CN102928514A - Frequency characteristic-based nondestructive detection method of stress waves of wood - Google Patents

Abstract

The invention discloses a frequency characteristic-based nondestructive detection method of stress waves of wood. The frequency characteristic-based nondestructive detection method comprises the following steps of: mounting piezoelectric type acceleration sensors around raw wood uniformly, knocking the No. 0 sensor by using a pulse hammer, and acquiring and storing output signals of the sensors by using a data acquisition card; carrying out K-point fast Fourier transform on the acquired signals and then solving an actual measurement frequency response function; establishing an observation matrix according to the actual measurement frequency response function and a frequency response function of healthy wood; carrying out blind source separation on the basis of second order statistic and establishing a frequency response function between a defect point and an observation point; and finally, judging whether the interior of the wood has defects as well as defect quantity and defect sizes according to a clustering result. The nondestructive detection method of the stress waves based on the frequency response function of the wood is not subject to interference from reflected wave and refracted wave signals and is more accurate in detection result, capable of automatically detecting information, such as whether the interior of the wood has defects as well as the defect sizes, and simple and convenient in detection process and strong in practicability.

Description

A kind of timber stress wave lossless detection method based on frequecy characteristic

Technical field

The invention belongs to lumber quality detection technique field, relate in particular to a kind of timber stress wave lossless detection method based on frequecy characteristic.

Background technology

The ultimate principle of timber stress wave Dynamic Non-Destruction Measurement as shown in Figure 1, utilize pulse hammer to knock timber, make its inner stress wave propagation that produces, by being installed in log sensor on every side, can collect the stress wave signal on each observation station, corresponding Reeb signal carries out time domain, frequency-domain analysis, determines the character (such as elastic modulus and defective etc.) of wood materials.

Have based on the timber Dynamic Non-Destruction Measurement of stress wave and to realize the advantages such as simple, easy to use, be widely applied, but be generally used for reflecting that the feature of Wood Inner Defect Testing is travel-time or the velocity of propagation of stress wave in timber, this method can not be judged position rotten in the timber, and the measurement result dispersiveness is larger, and testing result is difficult to reach pin-point accuracy.

Frequency response function is the ratio of the Fourier transform of system input signal and output signal, and it has comprised all information of timber internal physical parameter.Just be based on this character, the numerous researchers in China and foreign countries have proposed to utilize the Dynamic Non-Destruction Measurement of timber frequency-response characteristic.Stress wave Dynamic Non-Destruction Measurement based on frequency domain character has appearred subsequently, the people such as Bozhang have studied the timber frequency response characteristic, the decay of stress wave signal radio-frequency component is less in the healthy wood, can use this method and effectively identify rotten in early days in the timber, the people such as Yu Guanxia propose the spectral line area of stress wave signal frequency spectrum and the composition of timber has corresponding relation, can utilize the frequency spectrum size to represent what of corresponding timber composition, and test by the log sample to the rotten situation of known internal, confirmed that the method is feasible, and had the advantages such as test result is very directly perceived.

But these methods are not all considered the phenomenons such as the reflection, refraction, transmission of stress wave signal in the timber, in actual testing process, mix the signals such as a large amount of reflection waves, refraction wave in the stress wave signal that sensor observes, brought error to testing result.

Be illustrated in figure 2 as the xsect of a log, the A point is the pulse hammer beating point, and the B point is the installation site of observation sensor, because the existence of rotten some F, stress wave is except (supposing that this part stress wave signal is s along straight line AB to B the some propagation ₁(t)), the B point also receives reflected signal s ₂(t).

