CN103776901B - Based on the sticky cartridge clip Rotating fields ageing state recognition methods of vibratory response information - Google Patents

Abstract

The present invention discloses a kind of sticky cartridge clip Rotating fields ageing state recognition methods based on vibratory response information, first, by encouraging experiment to obtain the vibratory response information of structure at random; Then, based on the little ripple of the s-generation of signal waveform latent structure self-adaptation, and then self-adaptive redundant second generation wavelet packet transform method is proposed, and the pre-treatment for vibration response signal, then extract the frequency domain statistical nature composition primitive character collection of each band signal Hilbert envelope spectrum respectively; Then, minority sensitive features is selected based on distance assessment technology from a large amount of primitive character is concentrated; Finally, based on m ultiwavelet theory building m ultiwavelet kernel function, and combine with SVMs and propose m ultiwavelet support vector machine method, using the input as m ultiwavelet SVMs of the sensitive features selected, it is achieved the automatic identification of sticky cartridge clip Rotating fields ageing state.

Description

Based on the sticky cartridge clip Rotating fields ageing state recognition methods of vibratory response information

Technical field

The present invention relates to physical construction state of health recognition methods, in particular to the recognition methods of a kind of sticky cartridge clip Rotating fields ageing state.

Background technology

Sticky cartridge clip Rotating fields has excellent sealing, vibration damping, falls performances such as making an uproar, extensive in application in machine equipments. In life-time service process, due to the impact by the environmental factorss such as temperature, humidity, vibration and change thereof, viscoelastic material is by catabiosis such as unavoidably relaxing, harden, send out crisp so that the kinetic characteristic of sticky cartridge clip Rotating fields changes, and then affects the use properties of structure. Therefore, identify sticky cartridge clip Rotating fields ageing state, to raising structural reliability, improve structure practicality there is important meaning. Meanwhile, also it is the urgent needs of Efficient Evaluation equipment overall efficiency, for extension device work-ing life, ensures that its safety in utilization has important engineering use value.

Structure ageing state recognition methods conventional at present mainly contains the method based on model and the method based on vibratory response. But, the moiety of sticky cartridge clip Rotating fields is complicated, and the dynamics of viscoelastic material is inconstant, and this both increases the difficulty of model solution, constrains the accuracy rate of model solution. The vibratory response information identification structural performance that structure produces under outside is encouraged is utilized, for identifying that sticky cartridge clip Rotating fields ageing state provides conveniently, effective approach based on the method for vibratory response. When in various degree aging occurs in sticky cartridge clip Rotating fields, the dynamics of structure, as rigidity, damping etc. can change, shows as the change of structural vibration response information under outside is encouraged. Therefore, the otherness of vibratory response information under the different ageing state of the sticky cartridge clip Rotating fields of analysis, extracts and selects effective characteristic index describing this otherness, automatically identifies that structure ageing state is a kind of effective technological approaches based on intelligent classification algorithm.

Summary of the invention

It is an object of the invention to provide a kind of vibratory response information utilizing outside excitation to produce to identify the method for sticky cartridge clip Rotating fields ageing state. The method precision height, cost are low, simple and reliable, are convenient to be applied in engineering reality.

For reaching above object, the present invention takes following technical scheme to be achieved:

A kind of sticky cartridge clip Rotating fields ageing state recognition methods based on vibratory response information, it is characterised in that, comprise following step:

(1) acquisition of vibratory response information

By shaking table and Controlling System thereof, sticky cartridge clip Rotating fields is applied random excitation, after vibration is stable, gather vibratory response information;

(2) based on the feature extraction of self-adaptive redundant second generation wavelet packet conversion

First, based on the little ripple of the s-generation of signal waveform latent structure self-adaptation, vibration response signal x is carried out 1 layer of little wave conversion of the s-generation, taking the Hilbert envelope spectrum entropy index of detail signal d as objective function, utilize the predictor P that genetic algorithm for solving makes it minimum_opt; Assume detail signal { d_iHilbert envelope spectrum be { f_i, calculate the distribution probability density p of its spectrum in whole spectrum_i, the objective function of predictor is defined as:

