CN102879473B - System for recognition of fatigue damage state of AZ31 magnesium alloy based on PCA (principal component analysis) and TDF (tactical data fusion) - Google Patents

Abstract

The invention discloses a system for identifying the fatigue damage state of an AZ31 magnesium alloy based on PCA and TDF. The system consists of multiple acoustic emission transducers (6), multi-channel preamplifiers (5), and an acoustic emission instrument ( 4) and an AZ31 magnesium alloy fatigue damage detection unit (1). The AZ31 magnesium alloy fatigue damage detection unit (1) includes a filtering module (11), a primary data fusion module (12), and a secondary data fusion module (13). The present invention combines principal component analysis with fatigue damage categories, and performs neural network training in the data space to obtain the damage degree marks of each transducer in the data space, and then uses the damage degree marks to analyze the information of each acoustic emission transducer. Carry out local diagnosis; then use the output result of neural network to construct the basic probability value of data fusion; finally use the combined relationship of data fusion to diagnose the fatigue damage state. Using this system, the damage state of AZ31 magnesium alloy during the fatigue process can be identified and diagnosed, and then the basis for its reliable operation can be provided.

Description

Identification system of fatigue damage state of AZ31 magnesium alloy based on PCA and TDF

technical field

The invention relates to a method for identifying the fatigue damage state of an in-service AZ31 magnesium alloy during service. More specifically, it refers to a kind of data collected by acoustic emission transducers. Firstly, principal component analysis (PCA) is used to construct different data spaces according to their damage categories, and neural network training is carried out in the data space to obtain the The damage degree flag module in the data space, and then apply the module to perform two-level data fusion (TDF) on the acoustic emission data collected in real time for the in-service AZ31 magnesium alloy, so as to identify the fatigue damage state of the in-service AZ31 magnesium alloy.

Background technique

In recent years, the annual growth rate of magnesium alloy products in the world is as high as 20%, and it has become a material that has attracted much attention. Magnesium alloys have shown great application prospects in industrial fields such as transportation, electronics and communication products, aerospace, chemical and machinery. AZ31 magnesium alloy refers to a magnesium alloy containing 3% Al and 1% (wt%) Zn, and is the most widely used wrought magnesium alloy commercially. AZ31 magnesium alloy can be extruded into rods, pipes, and profiles, rolled into thin plates, thick plates, and processed into forgings. It is mainly used in automobiles, aerospace components, weapons, etc.

The damage of AZ31 magnesium alloy after service is one of the main reasons for its failure. Therefore, it is necessary to identify its damage state, evaluate the fatigue damage state of AZ31 magnesium alloy in a timely and correct manner, and provide the basis for its safe operation and life prediction.

Acoustic Emission Technique (Acoustic Emission Technique) has been widely used in the damage detection of structures and components due to its advantages of dynamic and real-time detection. Practice has shown that in different stages of the fatigue process, the acoustic emission characteristics of the material will undergo a series of different changes, that is to say, the AZ31 magnesium alloy will have different acoustic emission signals at different stages of fatigue damage, and the transition of the damage state often occurs It causes changes in multiple parameters of acoustic emission, and at the same time, a parameter change can be caused by multiple damage states, so it is necessary to use data fusion technology of multiple acoustic emission transducers, that is, to make full use of multiple acoustic emissions in different time and space. Emission transducer data resources, using computer technology to analyze, synthesize, control and use the multi-acoustic emission transducer observation data obtained in time series under certain criteria, to obtain consistent interpretation and description of the measured object, Then realize the corresponding decision-making and estimation, so that the system can obtain more sufficient information than its components. Therefore, the present invention introduces the data fusion technology into the AZ31 magnesium alloy damage state identification system, and establishes a principal component analysis, data fusion and artificial neural network combined diagnosis system to identify and diagnose the AZ31 magnesium alloy damage state.

Principal component analysis is also called principal component analysis. Principal component analysis uses the idea of dimensionality reduction to convert multiple indicators into several comprehensive indicators under the premise of losing little information. Usually, the converted indicators are called principal components, where each principal component is a linear combination of the original variables, and each principal component is independent of each other, so that the principal components have some superior performance than the original variables. "Principal Component Analysis" refers to the content introduction on pages 152 to 154 of "Multivariate Statistical Analysis" published by Renmin University of China Press, the second edition in September 2008.

A neural network is a nonlinear system that simulates human thinking. The radial basis function (RBF) neural network is proposed based on the receptive field with local regulation and overlap in the human cerebral cortex, also known as the local receptive field neural network. It is a three-layer feed-forward network model including input layer, hidden layer and output layer. Due to the simple structure of the RBF network, the ability to approximate any continuous function with arbitrary precision, and the fast learning rate, it is more and more widely used in various fields.

Contents of the invention

In order to reduce the personnel injury, equipment loss and economic loss caused by the sudden fracture of AZ31 magnesium alloy during use, the present invention proposes a method that combines principal component analysis, neural network and two-level data fusion to identify AZ31 magnesium in service Alloy fatigue damage state. The fatigue state recognition system first uses principal component analysis to construct two data spaces for different damage data in the training samples, and uses the neural network method to perform neural network training on the information collected by the multi-channel acoustic emission transducer in the two data spaces. Obtain the damage degree flag module for judging the different fatigue damage states of AZ31 magnesium alloy; then carry out the basic probability distribution of the neural network output of the module under the two data spaces, and then carry out the first-level data fusion, and then combine the multiple transducers in the first-level The data fusion results are subjected to secondary data fusion to obtain the data fusion module, and then the data fusion module is embedded in the AZ31 magnesium alloy fatigue identification system. The data fusion module embedded with the present invention can identify different fatigue damage states of the in-service AZ31 magnesium alloy under working conditions, and give early warning to the identified results.

The present invention is an identification system for identifying the fatigue damage state of AZ31 magnesium alloy by using principal component analysis, neural network and data fusion technology. The system includes multiple acoustic emission transducers (6), multiple The preamplifier (5), an acoustic emission instrument (4), is characterized in that: it also includes an AZ31 magnesium alloy fatigue damage detection unit (1);

The AZ31 magnesium alloy fatigue damage detection unit (1) includes a filtering module (11), a first-level data fusion module (12), and a second-level data fusion module (13), wherein the filtering module (11) has a data filtering processing module (11A ) and waveform filtering processing module (11B), the primary data fusion module (12) has a first data space (12A), a second data space (12B), a first damage degree flag module (12C), and a second damage degree flag Module (12D), D-S Evidence Portfolio Module (12E);

The acoustic emission transducer (6) is used in conjunction with the preamplifier (5), that is, the output end of each acoustic emission transducer (6) is connected to the input end of a preamplifier (5), and each preamplifier The output end of the amplifier (5) is connected to the information input interface of the acoustic emission instrument (4), and the information input interface is used to receive multiple channels of burst type amplification information The AZ31 magnesium alloy fatigue damage detection unit (1) is embedded in the memory of the acoustic emission instrument (4);

Acoustic emission transducer (6), used to collect burst information of in-service AZ31 magnesium alloy within the collection time T _X segment

Preamplifier (5), used for receiving burst-type information After amplifying 40dB, it becomes burst type amplification information

The acoustic emission device (4) is used to amplify the received burst information After A/D conversion, it becomes digital burst information Output to AZ31 magnesium alloy fatigue damage detection unit (1);

