A method, apparatus, equipment, and medium for predicting the aerodynamic damping distribution of a launch vehicle.
By combining convolutional autoencoders with long short-term memory neural networks, the problem of high computational cost in predicting the aerodynamic damping distribution of launch vehicles is solved, enabling fast and accurate prediction of aerodynamic damping distribution and improving the efficiency and generalization ability of aeroelastic stability analysis of launch vehicles.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NORTHWESTERN POLYTECHNICAL UNIV
- Filing Date
- 2026-05-27
- Publication Date
- 2026-07-31
AI Technical Summary
In the existing technology, the distribution prediction method of launch vehicle aerodynamic damping relies on sinusoidal forced motion training, which results in high computational cost and makes it difficult to meet the generalized aerodynamic damping assessment requirements of launch vehicles. It also fails to effectively quantify the contribution of local unsteady flow phenomena to aerodynamic damping characteristics.
By employing convolutional autoencoders and long short-term memory neural networks, and constructing an unsteady flow field time series dataset, the convolutional autoencoders and long short-term memory neural networks are trained to achieve dimensionality reduction of high-dimensional flow fields and prediction of potential flow field codes. This enables the rapid acquisition of aerodynamic damping distributions with low computational cost.
It enables rapid and accurate prediction of the aerodynamic damping distribution of launch vehicles, reduces computational costs, adapts to excitation signals of different structural frequencies, improves the generalization ability and iterative efficiency of aerodynamic damping assessment, and supports aeroelastic stability analysis in the transonic band.
Smart Images
Figure CN122287392B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aeroelastic instability prediction technology for launch vehicles, and in particular to a method, apparatus, equipment, and medium for predicting aerodynamic damping distribution of launch vehicles. Background Technology
[0002] Launch vehicles face the risk of aeroelastic instability during transonic flight. For the aeroelastic system of a launch vehicle, when the aerodynamic damping is negative, the system is unstable, i.e., aeroelastic instability will occur. Therefore, accurate calculation and prediction of the aerodynamic damping of the launch vehicle are necessary to avoid aeroelastic instability and ultimately structural failure. 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, both of which may contribute to aeroelastic stability.
[0003] Currently, to assess the contribution of local unsteady flows (such as shock wave oscillations and flow separation) to aerodynamic damping, researchers have proposed a distributed aerodynamic damping prediction method. This method uses a given sinusoidal forced motion signal and unsteady CFD (Computational Fluid Dynamics) simulation to obtain a time-series distribution of unsteady aerodynamic loads. Then, based on the principle of aerodynamic virtual work or modal coordinate transformation, it achieves a quantitative assessment of the spatial distribution of local aerodynamic damping characteristics.
[0004] The distributed aerodynamic damping prediction method relies on sinusoidal forced motion training to obtain the distribution of unsteady aerodynamic loads. However, forced motion can only simulate a single frequency signal. When it is necessary to change the structural frequency to predict aerodynamic damping, time-consuming unsteady CFD simulation and model training must be carried out again. This not only leads to extremely high computational costs, but also makes it difficult to meet the generalized aerodynamic damping evaluation requirements of launch vehicles, thus limiting the efficiency of aeroelastic design iteration. Summary of the Invention
[0005] Based on this, it is necessary to provide a method, device, equipment, and medium for predicting the aerodynamic damping distribution of launch vehicles to address the aforementioned technical problems. This would solve the issues in existing technologies, such as the unclear contribution of local flow phenomena to aerodynamic damping characteristics and the high time cost of forced motion training in obtaining damping distribution patterns. The goal is to achieve more generalized and accurate prediction of aerodynamic damping distribution of launch vehicles with lower computational costs.
[0006] The following technical solution is adopted in this specification: This specification provides a method for predicting the aerodynamic damping distribution of a launch vehicle, including: Construct a time-series dataset of unsteady flow fields, including high-dimensional flow field snapshots and corresponding forced motion excitation signals; A convolutional autoencoder and a long short-term memory neural network were constructed, and trained on an unsteady flow field time-series dataset, including: A high-dimensional flow field snapshot is input into the encoder of a convolutional autoencoder to generate a latent flow field code; the latent flow field code is input into the decoder of the convolutional autoencoder for decoding to obtain a reconstructed flow field; the convolutional autoencoder is trained by minimizing the difference between the high-dimensional flow field snapshot and the reconstructed flow field. The parameters of the convolutional autoencoder are frozen, and the forced motion excitation signal corresponding to the high-dimensional flow field snapshot is input into the long short-term memory neural network to generate the latent code of the predicted flow field. The latent code of the predicted flow field is input into the decoder of the convolutional autoencoder for decoding to generate the predicted reconstructed flow field. The long short-term memory neural network is trained to minimize the difference between the predicted reconstructed flow field and the reconstructed flow field. The forced motion excitation signal of the target to be predicted is input into the trained long short-term memory neural network to obtain the target prediction flow field latent code; the target prediction flow field latent code is input into the decoder in the trained convolutional autoencoder for decoding to obtain the target prediction reconstructed flow field; the aerodynamic damping distribution prediction result of the launch vehicle is obtained based on the target prediction reconstructed flow field.
