Multi-fidelity reduced-order model for GPU-enabled microfluidic concentration gradient design

This paper presents a multi-fidelity reduced-order model (MFROM) and global optimization method for rapid and accurate simulation and design of microfluidic concentration gradient generators (mu CGGs). It divides the entire process into two stages: the offline ROM construction and the online ROM-based design optimization. In the offline stage, proper orthogonal decomposition is used to obtain the low-dimensional representation of the high-fidelity CFD data and the low-fidelity physics-based component model (PBCM) data, and a kriging model is developed to bridge the fidelity gap between PBCM and CFD in the modal subspace, yielding compact MFROM applicable within broad trade space. The GPU-enabled genetic algorithm is utilized to optimize mu CGG design parameters through massively parallelized evaluation of the fast-running MFROM. The numerical results show that MFROM is a feasible and accurate multi-fidelity modeling approach to replace costly CFD simulation for rapid global optimization (up to 11 s/optimization). The design parameters obtained by MFROM-based optimization produce CGs that match the prescribed ones very well with an average error < 6%.

Automated summary

This paper introduces a multi-fidelity reduced-order model (MFROM) and a global optimization method for the simulation and design of microfluidic concentration gradient generators (mu CGGs). The method divides the process into two stages and uses proper orthogonal decomposition and a kriging model to bridge the fidelity gap. The results show that the MFROM-based optimization produces accurate designs with a significantly reduced computation time.