Rocket aircraft outflow real-time reconstruction method based on multi-source aerodynamic data fusion

Abstract

The invention discloses a rocket aircraft outflow real-time reconstruction method based on multi-source aerodynamic data fusion, and relates to the technical field of rocket aerodynamic tests, comprising: constructing a three-dimensional numerical simulation model of a rocket aircraft; cFD calculation software is used for conducting simulation calculation on the outflow flow field of the rocket aircraft under the specified working condition, and physical information of grid nodes at equal intervals is extracted; after sufficient CFD aerodynamic data is obtained, training a low-precision agent model; and performing secondary modeling based on real rocket flight test data and the characteristics of the prediction result of the low-precision agent model to obtain a high-precision agent model capable of quickly predicting real aerodynamic data, thereby realizing real-time reconstruction of the outflow of the rocket aircraft in the test process. According to the method, the prediction precision is effectively improved, the calculation time and cost are remarkably reduced, efficient real-time reconstruction is achieved, and meanwhile rapid evaluation and real-time feedback of the aerodynamic performance of the rocket in the initial stage of design are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of rocket aerodynamic testing technology, and in particular to a real-time reconfiguration method for rocket outflow based on multi-source aerodynamic data fusion. Background Technology

[0002] With the rapid development of aerospace technology, rocket testing methods are becoming increasingly intelligent and diversified. In this process, the accurate and rapid acquisition of aerodynamic data is crucial. However, current technologies primarily rely on ground-based wind tunnel tests and numerical simulations for aerodynamic data acquisition. Ground-based wind tunnel tests are costly, time-consuming, and unable to cover aerodynamic characteristics under all flight conditions. On the other hand, while traditional numerical simulation methods can simulate external flow fields to some extent, they are computationally expensive and time-consuming, making it difficult to meet real-time requirements.

[0003] In response to the above issues, scholars both at home and abroad have conducted extensive research:

[0004] First, Renganathan et al. proposed improving the accuracy of aerodynamic prediction through multi-fidelity data fusion and Bayesian inference (Renganathan, S. Ashwin, Harada, Kohei, and Mavris, Dimitri N. Aerodynamic Data Fusion Toward the Digital Twin Paradigm. United States: Np, 2020. Web.doi:10.2514 / 1.j059203). This multi-fidelity data fusion method effectively reduces computational costs while improving prediction efficiency. Bekemeyer et al. from the German Aerospace Center applied digital twin technology to the simulation of dynamic and complex flow fields, proposing a framework for real-time flow field prediction based on data-driven models and uncertainty management (Bekemeyer, C., & S.(2023). Data-driven aerodynamic modeling: Toward adigital twin for aerospace applications. Aerospace Science and Technology, 127, 107858. doi:10.1016 / j.ast.2023.107858.), by fusing real flight data and CFD simulation results to construct a high-precision surrogate model, the aerodynamic performance of rocket vehicles can be rapidly and reliably evaluated in the early stages of design; U. High-dimensional rocket flow field data is processed using convolutional neural networks (CNNs) to extract important physical features (U) in the flow field. and MUDemirezen, "Optimal Reusable Rocket Landing Guidance: ACutting-Edge Approach Integrating Scientific Machine Learning and Enhanced Neural Networks," in IEEE Access, vol. 12, pp. 16805-16829, 2024, doi:10.1109 / ACCESS.2024.3359417). This method integrates CFD data with actual flight data to construct a surrogate model that can adapt to various flight conditions. It can predict rocket aerodynamic changes in complex aerodynamic environments in real time, thereby improving data processing efficiency and prediction accuracy.

[0005] It can be seen that existing rocket aerodynamic testing methods mainly focus on the application of a single data source or the construction of a single-precision surrogate model. Summary of the Invention

[0006] This invention provides a real-time reconstruction method for the external flow field of a rocket based on multi-source aerodynamic data fusion. By fusing CFD calculation data and real experimental data, a low-precision surrogate model and a high-precision surrogate model are constructed, realizing real-time reconstruction of the external flow field of the rocket, improving prediction accuracy, significantly reducing computation time and cost, and accelerating the design and performance optimization process of the rocket.

