CN102495949A - Fault prediction method based on air data - Google Patents

Abstract

The invention discloses a fault prediction method based on air data. The method comprises the following steps of: establishing an air database for storing air data; establishing a knowledge base for storing a rule for fault analysis; removing noise of the air data by using a singular value decomposition method; performing chaotic characteristic judgment on the air data from which the noise is removed by using a maximum Lyapunov index method; realizing prediction of the air data by using a prediction algorithm; and performing fault analysis according to a predicted value and knowledge in the knowledge base, and finally outputting a predicted result. Due to the adoption of the fault prediction method, the influence of noise on prediction accuracy is reduced, selection of a more suitable prediction model is facilitated, high prediction accuracy is achieved, and the serviceable range of the method is widened.

Description

A kind of failure prediction method based on flying quality

Technical field

The invention belongs to artificial intelligence field, relate to failure prediction technology, noise management technique, time series forecasting technology and data mining technology, specifically, be meant a kind of failure prediction method.

Background technology

Along with the raising of aircraft robotization and level of intelligence, its potential possibility that breaks down is increasing.In a single day airborne equipment breaks down, and gently then reduces performance, influences flight safety, heavy then device damage, fatal crass.Existing timing maintenance and reaction equation diagnosis and repair mode, because of there being hysteresis quality, method falls behind; Specific aim and less economical; Workload is big, inefficiency, the demand for development of incompatibility aircraft industry; Therefore (Prognostics and Health Management, technology PHM) is arisen at the historic moment for failure prediction and health control.PHM is on advanced person's sensor acquisition technical foundation, the health status of predicting, monitor, diagnosing and managing airborne part of appliance by various algorithm and models.The failure prediction technology is one of core technology of PHM, has obtained unprecedented attention and development.

The data that aircraft produces in flight course (as are kept at the data on digital flight data recorder magnetic medium or the quick access recorder CD; Cruise data etc. in the report and the report of taking off) information that has comprised a large amount of aircraft states in; Utilize these data to carry out failure prediction, will important basis be provided for maintenance.

Inevitably can there be noise in flying quality in the process of gathering and transmitting, therefore when utilizing flying quality to carry out failure prediction, at first will dispel The noise, to improve final accuracy of predicting.Along with the progress of aeronautical technology, the standard of flight data recording appearance recording parameters is also improving constantly, and the kind of recording parameters also increases gradually.Each is different for the duty of these flight parameters and environmental factor, causes data characteristics separately also to exist than big-difference, if all adopt certainly will impact prediction with a kind of Forecasting Methodology accuracy and reliability.Therefore before predicting, need analyze, but going back the neither one method can take into full account on the basis of the problems referred to above, realize failure prediction based on flying quality to the characteristics of flight parameter.

Summary of the invention

In order to overcome the deficiency of prior art; The invention provides a failure prediction method based on flying quality; Not only can the noise in the flying quality be carried out pre-service; Also comprised two kinds of failure prediction algorithms, can select suitable Forecasting Methodology, ensured prediction accuracy and reliability according to the characteristics of flying quality.

The inventive method solves the technical scheme that its technical matters adopted and comprises following steps:

The first step makes up the flying quality storehouse, the storage flying quality.

Second step made up knowledge base, and storage is used for the rule of fault analysis, explained that aircraft breaks down if rule has represented that flying quality has met or exceeded certain value.

In the 3rd step, utilize the method for svd to dispel the noise of flying quality.

To flying quality sequence X={ x ₁, x ₂, Λ, x _NCarry out svd and may further comprise the steps, x wherein _iRepresent i flying quality, N representes natural number:

(1) structural matrix initial matrix

$A={\left[\begin{array}{cccc}{x}_1& x_2& \mathrm{\Λ}& x_k\\ x_2& x_3& \mathrm{\Λ}& x_{k+1}\\ M& M& O& M\\ x_n& x_{n+1}& \mathrm{\Λ}& x_N\end{array}\right]}_{n\×k}$

expression is not more than the integer of N+1/2; N=N-k+1, n and k are natural numbers.

