CN116127842A - Post-fault flight envelope online prediction method based on radial basis-counter propagation neural network - Google Patents

Post-fault flight envelope online prediction method based on radial basis-counter propagation neural network Download PDF

Info

Publication number
CN116127842A
CN116127842A CN202310062283.1A CN202310062283A CN116127842A CN 116127842 A CN116127842 A CN 116127842A CN 202310062283 A CN202310062283 A CN 202310062283A CN 116127842 A CN116127842 A CN 116127842A
Authority
CN
China
Prior art keywords
aircraft
fault
angle
coefficient
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310062283.1A
Other languages
Chinese (zh)
Inventor
吴震
尹楚
陆宇平
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.)
Nanjing University of Aeronautics and Astronautics
Original Assignee
Nanjing University of Aeronautics and Astronautics
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 Nanjing University of Aeronautics and Astronautics filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN202310062283.1A priority Critical patent/CN116127842A/en
Publication of CN116127842A publication Critical patent/CN116127842A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/15—Vehicle, aircraft or watercraft design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Computer Hardware Design (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Automation & Control Theory (AREA)
  • Medical Informatics (AREA)
  • Traffic Control Systems (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses an online safety envelope prediction method based on a radial basis-counter propagation neural network, which can prevent an aircraft from being out of control after sudden structural damage and pneumatic faults occur. Commands for the on-board flight controller are first sent to the aircraft actuators and sensors to monitor the health of the actuators. Then, the aircraft state and sensor bias are estimated based on the aircraft dynamics model using a Kalman filter and the stability derivative is estimated using a recursive least squares method. And then, carrying out aerodynamic effect modeling, and establishing an aircraft aerodynamic model under normal and fault states. And finally, by utilizing the calculated dimensionless force and moment, the output of the normal flight model is compared with the actual aircraft output measured value, and an online aerodynamic anomaly detection process is started. The potential fault location and scale are determined based on the identified stability derivatives. And (3) applying a database retrieval scheme and an interpolation algorithm according to the estimated damage condition to obtain the safety envelope under the current fault condition.

