CN114815592B - Fault-tolerant control method for delta-wing aircraft based on concurrent learning neural network - Google Patents
Fault-tolerant control method for delta-wing aircraft based on concurrent learning neural network Download PDFInfo
- Publication number
- CN114815592B CN114815592B CN202110121733.0A CN202110121733A CN114815592B CN 114815592 B CN114815592 B CN 114815592B CN 202110121733 A CN202110121733 A CN 202110121733A CN 114815592 B CN114815592 B CN 114815592B
- Authority
- CN
- China
- Prior art keywords
- data points
- value
- neural network
- aircraft
- database
- 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.)
- Active
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
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a fault-tolerant control method of a delta wing aircraft, which is characterized in that a fault-tolerant mechanism is introduced on the basis of a basic controller by establishing a concurrent learning neural network, so that the aircraft can be processed and solved on line after the unmanned aircraft fails, and continuous working under a certain acceptable state is realized. Meanwhile, the concurrent learning can record historical data, and ensure the convergence of the weight of the neural network under the condition of no continuous excitation, so that the task with higher requirements can be executed.
Description
Technical Field
The invention relates to a fault-tolerant control method of a delta wing aircraft, in particular to a fault-tolerant control method of a delta wing aircraft based on a concurrent learning neural network.
Background
Unmanned vehicles are widely used in national defense and socioeconomic construction. Compared with a piloted aircraft, the unmanned aircraft has the characteristics of flexible maneuvering, quick deployment, low cost, no risk of casualties and the like. In the fields of national defense and socioeconomic, unmanned aerial vehicles have demonstrated extremely broad application potential and development value.
However, unmanned aerial vehicles have problems in themselves while being applied on a large scale, and despite achieving a certain degree of autonomous flight, they have a gap from human intelligence in terms of fault handling. The unmanned aerial vehicle can be interfered by factors such as wind, engine vibration and the like in the flight process, and the unmanned aerial vehicle is easier to lose control after being failed due to the lack of manual intervention and environmental uncertainty. The automation degree of the unmanned aerial vehicle system is higher and higher, the scale is larger and larger, the investment on the unmanned aerial vehicle system is larger and larger, if the control system cannot eliminate or 'tolerate' faults when the aircraft breaks down in the flight process, the aircraft can be paralyzed and disabled, and further, casualties and important losses of property can be caused. In order to solve the problems, the patent proposes a fault-tolerant control method for an unmanned aerial vehicle, so that after the unmanned aerial vehicle fails, the unmanned aerial vehicle can solve the problems online and continue to work under a certain acceptable state.
During operation of unmanned aircraft, various emergency conditions may be encountered, such as pneumatic nonlinearity under large maneuvers, or actuator failure. The concurrency learning neural network can realize on-line adjustment of the control law under the condition of not knowing fault information so as to solve unknown nonlinear interference and realize fault-tolerant control of the aircraft. Meanwhile, the concurrent learning can estimate the aerodynamic parameters of the aircraft and the control efficiency of the actuating mechanism, and is used for solving the faults of partial actuating mechanism failure, aerodynamic parameter change and the like.
For the above reasons, the present inventors have made intensive studies on the existing fault-tolerant control method of the delta wing aircraft, in order to expect to design a delta wing aircraft fault-tolerant control method based on a concurrent learning neural network, which can solve the above problems.
Disclosure of Invention
In order to overcome the problems, the inventor performs intensive research and designs a delta wing aircraft fault-tolerant control method based on a concurrent learning neural network, wherein in the method, a fault-tolerant mechanism is introduced on the basis of a basic controller by establishing the concurrent learning neural network, so that after an unmanned aircraft breaks down, the aircraft can be processed and solved on line, and continuous working under a certain acceptable state is realized. Meanwhile, the concurrent learning can record historical data, ensure the convergence of the weight of the neural network under the condition of no continuous excitation, and further can execute tasks with higher requirements, thereby completing the invention.
Specifically, the invention aims to provide a delta wing aircraft fault-tolerant control method based on a concurrent learning neural network, which comprises the following steps of:
step 1, performing fault-tolerant processing on an output value of an aircraft basic controller to obtain a pseudo control quantity u;
The output value of the aircraft basic controller comprises an output value of concurrent learning pneumatic parameters and an output value of concurrent control efficiency processing;
And step 2, outputting the accurate pseudo control quantity u to an executing mechanism, and further controlling the attitude angle of the delta wing aircraft.
