Data-driven analysis of structural instabilities in electroactive polymer bilayers based on a variational saddle-point principle

In this work, we use a data-driven approach to model the onset of wrinkling in composite dielectric elastomer (DE) bilayer structures that are subjected to combined electro-mechanical loading conditions. Specifically, the critical surface-charge density required to activate wrinkling and the resulting number of half-waves are determined as functions of the tunable geometrical and material features of the DE bilayer using supervised machine learning (ML) models. The required data for the ML surrogates is generated using a finite-element-based framework for structural stability analysis of the DE bilayer, which is rooted in a variational saddle-point formulation for non-linear electro-elastostatics. Within the considered broad design space for the DE bilayer, the data points are sampled according to a Latin-hypercube-type design, following which training and test datasets containing the values of the target variables for the sampled input-feature vectors are generated using the adopted finite-element framework. Subsequently, we implement and compare the capabilities of different ML models to capture accurately the non-linear dependence of the wrinkling characteristics on the tunable features of the DE bilayer. •Variational saddle-point approach for modeling instabilities in soft dielectrics•Parametric study of wrinkling of dielectric bilayers under electro-mechanical loading•Efficient sampling of datasets for use in state-of-the-art machine learning methods•Development of surrogate models for predicting the buckling loads and mode shapes•Evaluation of the performances of the surrogates in predicting the buckling onset