(4dk) Leveraging Statistical Inference and Physical Modeling to Augment Electrochemical Analysis of Charge Storage Materials

I am a Ph.D. candidate working with Prof. Fikile Brushett in the Department of Chemical Engineering at MIT, and I plan to defend my dissertation in the spring of 2022. The energy infrastructure must undergo drastic changes to create a sustainable world, and I am motivated to contribute towards addressing this grand challenge through my research efforts. My doctoral research seeks to aid in the development of promising electrochemical devices by examining how voltammetry, combined with statistical inference and physical modeling, can be used to advance in situ techniques to study the behavior of soluble charge storage materials. In my postdoctoral research, I would like to expand upon the use of physics-based data analysis—for example, by developing or implementing machine learning methods with physical constraints—to elucidate the behavior of other systems (e.g., complete batteries). I would also like to learn more about state-of-the-art machine learning methods and tailor them to the field of electrochemistry to further bolster the inferential protocols I develop. Following my postdoctoral appointment, I intend to pursue a faculty position, where I will leverage my knowledge of electrochemistry and machine learning methods, along with my passion for teaching and mentoring, to support the growth of students, to build a caring community, and to aid in addressing the energy crisis.

Research Interests

The prospect of a renewably powered grid continues to grow as the cost of sustainable energy technologies decreases. However, intermittent resources (e.g., wind, solar photovoltaics) inhibits complete adoption of a clean grid infrastructure, necessitating energy storage systems that are able to operate across multiple time scales, like lithium-ion and redox flow batteries. The performance and durability of these devices are dictated by the complex interplay of many constituent components (e.g., electrodes, membranes/separators), and probing system behavior typically requires multiple ex situ techniques (e.g., nuclear magnetic resonance (NMR) and scanning electron microscopy), which, while informative, may be time-consuming, expensive, and may even provide incomplete information about the component(s) due to preparation procedures, such as dilution and deuteration. In contrast, further implementation of in situ or operando electrochemical methods may provide an inexpensive route to characterize device behavior; this, in turn, can also afford fewer and more targeted ex situ experiments to accelerate the development process. However, it can be challenging to extract information from in situ electrochemical methods because—unlike spectroscopic techniques—the signals from these experiments do not directly correspond with elemental, molecular, or structural features, encouraging more advanced data analysis to infer relevant information.

To this end, I have pursued the following overarching scientific question throughout my Ph.D. studies: How can physical modeling and machine learning methods be leveraged to uncover more information from in situ or operando electrochemical data? While I intend to develop a framework that can address this question for broad applications, I have restricted my methodological development thus far to the voltammetry of soluble redox couples in liquid phase electrolytes. I have explored this application through two projects, where I demonstrated that:

1) Physical models and Bayesian inference can be used to label redox couples from experimental voltammograms of a liquid-phase electrolyte [1]. My protocol references a library of physical descriptors catalogued for each candidate compound using experimental data and was validated using functionalized phenothiazines—redox couples explored for use in lithium-ion and redox flow batteries.

2) Steady-state models can be used to estimate relevant features from microelectrode voltammograms. In addition to automatically processing electrochemical data, this protocol demonstrates that utilizing the less-often used oblate spheroidal coordinate system greatly decreases the mathematical complexity of the problem and potentially increases the modeling capabilities of steady-state microelectrode voltammetry [2].

These two studies have revealed that the combination of physical data and statistical inference can unlock deeper information about soluble redox couples from voltammetric data, which is challenging to discern using more conventional approaches (e.g., Randles-Ševčik analysis). More generally, I have come to recognize the utility of inferential and machine learning protocols, and I am striving to grow in my understanding and fluency with these methods beyond my Ph.D. studies to contribute to the development of sustainable electrochemical technologies. I believe this drive, combined with my experience in analytical electrochemistry, will enable me to continue developing electrochemical methods that can reveal useful information. For example, I would be interested in developing and applying model reduction order methods to investigate concentrated electrolyte behavior in electrochemical systems for energy storage and conversion (e.g., flow batteries, carbon dioxide utilization, multivalent batteries). As another example, I would enjoy applying physics-based machine learning methods to analyze cell data for studying operational and failure modes in electrochemical devices.

Experience

I have both experimental and modeling experience, and I look forward to expanding my skillset during my postdoctoral studies. My primary expertise thus far is comprised of electrochemical experimentation and simulation. Over the past 4.5 years, I have employed an array of experimental techniques, including cyclic voltammetry, cyclic square wave voltammetry, microelectrode voltammetry, chronoamperometry, and electrolysis. I also have substantial experience conducting electrochemical experiments in a glovebox environment for the same period.

To complement my experiments, I have also developed in-house analytical and numerical frameworks over the past three years using MATLAB^®. With these models, I have simulated macro- and microelectrode voltammograms, homogeneous chemical reactions, and electrochemical cell behavior from mass conservation equations; I have additionally conducted numerical optimization using MATLAB^®’s built-in functions (specifically, fmincon) to estimate important parameters (e.g., electrochemical and mass transport descriptors).

In addition to my electrochemical expertise, I have sizeable experience performing NMR experiments (primarily ¹H and ¹³C, along with some experience with ¹H-¹H COSY) to analyze liquid-phase electrolytes post mortem over the past 2.5 years. I have also performed UV-Vis spectroscopy, Karl-Fischer titration, and X-ray diffraction for a variety of purposes throughout my doctoral research.

Selected works:

[1] A. M. Fenton Jr., F. R. Brushett, Using voltammetry augmented with physics-based modeling and Bayesian hypothesis testing to estimate electrolyte composition, submitted, https://doi.org/10.26434/chemrxiv.14479707.v1

[2] A. M. Fenton, Jr., B. J. Neyhouse, K. M. Tenny, Y. M. Chiang, F. R. Brushett, Analytical and Numerical Modeling of Microelectrode Voltammetry in Oblate Spheroidal Coordinates, 239th Electrochemical Society Meeting, Chicago, IL (virtual). May-June 2021 (Manuscript in prep)

[3] J. A. Kowalski, A. M. Fenton Jr., B. J. Neyhouse, F. R. Brushett, A Method for Evaluating Soluble Redox Couple Stability Using Microelectrode Voltammetry, Journal of The Electrochemical Society, 167(16), 160513, 2020

[4] B. J. Neyhouse, A. M. Fenton Jr., F. R. Brushett, Too Much of a Good Thing? Assessing Performance Tradeoffs of Two-Electron Compounds for Redox Flow Batteries, Journal of The Electrochemical Society, 168(5), 050501, 2021