A flight control method for a high-speed helicopter

Abstract

The invention discloses a flight control method for a high-speed helicopter, and belongs to the technical field of calculation, extrapolation or counting. Aiming at the multi-modal control of a high-speed helicopter, an incremental dynamic inverse control framework composed of an inner attitude control loop and an outer speed control loop is adopted, and a Pi‑Sigma neural network is used to compensate for the incremental dynamic inverse control error, so as to ensure the stability, rapidity and robustness of the control system.

Description

Technical Field

[0001] The present invention relates to high-speed helicopter control technology, and specifically discloses a flight control method for a high-speed helicopter, belonging to the technical field of calculation, extrapolation or counting. Background Art

[0002] The helicopter is one of the most successful bionic machines in human history. However, due to the structure of conventional helicopters, the maximum flight speed generally cannot exceed 350 m / h. Proposing a new type of high-speed helicopter suitable for changing external environments to break through the inherent limitations of conventional helicopters has become an important direction for the future development of helicopters. The successful test flights of the V-22 Osprey, Ka-92, and Eurocopter X3 mark that high-speed helicopters are gradually becoming an important armed force in various countries. Finding a control method for high-speed helicopters is of indispensable significance for improving flight safety performance.

[0003] The design method of flight control laws for specific flight states based on the small perturbation linearization model is the current common practice. After expanding the nonlinear function into a Taylor series near the selected equilibrium state or reference trajectory of the system and only retaining the linear terms, this method of linearizing the nonlinear model is called the small perturbation linearization theory. The small perturbation linearization theory is beneficial to the design of the controller by decoupling the input and output variables and reducing the order of the system at the same time. However, the small perturbation linearization theory also has some limitations, that is, the system must have sufficient stability margins to support the establishment of the linearization conditions. Then, when designing the controller in the neighborhood of the design point, in addition to paying attention to the performance itself, attention should also be paid to the relationship between the size of the neighborhood and the stability margin.

[0004] In recent years, in the command and tracking control of complex multi-modal aircraft, there have been various excellent methods for reference to solve similar problems, such as tilt-rotor aircraft, near-space aircraft, hypersonic aircraft, etc. Among these advanced control methods, the adaptive control of artificial neural networks (ANN) is an effective method for dealing with the complex disturbances and model uncertainties of multi-modal aircraft. Artificial neural networks have the ability to approximate continuous non-linear functions. Compared with simple lookup table methods, one advantage of neural networks is that it reduces the required memory and computing time. In addition, neural networks can provide interpolation between training points without additional computational effort. The Pi-Sigma neural network (PSNN) achieves fast non-linear approximation by introducing summation neurons and multiplication neurons. The PSNN has a simple structure, few hyperparameters, and high convergence efficiency. At present, PSNN has been widely used in various fields and is more suitable for solving the control problems of compound helicopters. In order to ensure the robustness against partial uncertain cross-coupling and model errors, the present invention aims to propose a flight control architecture based on incremental dynamic inversion and Pi-Sigma neural network. Summary of the Invention

[0005] The object of the present invention is to overcome the deficiencies of the above-mentioned background technology and propose a flight control method for a high-speed helicopter. This method is aimed at the multi-modal control of a high-speed helicopter and is based on an incremental dynamic inversion control framework composed of an inner attitude control loop and an outer speed control loop. The Pi-Sigma neural network is used to compensate for the incremental dynamic inversion control error to ensure the stability, rapidity, and robustness of the control system.

[0006] The present invention adopts the following technical solutions to achieve the above object:

[0007] A flight control method for a high-speed helicopter includes the following five major steps.

[0008] Step 1: Select the dynamic model of the high-speed helicopter as the research object, and obtain the desired speed tracking command of the flight system according to the flight mission of the high-speed helicopter. The desired speed tracking command of the flight system includes: the forward flight speed u c , the vertical speed v c , and the yaw speed w c , and use the desired speed tracking command of the flight system as the input quantity of the controller.

[0009] Step 2: Establish a non-affine non-linear dynamic system of the high-speed helicopter, including a speed loop sub-control system and an attitude loop sub-control system.