Description

Post-fault flight envelope online prediction method based on radial basis-counter propagation neural network
Technical Field
The invention belongs to the technical field of fault diagnosis, and particularly relates to an online prediction method of a post-fault flight envelope based on a radial basis-back propagation (Radial Basis Function-back propagation RBF-BP) neural network.
Background
Structural damage and aerodynamic faults are important factors threatening the safe flight of the aircraft, and the research of flight envelope prediction under the fault condition has important significance for ensuring the safe flight of the aircraft. The safety envelope may provide a reference to the pilot during normal flight. However, when sudden structural damage occurs to the aircraft, these envelopes may no longer be effective, as aircraft failure typically affects the aircraft's flight performance, resulting in a reduced safety flight envelope for the aircraft. Therefore, when an aircraft suffers structural damage, real-time online prediction of the safety envelope is required, and the resulting safety envelope can be transmitted to the pilot in real-time. However, the real-time performance of the safety envelope prediction in the prior art is insufficient, and a new safety envelope cannot be fed back to a pilot in time. Therefore, the invention provides the fault diagnosis and the safety envelope prediction based on the offline database, and the efficiency of the fault diagnosis and the safety envelope prediction is greatly improved.
The invention provides an online prediction method of a post-fault flight envelope based on a radial basis-counter propagation neural network. According to the method, based on an offline database obtained by modeling the aerodynamic effect of the damaged aircraft, the neural network is trained offline, so that the fault of the aircraft can be diagnosed online in real time, and a new safety envelope can be obtained online.
Disclosure of Invention
In order to complete the online prediction of the post-fault aircraft safety envelope, the invention provides an online prediction method of the post-fault flight envelope based on a radial basis-counter propagation neural network, which aims to solve the problem of difficult prediction of the damaged aircraft flight envelope in the prior art. With the help of the offline database, the challenges associated with the high computational cost of acquiring a global model and a safe envelope prediction of a damaged aircraft can be overcome. And detecting and identifying the damage state of the aircraft by using a mode classification technology and using the locally estimated stability derivative as a classification characteristic, so as to find a correct database index.
In order to achieve the above purpose, the invention adopts the following technical scheme:
an online prediction of post-fault flight envelope based on radial basis-counter propagating neural network, comprising the steps of:
s1, building an aircraft health state monitoring module;
s2, identifying pneumatic parameters;
s3, modeling the structural damage aerodynamic effect and constructing a fault database;
s4, training an rbf-bp neural network and diagnosing faults in real time;
and S5, establishing a flight envelope database.
Preferably, in the step S1, an aircraft health status monitoring module is established, including the following steps:
(1) Command delta from an on-board flight controller cmd Is received by an actuator;
(2) In the event of a fault, the actual output delta of the actuator abn And an expected value delta calculated from its mathematical model nom Abnormal residual errors between the two sensors are detected by the sensors;
(3) When a residual error occurs, an online aerodynamic anomaly detection process will be initiated, with new measurements of the flight status and response being sent to the system identification module.
Preferably, in the step S2, the identifying of the aerodynamic parameters includes the following steps:
(1) Calculating dimensionless forces and moments along each axis using the sensor measurements or state values estimated by the kalman filter, the dimensionless aerodynamic forces and moments of the aircraft being obtainable from equations (1) and (2);
Figure BDA0004061396790000021
Figure BDA0004061396790000022
wherein C is L 、C D 、C Y 、C l 、C m 、C n Respectively a lift coefficient, a drag coefficient, a side force coefficient, a roll moment coefficient, a pitch moment coefficient and a yaw moment coefficient, alpha is an attack angle, beta is a sideslip angle, m is aircraft mass, A X 、A Y 、A Z Acceleration components along X, Y, Z axis, ρ is air density, V is airspeed, S is wing area, I xx 、I yy 、I zz 、I xz For moment of inertia and moment of product, p, q, r are roll, yaw and pitch rates respectively,
Figure BDA0004061396790000023
derivatives for roll, yaw and pitch rates, b, etc.>
Figure BDA0004061396790000024
The span and average chord length are respectively;
(2) Estimating a stability derivative of the aircraft from the simulation data using a recursive least squares method by equation (3);
Figure BDA0004061396790000031
in the method, in the process of the invention,
Figure BDA0004061396790000032
the basic increment coefficient of the lifting force, the increment coefficient related to the lifting force and the attack angle, the increment coefficient related to the lifting force and the yaw angle and the increment coefficient related to the lifting force and the elevator are respectively;
(3) Parameter estimation is carried out by adopting a recursive least square method with forgetting factors, and the recursive least square method has the following structure:
Figure BDA0004061396790000033
Figure BDA0004061396790000034
for the parameter to be estimatedNumber lambda E [0,1]]As a forgetting factor, a variable forgetting factor is used to enhance the effect of new data when model parameters change suddenly and reduce covariance matrix saturation under steady-state conditions.