The data filtering processing module (11A) in the filtering module (11) of the AZ31 magnesium alloy fatigue damage detection unit (1) processes the received digital burst information Perform parameter filtering to filter out electromagnetic noise and environmental noise, and then purify to obtain preliminary information on acoustic emission fatigue damage Then the waveform filter processing module (11B) analyzes the preliminary information of acoustic emission fatigue damage Perform waveform filtering to obtain acoustic emission fatigue damage information $f_{11B}^{T_x}=(e,A,C,R,\mathrm{D.});$

Fatigue damage information received by the transducer for AZ31 magnesium alloy Carry out the accumulation processing within the acquisition time T _X period, and then normalize to obtain the normalized cumulative fatigue damage information f _11B ′, and project f _11B ′ in the first data space (12A) and the second data space (12B) respectively Obtain respective scoring matrices f _12A =(t _a1 ,t _a2 ,t _a3 ,t _a4 ,t _a5 ) and f _12B =(t _b1 ,t _b2 ,t _b3 ,t _b4 ,t _b5 ), scoring matrix f _12A = (t _a1 ,t _a2 ,t _a3 ,t _a4 ,t _a5 ) are processed by the first damage degree flag module (12C) to obtain the neural network output f _12C in the first data space =[f _ID,a (A ₁ ) f _ID,a (A ₂ )], scoring matrix f _12B =(t _b1 ,t _b2 ,t _b3 ,t _b4 ,t _b5 ) processed by the second damage degree flag module (12D) in the second data space The neural network output f _12D =[f _ID,b (A ₁ )f _ID,b (A ₂ )]; After assigning the correlation coefficients of each node to f _12C and f _12D , carry out basic probability distribution, and then carry out DS evidence combination The module (12E) processes and obtains the data fusion result m _ID (B _j ) of a single transducer, and then performs secondary data fusion (13) on the results of the DS evidence combination module (12E) of all transducers to obtain the data fusion result m( C _j ), the result is analyzed by the damage level assessment unit (2) and then output fatigue damage identification information D to the alarm unit (3).

According to the acoustic emission information, the present invention adopts the neural network to locally identify the damage state in the data space constructed by the principal component analysis, and then uses the data fusion technology to perform two-stage fusion of the neural network diagnosis results to identify and diagnose the AZ31 magnesium alloy The final fatigue damage state, the advantages of this identification system are:

(A) Acoustic emission information (energy e _S , measurement amplitude A _S , ringing count C _S , rise time R _S , duration D _S ) to collect, and input the relevant information as the information of the recognition system of the acoustic emission neural network, so that the present invention can perform acoustic emission transducer information through the acoustic emission instrument during the acoustic emission detection process Acquisition, and then according to the changes of acoustic emission information parameters and waveforms, it is identified whether it is damage information or noise information.

(B) According to the type of fatigue damage mode, principal component analysis is used to construct different data spaces, and the damage state is locally identified in each data space through the neural network. The difference in signal identifies the damage pattern.

(C) It can synthesize, control and use the monitoring data of multiple and various types of acoustic emission transducers, make full use of the information of each acoustic emission transducer, increase the reliability and accuracy of the diagnostic results, and improve the diagnostic quality. System adaptability.

(D) The comprehensive identification and diagnosis system combined with neural network and data fusion theory has a certain fault tolerance ability and can meet the requirements of damage diagnosis of complex steel structure systems.

Description of drawings

Figure 1 is a block diagram of the fatigue damage identification system for AZ31 magnesium alloy.

Fig. 2 is a structural block diagram of the AZ31 magnesium alloy fatigue damage detection unit of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

The fatigue damage state model includes: fatigue crack stable growth damage state, fatigue crack instability growth damage state. The load-bearing parts in service generally work under the state of stable fatigue crack expansion and damage. When the fatigue cracks are in a state of unstable expansion and damage, the load-bearing parts are seriously damaged, and the user should carry out real-time key detection, monitoring or replacement of the load-bearing parts. , so the detection of the fatigue damage state of the load-bearing parts can prevent and reduce the occurrence of accidents, so as to reduce the personal injury, equipment loss and economic loss caused by sudden fracture.

As shown in Figure 1 and Figure 2, the fatigue damage identification system for AZ31 magnesium alloy generally consists of multiple acoustic emission transducers 6 (also called sensors), multi-channel preamplifiers 5, an acoustic emission instrument 4, and an AZ31 magnesium alloy The alloy fatigue damage detection unit 1, the damage grade evaluation unit 2 and the alarm unit 3 are composed; wherein, the AZ31 magnesium alloy fatigue damage detection unit 1 includes a filtering module 11, a primary data fusion module 12, and a secondary data fusion module 13;

Wherein, the filtering module 11 includes a data filtering processing module 11A and a waveform filtering processing module 11B;

Among them, the primary data fusion module 12 includes a first data space 12A, a second data space 12B, a first damage degree marking module 12C, a second damage degree marking module 12D, and a D-S evidence combination module 12E.

AZ31 magnesium alloy fatigue damage detection unit 1 is developed using Matlab language (version R2011b). The AZ31 magnesium alloy fatigue damage detection unit 1 is embedded in the memory of the acoustic emission instrument 4 .

In the present invention, the acoustic emission transducer 6 is used in conjunction with the preamplifier 5, that is, the output end of each acoustic emission transducer 6 is connected to the input end of a preamplifier 5, and each preamplifier 5 The output end is connected to the information input interface of the acoustic emission device 4, and the information input interface is used to receive multiple channels of burst type amplification information

In the present invention, the acoustic emission instrument 4 selects the DiSP acoustic emission system produced by PAC Company of the United States, the acoustic emission transducer 6 selects the CZ series or the WD series acoustic emission transducer produced by the PAC Company of the United States, and the multi-channel preamplifier 5 selects 2/4/6 type preamplifier produced by American PAC company.

In the present invention, when the acoustic emission transducer 6 is used to collect information within the collection time _TX section, not only the damage information is collected, but also the noise (environmental noise, electromagnetic noise, mechanical friction noise) is collected (i.e. e _S , A _S , C _S , R _S , and D _S information contain noise), therefore, in the present invention, data filtering and waveform filtering are used to denoise the acquired information. The purpose of such denoising is to obtain the five parameters required by the present invention for fatigue damage monitoring: energy e _S , measurement amplitude A _S , ringing count C _S , rise time RS and duration D _{S .}

(1) Acoustic emission transducer 6

The acoustic emission transducer 6 is used to collect burst information S _n on the AZ31 magnesium alloy in service. In the present invention, the required number of acoustic emission transducers 6 is 40 cm to 100 cm per sensing range. The information collected by the acoustic emission transducer 6 is expressed in a collective form: burst information S _n = {e _S , A _S , C _S , R _S , D _S }, where e _S represents energy, and A _S represents measurement Amplitude, C _S for ring count, R _S for rise time, and D _S for duration.

In the present invention, the first acquisition time is marked as T ₁ , the second acquisition time is marked as T ₂ , ..., the last acquisition time T _X , for the convenience of description, the last acquisition time T _X is also referred to as any Acquisition time T _X .

