characteristic quantity matrix.

Multidimensional characteristic moment matrix is carried out dimension-reduction treatment, seek the average of the eigenvalue of all wavelet correspondence same positions, obtain one-dimensional signal characteristic quantity, one complicated characteristic quantity matrix can not analyze singular point preferably, on the basis of multidimensional characteristic moment matrix, try to achieve the quantitative information entropy average of the time window of corresponding same segment signal:

${\stackrel{\&OverBar;}{DH}}_i=\frac{1}{m}\left({\mathrm{DH}}_{1i}+{\mathrm{DH}}_{2i}+\mathrm{...}+{\mathrm{DH}}_{mi}\right),\left(i=1,2,\mathrm{...},n\right)$

Process obtains quantitative information entropy mean vector:

$\stackrel{\&OverBar;}{H}=\left\{{\stackrel{\&OverBar;}{DH}}_1,{\stackrel{\&OverBar;}{DH}}_2,{\stackrel{\&OverBar;}{DH}}_3,\mathrm{...},{\stackrel{\&OverBar;}{DH}}_n\right\}.$

The quantitative information entropy mean vector of structure, draws the dimension of the characteristic vector of measured signal, utilizes amplitude as the method for characteristic quantity according to conventional, constructs the amplitude Characteristics vector of equal dimension.

Performance reliability evaluation methodology based on threshold service is proposed, multiparameter association analysis method in application gray system theory, the degree of association of state model to be checked Yu standard normal condition pattern by calculating multiple failure characteristics amount, the size of the contrast degree of association judges the performance state of system, and then carry out performance failure analysis, the reliability of computing system performance.

The threshold service method of evaluating performance of gray system theory is as follows: assume a suspect system with the several characteristic parameter of n, its state Y to be checked_iState model vector be:

y⁽ⁱ⁾={ y₁ ⁽ⁱ⁾, y₂ ⁽ⁱ⁾, y₃ ⁽ⁱ⁾…y_n ⁽ⁱ⁾}^T

Standard normal condition Y₀State model vector be:

y⁽⁰⁾={ y₁ ⁽⁰ⁱ⁾, y₂ ⁽⁰ⁱ⁾, y₃ ⁽⁰ⁱ⁾…y_n ⁽⁰⁾}^T

If each characteristic parameter y_j ⁽⁰ⁱ⁾All obeying average isVariance isNormal distribution, define state Y to be checked_jFor standard normal condition Y₀Thresholding coefficient of association on jth characteristic parameter is:

${r_j}^{\left(i\right)}=\frac{{\mathrm{k\&sigma;}}_j^0}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

Then state Y to be checked_jFor standard normal condition Y₀Threshold service be:

$r^{\left(i\right)}=\frac{1}{{\mathrm{\&Sigma;}}_{j=1}^n{\mathrm{\&alpha;}}_j}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_j\frac{{\mathrm{k\&sigma;}}_j^o}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

The judgment criterion of state to be checked is: work as r_jWhen >=0.5, state to be checked belongs to standard normal condition; Work as r_jWhen≤0.5, state to be checked is not belonging to standard normal condition.

Based on the evaluation methodology of gray system theory and the one-dimensional component stress ripple signal characteristic of quantitative information entropy as it is shown in figure 1, it specifically comprises the following steps that

(1) primitive stress ripple signal is carried out pretreatment, carry out the decomposition of the suitable number of plies with the method correspondence wave signal of WAVELET PACKET DECOMPOSITION, obtain some wavelets.

(2) lift up out the concept of quantization comentropy at the conceptual foundation of comentropy, as the eigenvalue of stress wave signal, seek the quantitative information entropy of signal as follows:

$DH\left(X\right)=\underset{i=0}{\overset{D-1}{\&Sigma;}}{\mathrm{Num}}_i{p}_i{\log }_2\frac{1}{p_i}$

(3) construct the time window of fixed width as instrument, each wavelet signal moves extraction signal characteristic value by fixing stepping length, obtains the signal characteristic quantity of multidimensional. The value of stepping length B and time window width W meets W/8 < B < W/4.