Definition according to frequency response function:

$H_{\mathrm{AB}}\left(\mathrm{j\ω}\right)=\frac{X_B\left(\mathrm{j\ω}\right)}{X_A\left(\mathrm{j\ω}\right)}=\frac{S_1\left(\mathrm{j\ω}\right)+S_2\left(\mathrm{j\ω}\right)}{X_A\left(\mathrm{j\ω}\right)}=\frac{S_1\left(\mathrm{j\ω}\right)}{X_A\left(\mathrm{j\ω}\right)}+\frac{S_2\left(\mathrm{j\ω}\right)}{X_A\left(\mathrm{j\ω}\right)}=H_1\left(\mathrm{j\ω}\right)+H_2\left(\mathrm{j\ω}\right)---\left(1\right)$

In the formula: X _A(j ω), X _B(j ω) is respectively the Fourier transform of A point, B point observation signal;

S ₁(j ω), S ₂(j ω) is respectively signal s ₁(t), s ₂(t) Fourier transform.

That obviously, really reflect timberphysics characteristic between AB should be H ₁(j ω) is if simply with the ratio H of observation signal Fourier transform _AB(j ω) characterizes the physical characteristics of timber between AB, will inevitably cause larger error.

Summary of the invention

The invention provides a kind of timber stress wave lossless detection method based on frequecy characteristic, be intended to solve the lossless detection method that utilizes the timber frequency-response characteristic that prior art provides, all do not consider reflection, refraction, the transmission phenomenon of stress wave signal in the timber, a large amount of reflection waves, refraction wave signal have been mixed in the stress wave signal that sensor observes, the problem that the error of testing result is larger.

The object of the present invention is to provide a kind of timber stress wave lossless detection method based on frequecy characteristic, the method may further comprise the steps:

Step 1, I piezoelectric acceleration transducer that is connected to data collecting card by signal cable is installed around log equably, knock with pulse hammer and wherein to be numbered 0 sensor, then finish the collection of piezoelectric acceleration transducer output signal by data collecting card, save as x ₀(n), x ₁(n) ..., x _i(n) ..., x _I-1(n);

Step 2 is done the conversion of K point fast Fourier to the signal that collects, and obtains X ₀(k), X ₁(k) ..., X _i(k) ..., X _I-1(k), k=0,1, Λ K-1, then obtain the 0th sensor to the actual frequency response function between other each observation sensor:

$H_i\left(k\right)=\frac{X_i\left(k\right)}{X_0\left(k\right)},i=1,\mathrm{\ΛI}-1;---\left(2\right)$

Step 3, under definition the same terms in the clean timber the 0th sensor to the frequency response function between other each observation sensor be H _{0, i}(k), then with H _i(k) deduct H _{0, i}(k), obtain:

Y _i(k)＝H _i(k)-H _0，i(k)，i＝1，ΛI-1； (3)

Step 4: with Y=[Y ' ₁(k), Y ' ₂(k), Λ, Y ' _I-1(k)] ' and as observation matrix, it is done to separate based on the blind source of second-order statistic, estimate hybrid matrix W=[G ' _0,1(k), G ' _0,2(k), Λ G ' _{0, J}(k)] ' and source vector G _{J, i}(k), i=1, Λ I-1, j=1, Λ J, wherein G _{0, j}(k) be that the 0th sensor is to the frequency response function between inner j the defect point of timber, G _{J, i}(k) be j the frequency response function between defect point to the i observation station, detailed step is:

(1) correlation matrix of estimation Y

${\hat{R}}_Y\left(0\right)=\frac{1}{K}\underset{k=0}{\overset{K-1}{\mathrm{\Σ}}}Y\left(k\right)Y^{\′}\left(k\right)---\left(4\right)$

(2) right

Make Eigenvalues Decomposition (EVD)

${\hat{R}}_Y\left(0\right)=U_Y{\mathrm{\Σ}}_Y{V}_Y^{\′}=V_Y{\mathrm{\Λ}}_Y{V}_Y^{\′}=V_S{\mathrm{\Λ}}_S{V}_S^{\′}+V_N{\mathrm{\Λ}}_N{V}_N^{\′}---\left(5\right)$

