Multi-aircraft autonomous fault-tolerant propulsion control method and device

Abstract

The invention discloses a multi-aircraft autonomous fault-tolerant propulsion control method. Aiming at the technical problem that a multi-aircraft propulsion control method cannot adapt to different control surface failure conditions, the method comprises the following steps: firstly, performing online real-time compensation on an attitude angular acceleration instruction; then, based on an aircraft dynamic inverse flight control method, a failure control surface actuation instruction is solved; and then equivalent distribution of a failure control surface actuation instruction and an engine symmetric / differential thrust instruction is realized by using a series control distribution strategy, so that the equivalent symmetric / differential thrust instruction when the control surface fails is obtained, and the engine is quickly regulated and controlled to realize autonomous fault-tolerant propulsion control of the aircraft. The invention further discloses a multi-aircraft autonomous fault-tolerant propulsion control device. According to the invention, the propulsion control system can automatically adapt to the change of the dynamic characteristics of the aircraft under the condition that different control surfaces of the aircraft are damaged or failed, the symmetrical / differential thrust of the engine is autonomously distributed to generate torque to regulate and control the attitude, and the safety of the aircraft is ensured.

Description

Technical Field

[0001] This invention relates to an autonomous fault-tolerant propulsion control method for multi-engine aircraft, belonging to the field of integrated flight propulsion control technology. Background Technology

[0002] Large aircraft have repeatedly experienced serious accidents, even resulting in catastrophic crashes, due to the failure of conventional control surfaces, leading to loss of flight control and loss of life. To address this extreme situation, experienced pilots on multi-engine aircraft can manually adjust the throttle levers of each engine to apply symmetrical or differential thrust, thereby generating the necessary pitch, yaw, or roll moments. This backup flight control method, using engine thrust as a substitute for failed control surfaces, can achieve basic attitude control even when the primary flight control system fails, significantly improving the probability of a safe landing for out-of-control aircraft. It is one of the most promising emergency flight control technologies currently available.

[0003] Due to the complex flight environment, wide flight envelope, nonlinearity, and strong coupling characteristics of aircraft, the randomness and severity of failures of different control surfaces such as rudder, elevator, and ailerons are difficult to predict accurately. Current research assumes that the types of aircraft control surface failures are known and designs engine propulsion control methods to adjust attitude. While these methods achieve attitude control using symmetrical / differential engine thrust, they have poor adaptive capabilities and cannot tolerate failures of different control surfaces. In practical applications, pilots often need to rely on experience to judge and manually perform emergency flight control operations. Furthermore, this control method lacks autonomous fault tolerance and cannot autonomously respond to different flight control failures occurring randomly under different flight conditions, relying entirely on pilot experience and significantly increasing the pilot's workload. Therefore, achieving autonomous fault tolerance in multi-engine aircraft propulsion control methods remains a technical challenge in the face of emergency situations involving the failure of conventional control surfaces. Summary of the Invention

[0004] The technical problem to be solved by this invention is to overcome the shortcomings of existing multi-engine aircraft propulsion control technology in terms of poor adaptive capability and autonomous fault tolerance capability, and to provide a multi-engine aircraft autonomous fault-tolerant propulsion control method, which can enable the propulsion control system to automatically adapt to changes in the aircraft's dynamic characteristics under different control surface damage or failure conditions, and autonomously allocate the torque generated by the symmetrical / differential thrust of the engine to regulate the attitude and ensure the safety of the aircraft.

[0005] The present invention specifically adopts the following technical solutions to solve the above-mentioned technical problems: An autonomous fault-tolerant propulsion control method for multi-engine aircraft includes the following steps: S1. The command model calculates the corresponding attitude angular velocity command and reference attitude angular acceleration command based on the input flight command. S2. Based on the attitude angular velocity tracking error, a preliminary attitude angular acceleration compensation command is obtained through the PI error control method. The preliminary attitude angular acceleration compensation command is further compensated using a pre-trained neural network model. Then, the final attitude angular acceleration compensation command is used to compensate the reference attitude angular acceleration command to obtain the compensated attitude angular acceleration command. The input of the neural network model is the flight parameters and the attitude angular velocity tracking error, and the output is the attitude angular acceleration command compensation amount when the control surface fails. S3. Using the compensated attitude angular acceleration command and the current attitude angular velocity measurement value as input, the control surface operation command for controlling the attitude angular velocity is obtained through the dynamic inverse controller. S4. Use a series control distribution strategy to distribute the control surface operation commands: control surface operation commands that cannot be actuated by the control surface are converted into symmetrical / differential thrust commands and sent to the engine controller, while control surface operation commands that can be actuated by the control surface are sent directly to the control surface hydraulic actuation mechanism.