(4) stress wave quantitative information entropy matrix construction is as follows:

$H=\left[\begin{array}{cccc}{\mathrm{DH}}_{11}& {\mathrm{DH}}_{12}& ...& {\mathrm{DH}}_{1n}\\ {\mathrm{DH}}_{21}& {\mathrm{DH}}_{22}& ...& {\mathrm{DH}}_{2n}\\ ...& ...& ...& ...\\ {\mathrm{DH}}_{m1}& {\mathrm{DH}}_{m2}& ...& {\mathrm{DH}}_{mn}\end{array}\right]$

(5) multidimensional characteristic moment matrix being carried out dimension-reduction treatment, obtaining quantitative information entropy mean vector is:

$\stackrel{\&OverBar;}{H}=\left\{{\stackrel{\&OverBar;}{DH}}_1,{\stackrel{\&OverBar;}{DH}}_2,{\stackrel{\&OverBar;}{DH}}_3,\mathrm{...},{\stackrel{\&OverBar;}{DH}}_n\right\}$

(6) the threshold service method of evaluating performance of gray system theory is as follows: assume a suspect system with the several characteristic parameter of n, its state Y to be checked_iState model vector be:

y⁽ⁱ⁾={ y₁ ⁽ⁱ⁾, y₂ ⁽ⁱ⁾, y₃ ⁽ⁱ⁾…y_n ⁽ⁱ⁾}^T

Standard normal condition Y₀State model vector be:

y⁽⁰⁾={ y₁ ⁽⁰ⁱ⁾, y₂ ⁽⁰ⁱ⁾, y₃ ⁽⁰ⁱ⁾…y_n ⁽⁰⁾}^T

If each characteristic parameter y_j ⁽⁰ⁱ⁾All obeying average isVariance isNormal distribution, define state Y to be checked_jFor standard normal condition Y₀Thresholding coefficient of association on jth characteristic parameter is:

${r_j}^{\left(i\right)}=\frac{{\mathrm{k\&sigma;}}_j^0}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

Then state Y to be checked_jFor standard normal condition Y₀Threshold service be:

$r^{\left(i\right)}=\frac{1}{{\mathrm{\&Sigma;}}_{j=1}^n{\mathrm{\&alpha;}}_j}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_j\frac{{\mathrm{k\&sigma;}}_j^o}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

The judgment criterion of state to be checked is: work as r_jWhen >=0.5, state to be checked belongs to standard normal condition; Work as r_jWhen≤0.5, state to be checked is not belonging to standard normal condition.

Fig. 2 is that time window moves extraction signal characteristic value, and 4 wavelet signals are to be got through 2 layers of WAVELET PACKET DECOMPOSITION by stress wave signal, and wherein the relation of time window width W and stepping length B is: W/8≤B≤W/4.

Fig. 3 is the normalized vector curve of the quantitative information entropy that the present invention proposes; Curvilinear abscissa express time window figure place, vertical coordinate represents normalized quantitative information entropy average.

Fig. 4 is the normalized vector curve as characteristic quantity of the amplitude as a comparison; Curvilinear abscissa represents segments, and equal to time window figure place, vertical coordinate represents normalized quantitative information entropy average.

As shown in Figure 3, the stress wave signal choosing the known damage position of pile measurement gained is experimental subject, 3 layers of decomposition are carried out with sym8 wavelet packet, time window width W is taken as 1ms, in time window, temporally average sample is 100 points, stepping length B value is 0.2ms, i.e. 20 sampled points, and amplitude segments D is set to 10. Result is as shown in Figure 3.

With known based on utilizing the amplitude method as characteristic quantity structural feature vector, the present invention is by signal for quantitative information entropy characteristic quantity; Build time window is instrument, at Signal Move dynamic extraction characteristic quantity; By signal characteristic quantity dimensionality reduction, construct quantitative information entropy feature vector; The method proposing the threshold service of application gray system theory, evaluate the characteristic vector of inventive feature vector sum contrast, the extraction of the method correspondence Reeb signal characteristic quantity of the result display present invention obtains good effect with evaluation, screening for stress wave signal characteristic quantity provides new solution, for the later stage, the Non-Destructive Testing of one-dimensional component improves detection resolution, and noiseproof feature is excellent, do not occupy too much time resource, have good actual application value.

Claims (5)

1. the screening technique of one-dimensional component stress ripple signal characteristic, it is characterised in that: comprise the following steps,

Step one: primitive stress ripple signal is carried out pretreatment, decomposes with the method correspondence wave signal of WAVELET PACKET DECOMPOSITION, obtains m wavelet;

Step 2: using the quantitative information entropy characteristic quantity as one-dimensional component stress ripple signal;

Step 3: the time window of structure fixed width W, moves extraction signal characteristic value by stepping length B on each wavelet signal, obtains the stress wave quantitative information entropy matrix of multidimensional; The value of stepping length B meets W/8 < B < W/4;

Step 4: the stress wave quantitative information entropy matrix of multidimensional is carried out dimension-reduction treatment, seeks the average of the eigenvalue of all wavelet correspondence same positions, obtain one-dimensional quantitative information entropy mean vector;

Step 5: the multiparameter association analysis method in application gray system theory, the degree of association of state model to be checked Yu standard normal condition pattern by calculating multiple failure characteristics amount, the size of the contrast degree of association judges the performance state of system, completes the screening of stress wave characteristic quantity.