[0010] Step 3: The high-speed helicopter has three flight modes: high-speed mode, low-speed mode, and a transition mode between high and low speeds. For the multimodal control of the high-speed helicopter, a speed-loop incremental dynamic inversion control method is designed. By means of state transformation, the dynamic characteristics of the nonlinear system are transformed into the state characteristics of a linear system, and the given nonlinear system is compensated into a pseudo-linear control signal V L , and the speed virtual control command signal U 1 (t) is obtained. And the desired control commands φ c , θ c , ψ c of the attitude loop, as well as the ducted fan thrust vector T and the rotor collective pitch are obtained through control allocation. Similarly, an attitude-loop incremental dynamic inversion control method is designed to obtain the angular velocity virtual control command signal U 2 (t), and the actual control signals of the control surfaces are calculated through control allocation. According to the obtained actual control signals of the control surfaces, the control functions g 1 (.), g 2 (.) of the non-affine nonlinear dynamic system are updated, and then x 1 , x 2 are updated.

[0011] Step 4: Under the action of the inaccuracy of the high-speed helicopter model and various disturbances, the incremental dynamic inversion controller has a certain error. A Pi-Sigma neural network is designed to compensate for the incremental dynamic inversion model error. The input of the neural network realizes the fast approximation of the neural network by introducing summing neurons and multiplying neurons. The output of the Pi-Sigma neural network is V ad . The compensation value V ad1 of the neural network in the speed loop is obtained by the neural network predicting . The compensation value V ad2 of the neural network in the attitude loop is obtained by the neural network predicting .

[0012] Step 5: The speeds u, v, w, the attitude angles φ, θ, ψ, and the attitude angular velocities p, q, r of the compound rotorcraft are detected in real time dynamically, and Steps 1 to 5 are repeated.

[0013] Furthermore, in a flight control method of a high-speed helicopter, for the speed-loop incremental dynamic inversion control method designed in Step 3, the gradient of the control function of the speed-loop control subsystem in the direction of the change of the speed virtual control command is introduced, and the nonlinear part of the speed-loop control subsystem is eliminated through the inverse matrix of this gradient. A pseudo-linear control signal is introduced to convert the nonlinear speed-loop control subsystem into a linear system. The flight control law for designing the speed-loop incremental dynamic inversion control method is as follows: u 1 = u 1 0 + Δu 1 ,

[0014] Furthermore, in a flight control method for a high-speed helicopter, in the attitude loop incremental dynamic inversion control method designed in step three, the gradient of the control function of the attitude loop control subsystem in the direction of the change of the angular velocity virtual control command is introduced, and the nonlinear part of the attitude loop control subsystem is eliminated through the inverse matrix of this gradient. A pseudo-linear control signal is introduced to convert the nonlinear attitude loop control subsystem into a linear system. The flight control law for designing the attitude loop incremental dynamic inversion control method is: u 2 = u 2 0 + Δu 2 ,

[0015] Furthermore, in step four, the neural network compensation value V ad obtains the compensation value of the incremental dynamic inversion model error by predicting the input scalar through a Pi-Sigma neural network. According to the gradient descent method of the neural network, the update rate of the weight can be obtained:

[0016] The present invention adopts the above technical solutions and has the following beneficial effects:

[0017] (1) Aiming at the multi-modal maneuvers of a high-speed helicopter, the present invention regards the aerodynamic cross-coupling between control surfaces as partial uncertain disturbances, and proposes an incremental dynamic inversion framework composed of an inner attitude control loop and an outer speed control loop to decouple the over-actuated system, which can enable the high-speed helicopter to complete system command tracking within a set finite time and has good convergence.

[0018] (2) The present invention separately designs a speed outer loop and an attitude inner loop, and introduces an incremental dynamic inversion model into the control distribution ratio. The incremental dynamic inversion model accurately describes the change of the input quantity of the nonlinear dynamic system by describing the gradient change of the control function of the non-affine nonlinear dynamic system of the high-speed helicopter in the direction of the input quantity, improves the command tracking ability and anti-interference ability of the flight system, decouples each channel of the high-speed helicopter, and enhances the stability of the flight system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a block diagram of a flight control method for a high-speed helicopter according to the present invention.

[0020] Figures 2 to 4 is a time response diagram of the speed state variables of the high-speed helicopter according to the present invention.

[0021] Figures 5 to 7 It is the time response diagram of the Euler angle state variable of the high-speed helicopter of the present invention.

[0022] Figures 8 to 10 It is the time response diagram of the angular velocity state variable of the high-speed helicopter of the present invention.

[0023] Identifications in the figure: deg - degree (angle unit), t - time, s - second (time unit); m - meter (length unit), m·s -1 - meter per second (speed unit), deg·s -1 - degree per second (angular velocity unit). Detailed implementation manners

[0024] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings.

