Controlling Biofilm Transport with Porous Metamaterials Designed with Bayesian Learning

arXiv2305.08574MaRDI QIDQ6436623FDOQ6436623

Hanfeng Zhai, Jingjie Yeo

Publication date: 15 May 2023

Abstract: Biofilm growth and transport in confined systems is a common phenomenon. While machine learning and optimization have been extensively applied in materials design, there is still a scarcity of thorough evaluations regarding the optimization process, particularly for bio-porous materials design. We combined Bayesian optimization (BO) and individual-based modeling to conduct design optimizations for maximizing different porous materials' (PM) biofilm transporation capability. We first characterize the acquisition function in BO for designing 2-dimensional porous membranes. Results showed that the variance of the overall samples by the upper confidence bound (UCB) is 32.08% higher than that of the expected improvement (EI); the mean objective of the overall samples by the EI is 1.49% higher than that of the UCB. Given the predefined target region, the EI is 2.35% more efficient than the UCB compared with uniform grid search. We then use EI for designing lattice metamaterials (LM) and 3-dimensional porous media (3DPM). It is found that BO is 92.89% more efficient than the uniform grid search for LM and 223.04% more efficient for 3DPM. The selected characterization simulation tests match well with the Gaussian process regression approximated design spaces for three cases. We found that all the extracted optimal designs exhibit better biofilm growth and transportability than nonconfined vacuum space. Our comparison study shows that PM stimulates biofilm growth by taking up volumetric space and pushing biofilms' upward growth, as evidenced by a 20% increase in biofilms in vacuum space compared to porous materials, and 128% more biofilms in the target growth region for PM-induced biofilm growth compared with the vacuum space growth. Our work provides deeper insights into bio-porous materials design, optimization process characterizations, and extracting new mechanisms from the optimizations.

Has companion code repository: https://github.com/hanfengzhai/pylamdo

This page was built for publication: Controlling Biofilm Transport with Porous Metamaterials Designed with Bayesian Learning

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6436623)