In the present invention, the burst type information collected by the acoustic emission transducer 6 in the collection time _TX section is expressed as If the burst information obtained at the first acquisition time T ₁ is recorded as In the same way, it can be obtained that if the burst information obtained at the second collection time T ₂ is recorded as $S_{\mathrm{no}}^{T_2}=\{e_S,A_S,C_S{,R}_S,{\mathrm{D.}}_S\}.$

(2) Preamplifier 5

The preamplifier 5 is used for receiving the burst type information After amplifying 40dB, it becomes burst type amplification information

(3) Acoustic emission instrument 4

The acoustic emission instrument 4 is used to amplify the received burst type information After A/D conversion, it becomes digital burst information The output is sent to the AZ31 magnesium alloy fatigue damage detection unit 1, and the A/D converter is equipped in the acoustic emission instrument 4. de _S stands for Digital Energy, dA _S stands for Digital Measured Amplitude, dC _S stands for Digital Ring Count, dR _S stands for Digital Rise Time and dD _S stands for Digital Duration.

(4) Data filtering processing module 11A

In the present invention, the data filter processing module 11A processes the received digital burst information Perform parameter filtering, that is, filter out electromagnetic noise and environmental noise, and then purify and obtain preliminary fatigue damage information e0 refers to the energy of digital energy de _S after parameter filtering (referred to as parameter filtering energy), A ₀ refers to the measurement range of digital measurement range dA _S after parameter filtering (referred to as parameter filtering range), and C ₀ refers to digital dC _S is the ring count after parameter filtering (referred to as parameter filter ring count), R ₀ refers to the rise time of digital rise time dR _S after parameter filtering (referred to as parameter filter rise time), D ₀ It refers to the duration of the digital duration dD _S after the parameter filtering (referred to as the parameter filtering duration).

(5) Waveform filter processing module 11B

In the present invention, the waveform filter processing module 11B performs the preliminary information on the received fatigue damage Perform waveform filtering to obtain fatigue damage information e refers to the parameter filtering energy e ₀ after waveform filtering (abbreviated as waveform filtering energy), A refers to the parameter filtering amplitude A ₀ after waveform filtering amplitude (referred to as waveform filtering amplitude), C refers to the parameter filtering ringing count C ₀ ringing count after waveform filtering (abbreviated as waveform filtering ringing count), R refers to parameter filtering rise time R ₀ rising time after waveform filtering (abbreviated as waveform filtering rising time), D refers to parameter filtering The duration D ₀ is the duration after waveform filtering (referred to as the duration of waveform filtering).

In the present invention, fatigue damage information The first aspect is used for the first damage degree flag module 12C to perform RBF neural network training to obtain the first damage degree flag model; the second aspect is used for the second damage degree flag module 12D to perform RBF neural network training to obtain the second damage degree flag Model; the third aspect is used in the first data space 12A to perform projection processing to obtain the first score matrix; the fourth aspect is used in the second data space 12B to perform projection processing to obtain the second score matrix. Information for Fatigue Damage The RBF neural network training is performed before the first data space 12A and the second data space 12B.

(6) The first data space 12A

The first aspect of the first data space 12A is the received fatigue damage information Carry out cumulative processing within the acquisition time T _X segment to obtain the accumulated fatigue damage information ${\mathrm{AC}}_{11B}^{T_x}=(e,A,C,R,\mathrm{D.});$

The second aspect of the first data space 12A is for the accumulated fatigue damage information ${\mathrm{AC}}_{11B}^{T_x}=(e,A,C,R,\mathrm{D.})$ According to the normalized relationship ${f_{11B}}^{\′}=\frac{{\mathrm{AC}}_{11B}^{T_x}-\stackrel{\‾}{{\mathrm{AC}}_{11B}^{T_x}}}{{\mathrm{\σ}}_{{\mathrm{AC}}_{11B}^{T_x}}}$ processing to obtain the normalized cumulative fatigue damage information f _11B ′; the average value of the fatigue damage information Indicates the standard deviation of cumulative fatigue damage information;

The third plane of the first data space 12A projects the normalized cumulative fatigue damage information f _11B ′ according to the first projection relationship f _12A =f _11B ′×P _a to obtain the first score matrix f _12A =(t _a1 ,t _a2 ,t _a3 ,t _a4 ,t _a5 );

P _a represents the feature matrix of the first data space;

f _12A represents the score matrix of the sample to be diagnosed in the first data space, a is the code of the first data space, t _a1 , t _a2 , t _a3 , t _a4 , t _a5 represent the sample to be diagnosed in the first data space5 t _a1 represents the score vector of the first dimension of the sample to be diagnosed in the first data space; t _a2 represents the score vector of the second dimension of the sample to be diagnosed in the first data space; t _a3 represents the score vector of the sample to be diagnosed in the first data space The score vector of the third dimension of the diagnostic sample in the first data space; t _a4 represents the score vector of the fourth dimension of the sample to be diagnosed in the first data space; t _a5 represents the fifth dimension of the sample to be diagnosed in the first data space dimension score vector.

(7) The second data space 12B

The first aspect of the second data space 12B is for the received fatigue damage information Carry out cumulative processing within the acquisition time T _X segment to obtain the accumulated fatigue damage information ${\mathrm{AC}}_{11B}^{T_x}=(e,A,C,R,\mathrm{D.});$

The second aspect of the second data space 12B is the accumulated fatigue damage information ${\mathrm{AC}}_{11B}^{T_x}=(e,A,C,R,\mathrm{D.})$ According to the normalized relationship ${f_{11B}}^{\′}=\frac{{\mathrm{AC}}_{11B}^{T_x}-\stackrel{\‾}{{\mathrm{AC}}_{11B}^{T_x}}}{{\mathrm{\σ}}_{{\mathrm{AC}}_{11B}^{T_x}}}$ processing to obtain the normalized cumulative fatigue damage information f _11B ′; the average value of the fatigue damage information Indicates the standard deviation of cumulative fatigue damage information;

The third plane of the second data space 12B projects the normalized cumulative fatigue damage information f _11B ′ according to the second projection relationship f _12B =f _11B ′×P _b to obtain the second score matrix f _12B =(t _b1 ,t _b2 ,t _b3 ,t _b4 ,t _b5 );

P _b represents the feature matrix of the second data space;

f _12B represents the score matrix of the samples to be diagnosed in the second data space, b is the code of the second data space, t _b1 , t _b2 , t _b3 , t _b4 , t _b5 represent the samples to be diagnosed in the second data space5 t _b1 represents the score vector of the first dimension of the sample to be diagnosed in the second data space; t _b2 represents the score vector of the second dimension of the sample to be diagnosed in the second data space; t _b3 represents the score vector of the sample to be diagnosed in the second dimension The score vector of the third dimension of the diagnostic sample in the second data space; t _b4 represents the score vector of the fourth dimension of the sample to be diagnosed in the second data space; t _b5 represents the fifth dimension of the sample to be diagnosed in the second data space dimension score vector.

(8) The first damage degree flag module 12C

In the present invention, the main function of the fatigue damage identification system is to identify the fatigue crack stable growth stage and the fatigue crack unstable growth stage of the AZ31 magnesium alloy, so there are two types of damage data in the present invention, that is, the crack stability type ST and the crack failure type. The stability type UT; the crack stability type ST refers to the damage information of the fatigue crack stable growth stage; the crack instability type UT refers to the damage information of the fatigue crack instability growth stage.

