Projects

Skunkworks members work on a range of projects which we list here.

Fall 2018

Increasing Reliability of Detecting Bleed Sites in Angiograms (with Prof. Varun Jog, Dr. Mark Kleedehn, and Prof. Po-Ling Loh, Univ. of Wisconsin)
This project seeks to ihelp radiologists identify bleeds in angiograms to enable more effective embolization.
Determining Molecular Properties (with Luisa Meyer, Silatronix)
This project seeks to use standard organic compound images to predict compound properties for industrial applications.
Perovskite’s for Solar Photovoltaic and Fuel Cell Applications (with Dr. Ryan Jacobs and Mr. Benjamin Afflerbach, Univ. of Wisconsin)
This project seeks to predict new perovskites halides and oxides that are stable and have optimal electronic structure properties for light adsorption and oxygen catalysis.
Ductile to Brittle Temperature Transition Shifts in Reactor Pressure Vessel Steels (with Prof. G. Robert Odette, UCSB) 2015-ff
This project seeks to support safe life extension of nuclear reactors by predicting changes in the mechanical properties of steel components subject to long term irradiation.
Optimizing Concrete Composition (with Prof. Steve Cramer, Univ. of Wisconsin) 2016-ff
This project seeks to make cheaper and stronger concrete by understanding the influence of its ingredients on its mechanical properties.
Bulkier Metallic Glasses (with Prof. Izabela Szlufarska, Prof. John Perepezko, Univ. of Wisconsin, Dr. Logan Ward, Univ. of Chicago) 2015-ff
This project seeks to enable the formation of larger metallic glass components by mining data on existing metallic glasses.
Accelerating Electron Microscopy Simulations (with Prof. Paul Voyles, Univ. of Wisconsin) 2015-ff
This project applies informatics tools to enable rapid simulation of highly accurate electron microscopy images from atomic structure.
Automated Defect Characterization in STEM Images (with Dr. Kevin Field, ORNL) 2017-ff
This project explores seeks to accelerate nuclear materials development by automating the assessment of radiation damage in steels with machine vision.
Machine Learning software for materials science (MAST-ML package) (with Dr. Ryan Jacobs, Univ. of Wisconsin) 2016-ff
This project is developing new tools for machine learning applications for materials as part of the MAterials Simulation Toolkite – Machine Learning (MAST-ML).

Spring 2018

Ductile to Brittle Temperature Transition Shifts in Reactor Pressure Vessel Steels (with Prof. G. Robert Odette, UCSB) 2015-ff
This project seeks to support safe life extension of nuclear reactors by predicting changes in the mechanical properties of steel components subject to long term irradiation.
Optimizing Concrete Composition (with Prof. Steve Cramer, Univ. of Wisconsin) 2016-ff
This project seeks to make cheaper and stronger concrete by understanding the influence of its ingredients on its mechanical properties.
Bulkier Metallic Glasses (with Prof. Izabela Szlufarska, Prof. John Perepezko, Univ. of Wisconsin) 2015-ff
This project seeks to enable the formation of larger metallic glass components by mining data on existing metallic glasses.
Accelerating Electron Microscopy Simulations (with Prof. Paul Voyles, Univ. of Wisconsin) 2015-ff
This project applies informatics tools to enable rapid simulation of highly accurate electron microscopy images from atomic structure.
Automated Defect Characterization in STEM Images (with Dr. Kevin Field, ORNL) 2017-ff
This project explores seeks to accelerate nuclear materials development by automating the assessment of radiation damage in steels with machine vision.
Noninstrusive Learning of Dynamical-System Models (with Prof. Benjamin Peherstorfer, Univ. of Wisconsin) 2017-ff
This project focuses on obtaining reduced models of dynamical systems in order to improve speed and accuracy of modeling new systems.
Predicting nuclear fuel evolution (with Dr. Baptist Mouginot, Prof. Paul Wilson, Univ. of Wisconsin) 2017-ff
This project is using neural networks to massively accelerate simulations of element production during nuclear fuel burnup.
Machine Learning software for materials science (MAST-ML package) (with Dr. Ryan Jacobs, Univ. of Wisconsin) 2016-ff
This project is developing new tools for machine learning applications for materials as part of the MAterials Simulation Toolkite – Machine Learning (MAST-ML).

Fall 2017

Ductile to Brittle Temperature Transition Shifts in Reactor Pressure Vessel Steels (with Prof. G. Robert Odette, UCSB)
This project seeks to support safe life extension of nuclear reactors by predicting changes in the mechanical properties of steel components subject to long term irradiation.

Optimizing Concrete Composition
This project seeks to make cheaper and stronger concrete by understanding the influence of its ingredients on its mechanical properties.

Bulkier Metallic Glasses (with Prof. Izabela Szlufarska, Prof. John Perepezko, Univ of Wisconsin)
This project seeks to enable the formation of larger metallic glass components by mining data on existing metallic glasses.

Accelerating Electron Microscopy Simulations (with Prof. Paul Voyles, Univ of Wisconsin)
This project applies informatics tools to enable rapid simulation of highly accurate electron microscopy images from atomic structure.

Automated Defect Characterization in STEM Images (with Dr. Kevin Field, ORNL)
This project explores seeks to accelerate nuclear materials development by automating the assessment of radiation damage in steels with machine vision.

