CN101481019A

CN101481019A - Fault tolerant observing method of sensor for satellite attitude control system

Info

Publication number: CN101481019A
Application number: CNA2009100608162A
Authority: CN
Inventors: 魏蛟龙; 岑朝辉; 蒋睿
Original assignee: Huazhong University of Science and Technology
Current assignee: Huazhong University of Science and Technology
Priority date: 2009-02-20
Filing date: 2009-02-20
Publication date: 2009-07-15
Anticipated expiration: 2029-02-20
Also published as: CN101481019B

Abstract

The invention provides a sensor fault tolerance observing method for a satellite attitude control system. Firstly, the control system carries out the sensor fault detecting isolation; if a first sensor is judged to have faults according to an isolation result, a second KX observer is selected to observe; and if a second sensor is judged to have faults, a first KX observer is selected to observe. The first or second KX observer has the ensuring method that if the second or first sensor has faults, the control system is analyzed into a normal subsystem and a fault subsystem by using system feedback stable control as a target, and a low-dimension KX function observer is designed by respectively aiming at the two subsystems to combine and obtain a system KX observer with fault tolerance performance. The two parallel KX observers are designed and are fused into a controlling and switching flow; and when the partial sensor output is unreliable, the satellite attitude control system still can ensure the fault tolerance observation of system rest part observed quantity to ensure the closed loop control integrality of a failure system.

Description

Sensor fault-tolerant observation method for satellite attitude control system

Technical Field

The invention relates to the field of spacecraft fault diagnosis and fault-tolerant control, in particular to a fault-tolerant observer method for a sensor fault of a satellite attitude control system.

Background

As a typical dynamic control system, a satellite attitude control system is one of important components for determining the normal operation of a satellite in space. Due to the particularity of the field of aerospace engineering, the requirement on the reliability of the aerospace engineering is extremely high, and the safety of the whole system is possibly influenced when any link in the system breaks down, so that the satellite attitude control system is indispensable for monitoring the real-time state and ensuring the continuous operation (namely integrity) of the system when the satellite attitude control system breaks down. For a general control system, monitoring signals of the system mainly comprise control input and measurement output, and in order to realize state feedback control or form a residual error based on a state for diagnosis, a model observer method is generally adopted to track the state of the system based on the control input and the measurement output. For a satellite attitude control system, because the satellite attitude control system is limited by on-satellite conditions, the redundant sensor resources are less, and analytic redundant relations exist between the measured output quantities, when the traditional model observer method is used for observing the attitude state, the problem that observation is invalid due to the fact that partial sensors are prone to faults exists. Observation failure can cause incorrect state tracking, further leading to unstable closed loop feedback, and certainly, the remaining normal sensors cannot be utilized for fault isolation to ensure the integrity of the system. Therefore, a fault-tolerant observer method is designed, when the output of the closed-loop control system is unreliable due to the fault of part of sensors, the observer can still observe and obtain part of state quantity by using the output of the rest normal sensors, and therefore the system is stable, and part of diagnosis functions are realized. The method has important engineering application value for improving the fault tolerance of the satellite attitude control system and saving the redundant resources of the sensor hardware.

At present, in order to realize fault-tolerant observation of control system sensor faults, two types of approaches exist: firstly, estimating and compensating based on faults; and secondly, a method based on linear transformation. The first category of methods compensates the system state based on an estimate of the fault residual, and thus the fault must be limited to a specific fault and must be an accurate real-time estimate, which presupposes that the description of both the fault and the model in the observer must be accurate. For some sensor faults with nonlinear characteristics, the modeling process is complex, and the premise of identifying the fault size must be met, so that the method is not suitable for fault-tolerant observation of uncertain and complex sensor faults. The second method has the main idea that firstly, the system is linearly transformed into an equivalent triangle or a form capable of observing subsystem decomposition, and the observer design is carried out on the subsystem in the new system, so that the output of the observer is decoupled from the output of the fault sensor, and fault-tolerant observation can be further realized along with partial states of the system. If the original system is transformed and then the observer of the observable subsystem is designed, such as the dedicated observer method, although the requirement for detecting the fault residual error of the sensor can be met, the feedback stability of the closed loop under the fault condition of the whole original system is not necessarily met, namely the feedback constraint conditions of controllability and stability cannot be given, so that only fault detection and isolation (FDD) is realized, and the stability of the closed loop is not considered. If the original system is transformed into an equivalent standard triangle and then the observer design of the subsystem is carried out, such as a fault-tolerant dimension-reduction observer method, although the method is insensitive to the type of the sensor fault, the premise of carrying out the subsystem decomposition is that the sensor output is in fault, a transformation matrix is designed aiming at the fault sensor output, and the method is aiming at an open-loop system and is not applicable to feedback control.

