Predicting Flow and Heat Transfer in Stepped Channels with Multiple Obstacles: A Geometry Aware Physics Informed Neural Network Approach

This study presents a different approach to analyzing heat transfer of a reacting chemical in a stepped rectangular channel with multiple obstacles of varying sizes, shapes, and positions, utilizing finite element analysis and a Geometry Aware Physics Informed Neural Network (GAPINN) framework. Building on previous research, we extended the investigation to a more complex geometric configuration, predicting flow and heat transfer in diverse setups. The two-dimensional model includes circular, square, and elliptical heated obstacles in various positions, with laminar, steady-state flow and constant channel wall temperature. The Navier-Stokes, energy and transport equations are solved using COMSOL Multiphysics. The GAPINN framework, combining a Shape Encoding Network, a Physics Informed Neural Network, and a Boundary Constraining Network, is trained on finite element results for rapid exploration of geometric configurations, overcoming the computational limitations of traditional methods. Results show that obstacle placement and sizing can enhance heat transfer efficiency, revealing the importance of considering fluid flow and heat transfer interactions in reactor design. The GAPINN framework achieves excellent agreement with finite element results which significantly speeded up computations, providing a valuable tool for predicting reactor designs with applications in industries like chemical processing and environmental remediation.