CN110531737A - Satellite executing mechanism method for diagnosing faults, system and medium based on mixed model - Google Patents
Satellite executing mechanism method for diagnosing faults, system and medium based on mixed model Download PDFInfo
- Publication number
- CN110531737A CN110531737A CN201910803662.5A CN201910803662A CN110531737A CN 110531737 A CN110531737 A CN 110531737A CN 201910803662 A CN201910803662 A CN 201910803662A CN 110531737 A CN110531737 A CN 110531737A
- Authority
- CN
- China
- Prior art keywords
- satellite
- neural network
- output
- state observer
- actuator
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
- G05B23/0227—Qualitative history assessment, whereby the type of data acted upon, e.g. waveforms, images or patterns, is not relevant, e.g. rule based assessment; if-then decisions
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
Abstract
Satellite executing mechanism method for diagnosing faults, system and medium based on mixed model, belong to field of space technology.The present invention solves the problems, such as that conventional satellite actuator failure diagnostic method can not be detected actuator glitch by uncertain and disturbing influence, by designing one group of state observer sensitivity particular satellite actuator failures, preferably, design neural network constructs state observer residual error on this basis, effectively reduces uncertain and disturbing influence;Finally design the detection of two layers of failure decision logic realization satellite actuator glitch.
Description
Technical field
The present invention relates to satellite executing mechanism method for diagnosing faults, system and media based on mixed model, belong to space
Technical field.
Background technique
Executing agency is the key building block of satellite attitude control system, and the failure of satellite executing mechanism will lead to satellite
The exception of attitude control system, or even the in-orbit safety of satellite is endangered, lead to whole star mission failure.Therefore, satellite executing mechanism
Fault diagnosis has important engineering significance.
In recent years, although the fault diagnosis of satellite executing mechanism has obtained extensive research, but generally use based on quantitative
The method for diagnosing faults of model is single diagnostic method, and the history data of system is not made full use of in diagnostic design.When
Preceding diagnostic method still has following main problem: the influence such as disturbance, uncertainty that satellite system is subject to can not a) be effectively treated,
Accurate diagnostic model is constructed to have difficulties;B) satellite actuator glitch can not be detected in time, be usually only able to achieve for compared with
For the diagnosis of catastrophe failure.
However, realizing the diagnosis of executing agency's glitch, recovery measure is taken early, and there is prior engineering value.
Summary of the invention
It is held technical problem solved by the present invention is having overcome the deficiencies of the prior art and provide the satellite based on mixed model
Row mechanism-trouble diagnostic method, system and medium, solving conventional satellite actuator failure diagnostic method can not be in diagnostic model
Uncertain and disturbing influence problem is reduced in modeling process, comprehensively utilizes state observer and neural network carries out satellite and executes machine
Structure Fault diagnosis design, to realize the early detection of satellite executing mechanism glitch.
The technical solution of the invention is as follows: the satellite executing mechanism method for diagnosing faults based on mixed model, including such as
Lower step:
According to satellite attitude control system mathematical model, the satellite attitude control system power of satellite actuator failures is established
Learn model;
According to satellite attitude control system kinetic model design point observer group, and make each satellite actuator at least
It is corresponding with a state observer;
By the input of satellite attitude control system output and input as state observer group, then by attitude control of satellite
It is poor that the output of system processed and the output of state observer group are made, and obtains fault residual;
Judge whether satellite actuator breaks down according to the fault residual and Fisrt fault decision function.
Further, the state observer group is
Wherein,It is the estimated value of attitude dynamics system state variables, z (t) is the state variable of state observer,For the derivative of state variable, u (t)=[Mx,My,Mz]TFor the control moment of satellite executing mechanism output, Φ (x) is satellite
The nonlinear function item of attitude dynamics subsystem, y (t)=[ωx,ωy,ωz]TThe satellite triaxial attitude angle measured for gyro
Speed;
F, M, T, G and N are the parameter matrixs of state observer, they meet following condition:
Further, the Fisrt fault decision function is
Wherein,(t1,t2) it is the time window of assessment, for finite value;r
It (t) is state observer residual error; Event does not occur for satellite executing mechanism
Output estimation error when barrier, λmax(*) is maximum eigenvalue, and sup is max-value function, and Q is post filtering matrix.
Further, described to obtain to carry out in accordance with the following steps after fault residual:
When establishing a neural network respectively for each state observer, and not broken down according to satellite actuator
Historical data all neural networks are trained, obtain trained neural network group;
By the input of satellite attitude control system output and input as trained neural network group, then by failure
It is poor that the output of residual error and trained neural network group is made, and obtains updating fault residual;
Fault residual is replaced with fault residual is updated, and judges whether satellite actuator is sent out in conjunction with the second failure decision function
Raw failure.