[0025] As Figure 1 shown is the control architecture of a flight control method for a high-speed helicopter, and this method includes the following five steps.

[0026] Step 1: Select the dynamic model of the high-speed helicopter as the research object, and obtain the desired speed tracking command of the flight system according to the flight mission of the high-speed helicopter. The desired speed tracking command of the flight system includes: forward flight speed u c , climb speed v c and yaw speed w c , and use the desired speed tracking command of the flight system as the input quantity of the controller.

[0027] Step 2: Establish a non-affine nonlinear dynamic system of the high-speed helicopter, which is represented by two subsystems:

[0028]

[0029]

[0030] In Equations (1) and (2), x 1 is the state vector of the non-affine nonlinear dynamic system with relatively slow change, x 2 is the state vector of the non-affine nonlinear dynamic system with relatively fast change, x 1 = [u, v, w], u, v, w represent three velocity components in the body coordinate axis system, x 2 = [p, q, r], p, q, r represent three angular velocity components in the body coordinate axis system, represents the derivative of the state vectors x 1 , x 2 of the non-affine nonlinear dynamic system, u 1 , u2 represents the input signal of the non - affine nonlinear dynamic system, f(.), g(.) represent the system functions of the non - affine nonlinear dynamic system, and f 1 (.) is the state function of the outer speed loop, and g 1 (.) is the control function of the outer speed loop, and f 2 (.) is the state function of the inner attitude loop, and g 2 (.) is the control function of the inner attitude loop.

[0031] Step 3: The high - speed helicopter has three flight modes: high - speed mode, low - speed mode, and a transition mode between high - speed and low - speed. For the multi - modal control of the high - speed helicopter, a speed - loop incremental dynamic inversion control method is designed. Through state transformation, the dynamic characteristics of the nonlinear system are transformed into the state characteristics of a linear system, and the given nonlinear system is compensated into a pseudo - linear control signal V L , and the speed virtual control command signal U 1 (t) is obtained. U 1 (t) includes: speed increment d u , d v and d w , and the desired control commands φ c , θ c , ψ c of the attitude loop, as well as the ducted - fan thrust vector T and the rotor collective pitch Similarly, an attitude - loop incremental dynamic inversion control method is designed to obtain the angular - velocity virtual control command signal U 2 (t). U 2 (t) includes: angular - velocity increments dp, dq, dr, and the actual control signals of the control surfaces are calculated through control allocation. The actual control signals of the control surfaces include: lateral cyclic pitch A 1s , longitudinal cyclic pitch B 1s , right aileron deflection angle θ wr , left aileron deflection angle θ wl , the angle θ 1 between the ducted - fan thrust vector and the XOY plane in the body - fixed coordinate system, and the angle θ 2 between the projection of the ducted - fan thrust vector on the horizontal plane of the body - fixed coordinate system and the X - axis. According to the obtained actual control signals of the control surfaces, the control functions g 1 (.), g 2 (.) of the non - affine nonlinear dynamic system are updated, and then x 1 , x 2 are updated.

[0032] The control structures of the speed loop and the attitude loop are designed respectively. The speed loop is used as the outer loop to follow the speed command x 1c, and provide the desired control command x for the attitude loop 2c , the desired control command of the attitude loop includes: pitch angle φ c , yaw angle θ c , roll angle ψ c ; As the inner loop, the attitude loop can decouple each channel of the compound rotorcraft and enhance the stability of the system.

[0033] It can be known from the design of the speed loop incremental dynamic inversion control method that the flight control law is:

[0034]

[0035] u 1 = u 1 0 +Δu 1 (3),

[0036]

[0037] In formula (3), represents the gradient of the speed outer loop control function g 1 (.) in the direction of the change of u 1 , represents the derivative of the state vector of the desired control command of the speed outer loop, v L1 is the pseudo-linear control signal that compensates the speed outer loop incremental dynamic inversion control system into a system with a linear transfer relationship and completes decoupling, is the derivative of the state vector of the relatively slow-changing non-affine nonlinear dynamic system, u 1 is the speed virtual control command, u 1 0 is the sampling signal of the speed virtual control command, K P1 and K I1 are the linear control coefficients of the speed outer loop, V ad1 is the output of the neural network, used to compensate for model errors;

[0038] It can be known from the design of the attitude loop incremental dynamic inversion control method that the flight control law is:

