Sparse identification of particle-laden turbulence closures

In this talk, we present a sparse identification methodology for model closure of the multiphase Reynolds Average Navier—Stokes (RANS) equations, with specific emphasis on collisional fluid-particle flows. Sparse identification of multiphase RANS closures (SIMR) employs linear regression with a mixed penalty term. The penalty acts to ensure sparsity of the resultant model and bounded coefficients, thereby encouraging robust model selection. The outcome is a compact, algebraic RANS model, which is easily incorporated into existing CFD solvers and allows direct inferences about underlying physics to be drawn. We first demonstrate the SIMR methodology on single-phase homogeneous shear turbulence, and show improvement over existing turbulence models. We then demonstrate the method on highly-resolved data obtained from Eulerian-Lagrangian simulations of fully-developed cluster-induced turbulence (CIT), where fluctuations in particle concentration generate and sustain the carrier-phase turbulence. The numerical data is used as the training basis of the sparse identification regression to infer improved closure models for the Reynolds Stress Equations.