Wind speed operator network construction method suitable for motion-induced unsteady aerodynamic force
By constructing a wind speed operator network (VONet) and combining motion net and velocity net, the accuracy and efficiency problems of motion-induced unsteady aerodynamic force prediction were solved, achieving high-precision and fast prediction, revealing the physical laws in wind speed, and predicting stall flutter boundaries.
Patent Information
- Application Number
- CN202411157433.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-22
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-08-22
AI Technical Summary
Existing technologies struggle to predict motion-induced unsteady aerodynamic forces efficiently, quickly, and accurately. Traditional neural network architectures have weak generalization capabilities and cannot reveal physical laws.
The wind speed operator network construction method (VONet) is adopted. By combining motion net and velocity net neural networks, the concept of wind speed operator is introduced. Combined with CFD simulation or wind tunnel test data, a driving model is constructed and the aerodynamic work coefficient is solved by the complex Simpson integral method. Energy map is drawn to predict flutter instability boundary.
It significantly shortens the calculation time, improves the prediction accuracy, has clear physical meaning, can reveal the physical mechanism in wind speed, and can predict the boundary of stall flutter, thus expanding the application scenarios.
Smart Images

Figure CN119047370A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of aircraft aeroelasticity, and in particular is a method for constructing a wind speed operator network suitable for motion-induced unsteady aerodynamic forces. Background Art
[0002] Motion-induced unsteady aerodynamics is an important research branch in fluid mechanics.
[0003] Motion-induced unsteady aerodynamic forces depend not only on the current flow state and boundary conditions, but also on the flow history in the previous period. In addition to static nonlinear factors such as spatial discontinuity and flow separation, motion-induced unsteady aerodynamic forces are often accompanied by dynamic nonlinear effects in time and are coupled with the structural system to form complex aeroelastic phenomena. Due to its complexity and importance, this problem has received widespread attention in engineering applications in industries such as helicopters, turbomachinery, flapping wing dynamics, and wind turbines.
[0004] Since motion-induced unsteady aerodynamics often produce strong nonlinear and time-lag effects, it is difficult for traditional engineering algorithms to meet accuracy requirements. High-precision numerical simulation methods require a lot of computing power and cost. Therefore, the industry is in urgent need of a high-precision rapid prediction method.
[0005] In recent years, the reduced-order model based on neural network has provided a new idea for this problem. After wind tunnel experiments and numerical simulation data training, the neural network model can make high-precision and rapid predictions. However, due to the lack of prior knowledge and the integration of physical mechanisms, the generalization of traditional neural network architecture is weak and cannot reveal the physical laws contained in the flow.
[0006] Incorporating prior knowledge to enhance the generalization of the model and further explore hidden physical laws is a feasible solution. Summary of the invention
[0007] In view of the existing problems of motion-induced unsteady aerodynamics, the present invention proposes a method for constructing a wind speed operator network (VONet, Velocity Operation Net) suitable for motion-induced unsteady aerodynamics; by introducing the concept of wind speed operator, the traditional single network structure is changed into a combined network structure of motion net and velocity net by using dot multiplication, fully considering the physical effects of incoming wind speed and reduced wind speed in flow phenomena, which is equivalent to using wind speed as a function mapping operator. Compared with the traditional network architecture, the present invention has the advantages of strong interpretability, strong nonlinear capture ability and high generalization accuracy.
[0008] The wind speed operator network construction method applicable to motion-induced unsteady aerodynamic forces comprises the following specific steps:
[0009] Step 1: for the target airfoil or wing, given motion training signals at different reduced wind speeds, CFD simulation or wind tunnel test is performed to obtain the unsteady aerodynamic force corresponding to each motion signal.
[0010] Step 2: Build a driving model for unsteady aerodynamic data based on the VONet architecture;
[0011] The VONet architecture consists of two neural networks: motion net and velocity net:
[0012] The input of the motion net neural network is a motion time series vector [α t-n+1 ,α t-n+2 ,…,α t ], the output is a vector D of length n: D(α t-n+1 ,α t-n+2 ,…,α t )=[d1,d2,…,d n ];
[0013] The input of velocity net neural network is the incoming wind speed U ∞ and reduced wind speed U * , the output is a vector F of length n: F(U ∞ ,U * )=[f1,f2,…,f n ] T ;
[0014] The output vectors of the two neural networks are inner-producted to obtain the final output variable of the driving model: the unsteady aerodynamic torque coefficient C under the corresponding motion and wind speed m (t): C m (t) = D·F = f1d1+f2d2+…+f n d n ;
[0015] Step 3: Use each motion signal and the corresponding unsteady aerodynamic force to train the drive model, and obtain the optimal parameters of the drive model when the relative error is stable and less than 5%.