Further, the trained neural network group is
Wherein, Γ (κ)=1/ [1+exp (- κ)] is Sigmoid function;For neural network
Input vector,For all control moments in addition to i-th of actuator, yTIt (k) is the attitude angular velocity under normal condition;oj
(k) be j-th of hidden neuron output, a step be delayed olIt (k-1) is output after neural network structure unit, definition
Its initial value oj(0)=0;nxIt is the number of neural network input layer node;noIt is the number of neural network hidden node;wxIt is mind
Weight matrix through network input layer to hidden layer;wyIt is weight matrix of the neural network hidden layer to output layer;woIt is that neural network is hidden
Layer arrives the weight matrix of structural unit;bxAnd byIt is the threshold vector of neural network hidden layer and neural network output layer respectively.
Further, the second failure decision function isWherein,M is
Fixed sample size,The instruction of predicted boundary, J are passed through in kth sampled point for compensation residual errorthIt is worn for compensation residual error
The more number of predicted boundary ± l.
Further, the historical data when satellite actuator does not break down is the attitude angular velocity under normal condition
With all control moments in addition to i-th of actuator.
Satellite executing mechanism fault diagnosis system based on mixed model, for realizing the method, including
First module establishes the attitude control of satellite of satellite actuator failures according to satellite attitude control system mathematical model
System dynamics model processed;
Second module according to satellite attitude control system kinetic model design point observer group, and makes each satellite
Actuator is at least corresponding with a state observer;
Then the input of satellite attitude control system output and input as state observer group will by third module
It is poor that the output of satellite attitude control system and the output of state observer group are made, and obtains fault residual;
4th module judges whether satellite actuator occurs event according to the fault residual and Fisrt fault decision function
Barrier;
The state observer group isWherein,It is attitude dynamics system
The estimated value of state variable, z (t) are the state variables of state observer,For the derivative of state variable, u (t)=[Mx,My,
Mz]TFor the control moment of satellite executing mechanism output, Φ (x) is the nonlinear function item of Satellite Attitude Dynamics subsystem, y
(t)=[ωx,ωy,ωz]TThe satellite three-axis attitude angular speed measured for gyro;F, M, T, G, and N are the ginsengs of state observer
Matrix number, they meet following condition:
The Fisrt fault decision function isWherein,(t1,t2) it is the time window of assessment, for finite value;R (t) is state observation
Device residual error; Output estimation when not breaking down for satellite executing mechanism
Error, λmax(*) is maximum eigenvalue, and sup is max-value function, and Q is post filtering matrix.
Further, including
5th module establishes the attitude control of satellite of satellite actuator failures according to satellite attitude control system mathematical model
System dynamics model processed;
6th module according to satellite attitude control system kinetic model design point observer group, and makes each satellite
Actuator is at least corresponding with a state observer;
Then the input of satellite attitude control system output and input as state observer group will by the 7th module
It is poor that the output of satellite attitude control system and the output of state observer group are made, and obtains fault residual;
8th module establishes a neural network for each state observer respectively, and not according to satellite actuator
Historical data when breaking down is trained all neural networks, obtains trained neural network group;
9th module, by the input of satellite attitude control system output and input as trained neural network group,
Then it is poor the output of fault residual and trained neural network group to be made, and obtains updating fault residual;
Tenth module replaces fault residual with fault residual is updated, and judges that satellite is held in conjunction with the second failure decision function
Whether row device breaks down;
Historical data when the satellite actuator does not break down for the attitude angular velocity under normal condition and is removed i-th
All control moments other than actuator;
The trained neural network group is Wherein, Γ (κ)=1/ [1+exp (- κ)] is Sigmoid function;
For the input vector of neural network,For all control moments in addition to i-th of actuator, yT(k) under normal condition
Attitude angular velocity;oj(k) be j-th of hidden neuron output, a step be delayed olIt (k-1) is by neural network structure unit
Output afterwards defines its initial value oj(0)=0;nxIt is the number of neural network input layer node;noIt is neural network hidden node
Number;wxIt is weight matrix of the neural network input layer to hidden layer;wyIt is weight matrix of the neural network hidden layer to output layer;
Wo is weight matrix of the neural network hidden layer to structural unit;bxAnd byIt is neural network hidden layer and neural network output layer respectively
Threshold vector;
The second failure decision function isWherein,M is fixed sampling
Size,The instruction of predicted boundary, J are passed through in kth sampled point for compensation residual errorthPredicted boundary is passed through for compensation residual error
The number of ± l.
A kind of computer readable storage medium, is stored thereon with computer program, which is executed by processor
When, realize the method the step of.
