CN102879473A - 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 recognition of a fatigue damage state of an AZ31 magnesium alloy based on PCA (principal component analysis) and TDF (tactical data fusion). The system comprises a plurality of acoustic emission transducers (6), multiple preamplifiers (5), an acoustic emission device (4) and an AZ31 magnesium alloy fatigue damage detecting unit (1). The AZ31 magnesium alloy fatigue damage detecting unit (1) comprises a filtering module (11), a first-level data fusion module (12) and a second-level data fusion module (13). Principal component analysis and type of fatigue damage are combined, a neural network is trained in a data space to obtain a mark of damage degree in the data space of each transducer, and partial diagnosis is performed to each acoustic emission transducer by the mark of damage degree. An elementary probability value of data fusion is constructed by output result of the neural network. Finally, the fatigue damage state is diagnosed by a combination relation of data fusion. The damage state can be recognized and diagnosed in the fatigue process of the AZ31 magnesium alloy, and basis is further provided for stable operation of the system.

Description

The AZ31 magnesium alloy is carried out the recognition system of Fatigue Damage States based on PCA and TDF

Technical field

The present invention relates to a kind of method to identifying at labour AZ31 magnesium alloy Fatigue Damage States during one's term of military service.More particularly, refer to that a kind of data to the acoustic emission transducer collection at first adopt principal component analysis (PCA) (PCA) to make up different data spaces according to its damage type, the training of carrying out neural network in data space obtains the injury tolerance Sign module of each transducer in data space, then use this module to carrying out two-stage data fusion (TDF) in the acoustic emission data of labour AZ31 magnesium alloy Real-time Collection, thereby identify at labour AZ31 magnesium alloy to belong to which kind of Fatigue Damage States.

Background technology

In recent years, magnesium-alloy material output up to 20%, becomes the material that receives much concern in the annual growth in the whole world.Magnesium alloy has demonstrated great application prospect at industrial circles such as the vehicles, electronics and communication product, Aero-Space, chemical industry and machineries.The AZ31 magnesium alloy refers to contain the magnesium alloy of Al 3%, Zn 1% (wt%), is the most widely wrought magnesium alloy of present commercial applications.The AZ31 magnesium alloy can be squeezed into bar, tubing, section bar, can be rolled into thin plate, slab, also can be processed into forging, is mainly used in the aspects such as automobile, Aero-Space parts, weapons.

The AZ31 magnesium alloy under arms afterwards damage be the one of the main reasons that causes its inefficacy, to identify its faulted condition for this reason, in time, correctly estimate the Fatigue Damage States of AZ31 magnesium alloy, for its safe operation and life prediction provide foundation.

Acoustic emission (Acoustic Emission Technique) because of have dynamically, the advantage such as in real time detection, be widely used in the damage check of structure and member.Practice shows, material is in the different phase of fatigue process, a series of different variations can occur in its characteristics of Acoustic Emission, that is to say the fatigue damage stage that the AZ31 magnesium alloy is different, different acoustic emission signals will be arranged, and the transformation of faulted condition, often cause the variation of a plurality of parameters of acoustic emission, simultaneously a certain parameter changes and can be caused by multiple faulted condition again, so be necessary to adopt the Data fusion technique of many acoustic emission transducers, namely take full advantage of many acoustic emission transducers data resource in different time and space, adopt the many acoustic emission transducer observation data of computer technology to obtaining by the time sequence, under certain criterion, analyze, comprehensively, domination and use, acquisition is explained the consistance of measurand and is described, and then realize corresponding decision-making and estimate, make system obtain each ingredient information more fully than it.Therefore the present invention introduces Data fusion technique in the AZ31 magnesium alloy faulted condition recognition system, sets up principal component analysis (PCA), and the diagnostic system that data fusion and artificial neural network combine is identified, diagnosed AZ31 magnesium alloy faulted condition.

Principal component analysis (PCA) also claims principal component analysis.Principal component analysis (PCA) utilizes the thought of dimensionality reduction, is converted into the multivariate statistical method of several overall targets in lower a plurality of indexs of prerequisite of loss little information.Usually the index that changes into is called major component, wherein each major component is the linear combination of original variable, and uncorrelated mutually between each major component, makes major component have some more superior performance than original variable." principal component analysis (PCA) " quotes publishing house of the Renmin University of China in " multivariate statistical analysis " of the 2nd edition publication September in 2008, the 152nd page to the 154th page content introduction.

Neural network is an a kind of nonlinear system of the people's of simulation thinking.Radial basis function (RBF) neural network is that basis has local modulation in people's cortex and overlapping receptive field proposes, and is called again local receptive field neural network.It is a kind of three layers of feed-forward type network model that comprise input layer, hidden layer and output layer.Because the RBF network structure is simple, and has the ability of approaching the arbitrary continuation function with arbitrary accuracy, learning rate is fast, so more and more be widely used in every field.

Summary of the invention

The personnel's injury, equipment loss and the economic loss that in use happen suddenly and rupture and cause in order to reduce the AZ31 magnesium alloy, the present invention proposes a kind of employing principal component analysis (PCA), and the method that neural network and two-stage data fusion combine is identified in the Fatigue Damage States of labour AZ31 magnesium alloy.This fatigue state recognition system at first adopts principal component analysis (PCA) to make up two data spaces to different damage data in the training sample, adopt neural net method that the information that the multi-Channel Acoustic transmitting transducer collects is carried out neural metwork training under two data spaces, obtain to be used for judging the injury tolerance Sign module of AZ31 magnesium alloy different fatigue faulted condition; Then the neural network output of this module under two data spaces is carried out carrying out the one-level data fusion behind the basic probability assignment, a plurality of transducer one-level data fusion results being carried out secondary data merges again, obtain data fusion module, and then data fusion module is embedded in the AZ31 Fatigue of Magnesium Alloys recognition system.It is lower in working order to be embedded with data fusion module of the present invention, can be to identifying at labour AZ31 magnesium alloy different fatigue faulted condition, and the result who identifies is made early warning.

