CN117708985A - Machine learning-based high-speed double-spin unsteady aerodynamic prediction method - Google Patents

Abstract

Provides a machine learning-based unsteady aerodynamic prediction method for high-speed dual-rotation missiles, which belongs to the fields of flight mechanics and artificial intelligence technology. It includes generating a sample data set for high-speed dual-rotation missile aerodynamic modeling; dividing the data set into a training set and a test set to determine the model input and output; and using a neural network to establish a prediction model from the high-speed dual-rotation missile state quantity to the aerodynamic force and torque received. ;Use the model for rapid prediction of high-speed dual-rotation elastomeric aerodynamic forces and moments. This invention is based on machine learning technology and uses a nonlinear autoregressive neural network, which can well learn the mapping relationship between the flight state parameters of the double-rotating projectile and the aerodynamic force and torque it receives, and realize the unsteady aerodynamic force and torque of the high-speed double-rotating projectile. Fast and accurate prediction, reducing the time-consuming of large-scale coupled simulation. When modeling, the present invention considers the influence of the flight status at the current moment and the historical moment on the aerodynamic force and torque at the current moment, and also considers the influence of the aerodynamic force and moment at the historical moment, thereby effectively improving the prediction accuracy.

Description

A method for predicting unsteady aerodynamic force of high-speed dual-rotating projectiles based on machine learning

Technical field

The invention belongs to the technical fields of flight mechanics and artificial intelligence, and specifically relates to a method for predicting the unsteady aerodynamic force of a high-speed dual-rotation missile based on machine learning.

Background technique

In the design process of high-speed twin-spinning missiles, the calculation of its performance parameters is crucial to mastering the aerodynamic performance and flight performance of the twin-spinning missiles. The CFD/RBD coupling method can accurately calculate and simulate the aerodynamic characteristics, missile body attitude and motion trajectory of the aircraft. However, due to the need to repeatedly call CFD analysis in the CFD/RBD coupling calculation, this method is extremely inefficient and has high requirements on computing conditions and resources. High standard. Establish an efficient and accurate unsteady aerodynamic model and use it to replace the CFD module in coupled simulation to quickly and accurately predict aircraft flight performance. In fact, there is a certain mapping relationship between the flight state variables (such as speed, angle of attack, etc.) and the aerodynamic force and torque it receives. Using machine learning methods, you can learn and establish this mapping relationship and realize the unsteady aerodynamic force of the dual-rotation missile. rapid prediction. Therefore, it is necessary and widely needed to develop a method for predicting unsteady aerodynamic forces of high-speed dual-rotating missiles.

Contents of the invention

The technical problem solved by this invention is to provide a method for predicting the unsteady aerodynamic force of high-speed dual-rotating missiles based on machine learning. The purpose of the invention is to reduce the time-consuming problem of CFD/RBD coupling calculation in the design and performance evaluation of high-speed dual-rotating missiles. , and through this method, a mapping model is established between the flight state parameters of the dual-gyro missile and the aerodynamic force/torque it receives, to achieve rapid and accurate prediction of the aerodynamic force and torque of the high-speed dual-gyro missile, and to improve the efficiency of coupling calculations.

In order to achieve the above object, the technical solution adopted by the present invention is:

A method for predicting unsteady aerodynamic force of high-speed dual-rotating projectiles based on machine learning, including the following steps:

Step 1): Generate a high-speed dual-rotation projectile unsteady aerodynamic modeling data set;

Step 2): Determine the input and output parameters of the model. Since the aerodynamic force of the twin-spinning projectile has significant unsteady characteristics, the input parameters include the flight state quantities at the current moment and the previous m moments. The aerodynamic force/torque, the output is the aerodynamic force/torque exerted by the double spinning bomb at the current moment;

Step 3): Divide the data set in step 1) into a training set and a test set, where the training set is used to train the model, and the test set is used to test the prediction accuracy of the model;

Step 4): Use the NARX neural network to establish a prediction model from the input parameters in step 2) to the output parameters, and complete the construction of an aerodynamic model for rapid prediction of unsteady aerodynamic forces of high-speed dual-rotation missiles;

Step 5): Use the aerodynamic model in step 4) to quickly predict the unsteady aerodynamic force and torque of the high-speed double-rotating projectile.