$E_f=-\stackrel{n}{\mathrm{\Σ}}{p}_i\ln p_i$

In order to make approximation signal s can more feature except detail signal in original No. i, the letter of accurate characterization, with reconstructed error J_UAs the criterion upgrading device, solve and make J_UMinimum renewal device U_opt, J_UIt is defined as:

$J_U=E\left\{{({\hat{s}}^{\left(0\right)}-s^{\left(0\right)})}^2\right\}+E\left\{{({\hat{d}}^{\left(0\right)}-d^{\left(0\right)})}^2\right\}$

Wherein,WithIt is respectively the reconstruction signal as detail signal d=0Even sequence samples and strange sequence samples; E{ } represent mathematical expection;

Secondly, according to the adaptive prediction device P of structure_optWith adaptive updates device U_opt, utilize the method for interpolation zero padding to obtain the redundant prediction device of l layerDevice is upgraded with redundancyProposing self-adaptive redundant second generation wavelet packet transform method, its level discharge rating process is shown below respectively:

$\left\{\begin{array}{c}{c}_{l+\mathrm{1,2}}=c_{l,1}-P_{\mathrm{opt}}^{(l+1)}\left(c_{l,1}\right)\\ c_{l+\mathrm{1,1}}=C_{l,1}+U_{\mathrm{opt}}^{(l+1)}\left(c_{l+\mathrm{1,2}}\right)\\ ...\\ c_{{l+\mathrm{1,2}}^{l+1}}=c_{{\mathrm{1,2}}^l}-P_{\mathrm{opt}}^{(l+1)}\left(c_{{l,2}^t}\right)\\ c_{l+1,2^{l+1}-1}=c_{{\mathrm{1,2}}^l}+U_{\mathrm{opt}}^{(l+1)}\left(c_{l+1,2^{l+1}}\right)\end{array}\right.$

$\left\{\begin{array}{c}{c}_{l,1}=\frac{1}{2}(c_{l+\mathrm{1,1}}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+\mathrm{1,2}}\right)+c_{l+\mathrm{1,2}}+P_{\mathrm{opt}}^{(l+1)}(c_{l+\mathrm{1,1}}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+\mathrm{1,2}}\right)))\\ ...\\ c_{{l,2}^l}=\frac{1}{2}(c_{l+1,2^{l+1}-1}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+1,2^{l+1}}\right)+c_{l+1,2^{l+1}}+P_{\mathrm{opt}}^{(l+1)}(c_{l+1,2^{l+1}-1}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+1,2^{l+1}}\right)))\end{array}\right.$

Finally, carry out feature extraction based on the self-adaptive redundant second generation wavelet packet conversion proposed, vibration response signal is carried out at least three layers of WAVELET PACKET DECOMPOSITION, extract the frequency domain statistical nature composition primitive character collection of each band signal Hilbert envelope spectrum respectively;

(3) select based on the sensitive features apart from assessment technology

By the evaluation factor of jth feature, it is defined as:

${\mathrm{\α}}_j=d_j^{\left(b\right)}/d_j^{\left(w\right)}$

Wherein,Represent the mean value of the class spacing of jth feature C class;Represent the mean value of the inter-object distance of jth feature C class, ��_jSize reflect the complexity that C class classified by jth feature, ��_jMore big expression jth feature is more responsive, more easily C class is classified, and selects ��_jCorresponding structural feature sensitive features collection;

(4) based on the Classification and Identification of m ultiwavelet SVMs

First, according to SVMs kernel function condition and multi-wavelet transformation theory building m ultiwavelet kernel function, adopt GHM m ultiwavelet, it is to construct m ultiwavelet kernel function be defined as follows:

$K_1(x,x^{\′})=\underset{i=1}{\overset{d}{\mathrm{\Π}}}\mathrm{COR}\left({\mathrm{\ψ}}_1\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right)+\underset{i=1}{\overset{d}{\mathrm{\Π}}}\mathrm{COR}\left({\mathrm{\ψ}}_2\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right)$

$K_2(x,x^{\′})=\underset{i=1}{\overset{d}{\mathrm{\Π}}}(\mathrm{COR}\left({\mathrm{\ψ}}_1\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right)\×\mathrm{COR}\left({\mathrm{\ψ}}_2\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right))$