Disclosure of Invention

The invention provides a sensor fault-tolerant observation method for a satellite attitude control system, which not only keeps the stability of a closed-loop feedback system, but also has a high-efficiency sensor fault-tolerant observation effect.

A sensor fault-tolerant observation method for a satellite attitude control system specifically comprises the following steps: and a fault diagnosis module is adopted to detect and isolate the fault of the sensor of the satellite attitude control system, if the isolation result judges that the first sensor has the fault, a second KX observer is selected for observation, and if the second sensor has the fault, the first KX observer is selected for observation.

The first KX observer is generated as follows: taking the first sensor as a normal sensor and the second sensor as a fault, transforming and decomposing the satellite attitude control system into a normal subsystem and a fault subsystem, respectively constructing a normal KX observer and a fault KX observer aiming at the normal subsystem and the fault subsystem, and then combining the normal KX observer and the fault KX observer into a first KX observer;

the second KX observer is generated as follows: and generating a second KX observer in the same way as the first KX observer when the first sensor is in fault and the second sensor is normal.

The generating step of the first KX observer specifically comprises:

(1) a system decomposition step:

decomposing the satellite attitude control system to generate a normal subsystem

Is a normal subsystem state variable

Derivative of (2)

And normal subsystem state variables

The linear coefficient of (a) is,

is a normal subsystem state variableDerivative of (2)

Linear coefficients with the normal subsystem control input variable u,

measuring an output variable for a first sensor

And normal subsystem state variables

The linear coefficient of (a);

and a fault sub-system

As faulty subsystem state variablesDerivative of (2)

And fault subsystem state variables

The linear coefficient of (a) is,

as faulty subsystem state variables

Derivative of (2)

Linear coefficients with the fault subsystem control form input variable u',

outputting the variable for the second sensor form

And fault subsystem state variables

The linear coefficient of (a) is,

outputting the variable for the second sensor form

And normal subsystem state variables

The reciprocal value of the linear coefficient between the two, T represents the matrix transposition;

as described aboveEach coefficient satisfies the following relationship:

[\begin{matrix} {\tilde{y}}_{1} \\ {\tilde{y}}_{2} \end{matrix}] = (\begin{matrix} {\tilde{C}}_{1} & 0 \\ {\tilde{C}}_{2} & {\tilde{C}}_{3} \end{matrix}) [\begin{matrix} {\tilde{x}}_{1} \\ {\tilde{x}}_{2} \end{matrix}]

(2) and a normal KX observer generation step:

construction of normal KX observer for normal subsystem

z₁Is a state variable of a normal KX observer, omega₁Is the output variable of the normal KX observer,

derivative of state variable for normal KX observerAnd a state variable z₁The linear coefficient matrix of (a) is,

derivative of state variable for normal KX observer

Measuring an output variable with a first sensor

The linear coefficient matrix of (a) is,

derivative of state variable for normal KX observer

A linear coefficient matrix with the control input variable u of the normal subsystem,

for outputting variable omega of normal KX observer₁With the state variable z of the normal KX observer₁The linear coefficient matrix of (a) is,

for outputting variable omega of normal KX observer₁Measuring an output variable with a first sensor

A linear coefficient matrix of (a);

to restrain

The numerical relation among the coefficient matrixes meets the intermediate coefficient matrix of the KX observer condition;

(3) a step of generating a fault KX observer:

fault KX observer constructed for fault subsystem

z₂As state variables, ω, of fault KX observers₂Is the output variable of the fault KX observer,

derivative of state variable for fault KX observerAnd a state variable z₂The linear coefficient matrix of (a) is,

derivative of state variable for fault KX observer

And fault observer output variable omega₂The linear coefficient matrix of (a) is,

derivative of state variable for fault KX observer

A linear coefficient matrix with the control form input variable u' of the faulty subsystem,

outputting a variable ω for a fault observer₂With fault KX observer state variable z₂The linear coefficient matrix of (a) is,

outputting a variable ω for a fault observer₂Measuring a form output variable with a second sensor

