Distributed propulsion aircraft dynamics modeling method based on deep learning
Through deep learning-based methods, aerodynamic and dynamic models of distributed propulsion aircraft are established, which solves the problem that traditional methods are difficult to accurately model, and achieves more accurate dynamic modeling and more effective flight-thrust-control coupled control.
Patent Information
- Application Number
- CN202510217909.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-26
- Publication Date
- 2025-06-13
AI Technical Summary
Traditional methods have difficulty accurately modeling the dynamics of distributed propulsion aircraft, especially with limitations in power yaw and roll control.
Using a deep learning-based approach, aerodynamic model of distributed propulsion aircraft is established through wind tunnel experiment data and deep learning neural networks, and expanded it into a dynamic model. The method includes designing and making aircraft, conducting wind tunnel experiments, normalizing data processing, establishing deep learning neural network models, and optimizing model parameters through stochastic gradient descent algorithms.
It breaks through the limitations of traditional control methods, effectively solves the flight-thrust-control coupled control problem of distributed propulsion aircraft, provides a more accurate dynamic modeling basis, and has engineering practical value.
Smart Images
Figure CN120145547A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of flight control, and particularly relates to a dynamic modeling method for a distributed propulsion aircraft based on deep learning. Background Art
[0002] A distributed propulsion aircraft is a new layout and new concept aircraft, and its propulsion system consists of multiple electric propellers or small ducted fans distributed on the fuselage or wings. The adoption of the distributed propulsion method can reduce the design requirements and design difficulties of a single large engine, reduce aircraft noise, improve the anti-wind disturbance ability, improve the safety of the propulsion system, and shorten the takeoff and landing distance, and has wide applications in both commercial and military fields.
[0003] Dynamic modeling is an important basis for designing a flight controller, and the aerodynamic model is one of the key technologies of the dynamic model. Compared with traditional layout aircraft, a distributed propulsion aircraft has more power units, and its aerodynamic model is a multi-input multi-output system, and the various variables are closely coupled. It is difficult to ensure accuracy using traditional polynomial modeling or interpolation modeling methods.
[0004] Deep learning has good fitting ability for multi-input multi-output nonlinear systems and is an effective method for establishing the aerodynamic model of a distributed propulsion aircraft. At present, there is an urgent need to develop a dynamic modeling method for a distributed propulsion aircraft based on deep learning. Summary of the Invention
[0005] The technical problem to be solved by the present invention is to provide a dynamic modeling method for a distributed propulsion aircraft based on deep learning, so as to provide a model basis for the design of the controller of the distributed propulsion aircraft.
[0006] The dynamic modeling method for a distributed propulsion aircraft based on deep learning of the present invention includes the following steps:
[0007] S10. Design and manufacture a distributed propulsion aircraft;
[0008] The fuselage of the distributed propulsion aircraft is provided with symmetric wings, ailerons and a V-shaped tail on the left and right; distributed ducted fans are installed on the upper wing surfaces of both wings, and the distributed ducted fans include a plurality of ducted fans symmetrically distributed on the left and right;
[0009] S20. Conduct a wind tunnel test to obtain wind tunnel test data;
[0010] According to the preset wind tunnel test conditions, conduct a wind tunnel test to obtain wind tunnel test data;
[0011] S30. Perform normalization processing;
[0012] To make the input parameters of the deep learning neural network have a unified scale numerically, avoid excessive numerical differences between different input parameters, and improve the training effect of the deep learning neural network, the following normalization process is performed on the input parameters:
[0013]
[0014] In the formula, d p represents the original value of the input parameter, m p represents the sample mean of the input parameter, s p represents the sample variance of the input parameter, n p represents the value of the input parameter after normalization;
[0015] S40. Establish an aerodynamic model of the distributed propulsion aircraft;
[0016] Establish an aerodynamic model of the distributed propulsion aircraft using wind tunnel test data and a deep learning neural network;
[0017] Among them, the input parameters of the deep learning neural network are [T, α, V, β, δ l , δ r , δ a T , T represents the power of the ducted fan, α represents the angle of attack, V represents the speed, β represents the sideslip angle, δ l represents the deflection angle of the left V-tail rudder surface, δ r represents the deflection angle of the right V-tail rudder surface, δ a represents the aileron differential angle;
[0018] Use the stochastic gradient descent algorithm to optimize the input parameters of the deep learning neural network. The input layer contains 7 nodes, there are 4 hidden layers, and the number of nodes is 256, 1024, 1024, 256 respectively. The output layer contains 6 nodes;
[0019] The output parameters of the deep learning neural network are six-component aerodynamic data, which are the lift coefficient C L , the residual thrust coefficient C R , the residual thrust coefficient C R is the difference between the drag coefficient and the thrust coefficient, the side force coefficient C Y , the pitching moment coefficient C m , the rolling moment coefficient C l and the yaw moment coefficient C n ;
[0020] S50. Establish a dynamic model of the distributed propulsion aircraft;
[0021] The output parameters C L , C R , CY , C l , C m , C n , substitute into the following formula to obtain the dynamic model of the distributed propulsion aircraft;
[0022]
[0023] Wherein, V is the flight speed, β is the sideslip angle, α is the angle of attack, p is the roll rate, q is the pitch rate, r is the yaw rate, φ is the roll angle, θ is the pitch angle, and ψ is the yaw angle; is the rate of change of velocity, is the sideslip angular velocity, is the rate of change of the angle of attack, is the rate of change of the roll rate, is the rate of change of the pitch rate, is the rate of change of the yaw rate, is the rate of change of the roll angle, is the rate of change of the yaw angle; m is the mass, Q is the dynamic pressure, S is the wing area, b is the wingspan, c is the wing chord, and g is the acceleration due to gravity; I x is the moment of inertia about the x-axis, I y is the moment of inertia about the y-axis, I z is the moment of inertia about the z-axis, I xz is the product of inertia.