Wherein: K represents kernel function; D represents sample dimension; ��_k(), k=1,2 represents two GHM m ultiwavelet functions; $\mathrm{COR}\left({\mathrm{\ψ}}_k\right)={\∫}_{-\∞}^{\∞}{\mathrm{\ψ}}_k\left(v\right){\mathrm{\ψ}}_k(t+v)\mathrm{d\; v}$ Represent the auto-correlation of m ultiwavelet function,

Secondly, the m ultiwavelet kernel function of structure is combined with SVMs, it is achieved the Classification and Identification of m ultiwavelet SVMs;

Finally, m ultiwavelet SVMs exports the categorised decision function of following form:

$f\left(x\right)=\mathrm{sign}[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_i{y}_i{K}_k(x,x_i)+b]$

Wherein: x represents unknown sample; N represents learning sample number; K=1 or 2, represents two m ultiwavelet kernel functions of structure respectively, can automatically identify the ageing state of the sticky cartridge clip Rotating fields corresponding to unknown sample according to this categorised decision function.

In aforesaid method, described evaluation factor ��_jSelection is greater than a threshold value, and this threshold value is set as the mean value of all evaluation factors.

Present invention achieves modern signal processing, sensitive features selection and m ultiwavelet support vector cassification identification mixing algorithmically. Its advantage is, vibratory response information identification is utilized to glue cartridge clip Rotating fields ageing state, have simple, reliable, easy, cost is low, the feature such as real-time, is applicable to the ageing state of the sticky cartridge clip Rotating fields of on-the-spot Real time identification, is conducive to improving reliability and the security of structure, for the identification of small sample structure ageing state provides new approaches and novel method, there is important engineering practical value.

Accompanying drawing explanation

Below in conjunction with the drawings and the specific embodiments, the present invention is described in further detail.

Fig. 1 is the sticky cartridge clip Rotating fields ageing state identification process figure of the present invention.

Fig. 2 is the decomposition and reconstruction schematic diagram of the self-adaptive redundant second generation wavelet packet conversion proposed in Fig. 1. In figure, x represents original signal;Represent reconstruction signal;WithRepresent that the l layer redundant prediction device utilizing interpolation zero padding method to obtain and redundancy upgrade device respectively.

Fig. 3 is the m ultiwavelet kernel function figure of structure. Wherein, Fig. 3 (a) represents the figure of the m ultiwavelet kernel function 1 of structure; Fig. 3 (b) represents the figure of the m ultiwavelet kernel function 2 of structure.

Fig. 4 is the structure diagram of the sticky cartridge clip Rotating fields designed and produced. As shown in FIG., this sticky cartridge clip Rotating fields forms primarily of metallic substance and viscoelastic material (rubber), metal level and rubber layer is compressed by joint bolt, and for measuring bolt snap-in force, side arranges a pressure transmitter under the nut.

Fig. 5 is the power spectrum density that sticky cartridge clip Rotating fields is applied random excitation by shaking table, and excitation frequency scope is 10-2000Hz. Within the scope of 50-1000Hz, excitation energy maintains 0.005g²/ Hz. In figure, X-coordinate represents frequency, and unit is hertz (Hz); Ordinate zou represents power spectrum density, and unit is g²/Hz��

Fig. 6 is the time-domain diagram of the different ageing state vibration response signal under random external encourages of sticky cartridge clip Rotating fields. Wherein, Fig. 6 (a)��(m) is respectively the time domain beamformer of vibration response signal when structure is in 1��13 kind of ageing state. In figure, X-coordinate represents the time, and unit is s; Ordinate zou represents amplitude, and unit is g.

Fig. 7 is the sensitive features selection figure based on distance assessment technology. In figure, X-coordinate represents sample, ordinate zou representation feature evaluation factor.

Fig. 8 is the sticky cartridge clip Rotating fields ageing state recognition result adopting the inventive method. Wherein, Fig. 8 (a) is SVMs (the being abbreviated as MSVM1) recognition result using m ultiwavelet kernel function 1;Fig. 8 (b) is for using SVMs (the being abbreviated as MSVM2) recognition result of m ultiwavelet kernel function 2. In figure, X-coordinate represents the test sample book of 13 kinds of ageing states, and ordinate zou represents the class label of each test sample book.

