Data-Science-Informed Dislocation Network Theory for Precipitation-Strengthened Metals
Abstract
Crystal plasticity (CP) modeling, an approach for modeling plasticity within polycrystalline aggregates, is a powerful tool for predicting the performance of materials at length and time scales relevant to engineering applications. CP models are commonly used to simulate deformation in precipitation-strengthened metals, such as high-strength aluminum alloys. While based on plausible physical assumptions, the material models used in these simulations are not physics-based. As a result, many knowledge gaps exist regarding the behaviors of precipitation- strengthened metals. For example, in the case of aluminum alloys under high rate loading, which are the focus of this project, it is unclear why strain rate sensitivity varies widely across alloys and why some alloys exhibit adiabatic shear bandingÑan instability leading to highly localized deformation. To advance our understanding of these and other performance-critical alloys, predictive, physics-based CP models must be developed. In order to advance the development of physics-based models, new techniques for modeling and coarse-graining dislocation-based plasticity must be identified. The key challenges are: 1) constructing a dislocation-based CP model that is mathematically and computationally tractable but has enough complexity to resolve the underlying physics; and 2) coarse-graining dislocation processes contained in noisy datasets from lower-scale models (e.g., discrete dislocation dynamics (DDD)) in a manner that is useful to the CP model. This project tackles these challenges by 1) developing a new dislocation network-based model for crystal plasticity and 2) utilizing statistical machine learning techniques to accomplish coarse-graining. The project is comprised of three Research Tasks. In Task 1, the new dislocation network theory will be developed by constructing a governing partial differential equation (PDE) which characterizes dislocation network evolution. Within this theory, the various relevant dislocation processes (e.g., glide, junction formation) map directly onto terms in the governing PDE. For each term, a collection of constitutive functions must be defined to specify how the network evolves. Under Task 2, we will determine these constitutive functions by coarse-graining datasets obtained from DDD simulations. DDD simulations will be performed using a new algorithm for modeling dislocation-precipitate interactions with precipitate microstructures relevant to 2xxx series aluminum alloys. Specific features of dislocation network evolution relevant to the theory will be extracted from the simulation results and compiled. The resulting datasets will be noisy and stochastic, making them difficult to coarse-grain. To this end, we will employ Gaussian process regression (GPR), a non-parametric, Bayesian approach to regression wherein datasets are treated as random samples from a multivariate normal distribution. By training physics-informed GPR models, we can directly determine the constitutive functions necessary for the network model. And finally, in Task 3 we will implement the new data-science-informed CP model for use in the explicit dynamic finite element method. Predictions of the new finite element model will be compared with experimental measurements from high rate loading of aluminum alloys. The framework for constructing physics-based CP models developed under this project will be readily extensible to many other classes of alloys, including high entropy and additively manufactured alloys. If successful, the project stands to significantly advance the predictive capabilities of material models. Improved material models would enable reduced-cost and lighter weight engineering structures with less conservative designs thanks to reduced uncertainties in model predictions. Such physics-based models could also serve as guides for alloy and process design, enabling the next generation of high performance alloys.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 25, 2021
- Source ID
- W911NF2110086
Entities
People
- Ryan B. Sills
Organizations
- Army Contracting Command
- Rutgers University
- United States Army