Abstract

Data assimilation is the principle of combining uncertain measurements from experiments with an imperfect model to obtain a better prediction than either experiments or simulations can offer independently. Data assimilation removes noise, fills in missing experimental data and reduces uncertainty associated with ambiguous modelling parameters in simulations such as turbulence production or boundary conditions. In this talk, several methods for assimilating mean velocity measurements into low-fidelity computational fluid dynamics (CFD) are discussed. The experimental data are obtained from particle image velocimetry (PIV) and each method introduces an unknown forcing term into the Reynolds-averaged Navier-Stokes (RANS) equations. An optimisation problem is formulated whereby the unknown forcing is updated such that the discrepancy between the experimental mean velocity and the CFD is minimised. The talk will address how data assimilation can overcome several limitations of PIV such as sparsity, limited field of view and noise.