The present invention is a kind of employing principal component analysis (PCA), the technology that neural network and data fusion combine is carried out the recognition system that Fatigue Damage States is identified to the AZ31 magnesium alloy, this system includes a plurality of acoustic emission transducers (6), multichannel prime amplifier (5), an Acoustic radiating instrument (4), it is characterized in that: also include an AZ31 Fatigue of Magnesium Alloys damage check unit (1);

AZ31 Fatigue of Magnesium Alloys damage check unit (1) includes filtering module (11) and one-level data fusion module (12), secondary data Fusion Module (13), wherein, filtering module (11) has data filtering processing module (11A) and waveform filtering processing module (11B), one-level data fusion module (12) has the first data space (12A), the second data space (12B), the first injury tolerance Sign module (12C), the second injury tolerance Sign module (12D), D-S Evidence Combination Methods module (12E);

Acoustic emission transducer (6) and prime amplifier (5) are for supporting the use, the output terminal that is each acoustic emission transducer (6) is connected with the input end of a prime amplifier (5), the output terminal of each prime amplifier (5) is connected on the input information interface of Acoustic radiating instrument (4), and this input information interface is used for receiving multichannel demblee form amplification message

AZ31 Fatigue of Magnesium Alloys damage check unit (1) is embedded in the storer of Acoustic radiating instrument (4);

Acoustic emission transducer (6) is used for being captured in labour AZ31 magnesium alloy at acquisition time T _XDemblee form information burst type information in the section

Prime amplifier (5) is used for the demblee form information to receiving

Become the demblee form amplification message after amplifying 40dB

Acoustic radiating instrument (4) is used for the demblee form amplification message to receiving

After the A/D conversion, become digital demblee form information

Export to AZ31 Fatigue of Magnesium Alloys damage check unit (1);

The digital demblee form information of data filtering processing module (11A) in the filtering module (11) of AZ31 Fatigue of Magnesium Alloys damage check unit (1) to receiving

Carry out parametric filtering, filter electromagnetic noise and neighbourhood noise after, purifying obtains acoustic emission fatigue damage preliminary information

Then waveform filtering processing module (11B) is to acoustic emission fatigue damage preliminary information Carry out waveform filtering, obtain acoustic emission fatigue damage information $f_{11B}^{T_X}=(e,A,C,R,D);$

Transducer is received the fatigue damage information of AZ31 magnesium alloy

Carry out acquisition time T _XAccumulated process in the section, then normalization obtains normalization accumulation of fatigue damage information f _11B', with f _11B' obtain separately score matrix f in the lower projection of the first data space (12A) and the second data space (12B) respectively _12A=(t _A1, t _A2, t _A3, t _A4, t _A5) and f _12B=(t _B1, t _B2, t _B3, t _B4, t _B5), score matrix f _12A=(t _A1, t _A2, t _A3, t _A4, t _A5) process the neural network output f obtain in the first data space through the first injury tolerance Sign module (12C) _12C=[f _{ID, a}(A ₁) f _{ID, a}(A ₂)], score matrix f _12B=(t _B1, t _B2, t _B3, t _B4, t _B5) process the neural network output f that obtains in the second data space through the second injury tolerance Sign module (12D) _12D=[f _{ID, b}(A ₁) f _{ID, b}(A ₂)]; With f _12CAnd f _12DCarry out carrying out basic probability assignment after the related coefficient assignment of each node, then carry out D-S Evidence Combination Methods module (12E) and process the data fusion m as a result that obtains single transducer _ID(B _j), again D-S Evidence Combination Methods module (12E) result of all transducers is carried out secondary data and merge (13) and obtain as a result m (C of data fusion _j), this result exports fatigue damage identifying information D to alarm unit (3) after resolve damage grade evaluation unit (2).

The present invention is a kind of according to acoustic emission information, adopt neural network under the data space of principal component analysis (PCA) structure, faulted condition to be carried out part identification, then the application data integration technology is carried out the two-stage fusion to the Neural Network Diagnosis result, identifying and diagnosing goes out the final Fatigue Damage States of AZ31 magnesium alloy, and the advantage of this recognition system is:

(A) capture card in the employing Acoustic radiating instrument is to using acoustic emission information (the energy e of the acoustic emission transducer on the AZ31 magnesium alloy after a while _S, measuring amplitude A _S, Ring-down count C _S, rise time R _S, duration D _S) gather, and with the input information of this relevant information as the recognition system of acoustic emission neural network, so that the present invention is in the acoustic emission detection process, can be by Acoustic radiating instrument to the acoustic emission transducer information, then according to the variation of acoustic emission information parameter and waveform, identifying is damage information, or noise information.

(B) the damage mode type according to fatigue adopts the different data space of principal component analysis (PCA) structure, under each data space, faulted condition is carried out part identification by neural network, through two-stage data fusion output diagnostic result, the difference that takes full advantage of signal between different type of impairments is identified damage mode.

(C) can be comprehensive, arrange and use Monitoring Data a plurality of, the polytype acoustic emission transducer, take full advantage of the information of each acoustic emission transducer, increased reliability and the accuracy of diagnostic result, improved the adaptive faculty of diagnostic system.

(D) the comprehensive identifying and diagnosing system that combines of neural network and Data-Fusion theory has certain fault-tolerant ability, can satisfy the requirement of steel construction complication system damage.

Description of drawings

Fig. 1 is AZ31 Fatigue of Magnesium Alloys damage recognition system structured flowchart.

Fig. 2 is the structured flowchart of AZ31 Fatigue of Magnesium Alloys damage check of the present invention unit.

Embodiment

The present invention is described in further detail below in conjunction with drawings and Examples.

In the Fatigue Damage States pattern, include: the stable expansion of fatigue crack faulted condition, fatigue crack unstable propagation faulted condition.Generally under the stable expansion of fatigue crack faulted condition, work at the labour bearing member, when being in fatigue crack unstable propagation faulted condition, this bearing member damage is more serious, the user should carry out real-time emphasis to bearing member and detect, monitors or change, therefore bearing member being carried out Fatigue Damage States detects and can prevent and the generation of minimizing accident the personnel's injury, equipment loss and the economic loss that cause to reduce the burst fracture.

Referring to Fig. 1, shown in Figure 2, generally also claim sensor by a plurality of acoustic emission transducer 6(for the fatigue damage recognition system of AZ31 magnesium alloy), multichannel prime amplifier 5, an Acoustic radiating instrument 4, AZ31 Fatigue of Magnesium Alloys damage check unit 1, damage grade evaluation unit 2 and an alarm unit 3 form; Wherein, AZ31 Fatigue of Magnesium Alloys damage check unit 1 includes filtering module 11 and one-level data fusion module 12, secondary data Fusion Module 13;

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

Wherein, one-level data fusion module 12 includes the first data space 12A, the second data space 12B, the first injury tolerance Sign module 12C, the second injury tolerance Sign module 12D, D-S Evidence Combination Methods module 12E.

AZ31 Fatigue of Magnesium Alloys damage check unit 1 adopts Matlab language (version R2011b) exploitation.AZ31 Fatigue of Magnesium Alloys damage check unit 1 is embedded in the storer of Acoustic radiating instrument 4.

In the present invention, acoustic emission transducer 6 and prime amplifier 5 are for supporting the use, the output terminal that is each acoustic emission transducer 6 is connected with the input end of a prime amplifier 5, the output terminal of each prime amplifier 5 is connected on the input information interface of Acoustic radiating instrument 4, and this input information interface is used for receiving multichannel demblee form amplification message

In the present invention, Acoustic radiating instrument 4 is chosen the DiSP acoustic emission system that U.S. PAC company produces, acoustic emission transducer 6 is chosen CZ series or the WD series acoustic emission transducer that U.S. PAC company produces, and multichannel prime amplifier 5 is chosen the 2/4/6 type prime amplifier that U.S. PAC company produces.

