All of the selected research activities are designed to introduce undergraduates to the application and development of data science techniques for materials engineering problems. All research projects have an aspect related to MI. The students will be formally introduced to MI techniques through the boot camps, and spend most of the time interacting with their faculty and graduate student mentor.
Self-Driving Robo-Fluidic Lab for Accelerated Discovery of Clean Energy Materials
Milad Abolhasani, Chemical & Biomolecular Engineering Associate Professor
The discovery of new solution-processed photonic materials for energy technologies typically takes >10 years and >$100M of R&D cost, using current Edisonian chemical reaction development techniques. In this REU project, the student will work with a team of scientists in the Abolhasani Lab to integrate robotics, microfluidics, and colloidal nanoscience with data science to build a self-driving lab to accelerate the discovery of clean energy materials. The self-driving lab will utilize a robo-fluidic (robotics microfluidics) nanomaterial synthesizer operated by a machine learning (ML)-guided decision-making algorithm to accelerate synthetic path discovery and synthesis science studies of clean energy nanomaterials. Specifically, we will focus on metal halide perovskite (MHP) quantum dots (QDs), owing to their size- and composition-tunable optical and optoelectronic properties. With its autonomously reconfigurable flow reactors, the self-driving fluidic lab will rapidly discover the optimal formulation of high-performing MHP QD inks to be continuously manufactured using scalable reactors. The self-driving fluidic lab will enable closed-loop discovery and end-to-end manufacturing of MHP QD inks by addressing the following problems: (i) trial and error-based QD discovery methods, and (ii) poor size dispersity of batch-synthesized QDs, resulting in surface trap states. The resulting autonomous robotic experimentation strategy can reduce the discovery and synthesis optimization time of clean energy materials from years to months.
Robot-in-the-loop, data-driven experimental research in thin film photovoltaics to pinpoint the origins of degradation
Aram Amassian, MSE Professor
Organic photovoltaic (OPV) materials have recently achieved a certified power conversion efficiency (PCE) of 18.5% and there are several promising demonstrations of scalable coating of OPV devices and modules using green solvents. However, the stability of these solar cell materials and devices continues to plague this technology and is widely seen as a primary bottleneck to its market adoption1,2,3. Addressing this problem with the help of data science and informatics tools, including machine learning, may yield significant new breakthroughs where traditional approaches have so far failed. For this purpose, the Amassian group has recently automated the formulation and printing of OPV materials with the aid of robotics, and automated characterization using optical, Raman, and chemical spectroscopies. This allows for high throughput assessment of material degradation and pinpoint some of the mechanisms of degradation following exposure to various external stressors. The workflow automation should enable data digitization efforts which synergize with data science and ML techniques implemented with the help of the Yingling group. REU students will be exposed to various aspects of fundamental organic photovoltaics materials and device-related research, as well as materials and device degradation concepts, and will help design robot-in-the-loop experimental campaigns that will be augmented with data-science methods to help assess various hypotheses of degradation as well as identify new and emerging trends in material and device degradation.
Data-Driven Characterization of Energy Material Deformation During Ion Insertion
Veronica Augustyn, MSE Associate Professor
Ion insertion into transition metal oxides forms the basis of several technologies, including lithium-ion batteries and electrochromics. During ion insertion, the transition metal oxide can undergo stress that affects the lifetime and power capability of the electrode. The goal of this project is to utilize data science to understand the relationship between material deformation and ion insertion, including spatial heterogeneity, in polycrystalline thin film transition metal oxide electrodes.[1, 2] Operando atomic force microscopy dilatometry of transition metal oxides during ion insertion will be performed at different timescales and positions on the electrode. The purpose of this REU project is to utilize data science techniques to analyze the resulting operando atomic force microscopy data from several transition metal oxides undergoing ion insertion. Students will receive training in materials synthesis, electrochemistry, and data analysis. The student will also be exposed to other research projects in the Augustyn group, that range from CO2 capture to electrode architectures for energy storage.