[0039]

[0040] u 2 = u 2 0 +Δu 2 (4),

[0041]

[0042] In formula (4), represents the attitude inner loop control function g2 (.) The gradient in the u 2 change direction, represents the derivative of the desired control command state vector of the attitude inner loop, v L2 is to compensate the attitude inner loop incremental dynamic inverse control system into a pseudo-linear control signal with a linear transfer relationship and complete decoupling, is the derivative of the relatively fast-changing state vector of the non-affine nonlinear dynamic system, u 2 is the angular velocity virtual control command, u 2 0 is the sampling signal of the angular velocity virtual control command, K P2 and K I2 are the attitude inner loop linear control coefficients, V ad2 is the output of the neural network, used to compensate for the model error.

[0043] Step 4. Under the action of the inaccuracy of the high-speed helicopter model and various disturbances, the incremental dynamic inverse controller has a certain error. Design a Pi-Sigma neural network to compensate for the incremental dynamic inverse model error. The input of the neural network realizes the fast approximation of the neural network by introducing summing neurons and multiplying neurons. The output of the Pi-Sigma neural network is V ad , and the compensation value V of the neural network in the speed loop ad1 is obtained by predicting through the neural network for , and the compensation value V of the neural network in the attitude loop ad2 is obtained by predicting through the neural network for .

[0044] Step 4.1. The Pi-Sigma neural network is a single-hidden-layer feedforward network. The output layer is a product unit. The weights from the input layer to the hidden layer are adjustable, and the weights from the hidden layer to the output layer are fixed at 1. The input-output of the Pi-Sigma neural network can be expressed as:

[0045]

[0046]

[0047] In equations (5) and (6), w i is the adjustable weight of the i-th hidden layer of the neural network, x l is the scalar input of the l-th input node, K is the number of hidden layers, N is the number of input nodes, y is the scalar output of the Pi-Sigma neural network, h i is the output of the i-th hidden layer, and σ is the non-linear transfer function from the hidden layer to the output layer. Let the ideal output of the neural network be O l(1 ≤ l ≤ N), for convenience, let t be the function variable.

[0048] Step 4.2: Define the neural network error function as the traditional squared error function:

[0049]

[0050] In Equation (7), E(w) is the squared error function, O l is the ideal output of the neural network, and y is the actual output of the neural network.

[0051] The gradient of the squared error function E(w) with respect to the weight w i (1 ≤ i ≤ K) is:

[0052]

[0053] Step 4.3: According to the gradient descent method of the neural network, the update rate of the weight can be obtained:

[0054]

[0055] In Equation (9), η is the neural network learning rate, which is used to control the learning step size.

[0056] Step Five: Real-time dynamically detect the speeds u, v, w, attitude angles φ, θ, ψ, and attitude angular velocities p, q, r of the compound rotorcraft, and repeat Steps One to Five.

[0057] The present invention conducts command tracking simulation on a high-speed helicopter. In the simulation, an INDI controller without adaptive compensation and a single-layer perceptron (SHL) neural network adaptive controller combined with INDI are established, and the simulation process is carried out in MATLAB. Figure 2 It shows that the proposed controller successfully tracks the required forward speed command, and by designing a reference model according to the performance indicators in the time domain or frequency domain, it is possible to improve the tracking performance. However, the INDI controller and the SHL adaptive controller only barely follow the command. Figures 3 to 10 It shows the convergence performance related to the attitude and speed. It can be observed that even when the flight state changes rapidly, the dynamic model is inaccurate, and there are aerodynamic coupling disturbances, the attitude, angular velocity, and speed errors driven by the proposed control scheme asymptotically converge to the origin.