[0006] Furthermore, the aircraft dynamic matrix of the dynamic inverse controller is processed using the following method. A Control matrix B Perform online data-driven identification and updates: Based on a finite-dimensional dataset from a previous period at the current moment, a discrete state-space model is fitted using the dynamic mode decomposition method: , , ,

[0007] in express The pseudo-inverse, with subscripts 1, 2, ... m This is a finite-dimensional data index for a period of time prior to the current moment. m To identify the required data length, The dataset contains the state variables at the current moment. The dataset contains the state variables from the previous time step. The control data set from the previous moment. , , They are respectively The left singular matrix, diagonal matrix, and right singular matrix obtained by performing singular value decomposition; Based on the dimensions of state variables and control variables, Decomposition , matrix: ,in, , , for block matrix; Calculate the aircraft dynamic matrix at the current moment. A Control matrix B :

[0008] in, The sampling period for zero-order hold sampling. It is an identity matrix.

[0009] Preferably, the neural network model is an online learning model of a deep reinforcement network based on the temporal difference algorithm, and its model weights are updated online according to the current attitude angular velocity tracking error.

[0010] In one embodiment, the control surface operation commands include: an elevator deflection command controlling the pitch rate, a rudder deflection command controlling the yaw rate, and a differential aileron deflection command controlling the roll rate; the cascaded control allocation strategy is as follows: Elevator deflection command allocation: Elevator deflection command given by the dynamic inverse controller d elve After passing through the first saturation limiting module, and then... d elve Take the difference; if the difference is 0, then... d elve The signal is transmitted directly to the elevator actuation mechanism; otherwise, the difference Δ is transmitted. d elve Convert to symmetrical thrust command d F Then let d F After passing through the second saturation limiting module and d F Take the difference; if the difference is 0, then... d F Send to engine controller d F It is driven by symmetrical thrust; otherwise, the difference Δ d F It is converted into a pseudo-control protection command and sent to the engine controller, where it is compensated by the pseudo-control torque generated by the differential thrust; Rudder deflection command assignment: Rudder deflection command given by the dynamic inverse controller d drud After passing through the first saturation limiting module, and then... d drud Take the difference; if the difference is 0, then... d drud The signal is transmitted directly to the rudder actuation mechanism; otherwise, the difference Δ is transmitted. d drud Convert to differential thrust command d dF Then let d dF After passing through the second saturation limiting module and d dF Take the difference; if the difference is 0, then... d dF Send to engine controller d dF It is driven by symmetrical thrust; otherwise, the difference Δ d dF Allocation is disabled using saturation protection; the remaining instructions will not be executed. Differential aileron deflection command allocation: Differential aileron deflection command given by the dynamic inverse controller. d dail After passing through the saturation limiting module and d dail Take the difference; if the difference is 0, then... d dail The actuation value is directly transmitted to the differential aileron actuator; otherwise, the difference Δ is transmitted. d dail This is converted into the required roll angular acceleration increment, then multiplied by the yaw-based roll control gain. K ybrc Add it to the current rudder deflection command.

[0011] Based on the same inventive concept, the following technical solutions can also be obtained: An autonomous fault-tolerant propulsion control device for multi-engine aircraft, comprising: The command model is used to calculate the corresponding attitude angular velocity command and reference attitude angular acceleration command based on the input flight command. The real-time compensation module is used to obtain a preliminary attitude angular acceleration compensation command based on the attitude angular velocity tracking error through a PI error control method, and further compensate the preliminary attitude angular acceleration compensation command using a pre-trained neural network model. Then, the obtained final attitude angular acceleration compensation command is used to compensate the reference attitude angular acceleration command to obtain the compensated attitude angular acceleration command. The input of the neural network model is the flight parameters and the attitude angular velocity tracking error, and the output is the attitude angular acceleration command compensation amount when the control surface fails. The dynamic inverse controller is used to take the compensated attitude angular acceleration command and the current attitude angular velocity measurement value as inputs, and output the control surface operation command to control the attitude angular velocity. A serial control allocation strategy is used to allocate the control surface operation commands: control surface operation commands that cannot be actuated by the control surface are converted into symmetrical / differential thrust commands and sent to the engine controller, while control surface operation commands that can be actuated by the control surface are sent directly to the control surface hydraulic actuation mechanism.