Its embodiment

With reference to shown in Fig. 1, it is sticky cartridge clip Rotating fields ageing state identification process figure, by encouraging experiment to obtain the vibratory response information of sticky cartridge clip Rotating fields at random; Based on the little ripple of the s-generation of signal waveform latent structure self-adaptation, and then self-adaptive redundant second generation wavelet packet transform method is proposed, and the pre-treatment for vibration response signal, then extract the frequency domain statistical nature composition primitive character collection of each band signal Hilbert envelope spectrum respectively; Minority sensitive features is selected from a large amount of primitive character is concentrated based on distance assessment technology; Based on m ultiwavelet theory building m ultiwavelet kernel function, and combine with SVMs and propose m ultiwavelet support vector machine method, using the input of sensitive features as m ultiwavelet SVMs, it is achieved the automatic identification of sticky cartridge clip Rotating fields ageing state.

The present invention utilizes the sticky cartridge clip Rotating fields ageing state of vibratory response information identification to implement by following concrete steps:

(1) acquisition of vibratory response information

Sticky cartridge clip Rotating fields is installed on a vibration table, by shaking table and Controlling System thereof, structure is applied random excitation, after vibration is stable, gather the vibratory response information of structure.

(2) based on the feature extraction of self-adaptive redundant second generation wavelet packet conversion

With reference to figure 2, first, based on the little ripple of the s-generation of signal waveform latent structure self-adaptation. Vibration response signal x is carried out 1 layer of little wave conversion of the s-generation, taking the Hilbert envelope spectrum entropy index of detail signal d as objective function, utilizes the predictor P that genetic algorithm for solving makes it minimum_opt. Assume detail signal { d_iHilbert envelope spectrum be_fI}, calculates the distribution probability density p of its spectrum in whole spectrum_i, the objective function of predictor is defined as:

$E_f=-\underset{i}{\overset{n}{\mathrm{\Σ}}}{p}_i\ln p_i$

In order to make approximation signal s can more feature except detail signal in accurate characterization original signal, with reconstructed error J_UAs the criterion upgrading device, solve and make J_UMinimum renewal device U_opt, J_UIt is defined as:

$J_U=E\left\{{({\hat{s}}^{\left(0\right)}-s^{\left(0\right)})}^2\right\}+E\left\{{({\hat{d}}^{\left(0\right)}-d^{\left(0\right)})}^2\right\}$

Wherein,WithIt is respectively the reconstruction signal as detail signal d=0Even sequence samples and strange sequence samples; E{ } represent mathematical expection.

Secondly, according to the adaptive prediction device P of structure_optWith adaptive updates device U_opt, utilize the method for interpolation zero padding to obtain the redundant prediction device of l layerDevice is upgraded with redundancyProposing self-adaptive redundant second generation wavelet packet transform method, its level discharge rating process is shown below respectively:

$\left\{\begin{array}{c}{c}_{l+\mathrm{1,2}}=c_{l,1}-P_{\mathrm{opt}}^{(l+1)}\left(c_{l,1}\right)\\ c_{l+\mathrm{1,1}}=C_{l,1}+U_{\mathrm{opt}}^{(l+1)}\left(c_{l+\mathrm{1,2}}\right)\\ ...\\ c_{{l+\mathrm{1,2}}^{l+1}}=c_{{\mathrm{1,2}}^l}-P_{\mathrm{opt}}^{(l+1)}\left(c_{{l,2}^t}\right)\\ c_{l+1,2^{l+1}-1}=c_{{\mathrm{1,2}}^l}+U_{\mathrm{opt}}^{(l+1)}\left(c_{l+1,2^{l+1}}\right)\end{array}\right.$

$\left\{\begin{array}{c}{c}_{l,1}=\frac{1}{2}(c_{l+\mathrm{1,1}}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+\mathrm{1,2}}\right)+c_{l+\mathrm{1,2}}+P_{\mathrm{opt}}^{(l+1)}(c_{l+\mathrm{1,1}}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+\mathrm{1,2}}\right)))\\ ...\\ c_{{l,2}^l}=\frac{1}{2}(c_{l+1,2^{l+1}-1}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+1,2^{l+1}}\right)+c_{l+1,2^{l+1}}+P_{\mathrm{opt}}^{(l+1)}(c_{l+1,2^{l+1}-1}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+1,2^{l+1}}\right)))\end{array}\right.$

Finally, feature extraction is carried out based on the self-adaptive redundant second generation wavelet packet conversion proposed. Vibration response signal is carried out 3 layers of WAVELET PACKET DECOMPOSITION, extracts the frequency domain statistical nature composition primitive character collection of each band signal Hilbert envelope spectrum respectively.

