CN111897221A - Spacecraft fault diagnosis method based on combined observer - Google Patents

Spacecraft fault diagnosis method based on combined observer Download PDF

Info

Publication number
CN111897221A
CN111897221A CN202010783786.4A CN202010783786A CN111897221A CN 111897221 A CN111897221 A CN 111897221A CN 202010783786 A CN202010783786 A CN 202010783786A CN 111897221 A CN111897221 A CN 111897221A
Authority
CN
China
Prior art keywords
observer
neural network
spacecraft
fault
combined
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.)
Granted
Application number
CN202010783786.4A
Other languages
Chinese (zh)
Other versions
CN111897221B (en
Inventor
胡庆雷
李远东
陈绪宁
邵小东
郑建英
郭雷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beihang University
Original Assignee
Beihang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beihang University filed Critical Beihang University
Priority to CN202010783786.4A priority Critical patent/CN111897221B/en
Publication of CN111897221A publication Critical patent/CN111897221A/en
Application granted granted Critical
Publication of CN111897221B publication Critical patent/CN111897221B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance

Abstract

The invention relates to a spacecraft fault diagnosis method based on a combined observer, which comprises the following steps: firstly, establishing a spacecraft dynamics model considering the faults of an actuating mechanism; secondly, establishing a novel online training neural network observer based on a common fault detection observer and a neural network; then, combining the Lonberg observer and the novel neural network observer into a combined observer by using a self-adaptive threshold switching method; finally, performing autonomous diagnosis of the faults of the actuating mechanism by using the combined observer; the method can ensure that the autonomous fault diagnosis of the spacecraft attitude control system actuating mechanism is realized under the condition of limited calculated amount, and has the advantages of strong robustness and high diagnosis precision.

