Session: CIE-02 / 03 / 04 AMS: Joint Topics

Paper Number: 148072

148072 - Asynchronous Parallel Bayesian Optimization for Constitutive Model Calibration in Crystal Plasticity Finite Element Method 

Researching and developing predictive integrated computational materials engineering (ICME) models at multiple length-scales and time-scales has been conducted in the last two decades. To establish the predictive capability, the ICME model must be numerically verified and experimentally validated, in the spirit of uncertainty quantification. The development of ICME models as the third paradigm and scientific machine learning as the fourth paradigm is consistent with the Materials Genome Initiative, which has long been held as a cornerstone in materials science.

 

Constitutive model calibration has been fairly well-studied within the literature. Loosely speaking, calibrating a constitutive model for CPFEM can be formulated as an optimization problem, which in turn minimizes the misfit between the CPFEM and experimental data. However, most approaches, if not all as of this point, mainly utilize sequential optimization, which suffers from the high computational cost of the forward CPFEM model. Even though the high computational cost can be somewhat mitigated by message-passing interface (MPI) parallelism on multi-core high-performance computers, it also follows a diminishing return characterized by Amdahl's law~\cite{hill2008amdahl}. Modern optimization approaches often exploit both MPI (or MPI+X for heterogeneous) parallelism as well as optimization parallelism on high-performance computers and are capable of delivering a better solution in a shorter wallclock time. Compared to sequential optimization approaches, parallel optimization approaches can offer better performance in calibrating ICME models, especially for prototyping models that do not scale well with multi-cores and multi-threads on high-performance computing platforms. 

 

The crystal plasticity finite element model (CPFEM) is a powerful numerical simulation in the integrated computational materials engineering (ICME) toolboxes that relate microstructures to homogenized materials properties and establish the structure-property linkages in computational materials science. However, to establish the predictive capability, one needs to calibrate the underlying constitutive model, verify the solution, and validate the model prediction against experimental data. Bayesian optimization (BO) has stood out as a gradient-free efficient global optimization algorithm that is capable of calibrating constitutive models for CPFEM. In this talk, we apply a recently developed asynchronous parallel constrained BO algorithm to calibrate phenomenological constitutive models for stainless steel 304L, Tantalum, and Cantor high-entropy alloy.

Presenting Author: Anh Tran Sandia National Laboratories

Presenting Author Biography: Dr. Anh Tran is a Senior Member of Technical Staff at Sandia National Laboratories in Albuquerque, NM. Prior joining to the Scientific Machine Learning Department, he worked in the Uncertainty Quantification and Optimization Department. Anh obtained his B.S., M.S., and Ph.D. from the Georgia Institute of Technology in 2011 and 2018, respectively. He also held an M.S. in Applied Mathematics from Georgia Southern University in 2014. His research interests include machine learning, optimization, and uncertainty quantification, with applications to multi-scale computational materials science.

Authors:

Anh Tran Sandia National Laboratories
Hojun Lim Sandia National Laboratories