Wherein (I-1) * J ties up matrix V _M=[v ₁, v ₂, Λ v _I-1] be and J principal character value Λ that arranges by descending order _S=diag{ λ ₁〉=λ ₂Λ 〉=λ _JCorresponding eigenvector; (I-1) * (I-1-J) tie up matrix V _NComprise (I-1-J) individual noise characteristic Λ _N={ λ _J+1〉=Λ 〉=λ _I-1Corresponding noise characteristic vector, and λ _J＞λ _J+1

(3) white noise variance

Be estimated the average of (I-1-J) individual inessential eigenwert;

(4) carry out sane prewhitening conversion:

$\stackrel{\‾}{Y}\left(k\right)={\widehat{\mathrm{\Λ}}}_S^{-1/2}{V}_S^{\′}Y\left(k\right)=\mathrm{QY}\left(k\right)---\left(6\right)$

Wherein ${\widehat{\mathrm{\Λ}}}_S=\mathrm{diag}\{({\mathrm{\λ}}_1-{\widehat{\mathrm{\σ}}}_K^2),({\mathrm{\λ}}_2-{\widehat{\mathrm{\σ}}}_K^2),\mathrm{\Λ}({\mathrm{\λ}}_1-{\widehat{\mathrm{\σ}}}_K^2)\};$

(5) for given p ≠ 0, estimated vector

Covariance matrix, and carry out the svd of covariance matrix:

${\hat{R}}_{\stackrel{\‾}{Y}}\left(p\right)=\frac{1}{K}\underset{k=0}{\overset{K-1}{\mathrm{\Σ}}}\stackrel{\‾}{Y}\left(k\right){\stackrel{\‾}{Y}}^{\′}(k-p)=U_{\stackrel{\‾}{Y}}{\mathrm{\Σ}}_{\stackrel{\‾}{Y}}{V}_{\stackrel{\‾}{Y}}^{\′}---\left(7\right)$

(6) for given p, check diagonal matrix Whether all singular values are different, if identical, for different time lag p repeating steps (5), if singular value is different, and away from each other, then estimate source vector:

${\hat{G}}_j\left(k\right)=U_{\stackrel{\‾}{Y}}^{\′}{\widehat{\mathrm{\Λ}}}_S^{-1/2}{V}_S^{\′}Y\left(k\right),j=\mathrm{1,2},\mathrm{\Λ},J;---\left(8\right)$

Step 5, individual to the J that estimates

Carry out the k mean cluster analysis, specific practice is:

(1) for being the data set of J to sizing, makes L=1, choose 3 initial cluster center Z _i(L), i=1,2,3;

(2) calculate each data object Distance with k cluster centre

J=1,2, Λ, J, i=1,2,3, if satisfy $D({\hat{G}}_j\left(k\right),Z_i\left(L\right))=\min \{D({\hat{G}}_j\left(k\right),Z_i\left(L\right)),j=\mathrm{1,2},\mathrm{\Λ},J\},$ Then ${\hat{G}}_j\left(k\right)\∈C_k;$

(3) recomputate k new cluster centre

I=1,2,3, and the value of square error criterion function: $E(L+1)=\underset{m=1}{\overset{k}{\mathrm{\Σ}}}\underset{p\∈C_m}{\mathrm{\Σ}}{\left|\right|p-Z_i\left(m\right)\left|\right|}^2;$

(4) judge: if || E (L+1)-E (L) ||＜ε, then algorithm finishes, otherwise L=L+1 returns and continues to carry out (2).

Step 6, differentiating timber inside according to the result of cluster has zero defect, and categorical measure determines what of defect area, the size of this defect area of quantitaes of data object in every class.

Further, in step 3, H _{0, i}(k) reflection be stress wave signal along the propagation characteristic of sphere direction, irrelevant with defect information, can be drawn by the frequency response function in the healthy timber.

Further, in step 5, G _{J, i}(k) be j the frequency response function between defect point to the i observation station, comprised a large amount of information about defective.