A coefficient matrix of (a);

to restrain

(4) a first KX observer synthesis step:

will be normalKX observer and fault KX observer are synthesized into first KX observer

y is the measurement output of all sensors of the satellite attitude control system, z is the state variable of the first KX observer, omega is the output variable of the first KX observer, and the parameters F, G, H, M and N of the first KX observer satisfy the following relations: defining a normal matrix

F = [\begin{matrix} F_{1} & 0 \\ 0 & F_{2} \end{matrix}],

G = [\begin{matrix} G_{1} & 0 \\ G_{3} & G_{2} \end{matrix}],

H = [\begin{matrix} H_{1} \\ H_{2} \end{matrix}],

M＝[M₁M₂]，N＝[N₁N₂]；

F_{1} = {\tilde{F}}_{1}, F_{2} = {\tilde{F}}_{2};

G_{1} = {\tilde{G}}_{1}, G_{4} = {\tilde{G}}_{2};

M_{1} = {\tilde{M}}_{1}, M_{2} = {\tilde{M}}_{2};

N_{1} = {\tilde{N}}_{1}, N_{2} = {\tilde{N}}_{2};

The technical effects of the invention are as follows: the fault-tolerant observer method has substantive characteristics and remarkable progress, and is developed on the basis of a fault observer with good performance, a triangular standard transformation and subsystem decomposition, a KX observer technology and a self-adaptive control switching process. Compared with the prior art, the method has the advantages of three aspects of state observation fault tolerance (ensuring that the normal state is not influenced by fault output), integral fault tolerance of sensor faults (fault tolerance for all the sensor faults) and stability of satellite attitude closed-loop faults, and improves the fault tolerance of satellite attitude control to a certain extent.

Drawings

FIG. 1 is a flow chart of a fault-tolerant observation method according to the present invention.

Fig. 2 is a schematic structural diagram of a fault diagnosis module in the fault-tolerant observation process according to the present invention.

FIG. 3 is a detailed flow chart of the satellite attitude control fault-tolerant observation of the present invention.

Fig. 4 is a graph of an observed state curve and an actual state curve obtained by a luneberg (Luenberger) observer.

Fig. 5 is a graph of an observed state curve and an actual state curve obtained by the improved KX fault-tolerant observer in a state observation fault-tolerant performance comparison experiment.

Fig. 6 is a graph of an observed state curve and an actual state curve obtained by a KX fault-tolerant observer in a comparison experiment of the overall fault-tolerant performance of a sensor fault.

Fig. 7 is a graph of an observed state curve and an actual state curve obtained by the improved KX fault-tolerant observer in the overall fault-tolerant performance comparison experiment of the sensor fault.

Fig. 8 is a graph of an observed state curve and an actual state curve obtained by a KX fault-tolerant observer in a closed-loop fault stability performance comparison experiment.

Fig. 9 is a graph of an observed state curve and an actual state curve obtained by the improved KX fault-tolerant observer in the closed-loop fault stability performance comparison experiment.

Detailed Description

The method comprises the following steps: firstly, a fault diagnosis module is adopted for carrying out sensor fault detection and isolation on a control system, if the isolation result judges that a first sensor has a fault, a second KX observer is selected for observation, and if the second sensor has a fault, a first KX observer is selected for observation. The key point of the invention is that a corresponding conversion matrix is adopted for carrying out system transformation on a state equation of a control system, then a triangular standard-shaped subsystem is decomposed into two subsystems, a low-dimensional KX function observer is respectively designed for the two subsystems by taking system feedback stable control as a target, and then the system KX observers with fault-tolerant performance are obtained by combination. By designing two parallel system KX observers and integrating a control switching process, the satellite attitude control system can still ensure fault-tolerant observation of the observation quantity of the rest part of the system when the output of part of the sensors is unreliable, and the closed-loop control integrity of a fault system can be ensured based on a KX feedback mechanism.

