Data-Driven Prediction of Swirling Flow Fields via Feedforward Backpropagation Neural Networks

Abstract

Improving energy efficiency while minimizing environmental impact has long been a central challenge in combustion-based technologies. Swirl (vortex) flow, a key innovation for enhancing combustion, remains an active research focus. This study explores the use of artificial intelligence to predict swirl flow characteristics in a combustion chamber. Experimental positional and descriptive data served as inputs, with horizontal and vertical velocities and kinetic energy as outputs. The model accurately reproduced velocity density distributions and vortex center locations, closely matching experimental results. It demonstrated strong predictive performance on known datasets, effective reconstruction of the vortex flow field, and robust generalization to unseen cases. These results confirm AI’s potential for modeling complex combustion flows and suggest promising applications for predictive control and optimization in energy systems.