Further, for healthy timber, can be considered homogeneous timber, each G _{J, i}(k) numerical values recited equates; For defective timber, the G on the defect point _{J, i}(k) numerical value is different from other zones, and close defect point numerical value is roughly the same.

Timber stress wave lossless detection method based on frequecy characteristic provided by the invention, by the blind source separation method based on second-order statistic, obtain the frequency response function between defect point and the observation station, and take the frequency response function between defect point and the observation station as foundation, defects in timber are detected; This is according to the method for the stress wave Non-Destructive Testing of timber frequency response function, be not subjected to the interference of reflection wave, refraction wave signal, testing result is more accurate, and can automatically detect the information whether timber exists defective, defect size, testing process is easy, practical, have stronger propagation and employment and be worth.

Description of drawings

Fig. 1 is that the stress wave that utilizes that prior art provides is determined the basic principle schematic of wood materials character;

Fig. 2 is the basic principle schematic of the lossless detection method that utilizes the timber frequency-response characteristic that provides of prior art;

Fig. 3 is the principle schematic based on the timber stress wave lossless detection method of frequecy characteristic that the embodiment of the invention provides;

Fig. 4 is the wiring diagram of the embodiment of the invention.

Embodiment

Main thought of the present invention is as follows:

Referring to Fig. 3, be a defective log sectional view, dash area represents the foxiness cavity among the figure.Knock when being numbered 0 sensor when pulse hammer, stress wave can be at the timber internal communication.The stress wave signal x that each observation sensor receives _i(n) in, except directly propagating the stress wave signal of coming from beating point, also include the stress wave signal that reflects back from rotten cavity.J little defect point f regarded as in large rotten cavity ₁, f ₂..., f _JSet, have the characteristics of linearity according to Fourier transform, the signal spectrum that each sensor receives is:

$X_i\left(k\right)=X_0\left(k\right)H_{0,i}\left(k\right)+X_0\left(k\right)\underset{j=1}{\overset{J}{\mathrm{\Σ}}}{G}_{0,j}\left(k\right)G_{j,i}\left(k\right)---\left(9\right)$

In the formula, X ₀(k) be the 0th the stress wave signal frequency spectrum that sensor observes, H _{0, i}(k) frequency response function when being healthy timber between the 0th sensor of supposition and i the sensor, G _{0, j}(k) be the 0th frequency response function between sensor to the j defective unit, G _{J, i}(k) be j the frequency response function between defective unit to the i sensor.

First X in the right in the formula (9) ₀(k) H _{0, i}(k) for from healthy timber, being transmitted to the stress wave signal frequency spectrum of i sensor along sphere, second on equation the right

Reflex to the stress wave signal frequency spectrum of i sensor for each defect point.

With formula (9) the right and left simultaneously divided by X ₀(k), again with H _{0, i}(k) move to the equation another side, obtain:

$Y_i\left(k\right)=\underset{j=1}{\overset{J}{\mathrm{\Σ}}}{G}_{0,j}\left(k\right)G_{j,i}\left(k\right)=\frac{X_i\left(k\right)}{X_0\left(k\right)}-H_{0,i}\left(k\right)=H_i\left(k\right)-H_{0,i}\left(k\right)---\left(10\right)$

As previously mentioned, G _{J, i}(k) be frequency response function between i sensor of a j defective unit to the, comprised a large amount of information about defective.For healthy timber, can be considered as homogeneous timber, each G _{J, i}(k) numerical values recited equates; For defective timber, the G on the defect point _{J, i}(k) numerical value is different from other zones, and close defect point numerical value is roughly the same.Therefore can pass through G _{J, i}(k) in the identification timber zero defect being arranged, is that defective appears in single zone, or the size of defect area defective appears, in a plurality of zones.

Therefore the problem of Wood Defects Testing is converted to and asks for G _{J, i}(k), in formula (10), G _{0, j}(k) and G _{J, i}(k) be unknown message, unique utilizable information is exactly Y _i(k).The present invention is to Y _i(k) make blind source separation method based on second-order statistic, and then estimate G _{J, i}(k), obtain timber inherent vice information.

