CN105956977A - Grey correlation analysis based health grading evaluation method adopting all observation indexes - Google Patents

Abstract

The invention provides a grey correlation analysis based health grading evaluation method adopting all observation indexes, and belongs to the technical field of fault diagnosis. According to the invention, an all test point based weight improvement method is provided by using a non-healthy state-test correlation model, the grey correlation degrees are calculated according to an acquired percentage weight omega<k> of all observation indexes and a calculated grey correlation coefficient, the grey correlation degrees are sorted so as to form a grey correlation sequence, a factor behavior sequence with the maximum correlation degree in the grey correlation sequence is the optimal factor, and a mode to be inspected at present of the system is located in the factor, so that the health state of the current system can be judged, health grading evaluation is provided, and finally parameter measurement evaluation is provided for the health grading evaluation. The weight set in the invention more conforms to system characteristics, calculation of the grey correlation degree is more reasonable and more accurate, and a result of health grading evaluation is more accurate, so that the system state and system factors resulting in sub-health can be predicted more accurately.

Description

Use the grey correlation analysis health classification evaluation method of full observation index scheme

Technical field

The invention belongs to fault diagnosis technology field, the grey correlation being specific to relate to using full observation index scheme divides The healthy classification evaluation method of analysis.

Background technology

Each system or product high integrity increasingly, modularity in existing different field, function gets more and more, structure and layer Secondary become increasingly complex, wherein have several factors for system properly functioning for be not fatal, once these factors are sent out Raw exception or fault, system can enter " subhealth state " state.If in this state halt system run, may bring through Massive losses in Ji, and cause mission failure；If but not taking correction measure and allow system to continue to run with work, then Likely because this " subhealth state " state causes unforeseen heavy losses and harm.In light of this situation, system is carried out Healthy grading evaluation research has important real value.Therefore, system is carried out healthy grading evaluation research and have important Real value.

Unhealthy status test correlation models is to describe component units unhealthy status in three grades of health status systems The model of the correlation logic relation of (subhealth state or fault) and test, if this component units with test is in system information flows Up to, then they have dependency, otherwise do not have dependency.Similar with tradition correlation models, unhealthy status-test Correlation models (being abbreviated as NH-T correlation models) can also describe with correlation matrix (D matrix), as follows:

Wherein: row matrix U_i=[d_i1d_i2…d_in] give component units U_iWith each test point T_j(j=1,2 ..., n) Between dependency；Rectangular array T_j=[d_1jd_2j…d_mj]^TGive test point T_jWith each component units U_i(i=1,2 ..., m) it Between dependency；In matrix, the implication of cross term is as follows:

Gray system theory is that a kind of data amount is few, the new method of the uncertain problem of poor in information.By right Existing " part " Given information develops effectively, realizes the state to system, the effective mould of essence with this Fit control.Owing to, in the process of analysis and application, gray system theory makes the uncertain content of qualitative analysis more be close to Objective reality, can also make that the definitiveness content of quantitative analysis is more efficient meets subjective experience simultaneously, therefore applies the widest General.

Grey correlation refers to the uncertain association existed between things and things, or refer between system factor and the factor, The uncertain association existed between factor pair principal act.Grey Incidence Analysis utilizes sequence of grey correlation to describe different factor Between the power of relation, size, order, it comes the influence degree between descriptive system difference factor or difference by grey relational grade The factor contribution degree size to system principal act factor.The essential idea of grey relational grade is: with the data of system difference factor The sequence constituted is foundation, with the geometrical relationship between mathematical method research factor or between factor and principal act, i.e. according to song The geometry of line judges exchange premium degree, and then describes the grey relational grade between them.The most most widely used is Deng Shi Related degree model, its key equation calculating the degree of association is:

${\mathrm{\&gamma;}}_{0i}=\mathrm{\&gamma;}(X_0,X_i)=\frac{1}{n}\underset{k=1}{\overset{n}{\&Sigma;}}\mathrm{\&gamma;}\left(x_0\right(k),x_i(k\left)\right)$

γ(X₀,X_i) it is the comparative sequences X of system_iWith reference sequences X₀Grey relational grade, be abbreviated as γ_0i, γ (x₀ (k),x_i(k)) it is X_iWith X₀At the coefficient of association of observation index k, n represents observation index number.