Description

Spacecraft fault diagnosis method based on combined observer
Technical Field
The invention relates to a spacecraft fault diagnosis method based on a combined observer, which is mainly applied to fault diagnosis of an actuating mechanism of a spacecraft attitude control system under the conditions of model uncertainty, external interference and calculation amount limitation, and belongs to the technical field of spacecraft fault diagnosis.
Background
Spacecraft actuators, represented by reaction flywheels, are important components of spacecraft attitude control systems. Because the actuating mechanism works in a severe space for a long time, such as high and low temperature change, impact of high-energy particles, friction, corrosion, short circuit and other factors generated by long-time running of the motor, the actuating mechanism is easy to malfunction. If the fault of the executing mechanism cannot be diagnosed in time and corresponding measures are taken, the capability of the spacecraft for executing the task is reduced if the fault is not detected, and the spacecraft cannot complete the set task if the fault is detected, so that huge economic loss is brought. Therefore, the development of spacecraft actuator fault diagnosis research is of great significance. In addition, with the successive development of deep space exploration projects such as Mars exploration and Mars exploration in China, the distance between the spacecraft and the ground measurement and control station is longer and longer, and the communication delay is larger and larger. The conventional fault diagnosis scheme which depends on the expert judgment of the measurement and control station is difficult to diagnose the fault of the spacecraft actuating mechanism in time, which poses a great threat to the safe and reliable task execution of the spacecraft, so that the autonomous fault diagnosis of the spacecraft is very necessary. Due to the uncertainty of a spacecraft model and the factors of external interference, for the complex nonlinear system, the traditional observers such as the Luenberger observer, the self-adaptive observer and the like are difficult to carry out rapid, accurate and reliable fault diagnosis. The neural network has the characteristic of approximating any smooth nonlinear function, and plays a great role in many fields. The method can be combined with an observer to realize accurate fault diagnosis under the condition of model uncertainty and external interference. However, most of the existing neural network observers have the defects of complex structure and large calculation amount, and huge pressure is caused on the calculation amount, heat dissipation and power supply of the spacecraft attitude control computer. Therefore, the spacecraft fault diagnosis scheme which is small in calculation amount and capable of effectively resisting model uncertainty and external interference has great application value.
Disclosure of Invention
The technical problem of the invention is solved: aiming at the problems of calculation amount limitation, model uncertainty, external interference and the like of a spacecraft, the spacecraft fault diagnosis method based on the combined observer is provided, has strong robustness and strong anti-interference capability, can realize autonomous fault diagnosis of a spacecraft execution mechanism by using smaller calculation amount, solves the autonomous fault diagnosis problems of high precision and high reliability of the spacecraft under the conditions of limited calculation amount, external interference and model uncertainty, and improves the reliability of the spacecraft in executing tasks such as long-term deep space exploration.
The technical solution of the invention is as follows: a spacecraft fault diagnosis method based on a combined observer comprises the following implementation steps:
firstly, establishing a spacecraft dynamics model considering faults of an attitude control system actuating mechanism:
Figure BDA0002621169460000021
y(t)＝Cx(t)
wherein
Figure BDA0002621169460000022
State variables representing the attitude of the spacecraft, n being the system state dimension, t being the time,
Figure BDA0002621169460000023
is the input coefficient matrix of the system, m is the number of actuating mechanisms,
Figure BDA0002621169460000024
for the theoretical output torque of each actuator,
Figure BDA0002621169460000025
the subscript d represents interference, p is the number of interferences,
Figure BDA0002621169460000026
is the interference vector of the system and is,
Figure BDA0002621169460000027
is a fault distribution matrix of the system, r is the number of faults,
Figure BDA0002621169460000028
as a function of the failure of the system,
Figure BDA0002621169460000029
is the measured output vector of the system, c is the system output dimension,
Figure BDA00026211694600000210
for the output coefficient matrix of the system, Φ (x, t) is the nonlinear function term of the system:
Figure BDA00026211694600000211
wherein Ix、Iy、IzRepresenting the three-axis moment of inertia, x, of the spacecraft1、x2、x3And the three-axis attitude angular velocity of the spacecraft is represented. During the task execution period of the spacecraft, the flywheel does mechanical motion for a long time, and faults are easy to occur due to friction, temperature, corrosion and the like. Typical failure modes of actuators such as flywheels are:
(1) stuck-at fault
Figure BDA00026211694600000212
Wherein u isout(t) is the actual output torque of the momentum wheel, uin(t) output torque specified for the controller, tfAt the time of occurrence of the fault ukThe output torque is an arbitrary constant value and represents the actual output torque when the jamming fault occurs.
(2) Failure of efficiency drop
Figure BDA00026211694600000213
Where k is the rate of efficiency decrease.
(3) Friction torque increase failure
Figure BDA0002621169460000031
Wherein f isa(t) is a numerical function of the increase in friction torque.
(4) Short-circuit fault
Figure BDA0002621169460000032
Wherein t isf1、tf2、tf3Respectively the starting, middle and ending moments of the short circuit.
(5) Other faults
Figure BDA0002621169460000033
Where f (t) is a function of arbitrary form.
Secondly, establishing a detection observer according to a spacecraft and actuating mechanism dynamic model:
Figure BDA0002621169460000034
Figure BDA0002621169460000035
wherein
Figure BDA0002621169460000036
A state estimation variable representing the attitude of the spacecraft,
Figure BDA0002621169460000037
a vector is output for the estimation of the system,
Figure BDA0002621169460000038
is the observer gain. The remaining parameters are the same as those described in the first step.
The novel online neural network observer established based on the detection observer and the neural network is as follows:
Figure BDA0002621169460000039
Figure BDA00026211694600000310
wherein
Figure BDA00026211694600000311