In order to make purpose of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, the present invention is further described in detail.Should be appreciated that specific embodiment described herein only in order to explaining the present invention, and be not used in and limit invention.

A kind of embodiment of the present invention is: a kind of timber stress wave lossless detection method based on frequecy characteristic, and its specific practice is:

Step 1 is installed 6 piezoelectric acceleration transducers equably around log, acceleration sensor outputs signals is through charge amplifier access data capture card, and wiring diagram as shown in Figure 4.Referring to Fig. 3, knock the 0th sensor with pulse hammer, 6 sensors convert the stress wave signal that receives to electric signal input data collecting card.Under the control of upper computer software, finish the work such as conversion of data, and tracer signal x ₀(n), x ₁(n), x ₂(n), x ₃(n), x ₄(n), x ₅(n).The selective sampling cycle is 1 μ s, and tracer signal length is 10ms.

Step 2 is to x ₀(n), x ₁(n), x ₂(n), x ₃(n), x ₄(n), x ₅(n) make fast fourier transform, conversion is counted and is chosen as 1024, obtains frequency spectrum X ₀(k), X ₁(k), X ₂(k), X ₃(k), X ₄(k), X ₅(k); Then ask the 0th sensor to the frequency response function between other 5 sensors according to formula (2):

$H_1\left(k\right)=\frac{X_1\left(k\right)}{X_0\left(k\right)},$ $H_2\left(k\right)=\frac{X_2\left(k\right)}{X_0\left(k\right)},$ $H_3\left(k\right)=\frac{X_3\left(k\right)}{X_0\left(k\right)},$ $H_4\left(k\right)=\frac{X_4\left(k\right)}{X_0\left(k\right)},$ $H_5\left(k\right)=\frac{X_5\left(k\right)}{X_0\left(k\right)}.$

Step 3 is according to obtaining the 0th sensor to the transport function of other 5 sensors in the zero defect situation, H with stress wave signal in the healthy timber of seeds along sphere direction propagation characteristic _0,1(k), H _0,2(k), H _0,3(k), H _0,4(k), H _0,5(k); Again the actual transfer function in the step 2 is subtracted each other with it, obtains:

Y ₁(k)＝H ₁(k)-H _0，1(k)

Y ₂(k)＝H ₂(k)-H _0，2(k)

Y ₃(k)＝H ₃(k)-H _0，3(k)

Y ₄(k)＝H ₄(k)-H _0，4(k)

Y ₅(k)＝H ₅(k)-H _0，5(k)

Step 4, the Y in the step 3 _i(k) also be that the 0th sensor is to frequency response function G between defect point _{0, j}(k) and defect point to the frequency response function G between each observation station _{J, i}(k) instantaneous mixing, that is: At G _{0, j}(k) and G _{J, i}(k) all in the unknown situation, can adopt blind source separation method based on second-order statistic to carry out blind source and estimate.With Y ₁(k), Y ₂(k), Y ₃(k), Y ₄(k), Y ₅(k) be used as observation vector, adopt the blind source separation method based on second-order statistic, estimate G _{0, j}(k) and G _{J, i}(k).Specific practice is:

(1) estimated matrix Y=[Y ' ₁(k), Y ' ₂(k), Y ' ₃(k), Y ' ₄(k), Y ' ₅(k)] ' correlation matrix

For:

${\hat{R}}_Y\left(0\right)=\frac{1}{1024}\underset{k=0}{\overset{1023}{\mathrm{\Σ}}}Y\left(k\right)Y^{\′}\left(k\right)$

(2) right

Make Eigenvalues Decomposition (EVD):

${\hat{R}}_Y\left(0\right)=U_Y{\mathrm{\Σ}}_Y{V}_Y^{\′}=V_Y{\mathrm{\Λ}}_Y{V}_Y^{\′}=V_S{\mathrm{\Λ}}_S{V}_S^{\′}+V_N{\mathrm{\Λ}}_N{V}_N^{\′}$