The most this processing mode substantially thinks that health based on grey correlation analysis is divided by n observation index of system It is of equal value that level evaluates the effect played, i.e. weight is the most equal, not different with the difference of observation index k.This processing mode Only all symmetrical about all component units of system at all n observation indexs and be the most rationally suitable in the case of being independent of each other. For General System, its observation index generally also will not deliberately meet this symmetrical structure when choosing.Therefore, this calculating ash The method of the color degree of association has certain limitation, it is impossible to apply exactly in the health evaluating to system.

Summary of the invention

Present invention aim to address the existing circumscribed problem of calculating grey relational grade, system can be more accurately determined Health status, it is provided that a kind of grey correlation analysis health classification evaluation method using full observation index scheme.Due to non-strong Health state-test correlation models can provide each component units and the correlative relationship of each test in system exactly, therefore The D matrix intension utilizing unhealthy status test correlation models can provide weights based on full test point and determine scheme, Can preferably describe the percentage contribution that each test is played in the healthy grading evaluation of system, i.e. weights and more meet system spy Property, and then make healthy grading evaluation result more accurate.

The grey correlation analysis health classification evaluation method using full observation index scheme that the present invention provides, it is achieved step As follows:

Step one: set up the NH-T dependency graphical model of system；

If system factor sum has m, component units number has m ' individual, and observation index sum has n, component units set For { U₁,U₂,…,U_m', the collection of all observation indexs is combined into test point set { T₁,T₂,…,T_n}。

Step 2: released the D matrix of NH-T correlation models by the NH-T dependency graphical model of system；

I-th row jth column element d in D matrix_ijRepresent test point T_jWith component units U_iWhether it is correlated with, works as T_jU can be recorded_i's Represent relevant during non-health information, now d_ijValue is 1, works as T_jU can not be recorded_iNon-health information time represent uncorrelated, this Time d_ijValue is 0.

Step 3: use full observation index scheme to test；By the D matrix of NH-T correlation models, it is calculated kth (k=1,2 ..., n) the percentage ratio weights ω that system health grading evaluation is affected by individual observation index_k；

${\mathrm{\&omega;}}_k=W_k/\underset{j=1}{\overset{n}{\&Sigma;}}{W}_j\&times;100\%;W_k=\frac{1}{{\mathrm{\&alpha;}}_{ck}}\underset{i=1}{\overset{m}{\&Sigma;}}{d}_{ik};{\mathrm{\&alpha;}}_{ck}=C_k/\underset{j=1}{\overset{n}{\&Sigma;}}{C}_j;$

W_kRepresent the observation weights of kth observation index；C_kRepresent every correlative charges sum of kth observation index； α_ckRepresent the relative costs ratio of kth observation index.

Step 4: obtain characteristic behavior sequence and system all of factor behavior sequence；

If the observation data that i-th system factor is on kth observation index are x_i(k), the then row of i-th system factor It is shown as X for sequence table_i=(x_i(1),x_i(2),…,x_i(n))；Acquisition system treats the sight under all observation indexs of the investigation state Survey data, obtain characteristic behavior sequence X₀=(x₀(1),x₀(2),…,x₀(n))；The system that obtains under all factors is in all observations Observation data under index, determine system all of factor behavior sequence X₁,X₂,…X_m。

Step 5: obtain factor behavior sequence and the grey relational grade of characteristic behavior sequence；

To each factor behavior sequence, the observation data calculated in itself and characteristic behavior sequence under same observation index Absolute difference, and obtain

I-th factor behavior sequence X_iWith system features behavior sequence X₀Grey relational grade γ (X₀,X_i) it is:

$\mathrm{\&gamma;}(X_0,X_i)=\underset{k=1}{\overset{n}{\&Sigma;}}{\mathrm{\&omega;}}_k\mathrm{\&gamma;}\left(x_0\right(k),x_i(k\left)\right),i=1,2,\mathrm{...},m;$

γ(x₀(k),x_i(k)) it is i-th factor behavior sequence and characteristic behavior sequence Lycoperdon polymorphum Vitt pass under observation index k Contact number.