Machine Learning Methods for Calculating Vacancy Formation Energies in High Entropy Ceramics
Donald Brenner, MSE Professor
High entropy (HE) materials are typically defined by having five or more components in roughly equimolar proportions on a single lattice such that the configurational entropy contribution to the free energy becomes comparable to the enthalpic contribution.8 In this stability regime a given composition can have a stable crystal structure that is different from that of the constituents. It also leads to challenges in materials design. For example, HE carbides (HECs) are composed of two interpenetrating fcc lattices, one with carbon and one with five or more other cations arranged randomly. For a five species composition, a carbon vacancy has 210 possible near neighbor configurations; the number of possible configurations goes up to over 107 when also considering a second shell of cations. Clearly calculations of carbon vacancy concentrations, which are common defects that can affect a number of thermal-mechanical properties,9,10 are challenging with direct methods such as Density Functional Theory (DFT). REU students will be trained to run DFT calculations using VASP to find carbon vacancy formation energies for select configurations, and then use these calculations to help train a neural network (NN) from which energies can be calculated for other configurations. To start, we currently have a database of 40 carbon vacancy formation energies for nine different five-cation HEC compositions (a total 0f 360 energies), and a dedicated queue on the NC State High-Performance Computing Cluster to carry out additional calculations11. The goals of this research are for the students to (1) learn how to systematically carry out DFT calculations to expand this data base for NN fitting and model validation; (2) determine a database of not only energies but also properties such as valence electron concentrations and relative bond distances for species surrounding a carbon vacancy that optimize predictions of vacancy formation energies within a NN; and (3) understand how NNs are developed and used, as well as their limitations.
Using data to design LEDs emitting in the ultraviolet for disinfection applications
Ramon Collazo, MSE Associate Professor
When exposed to ultraviolet (UV) light with wavelength around 265 nm, DNA or RNA of bacteria, viruses, and fungi can be denatured. For this purpose, UV irradiation is commonly used in water treatment plants and more recently for surface disinfection in various environments. Today, mercury lamps are used as the dominant technology to provide disinfecting UV light12. However, despite strong research efforts and interest, state-of-the-art UV-LEDs suffer from low power output and low overall efficiency. Commercially available LEDs have external quantum efficiencies (EQE) < 3%, and even lower wall-plug efficiency. These numbers are especially low when compared to III-nitride-based blue LEDs, which have evolved to achieve > 80% EQE. The main challenges towards the realization of UVC LEDs for disinfection applications include control of defects in the epitaxy process, optimization of p- and n-contacts, device structure design and low light extraction efficiency on the substrate side. Prospective REU students will be able to address any of these challenges in the Collazo group where the complete UVC LED process cycle, from simulation to testing, has been implemented. Based on this the following sub-research projects are available to the REU students: (a) implement point defect control and doping schemes for AlGaN layers and devices during the metal-organic chemical vapor deposition process, (b) simulate and test novel approaches towards better carrier injection in UVC LEDs, (c) improve the performance of electrical contacts, (d) support the fabrication process for LEDs with the goal to improve reliability and performance, (e) develop new light extraction schemes including nanopatterning and photonic crystals, or (f) characterize the optoelectronic properties of these UVC LEDs. Any of these efforts offer the student the opportunity to work with experienced scientists and exposure to state-of-the-art semiconductor science and technology, including SILVACO simulation, MOCVD growth, clean room technology, lithography, and etching methods. Automated probes stations and the lithographic ability to provide multiple LED designs within a wafer allows for the generation of relevant data to determine efficiency, reproducibility, and reliability among a variety of designs and dimensionality. The collected data will be used to train relevant machine learning schemes for the generalization of design rules that could be applied in future TCAD simulations of these state-of-the-art UV LEDs.
Lightweight, high strength, and corrosion-resistant alloys at the low material cost
Rajeev Gupta, MSE Associate Professor
Materials beyond conventional property limits are needed to meet current technological, socio-economic, and environmental challenges. The properties, including strength and corrosion resistance of commercial alloys, are limited by conventional compositions and manufacturing technologies. Recent research on multi-principal element alloys (MPEAs) opened up a waste compositional space to be explored. Our ongoing research has shown that AlFeMnSi-based MPEAs exhibit high corrosion resistance, high strength, and low density at a lower material cost1. We have produced 15 different AlFeMnSi-based alloys and investigated the microstructure and corrosion performance, which depend on the composition. Experimental optimization of the composition is a cumbersome process. We propose to utilize machine learning tools to determine the alloy composition resulting in the highest corrosion performance. Our experimental results will be used in training the ML models. Selected alloy compositions will be produced using arc melting and their corrosion performance will be tested.