Preferably, in the step S3, the modeling of the structural damage aerodynamic effect and the construction of the fault database include the following steps:
(1) According to a series of wind tunnel experiments, each damage condition can generate a unique aerodynamic effect on the aircraft and change different stability derivatives. Horizontal stabilizer damage can lead to changes in longitudinal stability, which can be determined by
Figure BDA0004061396790000035
A representation; vertical tail wing tip damage mainly results in changes in lateral force and directional stability, which can be determined by
Figure BDA0004061396790000036
Is represented by a change value of (c).
(2) The change value at the damage degree can be obtained by the following formula;
Figure BDA0004061396790000037
in the formula, delta C, C d C is the degree of damage of the body, the dimensionless aerodynamic coefficient of the damaged body and the dimensionless aerodynamic coefficient of the complete body respectively;
(3) By analyzing the wind tunnel data, an approximate linear relationship between the range of variation of each aerodynamic coefficient and the percent loss of wing tip can be assumed; according to the calculated damage degree variation scale, a linear interpolation is used, a pneumatic damage model under each damage level can be established, and a fault pneumatic database comprising pneumatic data and corresponding stability derivative of the horizontal stabilizer damage and the vertical tail wing tip damage is generated.
Preferably, in the step S4, the rbf-bp neural network training and the real-time fault diagnosis include the following steps:
(1) Determining the input variable x as at each Mach
Figure BDA0004061396790000041
The output variable is the specific fault type and damage degree corresponding to the pneumatic data;
(2) Prior to rbf-bp neural network training, the dataset was run at 7:3, dividing the data into a training set and a testing set, and then normalizing the data to be between [0,1 ];
(3) Initializing an rbf network, selecting t different initial clustering centers, setting the iteration step number to be 1, and immediately selecting the input from the samples. The closest distance between the input sample and the cluster center can be obtained by the following formula:
iX k ＝min||X k -C i (m)||，i＝1,2,...,t，k＝1,2,...,n (6)
(4) Calculating and adjusting the center of hidden layer nodes:
Figure BDA0004061396790000042
when C i (m+1)＝C i And (m) ending the clustering process, and determining the width of the hidden layer node according to the clustering center. If the input sample and the center distance are not equal, the calculation of the input sample and the center distance is carried out again until the learning process is completed. The mean square error target was set to 0.05, the radial basis expansion rate was 1, and the maximum neuron number was set to 25 by default.
(5) For a bp neural network, setting the number of input neurons n=6, the number of output neurons m=10, the training frequency e=1000, the hidden layer neuron number h=13, and using log sig and purelin functions as an activation function, wherein the hidden layer neuron number h is obtained by the following formula:
Figure BDA0004061396790000043
where α is any integer between 1 and 10.
Preferably, in the step S5, a flight envelope database is established, including the following steps:
(1) The plane maximum speed can be obtained by:
Figure BDA0004061396790000044
T max for thrust, ρ (H) is the air density in relation to altitude, C xmin S is the wing area, which is the drag coefficient;
(2) The plane fly minimum speed can be obtained by:
Figure BDA0004061396790000051
g is the weight of the aircraft, C zmax Is the lift coefficient;
(3) The maximum altitude rise of an aircraft in constant velocity straight line fly can be obtained by:
Figure BDA0004061396790000052
v z the maximum rise rate of the aircraft is represented by K, the lift-drag ratio is represented by K, and the flat flying speed is represented by v;
(4) The step of establishing a flight envelope database is as follows:
constraint conditions:
Figure BDA0004061396790000053
flat flight conditions:
Figure BDA0004061396790000054
basic trim control parameters:
λ＝[θ,ψ,δ T ,δ e ,δ r ] (14)
optimizing an objective function:
Figure BDA0004061396790000055
wherein A-I is an optimized parameter, dr is resistance, L is lift force, T is thrust along the machine body, phi, psi and theta are rolling angle, yaw angle and pitch angle respectively,
Figure BDA0004061396790000056
the roll angle and yaw and pitch angle derivatives, respectively, p, q, r are roll, yaw and pitch rates, respectively,/->
Figure BDA0004061396790000057
Derivatives of roll, yaw and pitch rates, V is the flight speed, h is the altitude, V 0 And h 0 For a speed and height at a horizontal tail offset angle of 0 DEG, beta is the sideslip angle, +.>
Figure BDA0004061396790000058
Is the derivative of sideslip angle, alpha is the angle of attack, +.>
Figure BDA0004061396790000059
Is the derivative of angle of attack, m and g are the mass of the aircraft and the acceleration of gravity, delta T 、δ e 、δ r Rudder deflection angle caused by thrust, elevating aileron deflection angle, rudder deflection angle caused by resistance,/->
Figure BDA00040613967900000510
Is a high derivative>
Figure BDA00040613967900000511
Is the speed derivative;
when the aircraft fails, balancing the aircraft at each designated altitude and speed, a set of different drag and lift coefficients are obtained; from equations (1) - (3), the maximum and minimum speeds and the rise limit at this time can be estimated; based on a series of such estimations, a flight envelope for such fault conditions is obtained, which in turn results in a database of flight envelopes for various fault conditions.
Compared with the prior art, the invention has the following beneficial effects:
