Robust design optimisation of continuous flow polymerase chain reaction thermal flow systems

TL;DR

This paper introduces a robust optimization methodology for CFPCR devices that accounts for manufacturing uncertainties, improving thermal and pressure performance through surrogate modeling and probabilistic analysis.

Contribution

It combines Gaussian Process Regression and Polynomial Chaos Expansions to efficiently propagate uncertainties in the design optimization of CFPCR systems.

Findings

Validated noise-free simulation data for accurate modeling

Demonstrated the effectiveness of probabilistic models in robust design

Compared robust and deterministic optimization outcomes

Abstract

This paper presents an efficient methodology for the robust optimisation of Continuous Flow Polymerase Chain Reaction (CFPCR) devices. It enables the effects of uncertainties in device geometry, due to manufacturing tolerances, on the competing objectives of minimising the temperature deviations within the CFPCR thermal zones, together with minimising the pressure drop across the device, to be explored. We first validate that our training data from conjugate heat transfer simulations of the CFPCR thermal flow problems is noise free and then combine a deterministic surrogate model, based on the mean of a Gaussian Process Regression (GPR) simulator, with Polynomial Chaos Expansions (PCE) to propagate the manufacturing uncertainties in the geometry design variables into the optimisation outputs. The resultant probabilistic model is used to solve a series of robust optimisation problems. The influence of the robust problem formulation and constraints on the design conservatism of the robust optima in comparison with the corresponding deterministic cases is explored briefly.