In step 1, the pseudo control amount u is obtained by the following formula (one):
Wherein, Representing the flight state value of an ideal model, which is a matrix of ideal roll angle and ideal roll rate measured in real time by an aircraft in the ideal model,Representation matrixThe values of the second row, ax (2), represent the values of the second row of the matrix Ax;
x represents the actual flight status value of the vehicle, Wherein phi represents the aircraft roll angle and p represents the aircraft roll rate;
K P represents the proportional gain, K P = [ -1-1 ];
e represents the difference between the flight state value x ref of the ideal model and the actual state value x;
v ad denotes an output value processed by the concurrent learning neural network;
a represents the output value of the concurrently learned aerodynamic parameters,
The output value representing the concurrent control efficiency process is calculated from the control efficiency Λ of the actuator.
The concurrent learning neural network processing adopts a single hidden layer Radial Basis Function (RBF) neural network, and the output value v ad of the neural network is obtained by the following formula (II):
Wherein V represents the weight between the input layer and the hidden layer, and W represents the weight between the output layer and the hidden layer;
Sigma represents a nonlinear activation function of the hidden layer, preferably the nonlinear activation function is a gaussian basis function;
representing a neural network input layer;
The said Obtained by the following formula (III):
where b v denotes bias, set to constant 1;
x in is the input value of the input layer, and two input values are set here, which are phi and p actually measured by the aircraft.
The V and W are obtained by the following formulas (five) and (six):
Wherein, Representing the first derivative of W,Representing the first derivative of v;
i represents a unit matrix with proper dimension and is set as a 2-dimensional unit matrix;
Γ W denotes the gain of w, set to a constant 3;
Γ V denotes the gain of V, set to a constant 1;
Γ W1 represents the concurrence learning gain of W, set to a constant 1;
Γ V1 denotes the concurrent learning gain of V, set to a constant 1;
k represents the neural network damping gain, set to a constant;
I e i represents the norm of e, i.e. the norm of the difference between the flight state value of the ideal model and the actual flight state value;
Representing the residual signal;
Sigma' represents the derivative of the nonlinear activation function sigma of the hidden layer;
Neural network input layer representing records Is a value of (2);
V c represents an intermediate variable.
In the method, step 1 'is executed simultaneously with step 1, and step 1' comprises the following substeps:
Step 1' -1, synchronously starting to read data points when the aircraft starts to operate, reading every 0.005 seconds, and storing the read data points into a database;
Step 1' -2, when the database is fully stored, judging the data points read again, and listing the data points meeting the judging conditions as alternative data points;
step 1' -3, performing singular value maximization processing on the candidate data points and the data points in the database, and updating the data points in the database;
step 1' -4, obtaining residual signals in real time through a database
Preferably, the steps 1' -3 comprise the sub-steps,
A substep a, forming a data set from the alternative data points and 5 data points in the database;
step b, selecting 5 data points in the data set, and solving singular values of the data points;
Repeating the sub-step b 5 times, wherein each selected data point is not completely the same;
step d, sorting the obtained 6 singular values according to the size, and selecting the largest singular value;
and e, calling 5 data points corresponding to the maximum singular value, and updating a database by using the called 5 data points.
In step 1' -2, the determination conditions for the recorded data points are:
Wherein x (t) represents the flight state value of the data point read at the moment t, and x p represents the flight state value of the data point finally stored in the database;
the set determination condition is set to be constant 0.3.
5-10 Data points are recorded in the database at full load, preferably 5 data points are recorded in the database at full load.
The residual signal when the data points in the database are updatedObtained by the following formula (seven):
Where delta represents the calculated value of the model error,
Representing the second derivative of x i;
U i denotes a pseudo control amount U value recorded by the concurrent learning data recording process;
when the data points in the database are not updated, the residual signals The value is 0.
Output values of the concurrency control efficiency processObtained by the following formula (eight):
wherein Γ 2 represents a gain term, set to a constant of 1;
U i represents the pseudo control amount U value recorded by concurrent learning data recording process, B represents a matrix
Representation ofIs the first derivative of (a);
Y i represents the Y value recorded by the concurrent learning data recording process, wherein
Preferably, the a 11、a12、a21 and a 22 are obtained by the following formula (nine):
wherein k CL represents concurrent learning gain and is set to be constant 0.01;
Γ 1 denotes a gain matrix, set to a 4-dimensional identity matrix;
y (x) represents a state matrix and,
Y represents the integral of Y (x),Y i denotes y recorded by the concurrent learning data recording process;
θ represents a parameter of the aircraft model,
Representing the first derivative of θ.