[0012] Furthermore, the multi-engine aircraft autonomous fault-tolerant propulsion control device also includes: An online data-driven identification module is used to analyze the aircraft dynamic matrix of the dynamic inverse controller using the following methods. A Control matrix B Perform online data-driven identification and updates: Based on a finite-dimensional dataset from a previous period at the current moment, a discrete state-space model is fitted using the dynamic mode decomposition method: , , ,

[0013] in express The pseudo-inverse, with subscripts 1, 2, ... m This is a finite-dimensional data index for a period of time prior to the current moment. m To identify the required data length, The dataset contains the state variables at the current moment. The dataset contains the state variables from the previous time step. The control data set from the previous moment. , , They are respectively The left singular matrix, diagonal matrix, and right singular matrix obtained by performing singular value decomposition; Based on the dimensions of state variables and control variables, Decomposition , matrix: ,in, , , for block matrix; Calculate the aircraft dynamic matrix at the current moment. A Control matrix B :

[0014] in, The sampling period for zero-order hold sampling. It is an identity matrix.

[0015] Preferably, the neural network model is an online learning model of a deep reinforcement network based on the temporal difference algorithm, and its model weights are updated online according to the current attitude angular velocity tracking error.

[0016] In one embodiment, the control surface operation commands include: an elevator deflection command controlling the pitch rate, a rudder deflection command controlling the yaw rate, and a differential aileron deflection command controlling the roll rate; the cascaded control allocation strategy is as follows: Elevator deflection command allocation: Elevator deflection command given by the dynamic inverse controller d elve After passing through the first saturation limiting module, and then... d elve Take the difference; if the difference is 0, then... d elve The signal is transmitted directly to the elevator actuation mechanism; otherwise, the difference Δ is transmitted. d elve Convert to symmetrical thrust command d F Then let d F After passing through the second saturation limiting module and d F Take the difference; if the difference is 0, then... d F Send to engine controller d F It is driven by symmetrical thrust; otherwise, the difference Δ d F It is converted into a pseudo-control protection command and sent to the engine controller, where it is compensated by the pseudo-control torque generated by the differential thrust; Rudder deflection command assignment: Rudder deflection command given by the dynamic inverse controller d drud After passing through the first saturation limiting module, and then... d drud Take the difference; if the difference is 0, then... d drud The signal is transmitted directly to the rudder actuation mechanism; otherwise, the difference Δ is transmitted. d drud Convert to differential thrust command d dF Then let d dF After passing through the second saturation limiting module and d dF Take the difference; if the difference is 0, then... d dF Send to engine controller d dF It is driven by symmetrical thrust; otherwise, the difference Δ d dF Allocation is disabled using saturation protection; the remaining instructions will not be executed. Differential aileron deflection command allocation: Differential aileron deflection command given by the dynamic inverse controller. d dail After passing through the saturation limiting module and d dail Take the difference; if the difference is 0, then... d dail The actuation value is directly transmitted to the differential aileron actuator; otherwise, the difference Δ is transmitted. d dail This is converted into the required roll angular acceleration increment, then multiplied by the yaw-based roll control gain. K ybrc Add it to the current rudder deflection command.

[0017] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: This invention solves the technical problem that existing multi-engine aircraft propulsion control technology cannot adapt to different control surface failure situations by combining real-time online compensation of attitude angular acceleration and a control architecture that combines dynamic inverse controller and serial control allocation strategy. In the case of control surface failure, the system autonomously uses differential / symmetric thrust for attitude adaptive control to ensure stable flight of the aircraft without the need for fault diagnosis and identification, effectively improving the probability of safe landing of the aircraft in the case of control surface failure. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structural principle of a preferred embodiment of the autonomous fault-tolerant propulsion control device for multi-engine aircraft of the present invention; Figure 2 A schematic diagram illustrating the principle of a specific embodiment of a series control allocation strategy. Detailed Implementation

[0019] To address the technical challenge that existing multi-engine aircraft propulsion control technologies cannot adapt to different control surface failure scenarios, the present invention proposes a control architecture that combines real-time online compensation of attitude angular acceleration with a dynamic inverse controller and a series control allocation strategy. This architecture enables the system to autonomously utilize differential / symmetric thrust for attitude adaptive control in the event of control surface failure, ensuring stable flight of the aircraft without the need for fault diagnosis and identification.