[0016] Step 4: Using the driving model under the optimal parameters, the periodic aerodynamic work coefficient E of the dynamic stall process under different working conditions is solved by the complex Simpson integral method. f :
[0017]
[0018] in, is the pitch angular velocity during the motion, T is a motion cycle, and t is the motion moment.
[0019] Step 5: Use the aerodynamic work coefficients of different incoming wind speeds and motion amplitudes to construct a two-dimensional aerodynamic work matrix, draw an energy diagram of the incoming wind speed-motion amplitude, and output the energy diagram in combination with the aerodynamic damping coefficient to predict the flutter instability boundary.
[0020] Written as a two-dimensional aerodynamic work matrix E:
[0021]
[0022] The two-dimensional matrix is plotted as an energy diagram with the incoming wind speed as the horizontal axis and the motion amplitude as the vertical axis, so that the wind speed boundary where the aerodynamic work coefficient is 0 can be intuitively obtained from the diagram. Combined with the damping coefficient of the structure, the flutter instability boundary under different structural damping conditions can be determined.
[0023] Compared with the prior art, the present invention has the following advantages:
[0024] (1) The present invention is applicable to a method for constructing a wind speed operator network for motion-induced unsteady aerodynamics, which greatly shortens the computational time for motion-induced unsteady aerodynamics prediction and improves prediction accuracy. The network architecture only needs to perform CFD calculations at a few wind speeds to complete training, which greatly reduces computational time and cost, and can perform fast and accurate predictions at any wind speed and frequency within the training domain.
[0025] (2) The present invention is applicable to the wind speed operator network construction method for motion-induced unsteady aerodynamics. Compared with the traditional neural network architecture, the present invention has a clear physical meaning and can reveal the physical mechanism contained in the wind speed. Since the incoming wind speed and reduced wind speed are introduced as operators, the network architecture proposed by the present invention has a stronger generalization ability for variable wind speed and can capture the changes in the dynamic stall mechanism under the key reduced wind speed.
[0026] (3) The present invention is applicable to the method for constructing a wind speed operator network for motion-induced unsteady aerodynamic forces. Thanks to the good generalization and accuracy of the wind speed operator network architecture, the boundary of stall flutter can be predicted by combining the aerodynamic damping coefficient and the energy diagram, further expanding the application scenarios of the network architecture. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 It is a flow chart of the wind speed operator network construction method suitable for motion-induced unsteady aerodynamic force in the present invention.
[0028] Figure 2 It is a specific structural diagram of the VONet neural network used in the present invention. DETAILED DESCRIPTION
[0029] In order to make the technical solution of the present invention clearer and easier to understand, the present invention is further described in detail below with reference to the accompanying drawings.
[0030] Aiming at the problem of motion-induced unsteady aerodynamics, the present invention proposes a wind speed operator network construction method suitable for motion-induced unsteady aerodynamics, such as Figure 1 As shown, the specific steps are as follows:
[0031] Step 1: For the target airfoil or wing, given a forced motion training signal with clear physical information at different reduced wind speeds, CFD simulation or wind tunnel test is performed to obtain the unsteady aerodynamic lift or moment corresponding to each motion signal.
[0032] Step 2: Build an unsteady aerodynamic data-driven model based on the VONet architecture;
[0033] The VONet architecture consists of two neural network units: motion net and velocity net:
[0034] The motion net model selects a long short-term memory (LSTM) or gated recurrent unit (GRU) neural network to transform the motion time series vector [α t-n+1 ,α t-n+2 ,…,α t ] as input, representing the motion information from time t-n+1 to time t, and a vector D of length n as output: D(α t-n+1 ,α t-n+2 ,…,α t )=[d1,d2,,d n ];
[0035] The velocity net model uses a fully connected neural network (FCNN) to calculate the future wind speed U ∞ and reduced wind speed U * Two parameters are taken as input and a vector F of length n is output: F(U ∞ ,U * )=[f1,f2,…,f n ] T ;
[0036] The inner product of the vectors of the same length output by the two networks is used to obtain the final output variable of the VONet architecture: the unsteady aerodynamic torque coefficient C under the corresponding motion and wind speed m (t): C m (t) = D·F = f1d1+f2d2+…+f n d n ;
[0037] Inflow wind speed U ∞ and reduced wind speed U * Next, the motion time vector [αt-n+1 ,α t-n+2 ,…,α t ] to the dynamic stall torque coefficient C m The mapping relationship of (t) is: G(α)(U): <D(α t-n+1 ,α t-n+2 ,…,α t ),F(U ∞ ,U * )>→C m (t);
[0038] The VONet model is trained using optimization algorithms such as Adam. The trained model is tested using the input signal of any sinusoidal pitch motion within the training range and its corresponding aerodynamic data. The accuracy and generalization of the model are measured by the relative error of aerodynamic prediction. To prevent overfitting, the model parameters with the best generalization under different working conditions are selected and retained.