The invention has the beneficial effects that:
(1) According to the delta wing aircraft fault-tolerant control method based on the concurrent learning neural network, provided by the invention, the control law can be reconstructed on line under the condition that the fault information is unknown, and the unknown nonlinear faults are canceled, so that the performance of the controller is recovered to a certain extent, and the fault-tolerant control is realized;
(2) According to the delta wing aircraft fault-tolerant control method based on the concurrent learning neural network, which is provided by the invention, the neural network weight can be ensured to converge towards the true value under the condition of no continuous excitation, the fault-tolerant control efficiency of the neural network is improved, and when the fault is unchanged, the neural network self-adaptive control based on the concurrent learning does not need to learn every excitation;
(3) According to the delta wing aircraft fault-tolerant control method based on the concurrent learning neural network, which is provided by the invention, a large amount of real-time calculation is not needed, and the method is easy to realize under the condition that the aircraft calculation resources are limited;
(4) According to the delta wing aircraft fault-tolerant control method based on the concurrent learning neural network, the concurrent learning is utilized to estimate the efficiency and the pneumatic parameters of the actuating mechanism, and the estimation can be carried out under the condition of no continuous excitation, so that the faults of reduced control efficiency of the actuating mechanism and changed pneumatic parameters are solved well.
Drawings
FIG. 1 illustrates a delta wing aircraft fault tolerance control methodology framework of the present invention concurrently learning neural networks;
FIG. 2 shows the actual roll angle tracking effect after an error occurs in an aircraft provided with a fault-tolerant control method in an experimental example of the present invention;
fig. 3 shows a graph of an output value v ad of an aircraft concurrency learning neural network provided with a fault-tolerant control method in an experimental example of the invention;
FIG. 4 shows a fitted curve of a neural network to unknown nonlinear interference in an experimental example of the present invention;
Fig. 5 shows the actual roll angle tracking effect after an error occurs in an aircraft without the fault-tolerant control method in the experimental example of the present invention.
Detailed Description
The invention is further described in detail below by means of the figures and examples. The features and advantages of the present invention will become more apparent from the description.
The word "exemplary" is used herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. Although various aspects of the embodiments are illustrated in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated.
According to the fault-tolerant control method of the delta wing aircraft provided by the invention, in the method,
The method comprises the following steps:
step 1, performing fault-tolerant processing on an output value of an aircraft basic controller to obtain an accurate pseudo control quantity u;
The output value of the aircraft basic controller comprises an output value of concurrent learning pneumatic parameters and an output value of concurrent control efficiency processing;
And step 2, outputting the accurate pseudo control quantity u to an executing mechanism, and further controlling the attitude angle of the delta wing aircraft.
In step 1, the pseudo control amount is obtained by the following formula (one):
Wherein, The flight state value of the ideal model is represented, and the flight state value is a matrix of an ideal rolling angle and an ideal rolling rate measured in real time by an aircraft in the ideal model; Representation matrix The values of the second row, ax (2), represent the values of the second row of the matrix Ax;
the ideal model is generally referred to as a reference model, which can respond to instructions as desired;
x represents the actual flight state value, which can be expressed as Wherein φ represents aircraft roll angle (rad), and p represents aircraft roll rate (rad/s);
And phi and p are obtained by reading the sensor in real time, wherein the sensor is a rate gyro sensor, and the working frequency of the sensor is 200Hz.
K P represents the proportional gain, K P = [ -1-1 ];
e represents the flight state value of the ideal model The difference from the actual flight status value x;
v ad denotes an output value processed by the concurrent learning neural network;
a represents the output value of the concurrently learned aerodynamic parameters,
The output value representing the concurrent control efficiency process is calculated from the control efficiency Λ of the actuator.
The aircraft base controller is a simple proportional integral derivative controller (PID), and the delta wing aircraft roll dynamic model is expressed as
Wherein, Is the first derivative of x byCalculating to obtain; is directly measured by a sensor, Obtained by taking the first derivative of p.
Delta wing aircraft are open loop instability in performing high angle roll maneuvers, which is known as the "wing roll phenomenon". This is caused by an asymmetric aerodynamic effect acting on the delta wing, which can cause an unstable roll motion of the aircraft. Therefore, the pseudo control amount u needs to be calculated and controlled, so that the delta wing aircraft is actively controlled.
In the invention, the frame of the delta wing aircraft fault-tolerant control method of the concurrent learning neural network is shown in figure 1, the output value of the aircraft basic controller is counteracted through fault-tolerant processing, the accurate pseudo control quantity u is output after the elimination, and then the accurate pseudo control quantity u is output to an executing mechanism to control the unmanned aircraft, so that the influence of sudden errors on the flight of the aircraft is reduced.
According to a preferred embodiment of the present invention, the concurrent learning neural network process uses a single hidden layer Radial Basis Function (RBF) neural network, and the output value v ad of the concurrent learning neural network process is obtained by the following formula (two):
Wherein V represents the weight between the input layer and the hidden layer, and W represents the weight between the output layer and the hidden layer;
Sigma represents a nonlinear activation function of the hidden layer, preferably the nonlinear activation function is a gaussian basis function;
representing a neural network input layer;
Preferably, the said Obtained by the following formula (III):
where b v denotes bias, set to constant 1;
x in is the input value of the input layer, and two input values are set here, which are phi and p actually measured by the aircraft.
