A computational analysis of flow dynamics and heat transfer in a wavy patterned channel using physics-informed neural networks

The current study utilizes an unsupervised physics-informed neural network (PINN) that leverages heuristic optimization techniques to solve steady-state incompressible Navier–Stokes and energy equations in wavy channels for varying Reynolds numbers. Further, in asymmetric cases, in-phase and out-phase shifts between upper and lower channel walls are considered to mimic actual physiological structures occurring in many natural and industrial process. Moreover, the energy transfer equation is involved in the flow governing equation to accomplish the mixed convection flow for the fluid volume movement. Streamwise and cross-streamwise velocity components, pressure drop, and wall shear stresses are computed, and heat transfer characteristics are also analyzed. PINN predictions are compared with the conventional computational fluid dynamics results, ensuring excellent accuracy. The higher accuracy is obtained due to a new architecture of the deep learning model that ensures a conservative relationship between the stress and flow variables. Additionally, a novel scheduler algorithm is implemented to achieve convergence faster by updating the learning rate dynamically. Overall, the study demonstrates the capability of PINN (a more computationally feasible framework) for solving fluid dynamics and heat transfer problems and elucidating separation and reattachment flow dynamics in complex corrugated geometries. In uniform and non-uniform temperature distribution, the Nusselt number increases significantly with Reynolds number showing a heat transfer enhancement in corrugated geometries.