Step 6: carry out healthy grading evaluation；

Find and system features behavior sequence X₀The maximum factor behavior sequence of grey relational grade, by this factor behavior sequence The system health state of row correspondence and system factor are as X₀Corresponding system health state and system factor.

Advantages of the present invention with have the active effect that from system structure and testing cost, in conjunction with the survey of observation index Examination ability and testing expense, the weights that system health grading evaluation is affected by Optimal improvements observation index so that weights Determine and the most objective i.e. can preferably describe the percentage contribution that each test is played in the healthy grading evaluation of system comprehensively, I.e. weights more meet system performance, and then make the calculating of grey relational grade the most accurately, healthy grading evaluation result More accurate, and then can relatively accurately prognoses system state and cause the system factor of " subhealth state ".

Accompanying drawing explanation

Fig. 1 is the grey correlation analysis health classification evaluation method flow chart that the weights that the present invention uses improve；

Fig. 2 is a NH-T dependency figure model schematic；

Fig. 3 is the phantom figure of case signal conditioning circuit；

Fig. 4 is the NH-T dependency figure illustraton of model of case signal conditioning circuit.

Detailed description of the invention

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

The present invention utilizes unhealthy status test correlation models (NH-T correlation models) to provide based on full test point Weights improved method, according to the percentage ratio weights ω of obtained whole n observation indexs_k(k=1,2 ..., n) and The grey incidence coefficient obtained, utilizes formulaGrey relational grade γ can be calculated_0i.Ask Go out all m system factor behavior sequence X₁,X₂,…X_mWith system features behavior sequence X₀Grey relational grade, constitutes grey correlation Sequence.In sequence of grey correlation, the factor behavior sequence of degree of association maximum is optimum factor, and the system that is currently treats investigation pattern institute The factor at place, such that it is able to pass judgment on the health status of current system, provides healthy grading evaluation, finally gives healthy grading evaluation Go out parameter metric evaluation.

The present invention uses the weights improved method of full observation index scheme, healthy grading evaluation based on grey correlation analysis The flow process of method is as it is shown in figure 1, illustrate below in conjunction with embodiment.

Step one: set up the NH-T dependency graphical model of system.

(1.1) system thinking index, i.e. observation index are determined, numbered k, k=1,2 ..., n；N represents that observation index is total Number.

(1.2) the whole factor of system is determined, including all possible under health status, sub-health state and malfunction Factor, if system factor numbered i, i=1,2 ..., m；M represents system factor sum.

(1.3) component units set { U is jointly determined by the whole factor of system and system structure₁,U₂,…,U_m′, m ' expression Component units number, and 1≤m '≤m；The collection of all observation indexs is combined into test point set { T₁,T₂,…,T_n, by system structure Determine articulation set { E}.By { U₁,U₂,…,U_m′, { T₁,T₂,…,T_nAnd { E} sets up the NH-T dependency figure mould of system Type.

One NH-T dependency graphical model as in figure 2 it is shown, wherein, box indicating component units, circle represents test point, Oriented arrow represents connection.

Step 2: released the D matrix of NH-T correlation models by the NH-T dependency graphical model of system is as follows:

D matrix represents the dependency between component units and test point, the i-th row jth column element d in matrix_ijRepresent test Point T_jWith component units U_iWhether it is correlated with, works as T_jU can be recorded_iNon-health information time represent relevant, now d_ijValue is 1, works as T_j U can not be recorded_iNon-health information time represent uncorrelated, now d_ijValue is 0.