The single hidden layer Radial Basis Function (RBF) neural network is characterized in that an hidden layer is arranged between the input and the output, namely the output of the input layer is the input of the hidden layer, the product of the output of the hidden layer and the corresponding weight is the input of the output layer, and the output of the output layer is the final output.
The nonlinear activation function plays a very important role in learning and understanding very complex and nonlinear functions of the artificial neural network model. They introduce non-linear characteristics into our network, where the input values are summed by weighting and then acted upon by a function that is a non-linear activation function. The nonlinear activation function is introduced to increase the nonlinearity of the neural network model so that the neural network can approximate any nonlinear function, and thus the neural network can be applied to a plurality of nonlinear models. Gaussian base functions, also called radial base functions, are a commonly used type of kernel function, usually defined as a monotonic function of the euclidean distance from any point in space to a certain center.
In the present application, the nonlinear activation function is expressed as the following formula (four):
Where j=1..n 2,bj is a positive scalar representing the width of the gaussian basis function, c j is a center vector, the same as the input parameter vector x dimension, and the euclidean distance between the two is defined as ||x-c j |.
According to the invention, the acquisition of the weight value is an important link for ensuring that the concurrent learning and the neural network weight adjustment algorithm are combined to ensure that the weight converges to the true value, and the result can be ensured to converge to the true value even if no external excitation exists in the historical data and the estimated parameters of the instantaneous data are adopted, and the historical data contains enough information.
The V and W are obtained by the following formulas (five) and (six):
Wherein, Representing the first derivative of W,Representing the first derivative of V;
i represents a unit matrix with proper dimension and is set as a dimension unit matrix;
Γ W denotes the gain of W, set to a constant 3;
Γ V denotes the gain of V, set to a constant 1;
Γ W1 represents the concurrence learning gain of W, set to a constant 1;
Γ V1 denotes the concurrent learning gain of V, set to a constant 1;
k represents the neural network damping gain and is set to be constant 0.01;
I e i represents the norm of e, i.e. the norm of the difference between the flight state value of the ideal model and the actual flight state value;
Representing the residual signal;
Sigma' represents the derivative of the nonlinear activation function sigma of the hidden layer;
Neural network input layer representing records Is a value of (2);
V c represents an intermediate variable, has no specific physical meaning, and can be calculated according to an expression.
According to a preferred embodiment of the present invention, the concurrent learning neural network process further includes a concurrent learning data recording process, that is, recording and calculating data points, determining which data points can be recorded, and making a decision on subsequent data.
In the method, step 1 'is executed simultaneously with step 1, and step 1' comprises the following substeps:
Step 1' -1, synchronously starting to read data points when the aircraft starts to operate, reading every 0.005 seconds, and storing the read data points into a database;
Step 1' -2, when the database is fully stored, judging the data points read again, and listing the data points meeting the judging conditions as alternative data points;
step 1' -3, performing singular value maximization processing on the candidate data points and the data points in the database, and updating the data points in the database;
step 1' -4, obtaining residual signals in real time through a database
The data points are the aircraft roll angle phi and the aircraft roll rate p read by the sensor at the current moment.
Preferably, the steps 1' -3 comprise the sub-steps,
A substep a, forming a data set from the alternative data points and 5 data points in the database;
step b, selecting 5 data points in the data set, and solving singular values of the data points;
Repeating the sub-step b 5 times, wherein each selected data point is not completely the same;
step d, sorting the obtained 6 singular values according to the size, and selecting the largest singular value;
and e, calling 5 data points corresponding to the maximum singular value, and updating a database by using the called 5 data points.
According to the invention, 5-10 data points are recorded in the database at full load, preferably 5 data points are recorded in the database at full load.
Since the convergence rate of concurrent learning data recording processes is proportional to the minimum singular value of the historical data stack, it is necessary to ensure that 1) the rank condition is satisfied and 2) the minimum singular value is maximized in order to obtain a good result.
The rank condition refers to that the rank of the historical data stack Z is equal to the number of rows of the true weight matrix, i.e. a weight true value W *∈Rm ×n is set, and rank (Z) =m. Ensuring that the rank condition is satisfied can ensure that the tracking error converges to 0, and realizing faster convergence speed if the minimum singular value is maximized.
Although the more the record points in the database are, the better, the longer the calculation time of the calculation program is, the more the record points in the database cannot be set, the more the record points are, the less the data stored in the database cannot accurately judge the data value, the misjudgment can be caused, and the accurate control of the aircraft cannot be achieved.