Acquisition of two damage types RBF neural network training samples: when the in-service AZ31 magnesium alloy is only in the stage of crack fatigue crack stable growth, the fatigue damage information Crack stability type fatigue damage information When the in-service AZ31 magnesium alloy is only in the crack fatigue crack growth stage, the fatigue damage information Crack instability type fatigue damage information

The first damage degree flag module 12C first checks the received crack stability type fatigue damage information Carry out cumulative processing within the acquisition time T _X segment to obtain the accumulated fatigue damage information ${\mathrm{AC}}_{\mathrm{ST}}^{T_x}=(e,A,C,R,\mathrm{D.});$

The second aspect of the first damage degree flag module 12C is for the accumulated fatigue damage information ${\mathrm{AC}}_{\mathrm{ST}}^{T_x}=(e,A,C,R,\mathrm{D.})$ According to the normalized relationship ${f_{\mathrm{ST}}}^{\′}=\frac{{\mathrm{AC}}_{\mathrm{ST}}^{T_x}-\stackrel{\‾}{{\mathrm{AC}}_{\mathrm{ST}}^{T_x}}}{{\mathrm{\σ}}_{{\mathrm{AC}}_{\mathrm{ST}}^{T_x}}}$ After processing, the normalized cumulative fatigue damage information f _ST ′ of the crack stability type is obtained; the average value of the crack stability type cumulative fatigue damage information Indicates the standard deviation of the cumulative fatigue damage information of the crack stability type;

The first damage degree marking module 12C thirdly performs principal component analysis on the crack stability type normalized cumulative fatigue damage information f _ST ′, and obtains the characteristic matrix P _a =(p _a1 , p _a2 , p _a3 , p _a4 , p _a5 );

p _a1 , p _a2 , p _a3 , p _a4 , p _a5 represent the eigenvectors of ₅ dimensions in the first data space; p _a1 is the eigenvector of the first dimension in the first data space; eigenvectors of 2 dimensions; p _a3 the eigenvectors of the 3rd dimension in the first data space; p _a4 the eigenvectors of the 4th dimension in the first data space; p _a5 the features of the 5th dimension in the first data space vector;

In the present invention, when the fatigue damage identification system identifies the fatigue crack stable growth stage and the fatigue crack unstable growth stage of the AZ31 magnesium alloy, the characteristic matrix P _a of the first data space needs to be quoted in the first data space 12A for projection deal with.

In the fourth aspect, the first damage degree marking module 12C projects the normalized cumulative fatigue damage information f _ST ′ according to the third projection relationship Pf _12C =f _ST ′×P _a to obtain the third scoring matrix Pf _12C =(Pt _a1 ,Pt _a2 ,Pt _a3 ,Pt _a4 ,Pt _a5 );

Pt _a1 , Pt _a2 , Pt _a3 , Pt _a4 , Pt _a5 represent the score vectors of the five dimensions of the crack-stabilized training samples in the first data space; Pt _a1 represents the crack-stabilized training samples in the first data space 1-dimensional score vector; Pt _a2 represents the score vector of the crack-stabilized training sample in the second dimension in the first data space; Pt _a3 represents the crack-stabilized training sample in the third dimension of the first data space Score vector; Pt _a4 represents the score vector of the 4th dimension of the crack-stabilized training sample in the first data space; Pt _a5 represents the score vector of the 5th dimension of the crack-stabilized training sample in the first data space.

In the fifth aspect of the first damage degree flag module 12C, the third scoring matrix Pf _12A = (Pt _a1 , Pt _a2 , Pt _a3 , Pt _a4 , Pt _a5 ) is used as the input layer information of the RBF neural network, and the RBF neural network is set output layer information $o_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right],$ Start RBF neural network training to obtain the first damage degree mark model M _a ;

In the present invention, the output layer information of the RBF neural network $o_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]$ Among them, [1 0] means that the training sample is the sign code of the fatigue crack stable growth state, and [0 1] means the training sample is the sign code of the fatigue crack growth state.

The sixth aspect of the first damage degree flag module 12C uses M _a to process the first score matrix f _12A =(t _a1 , t _a2 , t _a3 , t _a4 , t _a5 ) to obtain the first damage degree parameter f _12C =[ f _ID,a (A ₁ )f _ID,a (A ₂ )].

In the present invention, the first damage degree flag module 12C obtains the first damage degree parameter after performing RBF neural network learning algorithm processing on the received f _12A =(t _a1 , t _a2 , t _a3 , t _a4 , t _a5 ) f _12C =[f _ID,a (A ₁ )f _ID,a (A ₂ )]; wherein f _ID,a (A ₁ ) indicates that the diagnostic sample of the transducer whose code is ID is the first in the first data space The neural network output of nodes; f _ID,a (A ₂ ) represents the neural network output of the second node in the first data space of the diagnostic sample of the transducer whose code is ID; ID represents the code of the acoustic emission transducer , a is the code of the first data space.

(9) The second damage degree flag module 12D

The second damage degree flag module 12D firstly checks the received crack instability type fatigue damage information Carry out cumulative processing within the acquisition time T _X segment to obtain the accumulated fatigue damage information ${\mathrm{AC}}_{\mathrm{UT}}^{T_x}=(e,A,C,R,\mathrm{D.});$

The second aspect of the first damage degree flag module 12C is for the accumulated fatigue damage information ${\mathrm{AC}}_{\mathrm{UT}}^{T_x}=(e,A,C,R,\mathrm{D.})$ According to the normalized relationship ${f_{\mathrm{UT}}}^{\′}=\frac{{\mathrm{AC}}_{\mathrm{UT}}^{T_x}-\stackrel{\‾}{{\mathrm{AC}}_{\mathrm{UT}}^{T_x}}}{{\mathrm{\σ}}_{{\mathrm{AC}}_{\mathrm{UT}}^{T_x}}}$ After processing, the normalized cumulative fatigue damage information f _UT ′ of the crack instability type is obtained; the average value of the fatigue damage information Indicates the standard deviation of the cumulative fatigue damage information of the crack instability type;

The second damage degree marking module 12D third party performs principal component analysis on the crack instability type normalized cumulative fatigue damage information f _UT ′, and obtains the characteristic matrix P _b =(p _b1 ,p _b2 ,p _b3 ,p _b4 ,p _b5 );

p _b1 , p _b2 , p _b3 , p _b4 , p _b5 represent the eigenvectors of the 5 dimensions _{in the second data space; p b1 the} eigenvectors of the first dimension in the second data space; 2-dimension eigenvectors; p _b3 eigenvectors of the 3rd dimension in the second data space; p _b4 eigenvectors of the 4th dimension in the second data space; p _b5 eigenvectors of the 5th dimension in the second data space vector;

In the present invention, when the fatigue damage identification system identifies the fatigue crack stable growth stage and the fatigue crack unstable growth stage of the AZ31 magnesium alloy, the characteristic matrix P _b of the second data space needs to be quoted in the second data space 12B for projection deal with.

