Launch vehicle distributed aerodynamic damping prediction method based on steady flow field characteristic modeling

Abstract

The present application belongs to the technical field of launch vehicle aeroelastic instability prediction, and particularly relates to a launch vehicle distributed aerodynamic damping prediction method based on steady flow field characteristic modeling, which comprises the following steps: obtaining the flow condition parameters, flow field parameters and structural parameters of the launch vehicle under the target working condition to be predicted; inputting the flow condition parameters, flow field parameters and structural parameters of the launch vehicle under the target working condition to be predicted into a trained multi-expert feature fusion neural network model to obtain the amplitude parameters and phase angle parameters of the unsteady aerodynamic force distribution; and inversely processing the predicted amplitude parameters and phase angle parameters of the unsteady aerodynamic force distribution to obtain the unsteady aerodynamic force distribution parameters; and determining the distributed aerodynamic damping of the launch vehicle based on the unsteady aerodynamic force distribution parameters and the structural modal shape. The present application quantifies the contribution of the unsteady flow phenomenon on the surface of the launch vehicle to the aeroelastic stability, and improves the accuracy of the evaluation.

Description

Technical Field

[0001] This invention relates to the field of aeroelastic instability prediction technology for launch vehicles, and more specifically, to a method for predicting distributed aerodynamic damping of launch vehicles based on steady flow field characteristic modeling. Background Technology

[0002] During the transonic flight phase of a launch vehicle, the rocket structure is highly susceptible to damage due to aeroelastic instability. Aerodynamic damping is one of the key parameters for measuring system stability. If the aerodynamic damping is negative, the system will be in an unstable state, which may lead to catastrophic consequences. Therefore, it is necessary to accurately calculate and predict the aerodynamic damping of the launch vehicle to avoid aeroelastic instability of the rocket body and ultimately structural damage.

[0003] Furthermore, due to the variable shape of launch vehicles, local unsteady flow phenomena such as shock waves and separated flows exist simultaneously on the rocket body, which may contribute to aeroelastic stability. However, traditional experimental or numerical simulation methods are difficult to effectively quantify the contribution of local flow phenomena to aeroelastic stability and have certain limitations. Therefore, the distributed aerodynamic damping prediction method is proposed. This method can effectively evaluate the contribution of local flow phenomena to aerodynamic damping characteristics.

[0004] Currently, commonly used aerodynamic damping prediction methods mainly rely on first obtaining the aerodynamic force distribution of a steady flow field, then adding sinusoidal forced motion training on the basis of the steady flow field aerodynamic force distribution to obtain an unsteady aerodynamic force distribution, and finally evaluating the aerodynamic damping characteristics by obtaining the local unsteady aerodynamic load distribution. However, this method is cumbersome, resulting in high time costs, low solution efficiency, and difficulty in effectively quantifying the unsteady flow phenomena on the launch vehicle surface, leading to inaccurate evaluation of the impact of complex flows and limiting the reliability of aeroelastic stability analysis. Summary of the Invention

[0005] In view of this, the present invention provides a method for predicting the distributed aerodynamic damping of launch vehicles based on steady flow field characteristic modeling, in order to solve the existing technical problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for predicting distributed aerodynamic damping of launch vehicles based on steady flow field characteristic modeling includes the following steps: Obtain the flow parameters, flow field parameters, and structural parameters of the launch vehicle under the target operating conditions to be predicted; The flow parameters, flow field parameters, and structural parameters of the launch vehicle under the target operating conditions to be predicted are input into a trained multi-expert feature fusion neural network model to obtain the amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution. The amplitude and phase angle parameters of the predicted unsteady aerodynamic force distribution are inverted to obtain the unsteady aerodynamic force distribution parameters; Based on unsteady aerodynamic distribution parameters and structural mode shapes, the distributed aerodynamic damping of the launch vehicle is determined; The construction of the multi-expert feature fusion neural network model includes the following steps: A feature sample set is constructed, which includes multiple sets of sample data. Each set of sample data includes: flow condition parameters, flow field parameters, structural parameters, aerodynamic distribution data of the steady flow field corresponding to the flow condition parameters, flow field parameters and structural parameters, amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution under the forced motion of the sinusoidal motion signal corresponding to the steady flow field; The structure of a multi-expert feature fusion neural network model is determined. The multi-expert feature fusion neural network model includes a flow parameter feature encoder, a spatial flow field feature encoder, a structural parameter feature encoder, and a fusion network. The inputs of the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder are parallel data input terminals. The input of the fusion network is connected to the output terminals of the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder. The multi-expert feature fusion neural network model is trained using a feature sample set until the prediction accuracy meets the prediction requirements, thus obtaining a well-trained multi-expert feature fusion neural network model.

[0007] Further, acquiring the aerodynamic distribution data of the steady flow field includes the following steps: Obtain the aerodynamic shape and dimensions of the launch vehicle; In the finite element analysis environment, the aerodynamic shape and dimensions of the launch vehicle are modeled and meshed for the aerodynamic shape of the launch vehicle. Extract the mode shapes and corresponding natural frequencies and generalized mass of the launch vehicle; Configure the solver parameters, and set the solver's solution method, numerical format, and boundary conditions. Configure the flow conditions in the solver, perform steady flow field calculations, and obtain the aerodynamic distribution data of the steady flow field.