Step 3: by the D matrix of NH-T correlation models, be calculated kth (k=1,2 ..., n) individual observation index is to being The percentage ratio weights ω of system healthy grading evaluation impact_k。

Whole test points (except redundancy testing point) that full test point scheme refers to system both participate in healthy grading evaluation, And these test points can the output of the whole component units of covering system.The test weights of each test point are at detection weights On the basis of, add the consideration on testing expense impact, determine that its test weight computing formula is:

$W_k=\frac{1}{{\mathrm{\&alpha;}}_{ck}}\underset{i=1}{\overset{m^{\&prime;}}{\&Sigma;}}{d}_{ik}$

${\mathrm{\&alpha;}}_{ck}=C_k/\underset{k=1}{\overset{n}{\&Sigma;}}{C}_k$

Wherein: W_kRepresent the observation weights of kth observation index；C_kRepresent kth observation index every correlative charges it With；α_ckRepresent the relative costs ratio of kth observation index.

With the test weights size of these all n observation indexs obtained, by these data normalizations, i.e.

${\mathrm{\&omega;}}_k=W_k/\underset{i=1}{\overset{n}{\&Sigma;}}{W}_i\&times;100\%$

Obtained ω_kExpression kth (k=1,2 ..., the n) percentage that system health grading evaluation is affected by individual observation index Compare weights.

Step 4: obtain characteristic behavior sequence and system all of factor behavior sequence.

(4.1) i-th system factor observation data on kth observation index are set as x_i(k) (k=1,2 ..., n), Then the behavior sequence of i-th system factor is expressed as X_i=(x_i(1),x_i(2),…,x_i(n)), the system under all factors that obtains exists Observation data under all observation indexs, determine system all of factor behavior sequence X₁,X₂,…X_m；

(4.2) acquisition system treats investigation state each observation data under all observation indexs, obtains characteristic of correspondence Behavior sequence X₀=(x₀(1),x₀(2),…,x₀(n))；

(4.3) to factor behavior sequence X₁,X₂,…X_mAnd system features behavior sequence X₀Carry out nondimensionalization process.If In each sequence, dimension or the order of magnitude of data are essentially identical, and this step can also be saved.

Step 5: obtain factor behavior sequence and the grey relational grade γ (X of characteristic behavior sequence₀,X_i)。

(5.1) to factor behavior sequence X_i=(x_i(1),x_i(2),…,x_i(n)), calculate and in characteristic behavior sequence same The absolute difference Δ of the observation data under one observation index_0i(k)=| x₀(k)-x_i(k) |, wherein, i=1,2 ..., m.

Ask for the two-stage lowest difference of characteristic behavior sequence and all factor behavior sequencesTwo-stage Maximum poor ${\mathrm{\&Delta;}}_{\max }=\underset{i}{\max }\underset{k}{\max }{\mathrm{\&Delta;}}_{0i}\left(k\right).$

Take resolution ratio ρ=0.5, calculate i-th factor behavior sequence according to formula and characteristic behavior sequence refers in observation Grey incidence coefficient γ under mark k_0i(k)；

${\mathrm{\&gamma;}}_{0i}\left(k\right)=\mathrm{\&gamma;}\left(x_0\right(k),x_i(k\left)\right)=\frac{{\mathrm{\&Delta;}}_{\min }+{\mathrm{\&rho;\&Delta;}}_{\max }}{{\mathrm{\&Delta;}}_{0i}\left(k\right)+{\mathrm{\&rho;\&Delta;}}_{\max }}$

(5.2) by the grey incidence coefficient γ under observation index k_0iK system health classification is commented by () and kth observation index The percentage ratio weights ω of valency impact_k, obtain the grey relational grade γ of i-th factor behavior sequence and system features behavior sequence (X₀,X_i)；

$\mathrm{\&gamma;}(X_0,X_i)=\underset{k=1}{\overset{n}{\&Sigma;}}{\mathrm{\&omega;}}_k\mathrm{\&gamma;}\left(x_0\right(k),x_i(k\left)\right)$

(5.3) selected next factor behavior sequence, enters (5.2) and performs, and continues to calculate grey incidence coefficient, until m Individual factor behavior sequence has calculated with the grey relational grade of system features behavior sequence.

Step 6: carry out healthy grading evaluation.