according to the method, aerodynamic effect modeling is carried out on the fault aircraft, aerodynamic data of the damaged aircraft are obtained, and a database is built. The radial basis-backpropagation neural network is trained in conjunction with a database to generate classifiers and to generate decision surfaces. And estimating the flight envelope in real time by applying the classification result, and generating a new flight envelope according to the offline database. Compared with the conventional dynamic envelope prediction method, the method adopts the mode of establishing an offline database, carrying out fault diagnosis and generating a safety envelope in real time, avoids dimension disasters and greatly reduces calculation time. Finally, the feasibility and the efficiency of the method are proved through simulation verification.
Drawings
FIG. 1 is an overall frame diagram of a flight envelope prediction system;
FIG. 2 is a pneumatic parameter identification flow chart;
FIG. 3 is a schematic diagram of a fault diagnosis flow;
FIG. 4 is a schematic diagram of a rbf-bp neural network learning process;
FIG. 5 is a schematic diagram of fault diagnosis accuracy for different sample numbers;
FIG. 6 is a schematic illustration of aerodynamic effects of different degrees of wing damage;
FIG. 7 is a diagram of an rbf-bp network versus different ma
Figure BDA0004061396790000061
Is a predictive diagram of (a).
Detailed Description
The invention will be further illustrated with reference to examples.
Aiming at the characteristic that the safety envelope of the fault aircraft is difficult to predict, the invention provides an online prediction method of the post-fault flight envelope based on an rbf-bp neural network, which can diagnose various faults on line in real time through an offline database and generate a new safety envelope. The method first utilizes sensors to monitor the health of the actuator. If any failure occurs, an abnormal residual between the actual output of the actuator and the expected value will soon detect the failure. The aircraft state and sensor bias are then estimated based on the aircraft dynamics model using a kalman filter or other advanced state estimator. And secondly, calculating dimensionless forces and moments on each shaft by using the estimated states and the sensor information, and providing input for the second step of the two-step method, namely estimating the stability derivative by using a recursive least squares method. And finally, by utilizing the calculated dimensionless force and moment, the output of the normal flight model is compared with the actual aircraft output measured value, and an online aerodynamic anomaly detection process is started. At the same time, an alarm is generated based on the newly identified stability derivative, a fault classification is triggered, and the likely location and scale of the fault is determined. Once the damage condition is estimated, this information is provided as an index to a database, and a database retrieval scheme and interpolation algorithm are applied to obtain the unique safe flight envelope closest to the current fault condition. The resulting safe flight envelope may be presented to the pilot and used by the fault tolerant controller to generate a new control law. Simulation results demonstrate the feasibility of this approach, successfully detecting and classifying two damage conditions.
Example 1
Step S1, building an aircraft health status monitoring module: input data delta nom 、δ abn For the expected values calculated from its normal pneumatic model and the actual output of the actuator, as shown in fig. 1. Command delta from an on-board flight controller cmd Is received by the actuator, and when a fault occurs, the actual output delta of the actuator abn And an expected value delta calculated from its mathematical model nom The anomaly residual between them can be detected quickly by the sensor. When the residual error occurs, an online aerodynamic anomaly detection process is started, and new measurement values of the flight state and response are sent to a system identification module.
Step S2, identifying pneumatic parameters: as shown in fig. 2, first, dimensionless aerodynamic force and moment of the aircraft are calculated by the formula (1) and the formula (2);
Figure BDA0004061396790000071
Figure BDA0004061396790000072
wherein C is L 、C D 、C Y C is the lift coefficient, the drag coefficient and the side force coefficient respectively l 、C m 、C n Respectively a roll moment coefficient, a pitch moment coefficient and a yaw moment coefficient, alpha is an attack angle, beta is a sideslip angle, m is aircraft mass, A X 、A Y 、A Z Acceleration components along X, Y, Z axis, ρ is air density, V is airspeed, S is wing area, I xx 、I yy 、I zz 、I xz For moment of inertia and moment of product, p, q, r are roll, yaw and pitch rates, b,
Figure BDA0004061396790000081
The span and average chord lengths, respectively. C (C) L 、C D 、C Y 、C l 、C m 、C n The remaining parameters are known or can be obtained by an on-board sensor for the required unknowns. Estimating a model structure based on the input-output model of formula (3):
Figure BDA0004061396790000082
in the method, in the process of the invention,
Figure BDA0004061396790000083
the basic increment coefficient of the lifting force, the increment coefficient related to the lifting force and the attack angle, the increment coefficient related to the lifting force and the yaw angle and the increment coefficient related to the lifting force and the elevator are respectively as above. The estimation method employed herein is the recursive least squares method. When model parameters change suddenly, a variable forgetting factor is used to enhance the effect of new data and reduce covariance matrix saturation under steady-state conditions. />
Figure BDA0004061396790000084
And as the parameter to be estimated, the value range of the forgetting factor alpha is 0.95-0.99, and the initial value is the stability derivative value of the aircraft in the normal state. The recursive least square method has the structure shown in the formula (4):
Figure BDA0004061396790000085
in step S3, modeling of structural damage aerodynamic effect and construction of a fault database:
