“Crystal plasticity” is a computationally intensive way of computing the behavior of materials undergoing large permanent deformations. Computation is very inhomogeneous: A large effort is expended everywhere, but only a small portion of the computational domain is doing anything interesting.
We use artificial intelligence to do fast “smart interpolation” in the unimportant parts. Because the function space is very high-dimensional, standard interpolation techniques will fail. Machine learning can optimize evaluation points and interpolants.
This technique was pioneered by S. Kalidindi at Georgia Tech. NIST can further this method with Object Oriented Finite Elements tool as a platform and deep expertise with uncertainty quantification.
Crystal plasticity finite element modeling can provide valuable structure-property insights, but is not used due to the computational expense, which mostly arises from extensive work evaluating constitutive rules at integration points. Most integration points, however, are very boring, just doing zero or near-zero yield, and in the general case, it’s hard to predict which ones these are. We can replace with those a fast interpolation.
- Problem: The domain is very high-dimensional.
- Solution: Machine learning is good for interpolating in high-dimensional spaces.