(6.1) sort according to grey relational grade size, obtain sequence of grey correlation；In sequence of grey correlation, the degree of association is maximum System factor behavior sequence be system and currently treat the factor residing for investigation pattern, provide healthy qualitative classification evaluation result.

(6.2) present invention is according to sequence of grey correlation result, calculates association area indexing and optimum factor confidence level, divides health Level result carries out parameter metric evaluation.

(6.2.1) association area indexing (Discrimination of Grey Relational Degree).

Association area indexing is used for describing in sequence of grey correlation certain two factor behavior sequence about system features behavior sequence Grey relational grade between differentiation degree, have a following two kinds expression way:

1. relative relationship discrimination (D_r).Relative relationship discrimination refers to two factor behavior sequences in sequence of grey correlation The relative difference of grey relational grade, represents with percent, and computing formula is as follows:

$D_{rij}=\frac{|{\mathrm{\&gamma;}}_{0i}-{\mathrm{\&gamma;}}_{0j}|}{{\mathrm{\&gamma;}}_{0i}}\&times;100\%$

Wherein, D_rijRepresent the jth factor behavior sequence relative relationship discrimination relative to i-th factor behavior sequence； γ_0iRepresent the grey relational grade γ (X of i-th factor behavior sequence₀,X_i)；γ_0jRepresent the Lycoperdon polymorphum Vitt of jth factor behavior sequence Degree of association γ (X₀,X_j)。

For relative relationship discrimination, the degree of association that most typically in sequence of grey correlation, numerical value is maximum and second largest pass Relative relationship discrimination between connection degree, it can be used for the district of which group of optimum factor that two groups of sequence of grey correlation of lateral comparison provide Indexing higher, the risk of i.e. optimum factor is less.

2. association area proportion by subtraction weight (D_p).Association area proportion by subtraction heavily refers to that in same sequence of grey correlation, two factor behavior sequences close Shared by differentiation degree between connection degree, the proportion of whole inteerelated order, represents with percent, and computing formula is as follows:

$D_{pij}=\frac{|{\mathrm{\&gamma;}}_{0i}-{\mathrm{\&gamma;}}_{0j}|}{{\mathrm{\&gamma;}}_{max}-{\mathrm{\&gamma;}}_{\min }}\&times;100\%$

Wherein, D_pijRepresent the association area proportion by subtraction weight between i-th factor behavior sequence and jth factor behavior sequence； γ_0iRepresent the grey relational grade of i-th factor behavior sequence；γ_0jRepresent the grey relational grade of jth factor behavior sequence； γ_maxRepresent the association angle value that in sequence of grey correlation, numerical value is maximum；γ_minRepresent the degree of association that in sequence of grey correlation, numerical value is minimum Value.

Utilize association area proportion by subtraction weight, can describe in same sequence of grey correlation between two factor behavior sequence degrees of association The proportion of whole inteerelated order shared by differentiation degree.Similarly, most typical association area proportion by subtraction is exactly heavily numerical value in sequence of grey correlation The maximum association area proportion by subtraction weight between the degree of association and second largest degree of association, it can be used to describe this inteerelated order and is given The discrimination of excellent factor and other factors how, i.e. the risk size of this optimum factor.

(6.2.2) optimum factor confidence level (Confidence of Optimal Factor).

Optimum factor confidence level is for describing the optimum factor determined based on maximum grey relational grade, and it is the most certain For the probability of optimum factor, or it is believed that this factor is exactly the probability of optimum factor in analysis.Order

Δγ_I-II=γ_I-γ_II

Wherein, Δ γ_I-IIRepresent most relevance degree and the difference of the second largest degree of association in sequence of grey correlation；γ_IRepresent that Lycoperdon polymorphum Vitt is closed Grey relational grade maximum in connection sequence；γ_IIRepresent the grey relational grade second largest in sequence of grey correlation.