(2) A is carried out svd

A＝UΓV ^T

U, V is orthogonal matrix, V ^TThe matrix transpose of expression V,

$\mathrm{\Γ}=\left[\begin{array}{cc}{\mathrm{\Π}}_{m\×m}& O_{m\×(k-m)}\\ O_{(n-m)\×m}& O_{(n-m)\×(k-m)}\end{array}\right]$

∏ wherein _{M * m}Be a diagonal form matrix, the element on the diagonal line is the singular value (α of matrix A ₁, α ₂, Λ, α _m).(α ₁, α ₂, Λ, α _m) satisfy α ₁>=α ₂>=Λ>=α _m>=0; O _{M * (k-m)}, O _{(n-m) * m}, O _{(n-m) * (k-m)}Be element and all be 0 matrix; M is a natural number, the order of representing matrix A.

(3) find the solution filtering matrix

Set filter threshold ε (0＜ε＜1), solving equation

$\frac{\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a_i}^2}{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{a_i}^2}=\mathrm{\ϵ}$

Can try to achieve positive integer l, the m-l that value is a less singular value is changed to 0 and obtains

${\mathrm{\Γ}}^1=\left[\begin{array}{cc}{\mathrm{\Π}}_{l\×l}& O_{l\×(k-l)}\\ O_{(n-l)\×l}& O_{(n-l)\×(k-l)}\end{array}\right],$

(4) filtering output

A ¹＝UΓ ¹V ^T

Matrix A then ¹In data be the flying quality of dispeling behind the noise.

In the 4th step, utilize largest Lyapunov exponent method to carry out chaotic characteristic and differentiate to dispeling flying quality behind the noise.

The 5th step, utilize prediction algorithm to realize prediction to flying quality, prediction algorithm has comprised autoregressive moving-average model and supporting vector machine model.When data are linear relationship, select autoregressive moving-average model; When data are nonlinear relationship, select supporting vector machine model.

Autoregressive moving-average model is shown in (1) formula:

S _t＝φ ₁S _t-1+φ ₂S _t-2+L+φ _pS _t-p+θ ₁ε _t-1+θ ₂ε _t-2+L+θ _qε _t-q+ε _t，t∈Z ⁺. (1)

Z wherein ⁺Be positive integer set, { ε _t, t ∈ Z ⁺Be the independent same distribution random series, ε _tNormal Distribution

The variance of expression normal distribution, p, q are model orders; Parameter phi ₁, φ ₂, L, φ _p, θ ₁, θ ₂, L, θ _q,

It is unknown parameter.

Flying quality prediction flow process based on autoregressive moving-average model may further comprise the steps:

(1) reading the flying quality of dispeling behind the noise, utilize distance of swimming method that flying quality is carried out stationary test, judge the stationarity of flying quality, if not steadily, must carry out difference processing to signal, is stably up to differentiated flying quality.

(2) autocorrelation function and the partial autocorrelation function of calculating flying quality, Application of B IC decides the rank criterion, confirms model order p, q.Use least square method estimation model parameter phi ₁, φ ₂, L, φ _p, θ ₁, θ ₂, L, θ _q,

(3) utilize next data value constantly of Model Calculation of setting up, finally export result of calculation.

Supporting vector machine model is shown in (2) formula:

$f\left(x\right)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_{i}K(x_i,x)+b---\left(2\right)$

λ wherein _i, i=1,2, L, N are the optimum solution of optimization problem (3).

min $\frac{1}{2}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{\mathrm{\λ}}_{j}K(x_i,x_j)+\mathrm{\ϵ}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|{\mathrm{\λ}}_i\right|-\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_i{\mathrm{\λ}}_i$

(3)

s.t. $\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_i=0,$ -C≤λ _i≤C，i＝1，2，...，N

$K(x,x^{\′})=\mathrm{Exp}(-\frac{{\left|\right|x-x^{\′}\left|\right|}^2}{{\mathrm{\σ}}^2}),$ $b=y_j-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{\mathrm{\λ}}_{i}K(x_i\·x_j)-\mathrm{\ϵ},$ ε is the insensitive loss factor, and C is the punishment parameter, and wherein σ is the yardstick of kernel function.

Flying quality prediction flow process based on SVMs may further comprise the steps:

(1) read the flying quality of dispeling behind the noise after, flying quality is carried out phase space reconfiguration.

(2) (Sequential Minimal Optimization, SMO) solving-optimizing problem (3) calculates λ to utilize the minimum optimization algorithm of sequence _i, i=1,2, L, N.

(3) utilize next numerical value constantly of Model Calculation of setting up, finally export result of calculation.

In the 6th step, carry out fault analysis according to the value and the knowledge in the knowledge base of prediction, the result of the final prediction of output.

The invention has the beneficial effects as follows:

1, utilizes the method for svd to dispel the noise in the flying quality, reduced the influence of noise precision of prediction.

