(532b) Can Deep Neural Networks Make Large-Scale Resolved Particle CFD a Reality?

Understanding the flow field and its interactions with transport phenomena and chemical reaction is the main challenge for design, scaling-up and optimization of most chemical processes particularly, fixed bed reactors. Nearly 20 years ago Dixon and colleagues established a new framework for studying fixed bed processes using Computational Fluid Dynamics (CFD). Since then they have led the field through extensive studies of these systems by integrating CFD with transport phenomena and surface chemistry, and addressing most of the mathematical and computational challenges. One of the persisting issues, however, is the computational cost of these simulations which blocks the pathway for large scale reactor simulations with complex geometries. For such studies the number of control volume cells (the size of volumetric mesh) exceeds the computational limitations and therefore, running the simulations becomes impossible.

However, recent advances in machine learning, specifically, deep learning, have shown promising frameworks for learning the computational structures (3D volumetric mesh) and their corresponding flow patterns. In this talk, we discuss the potential uses of Deep Neural Networks (DNNs) for learning and predicting the flow field within complex geometries such as fixed bed reactors and how this approach can revolutionize the field.