Discovery of Explicit Algebraic SGS Dissipation Model Using Gene Expression Programming

Abstract

Explicit algebraic subgrid-scale models (EASSMs) have shown better performance compared to eddy viscosity models (EVMs) due to the improved tensorial alignment to the subgrid-scale (SGS) stress tensor. Under the EASSM framework, the dissipation term requires closure. A number of models exist but some are unsuitable to be applied under the EASSM framework, due to their non-linearity in the SGS anisotropy tensor. The isotropic dissipation model is commonly applied model but studies have contradicted the isotropic assumption. We propose to discover a closure model suitable to be implemented under the EASSM framework using gene expression programming (GEP) as a symbolic regression tool. Techniques such as wall scaling constraints and numerical constant optimization from other works are also incorporated in the present work. In addition, we attempt to mix training data from homogeneous isotropic turbulence (HIT) and turbulent channel flow (TCF) within the training process in order to obtain a more robust model. Discovery of a scalar dissipation model is an intermediate step towards a dissipation tensor model, a priori evaluation shows better performance relative to a reference model. Validation of the scalar model was performed on a channel flow case which also showed improvements. Tensor model discovery produced a few models which provide better tensorial alignment only in selected cases relative to there being no model at all. A critical review of the present work recommended the investigation of implementing an alternate timescale for non-dimensionalization and a study on the dissipation effects that are contained within the pressure-strain model implemented in the framework.