other
Universität Stuttgart
Institut für Technische Verbrennung
In large eddy simulation (LES) of turbulent flows closures for the unresolved sub-grid scales are required. Recently, advanced deep-learning techniques have been used as deconvolution operators to reconstruct the lost sub-grid information by providing approximate inverse to the filter operator in LES such that unresolved terms in the governing equations can be computed explicitly in terms of deconvolved variables. In the context of deep learning methods, this deconvolution task is called super-resolution (SR) task. The networks applied for the SR task include deep residual-in-residual dense blocks (RRDB) in the framework of generative adversarial network (GAN) and transformer-based models including hybrid attention transformers. While these models have shown superior results in imaged-based SR tasks, in this project we aim at applying such models for the SR of flow fields to build sub-grid scale models for LES of pulverized biomass combustion.