[0024] The dynamic modeling method of the distributed propulsion aircraft based on deep learning of the present invention breaks through the limitation that traditional control methods are difficult to take into account dynamic yaw and roll, solves the problem of flight-thrust-control coupling control of the distributed propulsion aircraft, and has engineering practical value. Description of the Drawings
[0025] Figure 1 is the flowchart of the dynamic modeling method of the distributed propulsion aircraft based on deep learning of the present invention;
[0026] Figure 2 is the schematic diagram of the training process of the deep learning neural network adopted by the dynamic modeling method of the distributed propulsion aircraft based on deep learning of the present invention. Detailed Embodiments
[0027] The present invention will be described in detail below with reference to the drawings and embodiments.
[0028] Embodiment: The training parameters of this embodiment are shown in Table 1.
[0029] Table 1 Training Parameters
[0030]
[0031] Such as Figure 1As shown in the figure, the method for dynamic modeling of a distributed propulsion aircraft based on deep learning in this embodiment includes the following steps:
[0032] S10. Design and manufacture a distributed propulsion aircraft;
[0033] On the fuselage of the distributed propulsion aircraft, there are symmetrically arranged wings, ailerons and V-shaped tails on the left and right; Distributed ducted fans are installed on the upper wing surfaces of both wings, and the distributed ducted fans include several ducted fans symmetrically distributed on the left and right;
[0034] S20. Conduct a wind tunnel test to obtain wind tunnel test data;
[0035] According to the pre-set wind tunnel test conditions, conduct a wind tunnel test to obtain wind tunnel test data;
[0036] S30. Perform normalization processing;
[0037] In order to make the input parameters of the deep learning neural network have a unified scale numerically, avoid excessive numerical differences between different input parameters, and improve the training effect of the deep learning neural network, the following normalization processing is performed on the input parameters:
[0038]
[0039] In the formula, d p represents the original value of the input parameter, m p represents the sample mean of the input parameter, s p represents the sample variance of the input parameter, n p represents the value of the input parameter after normalization processing;
[0040] S40. Establish an aerodynamic model of the distributed propulsion aircraft;
[0041] As Figure 2 shown in the figure, use the wind tunnel test data and the deep learning neural network to establish an aerodynamic model of the distributed propulsion aircraft;
[0042] Among them, the input parameters of the deep learning neural network are [T, α, V, β, δ l , δ r , δ a T , T represents the power of the ducted fan, α represents the angle of attack, V represents the speed, β represents the sideslip angle, δ l represents the deflection angle of the left V-tail rudder surface, δ r represents the deflection angle of the right V-tail rudder surface, δ a represents the differential angle of the aileron;
[0043] The input parameters of the deep learning neural network are optimized using the stochastic gradient descent algorithm. The input layer contains 7 nodes, and there are 4 hidden layers with 256, 1024, 1024, and 256 nodes respectively. The output layer contains 6 nodes;
[0044] The output parameters of the deep learning neural network are six-component aerodynamic data, namely the lift coefficient C L , the residual thrust coefficient C R , the residual thrust coefficient C R is the difference between the drag coefficient and the thrust coefficient, the side force coefficient C Y , the pitching moment coefficient C m , the rolling moment coefficient C l and the yaw moment coefficient C n ;
[0045] S50. Establish the dynamic model of the distributed propulsion aircraft;
[0046] Substitute the output parameters C L , C R , C Y , C l , C m , C n of the deep learning neural network into the following formula to obtain the dynamic model of the distributed propulsion aircraft;
[0047]
[0048] where V is the flight speed, β is the sideslip angle, α is the angle of attack, p is the roll rate, q is the pitch rate, r is the yaw rate, φ is the roll angle, θ is the pitch angle, and ψ is the yaw angle; is the rate of change of velocity, is the sideslip angular velocity, is the rate of change of the angle of attack, is the rate of change of the roll rate, is the rate of change of the pitch rate, is the rate of change of the yaw rate, is the rate of change of the roll angle, is the rate of change of the yaw angle; m is the mass, Q is the dynamic pressure, S is the wing area, b is the wingspan, c is the wing chord, and g is the acceleration due to gravity; I x is the moment of inertia about the x-axis, I y is the moment of inertia about the y-axis, I z is the moment of inertia about the z-axis, I xz is the product of inertia.