[0007] The present invention adopts the following technical solution:

[0008] A method for real-time reconstructing of rocket outflow based on multi-source aerodynamic data fusion includes:

[0009] S1. Construct a three-dimensional numerical simulation model of the rocket, and define the rocket's geometric characteristics and physical boundary conditions;

[0010] S2 sets the simulated flight conditions and determines the flow characteristics under the defined geometric features and physical boundary conditions;

[0011] S3, based on flow characteristics, performs three-dimensional numerical simulation model mesh generation, and uses CFD to solve fluid dynamics equations to generate rocket external flow field data under specified conditions;

[0012] S4. Extract equally spaced grid node information from the CFD simulation of the rocket's external flow field to obtain the physical quantity distribution; construct a simulation dataset based on the physical quantity distribution for the initial training of the low-precision surrogate model;

[0013] S5 uses a radial basis function-based interpolation method to construct a low-precision surrogate model and uses the distribution of physical quantities in the simulation dataset to predict coarse aerodynamic distributions.

[0014] S6. Construct a high-precision proxy model. Based on the real test data of the rocket, use a multi-fidelity data fusion method to combine the prediction results of the low-precision proxy model with the real test data, and perform a weighted average through the data source weights to generate a weighted fusion prediction result.

[0015] S7, based on the Bayesian inference principle and the prediction results after weighted fusion, dynamically adjusts the data source weights of the high-precision surrogate model and optimizes the parameters of the high-precision surrogate model through maximum a posteriori probability estimation;

[0016] S8 compares the physical quantity distribution output by the high-precision surrogate model with the real experimental data. When the error is higher than the preset value, the parameters of the high-precision surrogate model are continuously adjusted until the error is lower than the preset value, and the trained high-precision surrogate model is obtained.

[0017] S9 will input real-time data collected by sparse pressure sensors into a trained high-precision surrogate model to achieve real-time reconstruction of the external flow field of the rocket.

[0018] Preferably, in S5, the coarse aerodynamic distribution f predicted by the low-precision surrogate model... CFD (x) represents the following:

[0019]

[0020] Where Φ is the radial basis function; x is the physical quantity distribution at the current prediction point; x i The sample points correspond to the flow field data calculated by CFD, where each sample point is a discrete point in a low-precision surrogate model; n is the total number of sample points; w i The weights corresponding to each sample point.

[0021] Preferably, in step S6, the weighted fusion prediction result generated based on the high-precision proxy model... It is expressed as follows:

[0022]

[0023] Where α is the data source weight; f CFD (x) represents the coarse aerodynamic distribution predicted by the low-precision surrogate model; f 试验 (x) represents the actual experimental data.

[0024] Preferably, in step S7, the optimized high-precision proxy model parameters It is expressed as follows:

[0025]

[0026] Where D is the dataset, θ is the current model parameter; D is the dataset composed of the weighted fusion prediction results; P(θ|D) is the posterior probability; P(D|θ) is the likelihood function; P(θ) is the prior probability.

[0027] Preferably, the physical boundary conditions include flight speed, angle of attack, and altitude.

[0028] Preferably, the distribution of physical quantities includes velocity, pressure, and temperature.

[0029] The beneficial effects of this invention are as follows:

[0030] (1) By integrating multi-level CFD calculation data and real flight / wind tunnel ground test data, this invention can build a more adaptable surrogate model under a wider range of flight conditions and optimize the prediction accuracy of the model through Bayesian inference. Compared with the single surrogate model of traditional methods, the secondary modeling of low-precision surrogate model and real test data can significantly improve the prediction accuracy of aerodynamic data and reduce the dependence on traditional wind tunnel tests and numerical simulations, thereby reducing test costs and time consumption.

[0031] (2) This invention combines low-precision proxy models and high-precision proxy models, which ensures efficient real-time reconstruction while ensuring rapid evaluation and real-time feedback of rocket aerodynamic performance in the early design stage, thus providing a more flexible, efficient and economical solution for rocket aerodynamic testing and flight control. Through multi-level data fusion and secondary modeling, it solves the shortcomings of traditional single data source methods, enhances the reliability and stability of the model under different flight conditions, and further promotes the intelligence and efficiency of rocket aerodynamic testing technology.

[0032] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments, but the method for real-time reconfiguration of rocket outflow based on multi-source aerodynamic data fusion is not limited to the embodiments. Attached Figure Description

[0033] Figure 1 This is a simplified flowchart of the real-time reconfiguration method for rocket outflow based on multi-source aerodynamic data fusion, as implemented in this invention.

[0034] Figure 2 A schematic diagram of a three-dimensional numerical simulation model of a rocket constructed for an embodiment of the present invention. Detailed Implementation

[0035] The present invention will be further described below through specific embodiments. It should be noted that the specific embodiments described herein are only for the convenience of illustrating and explaining the specific implementation of the present invention, and are not intended to limit the present invention.

[0036] To make the objectives and technical solutions of this invention clearer, the invention will be further described below with reference to the accompanying drawings and examples. It should be understood that the examples described herein are for illustrative purposes only and are not intended to limit the invention.