In the present invention, utilize acoustic emission transducer 6 at acquisition time T _XWhen carrying out information acquisition in the section, not only damage information is gathered, also noise (neighbourhood noise, electromagnetic noise, mechanical friction noise) being gathered simultaneously (is e _S, A _S, C _S, R _S, D _SIn the information be comprise noisy), therefore, in the present invention, adopted data filtering and waveform filtering that the information that gather to obtain has been carried out denoising.Its purpose of such denoising is to obtain five parameters that are used for carrying out the fatigue damage monitoring required for the present invention: i.e. energy e _S, measuring amplitude A _S, Ring-down count C _S, rise time R _SWith duration D _S

(1) acoustic emission transducer 6

Acoustic emission transducer 6 is used for being captured in the demblee form information S on the labour AZ31 magnesium alloy _nIn the present invention, individual take its sensing scope as 40cm～100cm/ for the number of acoustic emission transducer 6 required settings.The information that acoustic emission transducer 6 is collected adopts the set formal representation to be: demblee form information S _n={ e _S, A _S, C _S, R _S, D _S, e _SExpression energy, A _SExpression measuring amplitude, C _SExpression Ring-down count, R _SExpression rise time and D _SThe expression duration.

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

In the present invention, at acquisition time T _XThe demblee form information table that acoustic emission transducer 6 collects in the section is shown

If first acquisition time T ₁The demblee form information that obtains is designated as

In like manner can get, if second acquisition time T ₂The demblee form information that obtains is designated as ${\mathrm{<figure-callout\; id="2"\; label="S\; n\; T"\; filenames="HDA00002211417600011.png"\; state="\{\{state\}\}">S</figure-callout>}}_{n{T}_{}}^{{}_2}=\{e_S,A_S,C_S{,R}_S,D_S\}.$

(2) prime amplifier 5

The demblee form information that prime amplifier 5 is used for receiving Become the demblee form amplification message after amplifying 40dB

(3) Acoustic radiating instrument 4

Acoustic radiating instrument 4 is in the demblee form amplification message to receiving

After the A/D conversion, become digital demblee form information

Export to AZ31 Fatigue of Magnesium Alloys damage check unit 1, certainly have A/D converter in the Acoustic radiating instrument 4.De _SRepresentative digit formula energy, dA _SRepresentative digit formula measuring amplitude, dC _SRepresentative digit formula Ring-down count, dR _SRepresentative digit formula rise time and dD _SThe representative digit formula duration.

(4) data filtering processing module 11A

In the present invention, the digital demblee form information of data filtering processing module 11A to receiving

Carry out parametric filtering, namely filter electromagnetic noise and neighbourhood noise after, purifying obtains the fatigue damage preliminary information E0 refers to digital energy de _SEnergy behind parametric filtering (being called for short the parametric filtering energy), A ₀Refer to digital measurement amplitude dA _SMeasuring amplitude behind parametric filtering (being called for short the parametric filtering amplitude), C ₀Refer to digital Ring-down count dC _SRing-down count behind parametric filtering (being called for short the parametric filtering Ring-down count), R ₀Refer to digital rise time dR _SRise time behind parametric filtering (being called for short the parametric filtering rise time), D ₀Refer to digital duration dD _SDuration behind parametric filtering (being called for short the parametric filtering duration).

(5) waveform filtering processing module 11B

In the present invention, the fatigue damage preliminary information of waveform filtering processing module 11B to receiving

Carry out waveform filtering, obtain fatigue damage information

E refers to parametric filtering energy e ₀Through the filtered energy of waveform (being called for short waveform filtering energy), A refers to the parametric filtering amplitude A ₀Through the filtered amplitude of waveform (being called for short the waveform filtered amplitude), C refers to parametric filtering Ring-down count C ₀Through the filtered Ring-down count of waveform (referred to as waveform filtering Ring-down count), R refers to parametric filtering rise time R ₀Through the filtered rise time of waveform (referred to as the waveform filtering rise time), D refers to parametric filtering duration D ₀Through the filtered duration of waveform (being called for short the waveform filtering duration).

In the present invention, fatigue damage information

First aspect is used for the first injury tolerance Sign module 12C and carries out the RBF neural metwork training, obtains the first injury tolerance sign model; Second aspect is used for the second injury tolerance Sign module 12D and carries out the RBF neural metwork training, obtains the second injury tolerance sign model; The third aspect is used for the first data space 12A and carries out projection process, obtains the first score matrix; Fourth aspect is used for the second data space 12B and carries out projection process, obtains the second score matrix.For fatigue damage information

The RBF neural metwork training to carry out before the first data space 12A and the second data space 12B.

(6) first data space 12A

The fatigue damage information of the first data space 12A first aspect to receiving

Carry out place acquisition time T _XAccumulated process in the section obtains accumulating rear fatigue damage information ${\mathrm{ACf}}_{}^{}$ ${}_{11B}^{T_X}=(e,A,C,R,D);$

The first data space 12A second aspect is to accumulating rear fatigue damage information ${\mathrm{ACf}}_{}^{}$ ${}_{11B}^{T_X}=(e,A,C,R,D)$ According to Normalized Relation ${f_{11B}}^{\&prime;}=\frac{{\mathrm{ACf}}_{11B}^{T_X}-\stackrel{\&OverBar;}{{\mathrm{ACf}}_{11B}^{T_X}}}{{\mathrm{\&sigma;}}_{{\mathrm{<figure-callout\; id="11B"\; label="ACf"\; filenames="HDA00002211417600021.png"\; state="\{\{state\}\}">ACf</figure-callout>}}_{11B}^{T_X}}}$ Process, obtain normalization accumulation of fatigue damage information f _11B'; The mean value of fatigue damage information

The standard deviation of expression accumulation of fatigue damage information;

The first data space 12A third aspect is to normalization accumulation of fatigue damage information f _11B' according to the first projection relation f _12A=f _11B' * P _aCarry out projection, obtain the first score matrix f _12A=(t _A1, t _A2, t _A3, t _A4, t _A5);

P _aThe eigenmatrix that represents the first data space;

f _12ARepresent the score matrix of 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 _A5The score vector that represents sample to be diagnosed 5 dimensions in the first data space; t _A1The score vector that represents sample to be diagnosed the 1st dimension in the first data space; t _A2The score vector that represents sample to be diagnosed the 2nd dimension in the first data space; t _A3The score vector that represents sample to be diagnosed the 3rd dimension in the first data space; t _A4The score vector that represents sample to be diagnosed the 4th dimension in the first data space; t _A5The score vector that represents sample to be diagnosed the 5th dimension in the first data space.