[0008] Furthermore, the amplitude parameters and phase angle parameters of the unsteady aerodynamic force distribution corresponding to the steady flow field under the forced motion of the sinusoidal motion signal include the following steps: Based on the aerodynamic distribution data of the steady flow field, forced motion is applied to the launch vehicle. The type of forced motion is a sinusoidal motion signal. Unsteady aerodynamic distribution data under the forced motion of the sinusoidal motion signal corresponding to the steady flow field is obtained. Based on the above unsteady aerodynamic distribution data, feature extraction is performed to obtain the amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution.

[0009] Furthermore, the structure of the multi-expert feature fusion neural network model is determined, including the following steps: Based on the type and characteristics of the sample data in the feature sample set, it is classified into three categories: incoming flow parameters, spatial steady flow field parameters, and structural parameters. A fully connected neural network is used to capture the physical laws and interrelationships in the features of the sample data, forming a flow parameter feature encoder, a spatial flow field feature encoder, and a structural parameter feature encoder. The input parameters are compressed by the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder, and the output is the flow parameter feature, the spatial flow field feature, and the structural parameter feature. The output flow parameter characteristics, spatial flow field characteristics and structural parameter characteristics are fused and input into the fusion network. After processing by the fusion network, the output is the amplitude parameters and phase angle parameters of the predicted unsteady aerodynamic distribution.

[0010] Furthermore, before training the flow parameter feature encoder, spatial flow field feature encoder, structural parameter feature encoder, and fusion network using the feature sample set, the sample data in the feature sample set is normalized and mapped to a preset interval [0, 1] to obtain the preprocessed feature sample set.

[0011] Furthermore, the amplitude and phase parameters of the predicted unsteady aerodynamic distribution are inverted to obtain the unsteady aerodynamic distribution parameters, which are determined based on the following formula: , in, The parameters for predicting unsteady aerodynamic distributions are in Pascals. The pressure distribution is obtained from the steady-state solution, in Pascals. It is the phase angle parameter for predicting unsteady aerodynamic force distribution, in degrees. It is the amplitude parameter for predicting unsteady aerodynamic force distribution, in Pascals. The frequency of structural motion is expressed in Hertz.

[0012] Furthermore, the data from the feature sample set are sequentially input into the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder for training. The prediction errors of the amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution are used as the training basis to perform end-to-end joint training on the flow parameter feature encoder, the spatial flow field feature encoder, the structural parameter feature encoder, and the fusion network.

[0013] Furthermore, the prediction accuracy of the multi-expert feature fusion neural network model is determined based on the total loss function, which is determined based on the following formula: , in, , , This represents the total loss function value. Indicates amplitude prediction loss. This represents the phase angle prediction loss. Indicates the regularization loss. This represents the weighting coefficient corresponding to each loss term. This indicates the number of samples used in the loss function calculation. Indicates the first The amplitude parameters of the true unsteady aerodynamic distribution corresponding to each sample, in Pascals. Indicates the first The amplitude parameters of the predicted unsteady aerodynamic distribution corresponding to each sample, in Pascals. Indicates the first Phase angle parameters corresponding to the true unsteady aerodynamic distribution of each sample, in degrees. Indicates the first The phase angle parameters of the predicted unsteady aerodynamic distribution corresponding to each sample, in degrees.

[0014] Compared with existing technologies, this invention provides a method for predicting distributed aerodynamic damping of launch vehicles based on steady flow field feature modeling. It establishes a rapid and high-precision prediction chain from steady flow field features to unsteady aerodynamic force distribution and then to aerodynamic damping distribution. By constructing a multi-expert feature fusion neural network model, it can establish a mapping relationship between steady flow field features and corresponding unsteady aerodynamic responses. Furthermore, by employing a distributed aerodynamic damping prediction method, it spatially couples the predicted unsteady aerodynamic force distribution with structural mode shapes, thereby determining the distributed aerodynamic damping of the launch vehicle and evaluating the impact of different regions on aerodynamic damping of the launch vehicle's surface. The specific contribution of this method effectively quantifies the contribution of unsteady flow phenomena on the surface of launch vehicles to aeroelastic stability. This avoids the problem of traditional methods requiring time-consuming unsteady numerical simulations or experiments for each operating condition, simplifying the determination process. It can greatly improve solution efficiency and effectively reduce time costs. It can be applied not only to the aeroelastic stability assessment of complex flow stages such as transonic speeds of launch vehicles, timely detection of potential negative damping instability risks, but also to provide rapid feedback and data support for rocket body shape optimization and structural stiffness configuration, thereby improving the aeroelastic design level and flight safety of launch vehicles. Attached Figure Description

[0015] Figure 1 This is a flowchart of the present invention.

[0016] Figure 2 This is a schematic diagram illustrating the implementation of the launch vehicle distributed aerodynamic damping prediction method based on steady flow field characteristic modeling of the present invention.

[0017] Figure 3 A schematic diagram illustrating the process of constructing a multi-expert feature fusion neural network model.

[0018] Figure 4 This is a schematic diagram illustrating the acquisition and storage of simulation data according to the present invention.

[0019] Figure 5 This is a schematic diagram of the multi-expert feature fusion neural network model of the present invention.

[0020] Figure 6 This is a schematic diagram of the non-steady training results of the present invention.

[0021] Figure 7 This is a schematic diagram showing the distribution of aerodynamic amplitude and phase angle along the rocket axis in this invention.

[0022] Figure 8 This is a comparison chart of the simulation results and the modeled and predicted pressure coefficient distribution results of this invention.