In the fourth aspect, the second damage degree marking module 12D projects the normalized cumulative fatigue damage information f _UT ′ of the crack instability type according to the fourth projection relationship Pf _12D =f _UT ′×P _b to obtain the fourth scoring matrix Pf _12D =(Pt _b1 ,Pt _b2 ,Pt _b3 ,Pt _b4 ,Pt _b5 );

Pt _b1 , Pt _b2 , Pt _b3 , Pt _b4 , Pt _b5 represent the five-dimensional score vectors of the training samples of the crack instability type in the second data space; Pt _b1 represents the crack instability type training samples in the second data space The score vector of the first dimension in ; Pt _b2 indicates the score vector of the second dimension of the training sample of the crack instability type in the second data space; Pt _b3 indicates the second dimension of the training sample of the crack instability type in the second data space The score vector of three dimensions; Pt _b4 indicates the score vector of the fourth dimension of the training sample of the crack instability type in the second data space; Pt _b5 indicates the fifth dimension of the training sample of the crack instability type in the second data space Dimension score vector.

The fifth aspect of the second damage degree marking module 12D uses the fourth score matrix Pf _12D = (Pt _b1 , Pt _b2 , Pt _b3 , Pt _b4 , Pt _b5 ) as the input layer information of the RBF neural network, and sets the RBF neural network output layer information $o_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right],$ Start RBF neural network training to obtain the first damage degree mark model M _b ;

In the present invention, the output layer information of the RBF neural network $o_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]$ Among them, [1 0] means that the training sample is the sign code of the fatigue crack stable growth state, and [0 1] means the training sample is the sign code of the fatigue crack growth state.

In the sixth aspect, the second damage degree marking module 12D uses M _b to process the second score matrix f _12B =(t _b1 , t _b2 , t _b3 , t _b4 , t _b5 ), and obtains the second damage degree parameter f _12D =[ f _ID,b (A ₁ )f _ID,b (A ₂ )].

In the present invention, the first damage degree flag module 12C obtains the second damage degree parameter after performing RBF neural network learning algorithm processing on the received f _12B =(t _b1 ,t _b2 ,t _b3 ,t _b4 ,t _b5 ) f _12D =[f _ID,b (A ₁ )f _ID,b (A ₂ )]; wherein f _ID,b (A ₁ ) indicates that the diagnostic sample of the transducer whose code is ID is the first in the second data space The neural network output of the node; f _{ID, b} (A ₂ ) represents the neural network output of the second node in the second data space of the diagnosis sample of the transducer whose code is ID; ID represents the code of the acoustic emission transducer , b is the code of the second data space.

(10) D-S Evidence Combination Module 12E

In the present invention, the DS evidence combination module 12E first considers the received f _12C =[f _ID,a (A ₁ )f _ID,a (A ₂ )] and f _12D =[f _ID,b (A ₁ )f _ID,b (A ₂ )] according to the node correlation coefficient $Q_{\mathrm{ID},i}\left(A_j\right)=\left[\begin{array}{cc}\frac{|f_{\mathrm{ID},a}\left(A_1\right)-1|}{|f_{\mathrm{ID},a}\left(A_1\right)-1|+|f_{\mathrm{ID},a}\left(A_2\right)-1|}& \frac{|f_{\mathrm{ID},a}\left(A_2\right)-1|}{|f_{\mathrm{ID},a}\left(A_1\right)-1|+|f_{\mathrm{ID},a}\left(A_2\right)-1|}\\ \frac{|f_{\mathrm{ID},b}\left(A_1\right)-1|}{|f_{\mathrm{ID},b}\left(A_1\right)-1|+|f_{\mathrm{ID},b}\left(A_2\right)-1|}& \frac{|f_{\mathrm{ID},b}\left(A_2\right)-1|}{|f_{\mathrm{ID},b}\left(A_1\right)-1|+|f_{\mathrm{ID},b}\left(A_2\right)-1|}\end{array}\right]$ Perform processing to obtain the correlation coefficient Q _ID,i (A _j ) between the sample to be diagnosed and each node in the first data space and the second data space;

Q _ID,i (A _j ) represents the correlation coefficient between the to-be-diagnosed sample of the transducer with the code ID and the node j (j=1,2) in the data space with the code i (i=a,b).

DS Evidence Combination Module 12E Second Aspect Through Basic Probability Assignment Function $m_{\mathrm{ID},i}\left(A_j\right)=\frac{Q_{\mathrm{ID},i}\left(A_j\right)}{\underset{j=1}{\overset{N_c}{\mathrm{\Σ}}}{Q}_{\mathrm{ID},i}\left(A_j\right)+N_S\×(1-R_i)\×(1-{\mathrm{\α}}_i{\mathrm{\β}}_i)}$ The basic probability distribution is performed on each node j (j=1, 2), and the basic probability distribution value m _ID,i (A _j ) of the sample to be diagnosed with each node in the first data space and the second data space is obtained.

Where α _i =max{Q _ID,i (A _j )} represents the maximum correlation coefficient between the sample to be diagnosed and the node in the data space coded as i (i=a,b); Represents the distribution value of the correlation coefficient with the node in the data space whose code is i (i=a, b); Representation code is the reliability coefficient of the data space of i (i=a, b); In the present invention, N _c is the number of nodes, and N _s is the number of data spaces, both of which are 2; m _{ID, i} (A _j ) represents The sample to be diagnosed received by the transducer whose code is ID is assigned a value with the basic probability of node j (j=1,2) in the data space coded as i (i=a,b).

DS Evidence Combination Module 12E The third aspect uses DS evidence first combination relationship Data fusion is carried out between the samples to be diagnosed and the basic probability distribution values of each node in the two data spaces, and the first-level data fusion value m _ID (B _j ) is obtained.

N _S is the number of data spaces; m _ID (B _j ) represents the primary data fusion value of the transducer whose code is ID at node j (j=1,2);

(11) Second-level data fusion module 13

In the present invention, the secondary data fusion 13 pairs the received data fusion result m _ID (B _j ) according to the second combination relationship of DS evidence $m\left(C_j\right)=\frac{\underset{\∩B_j=C_j}{\mathrm{\Σ}}\underset{j\∈\mathrm{no}}{\mathrm{\Π}}{m}_{\mathrm{ID}}\left(B_j\right)}{1-\underset{\∩B_j=\mathrm{\φ}}{\mathrm{\Σ}}\underset{j\∈\mathrm{no}}{\mathrm{\Π}}{m}_{\mathrm{ID}}\left(B_j\right)}$ process to obtain the secondary data fusion result m(C _j ), where C _j represents the node identifier in the secondary data fusion module.

n is the number of acoustic emission transducers 6 set; m(C _j ) indicates that all acoustic emission transducers 6 in the fatigue damage identification system are in the form of nodes, and the secondary Data fusion value.

(12) Damage rating unit 2

In the present invention, the damage grade assessment unit 2 performs grade assessment on the received secondary data fusion result m(C _j ) to obtain the fatigue damage state of the in-service AZ31 magnesium alloy.

Rating conditions: when the maximum value of the node w=m(C ₁ ), the fatigue damage state is: the in-service AZ31 magnesium alloy is in the state of stable fatigue crack growth damage; when the maximum value of the node w=m(C ₂ ), the fatigue damage state is: The damage state is: the in-service AZ31 magnesium alloy is in the state of fatigue crack instability and propagation damage;

The node maximum value w=max(m(C ₁ ),m(C ₂ )), m(C ₁ ) represents the secondary Data fusion value, m(C ₂ ) represents the secondary data fusion value of all transducers in the fatigue damage identification system at node 2 (fatigue crack instability state). w represents the maximum value of the secondary data fusion values of all transducers in the fatigue damage identification system at two nodes.

