CN106599934A - Pod fault diagnosis method - Google Patents

Abstract

The invention discloses a pod fault diagnosis method. The method comprises the following steps of using integration test data of a pod with a preset fault to construct a fault mode characteristic subspace matrix database; using the integration test data of a real pod to construct a characteristic subspace, and according to a principal component model and Bayes posterior probability distribution, extracting a data estimation characteristic subspace; and carrying out similarity determination on the data estimation characteristic subspace and a characteristic subspace in the fault mode characteristic subspace matrix database, and according to a determination result, determining a fault mode of the real pod. In the invention, according to a geometric meaning of principal component analysis, a Bayes space estimation idea is introduced so that a characteristic subspace matrix of small sample data is estimated; and then through comparing a similarity of the characteristic subspace and the fault mode characteristic subspace, fault diagnosis is completed so that a pod fault diagnosis problem based on the small sample space is effectively solved.

Description

A kind of gondola method for diagnosing faults

Technical field

The present invention relates to communication technical field, more particularly to a kind of gondola method for diagnosing faults.

Background technology

As the continuous lifting of gondola technical merit, the ambit being related to are continuously increased, technical system framework is multiple all the more It is miscellaneous, failure mechanism and fault mode it is complex, belong to typical complex equipment fault diagnosis, fault diagnosis difficulty is larger.

And current gondola include it is various, for example catch control instruction gondola, targeting pod, Jamming pod, navigation pod, due to Gondola realizes that principle of work and power is different, and its fault mode and failure mechanism are different.At present, it is former in the failure for differentiating gondola Because when, fault diagnosis is realized using the first eigenvector direction of fault data and the matching degree of proper subspace generally, and This kind of method need to possess big-sample data condition, just accurately can be judged.And it is current, due to acquisition of information and data processing energy Power is not enough, using this kind of method, implements more difficult.And when gondola test data is limited, sample space is little, very Hardly possible goes out the status information comprising fault signature from limited extracting data, adopts Efficiency and positional accuracy, while also bringing white elephant for the data mining process work of diagnostic software.

Therefore, existing method for diagnosing faults is when test data sample is less, it is difficult to meet gondola Maintenance Support System Fault diagnosis demand.

The content of the invention

The embodiment of the present invention provides a kind of gondola method for diagnosing faults, to solve existing method for diagnosing faults when test When data sample is less, it is difficult to meet the needs of problems of the fault diagnosis of gondola Maintenance Support System.

In order to realize foregoing invention purpose, the present invention adopts following technical schemes：

According to one aspect of the present invention, there is provided a kind of gondola method for diagnosing faults, including：

Fault mode proper subspace matrix storehouse is built using the Synthetic Measuring Data of preset failure gondola；

By the Synthetic Measuring Data construction feature subspace of actual gondola, further according to principal component model and Bayes posterior probability Data estimation proper subspace is extracted in distribution；

By the proper subspace in the data estimation proper subspace and the fault mode proper subspace matrix storehouse Similarity judgement is carried out, and the fault mode of the actual gondola is determined according to judged result.

Further, the Synthetic Measuring Data of the use preset failure gondola builds fault mode proper subspace matrix Storehouse, specifically includes：

The Synthetic Measuring Data of the gondola of preset failure is built into fault mode data matrix；

Solve the characteristic vector of the covariance matrix of the fault mode data matrix under each fault mode, with feature to Measure fault signature subspace is built for substrate；

Carrying out after angle rotation to the fault signature subspace, adds to fault mode proper subspace matrix storehouse In.

Further, the characteristic vector for solving the covariance matrix of the data matrix under each fault mode, with Characteristic vector is that substrate builds fault signature subspace, is specifically included：

The covariance matrix for obtaining fault mode data matrix X is ∑ X, and computing formula is as follows：

Calculate the eigenvalue λ of the covariance matrix ∑ X_iAnd its corresponding characteristic vector b_i, formula is as follows：

Choose j (j<M) characteristic vector corresponding to individual eigenvalue of maximum, you can obtain the fault signature subspace B= [b1,b2,b3,…bj]。

Further, it is as follows to the computing formula for carrying out angle rotation of the fault signature subspace：

B^T(∑_X)^-1Bx=e²

Wherein, Bx is postrotational fault mode proper subspace.