Dynamic Interplay between Phase transformation and Material Degradation in Extreme Environments
Bharat Gwalani, MSE Assistant Professor
The complex and often interconnected processes that occur when materials are subjected to multi-extreme conditions such as high temperatures, mechanical stresses, high pressures, radiation, corrosion, and other harsh environments are difficult to understand and predict based on current knowledge base. This interplay can have significant implications for the performance, reliability, and durability of materials in various applications, including aerospace, nuclear, energy, and defense. Phase transformation refers to the change in the structure or composition of a material, often accompanied by changes in its physical, mechanical, and chemical properties. Material degradation, on the other hand, refers to the gradual deterioration of a material’s properties due to various mechanisms, such as mechanical wear, corrosion, radiation damage, thermal cycling, and fatigue. In extreme environments, phase transformation and material degradation can interact in complex ways, often influencing each other’s behavior. For example, the formation of new phases during phase transformation can affect material degradation by altering the material’s microstructure, grain boundaries, and defect concentrations, which can impact its mechanical and chemical properties. Similarly, material degradation processes can influence phase transformation by changing the availability of certain elements or altering the local conditions, such as temperature, pressure, and chemical environment, which can affect the kinetics and thermodynamics of phase transformations. Understanding the dynamic interplay between phase transformation and material degradation in extreme environments is crucial for designing materials with enhanced performance and durability. In this project, the student will study the changes in corrosion resistance of a metallic system undergoing thermo-mechanically induced phase transformation.
AI-Driven Control of Printed Organohydrogels through Ultrafast Imaging
Lilian C. Hsiao, CBE Assistant Professor
Global manufacturing of advanced materials in high-performance applications such as personalized medicine, soft machinery, and sustainable electronics is driven by ever-increasing efficiency through a combination of computation and real-time observations. These materials impart extreme performance in mechanical, electrical, and transport properties when nanoscale constituents can be precisely placed at specific locations. However, these extremely small components are sensitive to variations in processing conditions in additive manufacturing, migrating significantly with small fluctuations in shear stress and temperature and causing undesirable deviations in the product performance. Although optimal control platforms have been successfully implemented for a half century for macroscopic chemical plants, they have not been applied to maintain the stability and consistency of materials at the micron scale. This project aims to apply artificial intelligence-driven control algorithms to the in-line microstructural control of 3D-printed thermosensitive nanoemulsions used in drug delivery applications. Specifically, students will (1) build and characterize an in-line microscopy module for an extrusion-based printer with feedback control of temperature and pressure, (2) use deep learning to infer final product performance through rheological and cluster signatures in a flowing system, and (3) implement model predictive control of nanoscale constituents based on real-time image processing. This platform will be designed to intelligently extract the microstructure of the hydrogel ink components from their microscopy images, efficiently compute the relevant structural parameters and label their quality, and automatically adjust the process inputs to regulate the 3D printing process at its performance setpoint. Successful implementation of the work will have significant impact in the future manufacturing of a broad class of soft materials, such as composite elastomers, colloidal pigments, and cargo-laden hydrogels.
Materials Informatic Approach for Study of Cellulose Bioplastics
Melissa Pasquinelli, Forest Biomaterials Professor
Cellulose acetate (CA) is the most prolific cellulose ester, found in industrial and consumer products like films, cosmetics, and screens. Its diverse applications are credited to its superior characteristics which are highly dependent on acetyl group degree of substitution (DS). However, a major barrier in CA processing it’s thermal limitations. In pure forms, CA does not exhibit thermoplastic behavior, requiring chemical modifiers in order to be melt extruded. These modifiers, typically referred to as plasticizers (pz), expand CA’s thermal processing window by reducing thermal transition temperatures like glass transition temperature and melt temperature. Plasticizers change other properties; tensile strength, ductility, dielectric constant, etc., thus requiring extensive characterization. Ninety percent (90%) of plasticizers on the market are petrol-based with phthalates contributing up to 60% (by weight) of plastic products. However, phthalates are toxic and have been found to be weak endocrine disruptors and androgen blocking chemicals. This is concerning as plasticizer migration is known to occur at elevated temperature and is a poorly characterized feature of CA-pz systems. In order to uncover and examine non-toxic alternatives to phthalates, a materials informatic approach is required. This work will include (1) a data-driven approach for selection of new plasticizers, (2) the use of molecular dynamics simulations to model and characterize plasticizer migration, and (3) application of machine learning algorithms to aid in CA-pz property prediction.