(13) Alarm unit 3

In the present invention, the alarm module 3 does not alarm when w=m(C ₁ ), and activates an alarm signal when w=m(C ₂ ). The alarm unit 3 adopts a prompt sound alarm output in the form of a horn, a loudspeaker, or the like.

Embodiment 1: Acoustic emission detection is performed on a certain automobile hub.

The composition of the AZ31 magnesium alloy used in the hub is shown in Table 1:

Table 1 AZ31 magnesium alloy composition content

The detection equipment includes: (A) two R15 narrow-band acoustic emission transducers (CZ series of PAC company, response frequency is 100kHz ~ 400kHz, center frequency 150kHz) and two wide-band transducers (WD series of PAC company, Response frequency is 20kHz ~ 1MHz).

(B) Four PAC Corporation 2/4/6 type preamplifiers.

(C) The acoustic emission instrument is a full-digital 16-channel DiSP acoustic emission system from PAC Company of the United States. The threshold value of the acoustic emission instrument detection is 40dB, the acoustic emission peak definition time PDT is 300μs, the acoustic emission impact limit time HDT is 600μs, and the acoustic emission impact locking time HLT is 1000μs.

The acoustic emission information received by the four transducers in the AZ31 magnesium alloy within the sampling time T _X period is identified by the fatigue damage identification system, and the results of the fatigue damage detection unit are shown in Table 2.

Table 2A Basic probability distribution results of the acoustic emission information received by the four transducers in the fatigue damage detection unit

Table 2B The first-level data fusion results of the acoustic emission information received by the four transducers in the fatigue damage monitoring unit

transducer m _ID (B ₁ ) m _ID (B ₂ ) uncertainty 1 0.9323 0.0312 0.0365 2 0.7725 0.1120 0.1155 3 0.9849 0.0068 0.0084 4 0.9553 0.0203 0.0243

Table 2C The results of the secondary data fusion of the acoustic emission information received by the four transducers in the fatigue damage monitoring unit

m(C ₁ ) m(C ₂ ) uncertainty Secondary Fusion Results 1.0000 0.0000 0.0000

The fusion results in Table 2C are analyzed by the damage grade assessment unit 2, and the maximum value of the node can be obtained as m(C ₁ ), so the AZ31 magnesium alloy is in the stable growth stage of fatigue cracks, and no alarm is required.

It can be seen that data fusion can reduce the uncertainty of local fusion of neural network, which greatly improves the credibility of diagnostic decision-making. At the same time, it improves the fault-tolerant capability of the diagnosis system and can meet the requirements of damage diagnosis of complex steel structure systems.

The present invention adopts principal component analysis, neural network and two-level data fusion method to identify the fatigue damage state of AZ31 magnesium alloy, and establishes an identification and diagnosis system for fatigue damage state of AZ31 magnesium alloy: first, use the neural network model to analyze each acoustic emission The transducer information is locally diagnosed in different damage data spaces constructed by the principal component model; then the output results of the neural network in the same transducer are used to construct the basic probability value of the first-level data fusion module, and the first-level data fusion is performed; finally The first-level data fusion results of multiple sensors are combined with the second-level data, and then the fatigue damage state is diagnosed through the damage identification module. The model can be used to identify and diagnose the damage state of AZ31 magnesium alloy during the fatigue process, and then provide a basis for its reliable operation.

Claims (6)