Noninstrusive Learning of Dynamical-System Models
This project focuses on obtaining reduced models of dynamical systems in order to improve speed and accuracy of modeling new systems.

Predicting Perovskite Catalysts (with Dr. Ryan Jacobs, Univ of Wisconsin)
This project has explored a broad array of perovskite oxide catalysts seeking to find new stable materials for various catalysis applications.

Fall 2016 – Spring 2017

Predicting Novel Advanced Oxides
This project supports the development of new Perovskite oxides for advanced technologies (e.g. catalysis, piezoelectrics, and electronics) by replacing, time consuming computations with efficient machine learning algorithms.

Optimizing Concrete Composition
This project seeks to make cheaper and stronger concrete by understanding the influence of its ingredients on its mechanical properties.

Predicting High Melting Temperature Alloys (with data from Dr. Bryce Meredig, Citrine LLC)
This project attempts to predict alloys for high temperature applications by learning from existing data on the melting temperatures of materials.

Fall 2015 – Summer 2016

Atom Probe Image Reconstruction (Justin Schrimmer, Sam Gardner (UW-Milwaukee), with Prof. Rebecca Willett, Prof. Dane Morgan, started 12/2015). As scientist and researchers have begun to look at materials on the atomic scale, Atom Probe Tomography (APT) has come to the forefront of material analysis. This technology allows for single atoms to be separated from its material using electric fields with laser assistance. The atoms are then collected with a detector and mass spectrometer, and relatively simple physics models used to reconstruct form where the atoms originated. However, the accuracy of this reconstruction is rather uncertain, and atom position are often about 1 nm away from their true position, making it generally impossible to identify crystal planes, dislocation, and other atomic structures. In this project, we are using informatics tools such as deconvolution methods to take APT data sets and attempt to reconstruct the original structure with dramatically enhanced resolution.

Image Quality Assurance (Andrew Duplissis, with Prof. Tim Szycykutovicz, Prof. Rebecca Willett, started 02/2016): When dealing with computed tomography, or CT, scans, often times diagnoses are based solely on how the picture is presented. If you imagine a picture as a matrix, where each pixel has a certain coordinated and a different value stored in its location, then you would see that a green pixel has a much different value than a red pixel. This same general concept applies to CT images, except instead of the number corresponding to a color, it corresponds to a number with units of Hounsfield. When using these images as diagnostic tools a change as little as 32 Hounsfield units can result in an inability to see a lesion or tumor, this project is related to this; it has a goal to be able to detect when they have acquired a poor image and tell the technician such so the scan can be redone. We will be attempting to do this from a data base of 70,000 images all labeled with high or low image quality, and all having 20+ variables associated with them. We will be starting with the 6 we think to be most important and see if we can draw up a relationship between those 6 values and whether or not the image is of high or low quality.

Ductile to Brittle Temperature Transition Shifts in Reactor Pressure Vessel Steels (Josh Perry, Henry Wu, with Prof. Dane Morgan and Prof. Bob Odette (UCSB), started 09/2015)): As steel in a nuclear reactor is exposed to neutron radiation, the crystal structure degrades in a way that makes the metal more brittle, eventually to the point that it is in danger of cracking. This process is the main limit on how long reactors can operate before they need to be decommissioned, and is dependent on a variety of factors including amount of various impurities in the steel, the total radiation exposure, the rate of exposure, and the temperature of the steel when it was exposed. We are studying a dataset with 1550 data points and 9 independent variables, using Gaussian kernel ridge regression and other informatics methods, to predict the behavior of conditions that cannot be easily explored with present experiments.

Impurity Diffusion Informatics (Aren Lorenson, Liam Witteman, Haotian (Will) Wu, Ben Anderson, Henry Wu, with Prof. Dane Morgan, started 02/2015): The goal of the project is to predict impurity vacancy mediated diffusion energy barriers in binary FCC metal systems. This is a useful quantity to consider for many materials areas, e.g., when modelling sublattice ion defects in compound semiconductors, such as P in InP, or the microstructural evolution of steels shielding nuclear reactors subject to radiation accelerated Cr and Ni diffusion. In this project, we take existing, incomplete materials data generated by first-principles ab initio computations and achieving the same outcome in orders of magnitude less time.

Optimized Compositions for Al-Based Amorphous Alloys (Zach Jensen, with Prof. Izabela Szulfarska, Prof. Dane Morgan, and Prof. John Perepezko, started 9/2015) Amorphous (or glassy) metals represent a unusual form of metals with structures that are disordered, like window glass. Metallic glasses have many exceptional properties, including corrosion resistance, elastic response, wear resistance, and hardness. However, glasses are often hard to form, requiring very rapid quenching, allowing only small samples to be made. Aluminum (Al) glasses are particularly exciting due to their light weight and low cost. Researchers have experimentally examined many compositions to attempt to find good glass forming Al alloy using, e.g., techniques such as Differential Scanning Calorimetry. This project combines the results of all these experiments and looks for trends in glass forming properties such as the crystallization temperature using machine learning models. By analyzing the trends we can determine compositions that have strong glass forming properties as well as analyze the effects of individual elements and combinations of elements on the glass forming ability.

Other projects under development

New Thermoelectric Materials
Protein Structure Optimization
Steels Optimization for Improved Water Heaters (with AO Smith)