In the above step 1), the specific steps are as follows:

Step 1-1): Select a certain double-rotating bomb configuration and generate the aerodynamic calculation grid of the double-rotating bomb;

Step 1-2): Determine the initial state parameters of the double-rotating projectile, including initial velocity, angular velocity, angle of attack, sideslip angle and attitude angle;

Step 1-3): Use the aerodynamic calculation grid in step 1-1) to perform computational fluid dynamics CFD/rigid body dynamics RBD coupling simulation on the initial state in step 1-2) to obtain the unsteady aerodynamic force of the double-rotating projectile Modeling dataset.

In the above step 1-1), the configuration of the double-rotating bomb is a 155mm fixed-wing canard layout double-rotating missile; in the step 1-1), the aerodynamic calculation grid is a non-structural hybrid grid, and the entire set of grids has a total of Contains 60,000 surface grids and 6,613,683 individual grids; the boundary layer grid uses structural grids, the first layer height is 4e-7m, a total of 50 layers, and the growth rate is 1.17; other areas are unstructured grids.

In the above step 1-2), the initial state parameters of the dual-rotation bomb are as follows: Mach number is 1.08, front body roll angular velocity is 0, rear body roll angular velocity is 1570.7rad/s, pitch angular velocity and yaw are both 0, The pitch angle is 31.98°, the yaw angle is 0, the roll angles of the front and rear bodies are both 0, the angle of attack is 0°, and the sideslip angle is 0°.

In the above steps 1-3), the Roe format is used for spatial discretization in the CFD calculation, and the turbulence model is the S-A model; the seven-degree-of-freedom RBD equation is solved using the fourth-order variable step size Adams prediction correction method; for high-speed dual-rotation missile flight CFD/RBD coupling simulation is performed within 0-7 seconds, the simulation time step is 0.1ms, and a total of 70,000 samples are obtained.

In the above step 2), the input parameters include: the flight status quantity at the current moment and the previous 4 moments, the aerodynamic force/torque exerted by the dual-rotating bomb at the previous 4 moments, and the output is the aerodynamic force/torque exerted by the dual-rotating bomb at the current moment. moment.

Among them, the flight state quantities include: the three-axis speed, three-axis angular velocity, attitude angle, angle of attack and sideslip angle of the front and rear bodies respectively; the aerodynamic force/moment suffered by the double-rotating bomb includes the axial direction of the front and rear bodies. force, lateral force, normal force, roll moment, yaw moment and pitch moment.

In the above step 3), the samples in step 1) are divided into a training set and a test set, of which the training set accounts for 75% and the test set accounts for 25%.

In the above step 4), the NARX neural network is used, as described in step 2), and the flight state quantities at the current moment and the previous four moments, and the aerodynamic force and torque experienced by the double-rotating bomb at the previous four moments are used as inputs to the network. The aerodynamic force/torque experienced by the double-rotating projectile at the current moment is used as the output of the NARX network. An aerodynamic model for rapid prediction of the unsteady aerodynamic force of the high-speed double-rotating projectile is established and trained.

In the above step 5), the specific steps are as follows:

Step 5-1): Select input parameters from the test set data in step 3);

Step 5-2): Use the input parameters in step 5-1) as the input of the aerodynamic model trained in step 4), and predict the aerodynamic force and torque experienced by the dual-rotation projectile at the current moment.

Advantages of the present invention compared with existing technology:

1. This solution is based on machine learning technology and uses a nonlinear autoregressive neural network with external source input. It can establish a mapping model of the flight state parameters of the dual-rotary missile and the aerodynamic force/torque it receives, so that it can learn well The mapping relationship between the flight state parameters of the twin-spinning projectile and the aerodynamic force and torque it receives enables fast and accurate prediction of the unsteady aerodynamic force and torque of the high-speed twin-spinning projectile, improves the efficiency of coupling calculations, and reduces the cost of high-speed twin-spinning projectile design. And the time-consuming problem of CFD/RBD coupling simulation calculation in performance evaluation;

2. When establishing the aerodynamic prediction model for the unsteady aerodynamic prediction method of the double-rotating missile proposed in this plan, it not only considers the influence of the flight status at the current moment and the historical moment on the aerodynamic force and torque at the current moment, but also considers the historical moment. The influence of aerodynamic force and torque effectively improves the prediction accuracy;

3. In this plan, the mapping relationship between the state quantity of the double spin bomb and the aerodynamic force and torque it receives is established. Therefore, the established prediction model can replace the CFD module and the RBD module for coupled simulation.

