Scientific Events

Realizing physical discovery in imaging with machine learning

  • Date: Sep 2, 2021
  • Time: 05:00 PM c.t. - 06:00 PM (Local Time Germany)
  • Speaker: Dr Sergei V. Kalinin
  • The Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, USA
  • Location: Online
  • Host: Prof. Dierk Raabe
Machine learning and artificial intelligence (ML/AI) are rapidly becoming an indispensable part of physics research, with domain applications ranging from theory and materials prediction to high-throughput data analysis. In parallel, the recent successes in applying ML/AI methods for autonomous systems from robotics through self-driving cars are generating enthusiasm for the potential of these techniques to enable automated and autonomous experiment (AE) in areas ranging from materials synthesis to imaging. The implementation of this vision requires the synergy of three intertwined components, including the engineering controls of the physical tools, the algorithmic developments that allow to define non-trivial exploratory patterns, and physics-driven interpretation. In this presentation, I will discuss recent progress in automated experiment in electron and scanning probe microscopy, ranging from feature finding to physics discovery via active learning. The applications of classical deep learning methods in streaming image analysis are strongly affected by the out of distribution drift effects, and the approaches to minimize though are discussed. We further present invariant variational autoencoders as a method to disentangle affine distortions and rotational degrees of freedom from other latent variables in imaging and spectral data. The analysis of the latent space of autoencoders further allows establishing physically relevant transformation mechanisms. Extension of encoder approach towards establishing structure-property relationships will be illustrated on the example of plasmonic structures. Finally, I illustrate transition from post-experiment data analysis to active learning process. Here, the strategies based on simple Gaussian Processes often tend to produce sub-optimal results due to the lack of prior knowledge and very simplified (via learned kernel function) representation of spatial complexity of the system. Comparatively, deep kernel learning (DKL) methods allow to realize both the exploration of complex systems towards the discovery of structure-property relationship, and enable automated experiment targeting physics (rather than simple spatial feature) discovery. The latter is illustrated via experimental discovery of the edge plasmons in STEM/EELS and ferroelectric domain dynamics in PFM. This research is supported by the by the U.S. Department of Energy, Basic Energy Sciences, Materials Sciences and Engineering Division and the Center for Nanophase Materials Sciences, which is sponsored at Oak Ridge National Laboratory by the Scientific User Facilities Division, BES DOE. [more]

Modelling the combustion of metal powders in laminar and turbulent flames

Besides their ubiquitous use in load-bearing structures, metals also possess qualities of energetic materials. Lithium, for example, is a common fuel in batteries, while aluminum is frequently added to solid rocket propellants and used in pyrotechnics. At high temperatures, metal powders can be burned in air in a similar way to hydrocarbon fuels, releasing chemically stored energy as sensible heat. Contrary to hydrocarbon combustion, however, the main reaction products are solid oxide particles that can, in principle, be retrieved from the exhaust fumes. This amenability to oxide sequestration has stimulated the idea of harnessing metal powders as recyclable energy carriers which are burned, retrieved and, subsequently, recharged by a reduction process based on clean primary energy. Conceptually, the metal powders are akin to high-temperature batteries, serving as a means to buffer the large spatial and temporal intermittency associated with renewable energy sources. Motivated by the potential use of metal powders as recyclable fuels, we qualitatively discuss the physical and chemical processes involved in the combustion of a single metal particle and of metal dusts, respectively. Subsequently, a population balance model for predicting the size distribution of the oxide smoke precipitating in the vicinity of a single burning aluminum particle is presented. Here, we specifically focus on the kinetic rates that control the phase transitions and smoke dynamics, integrating recently developed detailed kinetics for gas phase and heterogeneous surface reactions. The population balance equation governing the oxide size distribution is solved with the aid of a tailored adaptive grid method. An alternative, potentially more economical solution approach we propose is based on an embedded reduced order representation of the particle size distribution that is informed by a training step. The accuracy and convergence properties of this method are investigated for a simplified test case involving particle growth and dispersion in a laminar plane jet. In the final part of the seminar, the physical description, from an Eulerian viewpoint, of metal powders is discussed with a particular emphasis on the ramifications of carrier flow turbulence. In order to account for the small-scale interactions between dispersed particles and the ambient gas phase, the population balance equation governing the metal powder or oxide smoke is integrated into a probabilistic description that naturally accounts for the variability among independent realizations of a turbulent, particle-laden flow. Owing to the high dimensionality of the resulting transport equations, a stochastic solution approach based on Eulerian stochastic fields is proposed for which we show preliminary accuracy and convergence analyses.

