POD-based surface pressure reconstructions from sparse sensing

Abstract

Aerodynamic load determination through integration of the surface pressure distribution is limited in accuracy by the discrete number of measurements taken. Therefore, typically relatively large numbers of spatially distributed measurements are required which increase setup complexity, latency, cost and weight while requiring physical access in and around the model to house transducers/taps. Especially in practical settings outside of a wind tunnel, such setups might be unfeasible and a reduction in the required number of measurements desired. In this thesis work, experimental measurements of the surface pressure distribution around a bluff body square cylinder model are combined with a modal decomposition method; Proper Orthogonal Decomposition (POD) to encode the system in a low-dimensional representation. This low-dimensional representation is used for the determination of optimal sensor placement which is in turn used for sparse surface pressure measurements on the model at various angles of attack. An extension to the POD, known as Gappy POD (GPOD), combines the low-dimensional representation with the sparse measurements to infer full state surface pressure reconstructions at a reduced number of sensors. Based on this outlined approach, experimental surface pressure distributions and drag coefficients have been reconstructed in reasonable accordance with corresponding references at sparsity levels up to 85%. GPOD + tailored sensor placement can therefore effectively be used to relieve the sensor requirements in practical settings if prior experimental ‘training’ is an option. To reduce reliance on experimental measurements, RANS simulations were used for sensor placement and dominant low-dimensional flow feature identification instead as well. RANS simulations lacked the accuracy to replace experimental training through POD as reconstructions could not consistently remain accurate.