The actual output torque of the spacecraft actuator estimated by the neural network observer. The concrete form is as follows:
Figure BDA0002621169460000041
wherein u isiFor nominal output torque of the actuator, NNiThe specific form of (t) is:
Figure BDA0002621169460000042
Figure BDA0002621169460000043
where in is the number of input layer nodes of the neural network, mid is the number of intermediate layer nodes of the neural network, and f (-) is the sigmoid activation function, i.e.
Figure BDA0002621169460000044
Wherein the input is
Figure BDA0002621169460000045
Comprises the following steps:
Ik(t)＝[e1(t-τ),e2(t-τ),e3(t-τ),…,e3(t-nuτ),NN1(t-τ),…,NN2m(t-nuτ)]T
wherein
Figure BDA0002621169460000046
eiI 1,2,3 stands for observer residual, NNiAnd i is 1, 2m represents the output of the neural network observer, τ is a delay constant, and nu is a historical data number, which is selected according to the required diagnosis precision and the performance of the spaceborne computer, wherein the larger the nu value is, the higher the diagnosis precision is, and the higher the requirement on the performance of the computer is. Cost function taking e in neural network training processm(t)＝e(t)Te (t). The cost function of the neural network selects a gradient descent method of the additional momentum.
Updating the weight omega of the middle layer of the neural network according to the cost functionkjOutput layer weight omegakiAnd intermediate layer threshold ajOutput layer threshold bi；
Figure BDA0002621169460000047
ωki(t)＝ωki(t-1)+η2Hjem+α2[ωki(t-1)-ωki(t-2)]
Figure BDA0002621169460000048
bi(t)＝bi(t-1)+η4em+α4[bi(t-1)-bi(t-2)]
Wherein etai,αiI is 1,2,3,4 is the gradient descent learning rate and the additional momentum learning rate of the neural network, respectively, IkFor neural network input, HjAs neural network intermediate parameters, emAs a cost function.
The neural network observer established based on the method can accurately diagnose the fault under the condition of ensuring that the calculated amount is small.
And thirdly, firstly, establishing an adaptive Luenberger observer. For different types of faults of each actuating mechanism, the Lonberg observer needs to establish different contribution observers. For the dead-jamming fault of the actuator, the observer is in the form of:
Figure BDA0002621169460000051
ri(t)＝Mzi(t)
Figure BDA0002621169460000052
wherein z isiRepresenting an observation state, i is 1,2, and M represents an actuator, and G and M are input and output gain matrixes of a lunberg observer respectively. RhoiTo the learning rate, riTo observe the output value, xiiFor a certain constant value of the constant,
Figure BDA0002621169460000053
bithe estimated values of the stuck position and whether the stuck position is completely failed or not are respectively obtained, and the other parameters have the same meanings as above.
For the efficiency drop fault of the actuator, the observer is in the form of:
Figure BDA0002621169460000054
ri(t)＝Mzi(t)
Figure BDA0002621169460000055
wherein
Figure BDA0002621169460000056
Is an estimate of the input matrix and,
Figure BDA0002621169460000057
for efficiency estimation, γiIs the learning rate, etaiThe other parameters are defined as above for a certain constant value.
And combining the Luenberger observer and the novel neural network observer into a combined observer by using an adaptive threshold switching method.
The formula of the combined observer is:
Figure BDA0002621169460000058
Figure BDA0002621169460000059
wherein G is1,G2To adapt the control gain of the lunberger observer to the matrix to be solved,
Figure BDA00026211694600000510
b,
Figure BDA00026211694600000511
estimating values of the adaptive Luenberger observer on efficiency faults, failure faults and deviation faults;
and only using the self-adaptive Luenberger observer under the normal operation state of the spacecraft. Due to the existence of external interference, model uncertainty and the like, the false alarm rate of the adaptive Luenberger observer is high, and therefore when the residual error exceeds a threshold value, the diagnosis of the adaptive Luenberger observer is not reliable. The novel neural network observer is activated when the adaptive lunberg observer diagnoses a fault or the lunberg observer residual exceeds a threshold but cannot diagnose a specific fault type, i.e., magnitude. The evaluation function for the residual of the lunberg observer is:
Figure BDA00026211694600000512
wherein T is T2-t1，||·||rmsTo calculate the root mean square value, r (t) is the observer residual, and the adaptive thresholds are:
Figure BDA0002621169460000061
wherein JthIs a threshold value, v (t) is an uncertainty vector, sup denotes the supremum,
Figure BDA0002621169460000062
q is the filter output, λmaxRepresenting the maximum eigenvalue; t is t2Representing the end of the threshold decision time window, t1The initial moment of the time window is determined on behalf of the threshold.
And during the switching process of the combined observer, the diagnostic data of the Lorber observer is adopted for rough fault-tolerant control, and after the switching process, the diagnostic data of the neural network observer is adopted for precise fault-tolerant control. The novel neural network observer established in the second step needs a small amount of calculation, and the neural network is not needed to be adopted for diagnosis when the spacecraft normally operates by using the combination mode of the Lorberg observer and the neural network observer, so that the requirement on the calculation amount is further reduced. It should be noted that the adaptive lunberger observer is applied here for fault diagnosis instead of the general detection observer, so that the diagnosis result of the adaptive lunberger observer can still be applied for rough control in the switching stage.
Compared with the prior art, the spacecraft fault diagnosis method based on the combined observer considering the existence of calculation amount limitation, model uncertainty and external interference has the advantages that:
(1) the invention designs a spacecraft fault diagnosis method based on a combined observer, aiming at the constraints of calculation amount limitation, model uncertainty, external interference action and the like, compared with the traditional neural network observer, the designed novel online training neural network observer has smaller calculation amount, better robustness, higher precision and simpler design process;
(2) the combined observer combines the novel neural network observer and the self-adaptive Lorberg observer, further reduces the requirement on a computer under the condition of normal operation, can carry out high-precision fault diagnosis under the condition of spacecraft fault, and improves the safety of the spacecraft in the process of executing remote tasks such as deep space exploration and the like;
drawings
FIG. 1 is a flow chart of a spacecraft fault diagnosis method based on a combined observer according to the invention;