In step 1' -2, the determination conditions for the recorded data points are:
Wherein x (t) represents the flight state value of the data point read at the moment t, and x p represents the flight state value of the data point finally stored in the database;
the set determination condition is set to be constant 0.3.
And if the current time data point meets the judging condition, listing the t time data point as an alternative data point, performing the steps 1-3, performing matrix singular value maximization processing on the alternative data, and simultaneously, recording the t time data point in the database only when the matrix singular value maximization of the database is required to be met.
If the current data point meets the judging condition and does not meet the judging condition or meets the judging condition but does not meet the matrix singular value maximization of the database, the current data point is not recorded in the database.
According to the application, the maximization of the singular value can ensure that the data in the database are sufficiently different, and the maximization of the matrix minimum singular value of the database can better reflect the data condition of the aircraft in actual flight.
In steps 1' -3, the matrix singular value maximization process is as follows, first recording points sufficiently different from the previous data, if the number of data points stored exceeds the maximum allowable number, the algorithm will merge new data points in a manner of maximizing the minimum singular value of the data stack, in order to achieve this, the algorithm sequentially replaces the data points in the historical data stack with the current data points, stores the resulting minimum singular value in a variable, finds the maximum minimum singular value, and replaces the new data points with the original data points, the specific algorithm is as follows:
According to a preferred embodiment of the invention, the residual signal The difference between the instantaneous estimate representing the model error and the stored estimate of the model error can be used directly to adaptively update the training information of the law.
The residual signal when the data points in the database are updatedObtained by the following formula (seven):
Where delta represents the calculated value of the model error,
Representing the second derivative of x i;
x i represents the flight state value of the data point i point which is updated and recorded recently by the data point, and is measured by a sensor in real time;
U i denotes a pseudo control amount U value recorded by the concurrent learning data recording process;
when the data points in the database are not updated, the residual signals The value is 0.
Through the method, the weight values V and W of the concurrent learning neural network can be obtained through the aircraft roll angle and the aircraft roll rate obtained in real time by the delta wing aircraft, the output value processed by the concurrent learning neural network is further obtained, and the output value is counteracted with error data in a basic controller of the aircraft, so that the fault tolerance effect is achieved.
According to a preferred embodiment of the invention, the fault tolerance processing further comprises concurrent learning control efficiency processing, and the calculated value of the control efficiency in the real-time flight process of the aircraft can be obtained through the concurrent learning control efficiency processing
Calculated value of the control efficiencyObtained by the following formula (eight):
wherein Γ 2 represents a gain term, set to a constant of 1;
U i represents the pseudo control amount U value recorded by concurrent learning data recording process, B represents a matrix
Representation ofIs the first derivative of (a);
Y i represents the Y value recorded by the concurrent learning data recording process, in which
The real-time control efficiency of the aircraft can be calculated and output through concurrent learning and control efficiency estimation processing, the output value of the concurrent learning and control efficiency processing is used for calculating the pseudo control quantity u, and the pseudo control quantity u which is closer to the actual situation can be obtained, so that when the aircraft encounters a fault, the aircraft can be accurately controlled through the method, the unmanned aircraft is prevented from being crashed, and basic work tasks can be completed.
Preferably, the a 11、a12、a21 and a 22 are obtained by the following formula (nine):
wherein k CL represents concurrent learning gain and is set to be constant 0.01;
Γ 1 denotes a gain matrix, set as a constant 4-dimensional identity matrix;
y (x) represents a state matrix and,
Y represents the integral of Y (x),Y i denotes y recorded by the concurrent learning data recording process;
θ represents a parameter of the aircraft model,
Representing the first derivative of θ.
The output value A of concurrent learning pneumatic parameter processing can be obtained by solving the pneumatic parameter theta.
The output value A of the concurrent learning aerodynamic parameters can be obtained in real time through the concurrent learning aerodynamic parameter processing, and when aerodynamics influence the flight of the aircraft, the aerodynamic parameters are adjusted through the output value A, so that the influence on the aircraft caused by environmental factors is reduced, and the probability of the fault of the aircraft is reduced.
According to the fault-tolerant control method of the delta wing aircraft, provided by the invention, the control law can be reconstructed on line under the condition that the fault information is unknown, and the unknown nonlinear faults are canceled, so that the performance of the controller is recovered to a certain extent, and the fault-tolerant control is realized. By utilizing concurrent learning to estimate the efficiency and the pneumatic parameters of the actuator, the actuator can be estimated under the condition of no continuous excitation, so that the faults of reduced control efficiency and changed pneumatic parameters of the actuator can be well solved.
