On the Stochastic Modeling of Viscoelastic Materials based on the RL-ABC Method
Please login to view abstract download link
The viscoelastic materials are a class of materials that undergo simultaneous elastic and viscous deformations when subjected to stress. However, most models currently used in the literature to characterize the properties of these materials are phenomenological and thus carry uncertainties, even with available data. In this context, this work proposes a transition from a deterministic perspective to a stochastic perspective concerning the definition of these materials' properties. Firstly, experimental data were collected for four viscoelastic materials (Bio-sourced Epoxy, Shape Memory Polymer filament for 3D printing, tBA / PEGDMA, Vitrimer). Then, a library of viscoelastic model candidates is proposed, considering the Zener and fractional Zener models, generalized maxwell model and the so-called 2-Springs-2-Parabolics-1-Dashpot model. Having the experimental master curves of the materials, the Reinforcement Learning - Approximate Bayesian Computation (RL-ABC) method is put forward to select the best model to describe a specific material, as well as the probabilistic estimate of the model parameters. It is interesting to note that the Bayesian approach favors less complex models that are capable of satisfactorily representing experimental data, providing a stochastic framework to improve model's reliability.