according to a series of wind tunnel experiments, each damage condition can generate a unique aerodynamic effect on the aircraft and change different stability derivatives. Horizontal stabilizer damage can lead to changes in longitudinal stability, which can be determined by
Figure BDA0004061396790000086
A representation; vertical tail wing tip damage mainly results in changes in lateral force and directional stability, which can be determined by
Figure BDA0004061396790000087
Is represented by a change value of (c). The scale of variation in the extent of damage can be obtained from the following formula;
Figure BDA0004061396790000088
in the formula, delta C, C d C is the degree of damage to the body, the dimensionless aerodynamic coefficient of the damaged body and the dimensionless aerodynamic coefficient of the complete body, respectively. By analyzing the wind tunnel data, an approximately linear relationship between the range of variation of each aerodynamic coefficient and the percent tip loss can be assumed. According to the calculated damage degree variation scale, a linear interpolation is used, a pneumatic damage model under each damage level can be established, and a fault pneumatic database comprising pneumatic data and corresponding stability derivative of the horizontal stabilizer damage and the vertical tail wing tip damage is generated. As shown in fig. 3, the established fault pneumatic database will be used to train the neural network.
S4, training an rbf-bp neural network and diagnosing faults in real time:
and determining the input variable x as the specific fault type and damage degree corresponding to the pneumatic data under each Mach number, and determining the output variable x as the specific fault type and damage degree corresponding to the pneumatic data. Prior to rbf-bp neural network training, the dataset was run at 7:3 is divided into a training set and a testing set, and then normalization processing is carried out on the data, so that the data is normalized to be between [0,1 ]. Initializing an rbf network, selecting t different initial clustering centers, setting the iteration step number to be 1, and immediately selecting the input from the samples. The closest distance between the input sample and the cluster center can be obtained by the following formula:
iX k ＝min||X k -C i (m)||，i＝1,2,...,t，k＝1，2，...，n (6)
the center of calculating and adjusting hidden layer nodes can be obtained by equation (7),
Figure BDA0004061396790000091
when C i (m+1)＝C i And (m) ending the clustering process, and determining the width of the hidden layer node according to the clustering center. If the input sample and the center distance are not equal, the calculation of the input sample and the center distance is carried out again until the learning process is completed. The neural network training process is shown in fig. 4. The mean square error target was set to 0.05, the radial basis expansion rate was 1, and the maximum neuron number was set to 25 by default. For bp neural network, set the input neuron number n=12, output neuron number m=10, training frequency e=1000, activation function uses log sig and purelin functions, hidden layer neuron number h=13 is obtained by the following formula:
Figure BDA0004061396790000092
where α is any integer between 1 and 10. The results shown in fig. 5, 6 and 7 show the effectiveness of the algorithm proposed by the present invention.
In step S5, a flight envelope database is established:
the maximum and minimum speeds of the plane and the maximum height rise of the plane in the constant speed straight line can be obtained by the formulas (6), (7) and (8):
Figure BDA0004061396790000101
Figure BDA0004061396790000102
Figure BDA0004061396790000103
T max for thrust, ρ (H) is the air density in relation to altitude, C xmin Is the resistance coefficient; g is the weight of the aircraft, C zmax Is the lift coefficient. v z K is the lift-drag ratio, which is the maximum lift-off rate of the aircraft.
And then establishing a flight envelope database:
constraint conditions:
Figure BDA0004061396790000104
flat flight conditions:
Figure BDA0004061396790000105
basic trim control parameters:
Figure BDA0004061396790000106
optimizing an objective function:
Figure BDA0004061396790000107
wherein A-I is an optimization parameter, dr is resistanceForce L is lift force, T is thrust along the machine body, phi and theta are roll angle, yaw angle and pitch angle respectively,
Figure BDA0004061396790000108
the roll angle and yaw and pitch angle derivatives, respectively, p, q, r are roll, yaw and pitch rates, respectively,/->
Figure BDA0004061396790000109
Derivatives of roll, yaw and pitch rates, V is the flight speed, h is the altitude, V 0 And h 0 For a speed and height at a horizontal tail offset angle of 0 DEG, beta is the sideslip angle, +.>
Figure BDA00040613967900001010
Is the derivative of sideslip angle, alpha is the angle of attack, +.>
Figure BDA00040613967900001011
Is the derivative of angle of attack, m and g are the mass of the aircraft and the acceleration of gravity, delta T 、δ e 、δ r Rudder deflection angle caused by thrust, elevating aileron deflection angle, rudder deflection angle caused by resistance,/->
Figure BDA00040613967900001012
Is a high derivative>
Figure BDA00040613967900001013
Is the speed derivative;
the trim of an aircraft is an optimization problem that satisfies initial values and constraint limits. When an aircraft fails, trimming the aircraft at each designated altitude and speed results in a set of different drag and lift coefficients. From equations (1) - (3), the maximum and minimum speeds and the rise limit at this time can be estimated. Based on a series of such estimations, a flight envelope for such fault conditions is obtained, which in turn results in a database of flight envelopes for various fault conditions.
The foregoing is only a preferred embodiment of the invention, it being noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the present invention, and such modifications and adaptations are intended to be comprehended within the scope of the invention.

