Frontiers of Materials Award Symposium Session: Data-Driven, Machine-learning Augmented Design and Novel Characterization for Nano-architectured Materials: A Special Award Session for Symposium: Self-organizing Nano-architectured Materials
Program Organizers: Yu-chen Karen Chen-Wiegart, Stony Brook University / Brookhaven National Laboratory
Wednesday 2:00 PM
March 2, 2022
Room: 260C
Location: Anaheim Convention Center
Session Chair: Yu-chen Karen Chen-Wiegart, Stony Brook University / Brookhaven National Laboratory; Ian McCue, Northwestern University
2:00 PM Invited
Designing Nano-architectured Materials with a Machine-learning Augmented Framework: Chonghang Zhao1; Cheng-Chu Chung1; Yu-chen Karen Chen-Wiegart2; 1Stony Brook University; 2Stony Brook University / Brookhaven National Laboratory
Design parameter space for nano-architectured materials is large and complex. Even with the guidance of simulation and theoretical computation, creating nano-architectured materials, despite being at the forefront of materials science, could suffer from the common issue of intuition-driven “trial-and-error” materials discovery. While there has been a significant increase in developing data-driven, machine-learning (ML) based methods for efficient materials design, only recently have researchers begun to integrate these approaches with advanced, high-throughput characterization tools. This talk aims to provide an overview on the latest development of this exciting direction in materials science and engineering. A use case of solid-state interfacial dealloying (SSID) creating bicontinuous nanostructures will also be presented; the example will highlight the use of ML-based models and algorithms to predict materials’ synthesibility as well as the promise to apply automated and autonomous experimental approaches at synchrotron facilities to further enhance and validate the ML models.
2:30 PM
“Big Data” Characterization of Material Properties and High Temperature Kinetics: James Horwath1; Peter Voorhees2; Eric Stach1; 1University of Pennsylvania; 2Northwestern University
Competition between various coarsening and degradation mechanisms in supported nanoparticles inhibits the performance and stability of heterogeneous catalysts. While mean field models provide a physical understanding of the average behavior of the system, these models neglect local effects which are important at the nanoscale. By pairing in situ Transmission Electron Microscopy with unsupervised machine learning for automated image analysis, we quantify the degradation of supported Au nanoparticles at high temperature in real time. After developing a model to predict average particle growth as a function of chamber pressure and temperature, we use evolution trajectories from hundreds of individual nanoparticles to retrieve values for physical properties of the system with high accuracy. Further, by comparing extracted property values with those found in the literature, we characterize the system in terms of properties which are impossible to measure through traditional experiments, such as the distance over which nanoparticles interact.
2:50 PM Invited
Discovery of Nanocomposite Phase Change Memory Materials via Closed-loop Autonomous Combinatorial Experimentation: Ichiro Takeuchi1; 1University of Maryland
We are incorporating active learning in screening of combinatorial libraries of functional materials. The array format with which samples of different compositions are laid out on combinatorial libraries is particularly conducive to active learning driven autonomous experimentation. For some physical properties, each characterization/measurement requires time/resources long/large enough that true "high"-throughput measurement is not possible. By incorporating active learning into the protocol of combinatorial characterization, we can streamline the measurement and the analysis process substantially. I will describe how we have discovered a new phase change memory (PCM) material using the closed-loop autonomous materials exploration and optimization (CAMEO) strategy. The discovered PCM material has been tested in various scaled-up device formats and continues to exhibit superior performance to industrial standards. This work is performed in collaboration with A. Gilad Kusne, H. Yu, M. Li, E. Pop, and A. Mehta. This work is funded by SRC, ONR, and NIST.
3:20 PM Break
3:40 PM Invited
Intelligent Design of Additively Manufactured Architected Materials: Mitra Taheri1; 1Johns Hopkins University
To fully realize nano-architectured materials, topological and chemical control are necessary. Unfortunately, the majority of synthesis pathways for these materials geometries are “blind.” This talk presents a step forward toward closed loop manufacturing, leveraging a combination of simulation, machine learning, and processing. The results provide an improvement in spatial and microstructural optimization of melt pools for complex, architected materials produced with laser additive manufacturing.
4:10 PM
Machine Learning Based Hierarchical Multi-scale Modeling of Mechanical Deformation for Metal-matrix-nano-composites: Md Shahrier Hasan1; Wenwu Xu1; 1San Diego State University
The mechanical properties of Metal-Matrix-Nano-Composites (MMNC) have demonstrated significant enhancement with the presence of a small fraction of nano-inclusions. To understand the effect of nano-inclusions on the macro-scale or continuum level properties of MMNCs, a multi-scale modeling is necessary, one that can pass the atomistic mechanism-dependent information to the continuum calculations. In this work, a novel Machine Learning (ML) enabled hierarchical multi-scale modeling was developed by coupling the atomistic Molecular Dynamics (MD) simulations with the macro-scale Finite Element Method (FEM) to understand and make predictions on how the nano-inclusions affect the macroscopic properties of MMNCs. At first, MD simulations at various loading conditions and nano-inclusion structures were conducted to generate constitutive data which are subsequently used to train various ML classification and regression models. These ML models are implemented as a constitutive material model in the macro-scale FEM, achieving the multi-scale modeling of mechanical deformation of MMNCs.
4:30 PM
Accelerated Discovery of Multi-phase Refractory Alloys through Machine Learning Surrogate Models of CALPHAD: Samuel Price1; Ian McCue1; Jonah Erlebacher2; 1Northwestern University; 2Johns Hopkins University
There is a great need for alloys that can operate at higher temperatures than Ni-base superalloys. Refractory alloys (e.g., consisting of Nb, Ta, Mo, and W) are promising alternatives, but their single-phase microstructures impair high-temperature mechanical performance. Thus, there is an impetus to design multi-phase refractory alloys with coherent precipitates that are stable to high temperatures. The vast compositional design space for such alloys necessitates the use of computational tools. One such tool, CAlculation of PHAse Diagrams (CALPHAD), can be used to screen alloys by predicting their equilibrium phases; however, CALPHAD is too slow to screen these large datasets. Here, we have solved this issue by developing accurate surrogate models of CALPHAD using artificial neural networks. Our new approach is 20x faster than the traditional “brute force” method, making it possible to search larger alloy systems in a matter of hours and days rather than weeks and months.