The advantages of the present invention over the prior art are that:
1) present invention is by one group of state observer of design and neural network, and utilizes historical data training neural network structure
Observer residual error is built, achievees the purpose that reducing Satellite Attitude Control System uncertainty and disturbance influences diagnosis;
2) present invention realizes the early detection of satellite executing mechanism glitch by two layers of failure decision logic of design.
Detailed description of the invention
Fig. 1 is the satellite executing mechanism fault diagnosis block diagram based on mixed model;
Fig. 2 is the state observer residual error under satellite actuator unfaulty conditions;
Fig. 3 is the compensation residual error under satellite actuator unfaulty conditions;
Fig. 4 is that state observer residual error assesses curve;
Fig. 5 is the assessment curve that neural network compensates residual error.
Specific embodiment
It is further described with reference to the accompanying drawings of the specification.
Such as Fig. 1, the satellite executing mechanism method for diagnosing faults based on mixed model, one of method includes the following steps:
One, according to satellite attitude control system mathematical model, the satellite attitude control system of satellite actuator failures is established
Kinetic model:
Satellite Attitude Dynamics state space equation when establishing satellite executing mechanism failure is as follows,
Wherein, x (t)=[ωx,ωy,ωz]TFor the state variable of Satellite Attitude Dynamics system;U (t)=[Mx,My,
Mz]TFor the control moment of satellite executing mechanism output;Y (t)=[ωx,ωy,ωz]TThe satellite triaxial attitude angle measured for gyro
Speed;D (t)=[Tdx,Tdy,Tdz]TIt is unknown disturbance torque caused by being influenced due to space environment;F (t)=[f1(t),f2
(t),f3(t)]TFor fault vectors, fi(t), i=1,2,3 is corresponding i-th axis actuator failures.η (x, u, t) and b (t) is respectively
For the uncertainty and gyroscopic drift error term of satellite dynamics system;Φ (x) is the non-thread of Satellite Attitude Dynamics subsystem
Property function item.
B, C, E, L are respectively the input matrix, output matrix, disturbance distribution matrix of Satellite Attitude Dynamics system, and
Failure distribution matrix:
Failure distribution matrix can further be expressed as L=[L1,L2,L3] form, wherein Li, i=1,2,3 is failure point
I-th column of cloth matrix L.
Two, according to satellite attitude control system kinetic model design point observer group, and make each satellite actuator
It is at least corresponding with a state observer:
Since the actuator on three axis of satellite is likely to occur failure, structure can be distinguished for each failure
A state observer is made, the state observer only other axial failures and disturbance decoupling to the axis Fault-Sensitive are made.
Specifically, for i-th of actuator failures, it can establish i-th of state observer as follows:
It is the estimated value of attitude dynamics system state variables;Z (t) is the state variable of state observer.
F, M, T, G and N are the parameter matrixs of state observer, they meet the following conditions:
Meanwhile to keep i-th of state observer sensitive to i-th of actuator failures and decoupling other axial failures and disturb
It is dynamic, constraint: TL need to be meti≠ 0 and TLj=0, j ≠ i, and then determine each parameter matrix of state observer.
State observer residual error can be defined asWherein Q represents post filtering matrix, it can be selected
It is taken as Q=TLi。
Three, by the input of satellite attitude control system output and input as state observer group, then by Satellite Attitude
It is poor that the output of state control system and the output of state observer group are made, and obtains fault residual;
Four, judge whether satellite actuator breaks down according to the fault residual and Fisrt fault decision function:
Although state observer residual error is influenced by uncertain and disturbance, for the detection sensitivity of glitch.But it is right
It can reliably be detected in more serious failure.The valuation functions of structural regime observer residual error are as follows:
Wherein, (t1,t2) it is the time window of assessment, for finite value.
Threshold value isWhereinDo not occur for satellite executing mechanism
Output estimation error when failure, λmax(*) is maximum eigenvalue, and sup is maximum.
Therefore, the failure decision logic based on state observer residual error is as follows:
Satellite executing mechanism method for diagnosing faults based on mixed model, the two of method are that event will be obtained in one of method
The step of after barrier residual error, replaces are as follows:
One, a neural network is established for each state observer respectively, and event is not occurred according to satellite actuator
Historical data when barrier is trained all neural networks, obtains trained neural network group:
Due to foundation state observer can not to all disturbances and it is uncertain decouple, so even if executing agency without reason
Barrier, obtained state observer residual error is also not zero due to being disturbed and uncertainty influences, to influence for executing
The detection of mechanism glitch.The residual error of state observer is modeled for this purpose, establishing neural network, reduce disturbance and is not known
The influence of property.
One neural network is established for each state observer respectively and is trained.Specifically, for i-th
State observer establishes the i-th neural networkWhereinIt is the input vector of neural network;θ is the design of neural network
Parameter, including weight matrix and threshold vector.