The invention has four parts and functions of a fault diagnosis module, a triangular standard shape transformation, a parallel KX observer and control switching (as shown in figure 1). The fault diagnosis module is used for detecting and isolating a fault sensor of the satellite attitude control system, namely, which output of the system is unreliable. The triangle standard shape transformation is completed to transform the system into a corresponding triangle standard shape, and the triangle standard shape is convenient for decomposing the system into a fault subsystem and a normal subsystem. And the parallel KX observer is used for observing a feedback state function KX of the transformed triangular standard system so as to give stable feedback of the system. And finally, the control switching determines to feed back the output of a corresponding KX observer in the parallel KX observers to the system control input according to the diagnosis result of the fault diagnosis module.

The following is a further description of the process of the invention:

1. fault diagnosis module

Considering a linear steady state control system, it can be described by the following state space model.

(1)

y(t)＝Cx(t)+f(t)

Wherein x ∈ Rⁿ(R represents a real number domain, n represents a real number domain space dimension) as a state vector, and u belongs to R^m(R represents a real number domain, m represents a real number domain space dimension) as a control vector, and y belongs to R^l(R represents a real number domain, l represents a real number domain spatial dimension) as an output vector, f () ∈ R^g(R represents a real number domain, g represents a real number domain space dimension) as a fault vector, and A, B, C as a system coefficient.

The fault diagnosis module is used for detecting and isolating which sensor output has faults, so that the fault diagnosis method based on the classical sensor fault residual observer is adopted in the invention:

l state observers are designed, and different sensor measurement signals are respectively used as input signals. As shown in fig. 2: the design steps are as follows.

1) Dividing the sensor output signal in dimension l:

y＝[y₁，y₂，...，y_l]^T(T denotes matrix transpose, the same applies hereinafter)

Make it y_iCorresponding to the output of the first sensor,

i

1, 2.

2) By y_iAnd u establishing l observers; the ith observer is composed of y only_iAnd u drive.

3) Obtaining l sets of state estimation values by the observerWherein

{\hat{x}}_{i} = {[{\hat{x}}_{i 1}, {\hat{x}}_{i 1}, . . ., {\hat{x}}_{in}]}^{T}

Is obtained by the ith observer,

i

1, 2

4) In the normal operation of the system,

(i ═ 1, 2.., l) should converge to the state x of the system; when the ith sensor fails and the remaining l-1 sensors operate normally, the state observations made by the ith observer

Will deviate from the true value x of the state and the diagnostic conclusion can be reached as follows.

Defining residual errors

r_{i} = y_{i} - C_{i} {\hat{x}}_{i}, i = 1,2, . . ., l

Wherein, C_iIs row i of the output matrix, defines a residual threshold e, if r_i<Epsilon, no fault exists; if r_i>Epsilon, then the ith sensor fails, and online fault detection and isolation can be performed.

2. Controlling handover

The method comprises the following steps of simultaneously observing a system by adopting a fault diagnosis module and parallel KX observers (a first KX observer and a second KX observer), acquiring residual errors output by each sensor after the fault observers output and observe the system, detecting and isolating a fault sensor, selecting the parallel KX observers according to a detection and isolation result by a control switching algorithm, wherein the basic flow of the control switching algorithm is as follows: if the sensor 2 fails, a first KX observer is selected, and if the sensor 1 fails, a second KX observer is selected. Switching the default output to the first KX observer.

After the fault diagnosis module detects and isolates the faults of the system sensor, the control switching module determines which part of KX observers in the parallel KX observers are selected to output, the parallel KX observers are respectively designed in a priori mode based on an assumed corresponding fault mode, and detailed design methods of the first KX observer and the second KX observer are respectively explained below.

3. Generation of the first KX observer

The first KX observer is designed based on the assumption that the sensor 2 is malfunctioning. For the default system (1), the system measures the output variable y ∈ R^lCan be described as y ═ y₁，y₂]，y₁Corresponding to the measured output, y, of the normal partial sensor (sensor 1)₂The corresponding failure portion sensor (sensor 2) measures an output.