FIG. 2 is a structural framework diagram of the novel online training neural network observer of the present invention;
FIG. 3 is a structural framework diagram of the combined observer of the present invention;
FIG. 4 is a result of a simulation of the combined observer for continuous fault diagnosis in accordance with the present invention;
FIG. 5 is a simulation result of the combined observer for diagnosing sudden-change faults according to the present invention.
Detailed Description
As shown in fig. 1, the spacecraft fault diagnosis method based on the combined observer of the present invention includes the steps of: firstly, establishing a spacecraft dynamics model considering the faults of an attitude control system actuating mechanism; secondly, establishing a novel online training neural network observer based on the detection observer and the neural network; then, combining the Lonberg observer and the novel neural network observer into a combined observer by using a self-adaptive threshold switching method; and finally, performing autonomous diagnosis of the fault by using the combined observer. The functional block diagram of the whole system is shown in fig. 1, and the specific implementation steps are as follows:
firstly, establishing a spacecraft dynamics model considering faults of an attitude control system actuating mechanism:
Figure BDA0002621169460000071
y(t)＝Cx(t)
wherein
Figure BDA0002621169460000072
State variables representing the attitude of the spacecraft, n being the system state dimension, t being the time,
Figure BDA0002621169460000073
is the input coefficient matrix of the system, m is the number of actuating mechanisms,
Figure BDA0002621169460000074
for the theoretical output torque of each actuator,
Figure BDA0002621169460000075
the subscript d represents interference, p is the number of interferences,
Figure BDA0002621169460000076
is the interference vector of the system and is,
Figure BDA0002621169460000077
is a fault distribution matrix of the system, r is the number of faults,
Figure BDA0002621169460000078
as a function of the failure of the system,
Figure BDA0002621169460000079
is the measured output vector of the system, c is the system output dimension,
Figure BDA00026211694600000710
for the output coefficient matrix of the system, Φ (x, t) is the nonlinear function term of the system:
Figure BDA00026211694600000711
wherein Ix、Iy、IzRepresenting the three-axis moment of inertia, x, of the spacecraft1、x2、x3And the three-axis attitude angular velocity of the spacecraft is represented. The main parameters of the spacecraft are selected as follows: i isx＝80kg·m、Iy＝90kg·m、Iz＝70kg·m、B＝I3×3、Ed＝I3×3、C＝I3×3、d(t)＝1×10-4[sin(t) sin(2t) cos(t)]TNm。
Secondly, establishing a detection observer model according to the spacecraft and actuating mechanism dynamic model as follows:
Figure BDA00026211694600000712
Figure BDA00026211694600000713
wherein
Figure BDA00026211694600000714
A state estimation variable representing the attitude of the spacecraft,
Figure BDA00026211694600000715
a vector is output for the estimation of the system,
Figure BDA00026211694600000716
is the vector of the measured output of the system,
Figure BDA00026211694600000717
is the observer gain. The rest of the parameter settings are consistent with the first step.
The block diagram of the novel online neural network observer established based on the detection observer and the neural network is shown in fig. 2: firstly, a controller generates an expected control moment through the difference between an expected attitude angle and an attitude angle measured by a sensor; then the executing mechanism receives the control signal to generate an actual control torque, and if the executing mechanism breaks down in the process, the actual control torque and the expected control torque generate deviation; and then the spacecraft dynamics generates attitude angular acceleration after receiving the actual control moment, and the attitude angular acceleration is measured by a sensor and then transmitted to a controller.
The detection observer obtains an estimated actual control moment by calculating an expected control moment and a fault estimation value, and drives an internal analysis model by using the moment and makes a residual error with the moment generated by an actual sensor. The residual error is transmitted to an improved BP neural network, and the neural network obtains an estimated value of the fault through calculation of a historical residual error value and a historical self value and transmits the estimated value to a detection observer. The above components together form a novel online training neural network observer.
The online neural network observer is in the form of:
Figure BDA0002621169460000081
Figure BDA0002621169460000082
wherein
Figure BDA0002621169460000083
The actual output torque of the spacecraft actuator estimated by the neural network observer. The concrete form is as follows:
Figure BDA0002621169460000084
wherein u isiTo implement the nominal output torque of the mechanism, NNiThe specific form of (t) is:
Figure BDA0002621169460000085
Figure BDA0002621169460000086
where in is the number of input layer nodes of the neural network, mid is the number of intermediate layer nodes of the neural network, and f (-) is the sigmoid activation function, i.e.
Figure BDA0002621169460000087
Wherein the input is
Figure BDA0002621169460000088
Comprises the following steps:
Ik(t)＝[e1(t-τ),e2(t-τ),e3(t-τ),…,e3(t-nuτ),NN1(t-τ),…,NN2m(t-nuτ)]T
wherein
Figure BDA0002621169460000089
eiI 1,2,3 stands for observer residual, NNi1, 2.. 2m represents the output of the neural network observer, and τ is the time delayAnd a constant nu is selected according to the required diagnosis precision and the performance of the spaceborne computer, and the larger the nu value is, the higher the diagnosis precision is, and the higher the requirement on the performance of the computer is. The cost function is equal to em(t)＝e(t)Te(t)。
Updating the weight omega of the middle layer of the neural network according to the cost functionkjOutput layer weight omegakiAnd intermediate layer threshold ajOutput layer threshold bi；
Figure BDA0002621169460000091
ωki(t)＝ωki(t-1)+η2Hjem+α2[ωki(t-1)-ωki(t-2)]
Figure BDA0002621169460000092
bi(t)＝bi(t-1)+η4em+α4[bi(t-1)-bi(t-2)]
Wherein etai,αiI is 1,2,3,4 is the gradient descent learning rate and the additional momentum learning rate of the neural network, respectively, IkFor neural network input, HjAs neural network intermediate parameters, emAs a cost function.
The online training neural network observer established based on the method can accurately diagnose the fault under the condition of ensuring that the calculated amount is small.
The observer gain in the simulation process is as follows: L-3I3×3. The number of nodes of the output layer of the neural network is as follows: 2m is 6. The delay coefficient is: τ is 1 s. The number of times of inputting the historical data is as follows: n is 5. The number of nodes of the input layer is calculated as follows: and in is 20.
