Model Order Reduction of Parameter-Dependent Particle Laden Flows

Abstract

In this contribution, we extend our previously-developed framework for model order reduction of transport-dominated systems towards parameterized cases. The method combines the shifted proper orthogonal decomposition with interpolation via artificial neural networks. The resulting framework is an a-posteriori one, i.e., for parameterized systems, time-dependent solutions for several parameter values are required for the construction of the reduced-order model (ROM). The ROM can then be cheaply evaluated to predict the solution for previously unseen parameter values. The method is suitable for industry-relevant transport-dominated systems, such as particle-laden flows, where traditional mode-based methods, such as proper orthogonal decomposition (POD), often fail.