Description of the drawings

Figure 1 is a flow chart of the modeling and prediction of the high-speed dual-rotation projectile unsteady aerodynamic force prediction method based on machine learning in the present invention;

Figure 2 is a schematic diagram of the input and output of the aerodynamic prediction model in the present invention;

Figure 3 is a schematic diagram of the geometric shape of the double-rotating bomb used in the present invention;

Figure 4 is a schematic diagram of the double-rotation aerodynamic calculation grid used in the present invention, (a) is a surface grid, (b) is an overall schematic diagram of the sliding grid;

Figure 5 is a partial prediction result of the aerodynamic force and torque of a high-speed dual-rotation projectile on the test set of the present invention. (a) is a comparison of the real and predicted results of the rear body normal force, and (b) is the real sum of the rear body lateral force. Comparison of predicted results, (c) is the comparison between the actual and predicted results of the rear body pitching moment, (d) is the comparison of the actual and predicted results of the rear body yaw moment.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Please refer to Figures 1-5 for a detailed description of the embodiments of the present invention.

A method for predicting unsteady aerodynamic force of high-speed dual-rotating projectiles based on machine learning, as shown in Figure 1, including the following steps:

Step 1): Generate a high-speed dual-rotation projectile unsteady aerodynamic modeling data set; the specific steps are as follows:

Step 1-1): Select a certain double-rotating bomb configuration and generate the aerodynamic calculation grid of the double-rotating bomb;

The configuration of the double-rotating missile selected in this embodiment is a 155mm fixed-wing canard layout double-rotating missile. The geometric shape diagram is shown in Figure 3; the aerodynamic calculation grid is a non-structural hybrid grid, as shown in Figure 4. The entire set of grids contains a total of 60,000 surface grids and 6,613,683 individual grids; the boundary layer grid uses a structural grid, with a first layer height of 4e-7m, a total of 50 layers, and a growth rate of 1.17; other areas are non-structural grids. grid.

Step 1-2): Determine the initial state parameters of the double-rotating projectile, including initial velocity, angular velocity, angle of attack, sideslip angle and attitude angle;

The initial state parameters of the dual-rotation bomb are as follows: Mach number is 1.08, front body roll angular velocity is 0, rear body roll angular velocity is 1570.7rad/s, pitch angular velocity and yaw are both 0, pitch angle is 31.98°, yaw The angle is 0, the roll angles of the front and rear bodies are both 0, the angle of attack is 0°, and the sideslip angle is 0°.

Step 1-3): Use the aerodynamic calculation grid in step 1-1) to perform computational fluid dynamics CFD/rigid body dynamics RBD coupling simulation on the initial state in step 1-2) to obtain the unsteady aerodynamic force of the double-rotating projectile Modeling dataset.

Specifically, the CFD/RBD coupling simulation was performed on the high-speed dual-rotary missile flight within 0-7 seconds using the initial state parameters in step 1-2). The simulation time step was 0.1ms, and a total of 70,000 samples were obtained; the CFD The Roe format is used for spatial discretization during calculation, and the turbulence model is the S-A model; the fourth-order variable step size Adams prediction correction method is used to solve the seven-degree-of-freedom RBD equation.

Step 2): Determine the input and output parameters of the model. Since the aerodynamic force of the twin-spinning projectile has significant unsteady characteristics, the input parameters include the flight state quantities at the current moment and the previous m moments. The aerodynamic force/torque of , the output is the aerodynamic force/torque exerted by the double-rotating bomb at the current moment, as shown in Figure 2;

The input parameters include: the flight state quantity at the current moment and the previous 4 moments, the aerodynamic force/torque experienced by the dual-rotating bomb at the previous 4 moments, and the output is the aerodynamic force/torque experienced by the dual-rotating bomb at the current moment.

