First Session September 30th 2022
AI-Ready materials-science FAIR data: methods and infrastructure.
Luca M. Ghiringhelli , Physics Department and IRIS Adlershof, Humboldt-Universität zu Berlin, Germany
To accelerate the identification and design of optimal materials for a desired property or process, strategies for a well-guided exploration of the materials space are highly needed. A desirable strategy would be to start from a consistent body of experimental or theoretical data, and by means of artificial-intelligence (AI), to identify yet unseen patterns in the data, and consequentially predictive, data-driven models. This leads to the identification of materials’ (properties) maps, where different regions correspond to materials with different properties. The main challenge on building such maps is to find the appropriate descriptive parameters (called descriptors) that define these regions of interest.
Here, I present recent updates on novel methods for the AI-aided identification of descriptors and materials maps, tailored to work (also) with “small-data”, and applied to important materials-science challenges such as the prediction of mechanical properties of perovskite materials, of catalytic properties of experimentally characterized materials, and more.
Furthermore, I will introduce the NOMAD AI toolkit, an online platform for publishing and sharing curated Jupyter notebooks for the tutorial introduction of old and new AI tools and for providing an interactive access to AI workflows as published in peer-reviewed journals. In this way, new and experienced researchers can fully benefit of the community’s advancements and reproducibility in science can meet its full potential.
Recent Advancements in Artificial Intelligence and the Impact on Tribology
Marius Stan, Chicago, IL, USA
Designing improved materials and processes requires a comprehensive evaluation of data and model quality. With the volume, variety and rate of data generation continuously increasing, human analysis becomes extremely difficult, if not impossible. In this talk, the concept of “intelligent software” is discussed. The software includes elements of Artificial Intelligence such as machine learning and computer vision, coupled with reduced-order modeling and Bayesian statistics. The value of the approach is illustrated using examples of material design and real-time optimization of manufacturing processes. Furthermore, a discussion of current and future applications of AI in tribology emphasizes the positive impact on the field.
Second Session, October 10th 2022
High-throughput screening of adhesive friction in solid-solid interfaces
Maria Clelia Righi, Department of Physics and Astronomy of Bologna University
First principles high throughput calculations have been successfully applied to screen the properties of hundreds of materials in an automatized way. However, implementing a workflow for high-throughput calculations is challenging and requires robust IT infrastructures to collect, analyze, and save the data. To this end, we have designed and developed an advanced software package to study solid materials with density functional theory (DFT). It allows to carry out high-throughput screening of bulk, surface, and interface properties. We used our package to calculate important tribological figures of merit connected to dry adhesive friction ofhomogeneous interfaces. In a more recent advancement, we created a database for heterogeneous interfaces focusing in particular to metal-on-metal surfaces relevant for technological applications. By analyzing the collected data with a machine learning approach, we identified a simple linear formula for predicting the ad-hesion energy before running expensive ab initio scalculations.
First principles simulations of epitaxial interface
Noa Marom, College of Engineering Department of Materials Science and Engineering Carnegie Mellon University
Inorganic interfaces lie at the heart of semiconductor, spintronic, and quantum devices. At a hybrid interface between two dissimilar materials (e.g., a ferromagnet and a semiconductor) physical properties and functionalities may arise, which do not exist in any of the isolated components in the bulk. The resulting device performance hinges on the electronic and magnetic properties of the interface, as well as on its quality. As devices become increasingly smaller, precise control over interface structure becomes increasingly critical. At the same time, the configuration space of possible inorganic interfaces is vast and largely underexplored, owing to the almost infinite number of ways different materials can be combined to form interfaces. The experiments required to fabricate high-quality defect-free interfaces and devices are costly and time consuming. Therefore, it is unfeasible to explore the space of possible structures and compositions by experimental means alone. Computer simulations may significantly accelerate the discovery and design of new inorganic interfaces with desirable properties.