Claims (6)

1. The post-fault flight envelope on-line prediction method based on the radial basis-counter propagation neural network is characterized by comprising the following steps of:
s1, building an aircraft health state monitoring module;
s2, identifying pneumatic parameters;
s3, modeling the structural damage aerodynamic effect and constructing a fault database;
s4, training an rbf-bp neural network and diagnosing faults in real time;
and S5, establishing a flight envelope database.
2. The method for online prediction of post-fault flight envelope based on radial basis-counter-propagating neural network according to claim 1, wherein in step S1, an aircraft health status monitoring module is established, comprising the steps of:
(1) Command delta from an on-board flight controller cmd Is received by an actuator;
(2) In the event of a fault, the actual output delta of the actuator abn And an expected value delta calculated from its mathematical model nom Abnormal residual errors between the two sensors are detected by the sensors;
(3) When a residual error occurs, an online aerodynamic anomaly detection process will be initiated, with new measurements of the flight status and response being sent to the system identification module.
3. The method for online prediction of post-fault flight envelope based on radial basis-counter-propagating neural network according to claim 1, wherein in step S2, the aerodynamic parameter identification comprises the following steps:
(1) Calculating dimensionless forces and moments along each axis using the sensor measurements or state values estimated by the kalman filter, the dimensionless aerodynamic forces and moments of the aircraft being derived from equations (1) and (2);
Figure FDA0004061396780000011
Figure FDA0004061396780000012
wherein C is L 、C D 、C Y 、C l 、C m 、C n Respectively a lift coefficient, a drag coefficient, a side force coefficient, a roll moment coefficient, a pitch moment coefficient and a yaw moment coefficient, alpha is an attack angle, beta is a sideslip angle, m is aircraft mass, A X 、A Y 、A Z Acceleration components along X, Y, Z axis, ρ is air density, V is airspeed, S is wing area, I xx 、I yy 、I zz 、I xz For moment of inertia and moment of product, p, q, r are roll, yaw and pitch rates respectively,
Figure FDA0004061396780000021
derivatives of roll, yaw and pitch rates, b +.>
Figure FDA0004061396780000022
The span and average chord length are respectively;
(2) Estimating a stability derivative of the aircraft from the simulation data using a recursive least squares method by equation (3);
Figure FDA0004061396780000023
in the method, in the process of the invention,
Figure FDA0004061396780000024
the basic increment coefficient of the lifting force, the increment coefficient related to the lifting force and the attack angle, the lifting force and the deflection are respectivelyIncrement coefficient related to the aviation angle and increment coefficient related to lift and elevator;
(3) Parameter estimation is carried out by adopting a recursive least square method with forgetting factors, and the recursive least square method has the following structure:
Figure FDA0004061396780000025
Figure FDA0004061396780000026
for the parameters to be estimated, lambda E [0,1]As a forgetting factor, a variable forgetting factor is used to enhance the effect of new data when model parameters change suddenly and reduce covariance matrix saturation under steady-state conditions.
4. The method for online prediction of post-fault flight envelope based on radial basis-counter-propagating neural network according to claim 1, wherein in step S3, the structural damage aerodynamic effect modeling and the fault database construction comprise the following steps:
(1) Horizontal stabilizer damage can lead to changes in longitudinal stability, due to
Figure FDA0004061396780000031
A representation; vertical tail wing tip damage causes a change in lateral force and directional stability, consisting of +.>
Figure FDA0004061396780000032
Is represented by a change value of (a);
(2) The change value under the damage degree is obtained by the following formula;
Figure FDA0004061396780000033
in the formula, delta C, C d C is the degree of injury of the body, the dimensionless aerodynamic coefficient of the damaged body and the dimensionless of the complete body respectivelyAerodynamic coefficient of the line;
(3) Assuming an approximately linear relationship between the range of variation of each aerodynamic coefficient and the percent loss of wing tip; and according to the calculated damage degree variation scale, using linear interpolation to establish a pneumatic damage model under each damage level, and generating a fault pneumatic database comprising pneumatic data and corresponding stability derivative of the horizontal stabilizer damage and the vertical tail wing tip damage.