Number of intermediate layer nodes
Figure BDA0002621169460000093
And thirdly, combining the neural network observer and the self-adaptive Luenberger observer built in the second step into a combined observer. The structure block diagram is shown in fig. 3. Firstly, a controller generates an expected control moment through the difference between an expected attitude angle and an attitude angle measured by a sensor; then the executing mechanism receives the control signal to generate an actual control torque, and if the executing mechanism breaks down in the process, the actual control torque and the expected control torque generate deviation; and then the spacecraft dynamics generates attitude angular acceleration after receiving the actual control moment, and the attitude angular acceleration is transmitted to the controller after being measured by the sensor.
The Longberger observer inputs expected control torque generated by the controller and actual attitude angular velocity measured by the sensor, and drives the internal model to generate observation residual errors and estimated fault types and sizes. And activating the novel neural network observer when the self-adaptive Lorber observer diagnoses a fault or the residual error of the Lorber observer exceeds a threshold value but cannot diagnose a specific fault form and size. It should be noted that the adaptive lunberger observer is applied here for fault diagnosis instead of the general detection observer, so that the diagnosis result of the adaptive lunberger observer can still be applied for rough control in the switching stage.
And establishing an adaptive Luenberger observer. For different types of faults of each actuating mechanism, the Lonberg observer needs to establish different contribution observers. For the dead-jamming fault of the actuator, the observer is in the form of:
Figure BDA0002621169460000094
ri(t)＝Mzi(t)
Figure BDA0002621169460000095
wherein z isiRepresenting an observation state, i is 1,2, and M represents an actuator, and G and M are input and output gain matrixes of a lunberg observer respectively. RhoiTo the learning rate, riTo observe the output value, xiiFor a certain constant value of the constant,
Figure BDA0002621169460000101
birespectively, the stuck position and an estimate of whether or not there was a complete failure.
For the efficiency drop fault of the actuator, the observer is in the form of:
Figure BDA0002621169460000102
ri(t)＝Mzi(t)
Figure BDA0002621169460000103
wherein
Figure BDA0002621169460000104
Is an estimate of the input matrix and,
Figure BDA0002621169460000105
for efficiency estimation, γiIs the learning rate, etaiIs a certain constant value.
And combining the Luenberger observer and the novel neural network observer into a combined observer by using an adaptive threshold switching method.
The formula of the combined observer is:
Figure BDA0002621169460000106
Figure BDA0002621169460000107
wherein G is1,G2To adapt the control gain of the lunberger observer to the matrix to be solved,
Figure BDA0002621169460000108
b,
Figure BDA0002621169460000109
in order to adapt the lunberger observer to the problems of efficiency faults, failure faults,an estimate of the deviation fault;
and only using the self-adaptive Luenberger observer under the normal operation state of the spacecraft. Due to the existence of external interference, model uncertainty and the like, the false alarm rate of the adaptive Luenberger observer is high, and therefore when the residual error exceeds a threshold value, the diagnosis result is not very reliable. And activating the novel neural network observer when the self-adaptive Lorber observer diagnoses a fault or the residual error of the Lorber observer exceeds a threshold value but cannot diagnose a specific fault form and size. The evaluation function for the residual of the lunberg observer is:
Figure BDA00026211694600001010
wherein T is T2-t1，||·||rmsTo calculate the root mean square value, r (t) is the observer residual, and the adaptive thresholds are:
Figure BDA00026211694600001011
wherein JthIs a threshold value, v (t) is an uncertainty vector, sup denotes the supremum,
Figure BDA00026211694600001012
q is the filter output, λmaxRepresenting the maximum eigenvalue; t is t2Representing the end of the threshold decision time window, t1The initial moment of the time window is determined on behalf of the threshold.
And during the switching process of the combined observer, the diagnostic data of the Lorber observer is adopted for rough fault-tolerant control, and after the switching process, the diagnostic data of the neural network observer is adopted for precise fault-tolerant control. The novel neural network observer established in the second step needs a small amount of calculation, and the neural network is not needed to be adopted for diagnosis when the spacecraft normally operates by using the combination mode of the Lorberg observer and the neural network observer, so that the requirement on the calculation amount is further reduced.
And carrying out simulation verification by using Matlab/Simulink, wherein the simulation step length is 0.01 second, and the simulation time is 1000 seconds. The training frequency of the neural network was 10 times/second. The simulation results are shown in fig. 4 and 5. The fault diagnosis can be completed within 10 seconds for both continuous faults and sudden faults, and the diagnosis speed is high. The precision of the fault diagnosis is within 0.01Nm for the variation, and within 0.001Nm for the constant fault, so that the precision is higher. Compared with an adaptive Luenberger observer, the combined observer can diagnose faults in any form, and an observer does not need to be designed for each flywheel independently. Compared with the traditional neural network observer, the combined observer has the advantages that the training output form is simpler, the training frequency is lower, and the calculated amount is smaller; and the glitch and ringing phenomena are not severe.
Those skilled in the art will appreciate that the invention may be practiced without these specific details. Although exemplary embodiments of the present invention have been described for illustrative purposes, those skilled in the art will appreciate that various modifications, additions, substitutions and the like can be made in form and detail without departing from the scope and spirit of the invention as disclosed in the accompanying claims, all of which are intended to fall within the scope of the claims, and that various steps in the various sections and methods of the claimed product can be combined together in any combination. Therefore, the description of the embodiments disclosed in the present invention is not intended to limit the scope of the present invention, but to describe the present invention. Accordingly, the scope of the present invention is not limited by the above embodiments, but is defined by the claims or their equivalents.