(7) second data space 12B

The fatigue damage information of the second data space 12B first aspect to receiving

Carry out place acquisition time T _XAccumulated process in the section obtains accumulating rear fatigue damage information ${\mathrm{ACf}}_{}^{}$ ${}_{11B}^{T_X}=(e,A,C,R,D);$

The second data space 12B second aspect is to accumulating rear fatigue damage information ${\mathrm{ACf}}_{}^{}$ ${}_{11B}^{T_X}=(e,A,C,R,D)$ According to Normalized Relation ${f_{11B}}^{\&prime;}=\frac{{\mathrm{ACf}}_{11B}^{T_X}-\stackrel{\&OverBar;}{{\mathrm{ACf}}_{11B}^{T_X}}}{{\mathrm{\&sigma;}}_{{\mathrm{<figure-callout\; id="11B"\; label="ACf"\; filenames="HDA00002211417600021.png"\; state="\{\{state\}\}">ACf</figure-callout>}}_{11B}^{T_X}}}$ Process, obtain normalization accumulation of fatigue damage information f _11B'; The mean value of fatigue damage information

The standard deviation of expression accumulation of fatigue damage information;

The second data space 12B third aspect is to normalization accumulation of fatigue damage information f _11B' according to the second projection relation f _12B=f _11B' * P _bCarry out projection, obtain the second score matrix f _12B=(t _B1, t _B2, t _B3, t _B4, t _B5);

P _bThe eigenmatrix that represents the second data space;

f _12BRepresent the score matrix of sample 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 _B5The score vector that represents sample to be diagnosed 5 dimensions in the second data space; t _B1The score vector that represents sample to be diagnosed the 1st dimension in the second data space; t _B2The score vector that represents sample to be diagnosed the 2nd dimension in the second data space; t _B3The score vector that represents sample to be diagnosed the 3rd dimension in the second data space; t _B4The score vector that represents sample to be diagnosed the 4th dimension in the second data space; t _B5The score vector that represents sample to be diagnosed the 5th dimension in the second data space.

(8) first injury tolerance Sign module 12C

In the present invention, fatigue damage recognition system Main Function is that the fatigue crack that identifies the AZ31 magnesium alloy is stablized extension phase and fatigue crack unstable propagation stage, so the damage data type has two classes among the present invention, i.e. crackle stability types ST and bursting type UT; Described crackle stability types ST refers to that fatigue crack stablizes the damage information of extension phase; Described bursting type UT refers to the damage information in fatigue crack unstable propagation stage.

Obtaining of two kinds of type of impairment RBF train samples: when only being in the Crack Fatigue Stable Crack Growth during stage at labour AZ31 magnesium alloy, with fatigue damage information

Be designated as crackle stability types fatigue damage information

When only being in the Crack Fatigue instable growth of crack during stage at labour AZ31 magnesium alloy, with fatigue damage information

Be designated as bursting type fatigue damage information

The crackle stability types fatigue damage information of the first injury tolerance Sign module 12C first aspect to receiving Carry out place acquisition time T _XAccumulated process in the section obtains accumulating rear fatigue damage information ${\mathrm{ACf}}_{\mathrm{ST}}^{T_X}=(e,A,C,R,D);$

The first injury tolerance Sign module 12C second aspect is to accumulating rear fatigue damage information ${\mathrm{ACf}}_{\mathrm{ST}}^{T_X}=(e,A,C,R,D)$ According to Normalized Relation ${f_{\mathrm{ST}}}^{\&prime;}=\frac{{\mathrm{ACf}}_{\mathrm{ST}}^{T_X}-\stackrel{\&OverBar;}{{\mathrm{ACf}}_{\mathrm{ST}}^{T_X}}}{{\mathrm{\&sigma;}}_{{\mathrm{ACf}}_{\mathrm{ST}}^{T_X}}}$ Process, obtain crackle stability types normalization accumulation of fatigue damage information f _ST'; The mean value of crackle stability types accumulation of fatigue damage information

The standard deviation of expression crackle stability types accumulation of fatigue damage information;

The first injury tolerance Sign module 12C third aspect is to crackle stability types normalization accumulation of fatigue damage information f _ST' carry out principal component analysis (PCA), obtain the eigenmatrix P of the first data space _a=(p _A1, p _A2, p _A3, p _A4, p _A5);

p _A1, p _A2, p _A3, p _A4, p _A5The proper vector that represents 5 dimensions in the first data space; p _A1The proper vector of the 1st dimension in the first data space; p _A2The proper vector of the 2nd dimension in the first data space; p _A3The proper vector of the 3rd dimension in the first data space; p _A4The proper vector of the 4th dimension in the first data space; p _A5The proper vector of the 5th dimension in the first data space;

In the present invention, the fatigue damage recognition system is stablized extension phase and fatigue crack unstable propagation during the stage at the fatigue crack of identification AZ31 magnesium alloy, the eigenmatrix P of this first data space _aNeed citation to the first data space 12A, to carry out projection process.

The first injury tolerance Sign module 12C fourth aspect is to normalization accumulation of fatigue damage information f _ST' according to the 3rd projection relation Pf _12C=f _ST' * P _aCarry out projection, obtain the 3rd score matrix Pf _12C=(Pt _A1, Pt _A2, Pt _A3, Pt _A4, Pt _A5);

Pt _A1, Pt _A2, Pt _A3, Pt _A4, Pt _A5The score vector of training sample 5 dimensions in the first data space of expression crackle stability types; Pt _A1The score vector of training sample the 1st dimension in the first data space of expression crackle stability types; Pt _A2The score vector of training sample the 2nd dimension in the first data space of expression crackle stability types; Pt _A3The score vector of training sample the 3rd dimension in the first data space of expression crackle stability types; Pt _A4The score vector of training sample the 4th dimension in the first data space of expression crackle stability types; Pt _A5The score vector of training sample the 5th dimension in the first data space of expression crackle stability types.

The first injury tolerance Sign module 12C the 5th aspect is with the 3rd score matrix Pf _12A=(Pt _A1, Pt _A2, Pt _A3, Pt _A4, Pt _A5) as the input layer information of RBF neural network, and set the output layer information of RBF neural network $O_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right],$ Beginning RBF neural metwork training obtains the first injury tolerance sign model M _a

In the present invention, the output layer information of RBF neural network $O_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]$ In [1 0] expression training sample be the identity code that fatigue crack stablize extended mode, [0 1] represent that training sample is the identity code of fatigue crack unstable propagation state.

The first injury tolerance Sign module 12C the 6th aspect adopts M _aTo the first score matrix f _12A=(t _A1, t _A2, t _A3, t _A4, t _A5) process, obtain the first injury tolerance parameter f _12C=[f _{ID, a}(A ₁) f _{ID, a}(A ₂)].

