Aerodynamic Modeling of a Missile Combining Transfer Learning and Dendritic Net

Abstract

During the initial stages of missile design, finding an efficient method for analyzing aerodynamic characteristics is crucial. This paper proposes a novel aerodynamic modeling method based on a limited computational fluid dynamics (CFD) dataset, combining transfer learning (TL) with Dendritic Net (DD). Initially, we employ CFD-calculated aerodynamic data to establish a pretrained model using DD. Subsequently, the model is adapted to the target domain by TL, predicting aerodynamic parameters under specific conditions. The overall aerodynamic parameters are utilized to generate a relational spectrum through DD’s white-box features, from which primary features are extracted to establish an aerodynamic polynomial model. Finally, the model’s practicality is validated by ballistic flight simulations. The innovation lies in leveraging DD to generate a relational spectrum of aerodynamic parameters, leading to a high-precision polynomial model. Research shows DD outperforms traditional cell-based networks in predicting aerodynamic parameters, and TL reduces the CFD computation workload in the target domain by 3/4 while maintaining prediction accuracy. The polynomial model exhibits superior accuracy compared to the empirical fitting formulas. The method reduces the computational workload for aerodynamic data collection, and through system identification, a high-precision polynomial model is obtained, which provides a reliable basis for missile controller design.