Claims (4)

1. A spacecraft fault diagnosis method based on a combined observer is characterized by comprising the following steps:
(1) establishing a spacecraft dynamics model considering faults of an attitude control system actuating mechanism;
(2) establishing a detection observer according to a spacecraft dynamics model, and establishing an online neural network observer based on the detection observer and a neural network;
(3) combining the Luenberger observer and the online neural network observer in the step (2) into a combined observer by using a self-adaptive threshold switching method;
(4) and (4) realizing accurate and reliable autonomous fault diagnosis for the spacecraft actuating mechanism under the limitation of the calculated amount by using the combined observer obtained in the step (3).
2. The combined observer-based spacecraft fault diagnosis method of claim 1, characterized in that: in the step (1), a spacecraft dynamics model considering faults of an attitude control system executing mechanism is established as follows:
Figure FDA0002621169450000011
y(t)＝Cx(t)
wherein
Figure FDA0002621169450000012
State variables representing the attitude of the spacecraft, n being the system state dimension, t being the time,
Figure FDA0002621169450000013
is the input coefficient matrix of the system, m is the number of actuating mechanisms,
Figure FDA0002621169450000014
for the theoretical output torque of each actuator,
Figure FDA0002621169450000015
the subscript d represents interference, p is the number of interferences,
Figure FDA0002621169450000016
is the interference vector of the system and is,
Figure FDA0002621169450000017
is a fault distribution matrix of the system, r is the number of faults,
Figure FDA0002621169450000018
as a function of the failure of the system,
Figure FDA0002621169450000019
is the measured output vector of the system, c is the system output dimension,
Figure FDA00026211694500000110
for the output coefficient matrix of the system, Φ (x, t) is the nonlinear function term of the system:
Figure FDA00026211694500000111
wherein Ix、Iy、IzRepresenting the three-axis moment of inertia, x, of the spacecraft1、x2、x3And the three-axis attitude angular velocity of the spacecraft is represented.
3. The combined observer-based spacecraft fault diagnosis method of claim 1, characterized in that: in the step (2), the established detection observer is:
Figure FDA0002621169450000021
Figure FDA0002621169450000022
wherein
Figure FDA0002621169450000023
A state estimation variable representing the attitude of the spacecraft,
Figure FDA0002621169450000024
a vector is output for the estimation of the system,
Figure FDA0002621169450000025
for observersGain;
the method comprises the following steps of establishing an online neural network observer based on a detection observer and a neural network:
Figure FDA0002621169450000026
Figure FDA0002621169450000027
wherein
Figure FDA0002621169450000028
The actual output torque of the spacecraft actuating mechanism estimated by the neural network observer is in the following specific form:
Figure FDA0002621169450000029
wherein u isiI 1,2, m is the nominal output torque of the actuator, NNiThe specific form of (t) is:
Figure FDA00026211694500000210
Figure FDA00026211694500000211
where in is the number of input layer nodes of the neural network, mid is the number of intermediate layer nodes of the neural network, and f (-) is the sigmoid activation function, i.e.
Figure FDA00026211694500000212
Wherein the input is
Figure FDA00026211694500000213
Comprises the following steps:
Ik(t)＝[e1(t-τ),e2(t-τ),e3(t-τ),…,e3(t-nuτ),NN1(t-τ),…,NN2m(t-nuτ)]T
wherein
Figure FDA00026211694500000214
Representing observer residual, NNiI is 1, 2m represents the output of the neural network observer, τ is a delay constant, and nu is a historical data number, and is selected according to the required diagnosis precision and the performance of the spaceborne computer; cost function taking e in neural network training processm(t)＝e(t)Te, (t), the neural network training frequency is selected according to the diagnosis precision requirement and the performance of the spaceborne computer;
updating the weight omega of the middle layer of the neural network according to the cost functionkjOutput layer weight omegakiAnd intermediate layer threshold ajOutput layer threshold bi；
Figure FDA0002621169450000031
ωki(t)＝ωki(t-1)+η2Hjem+α2[ωki(t-1)-ωki(t-2)]
Figure FDA0002621169450000032
bi(t)＝bi(t-1)+η4em+α4[bi(t-1)-bi(t-2)]
Wherein etai,αiI is 1,2,3,4 is the gradient descent learning rate and the additional momentum learning rate of the neural network, respectively, IkFor neural network input, HjAs neural network intermediate parameters, emAs a cost function.
4. The combined observer-based spacecraft fault diagnosis method of claim 1, characterized in that: in the step (3), the form of the combined observer is as follows:
Figure FDA0002621169450000033
Figure FDA0002621169450000034
wherein G is1,G2To adapt the control gain of the lunberger observer to the matrix to be solved,
Figure FDA0002621169450000035
b,
Figure FDA0002621169450000036
estimating values of the adaptive Luenberger observer on efficiency faults, failure faults and deviation faults;
when the adaptive Luenberger observer diagnoses a fault or the residual error of the Luenberger observer exceeds a threshold value but cannot diagnose a specific fault, activating the neural network observer, wherein the residual error of the Luenberger observer has an evaluation function as follows:
Figure FDA0002621169450000037
wherein T is T2-t1，||·||rmsTo calculate the root mean square value, r (t) is the observer residual, and the adaptive thresholds are:
Figure FDA0002621169450000038
wherein JthIs a threshold value, v (t) is an uncertainty vector, sup denotes the supremum,
Figure FDA0002621169450000039
q is the filter output, λmaxRepresenting the maximum eigenvalue; t is t2Representing the end of the threshold decision time window, t1Judging the initial moment of a time window by representing a threshold value;
and during the switching process of the combined observer, the diagnostic data of the Lorber observer is adopted for rough fault-tolerant control, and after the switching process, the diagnostic data of the neural network observer is adopted for precise fault-tolerant control.
CN202010783786.4A 2020-08-06 2020-08-06 Spacecraft fault diagnosis method based on combined observer Active CN111897221B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010783786.4A CN111897221B (en) 2020-08-06 2020-08-06 Spacecraft fault diagnosis method based on combined observer

