Feature-consistent field inversion and machine learning framework with regularized ensemble Kalman method for improving the <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>k</mml:mi></mml:math>-<i>ω</i> shear stress transport model i

Numerical simulations of separated turbulent flows are needed for practical applications of computational fluid dynamics. Due to the inherent Boussinesq assumption and initial development based on simplified flows, the commonly used RANS models often encounter errors and uncertainties when simulating complex turbulence, particularly in separated flows. To address this problem, a feature-consistent correction framework is presented in this work, using regularized ensemble Kalman inversion and machine learning. Insufficient prediction accuracy of RANS in simulating separated flows is addressed by incorporating DNS data and experimental measurements into the k-\ensuremath{\omega} shear stress transport model.