Further, the Synthetic Measuring Data construction feature subspace by actual gondola, further according to principal component model and Data estimation proper subspace is extracted in Bayes posterior probability distribution, is specifically included：

The Synthetic Measuring Data of actual gondola is obtained, and the Synthetic Measuring Data builds test data matrix；

The characteristic vector of the covariance matrix of the test data matrix is solved, and feature is set up by substrate of characteristic vector Space；

Carry out each pivot coordinate of orthogonal CS decomposition and inversion and pedestal target angular relationship to the proper subspace, and according to Bayesian probability is distributed and gibbs sampler obtains stochastic sampling matrix；

Actual gondola Synthetic Measuring Data is obtained through data moving average filter according to stochastic sampling matrix and estimates feature Subspace.

Further, it is described that each pivot coordinate of orthogonal CS decomposition and inversion and pedestal target angle are carried out to the proper subspace The formula of degree relation is as follows：

Wherein, B is characterized subspace；R and H₁It is the orthogonal matrix of n × m；H₂It is the semi-orthogonal matrix of (n-m) × m；N is Test item number, m are pivot number；C is diag (cos θ 1 ... cos θ m)；S for (sin θ 1 ... sin θ m)；θ m are principal component space With the angle number of degrees between the basis coordinates pivot characteristic vector of space.

Further, it is described that stochastic sampling matrix is obtained according to Bayesian probability distribution and gibbs sampler, specifically include：

The formula of the Bayes posterior probability of H1, H2 is as follows：

H is obtained by gibbs sampler₁ ⁽ⁿ⁾, H₂ ⁽ⁿ⁾；

H is obtained according to each pivot coordinate and pedestal target angular relationship, and gibbs sampler₁ ⁽ⁿ⁾And H₂ ^(n),, obtain with The formula of machine sampling matrix is as follows：

Further, it is described that actual gondola integration test is obtained through data moving average filter according to stochastic sampling matrix The formula of data estimation proper subspace is as follows：

Wherein, P_mTo take the front m principal component vector of matrix in bracket；B⁽ⁿ⁾For stochastic sampling matrix, N_iSlide for data and filter The window width of ripple, N_brRepresent data glide filter homing sequence numbering, N_iAnd N_brNumerical value generally will rule of thumb and data filtering Effect sets.

Further, in the data characteristicses subspace matrices and the fault mode proper subspace matrix storehouse When matrix carries out similarity and judges, specifically include：

Calculate each matrix in the data characteristicses subspace matrices and the fault mode proper subspace matrix storehouse The weighted sum of included angle cosine value on projecting direction；

When the value of the weighted sum is more than predetermined threshold value, then two similar matrixes, the failure of the actual gondola are judged Pattern is the corresponding fault mode of proper subspace matrix in the fault mode proper subspace matrix storehouse.

The present invention has the beneficial effect that：

Gondola method for diagnosing faults provided by the present invention, according to the geometric meaning of pivot analysis, introduces Bayes space The thought of estimation, estimates the proper subspace matrix of Small Sample Database；Then contrast characteristic subspace and fault mode are passed through The similarity of proper subspace, completes fault diagnosis, solves based on a small sample space gondola fault diagnosis difficult problem.

Described above is only the general introduction of technical solution of the present invention, in order to better understand the technological means of the present invention, And can be practiced according to the content of description, and in order to allow the above and other objects of the present invention, feature and advantage can Become apparent, below especially exemplified by the specific embodiment of the present invention.

Description of the drawings

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing Accompanying drawing to be used needed for having technology description is briefly described, it should be apparent that, drawings in the following description are only Some embodiments of the present invention, for those of ordinary skill in the art, without having to pay creative labor, also Other accompanying drawings can be obtained according to these accompanying drawings.

The flow chart of the gondola method for diagnosing faults that Fig. 1 is provided by the embodiment of the present invention.

Specific embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than the embodiment of whole.It is based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made Embodiment, belongs to the scope of protection of the invention.

Gondola method for diagnosing faults provided in an embodiment of the present invention, as shown in figure 1, specifically including following steps：

Step 1, builds fault mode proper subspace matrix storehouse using the Synthetic Measuring Data of preset failure gondola.It is optional , when the step is implemented, specifically include following steps：

Step 101, obtains the Synthetic Measuring Data of the gondola of preset failure, and the integration test of the gondola to preset failure Data normalization.

Wherein, Synthetic Measuring Data is standardized and the sample sequence of Synthetic Measuring Data is built into into n × m latitudes exactly Data matrix X=[x_m(1),x_m(2),x_m(3),…x_m(n)], wherein n is sampled point number, and m is test item number.

