Search for a command to run...
Abstract This research explores the aeroelastic behavior of a high aspect ratio composite plate made from glass fiber, a key material widely used in advanced engineering applications. The study involved subjecting the plate to subsonic wind tunnel testing, during which it exhibited Limit Cycle Oscillations (LCO). These oscillations underscore the need for advanced predictive methodologies to safeguard structural integrity and optimize performance under aerodynamic loading conditions. Experimental data capturing the time-domain displacement of the composite plate were utilized to develop a predictive framework. A Long Short-Term Memory (LSTM) neural network was implemented to forecast aeroelastic instability. The LSTM model was constructed with two layers, each comprising 50 units, and used batch normalization to improve stability and convergence during training. Sequential displacement data were preprocessed, normalized and segmented into input-output pairs through a sliding window approach, enabling the network to effectively learn the temporal dependencies and non-linear dynamics characteristic of aeroelastic phenomena. Complementing the LSTM approach, numerical analysis was performed using the SOL 145 aeroelastic solver with the PK method. This numerical technique offered precise insights into the stability margins and flutter characteristics of the plate under aerodynamic loading. By solving linearized aeroelastic equations, the PK method accurately determined critical flutter velocities and identified stability boundaries, serving as a robust validation tool for experimental and LSTM-based results. The LSTM model demonstrated superior adaptability to real-world conditions, leveraging its data-driven approach to achieve lower prediction errors and more accurate instability forecasts. This adaptability was evident in metrics like Root Mean Square Error (RMSE), where the LSTM effectively captured the plate’s nonlinear and temporal dynamics. Visual analyzes further confirmed that the LSTM model closely followed the experimental data and identified high-fidelity instability trends. This study underscores the complementary nature of machine learning and numerical analysis in addressing complex aeroelastic challenges. By integrating experimental observations, computational solvers, and data-driven deep learning techniques, this research establishes a robust and versatile framework for analyzing and predicting aeroelastic phenomena in composite structures, paving the way for safer and more efficient engineering systems.