Resolution ax γ of the degree of association in sequence of grey correlation is made to be:

$\mathrm{\&Delta;}\mathrm{\&gamma;}=\frac{{\mathrm{\&gamma;}}_{max}-{\mathrm{\&gamma;}}_{\min }}{m-1}\&times;\mathrm{\&lambda;}$

Wherein, γ_maxRepresent grey relational grade maximum in sequence of grey correlation, i.e. γ_I；γ_minRepresent in sequence of grey correlation Minimum grey relational grade；M is the number of system whole factor behavior sequence；λ is degree of association resolution adjustment coefficient, λ ＞ 0.

The optimum factor confidence level C of definition is:

$C=\left\{\begin{array}{cc}100\%& {\mathrm{\&Delta;\&gamma;}}_{I-II}\&GreaterEqual;\mathrm{\&Delta;}\mathrm{\&gamma;}\\ \frac{{\mathrm{\&Delta;\&gamma;}}_{I-II}}{\mathrm{\&Delta;}\mathrm{\&gamma;}}\&times;100\%& {\mathrm{\&Delta;\&gamma;}}_{I-II}<\mathrm{\&Delta;}\mathrm{\&gamma;}\end{array}\right.$

According to above-mentioned formula, the optimum factor confidence level in grey correlation analysis can be calculated.Wherein, for adjustment factor λ, its span is greater than the real number of zero.When λ=1, taking system relationship degree resolution is initial data, is not adjusted； When λ ∈ (0,1), result is to be adjusted to less by degree of association resolution ax γ, is equivalent to relax the requirement of confidence level, and gained is Excellent factor confidence level can be bigger；As λ ＞ 1, result is to be adjusted to degree of association resolution ax γ more greatly, be equivalent to confidence level Requiring tightened up, gained optimum factor confidence level can be less.Systematic analysis degree harsh to result is depended in the determination of λ value Requirement.

The embodiment that the present invention provides, selects certain signal conditioning circuit.The PSpice emulation of this signal conditioning circuit Model is as shown in Figure 3.Circuit is made up of four main functional modules:

(1)U₁: input filter circuit, one-level amplifying circuit；

(2)U₂: second amplifying circuit；

(3)U₃: three-stage amplifier；

(4)U₄: level Four amplifying circuit.

In phantom circuit, signal source S1 is direct current signal, and voltage range is 0 to 5.5mV.Circuit is provided with 4 electricity Pressure monitoring point, lays respectively at every grade of amplifying circuit output, such as T in Fig. 3₁～T₄.Circuit has 12 resistance R1～R12,9 electricity Hold C1～C9, it is assumed that each resistance or electric capacity have open circuit and two fault modes of short circuit and parameter drift 30% (rising) Subhealth state pattern, then circuit should have 42 malfunctions, 21 sub-health states and a health status.Arranging simulation step length is These 64 kinds of circuit states are simulated emulation by 0.1mV successively.If the tolerance of resistance and electric capacity is 5%, the emulation of every kind of state Repeatedly average.

Using the inventive method, concrete steps one～step 6 are as follows.

Step one: set up the NH-T dependency graphical model of system, as shown in Figure 4.

Component units collection is combined into { U₁,U₂,U₃,U₄, test point collection is combined into { T₁,T₂,T₃,T₄}.The system factor sum determined There are 23, illustrate in step 4 below.

Step 2: released the D matrix of NH-T correlation models by the NH-T dependency graphical model of system is as follows:

Step 3: by the D matrix of NH-T correlation models, is calculated each observation index to system health grading evaluation shadow The percentage ratio weights ω rung_k。

For four voltage monitoring points in embodiment of the present invention circuit, it is believed that they are all identical, because of The impact that weights are brought by this testing expense can be ignored.Can be referred in the hope of four observations by the D matrix of NH-T correlation models Target observation weights, i.e.Correspondingly the percentage ratio weights of they correspondences are

Step 4: obtain system all of factor behavior sequence and characteristic behavior sequence.