Examples:
the rolling dynamic model of the delta wing aircraft is selected as follows:
The calculation of the output value v ad of the concurrent learning neural network is carried out from the start time of the flight, v ad is 0 at the moment, and the failure occurs due to the influence of strong wind after 10s from the start of the flight.
The fault-tolerant control method is arranged in the aircraft, the aircraft obtains the phi and p values of the aircraft through detection, the output value v ad curve of the concurrent learning neural network is shown in figure 3, the fitting curve of the neural network to unknown nonlinear interference is shown in figure 4, and the output value of the concurrent control efficiency processing is obtainedAt the level of 0.75 of the total weight of the product,Kp=[-1 -1],
By passing throughThe pseudo control amount u is obtained in real time.
By the fault-tolerant control method in the aircraft, after the aircraft fails, the fault can be processed, the tracking of the roll angle is ensured, and the actual roll angle tracking effect is shown in fig. 2.
Comparative example:
The same general delta wing aircraft rolling dynamic model as the embodiment is selected, calculation of an output value v ad of the concurrent learning neural network is carried out from the starting moment of the flight, v ad is 0 at the moment, and after the beginning of the flight for 10 seconds, the aircraft breaks down due to the influence of strong wind.
The fault-tolerant control method is not arranged in the aircraft, the fault cannot be processed only through the basic controller after the fault occurs, and the actual roll angle tracking effect is shown in fig. 3.
In fig. 2 to 5, r represents a roll angle command, ref represents a roll angle in an ideal model, plant represents a roll angle during actual flight of the aircraft, and nonlinear represents a nonlinear fitted curve.
As can be seen from fig. 2, by the fault-tolerant control method in the application, the problem data can be processed in time, the pseudo control quantity of the aircraft can be corrected, the roll angle tracking can be ensured when an emergency is encountered, and further, the unmanned aerial vehicle is ensured not to crash under the fault condition, and the task can be continuously completed.
As can be seen from fig. 5, when an aircraft not provided with the fault-tolerant control method encounters an emergency, fault-tolerant processing cannot be performed, so that the unmanned aerial vehicle cannot track the roll angle, and the unmanned aerial vehicle may crash.
In summary, as can be seen from fig. 2 to fig. 5, the fault-tolerant control method in the present application can implement on-line adjustment of control law to solve unknown nonlinear interference and implement fault-tolerant control of an aircraft.
The invention has been described above in connection with preferred embodiments, which are, however, exemplary only and for illustrative purposes. On this basis, the invention can be subjected to various substitutions and improvements, and all fall within the protection scope of the invention.
Claims (9)
1. The delta wing aircraft fault-tolerant control method based on the concurrent learning neural network is characterized by comprising the following steps of:
step 1, performing fault-tolerant processing on an output value of an aircraft basic controller to obtain a pseudo control quantity u;
step 2, outputting the accurate pseudo control quantity u to an executing mechanism so as to control the attitude angle of the delta wing aircraft;
in step 1, the pseudo control amount u is obtained by the following formula (one):
Wherein, Representing the flight state value of an ideal model, which is a matrix of ideal roll angle and ideal roll rate measured in real time by an aircraft in the ideal modelRepresentation matrixThe values of the second row, ax (2), represent the values of the second row of the matrix Ax;
x represents the actual flight status value of the vehicle, Wherein phi represents the aircraft roll angle and p represents the aircraft roll rate;
K P represents the proportional gain, K P = [ -1-1 ];
e represents the flight state value of the ideal model The difference from the actual state value x;
v ad denotes an output value processed by the concurrent learning neural network;
a represents the output value of the concurrently learned aerodynamic parameters,
The output value representing the concurrent control efficiency process is calculated from the control efficiency Λ of the actuator.
2. The delta wing aircraft fault tolerance control method of claim 1,
The concurrent learning neural network processing adopts a single hidden layer Radial Basis Function (RBF) neural network, and the output value v ad of the neural network is obtained by the following formula (II):
Wherein V represents the weight between the input layer and the hidden layer, and W represents the weight between the output layer and the hidden layer;
Sigma represents a nonlinear activation function of an hidden layer, the nonlinear activation function being a gaussian basis function;
representing the neural network input layer.
3. The delta wing aircraft fault tolerance control method of claim 2,
The saidObtained by the following formula (III):
where b v denotes bias, set to constant 1;
x in is the input value of the input layer, and two input values are set here, which are phi and p actually measured by the aircraft.