To predict the structure of domain-matched epitaxial interfaces we use a combination of lattice matching and surface matching algorithms [1]. To study the electronic properties of interfaces we use density functional theory (DFT). Within DFT, the many-body interactions between electrons are described by approximate exchange-correlation functionals. The accuracy of the results hinges on an appropriate choice of functional. We have developed a method of machine learning the Hubbard U correction added to a DFT functional by Bayesian optimization (BO) [2]. The DFT+U(BO) method balances accuracy with computational cost, enabling unprecedented simulations of large surface and interface models of interest for applications in quantum computing. These include InAs and InSb surfaces [3], which are the substrates of choice for superconductor/semiconductor Majorana devices; the HgTe/CdTe and InAs/GaSb interfaces [4], in which a 2D topological insulator phase may arise, and the EuS/InAs interface [5], which is considered as a promising candidate for the realization of a ferromagnet-semiconductor-superconductor Majorana device, which does not require an external magnetic field.
[1] S. Moayedpour, D. Dardzinski, S. Yang, A. Hwang, N. Marom “Structure Prediction of Epitaxial Inorganic Interfaces by Lattice and Surface Matching with Ogre”, J. Chem. Phys., 155, 034111 (2021)
[2] M. Yu, S. Yang, C. Wu, and N. Marom “Machine Learning the Hubbard U Parameter in DFT+U Using Bayesian Optimization”, npj Computational Materials 6, 180 (2020)
[3] S. Yang, N. Schröter, S. Schuwalow, M. Rajpalk, K. Ohtani, P. Krogstrup, G. W. Winkler, J. Gukelberger, D. Gresch, G. Aeppli, V. N. Strocov, R. M. Lutchyn, N. Marom “Electronic Structure of InAs and InSb Surfaces: Density Functional Theory and Angle-Resolved Photoemission Spectroscopy” Advanced Quantum Technologies, 5, 2100033 (2022)
[4] S. Yang, D. Dardzinski, A. Hwang, D. I. Pikulin, G. W. Winkler, N. Marom “First Principles Feasibility Assessment of a Topological Insulator at the InAs/GaSb Interface”, Phys. Rev. Mater. 5, 084204 (2021)
[5] M. Yu, S. Moayedpour, S. Yang, D. Dardzinski, C. Wu, V. S. Pribiag, N. Marom “Dependence of the Electronic Structure of the EuS/InAs Interface on the Bonding Configuration”, Phys. Rev. Mater. 5, 064606 (2021)
Third Session October 14th 2022
Application of AI/ML methods for the analysis of tribological data sets – use cases and challenges
Georg Vorlaufer, AC2T research, Austria
In the wake of current developments which are often summarized as “digital transformation”, methods of data science, in particular machine learning (ML) and artificial intelligence (AI), have been receiving a lot of attention. In combination with sufficient data (“big data”), machines can be trained to predict customer decisions, create artwork based on verbal descriptions, or identify plants based on photographs. In the industrial sector, one of the most sought-after uses cases of AI/ML is the health monitoring and prediction of the remaining service life of machines based on on-line sensor data.
At AC2T research GmbH, we use data-driven approaches to, e.g., identify friction regimes in tribometer experiments, correlate chemical information from different lubricants or detect changes in the wear behaviour of specimens. Data are usually generated on-line by sensors during tribometer experiments or by ex-post analyses such as FTIR spectroscopy.
In this talk I will present some recent examples from our institute related to the topic at hand and highlight the main findings. I will also discuss some of the lessons learned from these activities.
Predicting EHL thickness parameters by machine learning approaches
Max Marian, Pontificia Universidad Católica de Chile, Chile
Non-dimensional similarity groups and analytically solvable proximity equations can be used to estimate integral fluid film parameters of elastohydrodynamically lubricated (EHL) contacts. In this webinar, it is shown how machine learning (ML) approaches can predict relevant film parameters more efficiently than sophisticated EHL simulations and with higher accuracy and more information than analytically solvable proximity equations.
WeSST 