OPTIMIZATION OF NANOPOROUS METALLIC ACTUATORS BY COMBINING MULTISCALE CALCULATIONS AND MACHINE LEARNING [Poster]

Nanoporous materials (NMs) in electrolytes can achieve approximately 1% deformation under low operating voltages of about 1 V, making them promising for use in artificial muscles. The multi-field and multi-scale nature of the NM electrochemical actuator makes simulation-based optimization extremely challenging. A computational framework was developed that combines joint density functional theory (JDFT), surface eigenstress model, symbolic regression, finite element methods (FEM), and surrogate modeling to perform both concurrent and sequential multi-scale calculations. Specifically, JDFT calculations were performed on Au thin films to obtain in-plane strain as a function of charge density and film thickness. The surface eigenstress and surface Young’s modulus of the Au nanofilm were then determined by fitting the surface eigenstress model to the JDFT data. Additionally, symbolic regression was used to obtain the constitutive equation of surface eigenstrain versus surface charge density, which realized macroscale FEM calculations. Finally, a mapping scheme was established between a given sequence of numbers and a particular structure of nanoporous Au, which allowed for the employment of Gaussian process regression surrogate models. These surrogate models were employed to accelerate the evaluation of actuation strain and effective Young’s modulus, and hence enable the exploration of the entire design space.
EXTENDED ABSTRACT