Physics-regularized data augmentation for effects on local mechanical properties of an epoxy resin ageing under thermo-oxidative environment

Nanoindentation tests are an effective experimental approach to characterize the effects on local mechanical properties of epoxy resins undergoing thermo-oxidative ageing. However, there is currently no physical model capable of accurately representing these effects. In this work, we propose a hybrid methodology combining experimental nanoindentation data with a chemical kinetic model of oxidation through Physics-Regularized Neural Networks (PRNNs). By using the thickness of the oxidized layer as a physically meaningful and common metric between mechanical data and chemical models, we augment limited nanoindentation data. The developed hybrid model demonstrates good generalization capability across different oxygen partial pressures and ageing times. Finally, the resulting hybrid twin is used to produce a reduced order separated representation of the solution field.