[0037] See Figure 1 As shown in the figure, this embodiment of a rocket spacecraft outflow real-time reconstruction method based on multi-source aerodynamic data fusion includes the following steps.

[0038] S1, Construct a three-dimensional numerical simulation model of the rocket (such as...) Figure 2 As shown, the geometric characteristics and physical boundary conditions of the rocket are defined, providing a basic model for subsequent CFD (Computational Fluid Dynamics) calculations.

[0039] Specifically, the physical boundary conditions include flight speed, angle of attack, and altitude; for example, flight speed V0 = 300 m / s, angle of attack α = 5°, and altitude h = 10 km. The set characteristics are defined according to engineering needs.

[0040] S2 sets the simulated flight conditions and determines the flow characteristics under defined geometric features and physical boundary conditions.

[0041] Specifically, simulated flight conditions are set, and dimensionless parameters such as Reynolds number are used to determine the flow characteristics under defined geometric features and physical boundary conditions, ensuring the applicability of the results under various flight conditions.

[0042] S3, based on flow characteristics, performs mesh generation for a three-dimensional numerical simulation model, and uses CFD to solve the fluid dynamics equations to generate rocket external flow field data under specified conditions.

[0043] Specifically, STAR-CCM+ was used to mesh the three-dimensional numerical simulation model, and FLUENT was used to solve the fluid dynamics equations to generate rocket external flow field data under defined geometric features and physical boundary conditions.

[0044] S4. Extract equally spaced grid node information, including data such as velocity, pressure, and temperature, from the external flow field data of the rocket simulated by CFD to obtain the distribution of physical quantities; construct a simulation dataset based on the distribution of physical quantities for the initial training of the low-precision surrogate model.

[0045] Specifically, equally spaced grid node information is extracted from the CFD simulation data of the rocket's external flow field to obtain a detailed distribution of physical quantities, including velocity, pressure, and temperature. Further, a simulation dataset is constructed based on this detailed distribution of physical quantities, and a rocket external flow simulation aerodynamic database is built. The data in this database is used for the initial training of subsequent low-precision surrogate models.

[0046] S5 uses a radial basis function-based interpolation method to construct a low-precision surrogate model and uses the distribution of physical quantities in the simulation dataset to predict coarse aerodynamic distributions.

[0047] Specifically, a low-precision surrogate model is constructed using an interpolation method based on radial basis functions. The model predicts a coarse aerodynamic distribution using data from the simulation dataset, which is used to quickly capture the basic trend of flow field changes, thereby shortening the prediction time and reducing the computational cost in complex flow field simulations.

[0048] The coarse aerodynamic distribution f predicted by the low-precision surrogate model CFD (x) represents the following:

[0049]

[0050] Where Φ is the radial basis function; x is the distribution of physical quantities at the current prediction point, usually certain specific conditions of the flow field (e.g., flight speed, temperature, pressure, etc.), which will serve as the input to the surrogate model; x i The sample points correspond to the flow field data calculated by CFD, where each sample point is a discrete point in a low-precision surrogate model; n is the total number of sample points; w i The weights assigned to each sample point reflect the importance of each sample point in the low-precision surrogate model. Sample points with larger weights have a greater impact on the model's prediction results.

[0051] It should be noted that, in order to correct the prediction results and improve the prediction accuracy of the low-precision surrogate model, the aerodynamic distribution f CFD A correction function p(x) can also be added to the right side of (x), such as the error function, gamma function, Bessel function, etc., as follows:

[0052]

[0053] S6. Construct a high-precision proxy model. Based on the real test data of the acquired rocket, use a multi-fidelity data fusion method to combine the prediction results of the low-precision proxy model with the real test data, and perform weighted averaging through the data source weights to generate a weighted fused prediction result.

[0054] Specifically, real test data of the rocket is obtained, and a multi-fidelity data fusion method is used to combine the prediction results of the low-precision surrogate model with the real test data to handle the accuracy differences between different data sources. A weighted average method is used to generate a weighted fused prediction result, mathematically expressed as follows:

[0055]

[0056] Where α is the data source weight; f CFD (x) represents the coarse aerodynamic distribution predicted by the low-precision surrogate model; f 试验 (x) represents the actual experimental data.

[0057] Specifically, S6 is implemented as follows.

[0058] (1) Obtain the prediction results f of the low-precision surrogate model CFD (x)

[0059] In S5, preliminary aerodynamic distribution prediction results have been obtained through low-precision surrogate models (such as interpolation methods based on radial basis functions). These results are obtained through CFD simulation calculations, but due to limited computational accuracy, the results are usually coarse.