[0039] Step 3: Use each motion signal and the corresponding unsteady aerodynamic force to train the drive model, and obtain multiple sets of drive model parameters after the relative error of the training data is stable and less than 5%.
[0040] The generalization of the trained driving model is tested using the input signal of any given motion within the training wind speed and oscillation frequency range and its corresponding aerodynamic data. The accuracy of the driving model is measured by the relative error of the aerodynamic prediction, and the model parameters with the best generalization under different working conditions are selected and retained for subsequent predictions.
[0041] Step 4: Using the driving model under the optimal parameters, the periodic aerodynamic work coefficient E of the dynamic stall process under different working conditions is solved by the complex Simpson integral method. f :
[0042] Firstly, the driving model is used to obtain the high-precision prediction results of the dynamic stall aerodynamic force with variable wind speed and amplitude. The periodic aerodynamic work coefficient E of the dynamic stall process under different working conditions is solved by the complex Simpson integral method. f :
[0043] The formula is as follows:
[0044]
[0045] in, is the pitch angular velocity during the motion, T is a motion cycle, and t is the motion moment.
[0046] Step 5: Use the aerodynamic work coefficients of different incoming wind speeds and motion amplitudes to construct a two-dimensional aerodynamic work matrix, draw an energy diagram of the incoming wind speed-motion amplitude, and output the energy diagram in combination with the aerodynamic damping coefficient to predict the wind speed at the flutter instability boundary.
[0047] The variation of aerodynamic work coefficient with reduced wind speed is analyzed to explore the changes in the physical mechanism of aerodynamic work under certain key reduced wind speeds, such as the transformation of aerodynamic work characteristics from linear to nonlinear, and from damping to excitation.
[0048] The VONet model is used to calculate different incoming wind speeds (U1-U p ) and the amplitude of motion (A1-A q ), revealing the physical mechanism of dynamic stall under different wind speeds, and written as a two-dimensional aerodynamic work matrix E:
[0049]
[0050] The two-dimensional matrix is plotted as an energy diagram with the incoming wind speed as the horizontal axis and the motion amplitude as the vertical axis, so that the wind speed boundary where the aerodynamic work coefficient is 0 can be intuitively obtained from the diagram. Combined with the damping coefficient of the structure, the flutter instability boundary under different structural damping conditions can be determined.
[0051] Example:
[0052] In this embodiment, step (1) designs a training signal for the VONet model and solves the corresponding unsteady aerodynamic force.
[0053] The training signal input to the VONet model consists of the swept frequency and swept amplitude motion signal of the airfoil or wing at different reduced wind speeds, and the output is the unsteady aerodynamic lift or moment coefficient corresponding to the motion.
[0054] The input training signal should cover a sufficiently wide range of motion frequencies and amplitudes, while also taking into account different incoming wind speeds, so as to ensure that accurate information on unsteady aerodynamic forces at different wind speeds can be obtained, and that it has a certain degree of generalization for different reduced wind speeds, incoming wind speeds, and motion amplitudes. The corresponding input training signal should be a swept frequency signal or a signal form with a known determined input rule. The designed training input signal is used as a given motion to perform CFD (computational fluid dynamics) simulation or wind tunnel test of the target airfoil or wing to obtain the unsteady aerodynamic forces corresponding to the training signal.
[0055] This data combination of "incoming flow velocity-unsteady motion-unsteady aerodynamic force" will serve as training data for the data-driven model.
[0056] Step (2): build an unsteady aerodynamic data-driven model based on the VONet architecture, and train and test the model for generalization.
[0057] Aiming at the generalization goal of variable frequency and variable amplitude unsteady aerodynamic force, two training strategies are proposed: pitching motion signal with fixed oscillation amplitude swept frequency and pitching motion signal with fixed frequency swept oscillation amplitude.