The target output of the neural network is the residual error under state observer i normal condition, and the input of neural network is positive
Attitude angular velocity y under normal stateT(k) all control moments and in addition to i-th of actuator.Note except i-th actuator with
Outer all control moments areThe then input of the neural network isThe target of neural network exports
For r (k), wherein k is the discretization of continuous time t, indicates sampled point.
Remember vectorI-th of element beThen establish neural networkMathematical expression it is as follows:
Wherein, Γ (κ)=1/ [1+exp (- κ)] is Sigmoid function;oj(k) be j-th of hidden neuron output,
Corresponding step delay oj(k-1) it is output after structural unit, defines its initial value oj(0)=0.nxIt is input layer
Number;noIt is the number of hidden node;wxIt is weight matrix of the input layer to hidden layer;wyIt is weight square of the hidden layer to output layer
Battle array;Wo is weight matrix of the hidden layer to structural unit;bxAnd byIt is the threshold vector of hidden layer and output layer respectively.
Two, by the input of satellite attitude control system output and input as trained neural network group, then will
It is poor that the output of fault residual and trained neural network group is made, and obtains updating fault residual:
Off-line training is carried out to neural network using the historical data under normal condition.It can be obtained after the completion of training all
Neural network design parameter estimated valueAnd the estimated value of state observer residual errorFrom
And available compensation residual error
Three, fault residual is replaced with update fault residual, and judges that satellite actuator is in conjunction with the second failure decision function
It is no to break down:
For the confidence level 1- α that neural network gives, the true value of state observer residual error is present in pre- with confidence level 1- α
It surveys in section, therefore
prob{-l≤re≤ l | f=0 }=1- α (8)
Wherein,It is predicted boundary.θ*The true value of representation parameter θ,It is θ*'s
Least-squares estimation can be calculated by the back-propagation algorithm of neural network.It is that freedom degree isT distribution
The quantile of α/2, whereinIt is the number of parameter θ,It is the number of trained sampled data;It is to be calculated by training data
Variances sigma2Unbiased esti-mator;It is the gradient function about θ * of neural network output;It is to be calculated by training data
The Jacobian matrix of the neural network model arrived.Mathematically, these amounts calculate as follows:
Since under executing agency's fault-free, compensation residual error, which still has the probability of α to pass through predicted boundary, to be caused to report by mistake.For this purpose, structure
The valuation functions for making compensation residual error are as follows:
Wherein, m is fixed sample size,Represent the finger that compensation residual error passes through predicted boundary in kth sampled point
Show, i.e.,
Therefore, the failure decision logic based on compensation residual error is as follows:
Wherein, JthIt is threshold value is to compensate residual error to pass through the number of predicted boundary ± l.
Satellite executing mechanism fault diagnosis system based on mixed model, for realizing one of above method, including
First module establishes the attitude control of satellite of satellite actuator failures according to satellite attitude control system mathematical model
System dynamics model processed;
Second module according to satellite attitude control system kinetic model design point observer group, and makes each satellite
Actuator is at least corresponding with a state observer;
Then the input of satellite attitude control system output and input as state observer group will by third module
It is poor that the output of satellite attitude control system and the output of state observer group are made, and obtains fault residual;
4th module judges whether satellite actuator occurs event according to the fault residual and Fisrt fault decision function
Barrier;
The state observer group isWherein,It is attitude dynamics system
The estimated value of state variable, z (t) are the state variables of state observer,For the derivative of state variable, u (t)=[Mx,My,
Mz]TFor the control moment of satellite executing mechanism output, Φ (x) is the nonlinear function item of Satellite Attitude Dynamics subsystem, y
(t)=[ωx,ωy,ωz]TThe satellite three-axis attitude angular speed measured for gyro;F, M, T, G, and N are the ginsengs of state observer
Matrix number, they meet following condition:
The Fisrt fault decision function isWherein,(t1,t2) it is the time window of assessment, for finite value;R (t) is state observation
Device residual error; Output when not breaking down for satellite executing mechanism is estimated
Count error, λmax(*) is maximum eigenvalue, and sup is max-value function, and Q is post filtering matrix.