2, flying quality chaos property is analyzed, helped selecting more suitable forecast model.

3, comprised autoregressive moving-average model that is applicable to the processing linear case and the supporting vector machine model that is applicable to the dealing with nonlinear situation in the method, made that not only method has than higher forecast precision, and widened the usable range of method.

Description of drawings

Fig. 1 is the structural drawing of the inventive method;

Fig. 2 is autoregressive moving-average model prediction process flow diagram;

Fig. 3 is the SVM prediction process flow diagram.

Embodiment

Below in conjunction with accompanying drawing the inventive method is elaborated.

As shown in Figure 1, a kind of failure prediction method based on flying quality may further comprise the steps:

(1) sets up the flying quality storehouse

The flying quality storehouse is used for storing the aircraft parameter value that aircraft cruises and reports, the aircraft parameter value is existed in the flying quality information table.The information of flying quality library storage comprises information such as the numerical value, sampling instant of unit, the parameter of title, the parameter of parameter aircraft possessed, parameter.

(2) set up knowledge base

Knowledge base is used for storing the rule that is used for fault analysis, and the representing with form of production of rule is i.e. " if regular prerequisite " then " rule conclusion ".If rule has been represented flying quality and has met or exceeded certain value then rule is left in the rule list that the design of rule list is as shown in table 1.

Table 1 rule list

(3) flying quality pre-service

Utilize the method for svd to dispel the noise of flying quality in the flying quality storehouse.If X={x ₁, x ₂, Λ, x _NBe a certain flying quality sequence, wherein x _iThe expression flying quality, N representes natural number.The process of X being carried out svd filtering is following:

(1) structural matrix initial matrix

$A={\left[\begin{array}{cccc}{x}_1& x_2& \mathrm{\Λ}& x_k\\ x_2& x_3& \mathrm{\Λ}& x_{k+1}\\ M& M& O& M\\ x_n& x_{n+1}& \mathrm{\Λ}& x_N\end{array}\right]}_{n\×k}$

expression is not more than the integer of N+1/2; N=N-k+1; N, k are natural numbers.

(2) A is carried out svd

A＝UΓV ^T

U, V is orthogonal matrix, V ^TThe matrix transpose of expression V,

$\mathrm{\Γ}=\left[\begin{array}{cc}{\mathrm{\Π}}_{m\×m}& O_{m\×(k-m)}\\ O_{(n-m)\×m}& O_{(n-m)\×(k-m)}\end{array}\right]$

∏ wherein _{M * m}Be a diagonal form matrix, the element on the diagonal line is the singular value (α of matrix A ₁, α ₂, Λ, α _m).(α ₁, α ₂, Λ, α _m) satisfy α ₁>=α ₂>=Λ>=α _m>=0; O _{M * (k-m)}, O _{(n-m) * m}, O _{(n-m) * (k-m)}Be element and all be 0 matrix; M is a natural number, the order of representing matrix A.

(3) find the solution filtering matrix

Set filter threshold ε (0＜ε＜1), solving equation

$\frac{\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a_i}^2}{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{a_i}^2}=\mathrm{\ϵ}$

Can try to achieve positive integer l, the m-l that value is a less singular value is changed to 0 and obtains

${\mathrm{\Γ}}^1=\left[\begin{array}{cc}{\mathrm{\Π}}_{l\×l}& O_{l\×(k-l)}\\ O_{(n-l)\×l}& O_{(n-l)\×(k-l)}\end{array}\right],$

(4) filtering output

A ¹＝UΓ ¹V ^T

Matrix A then ¹In data be the data of dispeling behind the noise.

(4) flying quality analysis

Utilize largest Lyapunov exponent method to carry out chaotic characteristic and differentiate to dispeling flying quality behind the noise.

(5) data prediction

Utilize prediction algorithm to realize prediction, in this patent, comprised autoregressive moving-average model and supporting vector machine model flying quality.When data are linear relationship, select autoregressive moving-average model; When data are nonlinear relationship, select supporting vector machine model.

Autoregressive moving-average model is shown in (1) formula:

S _t＝φ ₁S _t-1+φ ₂S _t-2+L+φ _pS _t-p+θ ₁ε _t-1+θ ₂ε _t-2+L+θ _qε _t-q+ε _t，t∈Z ⁺. (1)

Z wherein ⁺Be positive integer set, { ε _t, t ∈ Z ⁺Be the independent same distribution random series, suppose ε usually _tObey just too and distribute

The variance of expression normal distribution, p, q are model orders; Parameter phi ₁, φ ₂, L, φ _p, θ ₁, θ ₂, L, θ _q,

It is unknown parameter.