[0007] Furthermore, the frequency of the target forced motion excitation signal to be predicted is any frequency within the frequency band of interest for launch vehicle flutter analysis, including new frequencies not covered in the unsteady flow field time series data. The method for obtaining the predicted aerodynamic damping distribution of the launch vehicle based on the target prediction and reconstructed flow field specifically includes: Obtain the structural modal parameters of the target launch vehicle; Based on the target prediction, the reconstructed flow field is used to extract the wall aerodynamic load distribution of the target launch vehicle. The wall aerodynamic load distribution is then coupled with the structural modal parameters of the target launch vehicle to obtain the predicted aerodynamic damping distribution of the target launch vehicle.
[0008] Furthermore, the construction of the unsteady flow field time series dataset specifically includes: A geometric model was established based on the aerodynamic shape of the launch vehicle and a fluid dynamics mesh was generated. At the same time, the modal shapes, natural frequencies and generalized mass of the launch vehicle were extracted through structural finite element analysis. Configure the numerical simulation environment for unsteady flow fields: set the numerical calculation method, turbulence model, spatial discretization scheme and boundary conditions, and set the flow condition parameters according to the flight conditions; Several sets of forced motion excitation training signals are set up to simulate the rigid body motion excitation of the rocket body in aeroelastic vibration. Numerical simulation of unsteady flow field is carried out based on several sets of forced motion excitation training signals: the unsteady evolution of the flow field is driven sequentially based on each set of forced motion excitation training signals, and high-dimensional flow field snapshots and wall aerodynamic load distributions are collected at each time step to obtain unsteady flow field time series dataset.
[0009] Furthermore, the convolutional autoencoder includes a symmetrically arranged encoder and decoder, wherein the encoder includes multiple convolutional layers and pooling layers for nonlinearly reducing the high-dimensional flow field snapshot to the flow field latent code; the decoder includes multiple deconvolutional layers and upsampling layers for decoding the flow field latent code into a reconstructed flow field.
[0010] Furthermore, the long short-term memory neural network includes an input layer, a hidden layer, and an output layer. The hidden layer is configured with a gating mechanism to learn the time dependency of the forced motion excitation signal and establish a nonlinear mapping from the forced motion excitation signal to the latent encoding of the predicted flow field.
[0011] Furthermore, the step of training a long short-term memory neural network to minimize the difference between the predicted and reconstructed flow fields is implemented based on a time-sliding window method, specifically including: Set the time sliding window length and sliding step size, and divide the time series of the forced motion excitation signal into several input samples based on the time sliding window length and sliding step size; For each moment, the historical forced motion excitation signal within the time sliding window length before the current moment is extracted as the input feature of the long short-term memory neural network, and the predicted reconstructed flow field at the corresponding moment is used as the supervision label. The historical forced motion excitation signal is input into the long short-term memory neural network. The time dependency is learned through the hidden layer gating mechanism of the long short-term memory neural network, and the predicted flow field potential code at the current moment is output. The decoder of the convolutional autoencoder that inputs the latent code of the predicted flow field into the frozen parameters generates the predicted reconstructed flow field at the current time. The mean squared error loss between the predicted reconstructed flow field and the actual flow field reconstruction results is calculated, and the weight parameters of the long short-term memory neural network are updated through the backpropagation algorithm until convergence.
[0012] Furthermore, the step of training the convolutional autoencoder by minimizing the difference between the high-dimensional flow field snapshot and the reconstructed flow field is to minimize the mean square error between the high-dimensional flow field snapshot and the reconstructed flow field as the optimization objective, and to update the weight parameters of the convolutional autoencoder through the backpropagation algorithm until convergence.