Among them, the flight state quantities include: the three-axis velocity, three-axis angular velocity, attitude angle, angle of attack and sideslip angle of the front and rear bodies respectively; the aerodynamic force/moment experienced by the double-rotating bomb includes the axial force, lateral force, normal force, roll moment, yaw moment and pitch moment.

Step 3): Divide the data set in step 1-3) into a training set and a test set. The training set is used to train the model, and the test set is used to test the model prediction accuracy; the training set accounts for 75% and the test set accounts for 25%. .

Step 4): Use the NARX neural network to establish a prediction model from the input parameters in step 2) to the output parameters, and combine the flight state quantities at the current moment and the previous 4 moments, and the aerodynamic force experienced by the dual-spinning missile at the previous 4 moments. and torque are used as the input of the network, and the aerodynamic force/torque exerted by the double-rotating projectile at the current moment is used as the output of the NARX network. An aerodynamic model for rapid prediction of the unsteady aerodynamic force of the high-speed double-rotating projectile is established and trained, and the aerodynamic force model for the high-speed double-rotating projectile is completed. Construction of unsteady aerodynamic model for rapid prediction of elastomer force/torque.

Step 5): Use the aerodynamic model in step 4) to quickly predict the unsteady aerodynamic force and torque of the high-speed double-rotating projectile; the specific steps are as follows:

Step 5-1): Select input parameters from the test set data in step 3);

Step 5-2): Use the input parameters in step 5-1) as the input of the aerodynamic model trained in step 4), and predict the aerodynamic force and torque experienced by the dual-rotation projectile at the current moment. Referring to Figure 5, some prediction results of the aerodynamic force and torque of the high-speed double-rotating projectile on the test set are given (taking the rear body of the double-rotating projectile as an example): (a) Comparison of the real and predicted results of the normal force of the rear body , (b) is the comparison of the real and predicted results of the rear body lateral force, (c) is the comparison of the real and predicted results of the rear body pitching moment, (d) is the comparison of the real and predicted results of the rear body yaw moment.

This technical solution is based on machine learning technology and uses a nonlinear autoregressive neural network with external source input. It can well learn the mapping relationship between the flight state parameters of the dual-rotary missile and the aerodynamic force and torque it receives, and realize high-speed Fast and accurate prediction of unsteady aerodynamic forces and moments of dual-rotating projectiles. When establishing the dual-gyro aerodynamic prediction model, the present invention not only considers the influence of the flight status at the current moment and the historical moment on the aerodynamic force and torque at the current moment, but also considers the influence of the aerodynamic force and moment at the historical moment, effectively improving the prediction Accuracy.

It is obvious to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be regarded as illustrative and non-restrictive from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is therefore intended that all claims falling within the claims All changes within the meaning and scope of equivalent elements are included in the present invention. Any reference signs in the claims shall not be construed as limiting the claim in question.

In addition, it should be understood that although this specification is described in terms of implementations, not each implementation only contains an independent technical solution. This description of the specification is only for the sake of clarity, and those skilled in the art should take the specification as a whole. , the technical solutions in each embodiment can also be appropriately combined to form other implementations that can be understood by those skilled in the art.

Claims (10)