Step 102, fault data proper subspace build.Specifically, by solving the failure under each fault mode The characteristic vector of the covariance matrix of mode data matrix, builds fault signature subspace by substrate of characteristic vector.

When fault data proper subspace is built, the covariance matrix for obtaining fault mode data matrix X first is ∑ X, concrete formula are as follows：

Then, eigenvalue λ m of covariance matrix ∑ X, and its corresponding characteristic vector bi are obtained, concrete formula is as follows：

Then, j (j are chosen<M) characteristic vector corresponding to individual eigenvalue of maximum, you can obtain fault signature subspace B= [b1,b2,b3,…bj]。

Step 103, carries out angle rotation to fault signature subspace.Here angle rotation is exactly the space coordinates of data Carry out translating and rotationally-varying, make the new coordinate system of data match with data center of gravity, specific formula is as follows：

B^T(∑_X)^-1Bx=e² (3)

Wherein, Bx is the postrotational fault signature subspace of angle.

Step 104, builds fault mode proper subspace matrix storehouse.The substrate of different faults data is by characteristic vector institute Characterize, therefore the anglec of rotation of the substrate corresponding to different failures is also different, this means that different faults number It is also different according to corresponding proper subspace.By the postrotational fault signature subspace of angle under different faults pattern Add into fault mode proper subspace matrix storehouse.For n kind fault modes, i-th kind of fault mode is denoted as fi (1,2,3 ... N), thus occur different faults corresponding to proper subspace matrix be Bfi (1,2,3 ... n).

Step 2, by the Synthetic Measuring Data construction feature subspace of actual gondola, after principal component model and Bayes Test probability distribution and extract data estimation proper subspace.Specifically, comprise the steps：

Step 201, carries out integration test to certain instruction capture gondola, is standardized to obtaining Synthetic Measuring Data.Tool Body, Synthetic Measuring Data sample sequence is built into n × m latitude data matrix Y=[y_m(1),y_m(2),y_m(3),…y_m(n)], its Middle n is sampled point number, and m is number of checkpoints.

Step 202, according to Synthetic Measuring Data construction feature subspace.For building process can be found in above-mentioned step 102, that is, solve the covariance matrix ∑ of test data matrix Y_YCharacteristic vector by, with characteristic vector as substrate set up feature son Space B_Y, for building process, it will not be described here.

Proper subspace position is converted each pivot coordinate and pedestal target angular relationship by step 203.Due to characteristic vector Matrix B_YFor orthogonal matrix, carry out orthogonal CS and decompose achievable angular relationship conversion.The formula for obtaining C and S is as follows：

Wherein, H₁It is the orthogonal matrix of m × m with R, H₂It is the semi-orthogonal matrix of (n-m) × m；N is test item number；M is Pivot number；C is diag (cos θ₁,…cosθ_m)；S is (sin θ₁,…sinθ_m)；θ_mFor principal component space and space basis coordinates pivot The angle number of degrees between characteristic vector.

Step 204, according to Bayesian probability be distributed and gibbs sampler obtain stochastic sampling matrix.Specifically, according to master The principal component model of meta-analysis meets Gaussian distribution model and then introduces Bayesian probability distribution, therefore, H₁、H₂Bayesian posterior Probability distribution is shown in formula (5).

H is obtained by gibbs sampler₁ ⁽ⁿ⁾, H₂ ⁽ⁿ⁾。

By formula (4) the calculated C and S and calculated H of sampling₁ ⁽ⁿ⁾, H₂ ⁽ⁿ⁾, you can obtain the random of a sequence Sampling matrix B⁽ⁿ⁾, concrete formula is as follows：

According to stochastic sampling matrix through data moving average filter, actual gondola Synthetic Measuring Data can be obtained and estimate feature Subspace B*, concrete formula are as follows：

Wherein, P_mTo take the front m principal component vector of matrix in bracket；B⁽ⁿ⁾For stochastic sampling matrix, N_iSlide for data and filter The window width of ripple, N_brRepresent data glide filter homing sequence numbering, N_iAnd N_brNumerical value generally will rule of thumb and data filtering Effect sets.

Step 3, data estimation proper subspace is entered with the proper subspace in fault mode proper subspace matrix storehouse Row similarity judges, and the fault mode of actual gondola is determined according to judged result.