1) Transformation of triangular standard

The purpose of the trigonometric standard transformation is to transform the system (1) into a corresponding trigonometric standard form which facilitates the decomposition of the system into a normal subsystem and a faulty subsystem, but with the proviso that the measured output variable y for the system must be [ y ═ y [ ]₁，y₂]In, y₁Corresponding to the measured output, y, of the normal partial sensor (sensor 1)₂The corresponding failure portion sensor (sensor 2) measures an output. Because the system defaults to describing y₁Corresponding to the measured output, y, of the normal sensor 1₂The corresponding failure sensor 2 measures the output. Thus, the system trigonometric standard transformation and the subsystem decomposition can be directly performed on the default system when the first KX observer is designed.

Assuming that the system (1) is observable, its observability matrix V ═ C^TA^TC^T…(A^n-1)^TC^T]^TAnd rank (v) n (rank represents the rank of the matrix, hereinafter the same). Is provided with

Then there are:

definition 1. if order

i＝2，3，…，l

Then call { v_iI 1, 2, …, l is the set of trigonometric indices of system (1). It is clear that there is a need for,

based on the set of triangular canonical form indices, lemma 1 can be obtained.

Lemma 1. there is a linear coordinate transformation x ═ Px, which can transform the system (1) into the equivalent triangular standard form:

(2)

y (t) = \tilde{C} \tilde{x} (t)

the method for P is as follows:

P^{- 1} = [b_{1} A b_{1} . . . A^{v_{1} - 1} b_{1} b_{2} A b_{2} . . . A^{v_{2} - 1} b_{2} . . . b_{l} A b_{l} . . . A^{v_{l} - 1} b_{l}]

wherein b is_i(i ═ 1, 2.., l) is the solution of the following equation:

the value of (A) is the right side of the equation of the above formula [0.. 010.. 0 ]]Element 1 in (2) is located in the position number.

Let rank C ═ l, and let

Order to

{\tilde{x}}^{T} = {[{\tilde{x}}_{1}^{T} {\tilde{x}}_{2}^{T}]}^{},

{\tilde{y}}^{T} = {[{\tilde{y}}_{1}^{T} {\tilde{y}}_{2}^{T}]}^{},

Wherein

<math> <mrow> <msub> <mover> <mi>x</mi> <mo>~</mo> </mover> <mn>1</mn> </msub> <mo>&Element;</mo> <msub> <mi>R</mi> <msub> <mover> <mi>v</mi> <mo>~</mo> </mover> <mn>1</mn> </msub> </msub> <mo>,</mo> </mrow></math>

<math> <mrow> <msub> <mover> <mi>x</mi> <mo>~</mo> </mover> <mn>2</mn> </msub> <mo>&Element;</mo> <msub> <mi>R</mi> <msub> <mover> <mi>v</mi> <mo>~</mo> </mover> <mn>2</mn> </msub> </msub> <mo>,</mo> </mrow></math>

<math> <mrow> <msub> <mover> <mi>y</mi> <mo>~</mo> </mover> <mn>1</mn> </msub> <mo>&Element;</mo> <msub> <mi>R</mi> <mi>l</mi> </msub> <mo>,</mo> </mrow></math>

<math> <mrow> <msub> <mover> <mi>y</mi> <mo>~</mo> </mover> <mn>2</mn> </msub> <mo>&Element;</mo> <msub> <mi>R</mi> <mrow> <mi>l</mi> <mo>-</mo> <mover> <mi>l</mi> <mo>~</mo> </mover> </mrow> </msub> <mo>,</mo> </mrow></math>

While

{\tilde{v}}_{2} = n - {\tilde{v}}_{1},

The system (2) can then be represented as:

(3)

[\begin{matrix} {\tilde{y}}_{1} \\ {\tilde{y}}_{2} \end{matrix}] = (\begin{matrix} {\tilde{C}}_{1} & 0 \\ {\tilde{C}}_{2} & {\tilde{C}}_{3} \end{matrix}) [\begin{matrix} {\tilde{x}}_{1} \\ {\tilde{x}}_{2} \end{matrix}]

if it is a reissue

The system (3) can be decomposed into two subsystems as follows:

(4)

{\tilde{y}}_{1} = {\tilde{C}}_{1} {\tilde{x}}_{1}

(5)

2) KX observer

In state feedback, the control law can be generally expressed as u-Kx. In order to reduce the dimension of the observer, the invention adopts a KX observer method to directly carry out reconstruction observation on the function Kx of the state variable. The design method of the KX observer is shown as the formula (6):