2. the screening technique of one-dimensional component stress ripple signal characteristic according to claim 1, it is characterised in that: described quantitative information entropy is:

$DH\left(X\right)=\underset{i=0}{\overset{D-1}{\mathrm{\&Sigma;}}}{\mathrm{Num}}_i{p}_i{\log }_2\frac{1}{p_i}$

Wherein, D is the number that amplitude is interval, Num_iRepresent sampled point number in i-th amplitude interval, p_iFor the statistical probability of sampling number in i-th amplitude interval.

3. the screening technique of one-dimensional component stress ripple signal characteristic according to claim 1, it is characterised in that: described stress wave quantitative information entropy matrix is:

$H=\left[\begin{array}{cccc}{\mathrm{DH}}_{11}& {\mathrm{DH}}_{12}& \mathrm{...}& {\mathrm{DH}}_{1n}\\ {\mathrm{DH}}_{21}& {\mathrm{DH}}_{22}& \mathrm{...}& {\mathrm{DH}}_{2n}\\ \mathrm{...}& \mathrm{...}& \mathrm{...}& \mathrm{...}\\ {\mathrm{DH}}_{m1}& {\mathrm{DH}}_{m2}& \mathrm{...}& {\mathrm{DH}}_{mn}\end{array}\right]$

Where each row data represent the quantitative information entropy vector of a wavelet, and m wavelet constitutes m × n characteristic quantity matrix;

Each wavelet entropy vector is:

H_m={ DH_m1, DH_m2, DH_m3..., DH_mn(m=1,2 ... 2^k)

Wherein, stress wave, after wavelet packet k layer decomposes, obtains m=2^kIndividual signal wavelet, time window number of steps is n.

4. the screening technique of one-dimensional component stress ripple signal characteristic according to claim 1, it is characterised in that: described quantitative information entropy mean vector is:

$\stackrel{\&OverBar;}{H}=\left\{{\stackrel{\&OverBar;}{DH}}_1,{\stackrel{\&OverBar;}{DH}}_2,{\stackrel{\&OverBar;}{DH}}_3,\mathrm{...},{\stackrel{\&OverBar;}{DH}}_n\right\}$

${\stackrel{\&OverBar;}{DH}}_i=\frac{1}{m}\left({\mathrm{DH}}_{1i}+{\mathrm{DH}}_{2i}+\mathrm{...}+{\mathrm{DH}}_{mi}\right),\left(i=1,2,\mathrm{...},n\right).$

5. the screening technique of one-dimensional component stress ripple signal characteristic according to claim 1, it is characterised in that: in described step 5, state Y to be checked_iState model vector be:

y⁽ⁱ⁾={ y₁ ⁽ⁱ⁾, y₂ ⁽ⁱ⁾, y₃ ⁽ⁱ⁾...y_n ⁽ⁱ⁾}^T

Wherein suspect system has the several characteristic parameter of n;

Standard normal condition Y₀State model vector be:

y⁽⁰⁾={ y₁ ⁽⁰ⁱ⁾, y₂ ⁽⁰ⁱ⁾, y₃ ⁽⁰ⁱ⁾...y_n ⁽⁰⁾}^T

State Y to be checked_jFor standard normal condition Y₀Thresholding coefficient of association on jth characteristic parameter is:

${r_j}^{\left(i\right)}=\frac{{\mathrm{k\&sigma;}}_j^0}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

Then state Y to be checked_jFor standard normal condition Y₀Threshold service be:

$r^{\left(i\right)}=\frac{1}{{\mathrm{\&Sigma;}}_{j=1}^n{\mathrm{\&alpha;}}_j}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_j\frac{{\mathrm{k\&sigma;}}_j^0}{\left|{y_j}^{\left(i\right)}-{\mathrm{\&mu;}}_j^0\right|+{\mathrm{k\&sigma;}}_j^0}$

Work as r_jWhen >=0.5, state to be checked belongs to standard normal condition; Work as r_jWhen≤0.5, state to be checked is not belonging to standard normal condition.