Our work bringing together Convolutional Neural Networks and multi-physical Modeling was accepted on 2026-02-03
We are happy to announce that the joint work of Siddharth Sriram, Elten Polukhov, Felix Fritzen, and Marc-André Keip from the working groups for Materials Theory and Data Analytics in Engineering has been accepted for publication in Mechanics of Materials.
Magnetoelectric (ME) composites can convert magnetic fields into electrical signals and vice versa. ME coupling requires electrical poling of the ferroelectric phase to activate the piezoelectric effect by aligning electric dipoles. We trained convolutional neural networks on finite element simulation data for many different microstructures. The model was then used to identify micro-morphologies that maximize effective piezoelectric coefficients after poling. Using these optimized morphologies, we computed the effective ME modulus versus magnetic field strength for various constituent volume fractions, yielding trends that agree with experimental observations.
