Data-driven analysis of structural instabilities in electroactive polymer bilayers based on a variational saddle-point principle: Datasets and ML codes

The datasets and codes provided here are associated with our article entitled "Data-driven analysis of structural instabilities in electroactive polymer bilayers based on a variational saddle-point principle". The main idea of the work is to develop surrogate models using the concepts of machine learning (ML) to predict the onset of wrinkling instabilities in dielectric elastomer (DE) bilayers as a function of its tunable geometric and material parameters. The required datasets for building the surrogate models are generated using a finite-element-based framework for structural stability analysis of DE specimens that is rooted in a saddle-point-based variational principle. For a detailed description of this finite-element framework, the sampling of data points for the training/test sets and some brief notes regarding our implementation of the ML-based surrogates, kindly refer to our article mentioned above. Here, the datasets 'training_set.xlsx' and 'test_set.xlsx' contain the values of the critical buckling load (critical electric-charge density) and critical wrinkle count for the DE bilayer for the sampled data points, where each data point represents a unique set of four tunable input-feature values. The article above provides a description of these features, their physical units and their considered domain of values. The individual Jupyter notebooks import the training dataset and develop ML models for the different problems that are described in the article. The developed models are cross-validated and then tested on the test dataset. Extensive comments describing the ML workflow have been made in the notebooks for the user's reference. The conda environment containing all the necessary packages and dependencies for the execution of the Jupyter notebooks is provided in the file 'de_instabilities.yml'.