In the present invention, the first f of injury tolerance Sign module 12C to receiving _12A=(t _A1, t _A2, t _A3, t _A4, t _A5) carry out obtaining the first injury tolerance parameter f after the RBF Learning Algorithm processes _12C=[f _{ID, a}(A ₁) f _{ID, a}(A ₂)]; F wherein _{ID, a}(A ₁) the expression code is the neural network output of diagnostic sample the 1st node in the first data space of the transducer of ID; f _{ID, a}(A ₂) the expression code is the neural network output of diagnostic sample the 2nd node in the first data space of the transducer of ID; ID represents the code of acoustic emission transducer, and a is the code of the first data space.

(9) second injury tolerance Sign module 12D

The bursting type fatigue damage information of the second injury tolerance Sign module 12D first aspect to receiving

Carry out place acquisition time T _XAccumulated process in the section obtains accumulating rear fatigue damage information ${\mathrm{ACf}}_{\mathrm{UT}}^{T_X}=(e,A,C,R,D);$

The first injury tolerance Sign module 12C second aspect is to accumulating rear fatigue damage information ${\mathrm{ACf}}_{\mathrm{UT}}^{T_X}=(e,A,C,R,D)$ According to Normalized Relation ${f_{\mathrm{UT}}}^{\&prime;}=\frac{{\mathrm{ACf}}_{\mathrm{UT}}^{T_X}-\stackrel{\&OverBar;}{{\mathrm{ACf}}_{\mathrm{UT}}^{T_X}}}{{\mathrm{\&sigma;}}_{{\mathrm{ACf}}_{\mathrm{UT}}^{T_X}}}$ Process, obtain bursting type normalization accumulation of fatigue damage information f _UT'; The mean value of fatigue damage information

The standard deviation of expression bursting type accumulation of fatigue damage information;

The second injury tolerance Sign module 12D third aspect is to bursting type normalization accumulation of fatigue damage information f _UT' carry out principal component analysis (PCA), obtain the eigenmatrix P of the second data space _b=(p _B1, p _B2, p _B3, p _B4, p _B5);

p _B1, p _B2, p _B3, p _B4, p _B5The proper vector that represents 5 dimensions in the second data space; p _B1The proper vector of the 1st dimension in the second data space; p _B2The proper vector of the 2nd dimension in the second data space; p _B3The proper vector of the 3rd dimension in the second data space; p _B4The proper vector of the 4th dimension in the second data space; p _B5The proper vector of the 5th dimension in the second data space;

In the present invention, the fatigue damage recognition system is stablized extension phase and fatigue crack unstable propagation during the stage at the fatigue crack of identification AZ31 magnesium alloy, the eigenmatrix P of this second data space _bNeed citation to the second data space 12B, to carry out projection process.

The second injury tolerance Sign module 12D fourth aspect is to bursting type normalization accumulation of fatigue damage information f _UT' according to the 4th projection relation Pf _12D=f _UT' * P _bCarry out projection, obtain the 4th score matrix Pf _12D=(Pt _B1, Pt _B2, Pt _B3, Pt _B4, Pt _B5);

Pt _B1, Pt _B2, Pt _B3, Pt _B4, Pt _B5The score vector of training sample 5 dimensions in the second data space of expression bursting type; Pt _B1The score vector of training sample the 1st dimension in the second data space of expression bursting type; Pt _B2The score vector of training sample the 2nd dimension in the second data space of expression bursting type; Pt _B3The score vector of training sample the 3rd dimension in the second data space of expression bursting type; Pt _B4The score vector of training sample the 4th dimension in the second data space of expression bursting type; Pt _B5The score vector of training sample the 5th dimension in the second data space of expression bursting type.

The second injury tolerance Sign module 12D the 5th aspect is with the 4th score matrix Pf _12D=(Pt _B1, Pt _B2, Pt _B3, Pt _B4, Pt _B5) as the input layer information of RBF neural network, and set the output layer information of RBF neural network $O_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right],$ Beginning RBF neural metwork training obtains the first injury tolerance sign model M _b

In the present invention, the output layer information of RBF neural network $O_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]$ In [1 0] expression training sample be the identity code that fatigue crack stablize extended mode, [0 1] represent that training sample is the identity code of fatigue crack unstable propagation state.

The second injury tolerance Sign module 12D the 6th aspect adopts M _bTo the second score matrix f _12B=(t _B1, t _B2, t _B3, t _B4, t _B5) process, obtain the second injury tolerance parameter f _12D=[f _{ID, b}(A ₁) f _{ID, b}(A ₂)].

In the present invention, the first f of injury tolerance Sign module 12C to receiving _12B=(t _B1, t _B2, t _B3, t _B4, t _B5) carry out obtaining the second injury tolerance parameter f after the RBF Learning Algorithm processes _12D=[f _{ID, b}(A ₁) f _{ID, b}(A ₂)]; F wherein _{ID, b}(A ₁) the expression code is the neural network output of diagnostic sample the 1st node in the second data space of the transducer of ID; f _{ID, b}(A ₂) the expression code is the neural network output of diagnostic sample the 2nd node in the second data space of the transducer of ID; ID represents the code of acoustic emission transducer, and b is the code of the second data space.

(10) D-S Evidence Combination Methods module 12E

In the present invention, the f of D-S Evidence Combination Methods module 12E first aspect to receiving _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 related coefficient $Q_{\mathrm{ID},i}\left(A_j\right)=\left[\begin{array}{cc}\frac{|f_{\mathrm{ID},a}\left(A_1\right)-1|}{|f_{\mathrm{ID},a}\left(A_1\right)-1|+|f_{\mathrm{ID},a}\left(A_2\right)-1|}& \frac{|f_{\mathrm{ID},a}\left(A_2\right)-1|}{|f_{\mathrm{ID},a}\left(A_1\right)-1|+|f_{\mathrm{ID},a}\left(A_2\right)-1|}\\ \frac{|f_{\mathrm{ID},b}\left(A_1\right)-1|}{|f_{\mathrm{ID},b}\left(A_1\right)-1|+|f_{\mathrm{ID},b}\left(A_2\right)-1|}& \frac{|f_{\mathrm{ID},b}\left(A_2\right)-1|}{|f_{\mathrm{ID},b}\left(A_1\right)-1|+|f_{\mathrm{ID},b}\left(A_2\right)-1|}\end{array}\right]$ Process, obtain sample to be diagnosed in the first data space and the second data space with the related coefficient Q of each node _{ID, i}(A _j);

Q _{ID, i}(A _j) the expression code be ID transducer sample to be diagnosed code be in the data space of i (i=a, b) with the related coefficient of node j (j=1,2).

D-S Evidence Combination Methods module 12E second aspect is by the basic probability assignment function $m_{\mathrm{ID},i}\left(A_j\right)=\frac{Q_{\mathrm{ID},i}\left(A_j\right)}{\underset{j=1}{\overset{N_c}{\mathrm{\&Sigma;}}}{Q}_{\mathrm{ID},i}\left(A_j\right)+N_S\&times;(1-R_i)\&times;(1-{\mathrm{\&alpha;}}_i{\mathrm{\&beta;}}_i)}$ Each node j (j=1,2) is carried out basic probability assignment, obtain sample to be diagnosed in the first data space and the second data space with the basic probability assignment value m of each node _{ID, i}(A _j).

