Perspective on Mechanistic Modeling of Li-Ion Batteries

ConspectusThough relatively young, the mechanistic modeling approach has gained tremendous traction in the past decade as it was proven to be extremely versatile and effective for lithium-ion battery diagnosis and prognosis. The approach is relies on assembling digital twins by matching the individual voltage response of each electrode. Changing the matching, via scaling or translations, enables replication of the degradation modes electrochemical signature. Degradation modes comprise the loss of lithium inventory, the loss of active material, and kinetic changes and refer to the impact of degradation mechanisms on the electrodes rather than their root cause. Every degradation mechanism will affect, to some extent, the amount of material able to react, the amount of lithium able to go back and forth between the electrodes, and the overall reaction kinetics. Quantifying degradation modes open the gate for material-based diagnosis and prognosis without the need for complex models.This Account is first a reflection on a decade worth of discussion and validation of several key concepts about using digital twins for advanced diagnosis and prognosis since the seminal publications in the early 2010s. Since proposing our version of the framework, it has been used to diagnose the degradation of several hundred cells of multiple chemistries and blends. It made it possible to explain and predict the apparition of knees with the concept of hidden mechanisms and to emulate the impact of kinetics. The approach also proved useful to investigate overdischarge and overcharge and to generate big data with millions of synthetic voltage curves enabling the development of advanced diagnosis and prognosis tools. Herein, we will focus on the emulation of kinetic changes, of lithium plating both from rate-dependent and rate-independent origins, and on the utilization of synthetic datasets.This Account will also introduce the next decade with proof-of-concept implementations and simulations that will open new directions to enhance the modeling framework application to more complex case figures that could facilitate its use for deployed systems. This includes more varied synthetic data sets, blended and inhomogeneous electrodes and packs, voltage fade, and calculations outside of constant current. Blends and voltage fade will be simulated with a new paralleling model at the electrode level, inhomogeneities will be simulated with a new paralleling model at the cell level, and nonconstant duty cycles will be calculated by aggregating simulations at different rates. These new features could allow a much wider field implementation of better diagnosis and prognosis tools for deployed systems.