A priori sensor placement strategy for turbulent mean flow reconstruction using parametric model perturbations

This paper investigates an a priori approach for sparse sensor placement to generate experimental data for turbulent mean flow reconstruction—data assimilation—in the context of steady Reynolds-average Navier-Stokes (RANS) simulations. The strategy utilises perturbations of the turbulent model constants to generate a map of regions in the flow that are most sensitive to the turbulence model. Sensors are targeted at regions of highest sensitivity using a genetic-algorithm optimiser with a minimum distance constraint between any two sensors, to avoid clustering. The data assimilation approach is based on the adjoint-based field inversion, which modifies the transport equation(s) for an existing model with a spatial field, which is then iteratively optimised with the goal of reducing the error between the baseline model output and the high-fidelity data. The separated flow over the periodic hill is used as a test case, with the one-equation Spalart-Allmaras turbulence model. Direct numerical simulation data is used as surrogate experimental data to allow examining the effectiveness of the framework for various scenarios. Preliminary results show that errors in the streamwise velocity predictions can be reduced by over 30% with only five sensors, compared to over 74% reduction when using the data over the entire domain (i.e. data for the ∼ 14.7 × 103 cells in the mesh).