Modelling of aeroelastic flutter in T-shaped cantilever systems using extreme learning machine

Flutter-induced aeroelastic instabilities in flexible thin-walled structures are governed by strong fluid–structure coupling and motion-induced aerodynamic forces. The T-shaped cantilever plate with a tip mass represents a characteristic flutter configuration, exhibiting sensitive torsional modes and a rapid transition to self-excited oscillations once aerodynamic damping becomes negative. Its response depends on geometric and mechanical parameters, particularly the height of the vertical plate and the tip mass, while its ability to sustain oscillations over a wide range of wind speeds due to geometric nonlinearity makes it attractive for small-scale aeroelastic energy harvesting. This work proposes a data-driven framework for modelling motion-induced aerodynamic forces using high-resolution flow simulations with prescribed oscillations. These simulations provide training data capturing the relationship between structural motion and aerodynamic forces. An Extreme Learning Machine (ELM) is employed to efficiently learn this nonlinear mapping. Given the strong motion dependency of the aerodynamic forces, the framework also allows consideration of temporal motion information in the aerodynamic force prediction. The resulting surrogate model is intended to be coupled with a geometrically nonlinear structural solver to enable computationally efficient fluid–structure interaction simulations while retaining essential nonlinear aerodynamic effects. Preliminary investigations demonstrate promising agreement between the ELM-predicted and high-fidelity aerodynamic forces for different structural motion conditions and wind speeds. Ongoing work focuses on the fully coupled implementation of the proposed framework and the assessment of its capability to reproduce flutter onset, limit cycle oscillations, and post-critical aeroelastic response.