Spy in the gondola measured data to obtaining estimates proper subspace and fault mode proper subspace matrix storehouse When levying subspace and carrying out similarity and judge, mainly by asking the weighting of two matrix included angle cosine values on n projecting direction With judging：When weighted sum is closer to 1, then show that the similarity of two matrixes is higher.When similarity is more than predetermined threshold value When, then obtain and fault mode corresponding to proper subspace in fault mode proper subspace matrix storehouse, and judge the gondola Failure is the fault mode.

Understood based on above-mentioned, the gondola method for diagnosing faults in the embodiment of the present invention, the gondola integration test item for being adopted Mesh is less with data volume, according to the geometric meaning of pivot analysis, introduces CS and decomposes, by the think of using Bayes's Spatial outlier Think, covariance matrix problem is changed into into proper subspace estimation problem.The spy of Small Sample Database can be estimated using the present invention Levy subspace matrices；Then by contrast characteristic subspace and the similarity of fault mode proper subspace, fault diagnosis is completed, Solve based on a small sample space gondola fault diagnosis difficult problem.

Each embodiment in this specification is described by the way of progressive, what each embodiment was stressed be with The difference of other embodiment, between each embodiment identical similar part mutually referring to.For system embodiment For, due to itself and embodiment of the method basic simlarity, so description is fairly simple, portion of the related part referring to embodiment of the method Defend oneself bright.Also, herein, term " including ", "comprising" or its any other variant are intended to nonexcludability Include so that a series of process, method, article or equipment including key elements not only include those key elements, but also Including other key elements being not expressly set out, or also include intrinsic for this process, method, article or equipment wanting Element.In the absence of more restrictions, the key element for being limited by sentence "including a ...", it is not excluded that wanting including described The process of element, method, also there is other identical element in article or equipment.

In addition, one of ordinary skill in the art will appreciate that realizing that all or part of step of above-described embodiment can pass through Hardware is completing, it is also possible to instruct the hardware of correlation to complete by program, described program can be stored in a kind of computer In readable storage medium storing program for executing, storage medium mentioned above can be read only memory, disk or CD etc..

Obviously, those skilled in the art can carry out the essence of various changes and modification without deviating from the present invention to the present invention God and scope.So, if these modifications of the present invention and modification belong to the scope of the claims in the present invention and its equivalent technologies Within, then the present invention is also intended to comprising these changes and modification.

Claims (9)

1. a kind of gondola method for diagnosing faults, it is characterised in that include：

Fault mode proper subspace matrix storehouse is built using the Synthetic Measuring Data of preset failure gondola；

By the Synthetic Measuring Data construction feature subspace of actual gondola, further according to principal component model and Bayes posterior probability distribution Extract data estimation proper subspace；

The data estimation proper subspace is carried out with the proper subspace in the fault mode proper subspace matrix storehouse Similarity judges, and the fault mode of the actual gondola is determined according to judged result.

2. the method for claim 1, it is characterised in that the Synthetic Measuring Data of the use preset failure gondola builds Fault mode proper subspace matrix storehouse, specifically includes：

The Synthetic Measuring Data of the gondola of preset failure is built into fault mode data matrix；

The characteristic vector of the covariance matrix of the fault mode data matrix under each fault mode is solved, with characteristic vector is Substrate builds fault signature subspace；

Carrying out after angle rotation to the fault signature subspace, adds into fault mode proper subspace matrix storehouse.

3. method as claimed in claim 2, it is characterised in that the association of the data matrix under the solution each fault mode The characteristic vector of variance matrix, builds fault signature subspace by substrate of characteristic vector, specifically includes：

The covariance matrix for obtaining fault mode data matrix X is ∑ X, and computing formula is as follows：

${\mathrm{\&Sigma;}}_X=\frac{1}{n-1}{\mathrm{XX}}^T$

Calculate the eigenvalue λ of the covariance matrix ∑ X_iAnd its corresponding characteristic vector b_i, formula is as follows：

$\left\{\begin{array}{c}|{\mathrm{\&lambda;}}_{i}I-{\&Sigma;}_X|=0\\ ({\mathrm{\&lambda;}}_{i}I-{\&Sigma;}_X)b_i=0\end{array}\right.,i=1,2,3,\mathrm{...}m$

Choose j (j<M) characteristic vector corresponding to individual eigenvalue of maximum, you can obtain the fault signature subspace B=[b1, b2,b3,…bj]。

4. method as claimed in claim 2, it is characterised in that the meter for carrying out angle rotation to the fault signature subspace Calculate formula as follows：

B^T(∑_X)^-1Bx=e²

Wherein, Bx is postrotational fault mode proper subspace.

