Modeling and Design of Magnetic Soft Materials Accelerated by Neural Operators

Joint material and structural design of hard-magnetic soft materials, and the neural-operator surrogates being built to make that design loop tractable.

The difficulty is that the two choices are coupled. Where the material goes and what the material is, its density, its magnetic particle fraction and the direction its remanent magnetization points, determine the response together, so optimizing the geometry against a fixed material, or the reverse, finds a worse design than optimizing both. Joint optimization requires repeated nonlinear forward solves. The project therefore investigates neural-operator surrogates together with error estimation and correction.

The work is carried out in MatTO, and the first result, 21 constitutive models compared on actuation problems with experiment selecting the one carried into the joint optimization, is under review.