(4cf) Advancing Future Manufacturing By Integrating Experimental and Computational Data with Machine-Learning (ML)-Based Frameworks

Advancing future manufacturing by advancing scientific knowledge is my core motivation for the research trajectories lined up below. These trajectories span a range of materials, length-scales, properties/activities, and approaches that match my research and academic background in materials informatics and engineering physics, and share a common impetus: the understanding of fundamental physical/chemical processes for its own sake and to guide the optimization of existing technologies, and the development of new ones. In the upcoming years, my research focus will be on active learning design of experiments, synthesis and characterizations of novel crystalline materials, and data-driven atomistic modeling in advanced organic and inorganic materials. In particular, I plan to develop the machine-learning (ML)-based frameworks that integrate experimental and computational approaches for: (1) synchrotron characterization of low-dimensional materials, and (2) materials discovery via multi-fidelity Bayesian optimization. The motivation and strategies for each individual project are detailed below.

Synchrotron characterization of low-dimensional materials Interest in 2D materials and van der Waals solids is growing exponentially across various scientific and engineering disciplines owing to their fascinating electrical, optical, chemical, and thermal properties. My focus to date has been studying structural and electronic response in 2D materials [1-4] through a combination of synchrotron experiments and density functional theory (DFT). During my postdoctoral research at University of Wyoming (UWYO) I have acquired expertise on ML-based fabrication of laser-induced graphene for flexible electronics in sensing applications [5]. The work presents low-power solutions for in-space manufacturing and is thus attractive to NASA. I propose one direction to advance research in low-dimensional materials, with a focus on sensor applications:

During my PhD, I collaborated with researchers at BESSY and Australian Synchrotron to modify beamlines ad-hoc for magneto-optical spectroscopic investigation of graphene on nontransparent metallic substrates for spintronics applications [6, 7]. I aim to open these and further new collaborations, and implement the same autonomous intelligent process control done at AI Manufacturing (AIM) Center at UWYO to optimize composition-structure-property relationship of low-dimensional materials. In such study, every material in the array would be measured sequentially to look for the compound with the best properties in a fraction of a typical combinatorial study. This methodology is already developed e.g. at NIST, making it low-risk, yet it allows students to learn more about synchrotrons and automation.

Multi-fidelity modelling and Bayesian optimization Data-driven statistical approaches have been recognized as the fourth paradigm in materials science and are the focus of many funding agencies, such as DOE Integrated Computational and Data Infrastructure for Scientific Discovery, NSF Future Manufacturing, etc. While the development of many AI tools and strategies aim to a foster multidisciplinary collaborations, several bottlenecks have been identified. For instance, the diversity of measurements and simulations of materials, as well as experimental budget and time constraints are herculean challenges that need to be mitigated. It is thus critical to develop an integrated framework that efficiently shares data between experiments and simulations that helps establish peering partnerships with a focus around a scientific challenge.

I propose to develop a general framework that integrates evaluations of varying fidelities into hierarchies that share information between multiple sources and apply it to a wide range of materials discovery projects. Notably, the project will leverage adaptive tools deployed from my prior work in Bayesian optimization of laser-induced graphene [5], and multi-fidelity modelling of graphene-based single atom catalysts [8]. Preliminary investigations at UWYO have revealed in one case study that inexpensive low-fidelity data can be used to predict more costly high-fidelity properties; and in another, higher quality laser-induced graphene was produced significantly faster via ML than with human operators. The challenge is to combine the two machine learning approaches in the so-called Multi-Fidelity Bayesian Optimization (MFBO) framework. The vision is ambitious– integrating processes that are currently separate in order to leverage more information for better outcomes. This will require fundamental research in AI and ML, and the application of these novel approaches in materials science.

Teaching interests

I have taught at the pre-college, undergraduate, and graduate levels, and spoken to general audiences. I taught under-graduate and graduate-level classes while pursuing my education in Germany and Australia; and I privately tutored high school K-12 students in the German language. Pitching subject matter at the student level and motivating students were invaluable experiences to me. I strongly believe that outreach is essential to educate people about science and promoting science-based policies, and that science too stands to gain when scientists tune into people outside their field. I have been engaged in outreach since my undergraduate years. Most recently, I participated in science festivals, and gave general public talks (through the Pint of Science initiative in Germany, and TEDxCanberra in Australia. Below, I broach the subject of teaching as courses that I would like to teach.

I have experience in material informatics and engineering physics of low-dimensional structures, including both experimental work at synchrotron accelerators as well as condensed matter modelling. In addition to lower level physics courses, I would be interested to teach higher level undergraduate and graduate courses such as Data Science in Materials Science, Materials Informatics, Laser Physics, Laser Materials processing, Computational Chemistry and Modeling. If compatible with my research/course load, I would enjoy teaching a non-major “Machine Learning Materials” class as bridging the data and material science gap is crucial to enable the fourth paradigm of science and public support of science. I have already prepared and run such a course completely online on Jupyter notebook. In addition, it is a good recruiting tool for bringing students with coding experience into the materials engineering program. I would also be keen to develop an automated computations course if suitable to the department.

[1] H. Wahab, et al., The structural response of graphene on copper to surface- and interfacial-oxygen, Carbon 110 (2016) 414-425.

[2] H. Wahab, et al., Signatures of different carbon bonds in graphene oxide from soft x-ray reflectometry, X-Ray Spectrometry 44 (2015) 468– 473.

[3] H. Wahab, et al., The identification and characterisation of carbonaceous interface layers of graphene using polarisation-dependent X-ray reflectometry, Carbon 137 (2018) 252-265.

[4] H. Wahab, H.-C. Mertins, H. Timmers, T.J. Frankcombe, X-ray absorption fine structure of carboxyl and other adventitious moieties attached to copper-supported graphene, Carbon 141 (2019) 457-466.

[5] H. Wahab, et al., Machine-learning-assisted fabrication: Bayesian optimization of laser-induced graphene patterning using in-situ Raman analysis, Carbon 167 (2020) 609-619.

[6] M. Tesch, et al., Giant magneto-optical Faraday effect of graphene on Co in the soft x-ray range, Physical Review B 98(6) (2018) 064408.

[7] H.-C. Mertins, et al., Magneto-optical reflection spectroscopy on graphene/Co in the soft x-ray range, Journal of Physics: Conference Series 903(1) (2017) 012025.

[8] G.R. Hud Wahab, Patrick Johnson, Dilpuneet Aidhy, Lars Kotthoff, Multi-Fidelity Information Fusion DFT Study of Doped-Graphene Single Atom Catalysts Virtual MRS Spring Meeting and Exhibit, online, 2021.