Satellite executing mechanism fault diagnosis system based on mixed model, for realizing the two of method, including
5th module establishes the attitude control of satellite of satellite actuator failures according to satellite attitude control system mathematical model
System dynamics model processed;
6th module according to satellite attitude control system kinetic model design point observer group, and makes each satellite
Actuator is at least corresponding with a state observer;
Then the input of satellite attitude control system output and input as state observer group will by the 7th module
It is poor that the output of satellite attitude control system and the output of state observer group are made, and obtains fault residual;
8th module establishes a neural network for each state observer respectively, and not according to satellite actuator
Historical data when breaking down is trained all neural networks, obtains trained neural network group;
9th module, by the input of satellite attitude control system output and input as trained neural network group,
Then it is poor the output of fault residual and trained neural network group to be made, and obtains updating fault residual;
Tenth module replaces fault residual with fault residual is updated, and judges that satellite is held in conjunction with the second failure decision function
Whether row device breaks down;
Historical data when the satellite actuator does not break down for the attitude angular velocity under normal condition and is removed i-th
All control moments other than actuator;
The trained neural network group is Wherein, Γ (κ)=1/ [1+exp (- κ)] is Sigmoid function;
For the input vector of neural network,For all control moments in addition to i-th of actuator, yT(k) under normal condition
Attitude angular velocity;oj(k) be j-th of hidden neuron output, a step be delayed olIt (k-1) is by neural network structure unit
Output afterwards defines its initial value oj(0)=0;nxIt is the number of neural network input layer node;noIt is neural network hidden node
Number;wxIt is weight matrix of the neural network input layer to hidden layer;wyIt is weight matrix of the neural network hidden layer to output layer;
woIt is weight matrix of the neural network hidden layer to structural unit;bxAnd byIt is neural network hidden layer and neural network output layer respectively
Threshold vector;
The second failure decision function isWherein,M is fixed sampling
Size,The instruction of predicted boundary, J are passed through in kth sampled point for compensation residual errorthPredicted boundary is passed through for compensation residual error
The number of ± l.
A kind of computer readable storage medium, is stored thereon with computer program, which is executed by processor
When, realize the above method the step of.
A specific embodiment of the invention.
Below using certain low orbit satellite attitude control system as object, by simulation example, the verifying present invention executes machine to satellite
The diagnosis of structure glitch.
(1) state observer is established
According to the actuator in three axial directions of satellite, a state observer is established respectively, realizes the state observer pair
The Fault-Sensitive of corresponding axis actuator.By taking x-axis actuator failures as an example, according to Satellite Attitude Dynamics state space equation (1),
The 1st state observer is established, corresponding state Observer Design constraint is as follows:
Wherein,L1=[0.0011 0 0]T;L2
=[0 0.0013 0]T;L3=[0 0 0.0009]T;D (t)=[1.4 × 10-5,1.5×10-5,1.6×10-5]T×sinω0t
N·m;η (x, u, t)=[2 × 10-6sin(0.1t),3×10-6cos(0.5t),5×10-6cos(0.3t)]T;B (t)=[1 ,-
1,1]T×6/3600×pi/180。
Thus, it is possible to obtain state observer residual errorAs shown in Fig. 2, its
State observer residual error in the case of not breaking down for actuator.
(2) neural network is established
For the 1st state observer, corresponding neural network is established
It is sampled with data of the sampling time 0.1s to 500s to 2000s, and establishes data setWherein,
{ r (k), k=1,2 ... } is the residual error of state observer 1 under non-failure conditions, and the target as neural network exports;It is the control moment and attitude angle under non-failure conditions in addition to the 1st actuator
The vector that speed is constituted, the input as neural network;Neural network mathematical model (4) and (5) are carried out using these data
Training, wherein the hidden node number for choosing neural network is 5.
The neural network parameter that can be obtained after the completion of trainingAnd obtain compensation residual errorAs shown in figure 3, it is the compensation residual error of neural network in the case of actuator does not break down.
(3) failure decision is established
To state observer 1, the valuation functions of the residual error shaped like (6) are established | | r (t) | |RMS, and according to state observer
The failure decision logic (7) of residual error carries out the fault detection of actuator, and wherein threshold value is
When x-axis actuator is in tfGlitch f occurs when=3000s1(t)=0.8 × 10-5When, by the residual of state observer 1
The obtained failure detection result of difference is as shown in Figure 4.
For the corresponding neural network that state observer 1 is established, confidence alpha=0.01 is taken, it can be according to formula (8)~formula
(11) compensation residual error r is determinede, wherein
Further, the failure decision logic (14) according to the valuation functions (12) of compensation residual error and based on compensation residual error carries out
Actuator failures based on compensation residual error detect, wherein m=7, Jth=5.There are 5 in the 7 compensation residual values sampled in succession
Surmount predicted boundary, then shows that failure has occurred in actuator.The fault detection knot for thering is neural network compensation residual error to obtain accordingly
Fruit is as shown in Figure 5.
To sum up, from fig. 4, it can be seen that the failure decision logic based on state observer residual error fails obviously to detect satellite x-axis and hold
The glitch that row device occurs;And the failure decision logic based on compensation residual error as shown in Figure 5 effectively detects that satellite x-axis executes
The glitch that device occurs in 3000s.The above simulation result illustrates to execute based on state observer and the satellite of neural network
The validity of mechanism-trouble diagnostic method.
The content that description in the present invention is not described in detail belongs to the well-known technique of those skilled in the art.
