Physics Informed Neural Network based Modelling of Microchannel Flow Dynamics

The flow inside microchannels is experimentally studied using Particle Image Velocimetry (PIV) and numerically using computational fluid dynamics (CFD). However, depending on the model used for simulation, there is often a mismatch between the experimental data and the numerical counterpart. We devise a technique based on Physics-Informed Neural Networks (PINN) to solve the momentum equations by incorporating some data points from the experiments to get the most accurate model of the flow device/ channel. The loss function of the neural network is modified as per Navier-Stokes equations, and the continuity equation. The physical system modeling is done for the microchannel dimensions, for which computational cost is prohibitive at times, and experiments are not always feasible. We compare the effectiveness of our PINN model with CFD techniques to provide a robust alternative addressing experimental uncertainties. Such methodologies are effective in modeling the flow inside microdevices that have multifaceted implications in biosensing, drug discovery and delivery, point-of-care testing, and future energy devices.