Multifidelity aerodynamic flow field prediction using random forest-based machine learning

In this paper, a novel random forest (RF)-based multifidelity machine learning (ML) algorithm to predict the high-fidelity Reynolds-averaged Navier-Stokes (RANS) flow field is proposed. The RF ML used increase fidelity of low-fidelity potential field. Three cases are studied, first two consist past backward-facing step, and third, subsonic around an airfoil . case, data generated using ten different inlet velocities , in second six step heights, third 20 shapes parameterized B-spline curves. Input parameters case dependent. For x y cell-center locations corresponding velocities, along with specified velocity, used. second, values nondimensionalized height, height place velocity. remaining input features same as previous case. stream function velocity control point variables. outputs for all include RANS pressures, turbulent viscosities. results study compared those tensorFlowFoam (TFF) from directly solving equations. To quantify errors, absolute error relative L 2 norm metrics show that cases, consistently has 30 times lower TFF, only exception being viscosities better at predicting pressure skin friction coefficients RAE 2822 NACA 0012 airfoil. 1.67 1.19 coefficients, respectively.