If you use simple linear equation classification for big data analysis and classification modeling, the work efficiency is low, and the accuracy is also poor. For this reason, the thesis uses nonlinear differential equations to carry out computer-aided unsteady aerodynamic modeling. Based on the perspective of differential equations, the big data classification technology is studied, and the classification model is established. The article constructs the differential classification mathematical model by establishing the differential equation with second-order delay and the constraint conditions of the model specification set. The article identifies and identifies linear parameters such as characteristic time constants in the aerodynamic model. Research shows that the model can accurately predict unsteady aerodynamic characteristics under different maneuvers.
- Nonlinear differential equation
- big data
- Unsteady aerodynamic force
- Wind tunnel test
The problem of unsteady aerodynamic modeling at high angles of attack is one of the fourth-generation fighter jets and other problems with post-stall maneuverability. Establishing engineering practical unsteady aerodynamic models is the basis for conducting flight dynamics analysis, flight simulation, control law design, and verification research. Earlier studies have shown that the main cause of nonlinear and unsteady aerodynamic phenomena at high angles of attack is the flow separation on the upper surface of the wing and the flow field topology. Some scholars proposed a series of linear response functions, vortex sequence, and linear state space . These studies have achieved certain results in the unsteady aerodynamic modeling of small amplitude motion. The aerodynamic load for the large amplitude test also depends on the phenomenon of the forced oscillation amplitude. The current modeling methods are mainly divided into two categories: continuing to develop the previous methods. We propose the nonlinear response function method and its simplified method, nonlinear state-space method, etc. Another type of thinking is to use intelligent control methods such as neural networks or fuzzy logic to perform nonlinear algebraic fitting modeling directly.
However, the above modeling methods all have certain shortcomings. The response function method, the vortex sequence method, and the flight dynamics equation are combined into a “differential-integral” equation group. These are difficult to perform flight dynamics analysis. Predicting the unsteady aerodynamic force of small-amplitude motion with the nonlinear state-space method based on the results of the large-amplitude oscillating force measurement is far from the actual situation. Neural network and fuzzy logic algebraic fitting model generally need to fit the maneuvering process to constant amplitude oscillation motion. This will lead to the extraction of the reduced frequency as an input parameter, and unclear physical meaning will occur.
In this paper, the dynamic system modeling ideas research and develop nonlinear differential equation modeling and identification methods . The article establishes a nonlinear unsteady aerodynamic model that can accurately predict the movement of small amplitude and large amplitude at different frequencies simultaneously. The aerodynamic model established by this method has a simple form and automatically degenerates into a dynamic derivative model at a small angle of attack.
In the high angle of attack and small amplitude forced oscillation wind tunnel test, the unsteady characteristic is mainly manifested as the time history of aerodynamic load change is closely related to the oscillation frequency. Different dynamic wind tunnel tests show that the dynamic derivative obtained from the test within the range of 20°~50° angle of attack strongly depends on the oscillation frequency. The conventional linear superimposed aerodynamic derivative model based on the real-time motion state cannot reflect this unsteady characteristic . Introducing differential equations to describe the dynamic characteristics of unsteady aerodynamic forces has become one of the main modeling methods.
According to the absolute value of the eigenvalue of the pneumatic system matrix, the response of the linear system can be divided into two groups of dynamic characteristics of high-frequency mode and low-frequency mode. Since the dynamic response process of the high-frequency mode of the pneumatic system has little effect on the body's motion, it is usually only necessary to consider its steady-state output (steady aerodynamic force). The root cause of the unsteady aerodynamic phenomenon lies in a non-negligible low-frequency mode of the pneumatic system coupled with the motion frequency. Its intuitive performance is the hysteresis response of the vortex structure adjustment to the motion state . Therefore, the total aerodynamic force can be simplified into two parts, the steady aerodynamic force, and the unsteady aerodynamic force, directly according to the different time scales of the aerodynamic load in response to the movement change. The steady aerodynamic force is directly determined by the current motion state, reflecting the rapid response to a motion in the aerodynamic load. The unsteady aerodynamic force uses differential equations to describe the dominant low-order model of its dynamic characteristics. It reflects the comprehensive influence of the hysteresis response of the internal vortex topology adjustment of the flow field caused by the motion of the airframe on the aerodynamic load. The choice of the dominant low-order mode order is closely related to the configuration of the aircraft. Usually, only the dynamic influence of the first-order dominant mode needs to be considered near a certain angle of attack to obtain satisfactory unsteady aerodynamic modeling accuracy. Taking longitudinal motion as an example, the total aerodynamic load is decomposed into two parts: steady and unsteady. The aerodynamic coefficient is
Describe the hysteretic response of unsteady aerodynamic forces to motion at small amplitudes.