[0020] The autonomous fault-tolerant propulsion control method for multi-engine aircraft proposed in this invention includes the following steps: S1. The command model calculates the corresponding attitude angular velocity command and reference attitude angular acceleration command based on the input flight command. S2. Based on the attitude angular velocity tracking error, a preliminary attitude angular acceleration compensation command is obtained through the PI error control method. The preliminary attitude angular acceleration compensation command is further compensated using a pre-trained neural network model. Then, the final attitude angular acceleration compensation command is used to compensate the reference attitude angular acceleration command to obtain the compensated attitude angular acceleration command. The input of the neural network model is the flight parameters and the attitude angular velocity tracking error, and the output is the attitude angular acceleration command compensation amount when the control surface fails. S3. Using the compensated attitude angular acceleration command and the current attitude angular velocity measurement value as input, the control surface operation command for controlling the attitude angular velocity is obtained through the dynamic inverse controller. S4. Use a series control distribution strategy to distribute the control surface operation commands: control surface operation commands that cannot be actuated by the control surface are converted into symmetrical / differential thrust commands and sent to the engine controller, while control surface operation commands that can be actuated by the control surface are sent directly to the control surface hydraulic actuation mechanism.

[0021] The multi-engine aircraft autonomous fault-tolerant propulsion control device proposed in this invention includes: The command model is used to calculate the corresponding attitude angular velocity command and reference attitude angular acceleration command based on the input flight command. The real-time compensation module is used to obtain a preliminary attitude angular acceleration compensation command based on the attitude angular velocity tracking error through a PI error control method, and further compensate the preliminary attitude angular acceleration compensation command using a pre-trained neural network model. Then, the obtained final attitude angular acceleration compensation command is used to compensate the reference attitude angular acceleration command to obtain the compensated attitude angular acceleration command. The input of the neural network model is the flight parameters and the attitude angular velocity tracking error, and the output is the attitude angular acceleration command compensation amount when the control surface fails. The dynamic inverse controller is used to take the compensated attitude angular acceleration command and the current attitude angular velocity measurement value as inputs, and output the control surface operation command to control the attitude angular velocity. A serial control allocation strategy is used to allocate the control surface operation commands: control surface operation commands that cannot be actuated by the control surface are converted into symmetrical / differential thrust commands and sent to the engine controller, while control surface operation commands that can be actuated by the control surface are sent directly to the control surface hydraulic actuation mechanism.

[0022] To facilitate public understanding, the technical solution of the present invention will be described in detail below through a preferred embodiment and in conjunction with the accompanying drawings: like Figure 1As shown, the multi-engine aircraft autonomous fault-tolerant propulsion control device in this embodiment includes: a command model, a real-time compensation module, a dynamic inverse controller, a series control allocation strategy, and an online data-driven identification module. The command module is used to calculate the corresponding attitude angular velocity command and reference attitude angular acceleration command based on the input flight command. The real-time compensation module is used to obtain a preliminary attitude angular acceleration compensation command based on the attitude angular velocity tracking error through a PI error control method, and further compensate the preliminary attitude angular acceleration compensation command using a pre-trained neural network model, and then use the obtained final attitude angular acceleration compensation command... The reference attitude angular acceleration command is compensated to obtain a compensated attitude angular acceleration command. The dynamic inverse controller takes the compensated attitude angular acceleration command and the current attitude angular velocity measurement as inputs and outputs control surface operation commands to control the attitude angular velocity. A serial control allocation strategy is used to allocate the control surface operation commands: control surface operation commands that cannot be actuated by the control surfaces are converted into symmetrical / differential thrust commands and sent to the engine controller; control surface operation commands that can be actuated by the control surfaces are directly sent to the control surface hydraulic actuator. The online data-driven identification module is used to analyze the aircraft dynamic matrix of the dynamic inverse controller. A Control matrix B Perform online data-driven identification and updates.

[0023] The following is a more detailed explanation of the working process of this device: (a) such as Figure 1 As shown, the pilot operates the instrument control panel to input flight commands such as reference heading angle and track angle into the command model. The navigation controller in the command model calculates the attitude angular velocity command based on the error between the input reference heading angle and track angle and the feedback heading angle and track angle. The command solver further calculates the reference attitude angular acceleration command based on the attitude angular velocity command.

