Reduced-Order Modelling of an Airfoil with Fluid-Structure Interactions for Characterizing Physical Reservoir Computing

Modern flight systems are becoming increasingly sophisticated, demanding greater computational power for various tasks, including real-time control, navigation, and sensor processing. This need is particularly relevant for systems that leverage adaptive structures to optimize various performance characteristics. Increased degrees of freedom and complex fluid-structure interactions leave such systems susceptible to aeroelastic phenomena. Detecting and controlling such phenomena requires complex models or large numbers of sensors and additional computational power. Alternative computing architectures have been proposed as a means to address challenges associated with these increased computational costs. This includes leveraging the intrinsic fluid and structural dynamics through physical reservoir computing. Simulating the complex interaction of these physical domains is itself a computationally expensive task. This study explores the feasibility of using a simplified 2-D fluid-structure interaction model to explore the design space of an elastically deformable airfoil in an incompressible flow as a physical reservoir computer. Fluid dynamics are modeled using the unsteady lumped vortex method, while structural dynamics are modeled in the time domain using finite element analysis. The two physical domains are coupled using a loose scheme allowing for simple implementations of separate analysis codes. A spectral analysis is performed on strain readouts to examine the system's frequency content as material properties and flow speed are varied. Initial results show a relationship between frequency metrics associated with computational performance and the fluid and material properties of the system. However, closer inspection of the physical reservoir computer design space through task benchmarking suggests that careful selection of properties must be undertaken when considering application-relevant tasks. Some tasks, such as imitating the lift generation of the airfoil using strain readouts, are likewise impacted by the properties of the system, creating difficult tasks that even the higher performance reservoirs are ill-suited to handle. This suggests that fluid-structure interaction can be used to enhance the performance of a physical reservoir computer only by leveraging advantages of the respective domains, for instance, using a stiffer structure to simplify the complexity of the flow. The study also demonstrates the utility of this fluid-structure interaction model for exploring design spaces, enabling the identification of specific regions of interest for further development with high-fidelity modeling tools.