1. A recognition system based on PCA and TDF to the state of fatigue damage of AZ31 magnesium alloy, is characterized in that: the system includes a plurality of acoustic emission transducers (6), multi-channel preamplifiers (5), an acoustic emission transducer The launcher (4) is characterized in that: it also includes an AZ31 magnesium alloy fatigue damage detection unit (1); The AZ31 magnesium alloy fatigue damage detection unit (1) includes a filtering module (11), a first-level data fusion module (12), and a second-level data fusion module (13), wherein the filtering module (11) has a data filtering processing module (11A ) and waveform filter processing module (11B), the primary data fusion module (12) has the first data space (12A), the second data space (12B), the first damage degree flag module (12C), the second damage degree flag Module (12D), D-S Evidence Combination Module (12E); The acoustic emission transducer (6) is used to collect the burst-type information burst-type information of the in-service AZ31 magnesium alloy within the collection time T _X segment Preamplifier (5), used for receiving the burst type information After amplifying 40dB, it becomes burst type amplification information The acoustic emission instrument (4) is used to amplify the received burst type information After A/D conversion, it becomes digital burst information Output to AZ31 magnesium alloy fatigue damage detection unit (1); de _S means digital energy, dA _S means digital measurement amplitude, dC _S means digital ring count, dR _S means digital rise time and dD _S means digital duration; Data filter processing module (11A) is to the digital burst type information that receives Perform parameter filtering, that is, filter out electromagnetic noise and environmental noise, and then purify and obtain preliminary fatigue damage information The waveform filter processing module (11B) performs preliminary information on the received fatigue damage Perform waveform filtering to obtain fatigue damage information e ₀ refers to the energy of digital energy de _S after parameter filtering, A ₀ refers to the measurement amplitude of digital measurement range dA _S after parameter filtering, C ₀ refers to the digital ringing count dC _S after parameter filtering Ringing count, R ₀ refers to the rise time of the digital rise time dR _S after parameter filtering, D ₀ refers to the duration of the digital duration dD _S after parameter filtering; e refers to the parameter filtering energy e ₀ after waveform The energy after filtering, A refers to the parameter filtering amplitude A ₀ the amplitude after waveform filtering, C refers to the parameter filtering ring count C ₀ ringing count after waveform filtering, R refers to the parameter filtering rise time R ₀ after waveform filtering Rise time after filtering, D refers to the parameter filtering duration D ₀ duration after waveform filtering; Fatigue damage information received by the transducer for AZ31 magnesium alloy Carry out the accumulation processing within the acquisition time T _X period, and then normalize to obtain the normalized cumulative fatigue damage information f _11B ′, and project f _11B ′ in the first data space (12A) and the second data space (12B) respectively Obtain respective scoring matrices f _12A =(t _a1 ,t _a2 ,t _a3 ,t _a4 ,t _a5 ) and f _12B =(t _b1 ,t _b2 ,t _b3 ,t _b4 ,t _b5 ), scoring matrix f _12A = (t _a1 ,t _a2 ,t _a3 ,t _a4 ,t _a5 ) are processed by the first damage degree flag module (12C) to obtain the neural network output f _12C in the first data space =[f _ID,a (A ₁ ) f _{ID, a} (A ₂ )], f _{ID, a} (A ₁ ) represents the neural network output of the first node in the first data space of the diagnostic sample of the transducer whose code is ID; f _{ID, a} (A ₂ ) represent the neural network output of the diagnostic sample of the transducer whose code is ID in the first data space; ID represents the code of the acoustic emission transducer, a is the code of the first data space; score matrix f _12B = (t _b1 , t _b2 , t _b3 , t _b4 , t _b5 ) is processed by the second damage degree flag module (12D) to obtain the neural network output f in the second data space _12D = [f _ID,b ( A ₁ ) f _ID,b (A ₂ )], f _ID,b (A ₁ ) represents the neural network output of the first node in the second data space of the diagnostic sample of the transducer whose code is ID; f _{ID, b} (A ₂ ) represents the neural network output of the diagnostic sample of the transducer whose code is ID in the second data space; ID represents the code of the acoustic emission transducer, and b is the code of the second data space; Assign f _12C and f _12D to the correlation coefficients of each node, assign basic probability to f _12C and f _12D after assignment, and then perform DS evidence combination module (12E) processing to obtain the data fusion result m _ID (B _j ); t _a1 represents the score vector of the first dimension of the sample to be diagnosed in the first data space; t _a2 represents the score vector of the second dimension of the sample to be diagnosed in the first data space; t _a3 represents the score vector of the sample to be diagnosed in the first data space The score vector of the third dimension in the first data space; t _a4 represents the score vector of the sample to be diagnosed in the fourth dimension in the first data space; t _a5 represents the score vector of the sample to be diagnosed in the fifth dimension in the first data space Score vector; t _b1 represents the score vector of the sample to be diagnosed in the first dimension in the second data space; t _b2 represents the score vector of the sample to be diagnosed in the second dimension of the second data space; t _b3 represents the sample to be diagnosed in The score vector of the third dimension in the second data space; t _b4 indicates that the sample to be diagnosed is in the second data space The score vector of the fourth dimension; t _b5 represents the score vector of the fifth dimension of the sample to be diagnosed in the second data space; Secondary data fusion (13) for the received data fusion result m _ID (B _j ) according to the second combination relationship of DS evidence Process to obtain the secondary data fusion result m(C _j ), C _j represents the node identification in the secondary data fusion module; The damage grade assessment unit (2) performs grade assessment on the received secondary data fusion result m(C _j ), and obtains the fatigue damage state of the in-service AZ31 magnesium alloy; The alarm unit (3) receives the fatigue damage identification information output by the damage level assessment unit (2), and activates an alarm signal to alarm. 2. the recognition system that AZ31 magnesium alloy is carried out fatigue damage state based on PCA and TDF according to claim 1, is characterized in that: The first data space (12A) The first aspect is for the received fatigue damage information Carry out cumulative processing within the acquisition time T _X segment to obtain the accumulated fatigue damage information ${\mathrm{AC}}_{11B}^{T_x}=(e,A,C,R,\mathrm{D.});$ The first data space (12A) and the second aspect are for accumulated fatigue damage information ${\mathrm{AC}}_{11B}^{T_x}=(e,A,C,R,\mathrm{D.})$ According to the normalized relationship ${f_{11B}}^{\′}=\frac{{\mathrm{AC}}_{11B}^{T_x}-\stackrel{\‾}{{\mathrm{AC}}_{11B}^{T_x}}}{{\mathrm{\σ}}_{{\mathrm{AC}}_{11B}^{T_x}}}$ processing to obtain the normalized cumulative fatigue damage information f _11B ′; the average value of the fatigue damage information Indicates the standard deviation of cumulative fatigue damage information; The third plane of the first data space (12A) projects the normalized cumulative fatigue damage information f _11B ′ according to the first projection relationship f _12A =f _11B ′×P _a to obtain the first score matrix f _12A =(t _a1 , t _a2 , t _a3 , t _a4 , t _a5 ). 3. the recognition system that AZ31 magnesium alloy is carried out fatigue damage state based on PCA and TDF according to claim 1, is characterized in that: The first aspect of the second data space (12B) deals with the received fatigue damage information Carry out cumulative processing within the acquisition time T _X segment to obtain the accumulated fatigue damage information ${\mathrm{AC}}_{11B}^{T_x}=(e,A,C,R,\mathrm{D.});$ The second aspect of the second data space (12B) deals with accumulated fatigue damage information ${\mathrm{AC}}_{11B}^{T_x}=(e,A,C,R,\mathrm{D.})$ According to the normalized relationship ${f_{11B}}^{\′}=\frac{{\mathrm{AC}}_{11B}^{T_x}-\stackrel{\‾}{{\mathrm{AC}}_{11B}^{T_x}}}{{\mathrm{\σ}}_{{\mathrm{AC}}_{11B}^{T_x}}}$ processing to obtain the normalized cumulative fatigue damage information f _11B ′; the average value of the fatigue damage information Indicates the standard deviation of cumulative fatigue damage information; The second data space (12B) projects the normalized cumulative fatigue damage information f _11B ′ according to the second projection relationship f _12B =f _11B ′×P _b to obtain the second score matrix f _12B =(t _b1 , t _b2 , t _b3 , t _b4 , t _b5 ). 4. the recognition system that AZ31 magnesium alloy is carried out fatigue damage state based on PCA and TDF according to claim 1, is characterized in that: The first damage degree flag module (12C) first checks the received crack stability type fatigue damage information Carry out cumulative processing within the acquisition time T _X segment to obtain the accumulated fatigue damage information ${\mathrm{AC}}_{\mathrm{ST}}^{T_x}=(e,A,C,R,\mathrm{D.});$ The second aspect of the first damage degree flag module (12C) is for the accumulated fatigue damage information ${\mathrm{AC}}_{\mathrm{ST}}^{T_x}=(e,A,C,R,\mathrm{D.