α wherein _i=max{Q _{ID, i}(A _j) represent sample to be diagnosed code be in the data space of i (i=a, b) with the maximum correlation coefficient of node;

The expression code be in the data space of i (i=a, b) with the apportioning cost of node related coefficient;

The expression code is the safety factor of the data space of i (i=a, b); N in the present invention _cBe node number, N _sFor the data space number, be 2; m _{ID, i}(A _j) the expression code be the sample to be diagnosed that receives of the transducer of ID code be in the data space of i (i=a, b) with the basic probability assignment value of node j (j=1,2).

The D-S Evidence Combination Methods module 12E third aspect is used D-S evidence the first syntagmatic

Treat diagnostic sample and in two data spaces, carry out data fusion with the basic probability assignment value of each node, obtain one-level data fusion value m _ID(B _j).

N _SBe the data space number; m _ID(B _j) represent that code is that the transducer of ID is in the one-level data fusion value of node j (j=1,2);

(11) second level data fusion module 13

In the present invention, secondary data merges as a result m of 13 pairs of data fusion that receive _ID(B _j) according to D-S evidence the second syntagmatic $m\left(C_j\right)=\frac{\underset{\&cap;B_j=C_j}{\mathrm{\&Sigma;}}\underset{j\&Element;n}{\mathrm{\&Pi;}}{m}_{\mathrm{ID}}\left(B_j\right)}{1-\underset{\&cap;B_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}\underset{j\&Element;n}{\mathrm{\&Pi;}}{m}_{\mathrm{ID}}\left(B_j\right)}$ Process, obtain secondary data fusion results m (C _j), C _jNode identification in the expression secondary data Fusion Module.

N is the number of the acoustic emission transducer 6 of setting; M (C _j) in the expression fatigue damage recognition system all acoustic emission transducers 6 with joint form, the secondary data fusion value of locating at node j (j=1,2).

(12) damage grade evaluation unit 2

In the present invention, the secondary data fusion results m (C that receives of 2 pairs of damage grade evaluation unit _j) carry out ranking, obtain at the residing Fatigue Damage States of labour AZ31 magnesium alloy.

Ranking condition: as node maximal value w=m (C ₁) time, Fatigue Damage States is: be under the stable expansion of the fatigue crack faulted condition at labour AZ31 magnesium alloy; As node maximal value w=m (C ₂) time, Fatigue Damage States is: be under the fatigue crack unstable propagation faulted condition at labour AZ31 magnesium alloy;

Described node maximal value w=max (m (C ₁), m (C ₂)), m (C ₁) all transducers of expression fatigue damage recognition system are in node 1(fatigue crack steady state (SS)) the secondary data fusion value located, m (C ₂) all transducers of expression fatigue damage recognition system are at node 2(fatigue crack instability status) the secondary data fusion value located.W represents that all transducers of fatigue damage recognition system are in the maximal value of the secondary data fusion value of two nodes.

(13) alarm unit 3

In the present invention, work as w=m (C in the alarm module 3 ₁) time do not report to the police, as w=m (C ₂) time start alerting signal.Alarm unit 3 adopts the prompt tone warning output such as forms such as loudspeaker, loudspeakers.

Embodiment 1:Certain automotive hub is carried out acoustic emission detection.

The used AZ31 magnesium alloy composition of wheel hub sees Table 1:

Table 1AZ31 magnesium alloy component content

Detection has with equipment: (A) two the narrow frequency acoustic emission transducer of R15 (CZ series of PAC company, response frequency is 100kHz～400kHz, centre frequency 150kHz) and two wideband transducers (the WD series of PAC company, response frequency are 20kHz～1MHz).

(B) 2/4/6 type prime amplifier of four PAC companies.

(C) Acoustic radiating instrument is U.S. PAC company full digital 16 passage DiSP acoustic emission systems.Threshold value 40dB when Acoustic radiating instrument detects, acoustic emission peak value definition time PDT is 300 μ s, and acoustic emission bump limiting time HDT is 600 μ s, and acoustic emission bump blocking time HLT is 1000 μ s.

At the AZ31 magnesium alloy at sampling time T _XFour acoustic emission information process fatigue damage recognition system identifications that transducer receives in the section, the result is as shown in table 2 for the fatigue damage detecting unit.

The basic probability assignment result of acoustic emission information in the fatigue damage detecting unit that four transducers of table 2A receive

Acoustic emission information one-level data fusion result in the fatigue damage monitoring means that four transducers of table 2B receive

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

Acoustic emission information secondary data fusion results in the fatigue damage monitoring means that four transducers of table 2C receive

	m(C 1)	m(C 2)	Uncertainty
				The secondary fusion results	1.0000	0.0000	0.0000

Fusion results 2 parsings through the damage grade evaluation unit among the table 2C, can obtain the node maximal value is m (C ₁), so being in fatigue crack, this AZ31 magnesium alloy stablizes extension phase, need not to report to the police.

This shows that data fusion can reduce the local uncertainty that merges of neural network, so that the confidence level of diagnosis decision-making improves significantly.Simultaneously, improve the fault-tolerant ability of diagnostic system, and can satisfy the requirement of steel construction complication system damage.

The present invention adopts principal component analysis (PCA), the method that neural network and two-stage data fusion combine is identified AZ31 Fatigue of Magnesium Alloys faulted condition, has set up AZ31 Fatigue of Magnesium Alloys faulted condition identifying and diagnosing system: at first with neural network model each acoustic emission transducer information is carried out local diagnosis under the different damage data space of principal component model structure; Then with the elementary probability value of the structure of the neural network Output rusults in same transducer one-level data fusion module, carry out the one-level data fusion; One-level data fusion result with a plurality of sensors carries out the secondary data fusion at last, by the damage identification module Fatigue Damage States is diagnosed again.Utilize this model, can identify, diagnose the faulted condition in the AZ31 Fatigue of Magnesium Alloys process, and then provide foundation to its reliability service.

