Physics-Based Machine Learning Parametrization of Force Fields with in-situ TEM Experimental Validat
Abstract
-Approved for Public Release - Families of newly synthesize 2D materials are emerging with numerous applications including battery e,lectrodes, catalysts, sensors, pollution treatments, electro-magnetic shields, and structural materials. However, knowledge of their, mechanical properties, integrity, and durability is limited. This is the case because current progress in force field development,,for atomistic modeling, falls noticeably behind that of the synthesis of new 2D materials. A larger gap, in the discovery process, e,xists in terms of mechanical characterization using nanoscale experimentation. In this project we propose to address both shortcomin,g by (1) establishing a physics-based machine learning (ML) framework to select and parametrize force fields, (2) develop novel in s,itu HRTEM experiments (from RT to 500oC), and (3) explore the deformation and failure, including fracture, of a MXene material, Ti3C,2Tx, as a case study.We hypothesize that existing force fields, such as REBO, Tersoff, and ReaxFF are inaccurate when applied to mec,hanical properties of 2D materials, involving large deformations, e.g., in bending or at crack tips, because their parameterization,has not been optimized for such atomic configurations. We propose to establish ML frameworks, requiring much less domain knowledge a,nd prior experience in force field development, to parametrize these interatomic potentials using ab initio equilibrium and non-equi,librium data as ground truth. We will use genetic algorithms in the optimization step and evaluate accuracy using correlation and pr,incipal component analyses. These analyses will not only identify correlation relationships between properties and their redundancy,but also,ts using MEMS technology and employ image processing to compute deformation fields. Novel MEMS devices will be designed and built to, test deformation and failure of 2D materials, with Ti3C2Tx as a case study, in the temperature range of room temperature to 500oC.,Fracture tests, aiming to combine the power of high-resolution atomic imaging and high-resolution force/displacement measurement, wi,ll be performed to quantitatively study the effect of material chemical bonds and defects on fracture toughness of Ti3C2Tx. We will,use the combined computational-experimental approach to investigate:How do nonlinearities around the crack tip affect its propagatio,n? What is the role of lattice anisotropy on crack propagation mode?Is the agreement between MD predictions and experimental measure,ments of crack tip lattice reconstruction and interaction with defects qualitative or quantitative?For what crack sizes is the Griff,ith criterion valid? Under what conditions do nonlinearities at the crack tip become significant?The proposed research is well posit,ioned to advance ML physics-based parameterization of force fields, quantitative in situ TEM experimentation, over a range of temper,atures, and deliver fundamental knowledge about the fracture mechanism and fracture toughness of Ti3C2Tx. The computational approach,es would be readily transferable to newly discovered materials and their applications. In this regard, this project would pave the w,ay to establishing methodologies for reliability and life prediction analyses. Lastly but not least, the focus and depth of the proj,ect will be a unique opportunity for participating graduate students and post-docs to learn at the frontier of machine learning, com,putational science, and advanced materials. They will also benefit from interactions with scientists at Argonne National Laboratory,and national user facilities.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 16, 2022
- Source ID
- N000142212133
Entities
People
- Horacio D Espinosa
Organizations
- Northwestern University
- Office of Naval Research
- United States Navy