(3) select based on the sensitive features apart from assessment technology

The inter-object distance of the same class of a certain feature is more little, and the class spacing of inhomogeneity is more big, then this feature is more responsive. The evaluation factor of jth feature, is defined as:

${\mathrm{\α}}_j=d_j^{\left(b\right)}/d_j^{\left(w\right)}$

Wherein,Represent the mean value of the class spacing of jth feature C class;Represent the mean value of the inter-object distance of jth feature C class

��_jSize reflect the complexity that C class classified by jth feature. ��_jMore big expression jth feature is more responsive, more easily C class is classified.Select the �� being greater than certain threshold value_jCorresponding structural feature sensitive features collection, wherein this threshold value is set as the mean value of all evaluation factors.

(4) based on the Classification and Identification of m ultiwavelet SVMs

First, according to SVMs kernel function condition and multi-wavelet transformation theory building m ultiwavelet kernel function. In the present invention, m ultiwavelet elects the most frequently used GHM m ultiwavelet as, it is to construct m ultiwavelet kernel function be defined as follows:

$K_1(x,x^{\′})=\underset{i=1}{\overset{d}{\mathrm{\Π}}}\mathrm{COR}\left({\mathrm{\ψ}}_1\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right)+\underset{i=1}{\overset{d}{\mathrm{\Π}}}\mathrm{COR}\left({\mathrm{\ψ}}_2\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right)$ $K_2(x,x^{\′})=\underset{i=1}{\overset{d}{\mathrm{\Π}}}(\mathrm{COR}\left({\mathrm{\ψ}}_1\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right)\×\mathrm{COR}\left({\mathrm{\ψ}}_2\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right))$

Wherein: K represents kernel function; D represents sample dimension; ��_k(), k=1,2 represents two m ultiwavelet functions;Represent the auto-correlation of m ultiwavelet function. Two the m ultiwavelet kernel functions constructed are as shown in Figure 3.

Secondly, the m ultiwavelet kernel function of structure is combined with SVMs, it is achieved the Classification and Identification of m ultiwavelet SVMs. M ultiwavelet SVMs is substantially identical with the structure of traditional support vector machine, and difference is that the kernel function that they use is different.

Finally, m ultiwavelet SVMs exports the categorised decision function of following form:

$f\left(x\right)=\mathrm{sign}[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_i{y}_i{K}_k(x,x_i)+b]$

Wherein: x represents unknown sample; N represents learning sample number; K=1 or 2, represents two m ultiwavelet kernel functions of structure respectively.

Test according to m ultiwavelet SVMs exports the ageing state that can automatically identify the sticky cartridge clip Rotating fields corresponding to unknown sample.

Hereinafter provide an embody rule example procedure, the validity of simultaneous verification the present invention in engineer applied.

Designing and producing a kind of typical sticky cartridge clip Rotating fields, as shown in Figure 4, this structure forms primarily of metallic substance and viscoelastic material (rubber), and viscoelastic material is clipped in the middle of metallic substance, by joint bolt, structure is packed together to a structure entirety. First, viscoelastic material being put into a senile experiment case and carries out senile experiment, during experiment, temperature is 110 DEG C, and air circulation mode is for forcing air blast. Simulating 13 kinds of ageing states altogether, aging number of days corresponding to often kind of state is as shown in table 1. Therefrom it may be seen that from ageing state 1 to ageing state 13, the degree of aging of sticky cartridge clip Rotating fields is deepened successively.

Table 1 glues cartridge clip Rotating fields ageing state and mark

Secondly, being arranged on shaking table to encourage experiment at random by sticky cartridge clip Rotating fields, the preliminary tension of sticky cartridge clip Rotating fields is consistent: 7500N. The power spectrum density of exciting force is as shown in Figure 5. The vibratory response information of structure by acceleration transducer and data acquisition equipment collection and under storing 13 kinds of ageing states. Under sticky cartridge clip Rotating fields 13 kinds of ageing states, the time domain waveform of vibration response signal is as shown in Figure 6. As can be seen from Figure 6, under 13 kinds of states there is difference in the time domain waveform of structure dynamics response signal, but does not have regular difference, fails intuitively to reflect the ageing state of structure.

