Microfluidic Injector Models Based on Neural Networks
R. Magargle, J.F. Hoburg and T. Mukherjee
Carnegie Mellon University, US
microfluidic, electrokinetic, lab-on-a-chip, injector, neural network
An injector modeling methodology based on neural networks is presented. The new aspects of this approach are (1) the full description of the dynamic injector behavior and (2) the applicability of the approach to a much broader range of injector types. Examples are shown for the cross, doublt-tee, and gated-cross injectors. Accuracy is on the order of 10^-4 for mean squared error, with four orders of magnitude increase in speed relative to numerical simulation. These fast and accurate injector models present a much more feasible approach to CAD than numerical simulation, when interconnected with other microfluidic system component block models such as the mixer and separator.
Back to Program
Nanotech 2005 Conference Program Abstract