PET Programs
HPC Inspired Materials DesignBy Charles Cornwell, Sr. Research Scientist, CCM PET On-Site, HPTi |
 |
| The complete microstructure model of Aluminum with the associated finite element mesh. |
The ultimate goal of computer modeling of materials is to predict the properties and performance of materials through modeling and simulations. High performance computers allow materials scientist and engineers to define models with ever increasing size, resolution and more realistic computational algorithms. However, the increased complexity of the models and the time required to produce them places a practical limit on the use of modeling and simulation in materials research.
Help may be on the way. The Programming Environment and Training (PET) program on-sites Dr. Charles Cornwell and Dr. Ralph Noack are collaborating with Dr. Anthony Rollett at Carnegie Mellon University (CMU) to provide DoD material modelers access to accurate 3D digital microstructure models that are validated by direct comparison to experimental data and an automated microstructure and grid generation process to produce the complex microstructure models.
In recent years, significant progress in modeling the microstructure evolution of materials has been made in the area of application of continuum methods. These methods can be used to predict the final shape of a specimen following deformation and the temperature history of the material. However, these models have had little impact on the optimization of material properties because they cannot quantitatively determine, predict or manipulate the internal structure of the material. Novel materials are often manufactured via advanced processing techniques that greatly influence the microstructure arrangement of the material and thus its properties. Manipulating the microstructure of a material is the primary mechanism that materials scientists and engineers have to optimize the properties and performance of materials.
Three critical elements of this technology are needed for researchers to bring the high performance computing (HPC) capabilities of the DoD to bear on this problem 1) an efficient process for generating the microstructure arrangement of a material that makes a quantitative connection between the experimental materials and the abstract 3D digital representation used in the simulations, 2) an efficient method for generating the finite element mesh for the complex microstructure arrangement of the material at the grain level, 3) physics-based material models that have the accuracy required to predict the properties and performance of materials with computer simulations.
Researchers at CMU have developed a microstructure building process that employs a combination of experimental and computational techniques to capture the complex microstructure arrangement of materials. A Scanning Electron Microscope is used to examine the material and collect the data needed to characterize the microstructure. The data is used to construct distribution functions for the size, shape and orientation of the microstructure grains as well as the crystallographic orientation and miss-orientation between grains. It is the characterization of the material that provides the quantitative link between the experimental material and the abstract digital model. The experimental distribution functions are then used to guide the construction of the abstract 3D digital models. The characterization criterion provides a quantitative measure for gauging the accuracy and quality of the abstract model. This quantitative measure is used as the acceptance criteria in a Monte Carlo algorithm capable of sorting through tens of thousands of possible microstructure configurations to gradually refine the model until the distribution functions converge.
The grid generating algorithm decomposes the microstructure into its various components (nodes, edges, faces, cells and grains). It then starts with the basic components and begins meshing and reassembling the system. Information at the node level is used to split the edges into segments. The edges are then reassembled to form the perimeters of faces and the edge segments are used as input to a 2D advancing front grid generating algorithm to generate the mesh for the faces. Then the faces are reassembled to form the surface of the grains and this surface mesh is used as input to a 3D advancing front grid generating algorithm to mesh the grain volume. The grains are then reassembled to form the original system with the mesh. The entire process is automated and produces an accurate 3D digital representation of the microstructure of the material and the mesh needed for computational analyses of the material at the grain level.
There are a number of DoD research initiatives designed to study the evolutionary dynamics of materials that are developing new material models. At the Army Research Laboratory (ARL) Weapons and Materials Research Directorate (WMRD), the research of Scott Schoenfeld and John Clayton use microstructure models to understand the physical mechanisms responsible for the evolutionary dynamics of ductile polycrystalline materials subjected to large strains at high strain-rates (impact and penetration). Dr. Pat Baker’s group at WMRD needs microstructure models for their work simulating granular explosive materials. Similarly, the work of Robert Carter at WMRD requires meshing tools that can be applied to his work which is developing ceramic gun barrel materials. Research at the Air Force Research Lab (AFRL) in Dayton, Ohio use microstructure models to evaluate and calibrate sophisticated physics-based polycrystalline material models for the study of aircraft and engine structures. The research of Dr. Ozden Ochoa, Dr. Melanie Sarzynski and Dr. Benji Maruyama need accurate microstructure models to study the properties of carbon-foam materials. At the Naval Research Lab in Washington D. C., Andy Geltmacher and George Spanos use microstructure models to study the properties of non-magnetic steel materials that are of interest to the Navy. With all of these efforts, characterizing the complex conditions concerning the initial microstructure arrangement of the material at the grain level remains problematic.
The ARL MSRC and the PET program are committed to providing DoD researchers with both the latest in HPC resources as well as the support tools needed to make efficient use of the resources. The HPC resources extend the computational capability of DoD scientist and engineers and improve the quality of research conducted at DoD laboratories.
“These simulations (of the tungsten heavy alloy material) will enable us to model and probe local aspects of anisotropic material behavior associated with grain shapes and textures in a physically realistic manner. Such studies have not been undertaken in the past at ARL, nor have they been reported in the literature.” - John Clayton (ARL/WMRD)