[0058] First, a training strategy of frequency sweep with a given oscillation amplitude is given: determine the frequency range of the training signal [f0, f1]. Taking the stall flutter wind tunnel test results as an example, the stall flutter limit cycle oscillation frequency of this object is between 2Hz and 3Hz. Therefore, the designed training signal is in the form of a 2Hz-3Hz linear frequency sweep with a duration of 20 seconds. Select the amplitude A (40°) where a deep dynamic stall has occurred:
[0059] α(t)=Asin(2π(2+0.025t)t)
[0060] Where α is the angle of attack and t is the time. At different incoming flow velocities U = [6, 8, 10, 12] (m / s), the training signal of the above-mentioned sweeping motion is used to perform CFD calculations or wind tunnel tests to obtain the corresponding unsteady aerodynamic lift or moment coefficients.
[0061] The fixed frequency sweep oscillation amplitude training strategy gives a fixed pitch oscillation frequency and learns the aerodynamic characteristics of dynamic stall under different amplitudes by sinusoidally changing the oscillation amplitude:
[0062] α(t)=40sin(0.05πt)sin(5πt)
[0063] The given signal time length is 10s. Within this time range, the oscillation amplitude increases sinusoidally from 0° to 40°, and the pitch oscillation frequency remains constant at 2.5Hz.
[0064] Take the unsteady aerodynamic training of single-degree-of-freedom airfoil pitch oscillation as an example:
[0065] What VONet learns is that under a certain incoming wind speed U ∞ and reduced wind speed U * Next, the angle of attack time series α t-n+1 ,α t-n+2 ,…,α t To the unsteady aerodynamic moment coefficient C m The mapping relationship of (t) is:
[0066] G(α)(U): <d(α t-n+1 ,α t-n+2 ,…,α t ),f(U ∞ ,U * )>→C m (t)
[0067] like Figure 2As shown in Figure 1, the VONet architecture consists of two neural network units, motion net and velocity net. For the problem of motion-induced unsteady aerodynamic modeling, a long short-term memory (LSTM) or gated recurrent unit (GRU) neural network is selected as the motion net model, and the time series vector α of angle of attack with a length of n is transformed into t-n+1 ,α t-n+2 ,…,α t As input, it represents the motion information from time t-n+1 to time t, and outputs a vector d of length n;
[0068] Select the fully connected neural network (FCNN) as the velocity net model, and the future wind speed U ∞ and reduced wind speed U * The two parameters are used as input to the velocity net, and the vector f of length n is used as output;
[0069] The output vectors d and f of the two networks are inner-producted to obtain the final output variable of VONet, which is the unsteady aerodynamic torque coefficient C under the corresponding motion and wind speed. m (t).
[0070] The VONet model is trained using optimization algorithms such as Adam. A suitable learning rate is set, and the training is considered to have converged after the training error is observed to be stable. A sinusoidal test signal is used to test the generalization ability of the reduced-order model. If the requirements are met, the training of the VONet unsteady aerodynamic data-driven model is considered to have been completed.
[0071] In essence, the network decouples the information of unsteady aerodynamic motion changes and wind speed changes. The motion net unit with a given motion time series input encodes the general characteristics of the unsteady aerodynamic force, and the velocity net unit with wind speed input encodes the effect of wind speed on the phase and amplitude of the unsteady aerodynamic force. The output vectors d and f of the two network units are encoded by inner product to obtain the dynamic stall torque coefficient C at the corresponding time point. m (t).
[0072] The trained model is tested using input signals of arbitrary sinusoidal pitch motion within the training wind speed and frequency range and its corresponding aerodynamic data. The accuracy and generalization of the model are measured by the relative error of aerodynamic prediction. To prevent overfitting, the model parameters with the best generalization under different working conditions are selected and retained.
[0073] Step (3) The VONet unsteady aerodynamic model is used to output the dynamic stall aerodynamic torque with continuously variable frequency and amplitude, explore the changes in aerodynamic characteristics under different reduced frequencies, reveal the physical mechanism of dynamic stall under different wind speeds, and introduce the aerodynamic work coefficient to predict the stall flutter boundary based on the energy diagram.
[0074] Firstly, VONet is used to obtain the high-precision prediction results of the dynamic stall aerodynamic force with variable wind speed and amplitude. The aerodynamic work coefficient E of the dynamic stall process under different working conditions is solved by the complex Simpson integral method. f The aerodynamic damping coefficient represents the periodic work of the unsteady aerodynamic torque under sinusoidal motion. The formula is as follows:
[0075]
[0076] The change process of the aerodynamic work coefficient with the reduction frequency is further analyzed to explore the changes in the dynamic stall physical mechanism under certain key reduced wind speeds, such as the change of the aerodynamic work coefficient from linear to nonlinear, from damping to excitation, etc. Using the above calculation results, the energy diagram is further used to predict the incoming wind speed corresponding to the instability boundary. When the aerodynamic work coefficient changes from negative to positive, it means that the aerodynamic torque changes from damping properties to excitation properties, thereby inducing stall flutter.