[0049] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and embodiments. For those skilled in the art, without departing from the principle of the present invention, all features disclosed in the present invention, or steps in all methods or processes disclosed, except for mutually exclusive features and / or steps, can be combined in any way. The present invention is not limited to specific details and the illustrated examples herein.
Claims
1. A distributed propulsion aircraft dynamics modeling method based on deep learning, characterized in that: The distributed propulsion aircraft dynamics modeling method comprises the following steps: S10. Design and manufacture distributed propulsion aircraft; The fuselage of the distributed propulsion aircraft is provided with left-right symmetrical wings, ailerons and V-shaped tail; distributed ducted fans are installed on the upper wing surfaces of the wings on both sides, and the distributed ducted fans include a plurality of left-right symmetrically distributed ducted fans; S20. Conduct wind tunnel tests and obtain wind tunnel test data; Conduct wind tunnel tests according to pre-set wind tunnel test conditions and obtain wind tunnel test data; S30. performing normalization processing; In order to make the input parameters of the deep learning neural network have a unified scale in numerical value, avoid excessive numerical differences between different input parameters, and improve the training effect of the deep learning neural network, the input parameters are normalized as follows: Where, d pp Indicates the original value of the input parameter, m pp Represents the sample mean of the input parameter, s pp Represents the sample variance of the input parameter, n pp Indicates the value of the input parameter after normalization; S40. Establish an aerodynamic model of a distributed propulsion aircraft; Using wind tunnel test data and deep learning neural networks to establish an aerodynamic model of a distributed propulsion aircraft; Among them, the input parameters of the deep learning neural network are [T,α,V,β,δ ll ,δ rr ,δ aa ] TT , T represents the power of the ducted fan, α represents the angle of attack, V represents the speed, β represents the sideslip angle, δ ll Indicates the left V-tail rudder deflection angle, δ rr Indicates the right V-tail rudder deflection angle, δ aa Indicates the aileron differential angle; The stochastic gradient descent algorithm is used to optimize the input parameters of the deep learning neural network. The input layer contains 7 nodes, there are 4 hidden layers, the number of nodes are 256, 1024, 1024, 256 respectively, and the output layer contains 6 nodes; The output parameters of the deep learning neural network are six-component aerodynamic data, namely, lift coefficient C LL , residual thrust coefficient C RR , residual thrust coefficient C RR is the difference between the drag coefficient and the thrust coefficient, the side force coefficient C YY , pitch moment coefficient C mm , rolling moment coefficient C ll and the yaw moment coefficient C nn ; S50. Establish a dynamic model of a distributed propulsion aircraft; The output parameter C of the deep learning neural network LL ,C RR ,C YY ,C ll ,C mm ,C nn , substituting into the following formula, we can get the dynamic model of the distributed propulsion aircraft; Wherein, V is the flight speed, β is the sideslip angle, α is the angle of attack, p is the roll angle rate, q is the pitch angle rate, r is the yaw angle rate, φ is the roll angle, θ is the pitch angle, and ψ is the yaw angle; is the speed change rate, is the sideslip angular velocity, is the rate of change of angle of attack, is the rate of change of the roll angle, is the pitch angle rate of change, is the rate of change of yaw angle, is the rate of change of roll angle, is the rate of change of yaw angle; m is mass, Q is dynamic pressure, S is wing area, b is wing span, c is wing chord length, g is gravitational acceleration; I xx is the moment of inertia about the x-axis, I yy is the moment of inertia about the y-axis, I zz is the moment of inertia about the z-axis, I xxzz is the product of inertia.