Machine-Learning-Driven Predictive 3D Ramified Foam Fabrication and Mechanistic Understanding

Abstract

Hierarchical nanosuperstructures, ubiquitously found in nature, present dually enhanced mass transport and interfacial chemical reactions due to their unique 3D cascading features. Their man-made counterparts have demonstrated meritorious benefits toward electrocatalysis, flexible supercapacitors, and water disinfection. However, fabricating 3D superstructures with accurate structural characteristics remains exhaustive and challenging due to a multitude of variables in both experimental conditions and structural features. In this work, we explore three machine learning (ML) methods─linear regression, neural network regression, and Gaussian process regression─and, for the first time, realize accurate predictive fabrication of designed 3D microbranched foams by using a small training data set. Our findings demonstrate an advantageous accuracy of Gaussian process regression of over 87% across all benchmarks. We also effectively unravel the weighted roles of various experimental conditions, shedding light on the synthetic mechanisms. Overall, this work represents a new advance in the ML-enabled predictive fabrication of complex structures and materials with mechanistic elucidation.