Current M4L Research
Unraveling the structure-property relationships in fiber-composite materials using Machine Learning and Global Sensitivity Analysis
The development and deployment of advanced new materials are linked to the understanding of their structure-property relationships. Physically-based approaches have extensively been used for this purpose, but they present some limitations related to their computational cost and the communication of information between the multiple hierarchical length scales involved. For their part, data models are designed to be computationally efficient, but they are not necessarily formulated with an explicit knowledge of the physical behavior of the system under study.
In this work, we are developing a machine-learning based model that exploits a combination of the physical knowledge of the microstructure with data-driven techniques to predict the local strain field in the material. In particular, we are applying this method to extract the structure-property linkages in a two-dimensional metal matrix composite (MMC), by using a fully-connected neural network. As part of the model, global sensitivity analysis is also employed to identify the most prominent microstructural features that drive the mechanical behavior of the material.
Collaborators: Lori Graham-Brady (Johns Hopkins University), Francisco L. Jimenez (UC Boulder), Jessica A. Krogstad (University of Illinois at Urbana-Champaign), Michael D. Shields (Johns Hopkins University)
The effect of dislocation character on dislocation line tension in bcc tungsten and its impact on kink-pair enthalpy
In addition to the well-characterized elastic contribution, the energy of a dislocation contains an inelastic, or ‘core,’ term that reflects the loss of validity of elasticity theory at dislocation segments. While the elastic part is known to be symmetric about its maximum value for the edge orientation (minimum for screw), in bcc metals, the core energy displays an asymmetry than can be characterized using atomistic calculations. In kink-pair configurations on screw dislocations, this asymmetry leads to a difference in energy between ‘right’ and ‘left’ kinks that is not captured in elastic models. In this work, we are calculating dislocation segment self-energies as a function of dislocation character in bcc tungsten and kink-pair enthalpies as a function of stress. To avoid finite-size artifacts in atomistic simulations, we are developing continuum models of kink-pair configurations based on full elasticity and line tension approaches, parameterized with a substrate Peierls potential and dislocation self-energies obtained from atomistic calculations.
Collaborators: Marian Group (UC Los Angeles), Vasily Bulatov (Lawrence Livermore National Laboratory)