Utilize the method for the invention, first, based on the little ripple of the s-generation of signal waveform latent structure self-adaptation, vibration response signal adopts the conversion of self-adaptive redundant second generation wavelet packet carry out 3 layers of decomposition, extracts the frequency domain statistical nature composition primitive character collection of each band signal Hilbert envelope spectrum respectively; Then, select minority sensitive features based on distance assessment technology from a large amount of primitive character is concentrated, as shown in Figure 7; Finally, using the input as the m ultiwavelet SVMs proposed of the sensitive features selected, it is achieved the automatic identification of sticky cartridge clip Rotating fields ageing state. Wherein, the punishment factor of m ultiwavelet SVMs is set as 100, and kernel functional parameter adopts cross validation method search optimum value in [0.25,0.5,1��16].Recognition result, as shown in Fig. 8 and table 2, wherein uses the SVMs of m ultiwavelet kernel function 1 to be abbreviated as MSVM1, it may also be useful to the SVMs of m ultiwavelet kernel function 2 is abbreviated as MSVM2. As can be seen from Fig. 8 and table 2, MSVM1 and MSVM2 shows ageing state recognition result excellent equally, 390 test sample books of 13 kinds of states have only occurred 5 wrong part samples, recognition accuracy reaches 98.72%, thus demonstrates the present invention and identifying the validity in sticky cartridge clip Rotating fields ageing state.

Table 2 glues cartridge clip Rotating fields ageing state recognition accuracy (%)

Claims (2)

1. the sticky cartridge clip Rotating fields ageing state recognition methods based on vibratory response information, it is characterised in that, comprise following step:

(1) acquisition of vibratory response information

By shaking table and Controlling System thereof, sticky cartridge clip Rotating fields is applied random excitation, after vibration is stable, gather vibratory response information;

(2) based on the feature extraction of self-adaptive redundant second generation wavelet packet conversion

First, based on the little ripple of the s-generation of signal waveform latent structure self-adaptation, vibration response signal x is carried out 1 layer of little wave conversion of the s-generation, taking the Hilbert envelope spectrum entropy index of detail signal d as objective function, utilize the predictor P that genetic algorithm for solving makes it minimum_opt; Assume detail signal { d_iHilbert envelope spectrum be { f_i, calculate the distribution probability density p of its spectrum in whole spectrum_i, the objective function of predictor is defined as:

$E_f=-\stackrel{n}{\mathrm{\Σ}}{p}_i\ln p_i$

In order to make approximation signal s can more feature except detail signal in original No. i, the letter of accurate characterization, with reconstructed error J_UAs the criterion upgrading device, solve and make J_UMinimum renewal device U_opt, J_UIt is defined as:

$J_U=E\left\{{({\hat{s}}^{\left(0\right)}-s^{\left(0\right)})}^2\right\}+E\left\{{({\hat{d}}^{\left(0\right)}-d^{\left(0\right)})}^2\right\}$

Wherein,WithIt is respectively the reconstruction signal as detail signal d=0Even sequence samples and strange sequence samples; E{ } represent mathematical expection;

Secondly, according to the adaptive prediction device P of structure_optWith adaptive updates device U_opt, utilize the method for interpolation zero padding to obtain the redundant prediction device of l layerDevice is upgraded with redundancyProposing self-adaptive redundant second generation wavelet packet transform method, its level discharge rating process is shown below respectively:

$\left\{\begin{array}{c}{c}_{l+\mathrm{1,2}}=c_{l,1}-P_{\mathrm{opt}}^{(l+1)}\left(c_{l,1}\right)\\ c_{l+\mathrm{1,1}}=C_{l,1}+U_{\mathrm{opt}}^{(l+1)}\left(c_{l+\mathrm{1,2}}\right)\\ ...\\ c_{{l+\mathrm{1,2}}^{l+1}}=c_{{\mathrm{1,2}}^l}-P_{\mathrm{opt}}^{(l+1)}\left(c_{{l,2}^t}\right)\\ c_{l+1,2^{l+1}-1}=c_{{\mathrm{1,2}}^l}+U_{\mathrm{opt}}^{(l+1)}\left(c_{l+1,2^{l+1}}\right)\end{array}\right.$