[0024] (II) The attitude angular velocity command output by the command model is subtracted from the attitude angular velocity feedback value to obtain the attitude angular velocity tracking error, which is then input into the real-time compensation module. The PI error controller first generates an attitude angular acceleration compensation command based on the input attitude angular velocity tracking error. Then, a pre-trained neural network model is used to further compensate the initial attitude angular acceleration compensation command. Finally, the obtained final attitude angular acceleration compensation command is used to compensate the reference attitude angular acceleration command to obtain the compensated attitude angular acceleration command. The input of the neural network model is the flight parameters and the attitude angular velocity tracking error, and the output is the attitude angular acceleration command compensation amount when the control surface fails. To better adapt to different control surface failure conditions of the aircraft, the neural network model in this embodiment adopts a deep reinforcement network online learning model based on the temporal difference algorithm (TD3), and its model weights are updated online according to the attitude angular velocity tracking error at the current moment. The specific construction and online learning process of the TD3 model are as follows: Step 1: Construct a flight control simulation environment using aircraft modeling software to obtain corresponding flight state data (aircraft airspeed, altitude, angle of attack, sideslip angle, attitude angle) and attitude angle acceleration compensation datasets under different control surface failure conditions. Step 2: Use the above dataset to pre-train the TD3 learning model. The input parameters include flight parameters such as airspeed, altitude, angle of attack, sideslip angle, and attitude angular velocity tracking error. The output parameter is the attitude angular acceleration command compensation amount when the control surface fails. Step 3: During flight, set the current k Moment State s k (Flight status and attitude angular velocity tracking error) are input into the Actor network of the TD3 learning model to obtain the attitude angular acceleration compensation amount. a k ; Step 4: After the attitude angular acceleration compensation is applied to the system, construct a reward function based on the attitude angular velocity tracking error. r k ; Step 5: Obtain the attitude angular acceleration compensation amount applied to the system k +1 time state s k+1 At this time, ( s k , a k , r k , s k+1 Store it in the experience playback buffer D of the TD3 learning model; Step 6: Sample from the buffer and update the Critic network parameters online; Step 7: Periodically update the Actor network online using the gradient signal of the Critic network; Step 8: Repeat Step 3 through Step 7 until the flight is over.

[0025] (III) Compensated attitude angular acceleration command The current attitude angular velocity measurement is used as the input to the dynamic inverse controller, which outputs control surface operation commands to control the attitude angular velocity. In this embodiment, the control surface operation commands include: elevator deflection command to control the pitch rate, rudder deflection command to control the yaw rate, and differential aileron deflection command to control the roll rate.

[0026] Traditional dynamic inverse control methods typically require piecewise linearization of the flight envelope to construct locally linear motion equations. However, when control surfaces malfunction or are damaged, the aircraft's dynamic characteristics change significantly, potentially rendering the pre-established linear model inapplicable, severely impacting control performance and even jeopardizing flight safety. To address this issue, this embodiment employs an online data-driven identification module to analyze the aircraft dynamic matrix of the dynamic inverse controller. A Control matrix B Online data-driven identification and updating are performed to adaptively update the control law, effectively addressing uncertainties such as control surface malfunctions, aerodynamic parameter perturbations, and large-scale changes in flight status. The specific implementation principle of the online data-driven identification module is as follows: First, the kinematics of the aircraft can be represented as a continuous linear state-space model in the following form: (1) in, This refers to the state variable, namely the attitude angular velocity. p , q , r These are roll, pitch, and yaw angular velocities, respectively. This refers to the attitude angular acceleration; To control the quantity, , , These are the aileron deflection, elevator deflection, and rudder deflection, respectively. The controlled quantity; A For aircraft dynamic matrix, B For control matrix, It is the output matrix.

[0027] Secondly, based on the dynamic inverse algorithm, the aircraft dynamic inverse controller is expressed as follows: (2) in This is the attitude angle acceleration command. , , These are the desired roll, pitch, and yaw acceleration commands, respectively.

[0028] To obtain the control quantity described by the above dynamic inverse controller, we need to obtain... A , B For this matrix, an online data-driven identification method is used in this embodiment: First, the kinematic model of the discrete nonlinear aircraft system is described as follows: (3) subscript " k " indicates a discrete-time index.

[0029] Using the Koompan operator theory, the state variables x and control quantity u Augmented as augmented state quantity x aug : In closed-loop system control, since the control quantity is generated based on the process output or state, it is expressed as a control law that depends on the state function relationship. u = h ( x Therefore, the dynamic description of the augmented state is as follows: (4) in, It is a nonlinear functional relationship.

[0030] Secondly, according to the Koompan operator theory, the following equation holds. (5) in, For linear operators, It is a nonlinear observation function.

[0031] Next, in order to obtain A , B The matrix further describes the discrete linear state-space equations of aircraft kinematics as follows: (6) in, , These are the dynamic matrix and control matrix in the discrete state, respectively.

[0032] The above formula can be rearranged into the following form: (7) in, .

[0033] Based on a finite-dimensional dataset from a previous period at the current moment, a dynamic mode decomposition algorithm is used to fit a discrete state-space model: (8) (9) (10) (11) (12) in express The pseudo-inverse, with subscripts 1, 2, ... m This is a finite-dimensional data index for a period of time prior to the current moment. m To identify the required data length, The dataset contains the state variables at the current moment. The dataset contains the state variables from the previous time step. The control data set from the previous moment. , , They are respectively The left singular matrix, diagonal matrix, and right singular matrix obtained by performing singular value decomposition; Therefore, based on the dimensions of state variables and control variables, Decomposition , matrix: (13) final, A , B Matrix calculation is as follows (14) in, The sampling period for zero-order hold sampling. It is an identity matrix.