})$ According to the normalized relationship ${f_{\mathrm{ST}}}^{\′}=\frac{{\mathrm{AC}}_{\mathrm{ST}}^{T_x}-\stackrel{\‾}{{\mathrm{AC}}_{\mathrm{ST}}^{T_x}}}{{\mathrm{\σ}}_{{\mathrm{AC}}_{\mathrm{ST}}^{T_x}}}$ After processing, the normalized cumulative fatigue damage information f _ST ′ of the crack stability type is obtained; the average value of the crack stability type cumulative fatigue damage information Indicates the standard deviation of the cumulative fatigue damage information of the crack stability type; The third party of the first damage degree marking module (12C) conducts principal component analysis on the normalized cumulative fatigue damage information f _ST ′ of the crack stability type, and obtains the characteristic matrix P _a =(p _a1 ,p _a2 ,p _a3 ,p _a4 ,p _a5 ); p _a1 the eigenvector of the first dimension in the first data space; p _a2 the eigenvector of the second dimension in the first data space; p _a3 the third dimension in the first data space The eigenvector of p _a4 the eigenvector of the 4th dimension in the first data space; p _a5 the eigenvector of the 5th dimension in the first data space; The fourth aspect of the first damage degree marking module (12C) projects the normalized cumulative fatigue damage information f _ST ′ according to the third projection relationship Pf _12C =f _ST ′×P _a to obtain the third scoring matrix Pf _12C =( Pt _a1 , Pt _a2 , Pt _a3 , Pt _a4 , Pt _a5 ); Pt _a1 represents the score vector of the first dimension of the crack-stabilized training samples in the first data space; Pt _a2 represents the crack-stabilized training samples in the first dimension The score vector of the second dimension in a data space; Pt _a3 represents the score vector of the third dimension of the crack-stabilized training samples in the first data space; Pt _a4 represents the crack-stabilized training samples in the first data space The score vector of the 4th dimension; Pt _a5 represents the score vector of the 5th dimension of the training sample of the crack stability type in the first data space; The fifth aspect of the first damage degree flag module (12C) uses the third scoring matrix Pf _12A = (Pt _a1 , Pt _a2 , Pt _a3 , Pt _a4 , Pt _a5 ) as the input layer information of the RBF neural network, and sets the RBF neural network The output layer information of the network $o_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right],$ Start RBF neural network training to obtain the first damage degree mark model M _a ; Output layer information of RBF neural network $o_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]$ Among them, [1 0] means that the training sample is the sign code of the fatigue crack stable growth state, and [0 1] means the training sample is the sign code of the fatigue crack unstable growth state; The sixth aspect of the first damage degree flag module (12C) uses M _a to process the first score matrix f _12A = (t _a1 , t _a2 , t _a3 , t _a4 , t _a5 ) to obtain the first damage degree parameter f _12C =[f _{ID, a} (A ₁ ) f _{ID, a} (A ₂ )]. 5. the recognition system that AZ31 magnesium alloy is carried out fatigue damage state based on PCA and TDF according to claim 1, is characterized in that: The first aspect of the second damage degree flag module (12D) is to receive the crack instability type fatigue damage information Carry out cumulative processing within the acquisition time T _X segment to obtain the accumulated fatigue damage information ${\mathrm{AC}}_{\mathrm{UT}}^{T_x}=(e,A,C,R,\mathrm{D.});$ The second aspect of the first damage degree flag module (12D) is for the accumulated fatigue damage information ${\mathrm{AC}}_{\mathrm{UT}}^{T_x}=(e,A,C,R,\mathrm{D.})$ According to the normalized relationship ${f_{\mathrm{UT}}}^{\′}=\frac{{\mathrm{AC}}_{\mathrm{UT}}^{T_x}-\stackrel{\‾}{{\mathrm{AC}}_{\mathrm{UT}}^{T_x}}}{{\mathrm{\σ}}_{{\mathrm{AC}}_{\mathrm{UT}}^{T_x}}}$ After processing, the normalized cumulative fatigue damage information f _UT ′ of the crack instability type is obtained; the average value of the fatigue damage information Indicates the standard deviation of the cumulative fatigue damage information of the crack instability type; The third damage degree marking module (12D) conducts principal component analysis on the normalized cumulative fatigue damage information f _UT ′ of the crack instability type, and obtains the characteristic matrix P _b =(p _b1 ,p _b2 , p _b3 ,p _b4 ,p _b5 ); p _b1 is the eigenvector of the first dimension in the second data space; p _b2 is the eigenvector of the second dimension in the second data space; p _b3 is the third in the second data space The eigenvector of dimension; p _b4 the eigenvector of the 4th dimension in the second data space; p _b5 the eigenvector of the 5th dimension in the second data space; In the fourth aspect of the second damage degree marking module (12D), the normalized cumulative fatigue damage information f _UT ′ of the crack instability type is projected according to the fourth projection relationship Pf _12D =f _UT ′×P _b to obtain the fourth score matrix Pf _12D = (Pt _b1 , Pt _b2 , Pt _b3 , Pt _b4 , Pt _b5 ); Pt _b1 represents the score vector of the first dimension of the training samples of the crack instability type in the second data space; Pt _b2 represents the crack instability The score vector of the training sample of the type in the second data space in the second dimension; Pt _b3 represents the score vector of the training sample of the crack instability type in the third dimension in the second data space; Pt _b4 represents the score vector of the crack instability type The score vector of the fourth dimension of the training sample in the second data space; Pt _b5 represents the score vector of the fifth dimension of the training sample of the crack instability type in the second data space; The fifth aspect of the second damage degree flag module (12D) uses the fourth scoring matrix Pf _12D = (Pt _b1 , Pt _b2 , Pt _b3 , Pt _b4 , Pt _b5 ) as the input layer information of the RBF neural network, and sets the RBF neural network The output layer information of the network $o_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right],$ Start RBF neural network training to obtain the first damage degree mark model M _b ; Output layer information of RBF neural network $o_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]$ Among them, [1 0] means that the training sample is the sign code of the fatigue crack stable growth state, and [0 1] means the training sample is the sign code of the fatigue crack unstable growth state; The sixth aspect of the second damage degree flag module (12D) uses M _b to process the second score matrix f _12B = (t _b1 , t _b2 , t _b3 , t _b4 , t _b5 ) to obtain the second damage degree parameter f _12D =[f _ID,b (A ₁ ) f _ID,b (A ₂ )]. 6. the recognition system that AZ31 magnesium alloy is carried out fatigue damage state based on PCA and TDF according to claim 1, is characterized in that: The DS evidence combination module (12E) first regards the received f _12C =[f _ID,a (A ₁ ) f _ID,a (A ₂ )] and f _12D =[f _ID,b (A ₁ ) f _{ID ,b} (A ₂ )] According to the node correlation coefficient $o_{\mathrm{ID},i}\left(A_j\right)=\left[\begin{array}{cc}\frac{|f_{\mathrm{ID},a}\left(A_1\right)-1|}{|f_{\mathrm{ID},a}\left(A_1\right)-1|+|f_{\mathrm{ID},a}\left(A_2\right)-1|}& \frac{|f_{\mathrm{ID},a}\left(A_2\right)-1|}{|f_{\mathrm{ID},a}\left(A_1\right)-1|+|f_{\mathrm{ID},a}\left(A_2\right)-1|}\\ \frac{|f_{\mathrm{ID},b}\left(A_1\right)-1|}{|f_{\mathrm{ID},b}\left(A_1\right)-1|+|f_{\mathrm{ID},b}\left(A_2\right)-1|}& \frac{|f_{\mathrm{ID},b}\left(A_2\right)-1|}{|f_{\mathrm{ID},b}\left(A_1\right)-1|+|f_{\mathrm{ID},b}\left(A_2\right)-1|}\end{array}\right]$ Perform processing to obtain the correlation coefficient Q _ID,i (A _j ) between the sample to be diagnosed and each node in the first data space and the second data space; DS Evidence Combination Module (12E) The second aspect passes the basic probability assignment function $m_{\mathrm{ID},i}\left(A_j\right)=\frac{Q_{\mathrm{ID},i}\left(A_j\right)}{\underset{j=1}{\overset{N_c}{\mathrm{\Σ}}}{Q}_{\mathrm{ID},i}\left(A_j\right)+N_{\mathrm{the\; s}}\×(1-R_i)\×(1-{\mathrm{\α}}_i{\mathrm{\β}}_i)}$ Carry out basic probability distribution to each node j (j=1,2), obtain the basic probability distribution value m _ID,i (A _j ) of the sample to be diagnosed with each node in the first data space and the second data space; The third aspect of the DS evidence combination module (12E) uses the first combination relationship of DS evidence Carry out data fusion between the sample to be diagnosed and the basic probability distribution value of each node in the two data spaces, and obtain the first-level data fusion value m _ID (B _j ); Where α _i =max{Q _ID,i (A _j )} represents the maximum correlation coefficient between the sample to be diagnosed and the node in the data space coded as i (i=a,b); Represents the distribution value of the correlation coefficient with the node in the data space whose code is i (i=a, b); Representation code is the reliability coefficient of the data space of i (i=a, b); In the present invention, N _c is the number of nodes, and N _s is the number of data spaces, both of which are 2; m _{ID, i} (A _j ) represents The sample to be diagnosed received by the transducer whose code is ID is in the data space whose code is i (i=a,b) and the basic probability assignment value of node j (j=1,2); m _ID (B _j ) means The primary data fusion value of the transducer whose code is ID at node j (j=1, 2).