$\left\{\begin{array}{c}{c}_{l,1}=\frac{1}{2}(c_{l+\mathrm{1,1}}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+\mathrm{1,2}}\right)+c_{l+\mathrm{1,2}}+P_{\mathrm{opt}}^{(l+1)}(c_{l+\mathrm{1,1}}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+\mathrm{1,2}}\right)))\\ ...\\ c_{{l,2}^l}=\frac{1}{2}\left(c_{l+1,2^{l+1}-1}\right)-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+1,2^{l+1}}\right)+c_{l+1,2^{l+1}}+P_{\mathrm{opt}}^{(l+1)}(c_{l+1,2^{l+1}-1}-U_{\mathrm{opt}}^{(l+1)}\left(c_{l+1,2^{l+1}}\right))\end{array}\right.$

Finally, carry out feature extraction based on the self-adaptive redundant second generation wavelet packet conversion proposed, vibration response signal is carried out at least three layers of WAVELET PACKET DECOMPOSITION, extract the frequency domain statistical nature composition primitive character collection of each band signal Hilbert envelope spectrum respectively;

(3) select based on the sensitive features apart from assessment technology

By the evaluation factor of jth feature, it is defined as:

${\mathrm{\α}}_j=d_j^{\left(b\right)}/d_j^{\left(w\right)}$

Wherein,Represent the mean value of the class spacing of jth feature C class;Represent the mean value of the inter-object distance of jth feature C class, ��_jSize reflect the complexity that C class classified by jth feature, ��_jMore big expression jth feature is more responsive, more easily C class is classified, and selects the �� being greater than certain threshold value_jCorresponding structural feature sensitive features collection;

(4) based on the Classification and Identification of m ultiwavelet SVMs

First, according to SVMs kernel function condition and multi-wavelet transformation theory building m ultiwavelet kernel function, adopt GHM m ultiwavelet, it is to construct m ultiwavelet kernel function be defined as follows:

$K_1(x,x^{\′})=\underset{i=1}{\overset{d}{\mathrm{\Π}}}\mathrm{COR}\left({\mathrm{\ψ}}_1\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right)+\underset{i=1}{\overset{d}{\mathrm{\Π}}}\mathrm{COR}\left({\mathrm{\ψ}}_2\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right)$

$K_2(x,x^{\′})=\underset{i=1}{\overset{d}{\mathrm{\Π}}}(\mathrm{COR}\left({\mathrm{\ψ}}_1\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right)\×\mathrm{COR}\left({\mathrm{\ψ}}_2\left(\frac{x_i-{x_i}^{\′}}{a}\right)\right))$

Wherein: K represents kernel function; D represents sample dimension; ��_k(), k=1,2 represents two GHM m ultiwavelet functions; $\mathrm{COR}\left({\mathrm{\ψ}}_k\right)={\∫}_{-\∞}^{\∞}{\mathrm{\ψ}}_k\left(v\right){\mathrm{\ψ}}_k(t+v)\mathrm{d\; v}$ Represent the auto-correlation of m ultiwavelet function,

Secondly, the m ultiwavelet kernel function of structure is combined with SVMs, it is achieved the Classification and Identification of m ultiwavelet SVMs;

Finally, m ultiwavelet SVMs exports the categorised decision function of following form:

$f\left(x\right)=\mathrm{sign}[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_i{y}_i{K}_k(x,x_i)+b]$

Wherein: x represents unknown sample;N represents learning sample number; K=1 or 2, represents two m ultiwavelet kernel functions of structure respectively, can automatically identify the ageing state of the sticky cartridge clip Rotating fields corresponding to unknown sample according to this categorised decision function.

2. as claimed in claim 1 based on the sticky cartridge clip Rotating fields ageing state recognition methods of vibratory response information, it is characterised in that, described evaluation factor ��_jSelection is greater than a threshold value, and this threshold value is set as the mean value of all evaluation factors.