[0034] At each moment, based on the current measurement data... , , By updating, you can obtain the current aircraft information online in real time. A , B A matrix is ​​used to achieve adaptive dynamic inverse control of aircraft attitude.

[0035] (iv) Serial Control Allocation Strategy: Based on the output of the dynamic inverse controller, commands are allocated according to the elevator command controlling the pitch rate, the rudder deflection command controlling the yaw rate, and the differential aileron deflection command controlling the roll rate. Commands that cannot be actuated by the control surfaces are sent to the engine controller via symmetrical / differential thrust commands, while commands that can be actuated by the control surfaces are sent directly to the hydraulic control surface actuators. That is, based on the existing elevator, rudder, and aileron control surface actuation command allocation, a symmetrical / differential thrust allocation command is added in series. This controls the engine to provide corresponding symmetrical / differential thrust to generate pitch / yaw / roll moments, acting as a virtual "actuator" to replace the failed control surface actuations, thus achieving attitude control of the runaway aircraft. The serial control allocation strategy in this embodiment is as follows: Figure 2 As shown, the details are as follows: 1) Elevator command assignment: When the aircraft is operating normally, the dynamic inverse controller gives the elevator deflection command. d elve and d elve The difference Δ is obtained by subtracting the output values ​​after passing through the saturation limit module 1. d elve If the value is 0, no further instructions will be allocated. d elve The signal is transmitted directly to the elevator actuation mechanism; When the aircraft elevator is damaged or malfunctions d elve First of all d elve After subtracting the output values ​​of the saturation limit module 1, the remaining elevator commands that cannot be executed and need to be assigned are obtained as Δ. d elve Then, proceed in sequence. and The module obtains symmetrical thrust commands. d F ;then, d F and d F The difference between the output values ​​of the saturation-limited module 2 and the output values ​​of the module is used to obtain Δ. d F If Δ d F If the value is 0, then it will not be further allocated and will be sent directly to the engine controller. d F The motion is actuated by symmetrical thrust; if Δ d F If the value is not equal to 0, it means that the symmetrical thrust command exceeds the idle speed or maximum thrust limit. Therefore, the remaining part of the command, Δ... d F Will pass The module converts the command into a pseudo-control protection command and enters the pseudo-control protection module. It then uses the pseudo-control torque generated by the differential thrust to compensate for the deviation between the expected command and the actual response.

[0036] 2) Rudder command assignment: When the aircraft is operating normally, the dynamic inverse controller gives the rudder deflection command. d drud and d drud The difference between the output values ​​after passing through the saturation limit module 1 is used to obtain Δ. d drud If the value is 0, no further instructions will be allocated. d drud It is directly transmitted to the rudder actuator; When the aircraft rudder is damaged or malfunctions d drud First of all d drud The difference between the output values ​​after passing through the saturation limit module 1 is used to obtain the remaining unallocated commands Δ that the rudder cannot execute. d drud Then, Δ d drud Then pass through in sequence and The module receives differential thrust commands. d dF ;then, d dF and d dF Subtract the output values ​​after passing through the saturation-limited module 2 to obtain Δ. d dF If Δ d dF If the value is 0, then it will not be further allocated and will be sent directly to the engine controller. d dF The motion is actuated by symmetrical thrust; if Δ d dF If the value is not equal to 0, it means that the differential thrust command is limited by the current throttle position. Therefore, the remaining part of the command, Δ, is... d dF Will pass Module, enter the saturation protection module, and set the difference Δ d dF Allocation is prohibited by saturation protection; the remaining instructions will not be executed to avoid exceeding limits.

[0037] 3) Differential aileron command allocation: When the aircraft is operating normally, the dynamic inverse controller issues differential aileron deflection commands. ddail and d dail The difference between the output values ​​after passing through the saturation limiting module is used to obtain Δ. d dail If the value is 0, no further instructions will be allocated. d dail The signal is directly transmitted to the differential aileron actuator; When the aircraft's differential aileron is damaged or malfunctions d dail First of all d dail The difference between the output values ​​after passing through the saturation limit module 1 is used to obtain the remaining unassigned commands Δ that the differential ailerons cannot execute. d dail After The module is converted into the required roll angle acceleration increment, and then multiplied by the yaw-based roll control gain. K ybrc This is added to the current yaw rate control command, and then roll control is indirectly achieved through the elevator yaw command allocation strategy.