Claims (6)

1. recognition system of the AZ31 magnesium alloy being carried out Fatigue Damage States based on PCA and TDF, it is characterized in that: this system includes a plurality of acoustic emission transducers (6), multichannel prime amplifier (5), an Acoustic radiating instrument (4), it is characterized in that: also include a 16Mn steel force-bearing part fatigue damage detecting unit (1);

16Mn steel force-bearing part fatigue damage detecting unit (1) includes filtering module (11) and one-level data fusion module (12), secondary data Fusion Module (13), wherein, filtering module (11) has data filtering processing module (11A) and waveform filtering processing module (11B), one-level data fusion module (12) has the first data space (12A), the second data space (12B), the first injury tolerance Sign module (12C), the second injury tolerance Sign module (12D), D-S Evidence Combination Methods module (12E);

Acoustic emission transducer (6) is used for being captured in labour 16Mn steel force-bearing part at acquisition time T _XDemblee form information burst type information in the section

Prime amplifier (5) is used for the demblee form information to receiving

Become the demblee form amplification message after amplifying 40dB

Acoustic radiating instrument (4) is used for the demblee form amplification message to receiving

After the A/D conversion, become digital demblee form information

Export to 16Mn steel force-bearing part fatigue damage detecting unit (1);

The digital demblee form information of data filtering processing module (11A) to receiving

Carry out parametric filtering, namely filter electromagnetic noise and neighbourhood noise after, purifying obtains the fatigue damage preliminary information The fatigue damage preliminary information of waveform filtering processing module (11B) to receiving

Carry out waveform filtering, obtain fatigue damage information

Transducer is received the fatigue damage information of 16Mn steel force-bearing part Carry out acquisition time T _XAccumulated process in the section, then normalization obtains normalization accumulation of fatigue damage information f _11B', with f _11B' obtain separately score matrix f in the lower projection of the first data space (12A) and the second data space (12B) respectively _12A=(t _A1, t _A2, t _A3, t _A4, t _A5) and f _12B=(t _B1, t _B2, t _B3, t _B4, t _B5), score matrix f _12A=(t _A1, t _A2, t _A3, t _A4, t _A5) process the neural network output f obtain in the first data space through the first injury tolerance Sign module (12C) _12C=[f _{ID, a}(A ₁) f _{ID, a}(A ₂)], score matrix f _12B=(t _B1, t _B2, t _B3, t _B4, t _B5) process the neural network output f that obtains in the second data space through the second injury tolerance Sign module (12D) _12D=[f _{ID, b}(A ₁) f _{ID, b}(A ₂)]; With f _12CAnd f _12DCarry out the related coefficient assignment of each node, the f after the assignment _12CAnd f _12DCarry out basic probability assignment, then carry out D-S Evidence Combination Methods module (12E) and process the data fusion m as a result obtain single transducer _ID(B _j);

Secondary data merges (13) to the data fusion that receives m as a result _ID(B _j) according to D-S evidence the second syntagmatic Process, obtain secondary data fusion results m (C _j), C _jNode identification in the expression secondary data Fusion Module;

The secondary data fusion results m (C of damage grade evaluation unit (2) to receiving _j) carry out ranking, obtain at the residing Fatigue Damage States of labour AZ31 magnesium alloy;

Ranking condition: as node maximal value w=m (C ₁) time, Fatigue Damage States is: be under the stable expansion of the fatigue crack faulted condition at labour AZ31 magnesium alloy; As node maximal value w=m (C ₂) time, Fatigue Damage States is: be under the fatigue crack unstable propagation faulted condition at labour AZ31 magnesium alloy;

After alarm unit (3) receives the fatigue damage identifying information D of damage grade evaluation unit (2) output, as w=m (C ₁) time do not report to the police, as w=m (C ₂) time starts alerting signal and report to the police.

2. according to claim 1ly based on PCA and TDF the AZ31 magnesium alloy is carried out the recognition system of Fatigue Damage States, it is characterized in that:

The fatigue damage information of the first data space (12A) first aspect to receiving

Carry out place acquisition time T _XAccumulated process in the section obtains accumulating rear fatigue damage information ${\mathrm{ACf}}_{11B}^{T_X}=(e,A,C,R,D);$

The first data space (12A) second aspect is to accumulating rear fatigue damage information ${\mathrm{ACf}}_{11B}^{T_X}=(e,A,C,R,D)$ According to Normalized Relation ${f_{11B}}^{\&prime;}=\frac{{\mathrm{ACf}}_{11B}^{T_X}-\stackrel{\&OverBar;}{{\mathrm{ACf}}_{11B}^{T_X}}}{{\mathrm{\&sigma;}}_{{\mathrm{ACf}}_{11B}^{T_X}}}$ Process, obtain normalization accumulation of fatigue damage information f _11B'; The mean value of fatigue damage information

The standard deviation of expression accumulation of fatigue damage information;

The first data space (12A) third aspect is to normalization accumulation of fatigue damage information f _11B' according to the first projection relation f _12A=f _11B' * P _aCarry out projection, obtain the first score matrix f _12A=(t _A1, t _A2, t _A3, t _A4, t _A5).

3. according to claim 1ly based on PCA and TDF the AZ31 magnesium alloy is carried out the recognition system of Fatigue Damage States, it is characterized in that:

The fatigue damage information of the second data space (12B) first aspect to receiving

Carry out place acquisition time T _XAccumulated process in the section obtains accumulating rear fatigue damage information ${\mathrm{ACf}}_{11B}^{T_X}=(e,A,C,R,D);$

The second data space (12B) second aspect is to accumulating rear fatigue damage information ${\mathrm{ACf}}_{11B}^{T_X}=(e,A,C,R,D)$ According to Normalized Relation ${f_{11B}}^{\&prime;}=\frac{{\mathrm{ACf}}_{11B}^{T_X}-\stackrel{\&OverBar;}{{\mathrm{ACf}}_{11B}^{T_X}}}{{\mathrm{\&sigma;}}_{{\mathrm{ACf}}_{11B}^{T_X}}}$ Process, obtain normalization accumulation of fatigue damage information f _11B'; The mean value of fatigue damage information

The standard deviation of expression accumulation of fatigue damage information;

The second data space (12B) third aspect is to normalization accumulation of fatigue damage information f _11B' according to the second projection relation f _12B=f _11B' * P _bCarry out projection, obtain the second score matrix f _12B=(t _B1, t _B2, t _B3, t _B4, t _B5).