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010783786.4A CN111897221B (en) 2020-08-06 2020-08-06 Spacecraft fault diagnosis method based on combined observer

Publications (2)

Publication Number Publication Date
CN111897221A true CN111897221A (en) 2020-11-06
CN111897221B CN111897221B (en) 2021-10-15

Family

ID=73245874

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010783786.4A Active CN111897221B (en) 2020-08-06 2020-08-06 Spacecraft fault diagnosis method based on combined observer

Country Status (1)

Country Link
CN (1) CN111897221B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112947357A (en) * 2021-01-24 2021-06-11 西北工业大学 Aircraft intermittent fault diagnosis method based on double observers
CN112947392A (en) * 2021-04-05 2021-06-11 西北工业大学 Flight control system actuator and sensor composite tiny fault estimation method based on robust observer
CN114879512A (en) * 2022-07-06 2022-08-09 南京航空航天大学 Spacecraft formation orbit fault-tolerant control method based on learning neural network sliding mode
CN115046526A (en) * 2022-08-08 2022-09-13 东方空间技术(北京)有限公司 Aircraft fault diagnosis method and device, computer equipment and storage medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101481019A (en) * 2009-02-20 2009-07-15 华中科技大学 Fault tolerant observing method of sensor for satellite attitude control system
US7660774B2 (en) * 2005-05-31 2010-02-09 Honeywell International Inc. Nonlinear neural network fault detection system and method
CN102854874A (en) * 2012-06-18 2013-01-02 南京航空航天大学 A plurality of united observer based fault diagnosis and fault-tolerant control device and method
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

