Computational Fluid Dynamic Analysis and Optimisation of Microfluidic Systems

A novel Computational Fluid Dynamics-enabled multi-objective optimisation methodology for Polymerase Chain Reaction flow systems is proposed and used to explore the effect of geometry, material and flow variables on the temperature uniformity, pressure drop and heating power requirements, in a prototype three-zone thermal flow system. A conjugate heat transfer model for the three-dimensional flow and heat transfer is developed and solved numerically using COMSOL Multiphysics® and the solutions obtained demonstrate how the design variables affect each of the three performance parameters. These show that choosing a substrate with high conductivity and small thickness, together with a small channel area, generally improves the temperature uniformity in each zone, while channel area and substrate conductivity have the key influences on pressure drop and heating power respectively.
A multi-objective optimisation methodology, employing accurate surrogate modelling facilitated by Machine Learning via fully-connected Neural Networks, is used to create Pareto curves which demonstrate clearly the compromises that can be struck between temperature uniformity throughout the three zones and the pressure drop and heating power required. Besides the deterministic optimisation discussed above in which the input variables are assumed to be deterministic. A robust optimisation is conducted, considering the uncertainties in design parameters of the PCR device due to uncertain tolerances from the manufacturing process of PCR system. Monte Carlo simulations (MCS) are carried out to propagate and quantify the uncertainties, then the mean and standard deviation of the output obtained. The robust design is formulated based on the output of probabilistic solutions from MCS and solved to find optimal objective values (greatest temperature uniformity throughout the three zones and the smallest pressure drop) that exhibit minimum change as a result of variations in the design parameters.
The robust optimisation results demonstrate how the uncertainty in the design variables affect the objectives and that single optimum objective points are different from the deterministic optimum points. The robust optimum value depends on the design requirements, such as whether minimising variation in performance due to uncertainties is more important than optimising the objective function, and illustrative examples are provided. The Pareto front of multi-objective solutions demonstrate that the robust optimisation Pareto front is slightly shifted from that of the deterministic solution, and this shift increases with increasing uncertainty in the input design variables. Hence in the presence of uncertainty the deterministic Pareto front is not useful and its solution is restricted.