[0060] The predictions from low-precision surrogate models are aerodynamic distribution data obtained through CFD simulations under specific flight conditions. While typically less accurate than experimental data, they offer high computational efficiency, allowing for the acquisition of a large number of predictions in a short time.

[0061] (2) Obtain real experimental data f 试验 (x)

[0062] Real experimental data f 试验 (x) is obtained through actual testing (such as wind tunnel testing or flight testing). It is more practical and reliable.

[0063] Real-world test data, obtained through actual flight tests or wind tunnel tests, is typically highly accurate and reflects real aerodynamic characteristics. However, acquiring test data is costly and may not cover all possible flight conditions.

[0064] (3) Weight α allocation

[0065] To integrate data of varying precision, a weight α is assigned to each data source (the prediction results of the low-precision surrogate model and the actual experimental data). This weight reflects the relative importance of each data source in the final prediction, and α is a weighting coefficient between 0 and 1. A larger value means that the results of the low-precision surrogate model (CFD prediction) have a greater impact on the final prediction; conversely, a smaller value means that the experimental data is more important.

[0066] Weights can be selected in various ways, such as based on the reliability of experimental data, the accuracy of the CFD model, and the coverage of the data. Generally, higher accuracy of experimental data is desirable, but it is also more expensive; a balance needs to be found between the two.

[0067] (4) Weighted average fusion

[0068] Based on the weighted average method, the final fusion result It is a weighted average of the low-precision surrogate model and the experimental data. The specific formula is shown in Equation (2), which is the fused prediction result, representing the best estimate of aerodynamic characteristics under the current flight conditions. When α is close to 1, it means that the prediction result of the low-precision surrogate model dominates; when α is close to 0, the experimental data dominates.

[0069] (5) Handling differences in data source precision

[0070] Since low-precision surrogate models and experimental data may have different error characteristics, they usually need to be preprocessed or standardized to ensure that they are fused at the same scale.

[0071] This can be achieved through the following methods:

[0072] Data standardization unifies physical quantities (such as pressure, velocity, etc.) from different data sources into a standardized range.

[0073] Error estimation involves estimating the error of each data source using an error model or statistical methods, and adjusting the weight α accordingly so that data sources with smaller errors have a larger weight in the fusion process.

[0074] (6) Dynamically adjust weights

[0075] Throughout the simulation, the weight α can be dynamically adjusted based on real-time feedback. For example, when new experimental data is collected, the model can be re-evaluated, allowing the accuracy of the fusion model to continuously improve with the arrival of new data. This process typically involves Bayesian inference or other statistical methods to update and optimize the model.

[0076] (7) Result verification

[0077] After data fusion is completed, the final proxy model Its accuracy needs to be verified by comparing it with more experimental or simulation data. If there is a large deviation, it may be necessary to adjust the weight α or retrain the low-precision surrogate model to further improve the accuracy.

[0078] This step fuses the low-precision surrogate model and experimental data using a weighted average method to obtain a more reliable aerodynamic distribution prediction result. This process requires not only the appropriate selection of weights but also the handling of accuracy differences between data sources to ensure that the fused model can balance computational efficiency and prediction accuracy, and can adapt to aerodynamic changes under diverse flight conditions and complex flow environments.

[0079] S7 dynamically adjusts the data source weights of the high-precision surrogate model based on the Bayesian inference principle and the prediction results after weighted fusion, and optimizes the parameters of the high-precision surrogate model through maximum a posteriori probability estimation (MAP).

[0080] Optimized high-precision surrogate model parameters It is expressed as follows:

[0081]

[0082] Where D is the dataset, θ is the current model parameters; D is the dataset composed of weighted fusion prediction results; P(θ|D) is the posterior probability, which is the probability distribution of model parameters θ given observed data D; P(D|θ) is the likelihood function, which is the probability of observed data D given model parameters, and measures the likelihood that data D will be observed under specific model parameters θ; P(θ) is the prior probability, which is the hypothesis about parameter θ without any observed data.

[0083] S8 compares the physical quantity distribution output by the high-precision surrogate model with the real experimental data. When the error is higher than the preset value, the parameters of the high-precision surrogate model are continuously adjusted until the error is lower than the preset value, thus obtaining the trained high-precision surrogate model.

[0084] In this step, the physical quantity distribution output by the high-precision surrogate model is compared with the real experimental data to optimize the real-time prediction capability of the high-precision surrogate model and ensure the immediate availability of aerodynamic data during flight.