The forced oscillation wind tunnel test based on small amplitude and different frequencies can accurately identify the time constant
The phase relationship with the angle of attack in the forced motion,
From equation (4) and equation (5), it can be seen that the in-phase derivative is not equal to the static derivative, and the out-of-phase derivative is not equal to the dynamic derivative. Both of them add an item containing
If the unsteady aerodynamic response approximately has first-order linear dynamic characteristics, the data
The calculated time constant τ at different center angles of attack (Figure 2). Outside the range of
In areas where the unsteady effect is significant, the time constant is far greater than zero and exhibits nonlinear characteristics as the central angle of attack changes. This reflects that different vortex systems dominate the unsteady aerodynamic load at different angles of attack. When the angle of attack is below 35°, the unsteady effect is mainly caused by the movement of the front body vortex relative to the airframe. In comparison, the unsteady characteristics of the aerodynamic load near the angle of attack of 45° are dominated by the vortex separation movement from the trailing edge . The time constant describes the comprehensive response characteristics of different vortex systems to motion. Although only the dominant first-order mode is considered, it has higher accuracy. This method can avoid the identification of multiple state variables in complex flow fields.
The unsteady model formula (2) has a clear physical meaning. Its value is the dominant mode characteristic time of the unsteady hysteresis. This value quantitatively characterizes the significant degree of unsteady hysteresis. There is no unsteady aerodynamic phenomenon when it approaches zero. At this time, the conventional dynamic derivative aerodynamic model can be applied.
When the time constant
According to the above analysis, the parameters we need to identify are reduced to
This article uses the 2nd) method. The article uses a gradient-based static optimization algorithm to obtain
Integrate the obtained
In the range of angle of attack less than 20°, there is no obvious flow separation phenomenon, and Δ
After we have identified the two curves of
We compared the predicted value of the aerodynamic model with the wind tunnel test time history data to verify the model's accuracy. When predicting aerodynamic forces, it is necessary to determine the initial value
Because the unsteady model is directly based on the small-amplitude forced oscillation test results, the model prediction results are quite consistent with the small-amplitude test data. This paper uses this model to predict the aerodynamic force generated by the large-amplitude forced oscillation harmonic motion and compares it with the experimental data. At the same time, it is compared with the prediction results of the conventional linear superposition derivative model that depends on the angle of attack and the motion frequency to illustrate the effectiveness of the unsteady model. The linear superposition derivative model is
Figure 4 compares the pitch aerodynamic moment coefficient test data (TD) of the 40° central angle of attack, 10° amplitude, and 1.0 Hz frequency of the medium amplitude forced oscillating wind tunnel with the predicted results of the dynamic derivative model (SAMR) and the unsteady model The prediction result (UAMR). The article also presents the static test results (STD) and the static aerodynamic parameter identification results (AFIR) under the assumption of vortex non-separation obtained by the identification. It can be seen from Fig. 4 that the unsteady model equation (1) predicts the experimental results well and presents a full hysteresis loop. In contrast, the results predicted by the dynamic derivative model are far from the experimental data. Not only does the hysteresis loop produced by it are slender in shape, but it also produces physically unexplainable spikes.
Figure 5 shows the comparison results of the model's large-amplitude motion test value with the model's predicted value and the 30° central angle of attack, 20° amplitude, and 1.0 Hz frequency. It can be seen from Figure 5 that the unsteady model equation (1) predicted results are very close to the experimental data during the increase of the angle of attack. However, there is a certain gap between the predicted results and the experimental data in reducing the angle of attack.
Comparing Figure 4 and Figure 5, it can be seen that the use of linear differential equations to establish an unsteady aerodynamic model better predicts the results of wind-tunnel tests with amplitudes below 10°. However, the prediction accuracy of the large-amplitude test has decreased. This shows that the model does not fully reflect the influence of motion amplitude on unsteady aerodynamic characteristics. When the aerodynamic load
Based on the linear differential equation model, this paper introduces a nonlinear correction term to construct an unsteady aerodynamic model of the nonlinear differential equation. At the same time, the paper uses the results of large-amplitude oscillating wind tunnel tests to identify the nonlinear term coefficients.
We rewrite equation (2) as
In the formula:
The longitudinal large-amplitude test shows the obvious difference in aerodynamic characteristics in different directions of the angle of attack. We use a different coefficient
The article uses the minimum sum of squares of the difference between the aerodynamic time data predicted by the model and the test data as the objective function to identify the nonlinear correction term
We use four sets of forced oscillation data with different central angles of attack, including 37° central angle of attack, 20° amplitude, and 1.2 Hz, as identification samples. The initial population of the genetic algorithm is 75, and the objective function converges after 20 generations of the genetic algorithm. Compared with the initial random search results, the fitting effect of the genetic algorithm is improved by 36%.
We substitute genetic algorithm identification result
We compare the unsteady model with the 30° central angle of attack, 20° amplitude, and 1.0 Hz forced oscillation test data that have not been used for modeling. The result is shown in Figure 7. Results with the uncorrected model UAMREq. (1) Compared to the nonlinear correction result UAMREq. (3) The predicted results are more consistent with the experimental data.
(1) The article decomposes the high angle of aerodynamic attack force into two parts: steady and fast change and unsteady and slow change. At the same time, we use nonlinear differential equations to describe the unsteady aerodynamic dynamic characteristics, which can more accurately reflect the unsteady flow characteristics.
(2) The small-amplitude forced oscillation wind tunnel test data can accurately identify the time constant τ and the aerodynamic increment coefficient in the unsteady aerodynamic model of the linear differential equation. At the same time, it can reflect the main characteristics of unsteady aerodynamics. The established aerodynamic model is simple and effective. It can more accurately describe the aerodynamic characteristics under different maneuvering conditions and can be applied to research on flight dynamics analysis, flight simulation, and control law design.