Complas 2025

COMPLAS 2025

A neural network framework for thermoviscoplasticity

Jones, Reese (Sandia National Laboratories)
Fuhg, Jan (University of Texas at Austin)
Seidl, Tom (Sandia National Laboratories)

In session: 801C - Machine Learning for Computational Mechanics: Innovations and Applications III

Please login to view abstract download link

Building upon our previous work, we present a neural network model of thermoviscoplasticity which embeds thermodynamic principles. As with other modern approaches, we formulate the model entirely in terms of potentials with particular properties embedded in the neural network formulations. We contrast the adopted approach with other, related frameworks such as generalized standard models and variational plasticity theory. We demonstrate that the proposed model learns fundamental phenomenology including: • rate effects in evolution of internal state • the conversion of plastic work to heating using synthetic and experimental data.