Data-driven inverse design of metastructures for tailored quasi-zero-stiffness performance

Document Type

Article

Publication Date

6-1-2026

Abstract

Quasi-zero-stiffness (QZS) metastructures are promising for low-frequency vibration isolation in precision instruments and aerospace applications, but their inherent nonlinearities make tailoring performance with conventional methods difficult. This paper proposes a data-driven inverse design framework combining a backpropagation neural network with a genetic algorithm. A dataset of 1000 finite element simulations was generated by varying the geometric parameters of the metastructure. The neural network was trained to map these parameters to tangent stiffness and load capacity, and was then coupled with the genetic algorithm to inversely derive geometric configurations that meet target specifications. The neural network achieved excellent predictive accuracy (R2 = 0.99). For a target of 0.55 N load capacity and 0.01 N/mm tangent stiffness, experimental validation yielded 0.53 N and 0.009 N/mm, while simulations predicted 0.57 N and 0.013 N/mm. The maximum load deviation was 0.02 N (3.6% relative error). Sweep vibration tests showed an initial isolation frequency of 4.27 Hz and an isolation efficiency of 90% at 44.92 Hz. This machine-learning-based inverse design enables precise customization of QZS metastructures, offering an efficient alternative to trial-and-error methods.

Keywords

Backpropagation neural network, Genetic algorithm, Inverse design, Quasi-zero-stiffness metastructures, Vibration isolation

Publication Title

International Journal of Mechanics and Materials in Design

ISSN

1569-1713

DOI

10.1007/s10999-026-09897-2

Volume

22

Issue

2

Share

COinS