By to circuit theory and the analysis of emulation data, herein below can be summed up:

(1) R6 open circuit, the emulation of R9 open circuit and R12 open circuit can cause software to report an error, therefore not investigate；

(2) total undetectable state 23 kinds, the most undetectable malfunction 11 kinds, including: C1 opens a way, and C2 opens Road, C3 opens a way, and C4 opens a way, and C5 opens a way, and C6 opens a way, and C7 opens a way, and C8 opens a way, and C9 opens a way, R9 short circuit, R12 short circuit；Undetectable Sub-health state 12 kinds, including: C1 parameter drift, C2 parameter drift, C3 parameter drift, C4 parameter drift, C5 parameter drift, C6 Parameter drift, C7 parameter drift, C8 parameter drift, C9 parameter drift, R6 parameter drift, R9 parameter drift, R12 parameter drift. Simultaneously it can be seen that the open fault of electric capacity and parameter drift state are undetectable；

(3) total ambiguity group 9 groups, wherein fault ambiguity group has 8 groups, including: { R2 open circuit, R3 short circuit }, { R2 short circuit, C2 Short circuit, C3 short circuit }, { R4 open circuit, R5 short circuit }, { R4 short circuit, C4 short circuit, C5 short circuit }, { R7 open circuit, C6 short circuit, C7 short circuit, R8 Short circuit }, { R7 short circuit, R8 open circuit }, { C8 short circuit, C9 short circuit, R10 open circuit, R11 short circuit }, { R10 short circuit, R11 open circuit }；Sub-strong Health ambiguity group has 1 group, { R4 parameter drift, R8 parameter drift }.

Above-mentioned undetectable state being deleted from data list, each fuzzy group member merges in data list, I.e. can be obtained by the inventive method and can carry out circuit state and the emulation data of correspondence of healthy grading evaluation, be i.e. System factor behavior sequence, as shown in table 1 (S1=3mV).

Table 1 signal conditioning circuit emulation data (S1=3mV)

For verifying the effectiveness of healthy stage division based on grey correlation analysis, choose health, subhealth state and fault three The state of kind the most once emulates, and then emulation data is carried out grading evaluation, and observed result is the most correct.Wherein, circuit Health status is unique；Sub-health state randomly chooses one, have selected R7 parameter drift here；Malfunction randomly chooses Article one, have selected R5 open circuit here.These three state is the most once emulated, obtains characteristic behavior sequence to be fractionated, point It is not set to X₀₁,X₀₂,X₀₃, as shown in table 2.In this example, each observation data dimension is identical and gap little, does not the most do nondimensionalization Process.

Table 2 characteristic behavior to be fractionated sequence

Sequence number	State	T1(V)	T2(V)	T3(V)	T4(V)
						1	Health-X01	0.0331	-0.4539	2.241	-1.817
2	R7 parameter drift-X02	0.0323	-0.4513	2.021	-1.492
						3	R5 open circuit-X03	0.0318	0.0006	-0.0034	2.697

Step 5: obtain factor behavior sequence and the grey relational grade of characteristic behavior sequence.

The absolute difference such as table 3 of factor behavior sequence and the system features behavior sequence observation data under same observation index Shown in.Three system features behavior sequence X₀₁,X₀₂,X₀₃Grey relational grade respectively as shown in table 4, table 5 and table 6.

Table 3 X₀₁With X₁,X₂,…X₂₃Absolute difference

Table 4 X₀₁Grey incidence coefficient, grey relational grade and inteerelated order

Table 5 X₀₂Grey incidence coefficient, grey relational grade and inteerelated order

Table 6 X₀₃Grey incidence coefficient, grey relational grade and inteerelated order

Step 6: carry out healthy grading evaluation.

From the sequence of grey correlation in table 4, the system factor behavior sequence of degree of association maximum is X₁, i.e. health status shape State, with X₀₁State own is consistent.Therefore conclusion is that system is in health status.

From the sequence of grey correlation in table 5, the system factor behavior sequence of degree of association maximum is X₁₈, i.e. R7 parameter drift State, with X₀₂State own is consistent.Therefore conclusion is that system is in sub-health state, and subhealth state pattern is R7 parameter drift.

From the sequence of grey correlation in table 6, the system factor behavior sequence of degree of association maximum is X₁₃, i.e. R5 opens a way shape State, with X₀₃State own is consistent.Therefore conclusion is that system is in malfunction, and fault mode is R5 open circuit.