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7660774B2 (en) * 2005-05-31 2010-02-09 Honeywell International Inc. Nonlinear neural network fault detection system and method
CN101481019A (en) * 2009-02-20 2009-07-15 华中科技大学 Fault tolerant observing method of sensor for satellite attitude control system
CN102854874A (en) * 2012-06-18 2013-01-02 南京航空航天大学 A plurality of united observer based fault diagnosis and fault-tolerant control device and method
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

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
TARAK DAMAK: ""Fault diagnosis and isolation of a complex system using a neural network observer"", 《INT. J. AUTOMATION AND CONTROL》 *
陆玲莉 等: ""基于多观测器的无人机切换式故障诊断与调节"", 《东南大学学报(自然科学版)》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112947357A (en) * 2021-01-24 2021-06-11 西北工业大学 Aircraft intermittent fault diagnosis method based on double observers
CN112947392A (en) * 2021-04-05 2021-06-11 西北工业大学 Flight control system actuator and sensor composite tiny fault estimation method based on robust observer
CN112947392B (en) * 2021-04-05 2022-04-26 西北工业大学 Flight control system actuator and sensor composite tiny fault estimation method based on robust observer
CN114879512A (en) * 2022-07-06 2022-08-09 南京航空航天大学 Spacecraft formation orbit fault-tolerant control method based on learning neural network sliding mode
CN114879512B (en) * 2022-07-06 2022-11-08 南京航空航天大学 Spacecraft formation orbit fault-tolerant control method based on learning neural network sliding mode
CN115046526A (en) * 2022-08-08 2022-09-13 东方空间技术(北京)有限公司 Aircraft fault diagnosis method and device, computer equipment and storage medium
CN115046526B (en) * 2022-08-08 2022-10-25 东方空间技术(北京)有限公司 Aircraft fault diagnosis method and device, computer equipment and storage medium

Also Published As

Publication number Publication date
CN111897221B (en) 2021-10-15

Similar Documents

Publication Publication Date Title
CN111897221B (en) Spacecraft fault diagnosis method based on combined observer
Jianyong et al. Robust control for static loading of electro-hydraulic load simulator with friction compensation
CN107272639B (en) Detection, estimation and its adjusting method of rigid spacecraft reaction wheel failure
KR101021801B1 (en) Actuator fault diagnosis of UAVs using adaptive unknown input observers
CN111241748A (en) Wind driven generator fault diagnosis based on long-short term memory model recurrent neural network
JP2016511699A (en) Method for improving detection of collision between robot and its environment, system and computer program product for implementing the method
CN108427400B (en) Aircraft airspeed head fault diagnosis method based on neural network analytic redundancy
CN104049638A (en) Dynamic-actuator aircraft attitude distributed type fault-tolerant control system
CN115629547B (en) Control surface fault-oriented aircraft airborne fault-tolerant control method and system
Sotoudehnejad et al. Counteracting modeling errors for sensitive observer-based manipulator collision detection
Ossmann et al. Advanced sensor fault detection and isolation for electro-mechanical flight actuators
CN101241369A (en) Flight control system executor deadlock trouble real-time detection method
CN107703753A (en) A kind of fault tolerant control method of space manipulator
Qin et al. Sensor fault diagnosis of autonomous underwater vehicle based on LSTM
Han et al. Quadratic-Kalman-filter-based sensor fault detection approach for unmanned aerial vehicles
Hu et al. Dual neural networks based active fault-tolerant control for electromechanical systems with actuator and sensor failure
CN111173688A (en) Wind driven generator fault diagnosis and isolation method based on adaptive observer
Azam et al. In-flight fault detection and isolation in aircraft flight control systems
Kordestani et al. A distributed fault detection and isolation method for multifunctional spoiler system
CN104252575A (en) Behavior based UUV (unmanned underwater vehicle) propulsion operating system exception identification method
Al-Zyoud et al. Detection of actuator faults using a dynamic neural network for the attitude control subsystem of a satellite
CN112947357B (en) Aircraft intermittent fault diagnosis method based on double observers
Freeman et al. Analytical fault detection for a small UAV
Yu et al. Fault detection and isolation in a nonlinear electromechanical system
CN114217595B (en) X-type rudder AUV fault detection method based on interval observer

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
GR01 Patent grant
GR01 Patent grant