Wherein 5 * J ties up matrix V _M=[v ₁, v ₂, L v ₅] be with J by descending order arrange principal character value Λ _S=diag{ λ ₁〉=λ ₂L 〉=λ _JCorresponding eigenvector; 5 * (5-J) dimension matrix V _NComprise (5-J) individual noise characteristic Λ _N={ λ _J+1〉=L 〉=λ ₅Corresponding noise characteristic vector, and λ _J＞λ _J+1

(3) white noise variance Be estimated the average of (5-J) individual inessential eigenwert;

(4) carry out sane prewhitening conversion:

$\stackrel{\‾}{Y}\left(k\right)={\widehat{\mathrm{\Λ}}}_S^{-1/2}{V}_S^{\′}Y\left(k\right)=\mathrm{QY}\left(k\right)$

Wherein ${\widehat{\mathrm{\Λ}}}_S=\mathrm{diag}\{({\mathrm{\λ}}_1-{\widehat{\mathrm{\σ}}}_{1024}^2),({\mathrm{\λ}}_2-{\widehat{\mathrm{\σ}}}_{1024}^2),L({\mathrm{\λ}}_5-{\widehat{\mathrm{\σ}}}_{1024}^2)\};$

(5) for given p ≠ 0, estimated vector

Covariance matrix, and carry out the svd of covariance matrix, ${\hat{R}}_{\stackrel{\‾}{Y}}\left(p\right)=\frac{1}{1024}\underset{k=0}{\overset{1023}{\mathrm{\Σ}}}\stackrel{\‾}{Y}\left(k\right){\stackrel{\‾}{Y}}^{\′}(k-p)=U_{\stackrel{\‾}{Y}}{\mathrm{\Σ}}_{\stackrel{\‾}{Y}}{V}_{\stackrel{\‾}{Y}}^{\′};$

(6) for given p, check diagonal matrix Whether all singular values are different, if identical, for different time lag p repeating steps (5), if singular value is different, and away from each other, then estimate source vector ${\hat{G}}_j\left(k\right)=U_{\stackrel{\‾}{Y}}^{\′}{\widehat{\mathrm{\Λ}}}_S^{-1/2}{V}_S^{\′}Y\left(k\right),$ j＝1，ΛJ。

Step 5, individual to the J that estimates

Carry out the k mean cluster analysis, specific practice is:

(1) for being the data set of J to sizing, makes L=1, choose 3 initial cluster center Z _i(L), i=1,2,3;

(2) calculate each data object

Distance with k cluster centre

J=1,2, Λ, J, i=1,2,3, if satisfy $D({\hat{G}}_j\left(k\right),Z_i\left(L\right))=\min \{D({\hat{G}}_j\left(k\right),Z_i\left(L\right)),j=\mathrm{1,2},\mathrm{\Λ},J\},$ Then ${\hat{G}}_j\left(k\right)\∈C_k;$

(3) recomputate k new cluster centre I=1,2,3, and the value of square error criterion function: $E(L+1)=\underset{m=1}{\overset{k}{\mathrm{\Σ}}}\underset{p\∈C_m}{\mathrm{\Σ}}{\left|\right|p-Z_i\left(m\right)\left|\right|}^2;$

(4) judge: if || E (L+1)-E (L) ||＜ε, then algorithm finishes, otherwise L=L+1 returns and continues to carry out (2).

Step 6, differentiating timber inside according to the result of cluster has zero defect, and categorical measure determines what of defect area, the size of this defect area of quantitaes of data object in every class.