Claims (10)
1. the satellite executing mechanism method for diagnosing faults based on mixed model, which comprises the steps of:
According to satellite attitude control system mathematical model, the satellite attitude control system kinetic simulation of satellite actuator failures is established
Type;
According to satellite attitude control system kinetic model design point observer group, and make each satellite actuator at least with one
A state observer is corresponding;
By the input of satellite attitude control system output and input as state observer group, then by satellite gravity anomaly system
It is poor that the output of system and the output of state observer group are made, and obtains fault residual;
Judge whether satellite actuator breaks down according to the fault residual and Fisrt fault decision function.
2. the satellite executing mechanism method for diagnosing faults according to claim 1 based on mixed model, it is characterised in that: institute
Stating state observer group is
Wherein,It is the estimated value of attitude dynamics system state variables, z (t) is the state variable of state observer,For
The derivative of state variable, u (t)=[Mx,My,Mz]TFor the control moment of satellite executing mechanism output, Φ (x) is that the attitude of satellite is dynamic
The nonlinear function item of mechanics subsystem, y (t)=[ωx,ωy,ωz]TThe satellite three-axis attitude angular speed measured for gyro;
F, M, T, G and N are the parameter matrixs of state observer, they meet following condition:
3. the satellite executing mechanism method for diagnosing faults according to claim 2 based on mixed model, it is characterised in that: institute
Stating Fisrt fault decision function is
Wherein,T0=t2-t1, (t1,t2) it is the time window of assessment, for finite value;R (t) is
State observer residual error;When not breaking down for satellite executing mechanism
Output estimation error, λmax(*) is maximum eigenvalue, and sup is max-value function, and Q is post filtering matrix.
4. the satellite executing mechanism method for diagnosing faults according to claim 1 based on mixed model, which is characterized in that institute
It states to carry out in accordance with the following steps after obtaining fault residual:
Going through when establishing a neural network respectively for each state observer, and not broken down according to satellite actuator
History data are trained all neural networks, obtain trained neural network group;
By the input of satellite attitude control system output and input as trained neural network group, then by fault residual
It is poor that output with trained neural network group is made, and obtains updating fault residual;
Fault residual is replaced with fault residual is updated, and judges whether satellite actuator occurs event in conjunction with the second failure decision function
Barrier.
5. the satellite executing mechanism method for diagnosing faults according to claim 4 based on mixed model, it is characterised in that: institute
Stating trained neural network group is
Wherein, Γ (κ)=1/ [1+exp (- κ)] is Sigmoid function;For neural network input to
Amount,For all control moments in addition to i-th of actuator, yTIt (k) is the attitude angular velocity under normal condition;ojIt (k) is
The output of j hidden neuron, step delay ol(k-1) it is output after neural network structure unit, defines its initial value oj
(0)=0;nxIt is the number of neural network input layer node;noIt is the number of neural network hidden node;wxIt is that neural network is defeated
Enter layer to hidden layer weight matrix;wyIt is weight matrix of the neural network hidden layer to output layer;woIt is neural network hidden layer to structure
The weight matrix of unit;bxAnd byIt is the threshold vector of neural network hidden layer and neural network output layer respectively.
6. the satellite executing mechanism method for diagnosing faults according to claim 5 based on mixed model, it is characterised in that: institute
Stating the second failure decision function isWherein,M is fixed sample size,The instruction of predicted boundary, J are passed through in kth sampled point for compensation residual errorthPass through predicted boundary ± l's for compensation residual error
Number.
7. the satellite executing mechanism method for diagnosing faults according to claim 4 based on mixed model, it is characterised in that: institute
Historical data when satellite actuator does not break down is stated for the attitude angular velocity under normal condition and in addition to i-th of actuator
All control moments.
8. the satellite executing mechanism fault diagnosis system based on mixed model, described in any item for realizing claims 1 to 3
Method, it is characterised in that: including
First module establishes the satellite gravity anomaly system of satellite actuator failures according to satellite attitude control system mathematical model
System kinetic model;
Second module according to satellite attitude control system kinetic model design point observer group, and executes each satellite
Device is at least corresponding with a state observer;
Third module, by the input of satellite attitude control system output and input as state observer group, then by satellite
It is poor that the output of attitude control system and the output of state observer group are made, and obtains fault residual;
4th module judges whether satellite actuator breaks down according to the fault residual and Fisrt fault decision function;
The state observer group isWherein,It is attitude dynamics system mode
The estimated value of variable, z (t) are the state variables of state observer,For the derivative of state variable, u (t)=[Mx,My,Mz]T
For the control moment of satellite executing mechanism output, Φ (x) is the nonlinear function item of Satellite Attitude Dynamics subsystem, y (t)=
[ωx,ωy,ωz]TThe satellite three-axis attitude angular speed measured for gyro;F, M, T, G, and N are the parameter squares of state observer
Battle array, they meet following condition:
The Fisrt fault decision function isWherein,T0=
t2-t1, (t1,t2) it is the time window of assessment, for finite value;R (t) is state observer residual error;Output estimation error when not breaking down for satellite executing mechanism,
λmax(*) is maximum eigenvalue, and sup is max-value function, and Q is post filtering matrix.