Flying quality prediction flow process based on autoregressive moving-average model is as shown in Figure 2, and whole process is divided into 2 parts: training process and data prediction.Training process is to utilize pretreated flying quality to confirm model order p, q and unknown parameter φ ₁, φ ₂, L, φ _p, θ ₁, θ ₂, L, θ _q,

Data prediction is according to formula (1) and passes through p, q and the φ that training process obtains ₁, φ ₂, L, φ _p, θ ₁, θ ₂, L, θ _q,

Calculate next data value constantly.Detailed process is:

(1) reading pretreated flying quality, utilize distance of swimming method that flying quality is carried out stationary test, judge the stationarity of flying quality, if not steadily, must carry out difference processing to signal, is stably up to differentiated flying quality.

(2) autocorrelation function and the partial autocorrelation function of calculating flying quality, Application of B IC decides the rank criterion, confirms model order p, q.Use least square method estimation model parameter phi ₁, φ ₂, L, φ _p, θ ₁, θ ₂, L, θ _q,

(3) model that utilizes foundation is finally exported result of calculation to calculating next data value constantly.

Supporting vector machine model is shown in (2) formula:

$f\left(x\right)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_{i}K(x_i,x)+b---\left(2\right)$

λ wherein _i, i=1,2, L, N are the optimum solution of optimization problem (3).

min $\frac{1}{2}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{\mathrm{\λ}}_{j}K(x_i,x_j)+\mathrm{\ϵ}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|{\mathrm{\λ}}_i\right|-\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_i{\mathrm{\λ}}_i$

(3)

s.t. $\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_i=0,$ -C≤λ _i≤C，i＝1，2，...，N

$K(x,x^{\′})=\mathrm{Exp}(-\frac{{\left|\right|x-x^{\′}\left|\right|}^2}{{\mathrm{\σ}}^2}),$ $b=y_j-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{\mathrm{\λ}}_{i}K(x_i\·x_j)-\mathrm{\ϵ},$ ε is the insensitive loss factor, and C is the punishment parameter, and wherein σ is the yardstick of kernel function.

Flying quality prediction flow process based on SVMs is as shown in Figure 3, and whole process also is divided into 2 parts: training process and data prediction.Training process is to utilize pretreated flying quality to find the solution optimization problem (3), obtains λ _i, i=1,2, L, N.Data prediction is according to formula (2) and passes through the λ that training process obtains _i, i=1,2, L, the value of N is calculated next data value constantly.Detailed process is:

(1) read pretreated flying quality after, at first need carry out phase space reconfiguration to flying quality, in this patent, select to utilize pseudo-adjacent method that flying quality is carried out phase space reconfiguration.

(2) (Sequential Minimal Optimization, SMO) solving-optimizing problem (3) calculates λ to utilize the minimum optimization algorithm of sequence _i, i=1,2, L, N.

(3) utilize next numerical value constantly of Model Calculation of setting up, finally export result of calculation.

(6) carry out fault analysis according to the value and the knowledge in the knowledge base of prediction, the result of the final prediction of output.

Claims (1)

1. the failure prediction method based on flying quality is characterized in that comprising the steps:

The first step makes up the flying quality storehouse, the storage flying quality;

Second step made up knowledge base, and storage is used for the rule of fault analysis, explained that aircraft breaks down if rule has represented that flying quality has met or exceeded certain value;

In the 3rd step, utilize the method for svd to dispel the noise of flying quality;

To flying quality sequence X={ x ₁, x ₂, Λ, x _NCarry out svd and may further comprise the steps, x wherein _iRepresent i flying quality, N representes natural number:

(1) structural matrix initial matrix

$A={\left[\begin{array}{cccc}{x}_1& x_2& \mathrm{\Λ}& x_k\\ x_2& x_3& \mathrm{\Λ}& x_{k+1}\\ M& M& O& M\\ x_n& x_{n+1}& \mathrm{\Λ}& x_N\end{array}\right]}_{n\×k}$

expression is not more than the integer of N+1/2; N=N-k+1, n and k are natural numbers;

(2) A is carried out svd

A＝UΓV ^T

U, V is orthogonal matrix, V ^TThe matrix transpose of expression V,

$\mathrm{\Γ}=\left[\begin{array}{cc}{\mathrm{\Π}}_{m\×m}& O_{m\×(k-m)}\\ O_{(n-m)\×m}& O_{(n-m)\×(k-m)}\end{array}\right]$