(6)

ω＝Mz+Ny

the observer satisfies:

<math> <mrow> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <munder> <mi>lim</mi> <mrow> <mi>t</mi> <mo>&RightArrow;</mo> <mo>∞</mo> </mrow> </munder> <mi>ω</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>lim</mi> <mrow> <mi>t</mi> <mo>&RightArrow;</mo> <mo>∞</mo> </mrow> </munder> <mi>Kx</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mtd> </mtr> <mtr> <mtd> <munder> <mi>lim</mi> <mrow> <mi>t</mi> <mo>&RightArrow;</mo> <mo>∞</mo> </mrow> </munder> <mi>z</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>lim</mi> <mrow> <mi>t</mi> <mo>&RightArrow;</mo> <mo>∞</mo> </mrow> </munder> <mi>T</mi> <mo>′</mo> <mi>x</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>,</mo> </mrow></math>

then it is satisfied with the following conditions:

1) t ' a-FT ' ═ GC, T ' is a solid matrix;

2)H＝T′B

3) all eigenvalues of F have negative real parts.

4)MT+NC＝K

After a system triangular standard shape is obtained by transforming the original system with linear coordinates x-Px, T', F, G, H, M, N which meet the above conditions 1) -4) are selected according to a feedback parameter K, and then a system observer with Kx as an observation target can be designed.

In order to make the system observer have fault tolerance, the above parameters must be solved in a defined manner. The main solution thought is as follows: for the normal subsystem (4) and the fault subsystem (5), Kx sub-observers are respectively designed for the two subsystems, namely the constructed normal KX observer is

The fault KX observer is

The meaning of each parameter is explained by taking a normal KX observer as an example,

three coefficient matrixes determine the state variable derivative of the normal observer

With the normal observer state variable z₁The first sensor measures the output variable y₁And a linear numerical relationship between the normal observer control input variables u,

two coefficient matrixes determine the output variable omega of the normal observer₁With the normal observer variable z₁And the first sensor measuring the output variable

Linear numerical relationship between them. The meaning of each coefficient in the fault KX observer is substantially the same as that of the normal KX observer, and will not be described here.

Designed based on individual sub-observers

F_i，G_i，H_i，M_i，N_i(i ═ 1, 2) values, the values of T', F, G,h, M and N values. The parameters of the subsystem observer and the parameters of the first KX observer satisfy the following derivation relationship:

let the system Kx observer parameters be expressed as:

F = [\begin{matrix} F_{1} & 0 \\ 0 & F_{2} \end{matrix}],

G = [\begin{matrix} G_{1} & 0 \\ G_{3} & G_{4} \end{matrix}],

H = [\begin{matrix} H_{1} \\ H_{2} \end{matrix}],

M＝[M₁ M₂]，N＝[N₁ N₂]the parameters of the KX observer of the subsystem (4) are expressed asThe parameters of the KX observer of the subsystem (5) are expressed as

Then:

1)

2)

3)

F_{1} = {\tilde{F}}_{1}, F_{2} = {\tilde{F}}_{2}

4)

G_{1} = {\tilde{G}}_{1}, G_{4} = {\tilde{G}}_{2}

5)

6)

M_{1} = {\tilde{M}}_{1}, M_{2} = {\tilde{M}}_{2}

7)

N_{1} = {\tilde{N}}_{1}, N_{2} = {\tilde{N}}_{2}

8)G₃、

are arbitrarily selected but need to satisfy

4. Generation of the second KX observer

The generation of the second KX observer differs from the first KX observer only in that: the generation of the second KX observer is based on the assumption that the sensor 1 is faulty. In order to meet the premise that the triangle standard is transformed and decomposed into a normal subsystem and a fault subsystem, y of the default system (1) needs to be equal to [ y [ ]₁，y₂]In (3), y1 and y2 exchange positions to obtain a new positionThe sensor measurement output variable representation y ═ y₂，y₁]. Correspondingly adjusting all parameters and variable expressions in the corresponding system (1) to obtain a new satellite attitude control system expression form:

y′(t)＝C′x′(t)+f′(t)

the new sensor measurement output variable y' after adjustment is expressed as:

wherein,

in response to the second normal sensor measuring output,

measuring an output for the first failure sensor;

and generating a second KX observer according to the triangular standard shape transformation and the design process of the KX observer for the new system, wherein the principle and the steps are the same as those of the first KX observer.