9. the satellite executing mechanism fault diagnosis system based on mixed model, for realizing described in any one of claim 4~7
Method, it is characterised in that: including
5th module establishes the satellite gravity anomaly system of satellite actuator failures according to satellite attitude control system mathematical model
System kinetic model;
6th module according to satellite attitude control system kinetic model design point observer group, and executes each satellite
Device is at least corresponding with a state observer;
7th module, by the input of satellite attitude control system output and input as state observer group, then by satellite
It is poor that the output of attitude control system and the output of state observer group are made, and obtains fault residual;
8th module is established a neural network for each state observer respectively, and is not occurred according to satellite actuator
Historical data when failure is trained all neural networks, obtains trained neural network group;
9th module, by the input of satellite attitude control system output and input as trained neural network group, then
It is poor that the output of fault residual and trained neural network group is made, and obtains updating fault residual;
Tenth module replaces fault residual with fault residual is updated, and judges satellite actuator in conjunction with the second failure decision function
Whether break down;
Historical data when the satellite actuator does not break down is for the attitude angular velocity under normal condition and except i-th of execution
All control moments other than device;
The trained neural network group is Wherein, Γ (κ)=1/ [1+exp (- κ)] is Sigmoid function;
For the input vector of neural network,For all control moments in addition to i-th of actuator, yTIt (k) is the appearance under normal condition
State angular speed;oj(k) be j-th of hidden neuron output, a step be delayed olIt (k-1) is after neural network structure unit
Output, define its initial value oj(0)=0;nxIt is the number of neural network input layer node;noIt is neural network hidden node
Number;wxIt is weight matrix of the neural network input layer to hidden layer;wyIt is weight matrix of the neural network hidden layer to output layer;wo
It is weight matrix of the neural network hidden layer to structural unit;bxAnd byIt is neural network hidden layer and neural network output layer respectively
Threshold vector;
The second failure decision function isWherein,M is fixed sample size,The instruction of predicted boundary, J are passed through in kth sampled point for compensation residual errorthPass through predicted boundary ± l's for compensation residual error
Number.
10. a kind of computer readable storage medium, is stored thereon with computer program, it is characterised in that: the computer program quilt
When processor executes, the step of realizing one of claim 1~7 the method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910803662.5A CN110531737A (en) | 2019-08-28 | 2019-08-28 | Satellite executing mechanism method for diagnosing faults, system and medium based on mixed model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910803662.5A CN110531737A (en) | 2019-08-28 | 2019-08-28 | Satellite executing mechanism method for diagnosing faults, system and medium based on mixed model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110531737A true CN110531737A (en) | 2019-12-03 |
Family
ID=68665048
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910803662.5A Pending CN110531737A (en) | 2019-08-28 | 2019-08-28 | Satellite executing mechanism method for diagnosing faults, system and medium based on mixed model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110531737A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111290366A (en) * | 2020-02-12 | 2020-06-16 | 北京科技大学顺德研究生院 | Multi-fault diagnosis method for spacecraft attitude control system |
CN111472468A (en) * | 2020-04-17 | 2020-07-31 | 南通大学 | High-rise building damping control method based on distributed fault diagnosis and collaborative fault tolerance |
CN112036440A (en) * | 2020-07-31 | 2020-12-04 | 山东科技大学 | Satellite attitude control system fault diagnosis and early warning method based on random forest |
CN112947391A (en) * | 2021-04-05 | 2021-06-11 | 西北工业大学 | Flight control system actuator tiny fault diagnosis method based on TOMFIR residual error |
CN114115185A (en) * | 2021-11-15 | 2022-03-01 | 哈尔滨工业大学 | Fault detection threshold calculation method based on interval operation |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102176159A (en) * | 2011-02-28 | 2011-09-07 | 哈尔滨工业大学 | Satellite attitude control system failure diagnosis device and method based on state observer and equivalent space |
CN106292681A (en) * | 2016-09-19 | 2017-01-04 | 北京航空航天大学 | A kind of satellite Active Fault-tolerant Control Method distributed based on observer and On-line Control |
CN107121961A (en) * | 2017-05-25 | 2017-09-01 | 北京航空航天大学 | A kind of spacecraft attitude fault tolerant control method based on iterative learning interference observer |