[0023] Figure 9 This is a schematic diagram comparing the modeled damping distribution results with the simulation results of this invention. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0025] Example 1 like Figure 1 As shown, this invention provides a method for predicting distributed aerodynamic damping of launch vehicles based on steady flow field characteristic modeling, including the following steps: Obtain the flow parameters, flow field parameters, and structural parameters of the launch vehicle under the target operating conditions to be predicted; The flow parameters, flow field parameters, and structural parameters of the launch vehicle under the target operating conditions to be predicted are input into a trained multi-expert feature fusion neural network model to obtain the amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution. The amplitude and phase angle parameters of the predicted unsteady aerodynamic force distribution are inverted to obtain the unsteady aerodynamic force distribution parameters; Based on unsteady aerodynamic distribution parameters and structural mode shapes, the distributed aerodynamic damping of the launch vehicle is determined.

[0026] Specifically, in combination Figure 2 The above method is described in detail below: S1: Constructing the feature sample set Multiple sets of sample data are acquired. Each set of sample data includes: flow condition parameters, flow field parameters, structural parameters, aerodynamic distribution data of the steady flow field corresponding to the flow condition parameters, flow field parameters and structural parameters, amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution under the forced motion of the sinusoidal motion signal corresponding to the steady flow field. The above multiple sets of sample data constitute the feature sample set.

[0027] Reference Figure 4 In practical applications, the specific methods for obtaining aerodynamic distribution data of steady flow fields include the following steps: S11: Obtain the aerodynamic shape and dimensions of the launch vehicle. In the finite element analysis environment, model the aerodynamic shape and dimensions of the launch vehicle and perform mesh generation for the aerodynamic shape of the launch vehicle.

[0028] S12: Extract the modal shapes and corresponding natural frequencies and generalized masses of the launch vehicle using finite element method.

[0029] S13: Configure the solver parameters, and set the solver's solution method, numerical format and boundary conditions. Configure the flow conditions in the solver, perform steady flow field calculations and obtain the aerodynamic distribution data of the steady flow field.

[0030] The specific method for obtaining the amplitude and phase angle parameters of the unsteady aerodynamic force distribution under forced motion of a sinusoidal motion signal corresponding to a steady flow field includes the following steps: Based on the aforementioned techniques for obtaining aerodynamic distribution data of steady flow fields, a sinusoidal forced motion training signal is set up to conduct unsteady simulation solutions, and the amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution extracted by the simulation are stored in chronological order.

[0031] It should be further noted that the aforementioned flow parameters include at least the Mach number. and angle of attack The flow field parameters are steady flow field parameters related to the corresponding flow condition parameters. Structural parameters are used to characterize the structural state or structural features.

[0032] The system acquires the flow condition parameters, flow field parameters, and structural parameters corresponding to the launch vehicle. Based on these parameters, it performs simulations of the steady flow field and acquires and stores the aerodynamic distribution data of the steady flow field.

[0033] Based on the aerodynamic distribution data of the steady flow field, forced motion is applied to the launch vehicle. The type of forced motion is a sinusoidal motion signal. Unsteady aerodynamic distribution data under the forced motion of the sinusoidal motion signal corresponding to the steady flow field is obtained. Based on the above unsteady aerodynamic distribution data, feature extraction is performed to obtain the amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution.

[0034] Specifically, the above calculations of steady and unsteady flow fields are performed by discretizing the NS equations using the RANS method. When solving the NS equations, commonly used domestic and foreign commercial software (such as ANSYS Fluent) and open-source software (PHengLei) can be used. All of these software are existing mature technologies and will not be elaborated on here.

[0035] S2: Constructing a multi-expert feature fusion neural network model This invention establishes a multi-expert feature fusion neural network model, encodes and compresses sample data from feature sample sets of different sources and dimensions, and predicts the amplitude and phase parameters of unsteady aerodynamic distributions based on the encoded low-dimensional feature parameters.

[0036] Reference Figure 5 Furthermore, the specific method for constructing a multi-expert feature fusion neural network model includes the following steps: S21: Based on the type and characteristics of the sample data in the feature sample set, they are classified into three categories: incoming flow parameters, spatial steady flow field parameters, and structural parameters, such as... Figure 3 As shown, a three-feature multi-expert feature fusion neural network model is established. The fully connected neural network is used to capture the physical laws and interrelationships in the sample data features of the feature sample set, forming a flow parameter feature encoder, a spatial flow field feature encoder, and a structural parameter feature encoder. The input parameters are compressed by the three-feature multi-expert feature fusion neural network model, and the compressed parameter outputs are flow parameter features, spatial flow field features, and structural parameter features.

[0037] S22: The flow parameter features, spatial flow field features and structural parameter features output in step S21 are fused and input into the fusion network. After processing by the fusion network, the amplitude parameters and phase angle parameters of the predicted unsteady aerodynamic distribution are output.

[0038] S23: Compare the amplitude and phase parameters of the unsteady aerodynamic distribution predicted by the fusion network in step S22 with the amplitude and phase parameters of the unsteady aerodynamic distribution obtained in step S1, calculate the prediction error. If the error is greater than the preset threshold, adjust the parameters of the multi-expert feature fusion neural network model and repeat the training process of steps S21 and S22. If the error is less than or equal to the preset threshold, the multi-expert feature fusion neural network model is considered to have been trained and a multi-expert feature fusion neural network model that can be used for fast prediction is obtained.

[0039] The multi-expert feature fusion neural network model includes a flow parameter feature encoder, a spatial flow field feature encoder, a structural parameter feature encoder, and a fusion network. The outputs of the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder are used as inputs to the fusion network. The feature parameters output by the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder are input into the fusion network. The aim is to predict the amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution through the fusion network.