[0013] This specification provides a device for predicting the aerodynamic damping distribution of a launch vehicle, including: The dataset construction module is used to construct unsteady flow field time series datasets, including high-dimensional flow field snapshots and corresponding forced motion excitation signals; The model training module is used to construct a convolutional autoencoder and a long short-term memory neural network. It trains the convolutional autoencoder and the long short-term memory neural network based on an unsteady flow field time-series dataset, including: A high-dimensional flow field snapshot is input into the encoder of a convolutional autoencoder to generate a latent flow field code; the latent flow field code is input into the decoder of the convolutional autoencoder for decoding to obtain a reconstructed flow field; the convolutional autoencoder is trained by minimizing the difference between the high-dimensional flow field snapshot and the reconstructed flow field. The parameters of the convolutional autoencoder are frozen, and the forced motion excitation signal corresponding to the high-dimensional flow field snapshot is input into the long short-term memory neural network to generate the latent code of the predicted flow field. The latent code of the predicted flow field is input into the decoder of the convolutional autoencoder for decoding to generate the predicted reconstructed flow field. The long short-term memory neural network is trained to minimize the difference between the predicted reconstructed flow field and the reconstructed flow field. The aerodynamic damping distribution prediction module is used to input the forced motion excitation signal of the target to be predicted into the trained long short-term memory neural network to obtain the target prediction flow field latent code; input the target prediction flow field latent code into the decoder in the trained convolutional autoencoder for decoding to obtain the target prediction reconstructed flow field; and obtain the aerodynamic damping distribution prediction result of the launch vehicle based on the target prediction reconstructed flow field.
[0014] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting the aerodynamic damping distribution of a launch vehicle.
[0015] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for predicting the aerodynamic damping distribution of a launch vehicle.
[0016] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: This invention generates a low-dimensional flow field latent code by inputting a high-dimensional flow field snapshot into the encoder of a convolutional autoencoder. The convolutional autoencoder is then trained with the goal of minimizing the difference between the high-dimensional flow field snapshot and the reconstructed flow field. This enables the autoencoder to accurately extract core features of the high-dimensional flow field and complete flow field encoding and decoding, achieving efficient dimensionality reduction and feature preservation of high-dimensional flow field information without the need for repeated, costly unsteady CFD simulations for different structure frequencies to generate specific flow field data. The convolutional autoencoder parameters are frozen, and a long short-term memory neural network is trained using a forced motion excitation signal as input. This allows the network to learn the temporal correlation between the excitation signal and the flow field latent code. The model is then optimized with the goal of minimizing the difference between the predicted and reconstructed flow field. Long Short-Term Memory (LSTM) neural networks can accurately predict the potential codes of the flow field corresponding to different excitation signals and adapt to forced motion excitation signals corresponding to different structural frequencies. There is no need to retrain the time-consuming model to change the structural frequency. The aerodynamic damping distribution prediction model that has been trained can directly complete the prediction of the potential codes of the flow field and the reconstruction of the flow field based on the input excitation signal, and then carry out the prediction of the aerodynamic damping distribution of the launch vehicle. The entire process does not require repeated high-cost unsteady CFD simulations and model retraining, which greatly reduces the computational cost. At the same time, the model can adapt to excitation signals corresponding to different structural frequencies, which effectively improves the evaluation capability of generalized aerodynamic damping of launch vehicles and meets the engineering requirements for rapid iterative evaluation of aerodynamic damping distribution. Attached Figure Description
[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is one of the flowcharts illustrating a method for predicting the aerodynamic damping distribution of a launch vehicle provided in this specification; Figure 2 This is the second flowchart illustrating a method for predicting the aerodynamic damping distribution of a launch vehicle, as provided in this specification. Figure 3 A schematic diagram of a launch vehicle aerodynamic damping distribution prediction device provided in this specification; Figure 4 This is a schematic diagram of a computer device provided for this specification. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.