The timber stress wave lossless detection method based on frequecy characteristic that the embodiment of the invention provides, by the blind source separation method based on second-order statistic, obtain the frequency response function between defect point and the observation station, and take the frequency response function between defect point and the observation station as foundation, defects in timber are detected; This is according to the method for the stress wave Non-Destructive Testing of timber frequency response function, do not eliminated the interference of reflection wave, refraction wave signal, testing result is more accurate, and can automatically detect the information whether timber exists defective, defect size, testing process is easy, practical, have stronger propagation and employment and be worth.

The above only is preferred embodiment of the present invention, not in order to limiting the present invention, all any modifications of doing within the spirit and principles in the present invention, is equal to and replaces and improvement etc., all should be included within protection scope of the present invention.

Claims (6)

1. timber stress wave lossless detection method based on frequecy characteristic is characterized in that the method may further comprise the steps:

Step 1, I piezoelectric acceleration transducer that is connected to data collecting card by signal cable is installed around log equably, knock with pulse hammer and to be numbered 0 sensor, then finish the collection of piezoelectric acceleration transducer output signal by data collecting card, save as x ₀(n), x ₁(n) ..., x _i(n) ..., x _I-1(n);

Step 2 is done the conversion of K point fast Fourier to the signal that collects, and obtains X ₀(k), X ₁(k) ..., X _i(k) .X _I-1(k), k=0,1, Λ K-1, then obtain the 0th sensor to the actual frequency response function between other each observation sensor:

$H_i\left(k\right)=\frac{X_i\left(k\right)}{X_0\left(k\right)},i=1,\mathrm{\ΛI}-1;$

Step 3, under definition the same terms in the healthy timber the 0th sensor to the frequency response function between other each observation sensor be H _{0, i}(k), then with H _i(k) deduct H _{0, i}(k), obtain:

Y _i(k)＝H _i(k)-H _0，i(k)，i＝1，ΛI-1；

Step 4: make up Y=[Y ' ₁(k), Y ' ₂(k), Λ, Y ' _I-1(k)] ' and as observation matrix, it is done to separate based on the blind source of second-order statistic, estimate hybrid matrix W=[G ' _0,1(k), G ' _0,2(k), Λ G ' _{0, J}(k)] ' and source vector G _{J, i}(k), i=1, Λ I-1, j=1,2 Λ, J, wherein G _{0, j}(k) be that the 0th sensor is to the frequency response function between inner j the defect point of timber, G _{J, i}(k) be j the frequency response function between defect point to the i observation station;

Step 5, individual to the J that estimates

Carry out the k mean cluster analysis;

Step 6, differentiating timber inside according to the result of cluster has zero defect, and categorical measure determines what of defect area, the size of this defect area of quantitaes of data object in every class.

2. the method for claim 1 is characterized in that, in step 3, and H _{0, i}(k) reflection be stress wave signal along the propagation characteristic of sphere direction, irrelevant with defect information, can be drawn by the frequency response function in the healthy timber.

3. the method for claim 1 is characterized in that, in step 5, and G _{J, i}(k) be j the frequency response function between defect point to the i observation station, comprised a large amount of information about defective.

4. method as claimed in claim 3 is characterized in that, for healthy timber, can be considered homogeneous timber, each G _{J, i}(k) numerical values recited equates; For defective timber, the G on the defect point _{J, i}(k) numerical value is different from other zones, and close defect point numerical value is roughly the same.

5. the method for claim 1 is characterized in that, with Y=[Y ' ₁(k), Y ' ₂(k), Λ, Y _I-1(k)] ' and as observation matrix, it is done to separate based on the blind source of second-order statistic, estimate hybrid matrix

W=[G ' _0,1(k), G ' _0,2(k), Λ G ' _{0, J}(k)] ' and source vector G _{J, i}(k), i=1, Λ I-1, j=1,2, Λ, J, wherein G _{0, j}(k) be that the 0th sensor is to the frequency response function between inner j the defect point of timber, G _{J, i}(k) be j the frequency response function between defect point to the i observation station, detailed step is:

