Prediction of aerodynamic loads and aeroelastic responses in solving dynamic aeroelasticity problems

In the process of developing modern aviation technology, the importance of aeroelasticity in the design of aircraft is constantly increasing, but the existing methods of aeroelastic analysis and optimization are not sufficiently effective, which hinders their practical engineering application. The aim of this study is to analyze these methods and evaluate their effectiveness in the context of solving the problems of aeroelasticity of an aircraft. Effective solution of these problems requires application of the advanced deep learning neural networks capable of modeling complex nonlinearities and managing large amounts of data. The existing neural network-based models have limitations, in particular, in modeling unsteady flows. Traditional models lack stability when extrapolating, so incorporating large-scale data processing into the model represents a significant advancement in predictive capabilities. The methodological approach includes data collection, model training using a long short-term memory network, and model validation. This study presents a reduced-order unsteady aerodynamic model using a long short-term memory network. A comparison of the efficiency of the proposed model with traditional approaches is provided, which emphasizes its applicability in real-world scenarios. If deep learning only facilitates the management of aerodynamic complexities, then long short-term memory networks additionally take into account the dynamic, nonlinear behavior on extended data sets, which increases the reliability of forecasting. The emergence of complex neural network models heralds a new era in aeroelastic analysis, overcoming the limitations of shallow models.