[0019] The technical solution provided by this invention can be applied to the aeroelastic stability analysis and design scenarios of launch vehicles in the transonic flight phase. Addressing the difficulties in quantifying and identifying the contribution of local unsteady flow phenomena (such as shock wave oscillations and flow separation) to aerodynamic damping characteristics in existing technologies, and the high time cost resulting from traditional distributed aerodynamic damping prediction methods relying on sinusoidal forced motion training at specific frequencies, this invention provides a rapid prediction method for distributed aerodynamic damping of launch vehicles based on CAE (Convolutional Auto Encoder) and LSTM (Long Short-Term Memory Neural Network). This method is based on unsteady flow field reconstruction technology, establishing a nonlinear mapping relationship between motion excitation signals and the evolution of potential flow field features and aerodynamic load distribution by constructing a cascaded convolutional autoencoder and long short-term memory neural network model. Using this model, only a sinusoidal forced motion signal of any target frequency needs to be input to quickly predict the corresponding unsteady flow field structural evolution and the temporal distribution of wall aerodynamic loads. Furthermore, through coupled analysis with structural mode shapes, the spatial distribution of aerodynamic damping along the rocket body surface at that frequency can be accurately extracted. Compared to traditional methods that require repeated, time-consuming high-fidelity unsteady numerical simulations at different structural natural frequencies, this invention achieves rapid acquisition of aerodynamic damping distribution at any frequency under single-degree-of-freedom vibration conditions. It can effectively quantify the contribution of unsteady flow phenomena on the launch vehicle surface to aeroelastic stability, providing theoretical guidance for improving aerodynamic shape and enhancing flow control to increase aeroelastic stability margins. Moreover, it can quickly acquire the unsteady load distribution under sinusoidal forced motion signals of any frequency, thereby rapidly obtaining the aerodynamic damping distribution corresponding to different structural natural frequencies. This improves the efficiency of aeroelastic stability analysis of launch vehicles and provides a reliable technical approach for evaluating transonic aerodynamic damping characteristics and predicting flutter boundaries.
[0020] The following is combined Figures 1-2 The present invention describes a method for predicting the aerodynamic damping distribution of a launch vehicle.
[0021] Figure 1 This is one of the flowcharts illustrating a method for predicting the aerodynamic damping distribution of a launch vehicle provided in this specification, such as... Figure 1 As shown, the method includes: S1. Construct an unsteady flow field time series dataset, including high-dimensional flow field snapshots and corresponding forced motion excitation signals.
[0022] S2. Construct a convolutional autoencoder and a long short-term memory neural network. Train the convolutional autoencoder and the long short-term memory neural network based on an unsteady flow field time-series dataset, including: A high-dimensional flow field snapshot is input into the encoder of a convolutional autoencoder to generate a latent flow field code; the latent flow field code is input into the decoder of the convolutional autoencoder for decoding to obtain a reconstructed flow field; the convolutional autoencoder is trained by minimizing the difference between the high-dimensional flow field snapshot and the reconstructed flow field. The parameters of the convolutional autoencoder are frozen, and the forced motion excitation signal corresponding to the high-dimensional flow field snapshot is input into the long short-term memory neural network to generate the latent code of the predicted flow field. The latent code of the predicted flow field is input into the decoder of the convolutional autoencoder for decoding to generate the predicted reconstructed flow field. The long short-term memory neural network is trained to minimize the difference between the predicted reconstructed flow field and the reconstructed flow field.
[0023] S3. Input the excitation signal of the forced motion of the target to be predicted into the trained long short-term memory neural network to obtain the latent code of the target prediction flow field; input the latent code of the target prediction flow field into the decoder in the trained convolutional autoencoder for decoding to obtain the target prediction reconstructed flow field; obtain the prediction result of the aerodynamic damping distribution of the launch vehicle based on the target prediction reconstructed flow field.
[0024] Based on the above, Figure 1 In the illustrated embodiment, for example, step S1 above acquires and stores the unsteady numerical simulation flow field data of the launch vehicle. First, the aerodynamic shape and dimensions of the launch vehicle are defined, and a mesh is generated based on the aerodynamic shape. Mode shapes and corresponding parameters are extracted using finite element method (FEM) technology. Solver parameters are configured, and boundary conditions, physical models, and numerical algorithms for the numerical simulation are set, along with different forced motion signals. Finally, the simulation-derived flow field snapshots and response results are stored in chronological order. This may include the following steps: S11: A geometric model is established based on the aerodynamic shape of the launch vehicle, and a hydrodynamic mesh is generated. Simultaneously, the modal shapes, natural frequencies, and generalized mass of the launch vehicle are extracted through structural finite element analysis. Specifically, the aerodynamic shape and dimensions of the launch vehicle are defined. A multi-region structured / unstructured hybrid mesh is generated for the aerodynamic shape, with a boundary layer mesh set near the wall to generate a computational domain containing multiple mesh elements. Simultaneously, a finite element model of the rocket body structure is established using finite element technology, and the elastic modal shapes and corresponding natural frequencies and generalized mass of the launch vehicle are extracted as structural dynamic input parameters for subsequent aeroelastic coupling analysis.
