Development of a CFD Based Artificial Neural Network Metamodel for the Design of Micromixer

Microfluidics has enabled researchers to explore the physics of fluid in the regions of micro in size. Use of Machine Learning techniques in microfluidics to predict flow behavior and reduce time is an unexplored area. A Y-shaped micromixer was designed and its five design variables were identified. A sample size was designed by general factorial method and the simulation experiments were conducted using a CFD solver. The evaluation criteria was mixing index at the outlet of Y-shaped micromixer. Based on results of 1024 simulations, an Artificial Neural Network (ANN) metamodel was developed. Finally the metamodel was validated against the simulation results for the values of design variables outside the full factorial sample. The results show that the metamodel shows agreement with the simulation results with 0.6% difference.