CN109808918A (en) * | 2019-01-30 | 2019-05-28 | 上海卫星工程研究所 | Double super satellite load cabin interference compensation methods neural network based |
-
2019
- 2019-08-28 CN CN201910803662.5A patent/CN110531737A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102176159A (en) * | 2011-02-28 | 2011-09-07 | 哈尔滨工业大学 | Satellite attitude control system failure diagnosis device and method based on state observer and equivalent space |
CN106292681A (en) * | 2016-09-19 | 2017-01-04 | 北京航空航天大学 | A kind of satellite Active Fault-tolerant Control Method distributed based on observer and On-line Control |
CN107121961A (en) * | 2017-05-25 | 2017-09-01 | 北京航空航天大学 | A kind of spacecraft attitude fault tolerant control method based on iterative learning interference observer |
CN109808918A (en) * | 2019-01-30 | 2019-05-28 | 上海卫星工程研究所 | Double super satellite load cabin interference compensation methods neural network based |
Non-Patent Citations (1)
Title |
---|
程瑶: "卫星姿态控制***的混合故障诊断方法研究", 《中国博士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111290366A (en) * | 2020-02-12 | 2020-06-16 | 北京科技大学顺德研究生院 | Multi-fault diagnosis method for spacecraft attitude control system |
CN111290366B (en) * | 2020-02-12 | 2022-05-27 | 北京科技大学顺德研究生院 | Multi-fault diagnosis method for attitude control system of spacecraft |
CN111472468A (en) * | 2020-04-17 | 2020-07-31 | 南通大学 | High-rise building damping control method based on distributed fault diagnosis and collaborative fault tolerance |
CN112036440A (en) * | 2020-07-31 | 2020-12-04 | 山东科技大学 | Satellite attitude control system fault diagnosis and early warning method based on random forest |
CN112947391A (en) * | 2021-04-05 | 2021-06-11 | 西北工业大学 | Flight control system actuator tiny fault diagnosis method based on TOMFIR residual error |
CN114115185A (en) * | 2021-11-15 | 2022-03-01 | 哈尔滨工业大学 | Fault detection threshold calculation method based on interval operation |
CN114115185B (en) * | 2021-11-15 | 2023-07-18 | 哈尔滨工业大学 | Fault detection threshold value calculation method based on interval operation |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110531737A (en) | Satellite executing mechanism method for diagnosing faults, system and medium based on mixed model | |
CN101697079B (en) | Blind system fault detection and isolation method for real-time signal processing of spacecraft | |
Odendaal et al. | Actuator fault detection and isolation: An optimised parity space approach | |
Napolitano et al. | Kalman filters and neural-network schemes for sensor validation in flight control systems | |
Cheng et al. | A combined model-based and intelligent method for small fault detection and isolation of actuators | |
CN111290366B (en) | Multi-fault diagnosis method for attitude control system of spacecraft | |
Sobhani-Tehrani et al. | Hybrid fault diagnosis of nonlinear systems using neural parameter estimators | |
CN108540311B (en) | Fault detection deep learning network processing method and device of satellite actuating mechanism | |
CN115857529B (en) | Method for reconstructing actuator fault of spacecraft attitude control system | |
CN109813309A (en) | A kind of six gyro redundance type Strapdown Inertial Navigation System Dual Failures partition methods | |
Le et al. | A residual based adaptive unscented Kalman filter for fault recovery in attitude determination system of microsatellites | |
Yuan et al. | Fault detection, isolation, and reconstruction for satellite attitude sensors using an adaptive hybrid method | |
CN106742068A (en) | A kind of method for diagnosing satellite attitude control system unknown failure | |
Barzegar et al. | Fault diagnosis and prognosis for satellite formation flying: A survey | |
Xin et al. | A Feed‐Forward Wavelet Neural Network Adaptive Observer‐Based Fault Detection Technique for Spacecraft Attitude Control Systems | |
Li et al. | Aircraft sensor fault diagnosis method based on residual antagonism transfer learning | |
CN116678434A (en) | Three-self laser inertial measurement unit fault detection method based on interval value confidence rule base | |
CN109753045B (en) | Spacecraft attitude control system fault diagnosis method based on BP neural network | |
Okatan et al. | Kalman filter innovation sequence based fault detection in LEO satellite attitude determination and control system | |
Lee et al. | Migrating fault trees to decision trees for real time fault detection on international space station | |
Yen et al. | Winner take all experts network for sensor validation | |
Shi et al. | Research on Fault Diagnosis of Three Degrees of Freedom Gyroscope Redundant System | |
Li et al. | Fault diagnosis and reconstruction for sensor of aeroengine control system based on AANN network | |
Patton et al. | A Monte Carlo analysis and design for FDI of a satellite attitude control system | |
Notaro et al. | HW VS SW sensor redundancy: Fault detection and isolation observer based approaches for inertial measurement units |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191203 |