Treffer: Predicting the response of aqueous droplets in a microfluidic trap using neural network modeling

Simulations and predictions are now the essential driving tools for design of materials, processes, and products. Rapid digitalization of information has allowed virtual models to be built for cost-effective testing of functional capabilities and validating operating parameters to achieve optimal efficiencies in dynamic environments. These models are usually derived from the laws of physics, mathematical equations, or empirical relations from the observed behavior of the system in the form of data. Various constraints could cause hindrance to generation of data beyond the safe limits of operation, which insists for model-based predictions. Presented here is a similar attempt to develop a Neural Network to describe a MEMS system with limited dataset of simulated outcomes, where aqueous droplets are aimed to be trapped at a designated zone for interaction studies in a microchip. A classical neural network built using Keras Sequential library in Python gave a maximum fitting accuracy of 97.22%. Although the main objective of the study is to understand the sensitivity of geometry, owing to the manufacturing tolerances, in deciding the fate of droplet-trapping, this approach of neural networks paved path for a different advantage.