Our research will be organized into five major thrusts according to differing length and time scales, with the following overall plans and collaborations:

Thrust 1: Primary particles & crystal lattice
The goal of this thrust is to generate large data sets for single primary particles from phase-field simulations and time and nano-scale resolved measurements. Subsequently, machine learning will be applied to these datasets to extract fundamental properties of electrode materials (composition-dependent free energy, reaction kinetics, and diffusion coefficients, etc.). At the same time, we will also solve the local atomic structures by seeking machine-learned models that relate nonstructural experimental probes to structural features of the crystal lattice. Such models aim to reduce immense configuration search space, enabling strategic DFT calculations to map nonstructural experimental data to the detailed atomistic structure. The expected outcomes are new and/or improved atomistic and phase-field models for the electrode materials and fundamental understanding of transport, strain, and phase transformations.


Thrust 2: Interfaces and interphases
Large data sets on interfacial intercalation kinetics and interphase growths (e.g., of SEI and lithium plating in graphite) will be obtained experimentally and via phase-field and continuum models spanning a large number of parameters such as electrode and electrolyte composition and electrochemical cycling conditions. Chemo-mechanics of the interphases, especially how it couples with the primary particle, will also be considered. Reaction kinetics of intercalation and SEI growth will also be studied as a function of interfacial voltage, surface coating (carbon, ceramic) and electrolyte, and the empirical Butler-Volmer equation will be tested against generalized phase-field kinetics and Marcus theory of electron transfer. Machine learning on these experiments and on battery cycling data conducted across a large parameter space will reveal how these mechanisms relate to capacity fade and lifetime statistics. The expected engineering outcomes are electrodes with improved round-trip efficiency, lower capacity fading and lithium plating risk in graphite, primarily through rationally engineered electrochemical cycling protocols, microstructure, and materials compositions.


Thrust 3: Chemo-mechanics of secondary particles
This thrust will focus on generating and analyzing large 3D data sets on secondary particles generated via X-ray tomography experiments and via 3D phase-field models. Chemo-mechanical coupling between strain, voltage, kinetics, and fracture will be probed at the microstructural level as well as at the atomistic level (Reed). Machine learning at the micrometer scale will be used correlate microstructural/topological features in the secondary particles to performance degradation, primarily using a large data set consisting of computer generated microstructures with varying size, shape, porosity and tortuosity. Similarly, machine learning on literature and experimental chemical expansion data will be used to predict compositions with desired chemo-mechanical properties at the atomistic level. The expected engineering outcomes of this thrust are electrode compositions and microstructures that minimize mechanical failure.


Thrust 4: Data-driven porous electrode models
This thrust will develop a general simulation framework and software for Multiphase Porous Electrode Theory (MPET) that combines our new data-driven models of active materials, interfaces, secondary particles, and degradation with traditional PET developed by Newman and collaborators. The key innovation in MPET is the modeling of phase transformations, which will be used to predict “mosaic instabilities” of discrete, stochastic particle transformations in composite porous electrodes and related heterogeneous reaction kinetics of intercalation and SEI growth. The code will be tuned/validated against “bottom up” 3D phase field simulations and “top down” battery cycling experiments.


Thrust 5: Optimization of battery management systems
This thrust represents the “top down” approach of generating and analyzing massive data sets from battery cycling, and use them to optimize battery management systems. We will employ both model-less data mining and optimal design of experiments and model-based optimization and control. Simplified versions of MPET and circuit models will be  eveloped that capture both top-down experimental trends and bottom-up theoretical predictions. The expected engineering outcomes of this thrust are optimized protocols and processes for battery charging, health estimation, and formation.