Electrochemical Capacitance under Confinement: Implications for Electrochemical Energy Storage and Conversion

Electrochemical Capacitance under Confinement: Implications for Electrochemical Energy Storage and Conversion
Abstract: Many layered materials of interest for electrochemical energy storage and conversion applications are flexible hosts whose interlayers can be expanded to accommodate not just ions but also solvents, organic molecules, polymers, and organometallics. When these “hybrid” materials are placed into an electrochemical environment, the distinction between surface and bulk becomes blurred since the electrochemical interface can now be viewed to extend into the interlayer. During this seminar, I will discuss fundamental aspects of charge storage at electrochemical interfaces and how interfacial charge storage and reactivity change under confinement. I will also describe synthesis of hybrid layered materials and the use of in situ and operando characterization to understand the relationships between structure and composition and the resulting electrochemical reactivity. [more]

Artificial Intelligence for Engineering Design and Computational Mechanics

Engineered systems are an indispensable part of our modern life with far-reaching applications that include aerial and ground transportation, electronics, large-scale structures, and medicine. The ever-evolving societal, environmental, and cultural awareness calls for significantly complex systems with unprecedented properties that reliably meet stakeholders’ demands under extreme conditions. To accelerate the design and deployment of such systems while reducing the reliance on costly and time-consuming experiments, it is necessary to develop advanced computational methods that streamline their design and analysis process. In this talk, I will present some of our recent works for solving challenging problems in engineering design, solid mechanics, and fluid dynamics. In particular, I will demonstrate how we can (1) accelerate multiscale simulations of casting materials ten times via mechanistic reduced order models, (2) surrogate plastic and history dependent deformation of fiber composites with deep learning, (3) optimize material composition with latent map Gaussian processes and Bayesian optimization, and (4) solve partial differential equations with transfer learning. [more]

STZ vortex unit – the key to understand and control shear banding in metallic glasses

Transition to a Group Leader position at a German University

Where: virtual on Zoom (link follows) [more]

Metal energy carriers: renewable fuels of the future

Metal powder has superior energy density compared to fossil fuels and hydrogen. Therefore, metal powders have gained interest as a material for energy storage. The main benefits of metal fuels are that they do not produce CO2 emissions during combustion, they have the potential to be retrofitted in existing coal power plants and they can fit into the existing fuel transportation infrastructure. Furthermore, this enables the production of sustainable energy since metal fuels can be regenerated from metal oxides, using hydrogen from renewable sources. In this presentation, the main characteristics of metal fuels are presented with a final focus on clean combustion. A series of burners has been developed: - single particle or fuel jet in a micro burner to study single particle combustion and particle-particle interaction - Bunsen-type burner for stabilizing laminar and weakly turbulent premixed flames - Tornado-swirl burner First numerical studies are also started for comparison. Furthermore, a 100 KW demonstrator set-up is developed to demonstrate clean combustion to produce steam (placed at Swinkels brewery and Metalot centre). Studies to scale up are also conducted. The main objective of this practical systems is the development of an integrated flexible metal fuel burner with a capacity of 100 KW (TRL5). This is an essential step towards implementation of this sustainable technology. This project forms the basis to further develop full scale burners with a capacity of 10 MWth. The development of the prototype burner is executed by a consortium which covers the entire supply chain. This includes the production of metal powder, fuel preparation, burner and combusted product handling. The industrial partners have broad experience in metal powder supply, dense energy carriers and operating coal fired power plants. Furthermore, techno-economic analyses and the assessment of retrofit potential to existing assets will be carried out. Status-quo will be presented [more]
Show more
Go to Editor View