[0038] By adopting the above technical solution, when different flight control surfaces of the aircraft malfunction, the system can use differential / symmetric thrust for attitude control to ensure stable flight of the aircraft without the need for fault diagnosis and identification.

Claims

1. A method for autonomous fault-tolerant propulsion control of multi-engine aircraft, characterized in that, Includes the following steps: S1. The command model calculates the corresponding attitude angular velocity command and reference attitude angular acceleration command based on the input flight command. S2. Based on the attitude angular velocity tracking error, a preliminary attitude angular acceleration compensation command is obtained through the PI error control method. The preliminary attitude angular acceleration compensation command is further compensated using a pre-trained neural network model. Then, the final attitude angular acceleration compensation command is used to compensate the reference attitude angular acceleration command to obtain the compensated attitude angular acceleration command. The input of the neural network model is the flight parameters and the attitude angular velocity tracking error, and the output is the attitude angular acceleration command compensation amount when the control surface fails. S3. Using the compensated attitude angular acceleration command and the current attitude angular velocity measurement value as input, the control surface operation command for controlling the attitude angular velocity is obtained through the dynamic inverse controller. S4. Use a series control distribution strategy to distribute the control surface operation commands: control surface operation commands that cannot be actuated by the control surface are converted into symmetrical / differential thrust commands and sent to the engine controller, while control surface operation commands that can be actuated by the control surface are sent directly to the control surface hydraulic actuation mechanism.

2. The autonomous fault-tolerant propulsion control method for multi-engine aircraft as described in claim 1, characterized in that, The aircraft dynamic matrix of the dynamic inverse controller is processed using the following method. A Control matrix B Perform online data-driven identification and updates: Based on a finite-dimensional dataset from a previous period at the current moment, a discrete state-space model is fitted using the dynamic mode decomposition method: , , , in express The pseudo-inverse, with subscripts 1, 2, ... m This is a finite-dimensional data index for a period of time prior to the current moment. m To identify the required data length, The dataset contains the state variables at the current moment. The dataset contains the state variables from the previous time step. The control data set from the previous moment. , , They are respectively The left singular matrix, diagonal matrix, and right singular matrix obtained by performing singular value decomposition; Based on the dimensions of state variables and control variables, Decomposition , matrix: ,in, , , for block matrix; Calculate the aircraft dynamic matrix at the current moment. A Control matrix B : in, The sampling period for zero-order hold sampling. It is an identity matrix.

3. The autonomous fault-tolerant propulsion control method for multi-engine aircraft as described in claim 1, characterized in that, The neural network model is a deep reinforcement network online learning model based on the temporal difference algorithm, and its model weights are updated online according to the current attitude angular velocity tracking error.

4. The autonomous fault-tolerant propulsion control method for multi-engine aircraft as described in claim 1, characterized in that, The control surface operation commands include: elevator deflection commands to control pitch rate, rudder deflection commands to control yaw rate, and differential aileron deflection commands to control roll rate; the specific cascaded control allocation strategy is as follows: Elevator deflection command allocation: Elevator deflection command given by the dynamic inverse controller δ elve After passing through the first saturation limiting module, and then... δ elve Take the difference; if the difference is 0, then... δ elve The signal is transmitted directly to the elevator actuation mechanism; otherwise, the difference Δ is transmitted. δ elve Convert to symmetrical thrust command δ F Then let δ F After passing through the second saturation limiting module and δ F Take the difference; if the difference is 0, then... δ F Send to engine controller δ F It is driven by symmetrical thrust; otherwise, the difference Δ δ F It is converted into a pseudo-control protection command and sent to the engine controller, where it is compensated by the pseudo-control torque generated by the differential thrust; Rudder deflection command assignment: Rudder deflection command given by the dynamic inverse controller δ drud After passing through the first saturation limiting module, and then... δ drud Take the difference; if the difference is 0, then... δ drud The signal is transmitted directly to the rudder actuation mechanism; otherwise, the difference Δ is transmitted. δ drud Convert to differential thrust command δ dF Then let δ dF After passing through the second saturation limiting module and δ dF Take the difference; if the difference is 0, then... δ dF Send to engine controller δ dF It is driven by symmetrical thrust; otherwise, the difference Δ δ dF Allocation is disabled using saturation protection; the remaining instructions will not be executed. Differential aileron deflection command allocation: Differential aileron deflection command given by the dynamic inverse controller. δ dail After passing through the saturation limiting module and δ dail Take the difference; if the difference is 0, then... δ dail The actuation value is directly transmitted to the differential aileron actuator; otherwise, the difference Δ is transmitted. δ dail This is converted into the required roll angular acceleration increment, then multiplied by the yaw-based roll control gain. K ybrc Add it to the current rudder deflection command.