[0040] After the multi-expert feature fusion neural network model is built, it needs to be trained to ensure that it can meet the accuracy requirements for use.

[0041] Before training, the sample data in the feature sample set is preprocessed to obtain the preprocessed feature sample set.

[0042] Specifically, preprocessing includes: Mach number in the flow condition parameters and angle of attack Normalization is performed, mapping the result to a preset interval [0, 1], to obtain the normalized Mach number. and normalized angle of attack .

[0043] Similarly, the flow field parameters and structural parameters are also normalized and mapped to a preset interval [0, 1] to eliminate the influence of different dimensions and numerical ranges on the training of the multi-expert feature fusion neural network model.

[0044] The sample data in the preprocessed feature sample set is divided into a training set and a validation set according to a preset ratio, with the training set accounting for 80% and the validation set accounting for 20%. The multi-expert feature fusion neural network model is trained using the training set, and the validation set is used to confirm the prediction accuracy of the multi-expert feature fusion neural network model. When the total loss function value on the validation set is lower than a preset threshold, or when the decrease in the total loss function is less than the preset threshold in several consecutive training rounds, the multi-expert feature fusion neural network model is considered to have completed training, and the multi-expert feature fusion neural network model is obtained.

[0045] Specifically, the aforementioned flow parameter feature encoder, spatial flow field feature encoder, and structural parameter feature encoder all belong to parameter encoders. Their function is to map the sample data in the corresponding feature sample set from a high-dimensional or multi-dimensional form to low-dimensional feature parameters.

[0046] It should be noted that in this embodiment, the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder are implemented using a fully connected neural network (FCNN). That is, a feedforward fully connected network consisting of an input layer, one or more hidden layers, and an output layer is used to perform nonlinear mapping and feature compression on the sample data in the feature sample set to form feature parameters.

[0047] It should be noted that the aforementioned fully connected neural network (FCNN) is a multiple-input multiple-output structure. Feature compression of the input parameters specifically refers to using a fully connected neural network (FCNN) to transform multiple parameters into a single parameter.

[0048] Specifically, when compressing sample data in a feature sample set, for flow condition parameters, including Mach number... and angle of attack The normalized flow condition parameters are encoded using a flow parameter feature encoder, transforming them into a low-dimensional flow parameter feature. This is used to characterize flow condition information, and its expression is: , in, These are the output parameters of the flow parameter characteristic encoder. This represents a flow parameter characteristic encoder. It is the normalized Mach number. It is the angle of attack after normalization.

[0049] Similarly, spatial flow field parameters and structural parameters can also be compressed into a single parameter using a corresponding encoder. , .

[0050] The fully connected neural network (FCNN) structure of the flow parameter feature encoder described above includes: The input layer has 2 neurons, the first hidden layer has 16 neurons with ReLU activation, the second hidden layer has 8 neurons with ReLU activation, and the output layer has 1 neuron to output the flow parameter features. .

[0051] A spatial flow field feature encoder is used to encode aerodynamic shape parameters and flow field parameters into low-dimensional feature parameters to characterize flow field information. The spatial flow field feature encoder is also implemented using a fully connected neural network (FCNN) structure. The flow field parameters are input into the spatial flow field feature encoder to obtain the spatial flow field features. Its expression is: , in, Encoder representing spatial flow field characteristics. This represents the low-dimensional feature parameters obtained after encoding the flow field parameters. These represent the normalized aerodynamic shape parameters and flow field parameters, respectively.

[0052] The fully connected neural network (FCNN) structure of the spatial flow field feature encoder includes: an input layer (the number of neurons is determined according to the dimension of the flow field parameters), a first hidden layer with 128 neurons and ReLU activation function, a second hidden layer with 64 neurons and ReLU activation function, a third hidden layer with 32 neurons and ReLU activation function, and an output layer with 1 neuron for outputting the spatial flow field features. .

[0053] It should be noted that since the dimension of flow field parameters is usually higher than that of flow condition parameters, spatial flow field feature encoders can be configured with more hidden layers or more neurons to enhance the ability to compress and express high-dimensional flow field features.

[0054] A structural parameter feature encoder (SMR) is used to encode structural parameters into low-dimensional feature parameters to represent structural information. The SMR is implemented using a fully connected neural network (FCNN) architecture. The structural parameters are input into the SMR to obtain the structural parameter features. Its expression is: , in, This represents a structural parameter feature encoder. This represents the low-dimensional feature parameters obtained after encoding the structural parameters. These represent the normalized natural frequency, generalized mass, and mode shape of the structure, respectively.

[0055] The fully connected neural network (FCNN) architecture for structural parametric feature encoders includes: The number of neurons in the input layer is determined by the structural parameter dimension. The first hidden layer has 32 neurons with the ReLU activation function. The second hidden layer has 16 neurons with the ReLU activation function. The output layer has 1 neuron and is used to output the structural parameter features. .

[0056] In obtaining flow parameter characteristics Spatial flow field characteristics and structural parameter characteristics Then, the three vectors are concatenated to form a fused feature vector. Its expression is: , Here, Concat represents the vector concatenation operation. This represents the combined feature vector after fusion.

[0057] The fused feature vector is input into the fusion network to predict the amplitude and phase parameters of the unsteady aerodynamic distribution, and its expression is as follows: , , in, , These represent the output mapping relationships for amplitude and phase parameters, respectively. It is the phase angle parameter for predicting unsteady aerodynamic force distribution, in degrees. It is the amplitude parameter for predicting unsteady aerodynamic force distribution, in Pascals.