4. A delta wing aircraft fault tolerance control method according to claim 3,
The V and W are obtained by the following formulas (five) and (six):
Wherein, Representing the first derivative of W,Representing the first derivative of V;
i represents a unit matrix with proper dimension and is set as a 2-dimensional unit matrix;
Γ W denotes the gain of W, set to a constant 3;
Γ V denotes the gain of V, set to a constant 1;
Γ W1 represents the concurrence learning gain of W, set to a constant 1;
Γ V1 denotes the concurrent learning gain of V, set to a constant 1;
kappa represents the neural network damping gain, set to a constant of 0.01;
I e i represents the norm of e, i.e. the norm of the difference between the flight state value of the ideal model and the actual flight state value;
Representing the residual signal;
Sigma' represents the derivative of the nonlinear activation function sigma of the hidden layer;
Neural network input layer representing records Is a value of (2);
V c represents an intermediate variable.
5. The delta wing aircraft fault tolerance control method of claim 1,
In the method, step 1 'is executed simultaneously with step 1, and step 1' comprises the following substeps:
Step 1' -1, synchronously starting to read data points when the aircraft starts to operate, reading every 0.005 seconds, and storing the read data points into a database;
Step 1' -2, when the database is fully stored, judging the data points read again, and listing the data points meeting the judging conditions as alternative data points;
step 1' -3, performing singular value maximization processing on the candidate data points and the data points in the database, and updating the data points in the database;
step 1' -4, obtaining residual signals in real time through a database
The steps 1' -3 comprise the sub-steps of,
A substep a, forming a data set from the alternative data points and 5 data points in the database;
step b, selecting 5 data points in the data set, and solving singular values of the data points;
repeating the sub-step b 5 times, wherein each selected data point is not completely the same;
step d, sorting the obtained 6 singular values according to the size, and selecting the largest singular value;
and e, calling 5 data points corresponding to the maximum singular value, and updating a database by using the called 5 data points.
6. The delta wing aircraft fault tolerance control method of claim 5,
In step 1' -2, the determination conditions for the recorded data points are:
Wherein x (t) represents the flight state value of the data point read at the moment t, and x p represents the flight state value of the data point finally stored in the database;
the set determination condition is set to be constant 0.3.
7. The delta wing aircraft fault tolerance control method of claim 5,
And 5-10 data points are recorded when the database is full.
8. The delta wing aircraft fault tolerance control method of claim 4,
The residual signal when the data points in the database are updatedObtained by the following formula (seven):
Where delta represents the calculated value of the model error,
Representing the second derivative of x i;
U i denotes a pseudo control amount U value recorded by the concurrent learning data recording process;
when the data points in the database are not updated, the residual signals The value is 0.
9. The delta wing aircraft fault tolerance control method of claim 8,
Output values of the concurrency control efficiency processObtained by the following formula (eight):
wherein Γ 2 represents a gain term, set to a constant of 1;
b represents a matrix
Representation ofIs the first derivative of (a);
Y i represents the Y value recorded by the concurrent learning data recording process, in which
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110121733.0A CN114815592B (en) | 2021-01-28 | 2021-01-28 | Fault-tolerant control method for delta-wing aircraft based on concurrent learning neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110121733.0A CN114815592B (en) | 2021-01-28 | 2021-01-28 | Fault-tolerant control method for delta-wing aircraft based on concurrent learning neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114815592A CN114815592A (en) | 2022-07-29 |
| CN114815592B true CN114815592B (en) | 2025-06-17 |
Family
ID=82526179
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110121733.0A Active CN114815592B (en) | 2021-01-28 | 2021-01-28 | Fault-tolerant control method for delta-wing aircraft based on concurrent learning neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114815592B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110673616A (en) * | 2019-08-30 | 2020-01-10 | 张超凡 | Limited time fault-tolerant control method for fixed-wing unmanned aerial vehicle based on self-adaptive sliding mode |
Family Cites Families (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6092919A (en) * | 1995-08-01 | 2000-07-25 | Guided Systems Technologies, Inc. | System and method for adaptive control of uncertain nonlinear processes |
| US8489528B2 (en) * | 2009-07-28 | 2013-07-16 | Georgia Tech Research Corporation | Systems and methods for training neural networks based on concurrent use of current and recorded data |