5. An autonomous fault-tolerant propulsion control device for multi-engine aircraft, characterized in that, include: The command model is used to calculate the corresponding attitude angular velocity command and reference attitude angular acceleration command based on the input flight command. The real-time compensation module is used to obtain a preliminary attitude angular acceleration compensation command based on the attitude angular velocity tracking error through a PI error control method, and further compensate the preliminary attitude angular acceleration compensation command using a pre-trained neural network model. Then, the obtained final attitude angular acceleration compensation command is used to compensate the reference attitude angular acceleration command to obtain the compensated attitude angular acceleration command. The input of the neural network model is the flight parameters and the attitude angular velocity tracking error, and the output is the attitude angular acceleration command compensation amount when the control surface fails. The dynamic inverse controller is used to take the compensated attitude angular acceleration command and the current attitude angular velocity measurement value as input, and output the control surface operation command to control the attitude angular velocity. A serial control allocation strategy is used to allocate the control surface operation commands: control surface operation commands that cannot be actuated by the control surface are converted into symmetrical / differential thrust commands and sent to the engine controller, while control surface operation commands that can be actuated by the control surface are sent directly to the control surface hydraulic actuation mechanism.

6. The autonomous fault-tolerant propulsion control device for multi-engine aircraft as described in claim 5, characterized in that, Also includes: An online data-driven identification module is used to analyze the aircraft dynamic matrix of the dynamic inverse controller using the following methods. A Control matrix B Perform online data-driven identification and updates: Based on a finite-dimensional dataset from a previous period at the current moment, a discrete state-space model is fitted using the dynamic mode decomposition method: , , , in express The pseudo-inverse, with subscripts 1, 2, ... m This is a finite-dimensional data index for a period of time prior to the current moment. m To identify the required data length, The dataset contains the state variables at the current moment. The dataset contains the state variables from the previous time step. The control data set from the previous moment. , , They are respectively The left singular matrix, diagonal matrix, and right singular matrix obtained by performing singular value decomposition; Based on the dimensions of state variables and control variables, Decomposition , matrix: ,in, , , for block matrix; Calculate the aircraft dynamic matrix at the current moment. A Control matrix B : in, The sampling period for zero-order hold sampling. It is an identity matrix.

7. The autonomous fault-tolerant propulsion control device for multi-engine aircraft as described in claim 5, characterized in that, The neural network model is a deep reinforcement network online learning model based on the temporal difference algorithm, and its model weights are updated online according to the current attitude angular velocity tracking error.

8. The autonomous fault-tolerant propulsion control device for multi-engine aircraft as described in claim 5, characterized in that, The control surface operation commands include: elevator deflection commands to control pitch rate, rudder deflection commands to control yaw rate, and differential aileron deflection commands to control roll rate; the specific cascaded control allocation strategy is as follows: Elevator deflection command allocation: Elevator deflection command given by the dynamic inverse controller δ elve After passing through the first saturation limiting module, and then... δ elve Take the difference; if the difference is 0, then... δ elve The signal is transmitted directly to the elevator actuation mechanism; otherwise, the difference Δ is transmitted. δ elve Convert to symmetrical thrust command δ F Then let δ F After passing through the second saturation limiting module and δ F Take the difference; if the difference is 0, then... δ F Send to engine controller δ F It is driven by symmetrical thrust; otherwise, the difference Δ δ F It is converted into a pseudo-control protection command and sent to the engine controller, where it is compensated by the pseudo-control torque generated by the differential thrust; Rudder deflection command assignment: Rudder deflection command given by the dynamic inverse controller δ drud After passing through the first saturation limiting module, and then... δ drud Take the difference; if the difference is 0, then... δ drud The signal is transmitted directly to the rudder actuation mechanism; otherwise, the difference Δ is transmitted. δ drud Convert to differential thrust command δ dF Then let δ dF After passing through the second saturation limiting module and δ dF Take the difference; if the difference is 0, then... δ dF Send to engine controller δ dF It is driven by symmetrical thrust; otherwise, the difference Δ δ dF Allocation is disabled using saturation protection; the remaining instructions will not be executed. Differential aileron deflection command allocation: Differential aileron deflection command given by the dynamic inverse controller. δ dail After passing through the saturation limiting module and δ dail Take the difference; if the difference is 0, then... δ dail The actuation value is directly transmitted to the differential aileron actuator; otherwise, the difference Δ is transmitted. δ dail This is converted into the required roll angular acceleration increment, then multiplied by the yaw-based roll control gain. K ybrc Add it to the current rudder deflection command.

Images