[0025] S12: Configure the unsteady flow field numerical simulation environment: Set the numerical calculation method, turbulence model, spatial discretization scheme, and boundary conditions, and set the flow condition parameters according to the flight conditions. Specifically, set the calculation method, turbulence model, numerical scheme, and boundary conditions of the unsteady flow solver; configure the flow conditions in the solver, including the incoming Mach number, Reynolds number, and angle of attack. The specific data of the above numerical simulation environment can be set according to the actual situation, and this embodiment of the application does not impose any limitations.
[0026] S13: Several sets of forced motion excitation training signals are set up to simulate the rigid body motion excitation of the rocket body in aeroelastic vibration; unsteady flow field numerical simulation is carried out based on several sets of forced motion excitation training signals: the unsteady evolution of the flow field is driven sequentially based on each set of forced motion excitation training signals, and high-dimensional flow field snapshots and wall aerodynamic load distributions are collected at each time step to obtain an unsteady flow field time series dataset. Specifically, several sets of forced motion training signals are set up for unsteady solution. Dynamic mesh technology is used to handle the boundary motion of the object surface in the simulation. The unsteady flow field (including high-dimensional snapshot data of pressure, velocity, and density fields, with the dimension of a single frame data being the number of mesh nodes × the number of physical quantities) and the wall aerodynamic load distribution (wall pressure coefficient distribution) at each unsteady time step obtained in the simulation are saved in chronological order to construct an unsteady flow field time series dataset containing the input excitation-output response mapping relationship.
[0027] Based on any of the above embodiments, for example, in step S2, the convolutional autoencoder includes a symmetrically arranged encoder and decoder. The encoder includes multiple convolutional layers and pooling layers to nonlinearly reduce the high-dimensional flow field snapshot to a latent flow field code. The decoder includes multiple deconvolutional layers and upsampling layers to decode the latent flow field code into a reconstructed flow field. The long short-term memory neural network includes an input layer, a hidden layer, and an output layer. The hidden layer is configured with a gating mechanism to learn the temporal dependency of the forced motion excitation signal, establishing a nonlinear mapping from the forced motion excitation signal to the predicted latent flow field code.
[0028] Based on any of the above embodiments, for example, step S2 employs a flow field reconstruction method to achieve unsteady flow field modeling: A CAE model is trained to perform nonlinear dimensionality reduction on existing flow field snapshots to obtain the latent codes corresponding to the flow fields at different time steps. An LSTM model is trained in the encoded latent space to establish a nonlinear dynamic mapping from motion signals to latent codes. This completes the construction of unsteady flow field reconstruction modeling. This may include the following steps: S21: Using the unsteady flow field snapshot and aerodynamic load distribution saved in step S13 as input samples, the CAE model is trained to perform nonlinear dimensionality reduction on the existing flow field snapshot and aerodynamic load distribution to obtain the latent encoding corresponding to the flow field at different time steps. The encoder extracts spatial features step by step through convolutional layers and pooling layers, compressing the original flow field data into a low-dimensional encoding vector in the latent space; the decoder reconstructs the flow field through deconvolutional layers and upsampling layers. During training, the mean squared error loss function is used, and the network parameters are optimized through backpropagation until the error between the reconstructed flow field and the original flow field is less than a set threshold.
[0029] S22: Train an LSTM model in the encoded latent space to establish a nonlinear dynamic mapping from motion signals to latent codes. Construct an LSTM neural network whose input layer receives sequences of motion parameters from the current and historical time steps, whose hidden layers learn time dependencies through gating mechanisms, and whose output layer predicts the latent code vector for the next time step. Use a time-sliding window method to organize the training data, optimizing the LSTM network weights by minimizing the error between the predicted and actual codes. The time-sliding window method includes: The time-sliding window length and sliding step size are set, and the time series of the forced motion excitation signal is divided into several input samples based on the time-sliding window length and sliding step size. For each time step, the historical forced motion excitation signal within the time-sliding window length before the current time step is extracted as the input feature of the long short-term memory neural network, and the predicted reconstructed flow field at the corresponding time step is used as the supervision label. The historical forced motion excitation signal is input into the long short-term memory neural network, and the time dependency is learned through the hidden layer gating mechanism of the long short-term memory neural network to output the latent code of the predicted flow field at the current time step. The latent code of the predicted flow field is input into the decoder of the convolutional autoencoder with frozen parameters to generate the predicted reconstructed flow field at the current time step. The mean squared error loss between the predicted reconstructed flow field and the actual flow field reconstruction results is calculated, and the weight parameters of the long short-term memory neural network are updated through the backpropagation algorithm until convergence.
