A Novel Gradient Enhanced Gaussian Predictor Framework for the Formulation of Hyperelastic Electromechanical Models
Please login to view abstract download link
Robotics is undergoing a paradigm shift from hard to soft robotics, emphasising the use of metamaterials such as Electroactive Polymers (EAPs) that adapt to their environments. EAPs, known for their high energy density, rapid response, and low weight, often feature complex microstructures requiring modelling which incorporates multi-scale homogenisation, such as FEM2, to yield macro-scale responses. However, this process is computationally intensive, and exploration of acceleration provides an opportunity for Machine Learning (ML) techniques. Gaussian Process Regression, also referred to as Kriging, offers a promising alternative by learning the homogenised material response directly from data, addressing inefficiencies in multi-scale modelling. This work explores the application of Gradient Enhanced Kriging to electromechanical constitutive models, utilising adapted correlation functions to uphold physical symmetry constraints through invariants of strain tensors.