Data Driven Quantum Tunneling Molecular Devices
Martin Thuo, MSE Professor
Advances in miniaturization of electronics are predicted to slow down in part due to challenges in fabrication of sub-nanometer components and associated high energy needs. The latter is predicted to limit our advances by 2035. One strategy to overcome this challenge is to exploit low-energy quantum charge transport processes like tunneling. These processes rely on lowering the charge transport distant, a fete often achieved by placing a single molecule between two electrodes. Unimolecular devices are appealing, in part due to; i) an extensive understanding of single molecule properties (through a diverse set of characterization tools), ii) synthesis-drive wave function tuning, and iii) a plethora of theoretical tools to predict molecular properties. A major caveat in this approach is the need to adopt an ensemble (self-assembled monolayers) in lieu of a single molecule given challenges in fabricating devices with the latter. To understand such an ensemble, and associated conformational space, there is a need to further interrogate data from different characterization methods. We are interested in looking for correlation in data across a homologous series and different characterization methods. The resulting analysis will lead to data-driven design rules that enable in silico predictive models in the design of functional tunneling-based molecular electronics.
Data Driven Design of NP-based Materials
Yaroslava Yingling, MSE, Biomedical Engineering, and Physics Professor
Yingling’s group uses a large-scale all-atom molecular dynamics (MD) simulation approach to predict physical, chemical, optical, and electromagnetic behavior of a wide range of organic and inorganic materials. As an example, iron nanoparticles with ligand surface functionalization can develop a certain level of biocompatibility or express controlled polarizability under external magnetic field application. Similarly, a lot of interest attracts cellulose nanocrystals (CNC)-based materials, which are well-known for being chiral materials with promising optical and mechanical tunability depending on the assembly and ordering parameters. Such materials with peculiar chiral multilayer ordered structure offer brilliant, iridescent colors typically observed in birds, butterflies, fish, beetles, veggies and fruits.
Another important aspect of our research is studying ligands. Ligands are organic molecules or ions that bind to the surface of nanoparticles, forming a layer that stabilizes and modifies their surface properties. Nanoparticle ligand chemistry involves the study of various factors such as ligand structure, binding strength, and orientation, as well as their interaction with the nanoparticle core and the surrounding environment. Thus, understanding the chemistry of nanoparticle ligands is crucial for harnessing the full potential of NP-based materials. Wide portfolio of available types and configurations of MD-simulated systems allow us to work on complex approaches towards development of sustainable, environmentally friendly materials with controlled behavior and characteristics on-demand.
The REU student will use the data generated by in-house developed models like, for example, bond-polarizability calculation approach, which involves the calculation of system polarizability tensors, refractive index tensors, and in-plane and out-of-plane birefringence values. The measured material’s performance and related properties will be used to train the MI model. MI structure-processing-property relations will be suggested by analyzing materials data sets from experimental and computational studies with statistical ML algorithms and propose optimal de novo materials design with tailored properties. The predicted materials will then be tested and used to validate or adjust the Materials Informatics model.
Research Group Website
Machine Learning for Predicting Microstructural Behavior
Mohammed Zikry, MAE Zan Prevost Smith Professor
This project will involve using computational data from finite element predictions and machine learning approaches to predict and optimize microstructural behavior related to grain boundaries, grain orientations, and defects, such as dislocations, vacancies, and stacking faults.
(1) Ghasemi, M.; Balar, N.; Peng, Z.; Hu, H.; Qin, Y.; Kim, T.; Rech, J. J.; Bidwell, M.; Mask, W.; McCulloch, I.; et al. A Molecular Interaction–Diffusion Framework for Predicting Organic Solar Cell Stability. Nat. Mater. 2021, 20 (4). https://doi.org/10.1038/s41563-020-00872-6.
(2) Soon, Y. W.; Cho, H.; Low, J.; Bronstein, H.; McCulloch, I.; Durrant, J. R. Correlating Triplet Yield, Singlet Oxygen Generation and Photochemical Stability in Polymer/Fullerene Blend Films. Chem. Commun. 2013, 49 (13). https://doi.org/10.1039/c2cc38243a.
(3) Clarke, A. J.; Luke, J.; Meitzner, R.; Wu, J.; Wang, Y.; Lee, H. K. H.; Speller, E. M.; Bristow, H.; Cha, H.; Newman, M. J.; et al. Non-Fullerene Acceptor Photostability and Its Impact on Organic Solar Cell Lifetime. Cell Reports Phys. Sci. 2021, 2 (7). https://doi.org/10.1016/j.xcrp.2021.100498.