[0030] Based on any of the above embodiments, for example, step S3 above uses a damping distribution analysis method to predict the aerodynamic damping distribution of the launch vehicle: A sinusoidal signal of arbitrary frequency is provided, and an LSTM model is used for prediction to obtain the evolution of the potential encoding time dimension of the current signal. Using the decoding part of CAE, the potential encoding corresponding to the current time step is decoded to obtain the flow field and unsteady load distribution of the current time step. Based on the obtained unsteady load distribution, it is coupled with the structural mode shape to realize the prediction of the aerodynamic damping distribution of the launch vehicle. This may include the following steps: S31: Obtain the structural modal parameters of the target launch vehicle and the forced motion excitation signal of the target frequency. The target frequency is any frequency within the frequency band of interest for launch vehicle flutter analysis, including new frequencies not covered by training. Specifically, provide a sinusoidal signal of any fixed frequency containing several cycles. Using the LSTM model trained in S2, obtain the target predicted flow field potential code corresponding to the sinusoidal signal that evolves in time sequence.
[0031] S32: Using the decoding module of the CAE model trained in S21, the obtained encoding is decoded to obtain the target predicted reconstructed flow field corresponding to each unsteady time step of the sinusoidal signal in S31. Each target predicted reconstructed flow field corresponds to an unsteady flow field information. After training, this model can be used to achieve rapid aerodynamic damping distribution prediction at any frequency. Specifically, based on the predicted reconstructed unsteady flow field time series distribution, the launch vehicle wall pressure time series distribution is extracted. The extraction process is automatically executed by the post-processing module. According to the pre-stored wall mesh topology index, the boundary node pressure values are selected from the reconstructed flow field volume data and organized into a structured array according to the time series. By performing surface integration on the wall pressure time series distribution, the unsteady aerodynamic load time series distribution is generated.
[0032] S33: Based on the target prediction and reconstructed flow field, the wall aerodynamic load distribution is extracted. The wall aerodynamic load distribution is coupled with the structural modal parameters of the target launch vehicle, and a distributed aerodynamic damping prediction method is used to predict the aerodynamic damping distribution of the launch vehicle. Specifically, the time-series distribution of unsteady aerodynamic loads is aeroelastically coupled with the structural modal parameters of the launch vehicle. Based on the principle of aerodynamic virtual work, the virtual work done by the unsteady aerodynamic forces on the generalized coordinates of each mode within a complete vibration cycle is calculated. Based on the virtual work calculation results, the aerodynamic damping coefficients corresponding to each mode are determined, and the spatial distribution characteristics of the aerodynamic damping coefficients along the surface of the launch vehicle are analyzed.
[0033] Based on any of the above embodiments, for example, Figure 2 This is the second flowchart illustrating a method for predicting the aerodynamic damping distribution of a launch vehicle, as provided in this specification. Figure 2As shown, the above method is divided into two stages: model training and aerodynamic damping prediction. In the training stage, a snapshot dataset of unsteady aerodynamic force distribution is first generated using multiple sets of random training signals. This dataset is then used to train a convolutional neural network to obtain a convolutional autoencoder, and a latent coding dataset is generated. Subsequently, the latent coding dataset and multiple sets of random training signals are input together to train a long short-term memory neural network to obtain a long short-term memory neural network model. In the prediction stage, a sinusoidal motion signal is input to the trained long short-term memory neural network model to generate a latent coding of the target signal. This latent coding is decoded by a convolutional autoencoder to obtain the unsteady aerodynamic force distribution, and finally, the aerodynamic damping distribution is obtained by processing it using an aerodynamic damping distribution method.
[0034] The aerodynamic damping distribution prediction device for launch vehicles provided by the present invention is described below. The aerodynamic damping distribution prediction device for launch vehicles described below can be referred to in correspondence with the aerodynamic damping distribution prediction method for launch vehicles described above.
[0035] Figure 3 This specification provides a schematic diagram of the structure of a launch vehicle aerodynamic damping distribution prediction device. For an example, please refer to [link to schematic diagram]. Figure 3 As shown, the launch vehicle aerodynamic damping distribution prediction device may include: The dataset construction module is used to build unsteady flow field time series datasets, including high-dimensional flow field snapshots and corresponding forced motion excitation signals.