[0085] Specifically, the real-time predictive capabilities of the high-precision surrogate model are optimized based on data comparison, primarily relying on data feedback mechanisms and model evaluation strategies to dynamically adjust and update the model. After obtaining real-time data, the high-precision surrogate model is used to predict the current flight conditions. Then, the model prediction results are compared with the actual observed flight data, the errors and deviations are calculated, and the initial accuracy of the high-precision surrogate model is optimized.

[0086] Furthermore, based on dynamic feedback from experimental data, the parameters in the high-precision surrogate model are gradually adjusted to optimize prediction accuracy in real time, thereby further enhancing the model's adaptability and flexibility.

[0087] By employing real-time data feedback, incremental learning, and Bayesian optimization, the parameters in the high-precision surrogate model are dynamically adjusted to further optimize the model's predictive capabilities. By continuously acquiring real-time experimental data and adjusting the high-precision surrogate model's parameters, it is ensured that the high-precision surrogate model can continuously adapt to new flight conditions and environmental changes during flight, maintaining high predictive accuracy.

[0088] S9 will input real-time data collected by sparse pressure sensors into a trained high-precision surrogate model to achieve real-time reconstruction of the external flow field of the rocket.

[0089] It should be understood that those skilled in the art can make improvements and modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for real-time reconstructing of rocket outflow based on multi-source aerodynamic data fusion, characterized in that, include: S1. Construct a three-dimensional numerical simulation model of the rocket, and define the rocket's geometric characteristics and physical boundary conditions; S2 sets the simulated flight conditions and determines the flow characteristics under the defined geometric features and physical boundary conditions; S3, based on flow characteristics, performs three-dimensional numerical simulation model mesh generation, and uses CFD to solve fluid dynamics equations to generate rocket external flow field data under specified conditions; S4. Extract equally spaced grid node information from the CFD simulation of the rocket's external flow field to obtain the physical quantity distribution; construct a simulation dataset based on the physical quantity distribution for the initial training of the low-precision surrogate model; S5 uses a radial basis function-based interpolation method to construct a low-precision surrogate model and uses the distribution of physical quantities in the simulation dataset to predict coarse aerodynamic distributions. S6. Construct a high-precision proxy model. Based on the real test data of the rocket, use a multi-fidelity data fusion method to combine the prediction results of the low-precision proxy model with the real test data, and perform a weighted average through the data source weights to generate a weighted fusion prediction result. S7, based on the Bayesian inference principle and the prediction results after weighted fusion, dynamically adjusts the data source weights of the high-precision surrogate model and optimizes the parameters of the high-precision surrogate model through maximum a posteriori probability estimation; S8 compares the physical quantity distribution output by the high-precision surrogate model with the real experimental data. When the error is higher than the preset value, the parameters of the high-precision surrogate model are continuously adjusted until the error is lower than the preset value, and the trained high-precision surrogate model is obtained. S9 will input real-time data collected by sparse pressure sensors into a trained high-precision surrogate model to achieve real-time reconstruction of the external flow field of the rocket.

2. The method for real-time reconstructing of rocket outflow based on multi-source aerodynamic data fusion according to claim 1, characterized in that, In S5, the coarse aerodynamic distribution f predicted by the low-precision surrogate model. CFD (x) represents the following: Where Φ is the radial basis function; x is the physical quantity distribution at the current prediction point; x i The sample points correspond to the flow field data calculated by CFD, where each sample point is a discrete point in a low-precision surrogate model; n is the total number of sample points; w i The weights corresponding to each sample point.

3. The method for real-time reconstructing of rocket outflow based on multi-source aerodynamic data fusion according to claim 1, characterized in that, In step S6, the weighted fusion prediction result generated based on the high-precision proxy model It is expressed as follows: Where α is the data source weight; f CFD (x) represents the coarse aerodynamic distribution predicted by the low-precision surrogate model; f 试验 (x) represents the actual experimental data.

4. The method for real-time reconstructing of rocket outflow based on multi-source aerodynamic data fusion according to claim 1, characterized in that, In S7, the optimized high-precision proxy model parameters It is expressed as follows: Where D is the dataset, θ is the current model parameter; D is the dataset composed of the weighted fusion prediction results; P(θ|D) is the posterior probability; P(D|θ) is the likelihood function; P(θ) is the prior probability.

5. The method for real-time reconfiguration of rocket outflow based on multi-source aerodynamic data fusion according to claim 1, characterized in that, The physical boundary conditions include flight speed, angle of attack, and altitude.

6. The method for real-time reconfiguration of rocket outflow based on multi-source aerodynamic data fusion according to claim 1, characterized in that, The distribution of physical quantities includes velocity, pressure, and temperature.

Images