4. according to claim 1ly based on PCA and TDF the AZ31 magnesium alloy is carried out the recognition system of Fatigue Damage States, it is characterized in that:

The crackle stability types fatigue damage information of the first injury tolerance Sign module (12C) first aspect to receiving

Carry out place acquisition time T _XAccumulated process in the section obtains accumulating rear fatigue damage information ${\mathrm{ACf}}_{\mathrm{ST}}^{T_X}=(e,A,C,R,D);$

The first injury tolerance Sign module (12C) second aspect is to accumulating rear fatigue damage information ${\mathrm{ACf}}_{\mathrm{ST}}^{T_X}=(e,A,C,R,D)$ According to Normalized Relation ${f_{\mathrm{ST}}}^{\&prime;}=\frac{{\mathrm{ACf}}_{\mathrm{ST}}^{T_X}-\stackrel{\&OverBar;}{{\mathrm{ACf}}_{\mathrm{ST}}^{T_X}}}{{\mathrm{\&sigma;}}_{{\mathrm{ACf}}_{\mathrm{ST}}^{T_X}}}$ Process, obtain crackle stability types normalization accumulation of fatigue damage information f _ST'; The mean value of crackle stability types accumulation of fatigue damage information

The standard deviation of expression crackle stability types accumulation of fatigue damage information;

First injury tolerance Sign module (12C) third aspect is to crackle stability types normalization accumulation of fatigue damage information f _ST' carry out principal component analysis (PCA), obtain the eigenmatrix P of the first data space _a=(p _A1, p _A2, p _A3, p _A4, p _A5);

The first injury tolerance Sign module (12C) fourth aspect is to normalization accumulation of fatigue damage information f _ST' according to the 3rd projection relation Pf _12C=f _ST' * P _aCarry out projection, obtain the 3rd score matrix Pf _12C=(Pt _A1, Pt _A2, Pt _A3, Pt _A4, Pt _A5);

The first injury tolerance Sign module (12C) the 5th aspect is with the 3rd score matrix Pf _12A=(Pt _A1, Pt _A2, Pt _A3, Pt _A4, Pt _A5) as the input layer information of RBF neural network, and set the output layer information of RBF neural network $O_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right],$ Beginning RBF neural metwork training obtains the first injury tolerance sign model M _a

The output layer information of RBF neural network $O_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]$ In [1 0] expression training sample be the identity code that fatigue crack stablize extended mode, [0 1] represent that training sample is the identity code of fatigue crack unstable propagation state.

The first injury tolerance Sign module (12C) the 6th aspect adopts M _aTo the first score matrix f _12A=(t _A1, t _A2, t _A3, t _A4, t _A5) process, obtain the first injury tolerance parameter f _12C=[f _{ID, a}(A ₁) f _{ID, a}(A ₂)].

5. according to claim 1ly based on PCA and TDF the AZ31 magnesium alloy is carried out the recognition system of Fatigue Damage States, it is characterized in that:

The bursting type fatigue damage information of the second injury tolerance Sign module (12D) first aspect to receiving

Carry out place acquisition time T _XAccumulated process in the section obtains accumulating rear fatigue damage information ${\mathrm{ACf}}_{\mathrm{UT}}^{T_X}=(e,A,C,R,D);$

The first injury tolerance Sign module (12D) second aspect is to accumulating rear fatigue damage information ${\mathrm{ACf}}_{\mathrm{UT}}^{T_X}=(e,A,C,R,D)$ According to Normalized Relation ${f_{\mathrm{UT}}}^{\&prime;}=\frac{{\mathrm{ACf}}_{\mathrm{UT}}^{T_X}-\stackrel{\&OverBar;}{{\mathrm{ACf}}_{\mathrm{UT}}^{T_X}}}{{\mathrm{\&sigma;}}_{{\mathrm{ACf}}_{\mathrm{UT}}^{T_X}}}$ Process, obtain bursting type normalization accumulation of fatigue damage information f _UT'; The mean value of fatigue damage information

The standard deviation of expression bursting type accumulation of fatigue damage information;

Second injury tolerance Sign module (12D) third aspect is to bursting type normalization accumulation of fatigue damage information f _UT' carry out principal component analysis (PCA), obtain the eigenmatrix P of the second data space _b=(p _B1, p _B2, p _B3, p _B4, p _B5);

The second injury tolerance Sign module (12D) fourth aspect is to bursting type normalization accumulation of fatigue damage information f _UT' according to the 4th projection relation Pf _12D=f _UT' * P _bCarry out projection, obtain the 4th score matrix Pf _12D=(Pt _B1, Pt _B2, Pt _B3, Pt _B4, Pt _B5);

The second injury tolerance Sign module (12D) the 5th aspect is with the 4th score matrix Pf _12D=(Pt _B1, Pt _B2, Pt _B3, Pt _B4, Pt _B5) as the input layer information of RBF neural network, and set the output layer information of RBF neural network $O_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right],$ Beginning RBF neural metwork training obtains the first injury tolerance sign model M _b

The output layer information of RBF neural network $O_{\mathrm{ID},\mathrm{tr}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]$ In [1 0] expression training sample be the identity code that fatigue crack stablize extended mode, [0 1] represent that training sample is the identity code of fatigue crack unstable propagation state.

The second injury tolerance Sign module (12D) the 6th aspect adopts M _bTo the second score matrix f _12B=(t _B1, t _B2, t _B3, t _B4, t _B5) process, obtain the second injury tolerance parameter f _12D=[f _{ID, b}(A ₁) f _{ID, b}(A ₂)].

6. according to claim 1ly based on PCA and TDF the AZ31 magnesium alloy is carried out the recognition system of Fatigue Damage States, it is characterized in that:

The f of D-S Evidence Combination Methods module (12E) first aspect to receiving _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 related coefficient $Q_{\mathrm{ID},i}\left(A_j\right)=\left[\begin{array}{cc}\frac{|f_{\mathrm{ID},a}\left(A_1\right)-1|}{|f_{\mathrm{ID},a}\left(A_1\right)-1|+|f_{\mathrm{ID},a}\left(A_2\right)-1|}& \frac{|f_{\mathrm{ID},a}\left(A_2\right)-1|}{|f_{\mathrm{ID},a}\left(A_1\right)-1|+|f_{\mathrm{ID},a}\left(A_2\right)-1|}\\ \frac{|f_{\mathrm{ID},b}\left(A_1\right)-1|}{|f_{\mathrm{ID},b}\left(A_1\right)-1|+|f_{\mathrm{ID},b}\left(A_2\right)-1|}& \frac{|f_{\mathrm{ID},b}\left(A_2\right)-1|}{|f_{\mathrm{ID},b}\left(A_1\right)-1|+|f_{\mathrm{ID},b}\left(A_2\right)-1|}\end{array}\right]$ Process, obtain sample to be diagnosed in the first data space and the second data space with the related coefficient Q of each node _{ID, i}(A _j);

D-S Evidence Combination Methods module (12E) second aspect is by the basic probability assignment function $m_{\mathrm{ID},i}\left(A_j\right)=\frac{Q_{\mathrm{ID},i}\left(A_j\right)}{\underset{j=1}{\overset{N_c}{\mathrm{\&Sigma;}}}{Q}_{\mathrm{ID},i}\left(A_j\right)+N_S\&times;(1-R_i)\&times;(1-{\mathrm{\&alpha;}}_i{\mathrm{\&beta;}}_i)}$ Each node j (j=1,2) is carried out basic probability assignment, obtain sample to be diagnosed in the first data space and the second data space with the basic probability assignment value m of each node _{ID, i}(A _j).

D-S Evidence Combination Methods module (12E) third aspect is used D-S evidence the first syntagmatic

Treat diagnostic sample and in two data spaces, carry out data fusion with the basic probability assignment value of each node, obtain one-level data fusion value m _ID(B _j).

Images