(1) correlation matrix of estimation Y

${\hat{R}}_Y\left(0\right)=\frac{1}{K}\underset{k=0}{\overset{K-1}{\mathrm{\Σ}}}Y\left(k\right)Y^{\′}\left(k\right)$

(2) right

Make Eigenvalues Decomposition (EVD)

${\hat{R}}_Y\left(0\right)=U_Y{\mathrm{\Σ}}_Y{V}_Y^{\′}=V_Y{\mathrm{\Λ}}_Y{V}_Y^{\′}=V_S{\mathrm{\Λ}}_S{V}_S^{\′}+V_N{\mathrm{\Λ}}_N{V}_N^{\′}$

Wherein (I-1) * J ties up matrix V _M=[v ₁, v ₂, Λ v _I-1] be with J by descending order arrange principal character value Λ _S=diag{ λ ₁〉=λ ₂L 〉=λ _JCorresponding eigenvector; (I-1) * (I-1-J) tie up matrix V _NComprise (I-1-J) individual noise characteristic Λ _N={ λ _J+1〉=L 〉=λ _I-1Corresponding noise characteristic vector, and λ _J＞λ _J+1

(3) white noise variance

Be estimated the average of (I-1-J) individual inessential eigenwert;

(4) carry out sane prewhitening conversion:

$\stackrel{\‾}{Y}\left(k\right)={\widehat{\mathrm{\Λ}}}_S^{-1/2}{V}_S^{\′}Y\left(k\right)=\mathrm{QY}\left(k\right)$

Wherein ${\widehat{\mathrm{\Λ}}}_S=\mathrm{diag}\{({\mathrm{\λ}}_1-{\widehat{\mathrm{\σ}}}_K^2),({\mathrm{\λ}}_2-{\widehat{\mathrm{\σ}}}_k^2)\};$

(5) for given p ≠ 0, estimated vector

Covariance matrix, and carry out the svd of covariance matrix:

${\hat{R}}_{\stackrel{\‾}{Y}}\left(p\right)=\frac{1}{K}\underset{k=0}{\overset{K-1}{\mathrm{\Σ}}}\stackrel{\‾}{Y}\left(k\right){\stackrel{\‾}{Y}}^{\′}(k-p)=U_{\stackrel{\‾}{Y}}{\mathrm{\Σ}}_{\stackrel{\‾}{Y}}{V}_{\stackrel{\‾}{Y}}^{\′}$

(6) for given p, check diagonal matrix

Whether all singular values are different, if identical, for different time lag p repeating steps (5), if singular value is different, and away from each other, then estimate source vector:

${\hat{G}}_j\left(k\right)=U_{\stackrel{\‾}{Y}}^{\′}{\widehat{\mathrm{\Λ}}}_S^{-1/2}{V}_S^{\′}Y\left(k\right),j=\mathrm{1,2},\mathrm{\Λ},J.$

6. the method for claim 1 is characterized in that, and is individual to the J that estimates

Carry out the k mean cluster analysis, specific practice is:

(1) for being the data set of J to sizing, makes L=1, choose 3 initial cluster center Z _i(L), i=1,2,3;

(2) calculate each data object

Distance with k cluster centre

J=1,2, Λ, J, i=1,2,3, if satisfy $D({\hat{G}}_j\left(k\right),Z_i\left(L\right))=\min \{D({\hat{G}}_j\left(k\right),Z_i\left(L\right)),j=\mathrm{1,2},\mathrm{\Λ},J\},$ Then ${\hat{G}}_j\left(k\right)\∈C_k;$

(3) recomputate k new cluster centre

I=1,2,3, and the value of square error criterion function: $(L+1)=\underset{m=1}{\overset{k}{\mathrm{\Σ}}}\underset{p\∈C_m}{\mathrm{\Σ}}{\left|\right|p-Z_i\left(m\right)\left|\right|}^2;$

(4) judge: if || E (L+1)-E (L) ||＜ε, then algorithm finishes, otherwise L=L+1 returns and continues to carry out (2).

Images