[0058] The fusion network adopts a fully connected neural network (FCNN) structure, including: an input layer with 3 neurons, a first hidden layer with 64 neurons and ReLU activation function, a second hidden layer with 32 neurons and ReLU activation function, a third hidden layer with 16 neurons and ReLU activation function, and an output layer with 2 neurons, corresponding to the amplitude parameter and phase angle parameter of the unsteady aerodynamic distribution, respectively.

[0059] It should be noted that fully connected neural networks support multiple inputs and multiple outputs, so fusion networks can simultaneously predict the amplitude and phase parameters of unsteady aerodynamic distributions under the same network structure.

[0060] Due to the characteristics of flow parameters Spatial flow field characteristics and structural parameter characteristics All of these are intermediate low-dimensional representation parameters learned by each parameter encoder through FCNN, and there are no pre-set standard label values. Therefore, this invention does not use the feature parameters themselves as the supervision target, but uses the prediction error of the amplitude parameter and phase angle parameter of the unsteady aerodynamic distribution as the training basis to perform end-to-end joint training on the flow parameter feature encoder, the spatial flow field feature encoder, the structural parameter feature encoder and the fusion network.

[0061] Specifically, the total loss function during training can be expressed as: , in, This represents the total loss function value. Indicates amplitude prediction loss. This represents the phase angle prediction loss. Indicates the regularization loss. This represents the weighting coefficient corresponding to each loss term.

[0062] Among them, amplitude prediction loss It can be expressed in the form of mean square error as: , in, This indicates the number of samples used in the loss function calculation. Indicates the first The amplitude parameters of the true unsteady aerodynamic distribution corresponding to each sample, in Pascals. Indicates the first The amplitude parameters of the predicted unsteady aerodynamic distribution corresponding to each sample, in Pascals.

[0063] Phase Angle Prediction Loss Determined based on the following formula: , in, Indicates the first Phase angle parameters corresponding to the true unsteady aerodynamic distribution of each sample, in degrees. Indicates the first The phase angle parameters of the predicted unsteady aerodynamic distribution corresponding to each sample, in degrees.

[0064] Regularization loss Used to constrain network parameters and reduce the risk of overfitting in multi-expert feature fusion neural network models.

[0065] S3: Based on a multi-expert feature fusion neural network model, predict the unsteady aerodynamic distribution parameters of the launch vehicle under target operating conditions. First, the flow parameters, flow field parameters, and structural parameters of the launch vehicle under the target operating condition to be predicted are input into the multi-expert feature fusion neural network model to predict the amplitude parameters and phase angle parameters of the unsteady aerodynamic force distribution. Then, the predicted amplitude parameters and phase angle parameters of the unsteady aerodynamic force distribution are inverted to obtain the unsteady aerodynamic force distribution parameters.

[0066] Specifically, refer to Figure 2 The specific method for predicting the unsteady aerodynamic distribution parameters of a launch vehicle under target operating conditions includes the following steps: S31: Given the incoming flow parameters and structural parameters of the launch vehicle to be predicted, perform numerical calculations of the steady flow field, and obtain and retain the steady aerodynamic distribution parameters on the surface of the launch vehicle. S32: Input the incoming flow parameters, structural parameters, and steady aerodynamic force distribution parameters calculated in step S31 of the launch vehicle to be predicted into the trained multi-expert feature fusion neural network model to quickly predict the amplitude parameters and phase angle parameters of the unsteady aerodynamic force distribution corresponding to the sinusoidal forced motion of the launch vehicle to be predicted. S33: Based on the amplitude and phase angle parameters of the unsteady aerodynamic force distribution, and combined with the sinusoidal motion law, the unsteady aerodynamic force distribution parameters of the launch vehicle to be predicted are derived in reverse. S34: Based on the obtained unsteady aerodynamic distribution parameters and structural mode shapes, determine the distributed aerodynamic damping of the launch vehicle.

[0067] It should be noted that the inversion of the amplitude and phase parameters of the predicted unsteady aerodynamic distribution follows these steps: Premise: This method is derived by introducing the first harmonic assumption based on the distributed damping prediction method. It only requires the amplitude and phase angle of the first harmonic component of the unsteady aerodynamic force. Based on the frequency locking assumption (one of the premises of this method is that the aerodynamic force frequency is equal to the natural frequency of the structure, i.e. the frequency is known), the first harmonic of the aerodynamic force can be recovered.

[0068] During the forced motion, since the motion is simple harmonic motion under a sinusoidal signal, the wall pressure distribution on the surface mesh is shown here. There is a phase angle between it and the local displacement signal. , It can be written as: , in, Wall pressure distribution, unit: Pascal. This represents the time-averaged pressure distribution, in Pascals. The amplitude of the first harmonic term, in Pascals (Pa). The frequency of structural motion, measured in Hertz. The phase angle is expressed in degrees.

[0069] Then, when the predicted phase angle parameters are obtained through the fusion network... and predicted amplitude parameters Replace phase angle Amplitude of the first harmonic term Pressure distribution of steady solution Replacement mean This allows us to inverse the unsteady aerodynamic distribution parameters. ,at this time: , in, The parameters for predicting unsteady aerodynamic distributions are in Pascals. The pressure distribution obtained from the steady-state solution, in Pascals, is used to replace the time-averaged pressure distribution. , and The predicted phase angle and amplitude parameters are given in Pascals and degrees, respectively. The frequency of structural motion is expressed in Hertz.