| CN104035447A (en) * | 2014-06-27 | 2014-09-10 | 金陵科技学院 | Dynamic control reallocation based unmanned aerial vehicle attitude fault-tolerance control method |
| CN106843254B (en) * | 2017-03-08 | 2019-08-09 | 北京航天自动控制研究所 | A Real-Time Active Reconfiguration Fault-Tolerant Control Method |
| CN108107899B (en) * | 2018-01-19 | 2020-09-18 | 北京航空航天大学 | Energy-optimized large-angle attitude maneuver control method for linear pseudospectral outer space vehicle |
| CN108375907B (en) * | 2018-03-28 | 2020-11-10 | 北京航空航天大学 | Adaptive compensation control method of hypersonic aircraft based on neural network |
| CN108628167B (en) * | 2018-05-08 | 2021-11-12 | 南京邮电大学 | Fault-tolerant control method of flight control system based on adaptive distribution technology |
| CN109164708B (en) * | 2018-10-31 | 2020-08-04 | 南京航空航天大学 | Neural network self-adaptive fault-tolerant control method for hypersonic aircraft |
| FR3090851B1 (en) * | 2018-12-20 | 2021-03-19 | Thales Sa | AUTOMATIC LEARNING IN AVIONICS |
| CN110320927A (en) * | 2019-07-26 | 2019-10-11 | 中国人民解放军国防科技大学 | Flight control method and system of intelligent deformable aircraft |
| CN110824925A (en) * | 2019-11-29 | 2020-02-21 | 天津大学 | Adaptive robust fault-tolerant control method for tilting type three-rotor unmanned aerial vehicle |
| CN111474922B (en) * | 2020-05-08 | 2021-07-16 | 哈尔滨工业大学 | A Controller Construction Method for Continuous Nonlinear Systems |
| CN111781942B (en) * | 2020-06-23 | 2021-10-12 | 南京航空航天大学 | Fault-tolerant flight control method based on self-constructed fuzzy neural network |
-
2021
- 2021-01-28 CN CN202110121733.0A patent/CN114815592B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110673616A (en) * | 2019-08-30 | 2020-01-10 | 张超凡 | Limited time fault-tolerant control method for fixed-wing unmanned aerial vehicle based on self-adaptive sliding mode |
Non-Patent Citations (2)
| Title |
|---|
| A Singular Value Maximizing Data Recording Algorithm for Concurrent Learning;G. Chowdhary et al.;ISI_Web of Science;20110701;3547-3552 * |
| Concurrent Learning for Improved Convergence in Adaptive Flight Control;Girish Chowdhary et al.;AEROSPACE RESEARCH CENTRAL;20100805;第2-7, 11-19页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114815592A (en) | 2022-07-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11772264B2 (en) | Neural network adaptive tracking control method for joint robots | |
| Venugopal et al. | On-line learning control of autonomous underwater vehicles using feedforward neural networks | |
| WO2020142984A1 (en) | Active fault tolerant control method of aero-engine based on error interval observer | |
| CN111781942B (en) | Fault-tolerant flight control method based on self-constructed fuzzy neural network | |
| CN113568423A (en) | An intelligent fault-tolerant control method for quadrotor UAV considering motor failure | |
| CN110727198B (en) | Fault-tolerant control method for missile multi-drive-state actuating mechanism faults | |
| CN104076690B (en) | The adaptive location of a kind of nonlinear ship dynamical system follows the tracks of fault tolerant control method | |
| CN117369267B (en) | Method and system for controlling underwater helicopter surrounding formation under event triggering frame | |
| CN108764122A (en) | A kind of optimization of AUV control systems and fault monitoring method based on deep learning | |
| CN111474922A (en) | A Controller Construction Method for Continuous Nonlinear Systems | |
| CN116088556A (en) | Intelligent fault-tolerant control method for aircraft based on deep reinforcement learning | |
| CN117452827B (en) | Under-actuated unmanned ship track tracking control method | |
| CN114815592B (en) | Fault-tolerant control method for delta-wing aircraft based on concurrent learning neural network | |
| CN115659502A (en) | Control surface fault diagnosis method of flying wing unmanned aerial vehicle based on neural network adaptive observer | |
| CN114489057A (en) | Data-driven multi-unmanned surface vessel containing control method and system and storage medium | |
| Fravolini et al. | Minimal resource allocating networks for aircraft SFDIA | |
| Sharma et al. | A model-free deep reinforcement learning approach to maneuver A quadrotor despite single rotor failure | |
| CN119270846A (en) | Robust control method, device, system and storage medium for underactuated unmanned boat | |
| CN117539239A (en) | Unmanned ship track tracking controller structure based on meta-learning and control method | |
| Hussain et al. | Aircraft sensor estimation for fault tolerant flight control system using fully connected cascade neural network | |
| CN116088309A (en) | Compound learning fault-tolerant control method for surface ship based on fault identification | |
| CN116880182A (en) | Intelligent control method for aircraft model with full-state constraint and actuator fault | |
| Yuan et al. | Intelligent Event-triggering Fault Estimation and Fault Tolerant Control of A Quadrotor UAV with Actuator Fault | |
| CN115097854A (en) | An attitude adaptive control method for fixed-wing UAV based on model correction | |
| Yang et al. | Reinforcement learning-based position control for a disc-shaped underwater remotely operated vehicle |
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 |