∏ wherein _{M * m}Be a diagonal form matrix, the element on the diagonal line is the singular value (α of matrix A ₁, α ₂, Λ, α _m); (α ₁, α ₂, Λ, α _m) satisfy α ₁>=α ₂>=Λ>=α _m>=0; O _{M * (k-m)}, O _{(n-m) * m}, O _{(n-m) * (k-m)}Be element and all be 0 matrix; M is a natural number, the order of representing matrix A;

(3) find the solution filtering matrix

Set filter threshold ε (0＜ε＜1), solving equation

$\frac{\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a_i}^2}{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{a_i}^2}=\mathrm{\ϵ}$

Can try to achieve positive integer l, the m-l that value is a less singular value is changed to 0 and obtains

${\mathrm{\Γ}}^1=\left[\begin{array}{cc}{\mathrm{\Π}}_{l\×l}& O_{l\×(k-l)}\\ O_{(n-l)\×l}& O_{(n-l)\×(k-l)}\end{array}\right];$

(4) filtering output

A ¹＝UΓ ¹V ^T

Matrix A then ¹In data be the flying quality of dispeling behind the noise;

In the 4th step, utilize largest Lyapunov exponent method to carry out chaotic characteristic and differentiate to dispeling flying quality behind the noise;

The 5th step, utilize prediction algorithm to realize prediction to flying quality, prediction algorithm has comprised autoregressive moving-average model and supporting vector machine model; When data are linear relationship, select autoregressive moving-average model; When data are nonlinear relationship, select supporting vector machine model;

Autoregressive moving-average model is shown in (1) formula:

S _t＝φ ₁S _t-1+φ ₂S _t-2+L+φ _pS _t-p+θ ₁ε _t-1+θ ₂ε _t-2+L+θ _qε _t-q+ε _t，t∈Z ⁺. (1)

Z wherein ⁺Be positive integer set, { ε _t, t ∈ Z ⁺Be the independent same distribution random series, ε _tNormal Distribution

The variance of expression normal distribution, p, q are model orders; Parameter phi ₁, φ ₂, L, φ _p, θ ₁, θ ₂, L, θ _q,

It is unknown parameter;

Flying quality prediction flow process based on autoregressive moving-average model may further comprise the steps:

(1) reading the flying quality of dispeling behind the noise, utilize distance of swimming method that flying quality is carried out stationary test, judge the stationarity of flying quality, if not steadily, must carry out difference processing to signal, is stably up to differentiated flying quality;

(2) autocorrelation function and the partial autocorrelation function of calculating flying quality, Application of B IC decides the rank criterion, confirms model order p, q; Use least square method estimation model parameter phi ₁, φ ₂, L, φ _p, θ ₁, θ ₂, L, θ _q,

(3) utilize next data value constantly of Model Calculation of setting up, finally export result of calculation;

Supporting vector machine model is shown in (2) formula:

$f\left(x\right)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_{i}K(x_i,x)+b---\left(2\right)$

λ wherein _i, i=1,2, L, N are the optimum solution of optimization problem (3).

min $\frac{1}{2}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{\mathrm{\λ}}_{j}K(x_i,x_j)+\mathrm{\ϵ}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|{\mathrm{\λ}}_i\right|-\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_i{\mathrm{\λ}}_i$

(3)

s.t. $\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_i=0,$ -C≤λ _i≤C，i＝1，2，...，N

$K(x,x^{\′})=\mathrm{Exp}(-\frac{{\left|\right|x-x^{\′}\left|\right|}^2}{{\mathrm{\σ}}^2}),$ $b=y_j-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{\mathrm{\λ}}_{i}K(x_i\·x_j)-\mathrm{\ϵ},$ ε is the insensitive loss factor, and C is the punishment parameter, and wherein σ is the yardstick of kernel function;

Flying quality prediction flow process based on SVMs may further comprise the steps:

(1) read the flying quality of dispeling behind the noise after, flying quality is carried out phase space reconfiguration;

(2) utilize the minimum optimization algorithm solving-optimizing problem (3) of sequence, calculate λ _i, i=1,2, L, N;

(3) utilize next numerical value constantly of Model Calculation of setting up, finally export result of calculation;

In the 6th step, carry out fault analysis according to the value and the knowledge in the knowledge base of prediction, the result of the final prediction of output.

Images