[0070] Specifically, the distributed aerodynamic damping prediction method used in this invention, based on the forced motion method, employs an aerodynamic damping distribution method to analyze the contribution of local aerodynamic forces to the aeroelastic stability of the launch vehicle.

[0071] For the j-th order structural mode, the Lagrange equations of motion can be transformed into: , in, , , These represent the generalized displacement, generalized velocity, and generalized acceleration corresponding to the structural modes, respectively. Here, represents the structural damping coefficient. , Let be the natural frequency of the j-th structural mode, in Hertz. Mass in a general sense, unit: kilogram. For generalized aerodynamic force, the unit is Newton.

[0072] The generalized aerodynamic force corresponding to the j-th order structural mode Determined based on the following formula: , in, Pressure distribution on the structural wall, unit: Pascal. The surface vector of the structural wall. Let be the structural mode shape vector corresponding to the j-th mode. For generalized aerodynamic forces, the unit is Newton. Divide the wall area into regions.

[0073] Under forced motion conditions, the forced aerodynamic force on the right-hand side of the structural motion equation can be equivalently expressed as an aerodynamic damping term and an aerodynamic stiffness term, namely: , Omitting subsequent higher-order terms, in the formula This represents the corresponding aerodynamic damping coefficient. Let be the corresponding aerodynamic stiffness coefficient. Since frequency locking occurs when a launch vehicle experiences flutter, the aerodynamic stiffness effect generated by the generalized aerodynamic force is extremely small and negligible relative to the structural stiffness. Therefore, the generalized aerodynamic force can ultimately be equivalent to only an aerodynamic damping term, i.e.: , Since the structural modes have very small aerodynamic damping during free response, they can be considered as simple harmonic motions of constant amplitude: , Where A is the amplitude of the simple harmonic motion signal, in meters.

[0074] Substituting into the above equation and performing the operation, we get: , , , Within a sinusoidal period T, from to By performing orthogonal integration at each time step, we can obtain: , , After simplification, we get: , in, This is the aerodynamic damping coefficient. For structural generalized mass, the unit is kilogram. The amplitude of the simple harmonic motion signal is expressed in meters. The structure's natural frequency, measured in Hertz. The period of the sine wave is in seconds. For unsteady generalized aerodynamic forces, the unit is Newton.

[0075] Because the launch vehicle has a spin-shaped configuration, its flow characteristics are distributed along the axial direction. This results in the generalized aerodynamic forces and the resulting aerodynamic damping also having an axial distribution. The predicted pressure distribution can be obtained through a multi-expert feature fusion neural network model. The unit is Pascal. By integrating this force distribution along the cross-sectional direction (y, z directions), the predicted generalized aerodynamic force distribution along the axial direction is obtained. The unit is cow.

[0076] The above-predicted aerodynamic damping distribution It can be represented as , Here For the predicted aerodynamic damping distribution, For structural generalized mass, the unit is kilogram. The amplitude of the simple harmonic motion signal is expressed in meters. The structure's natural frequency, measured in Hertz. The period of the sine wave is in seconds. The generalized aerodynamic force distribution along the axial direction is predicted, in Newtons per meter.

[0077] Therefore, the overall predicted aerodynamic damping is expressed as: , in, This represents the overall predicted aerodynamic damping. The length of the arrow is in meters.