The first KX observer and the second KX observer are used for parallel observation system control input and measurement output to form a parallel KX observer and a fault diagnosis module to perform online observation at the same time, and the control switching module is used for determining which path of KX observer in the parallel KX observer outputs according to a fault diagnosis result.

The following examples are further provided in connection with the present invention:

the invention is applied to the attitude air injection control of the earth orientation three-axis stable satellite, and if the satellite is regarded as a rigid body, the pitching channel is decoupled and can be independently designed. Hence here we only consider the state space form of the two orbits rolling and yawing:

l in the formula (7)_x、L_y、L_zRepresenting three components of external moment in a satellite body coordinate system; i is_x、I_y、I_zThree principal inertias representing satellites; omega₀Representing the orbital angular velocity of the satellite.

For the convenience of the following discussion, we have here:

u = [\begin{matrix} L_{x} \\ L_{z} \end{matrix}],

C＝I_2×2

B = [\begin{matrix} L_{x} \\ L_{z} \end{matrix}]

the jet attitude control system can be expressed in the form of an equation of state in the form of system (1).

A fault-tolerant observation and feedback stabilization control block diagram of a rolling/yawing loop is shown in fig. 3, a fault observer and a KX observer are adopted to observe a system at the same time, the fault observer acquires residual errors output by each sensor after observing the output of the system, a fault sensor is detected and isolated, a control switching algorithm selects the KX observer according to a detection isolation result, and the basic flow of the control switching algorithm is as follows: if the sensor 1 fails, the KX observer 1 is selected, and if the sensor 2 fails, the KX observer 2 is selected. The default output is switched to KX observer 1.

In order to compare the superiority of the fault-tolerant observation scheme provided by the invention with other schemes, for the jet control system, comparison experiments are respectively carried out on the three aspects of state observation fault tolerance (whether the normal state is influenced by fault output), the integral fault tolerance of sensor faults (whether fault tolerance is carried out on all the sensor faults) and the satellite attitude closed-loop fault stability.

1) State observation fault tolerance comparison

The observer method and the classical Luenberger observer method are compared by state tracking aiming at a system (7) under the sensor fault, and the fault of the sensor 1 is set to be a step type failure fault and occurs in 1 s. The observed state curves obtained by both observers are shown in fig. 4 and 5, in which the abscissa represents time (time):

from fig. 4 and 5, it can be seen that when the step fault of the sensor 1 occurs at 1s, all the state estimation x of the Luenberger observer deviates progressively from the actual system state x, while the KX fault-tolerant observer used herein excludes the state estimation variable affected by the sensor output

Deviation from actual state value x₃、x₄In addition to the above-mentioned problems,

can still track x₁、x₂Therefore, compared with the traditional observer method, the KX fault-tolerant observer can still track the rest part of normal states when part of output is unreliable.

2) Overall fault tolerance comparison of sensor faults

The improved KX observer method and the improved KX observer method are subjected to state tracking comparison under different sensor faults, and it is set that the faults of the

sensors

1 and 2 are intermittent failure faults, the fault of the sensor 2 occurs in 0.6-0.8 second, and the fault of the sensor 1 occurs in 1.2-1.4 second. The observed state curves obtained by both observers are shown in fig. 6 and 7 as follows:

as can be seen from FIGS. 6 and 7, when the sensor 2 has a fault within 0.6-0.8 seconds, both the KX observer and the improved KX fault-tolerant observer can track part of the normal state x in the fault occurrence stage₁、x₂Both can track all normal states after the fault disappears; when the fault of the sensor 1 occurs in 1.2-1.4 seconds, the KX observer can not track all the states of the system any more, and the improved KX observer can still track the actual state value x₃、x₄After the fault disappears, the KX observer still can not recover the state tracking, and the improved KX observer method can recover and track all states.