[0036] The model training module is used to construct a convolutional autoencoder and a long short-term memory neural network. It trains the convolutional autoencoder and the long short-term memory neural network based on an unsteady flow field time-series dataset, including: A high-dimensional flow field snapshot is input into the encoder of a convolutional autoencoder to generate a latent flow field code; the latent flow field code is input into the decoder of the convolutional autoencoder for decoding to obtain a reconstructed flow field; the convolutional autoencoder is trained by minimizing the difference between the high-dimensional flow field snapshot and the reconstructed flow field. The parameters of the convolutional autoencoder are frozen, and the forced motion excitation signal corresponding to the high-dimensional flow field snapshot is input into the long short-term memory neural network to generate the latent code of the predicted flow field. The latent code of the predicted flow field is input into the decoder of the convolutional autoencoder for decoding to generate the predicted reconstructed flow field. The long short-term memory neural network is trained to minimize the difference between the predicted reconstructed flow field and the reconstructed flow field.
[0037] The aerodynamic damping distribution prediction module is used to input the forced motion excitation signal of the target to be predicted into the trained long short-term memory neural network to obtain the target prediction flow field latent code; input the target prediction flow field latent code into the decoder in the trained convolutional autoencoder for decoding to obtain the target prediction reconstructed flow field; and obtain the aerodynamic damping distribution prediction result of the launch vehicle based on the target prediction reconstructed flow field.
[0038] Specific limitations regarding the launch vehicle aerodynamic damping distribution prediction device can be found in the above-mentioned limitations on launch vehicle aerodynamic damping distribution prediction, and will not be repeated here. Each module in the aforementioned launch vehicle aerodynamic damping distribution prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.
[0039] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 A method for predicting the aerodynamic damping distribution of launch vehicles is provided.
[0040] This instruction manual also provides Figure 4 The schematic diagram of the computer device shown is as follows: Figure 4 At the hardware level, the computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 A method for predicting the aerodynamic damping distribution of launch vehicles is provided.
[0041] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.
[0042] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
Claims
1. A method for predicting the aerodynamic damping distribution of a launch vehicle, characterized in that, include: Construct a time-series dataset of unsteady flow fields, including high-dimensional flow field snapshots and corresponding forced motion excitation signals; A convolutional autoencoder and a long short-term memory neural network were constructed, and trained on an unsteady flow field time-series dataset, including: A high-dimensional flow field snapshot is input into the encoder of a convolutional autoencoder to generate a latent flow field code; the latent flow field code is input into the decoder of the convolutional autoencoder for decoding to obtain a reconstructed flow field; the convolutional autoencoder is trained by minimizing the difference between the high-dimensional flow field snapshot and the reconstructed flow field. The parameters of the convolutional autoencoder are frozen, and the forced motion excitation signal corresponding to the high-dimensional flow field snapshot is input into the long short-term memory neural network to generate the latent code of the predicted flow field. The latent code of the predicted flow field is input into the decoder of the convolutional autoencoder for decoding to generate the predicted reconstructed flow field. The long short-term memory neural network is trained to minimize the difference between the predicted reconstructed flow field and the reconstructed flow field. The forced motion excitation signal of the target to be predicted is input into the trained long short-term memory neural network to obtain the target prediction flow field latent code; the target prediction flow field latent code is input into the decoder in the trained convolutional autoencoder for decoding to obtain the target prediction reconstructed flow field; the aerodynamic damping distribution prediction result of the launch vehicle is obtained based on the target prediction reconstructed flow field.
2. The launch vehicle aerodynamic damping distribution prediction method of claim 1, wherein, The frequency of the target forced motion excitation signal to be predicted is any frequency within the frequency band of interest for launch vehicle flutter analysis, including new frequencies not covered in the unsteady flow field time series data. The method for obtaining the predicted aerodynamic damping distribution of the launch vehicle based on the target prediction and reconstructed flow field specifically includes: Obtain the structural modal parameters of the target launch vehicle; Based on the target prediction, the reconstructed flow field is used to extract the wall aerodynamic load distribution of the target launch vehicle. The wall aerodynamic load distribution is then coupled with the structural modal parameters of the target launch vehicle to obtain the predicted aerodynamic damping distribution of the target launch vehicle.
3. The method for predicting the aerodynamic damping distribution of a launch vehicle according to claim 1, characterized in that, The construction of the unsteady flow field time series dataset specifically includes: A geometric model was established based on the aerodynamic shape of the launch vehicle and a fluid dynamics mesh was generated. At the same time, the modal shapes, natural frequencies and generalized mass of the launch vehicle were extracted through structural finite element analysis. Configure the numerical simulation environment for unsteady flow fields: set the numerical calculation method, turbulence model, spatial discretization scheme and boundary conditions, and set the flow condition parameters according to the flight conditions; Several sets of forced motion excitation training signals are set up to simulate the rigid body motion excitation of the rocket body in aeroelastic vibration. Numerical simulation of unsteady flow field is carried out based on several sets of forced motion excitation training signals: the unsteady evolution of the flow field is driven sequentially based on each set of forced motion excitation training signals, and high-dimensional flow field snapshots and wall aerodynamic load distributions are collected at each time step to obtain unsteady flow field time series dataset.