[0078] In practical operation, the method for predicting distributed aerodynamic damping of a launch vehicle based on steady flow field characteristic modeling provided by this invention includes the following steps: S100: First, based on the launch vehicle model to be analyzed, an accurate aerodynamic geometric model is established. Then, computational fluid dynamics (CFD) preprocessing software (such as Pointwise, ICEM CFD, etc.) is used to perform structured or unstructured mesh generation on the computational domain. The mesh needs to be refined on the rocket surface, in areas where shock waves may occur, and in flow separation zones to ensure that key flow characteristics can be accurately captured. The generated mesh is then exported in a format that the solver can recognize. S101: Establish a structural finite element model of the launch vehicle using finite element analysis software (such as ANSYS, Nastran, etc.), perform modal analysis, and extract the mode shapes, corresponding natural frequencies, and generalized mass of interest. S102: Configure the CFD solver (such as Fluent, CFX, etc.), set the solution method (such as pressure-based solver or density-based solver), turbulence model (such as one-equation SA model or two-equation SST k-ω model; the choice should be made according to the specific situation in the specific case. In the rocket case, the SA model is preferred), spatial and temporal discretization format, and set a series of different combinations of incoming flow parameters according to the operating condition range to be covered by the training feature sample set, including but not limited to Mach number, angle of attack, etc. Perform steady flow field calculation for each operating condition. After the calculation is completed, extract and store the steady aerodynamic distribution data of the entire surface (or area of ​​interest) of the launch vehicle. At the same time, record the incoming flow parameters and structural parameters corresponding to the operating condition. S103: For the same batch of working conditions for which the steady flow field has been calculated in step S102 above, unsteady simulation is carried out based on the steady flow field solution. Specifically, the launch vehicle model is set to perform sinusoidal forced vibration according to the mode shape extracted above, with a specified amplitude and a series of different frequencies (distributed around the natural frequency), and unsteady solution is performed. During the calculation process, the data of the unsteady aerodynamic force distribution on the rocket surface changing with time at each unsteady time step is recorded. S104: Post-process the unsteady aerodynamic time series data stored in step S103 above. For the aerodynamic data at each position on the surface, extract the amplitude of the unsteady aerodynamic force at that position under forced vibration and the phase angle relationship between the unsteady aerodynamic force distribution and the motion signal to obtain a complete set of amplitude parameters and phase angle parameters of the spatially distributed unsteady aerodynamic force distribution. S105: Based on the physical characteristics and dimensions of the input data, three independent expert neural networks are designed, all of which adopt a fully connected neural network structure. They are: a flow parameter feature encoder (input is the incoming flow parameter vector); a spatial flow field feature encoder (input is high-dimensional steady aerodynamic distribution data); and a structural parameter feature encoder (input is the structural parameter vector). S106: The feature vectors output by the above three expert neural networks are concatenated to form a fused feature vector. This fused feature vector is then input into a fusion network. The output layer of the fusion network consists of two nodes, which correspond to the amplitude parameters and phase angle parameters of the predicted unsteady aerodynamic distribution, respectively. S107: Divide the feature sample set prepared in S100-S106 into a training set, a validation set, and a test set. The training set is used to optimize the network weights, the validation set is used to adjust hyperparameters and prevent overfitting, and the test set is used to finally evaluate the performance of the multi-expert feature fusion neural network model. Then, by comparing the amplitude and phase angle of the unsteady aerodynamic distribution predicted by the fusion network in step S106 with the true values ​​extracted from the unsteady simulation data, the prediction error is calculated. If the error is greater than a preset threshold, the parameters of the multi-expert feature fusion neural network model are adjusted and the training process of steps S105 and S106 is repeated. If the error is less than or equal to the preset threshold, the multi-expert feature fusion neural network model is considered to have been trained and a multi-expert feature fusion neural network model that can be used for fast prediction is obtained. S108: For new operating conditions that require prediction of aerodynamic damping, only one steady-state calculation is needed first. This calculation takes much less time than the unsteady-state calculation. After the calculation is completed, save the steady-state aerodynamic force distribution data under this operating condition. S109: Input the incoming flow parameters of the new operating condition, the calculated steady aerodynamic force distribution, and the structural parameters into the trained multi-expert feature fusion neural network model. Use the multi-expert feature fusion neural network model to predict the amplitude and phase angle parameters of the unsteady aerodynamic force distribution. Then, based on the obtained amplitude and phase angle parameters of the unsteady aerodynamic force distribution, invert the unsteady aerodynamic force distribution. Couple it with the structural mode shape to form a generalized aerodynamic force distribution, and finally determine the distributed aerodynamic damping of the launch vehicle.

[0079] Figures 6 to 9 This is a schematic diagram of the relevant results, where, Figure 6 This is a schematic diagram of the non-steady training results of the present invention. Figure 7 This is a schematic diagram showing the distribution of aerodynamic amplitude and phase angle along the rocket axis in this invention. Figure 8 This is a comparison chart of the pressure coefficient distribution results from the simulation and modeling predictions of this invention. Figure 9The diagram shows a comparison between the modeled damping distribution results and the simulation results of this invention. As can be seen from the diagram, the simulation results and the modeling results have a high degree of fit. Therefore, the method provided by this invention can effectively predict the aerodynamic damping distribution.

[0080] It should be further explained that the above method can also be implemented using electronic devices. The structure of such electronic devices specifically includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The processor executes the computer program to implement the above-described method for predicting the distributed aerodynamic damping of a launch vehicle based on steady flow field characteristics modeling.

[0081] A communication interface is set up between the memory and the processor to realize the signal connection between the two and realize data transmission. The communication interface can be either a serial interface or a parallel interface.

[0082] The processor may be a central processing unit (CPU), a specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0083] The above method can also be implemented using a computer-readable storage medium that stores a computer program, which is executed by a processor to implement the above-described method for predicting distributed aerodynamic damping of a launch vehicle based on steady flow field characteristics modeling.

[0084] This invention provides a method for predicting the distributed aerodynamic damping of launch vehicles based on steady flow field feature modeling. By constructing a multi-expert feature fusion network model, a mapping relationship is established between steady flow field features and corresponding unsteady aerodynamic responses. This avoids the problem of traditional methods requiring time-consuming unsteady numerical simulations or experiments for each operating condition, which greatly improves solution efficiency and effectively reduces time costs. At the same time, the predicted aerodynamic damping distribution of the launch vehicle can assess the specific contribution of different regions on the rocket surface to aerodynamic damping, thereby effectively quantifying the contribution of unsteady flow phenomena on the launch vehicle surface to aeroelastic stability. This provides accurate and reliable theoretical guidance for subsequent active aeroelastic stability margin improvement methods such as improving aerodynamic shape and increasing flow control.

[0085] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting distributed aerodynamic damping of launch vehicles based on steady flow field characteristic modeling, characterized in that, Includes the following steps: The flow parameters, flow field parameters, and structural parameters of the launch vehicle under the target operating conditions to be predicted are obtained and input into a trained multi-expert feature fusion neural network model to obtain the amplitude and phase angle parameters of the unsteady aerodynamic force distribution. The amplitude and phase angle parameters of the predicted unsteady aerodynamic force distribution are inverted to obtain the unsteady aerodynamic force distribution parameters. Based on the unsteady aerodynamic force distribution parameters and structural mode shapes, the distributed aerodynamic damping of the launch vehicle is determined. The construction of the multi-expert feature fusion neural network model includes the following steps: A feature sample set is constructed, which includes multiple sets of sample data. Each set of sample data includes: flow condition parameters, flow field parameters, structural parameters, aerodynamic distribution data of the steady flow field corresponding to the flow condition parameters, flow field parameters and structural parameters, amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution under the forced motion of the sinusoidal motion signal corresponding to the steady flow field; The structure of a multi-expert feature fusion neural network model is determined. The multi-expert feature fusion neural network model includes a flow parameter feature encoder, a spatial flow field feature encoder, a structural parameter feature encoder, and a fusion network. The inputs of the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder are parallel data input terminals. The input of the fusion network is connected to the output terminals of the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder. The multi-expert feature fusion neural network model is trained using a feature sample set until the prediction accuracy meets the prediction requirements, thus obtaining a well-trained multi-expert feature fusion neural network model.

