AACE-AFM-378: Determining Closure Coefficients of Turbulence Models Using Machine Learning …
The solution of the Reynolds-averaged Navier-Stokes (RANS) equations employs an appropriate set of equations for the turbulence modelling. The closure coefficients of the turbulence model have been calibrated using empiricism and arguments of dimensional analysis. Those coefficients are considered ‘universal’, but there is no guarantee this is correct for test cases other than those used in the calibration process. This work aims at revisiting the ‘universality’ of closure coefficients using machine learning, adaptive design of experiments (ADOE) and a high-performance computing facility.The technical objectives of this research activity are to: a) assess the impact of the closure coefficients of the turbulence model (e.g. Spalart-Allmaras) on the prediction of the aerodynamic characteristics; b) calibrate these coefficients by minimizing the deviations of the numerical results from a set of experimental data; and c) verify the independence of the optimized coefficients from spatial discretization errors and specific solver implementations.Our approach entails the exploration of the design space using an efficient and accurate ADOE method. The ADOE algorithm consists of an iterative, machine-learning based technique capable of: a) identifying the regions of the design space that are characterized by stronger nonlinearities; b) adaptively selecting the location of the design points in order to maximize…
Link to Full Article: AACE-AFM-378: Determining Closure Coefficients of Turbulence Models Using Machine Learning …
AACE-AFM-378: Determining Closure Coefficients of Turbulence Models Using Machine Learning …