4. The method for predicting the aerodynamic damping distribution of a launch vehicle according to claim 1, characterized in that, The convolutional autoencoder includes a symmetrically arranged encoder and decoder. The encoder contains multiple convolutional layers and pooling layers to nonlinearly reduce the high-dimensional flow field snapshot to the latent flow field code. The decoder contains multiple deconvolutional layers and upsampling layers to decode the latent flow field code into a reconstructed flow field.
5. The method for predicting the aerodynamic damping distribution of a launch vehicle according to claim 1, characterized in that, The long short-term memory neural network includes an input layer, a hidden layer, and an output layer. The hidden layer is configured with a gating mechanism to learn the time dependence of the forced motion excitation signal and establish a nonlinear mapping from the forced motion excitation signal to the latent encoding of the predicted flow field.
6. The method for predicting the aerodynamic damping distribution of a launch vehicle according to claim 5, characterized in that, The method of training a long short-term memory neural network to minimize the difference between the predicted and reconstructed flow fields is based on a time-sliding window method, specifically including: Set the time sliding window length and sliding step size, and divide the time series of the forced motion excitation signal into several input samples based on the time sliding window length and sliding step size; For each moment, the historical forced motion excitation signal within the time sliding window length before the current moment is extracted as the input feature of the long short-term memory neural network, and the predicted reconstructed flow field at the corresponding moment is used as the supervision label. The historical forced motion excitation signal is input into the long short-term memory neural network. The time dependency is learned through the hidden layer gating mechanism of the long short-term memory neural network, and the predicted flow field potential code at the current moment is output. The decoder of the convolutional autoencoder that inputs the latent code of the predicted flow field into the frozen parameters generates the predicted reconstructed flow field at the current time. The mean squared error loss between the predicted reconstructed flow field and the actual flow field reconstruction results is calculated, and the weight parameters of the long short-term memory neural network are updated through the backpropagation algorithm until convergence.
7. The method for predicting the aerodynamic damping distribution of a launch vehicle according to claim 1, characterized in that, The method of training a convolutional autoencoder by minimizing the difference between the high-dimensional flow field snapshot and the reconstructed flow field is to take minimizing the mean square error between the high-dimensional flow field snapshot and the reconstructed flow field as the optimization objective, and update the weight parameters of the convolutional autoencoder through the backpropagation algorithm until convergence.
8. A device for predicting the aerodynamic damping distribution of a launch vehicle, characterized in that, include: The dataset construction module is used to construct unsteady flow field time series datasets, including high-dimensional flow field snapshots and corresponding forced motion excitation signals; The model training module is used to construct a convolutional autoencoder and a long short-term memory neural network. It trains the convolutional autoencoder and the long short-term memory neural network based on an unsteady flow field time-series dataset, including: A high-dimensional flow field snapshot is input into the encoder of a convolutional autoencoder to generate a latent flow field code; the latent flow field code is input into the decoder of the convolutional autoencoder for decoding to obtain a reconstructed flow field; the convolutional autoencoder is trained by minimizing the difference between the high-dimensional flow field snapshot and the reconstructed flow field. The parameters of the convolutional autoencoder are frozen, and the forced motion excitation signal corresponding to the high-dimensional flow field snapshot is input into the long short-term memory neural network to generate the latent code of the predicted flow field. The latent code of the predicted flow field is input into the decoder of the convolutional autoencoder for decoding to generate the predicted reconstructed flow field. The long short-term memory neural network is trained to minimize the difference between the predicted reconstructed flow field and the reconstructed flow field. The aerodynamic damping distribution prediction module is used to input the forced motion excitation signal of the target to be predicted into the trained long short-term memory neural network to obtain the target prediction flow field latent code; input the target prediction flow field latent code into the decoder in the trained convolutional autoencoder for decoding to obtain the target prediction reconstructed flow field; and obtain the aerodynamic damping distribution prediction result of the launch vehicle based on the target prediction reconstructed flow field.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for predicting the aerodynamic damping distribution of a launch vehicle as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for predicting the aerodynamic damping distribution of a launch vehicle as described in any one of claims 1 to 7.