5. The method for online prediction of post-fault flight envelope based on radial basis-counter-propagating neural network according to claim 4, wherein in step S4, rbf-bp neural network training and real-time fault diagnosis comprises the following steps:
(1) Determining the input variable x as at each Mach
Figure FDA0004061396780000034
The output variable is the specific fault type and damage degree corresponding to the pneumatic data;
(2) Before rbf-bp neural network training is carried out, dividing a data set into a training set and a testing set, and then carrying out normalization processing on the data to normalize the data to be between [0,1 ];
(3) Initializing an rbf network, selecting t different initial clustering centers, setting iteration steps, and inputting and then selecting from samples; the closest distance between the input sample and the cluster center is obtained by the following formula:
iX k ＝min||X k -C i (m)||，i＝1,2,...,t，k＝1,2,...,n (6)
(4) Calculating and adjusting the center of hidden layer nodes:
Figure FDA0004061396780000035
when C i (m+1)＝C i (m) when the clustering process is finished, determining the width of hidden layer nodes according to a clustering center; if the two are not equal, the input sample and the center distance are re-performedCalculating the distance until the learning process is completed;
(5) For the bp neural network, setting the number n of input neurons, the number m of output neurons and the training frequency E, wherein the activation function uses log sig and purelin functions, and the number h of hidden layer neurons is obtained by the following formula:
Figure FDA0004061396780000041
where α is any integer between 1 and 10.
6. The method for online prediction of post-fault flight envelope based on radial basis-counter-propagating neural network according to claim 1, wherein in step S5, a flight envelope database is built, comprising the steps of:
(1) The plane maximum speed is obtained by the following formula:
Figure FDA0004061396780000042
T max for thrust, ρ (H) is the air density in relation to altitude, C xmin Is the resistance coefficient;
(2) The plane fly minimum speed is obtained by:
Figure FDA0004061396780000043
g is the weight of the aircraft, C zmax Is the lift coefficient;
(3) The maximum altitude rise limit for an aircraft to maintain a constant velocity straight line flat flight is obtained by:
Figure FDA0004061396780000044
v z the maximum rise rate of the aircraft is shown, and K is the lift-drag ratio;
(4) The step of establishing a flight envelope database is as follows:
constraint conditions:
Figure FDA0004061396780000045
flat flight conditions:
Figure FDA0004061396780000046
/>
basic trim control parameters:
Figure FDA0004061396780000047
optimizing an objective function:
Figure FDA0004061396780000048
wherein A-I is an optimized parameter, dr is resistance, L is lift force, T is thrust along the machine body, phi, psi and theta are rolling angle, yaw angle and pitch angle respectively,
Figure FDA0004061396780000051
the roll angle and yaw and pitch angle derivatives, respectively, p, q, r are roll, yaw and pitch rates, respectively,/->
Figure FDA0004061396780000052
Derivatives of roll, yaw and pitch rates, V is the flight speed, h is the altitude, V 0 And h 0 For a speed and height at a horizontal tail offset angle of 0 DEG, beta is the sideslip angle, +.>
Figure FDA0004061396780000053
Is the derivative of sideslip angle, alpha is the angle of attack, +.>
Figure FDA0004061396780000056
Is the derivative of angle of attack, m and g are the mass of the aircraft and the acceleration of gravity, delta T 、δ e 、δ r Is the rudder deflection angle caused by thrust, the aileron deflection angle and the rudder deflection angle caused by resistance,
Figure FDA0004061396780000054
is a high derivative>
Figure FDA0004061396780000055
Is the speed derivative;
when the aircraft fails, balancing the aircraft at each designated altitude and speed, a set of different drag and lift coefficients are obtained; estimating the maximum and minimum speeds and the rise limit at this time according to the formulas (1) - (3); based on a series of such estimations, a flight envelope for such fault conditions is obtained, which in turn results in a database of flight envelopes for various fault conditions.
CN202310062283.1A 2023-01-17 2023-01-17 Post-fault flight envelope online prediction method based on radial basis-counter propagation neural network Pending CN116127842A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310062283.1A CN116127842A (en) 2023-01-17 2023-01-17 Post-fault flight envelope online prediction method based on radial basis-counter propagation neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310062283.1A CN116127842A (en) 2023-01-17 2023-01-17 Post-fault flight envelope online prediction method based on radial basis-counter propagation neural network