5. the method for claim 1, it is characterised in that Synthetic Measuring Data construction feature by actual gondola Space, extracts data estimation proper subspace further according to principal component model and Bayes posterior probability distribution, specifically includes：

The Synthetic Measuring Data of actual gondola is obtained, and the Synthetic Measuring Data builds test data matrix；

The characteristic vector of the covariance matrix of the test data matrix is solved, feature is set up as substrate with characteristic vector empty Between；

The each pivot coordinate of orthogonal CS decomposition and inversion and pedestal target angular relationship are carried out to the proper subspace, and according to pattra leaves This probability distribution and gibbs sampler obtain stochastic sampling matrix；

Actual gondola Synthetic Measuring Data is obtained through data moving average filter according to stochastic sampling matrix and estimates that feature is empty Between.

6. method as claimed in claim 5, it is characterised in that described that orthogonal CS decomposition and inversion is carried out to the proper subspace Each pivot coordinate is as follows with the formula of pedestal target angular relationship：

$B=\left[\begin{array}{c}{H}_{1}C\\ H_{2}S\end{array}\right]R^T$

Wherein, B is characterized subspace；R and H₁It is the orthogonal matrix of n × m；H₂It is the semi-orthogonal matrix of (n-m) × m；N is test Item number, m are pivot number；C is diag (cos θ 1 ... cos θ m)；S for (sin θ 1 ... sin θ m)；θ m are principal component space and sky Between basis coordinates pivot characteristic vector between the angle number of degrees.

7. method as claimed in claim 6, it is characterised in that described to be obtained according to Bayesian probability distribution and gibbs sampler Stochastic sampling matrix, specifically includes：

The formula of the Bayes posterior probability of H1, H2 is as follows：

$p\left(y_1\right|H_1,H_2)\&Proportional;e^{-({\mathrm{\&alpha;}}_i-{\mathrm{\&gamma;}}_i)+2{\mathrm{\&beta;}}_i{y_i}^{\frac{1}{2}}{(1-y_i)}^{\frac{-1}{2}}}{y_i}^{\frac{-1}{2}}{(1-y_i)}^{\frac{-1}{2}}\&lsqb;0,y_{max}\&rsqb;\left(y_i\right)$

H is obtained by gibbs sampler₁ ⁽ⁿ⁾, H₂ ⁽ⁿ⁾；

H is obtained according to each pivot coordinate and pedestal target angular relationship, and gibbs sampler₁ ⁽ⁿ⁾And H₂ ^(n),, obtain and take out at random The formula of sample matrix is as follows：

$B^{\left(n\right)}=\left[\begin{array}{c}{H_1}^{\left(n\right)}{C}^{\left(n\right)}\\ {H_2}^{\left(n\right)}{S}^{\left(n\right)}\end{array}\right]$

8. method as claimed in claim 7, it is characterised in that described to be filtered through data moving averages according to stochastic sampling matrix Ripple obtains actual gondola Synthetic Measuring Data and estimates that the formula of proper subspace is as follows：

$B*=P_m\left\{\frac{1}{N_i}\underset{n=N_{br}+1}{\overset{N_{br}+N_r}{\&Sigma;}}{B}^{\left(n\right)}{\left(B^{\left(n\right)}\right)}^T\right\}$

Wherein, P_mTo take the front m principal component vector of matrix in bracket；B⁽ⁿ⁾For stochastic sampling matrix, N_iFor data glide filter Window width, N_brRepresent data glide filter homing sequence numbering, N_iAnd N_brNumerical value generally will rule of thumb with data filter effect Setting.

9. the method for claim 1, it is characterised in that to the data characteristicses subspace matrices and the failure mould When matrix in formula proper subspace matrix storehouse carries out similarity and judges, specifically include：

Calculate the data characteristicses subspace matrices throwing with each matrix in the fault mode proper subspace matrix storehouse The weighted sum of included angle cosine value on shadow direction；

When the value of the weighted sum is more than predetermined threshold value, then two similar matrixes, the fault mode of the actual gondola are judged For the corresponding fault mode of proper subspace matrix in the fault mode proper subspace matrix storehouse.