(1) Wang, R.; Mitchell, J.B.; Gao, Q.; Tsai, W.-Y.; Boyd, S.; Pharr, M.; Balke, N.; Augustyn, V.Operando Atomic Force Microscopy Reveals Mechanics of Structural Water Driven Battery-to-Pseudocapacitor Transition. ACS Nano 2018, 12 (6) 6032.
(2) Augustyn, V.; Wang, R.; Balke, N.; Pharr, M.; Arnold, C.B. Deformation during Electrosorption and Insertion-type Charge Storage: Origins, Characterization, and Design of Materials for High Power. ACS Energy Letters 2020, 5, 3548.
(1) Feng, L.; Fahrenholtz, W. G.; Brenner, D. W. High-Entropy Ultra-High-Temperature Borides and Carbides: A New Class of Materials for Extreme Environments. Annu. Rev. Mater. Res. 2021, 51 (1). https://doi.org/10.1146/annurev-matsci-080819-121217.
(2) Citrine Informatics: The AI Platform for Materials Development https://citrine.io/ (accessed Aug 25, 2021).
(3) Rost, C. M.; Borman, T.; Hossain, M. D.; Lim, M.; Quiambao-Tomko, K. F.; Tomko, J. A.; Brenner, D. W.; Maria, J.-P.; Hopkins, P. E. Electron and Phonon Thermal Conductivity in High Entropy Carbides with Variable Carbon Content. Acta Mater. 2020, 196. https://doi.org/10.1016/j.actamat.2020.06.005.
(4) Hossain, M. D.; Borman, T.; Kumar, A.; Chen, X.; Khosravani, A.; Kalidindi, S. R.; Paisley, E. A.; Esters, M.; Oses, C.; Toher, C.; et al. Carbon Stoichiometry and Mechanical Properties of High Entropy Carbides. Acta Mater. 2021, 215. https://doi.org/10.1016/j.actamat.2021.117051.
(5) Daigle, S. E.; Brenner, D. W. Statistical Approach to Obtaining Vacancy Formation Energies in High-Entropy Crystals from First Principles Calculations: Application to a High-Entropy Diboride. Phys. Rev. Mater. 2020, 4 (12). https://doi.org/10.1103/PhysRevMaterials.4.123602.
(1) Chen, J.; Loeb, S.; Kim, J.-H. LED Revolution: Fundamentals and Prospects for UV Disinfection Applications. Environ. Sci. Water Res. Technol. 2017, 3 (2), 188–202. https://doi.org/10.1039/C6EW00241B.
(1) S.P. O’Brien, J. Christudasjustus, L. Esteves, S. Vijayan, J.R. Jinschek, N. Birbilis, R.K. Gupta, A low-cost, low-density, and corrosion resistant AlFeMnSi compositionally complex alloy, npj Materials Degradation, 5 (2021), 1
(1) Xiong, R.; Kim, H. S.; Zhang, S.; Kim, S.; Korolovych, V. F.; Ma, R.; Yingling, Y. G.; Lu, C.; Tsukruk, V. V. Template-Guided Assembly of Silk Fibroin on Cellulose Nanofibers for Robust Nanostructures with Ultrafast Water Transport. ACS Nano 2017, 11 (12). https://doi.org/10.1021/acsnano.7b04235.
(2) Xiong, R.; Kim, H. S.; Zhang, L.; Korolovych, V. F.; Zhang, S.; Yingling, Y. G.; Tsukruk, V. V. Wrapping Nanocellulose Nets around Graphene Oxide Sheets. Angew. Chem. Int. Ed. Engl. 2018, 57 (28), 8508–8513. https://doi.org/10.1002/anie.201803076.
(3) Grant, A. M.; Kim, H. S.; Dupnock, T. L.; Hu, K.; Yingling, Y. G.; Tsukruk, V. V. Bionanocomposites: Silk Fibroin–Substrate Interactions at Heterogeneous Nanocomposite Interfaces (Adv. Funct. Mater. 35/2016). Adv. Funct. Mater. 2016, 26 (35). https://doi.org/10.1002/adfm.201670231.
(4) Stuart, M. A. C.; Huck, W. T. S.; Genzer, J.; Müller, M.; Ober, C.; Stamm, M.; Sukhorukov, G. B.; Szleifer, I.; Tsukruk, V. V.; Urban, M.; et al. Emerging Applications of Stimuli-Responsive Polymer Materials. Nat. Mater. 2010, 9 (2). https://doi.org/10.1038/nmat2614.
Questions? Email email@example.comProgram Details Application Requirements