2. The method for predicting distributed aerodynamic damping of a launch vehicle based on steady flow field characteristic modeling according to claim 1, characterized in that, Obtaining the aerodynamic distribution data of the steady flow field includes the following steps: Obtain the aerodynamic shape and dimensions of the launch vehicle; In the finite element analysis environment, the aerodynamic shape and dimensions of the launch vehicle are modeled and meshed for the aerodynamic shape of the launch vehicle. Extract the mode shapes and corresponding natural frequencies and generalized mass of the launch vehicle; Configure the solver parameters, and set the solver's solution method, numerical format, and boundary conditions. Configure the flow conditions in the solver, perform steady flow field calculations, and obtain the aerodynamic distribution data of the steady flow field.

3. The method for predicting distributed aerodynamic damping of a launch vehicle based on steady flow field characteristic modeling according to claim 2, characterized in that, The amplitude and phase parameters of the unsteady aerodynamic force distribution corresponding to the steady flow field under the forced motion of the sinusoidal motion signal include the following steps: Based on the aerodynamic distribution data of the steady flow field, forced motion is applied to the launch vehicle. The type of forced motion is a sinusoidal motion signal. Unsteady aerodynamic distribution data under the forced motion of the sinusoidal motion signal corresponding to the steady flow field is obtained. Based on the above unsteady aerodynamic distribution data, feature extraction is performed to obtain the amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution.

4. The method for predicting distributed aerodynamic damping of a launch vehicle based on steady flow field characteristic modeling according to claim 1, characterized in that, Determine the structure of the multi-expert feature fusion neural network model, including the following steps: Based on the type and characteristics of the sample data in the feature sample set, it is classified into three categories: incoming flow parameters, spatial steady flow field parameters, and structural parameters. A fully connected neural network is used to capture the physical laws and interrelationships in the features of the sample data, forming a flow parameter feature encoder, a spatial flow field feature encoder, and a structural parameter feature encoder. The input parameters are compressed by the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder, and the output is the flow parameter feature, the spatial flow field feature, and the structural parameter feature. The output flow parameter characteristics, spatial flow field characteristics and structural parameter characteristics are fused and input into the fusion network. After processing by the fusion network, the output is the amplitude parameters and phase angle parameters of the predicted unsteady aerodynamic distribution.

5. The method for predicting distributed aerodynamic damping of a launch vehicle based on steady flow field characteristic modeling according to claim 4, characterized in that, Before training the flow parameter feature encoder, spatial flow field feature encoder, structural parameter feature encoder and fusion network using the feature sample set, the sample data in the feature sample set is normalized and mapped to a preset interval [0, 1] to obtain the preprocessed feature sample set.

6. The method for predicting distributed aerodynamic damping of a launch vehicle based on steady flow field characteristic modeling according to claim 5, characterized in that, The amplitude and phase parameters of the predicted unsteady aerodynamic force distribution are inverted to obtain the unsteady aerodynamic force distribution parameters, which are determined based on the following formula: ， in, For the predicted unsteady aerodynamic distribution parameters, The pressure distribution obtained from the steady-state solution, It is a phase angle parameter for predicting unsteady aerodynamic force distribution. It is the amplitude parameter for predicting unsteady aerodynamic force distribution, in Pascals. The frequency of structural motion is expressed in Hertz.

7. The method for predicting distributed aerodynamic damping of a launch vehicle based on steady flow field characteristic modeling according to claim 1, characterized in that, The data from the feature sample set are sequentially input into the flow parameter feature encoder, the spatial flow field feature encoder, and the structural parameter feature encoder for training. The prediction errors of the amplitude parameters and phase angle parameters of the unsteady aerodynamic distribution are used as the training basis, and the flow parameter feature encoder, the spatial flow field feature encoder, the structural parameter feature encoder, and the fusion network are jointly trained end-to-end.

8. The method for predicting distributed aerodynamic damping of a launch vehicle based on steady flow field characteristic modeling according to claim 7, characterized in that, The prediction accuracy of the multi-expert feature fusion neural network model is determined based on the total loss function, which is determined based on the following formula: ， in, , ， This represents the total loss function value. Indicates amplitude prediction loss. This represents the phase angle prediction loss. Indicates the regularization loss. This represents the weighting coefficient corresponding to each loss term. This indicates the number of samples used in the loss function calculation. Indicates the first The amplitude parameters of the true unsteady aerodynamic distribution corresponding to each sample, in Pascals. Indicates the first The amplitude parameters of the predicted unsteady aerodynamic distribution corresponding to each sample, in Pascals. Indicates the first Phase angle parameters corresponding to the true unsteady aerodynamic distribution of each sample, in degrees. Indicates the first The phase angle parameters of the predicted unsteady aerodynamic distribution corresponding to each sample, in degrees.

Images