3) Closed loop fault stability performance comparison

The improved KX observer method and the improved KX observer method of the invention are compared with the stable control effect in a closed-loop feedback control system, and the sensor 1s is still set to have step type failure fault. The observed state curves obtained by both observers are shown in fig. 8 and 9 below:

from fig. 8 and 9, when the step fault of the sensor 1 occurs at 1s, the KX observer can still track part of the normal state x₁、x₂But the system state begins to deviate from stability, and the system feedback control fails; the improved KX fault-tolerant observer can also track part of the normal state x₁、x₂But the system is asymptotically stable.

The analysis shows that the improved KX fault-tolerant observer method can still estimate the residual normal state of the system when any part of sensors have faults, can ensure the closed-loop feedback control of the system to be stable, and has better fault-tolerant observation performance and feedback system stability performance than other observer methods.

Claims

1. A sensor fault-tolerant observation method for a satellite attitude control system specifically comprises the following steps: and a fault diagnosis module is adopted to detect and isolate the fault of the sensor of the satellite attitude control system, if the isolation result judges that the first sensor has the fault, a second KX observer is selected for observation, and if the second sensor has the fault, the first KX observer is selected for observation.

2. The method according to claim 1, wherein the step of generating the first KX observer specifically comprises:

(1) a system decomposition step:

Is a normal subsystem state variableDerivative of (2)

And normal subsystem state variables

The linear coefficient of (a) is,

is a normal subsystem state variable

Derivative of (2)

Linear coefficients with the normal subsystem control input variable u,measuring an output variable for a first sensor

And normal subsystem state variablesThe linear coefficient of (a);

and a fault sub-system

As faulty subsystem state variables

Derivative of (2)

And fault subsystem state variables

The linear coefficient of (a) is,

as faulty subsystem state variables

Derivative of (2)

Linear coefficients with the fault subsystem control form input variable u',

outputting the variable for the second sensor form

And fault subsystem state variables

The linear coefficient of (a) is,

outputting the variable for the second sensor formAnd normal subsystem state variables

In betweenThe linear coefficient is inverted, and T represents matrix transposition;

the above coefficients satisfy the following relations:

[\begin{matrix} {\tilde{y}}_{1} \\ {\tilde{y}}_{2} \end{matrix}] = (\begin{matrix} {\tilde{C}}_{1} & 0 \\ {\tilde{C}}_{2} & {\tilde{C}}_{3} \end{matrix}) [\begin{matrix} {\tilde{x}}_{1} \\ {\tilde{x}}_{2} \end{matrix}];

(2) and a normal KX observer generation step:

construction of normal KX observer for normal subsystem

z₁Is a state variable of a normal KX observer, omega₁Is the output variable of the normal KX observer,derivative of state variable for normal KX observerAnd a state variable z₁The linear coefficient matrix of (a) is,

derivative of state variable for normal KX observer

Measuring an output variable with a first sensorThe linear coefficient matrix of (a) is,

derivative of state variable for normal KX observer

A linear coefficient matrix of (a);

to restrain

(3) a step of generating a fault KX observer:

fault KX observer constructed for fault subsystem

derivative of state variable for fault KX observer

And a state variable z₂The linear coefficient matrix of (a) is,

derivative of state variable for fault KX observer

derivative of state variable for fault KX observer

outputting a variable ω for a fault observer₂Measuring a form output variable with a second sensorA coefficient matrix of (a);

to restrain

(4) a first KX observer synthesis step:

synthesizing a normal KX observer and a fault KX observer into a first KX observer

F = [\begin{matrix} F_{1} & 0 \\ 0 & F_{2} \end{matrix}],

G = [\begin{matrix} G_{1} & 0 \\ G_{3} & G_{2} \end{matrix}],

H = [\begin{matrix} H_{1} \\ H_{2} \end{matrix}],

M＝[M₁ M₂]，N＝[N₁ N₂]；

F_{1} = {\tilde{F}}_{1}, F_{2} = {\tilde{F}}_{2};

G_{1} = {\tilde{G}}_{1}, G_{4} = {\tilde{G}}_{2};

M_{1} = {\tilde{M}}_{1}, M_{2} = {\tilde{M}}_{2};

N_{1} = {\tilde{N}}_{1}, N_{2} = {\tilde{N}}_{2};

3. The method as claimed in claim 1 or 2, wherein the satellite attitude control system is decomposed by transformation in a trigonometric standard form.