1. A method for predicting unsteady aerodynamic force of high-speed dual-rotating projectiles based on machine learning, which is characterized by including the following steps: Step 1): Generate a high-speed dual-rotation projectile unsteady aerodynamic modeling data set; Step 2): Determine the input and output parameters of the model. Since the aerodynamic force of the twin-spinning projectile has significant unsteady characteristics, the input parameters include the flight state quantities at the current moment and the previous m moments. The aerodynamic force/torque, the output is the aerodynamic force/torque exerted by the double spinning bomb at the current moment; Step 3): Divide the data set in step 1) into a training set and a test set, where the training set is used to train the model, and the test set is used to test the prediction accuracy of the model; Step 4): Use the NARX neural network to establish a prediction model from the input parameters in step 2) to the output parameters, and complete the construction of an aerodynamic model for rapid prediction of unsteady aerodynamic forces of high-speed dual-rotation missiles; Step 5): Use the aerodynamic model in step 4) to quickly predict the unsteady aerodynamic force and torque of the high-speed double-rotating projectile. 2. A method for predicting unsteady aerodynamic force of high-speed double-rotating projectiles based on machine learning according to claim 1, characterized in that: in the above step 1), the specific steps are as follows: Step 1-1): Select a certain double-rotating bomb configuration and generate the aerodynamic calculation grid of the double-rotating bomb; Step 1-2): Determine the initial state parameters of the double-rotating projectile, including initial velocity, angular velocity, angle of attack, sideslip angle and attitude angle; Step 1-3): Use the aerodynamic calculation grid in step 1-1) to perform computational fluid dynamics CFD/rigid body dynamics RBD coupling simulation on the initial state in step 1-2) to obtain the unsteady aerodynamic force of the double-rotating projectile Modeling dataset. 3. A method for predicting unsteady aerodynamic force of high-speed dual-rotation missiles based on machine learning according to claim 2, characterized in that: in the above-mentioned step 1-1), the configuration of the dual-rotation missile is a 155mm fixed-wing canard. The aerodynamic calculation grid is a non-structural hybrid grid, and the whole set of grids contains a total of 60,000 surface grids and 6,613,683 individual grids; the boundary layer grid uses a structural grid, and the first layer height is 4e-7m, 50 floors in total, growth rate is 1.17; other areas are unstructured grids. 4. A method for predicting the unsteady aerodynamic force of a high-speed double-rotating projectile based on machine learning according to claim 2, characterized in that: in the above steps 1-2), the initial state parameters of the double-rotating projectile are as follows: the Mach number is 1.08. The roll angular speed of the front body is 0, the roll angular speed of the rear body is 1570.7rad/s, the pitch angular speed and yaw are both 0, the pitch angle is 31.98°, the yaw angle is 0, the roll angles of the front and rear bodies are all 0, the angle of attack is 0°, and the sideslip angle is 0°. 5. A method for predicting unsteady aerodynamic force of high-speed double-rotating missiles based on machine learning according to claim 2, characterized in that: in the above-mentioned steps 1-3), the spatial discretization of the CFD calculation adopts the Roe format, and the turbulence The model is an S-A model; the seven-degree-of-freedom RBD equation is solved using the fourth-order variable step Adams prediction correction method; a CFD/RBD coupling simulation is performed within 0-7 seconds of the high-speed double-rotating missile flight, and the simulation time step is 0.1ms. A total of 70,000 samples were obtained. 6. A method for predicting unsteady aerodynamic force of high-speed dual-rotating missiles based on machine learning according to claim 1, characterized in that: in the above-mentioned step 2), the input parameters include: the flight status of the current moment and the previous four moments. The output is the aerodynamic force/torque experienced by the double-rotating projectile at the current moment. 7. A method for predicting unsteady aerodynamic force of high-speed dual-rotating missiles based on machine learning according to claim 6, characterized in that: the flight state quantity includes: three-axis velocity, three-axis angular velocity, and attitude of the front and rear bodies respectively. Angle, angle of attack and sideslip angle; the aerodynamic force/torque experienced by the double-spinning projectile includes the axial force, lateral force, normal force, roll moment, yaw moment and pitching moment of the front and rear bodies respectively. 8. A method for predicting unsteady aerodynamic force of high-speed dual-rotating missiles based on machine learning according to claim 1, characterized in that in the above step 3), the training set accounts for 75% and the test set accounts for 25%. 9. A method for predicting the unsteady aerodynamic force of high-speed dual-rotating missiles based on machine learning according to claim 6, characterized in that in the above-mentioned step 4), the NARX neural network is used to calculate the current aerodynamic force as described in step 2). time and the flight state quantities of the previous 4 moments, the aerodynamic force and torque experienced by the dual-rotating bomb at the previous 4 moments are used as the input of the network, and the aerodynamic force/torque experienced by the dual-rotating bomb at the current moment are used as the output of the NARX network to establish a high-speed Aerodynamic model for rapid prediction of unsteady aerodynamic forces of double-rotating missiles and training. 10. A method for predicting unsteady aerodynamic force of high-speed dual-rotating projectiles based on machine learning according to claim 1, characterized in that: in the above-mentioned step 5), the specific steps are as follows: Step 5-1): Select input parameters from the test set data in step 3); Step 5-2): Use the input parameters in step 5-1) as the input of the aerodynamic model trained in step 4), and predict the aerodynamic force and torque experienced by the dual-rotation projectile at the current moment.