Metric parameter computing formula according to healthy classifying capability and X₀₁X₀₂X₀₃Sequence of grey correlation, investigate high point Connection degree and the discrimination of the second largest degree of association, take degree of association resolution adjustment coefficient lambda=1, and this three stack features behavior sequence is corresponding Metric parameter as shown in table 7.

Metric parameter under the full observation index of table 7

Shown in the result in table 7, the numerical value of the first row is less, illustrates at X₀₁The conclusion obtained in characteristic behavior sequence can The highest by degree, and at X₀₂The conclusion reliability obtained in characteristic behavior sequence can be more relatively high.And X₀₃In characteristic behavior sequence The conclusion reliability obtained is lower, and the conclusion the most finally drawn should comprise multiple possible fault mode, obtain more The result of body needs further to analyze.

Claims (1)

1. the grey correlation analysis health classification evaluation method using full observation index scheme, it is characterised in that include step Rapid:

Step one: set up the NH-T dependency graphical model of system；

If system factor sum has m, component units number has m ' individual, and observation index sum has n, and component units collection is combined into {U₁,U₂,…,U_m′, the collection of all observation indexs is combined into test point set { T₁,T₂,…,T_n}；

Step 2: released the D matrix of NH-T correlation models by the NH-T dependency graphical model of system；

I-th row jth column element d in D matrix_ijRepresent test point T_jWith component units U_iWhether it is correlated with, works as T_jU can be recorded_iNon-strong Represent relevant during health information, now d_ijValue is 1, works as T_jU can not be recorded_iNon-health information time represent uncorrelated, now d_ij Value is 0；I=1,2 ... m '；J=1,2 ... n；

Step 3: use full observation index scheme to test；By the D matrix of NH-T correlation models, it is calculated kth (k= 1,2 ..., n) the percentage ratio weights ω that system health grading evaluation is affected by individual observation index_k；

${\mathrm{\&omega;}}_k=W_k/\underset{k=1}{\overset{n}{\&Sigma;}}{W}_k\&times;100\%;W_k=\frac{1}{{\mathrm{\&alpha;}}_{ck}}\underset{i=1}{\overset{m^{\&prime;}}{\&Sigma;}}{d}_{ik},{\mathrm{\&alpha;}}_{ck}=C_k/\underset{k=1}{\overset{n}{\&Sigma;}}{C}_k;$

W_kRepresent the observation weights of kth observation index；C_kRepresent every correlative charges sum of kth observation index；α_ckRepresent The relative costs ratio of kth observation index；

Step 4: obtain characteristic behavior sequence and system all of factor behavior sequence；

If the observation data that i-th system factor is on kth observation index are x_i(k), the then row of i-th system mode pattern It is shown as X for sequence table_i=(x_i(1),x_i(2),…,x_i(n))；Acquisition system treats the sight under all observation indexs of the investigation state Survey data, obtain characteristic behavior sequence X₀=(x₀(1),x₀(2),…,x₀(n))；The system under all state modeles that obtains is owning Observation data under observation index, determine system all of factor behavior sequence X₁,X₂,…X_m；

Step 5: obtain factor behavior sequence and the grey relational grade of characteristic behavior sequence；

I-th factor behavior sequence X_iWith system features behavior sequence X₀Grey relational grade γ (X₀,X_i) it is:

$\mathrm{\&gamma;}(X_0,X_i)=\underset{k=1}{\overset{n}{\&Sigma;}}{\mathrm{\&omega;}}_k\mathrm{\&gamma;}\left(x_0\right(k),x_i(k\left)\right),i=1,2,...,m;$

γ(x₀(k),x_i(k)) it is i-th factor behavior sequence and characteristic behavior sequence grey correlation system under observation index k Number；

Step 6: carry out healthy grading evaluation；

Find and system features behavior sequence X₀The maximum factor behavior sequence of grey relational grade, by this factor behavior sequence pair The system health state answered and system factor are as X₀Corresponding system health state and system factor.