[0077] The wind speed operator network (VONet) architecture proposed in the present invention is divided into two parts: motion net and velocity net. The two networks are connected by output dot product. The framework uses LSTM or GRU neural network as motion net, and a given motion time series as its input; uses a fully connected neural network as velocity net, and uses the steady incoming wind speed and the reduced wind speed that measures the unsteady effect as its input; and uses the unsteady aerodynamic lift or torque coefficient under the corresponding motion and wind speed as the output after the final dot product. The network structure of the proposed wind speed operator network architecture is lightweight and compact, with good interpretability, strong nonlinear capture ability, and high generalization accuracy. It uses wind speed operator learning to predict motion-induced unsteady aerodynamic forces, which can explore the physical mechanism contained in wind speed changes, has a stronger generalization ability for variable wind speed, and can be further combined with energy graphs to predict fluid-solid coupling instability boundaries.
Claims
1. A wind speed operator network construction method suitable for motion-induced unsteady aerodynamics, It is characterized in that The specific steps are as follows: Step 1: for the target airfoil or wing, given motion training signals at different reduced wind speeds, CFD simulation or wind tunnel test is performed to obtain the unsteady aerodynamic force corresponding to each motion signal; Step 2: Build a driving model for unsteady aerodynamic data based on the VONet architecture; The VONet architecture consists of two neural networks: motion net and velocity net: The input of the motion net neural network is a motion time series vector [α t-n+1 ,α t-n+2 ,…,α t ], the output is a vector D of length n: D(α t-n+1 ,α t-n+2 ,…,α t )=[d 1 ,d 2 ,…,d n ]; The input of velocity net neural network is the incoming wind speed U ∞ and reduced wind speed U * , the output is a vector F of length n: F(U ∞ ,U * )=[f 1 ,f 2 ,…,f n ] T ; The output vectors of the two neural networks are inner-producted to obtain the final output variable of the driving model: the unsteady aerodynamic torque coefficient C under the corresponding motion and wind speed m (t): C m (t) = D·F = f 1 d 1 +f 2 d 2 +…+f n d n ; Step 3: Use each motion signal and the corresponding unsteady aerodynamic force to train the drive model, and obtain the optimal parameters of the drive model when the relative error is stable and less than 5%; Step 4: Using the driving model under the optimal parameters, the periodic aerodynamic work coefficient of the dynamic stall process under different working conditions is solved by the complex Simpson integral method; Step 5: Use the aerodynamic work coefficients of different incoming wind speeds and motion amplitudes to construct a two-dimensional aerodynamic work matrix, draw an energy diagram of the incoming wind speed-motion amplitude, and output the energy diagram in combination with the aerodynamic damping coefficient to predict the flutter instability boundary.
2. A method for constructing a wind speed operator network suitable for motion-induced unsteady aerodynamic forces as claimed in claim 1, It is characterized in that The periodic aerodynamic work coefficient E in step 4 f : in, is the pitch angular velocity during the motion, T is a motion cycle, and t is the motion moment.
3. A method for constructing a wind speed operator network suitable for motion-induced unsteady aerodynamic forces as claimed in claim 1, It is characterized in that The two-dimensional aerodynamic work matrix E in step 5 is: U 1 -U p For different incoming wind speeds, A 1 -A q is the corresponding motion amplitude; the two-dimensional matrix is plotted as an energy diagram with the incoming wind speed as the horizontal axis and the motion amplitude as the vertical axis, so that the wind speed boundary with the aerodynamic work coefficient of 0 can be intuitively obtained from the diagram. Combined with the damping coefficient of the structure, the flutter instability boundary under different structural damping conditions can be determined.
4. A method for constructing a wind speed operator network suitable for motion-induced unsteady aerodynamic forces as claimed in claim 1, It is characterized in that In step 2, the incoming wind speed U ∞ and reduced wind speed U * Next, the motion time vector [α t-n+1 ,α t-n+2 ,…,α t ] to the dynamic stall torque coefficient C m The mapping relationship of (t) is: G(a)(U): <D(α t-n+1 ,a t-n+2 ,…,a t ),F(U ∞ ,U * )>→C m (t)。
Citation Information
Patent Citations
Fixed-wing aircraft element flight aerodynamic modeling method considering wind interference
CN117195763A
Wind turbine blade dynamic stall aerodynamic force intelligent prediction method
CN117992761A
Generation of CFD-based structurally independent aerodynamic influence coefficient matrix
US20240068903A1