Publications (1)

Publication Number Publication Date
CN116127842A true CN116127842A (en) 2023-05-16

Family

ID=86295211

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310062283.1A Pending CN116127842A (en) 2023-01-17 2023-01-17 Post-fault flight envelope online prediction method based on radial basis-counter propagation neural network

Country Status (1)

Country Link
CN (1) CN116127842A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116552804A (en) * 2023-07-10 2023-08-08 四川腾盾科技有限公司 Test flight planning method for verifying practical limit rise index of low-speed unmanned aerial vehicle
CN116702030A (en) * 2023-05-31 2023-09-05 浙江大学 Blast furnace state monitoring method and device based on sensor reliability analysis

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116702030A (en) * 2023-05-31 2023-09-05 浙江大学 Blast furnace state monitoring method and device based on sensor reliability analysis
CN116702030B (en) * 2023-05-31 2024-01-23 浙江大学 Blast furnace state monitoring method and device based on sensor reliability analysis
CN116552804A (en) * 2023-07-10 2023-08-08 四川腾盾科技有限公司 Test flight planning method for verifying practical limit rise index of low-speed unmanned aerial vehicle

Similar Documents

Publication Publication Date Title
CN108594788B (en) Airplane actuator fault detection and diagnosis method based on deep random forest algorithm
CN107608335B (en) Data driving method for fault detection and fault separation of unmanned aerial vehicle flight control system
CN116127842A (en) Post-fault flight envelope online prediction method based on radial basis-counter propagation neural network
EP2570880B1 (en) A method for real-time model based structural anomaly detection
CN104773304B (en) Load estimation system for aerodynamic structures
US9221555B2 (en) Structural health management with active control using integrated elasticity measurement
US20180335018A1 (en) Turbine Loads Determination and Condition Monitoring
CN108170127A (en) A kind of fault detection method of UAV Flight Control System
CN112884051B (en) Data-driven multi-component online complex fault diagnosis method for lightweight unmanned aerial vehicle
CN113128035B (en) Civil aircraft flight control sensor signal reconstruction fault-tolerant control method
CN113568423B (en) Intelligent fault-tolerant control method of four-rotor unmanned aerial vehicle considering motor faults
CN115320886A (en) Real-time monitoring method and system for aircraft control surface fault
CN112046761B (en) Airplane icing on-line detection method based on statistical test and filtering
Caliskan et al. Aircraft icing detection, identification, and reconfigurable control based on Kalman filtering and neural networks
Aykan et al. Kalman filter and neural network‐based icing identification applied to A340 aircraft dynamics
Aykan et al. Aircraft icing detection, identification and reconfigurable control based on Kalman filtering and neural networks
Miao et al. Dynamics and adaptive fault-tolerant flight control under structure damage of horizontal stabilizer
WO2023101757A1 (en) Multicopter online rotor fault diagnosis system
Van Eykeren et al. Actuator fault detection by aerodynamic model identification
Sahwee et al. Wind tunnel evaluation for control transition from elevator to stabilator of small UAV
Dutta et al. Time-series assisted machine learning framework for probabilistic rotor fault diagnosis on multicopters under varying operating conditions
Xiaoxiong et al. A method of fault diagnosis and flight envelop assessment for flight control systems
Goswami et al. Incipient Stall Detection Using the Real-Time Aerodynamic Load Estimation from a Distributed FADS Sensor Network
Aykan et al. F16 icing identification based on neural networks
Aykan et al. Kalman Filter and Neural Network Based Icing Detection Applied to A340 Aircraft Dynamics

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