Claims

1. A flight control method for a high-speed helicopter, characterized in that: A high-speed helicopter non-affine nonlinear dynamic system is established in which the system control function is related to the actual control signal of the rudder. The high-speed helicopter non-affine nonlinear dynamic system updates the speed, attitude angle and attitude angular velocity of the high-speed helicopter in real time according to the actual control signal of the rudder output by the attitude inner loop. The high-speed helicopter non-affine nonlinear dynamic system includes two subsystems. Among them, x1 is the state vector of the non-affine nonlinear dynamic system that changes relatively slowly, x2 is the state vector of the non-affine nonlinear dynamic system that changes relatively quickly, x1 = [u, v, w], u, v, w are the three velocity components in the body coordinate axis system, x2 = [p, q, r], p, q, r are the three angular velocity components in the body coordinate axis system, is the derivative of the state vector x1, x2 of the non-affine nonlinear dynamic system, u1 is the speed virtual control instruction, u2 is the angular velocity virtual control instruction, f1(.) is the state function of the speed outer loop, g1(.) is the control function of the speed outer loop, f2(.) is the state function of the attitude inner loop, g2(.) is the control function of the attitude inner loop; The speed outer loop that follows the expected speed tracking instruction of the high-speed helicopter flight system is established. The speed outer loop filters the speed, attitude angle, and attitude angular velocity of the high-speed helicopter and then performs control allocation based on the incremental dynamic inverse model. The flight control law is: u1=u1 0 +Δu1, Among them, u1 is the speed virtual control instruction, u1 0 is the sampling signal of the speed virtual control instruction, Δu1 is the increment of u1, is the gradient of the speed outer loop control function g1(.) in the direction of u1 change, is the desired control command state vector of the speed outer loop [u c ,v c ,w c ] is the first-order derivative, u c is the forward flight speed, v c is the lifting speed, w c is the yaw speed, v L1 is the first pseudo-linear control signal, u, v, w are the real-time three-axis velocity components in the body coordinate system, is the derivative of the relatively slow-changing state vector of a non-affine nonlinear dynamic system, K P1 and K I1 is the linear control coefficient of the speed outer loop, V ad1 The first pseudo-linear control signal is a compensation signal of the first pseudo-linear control signal, and the desired control instruction of the attitude inner loop is outputted. The first pseudo-linear control signal that compensates the speed outer loop control system to a linear transfer relationship is introduced into the control allocation based on the incremental dynamic inverse model, and the derivatives of each component of the desired control instruction of the speed outer loop, the speed of the high-speed helicopter, the attitude angle of the high-speed helicopter, and the attitude angular velocity of the high-speed helicopter are predicted by the Pi-Sigma neural network to obtain the compensation signal of the first pseudo-linear control signal; Establish an attitude inner loop that follows the expected control command of the attitude inner loop. The attitude inner loop filters the speed, attitude angle, and attitude angular velocity of the high-speed helicopter and then performs control allocation based on the incremental dynamic inverse model. The flight control law is: u2=u2 0 +Δu2, Among them, u2 is the angular velocity virtual control instruction, u2 0 is the sampling signal of the angular velocity virtual control command, Δu2 is the increment of u2, is the gradient of the attitude inner loop control function g2(.) in the direction of u2 change, is the attitude loop desired control command state vector [φ c ,θ c ,ψ c ], φ c is the pitch angle, θ c is the yaw angle, ψ c is the roll angle, [p c ,q c ,r c ] is [φ c ,θ c ,ψ c ], v L2 is the second pseudo linear control signal, is the derivative of the relatively fast changing state vector of a non-affine nonlinear dynamic system, K P2 and K I2 is the linear control coefficient of the attitude inner loop, V ad2 The compensation signal of the second pseudo-linear control signal is output as the actual control signal of the control surface. The second pseudo-linear control signal which compensates the attitude inner loop control system to a linear transfer relationship is introduced into the control allocation based on the incremental dynamic inverse model. The derivative of each component of the input attitude inner loop desired control instruction, the speed of the high-speed helicopter, the attitude angle of the high-speed helicopter, and the attitude angular velocity of the high-speed helicopter are predicted by the Pi-Sigma neural network to obtain the compensation signal of the second pseudo-linear control signal. The compensation signal of the first pseudo-linear control signal and the second pseudo-linear control signal are expressed as follows: Prediction, where V ad is the prediction result of the Pi-Sigma neural network for the scalar input, σ is the nonlinear transfer function from the hidden layer to the output layer, and h i is the output of the i-th hidden layer, K is the number of hidden layers, N is the number of input nodes, w i is the adjustable weight of the i-th hidden layer, x l is the scalar input to the lth input node.

2. The flight control method of a high-speed helicopter according to claim 1, characterized in that: The weights of the Pi-Sigma neural network are updated using the gradient descent method.

3. The flight control method of a high-speed helicopter according to claim 2, characterized in that: The expression for updating the weights of the Pi-Sigma neural network is: Among them, w i k+1 、w i k Adjustable for the i-th hidden layer The k+1th and kth update values ​​of the weights, η is the learning rate of the Pi-Sigma neural network, is the error between the actual output of the hidden layer and the ideal output.

Images