Machine-Learning-Driven Advanced Characterization of Battery Electrodes

Abstract

Materials characterization is fundamental to our understanding of lithium ion battery electrodes and their performance limitations. Advances in laboratory-based characterization techniques have yielded powerful insights into the structure–function relationship of electrodes, yet there is still far to go. Further improvements rely, in part, on gaining a deeper understanding of complex physical heterogeneities in the materials. However, practical limitations in characterization techniques inhibit our ability to combine data directly. For example, some characterization techniques are destructive, thus preventing additional analyses on the same region. Fortunately, artificial intelligence (AI) has shown great potential for achieving representative, 3D, multi-modal datasets by leveraging data collected from a range of techniques. In this Perspective, we give an overview of recent advances in lab-based characterization techniques for Li-ion electrodes. We then discuss how AI methods can combine and enhance these techniques, leading to substantial acceleration in our understanding of electrodes.

MatGPT: A Vane of Materials Informatics from Past, Present, to Future

“…[155] In addition to electrolyte engineering, AI has also played a key role in electrode material design. [144,[156][157][158] For example, Choy et al applied six ML regression models to study the correlations of the structural, elemental feature of 168 distinct doped nickelcobalt-manganese (NCM) systems. They found that GBDT was the best prediction power for both initial discharge capacity and 50th cycle discharge capacity (EC), with the root-mean-square errors calculated to be 16.66 and 18.59 mAh g −1 , respectively.…”

“…In addition to electrolyte engineering, AI has also played a key role in electrode material design. [ 144,156–158 ] For example, Choy et al. applied six ML regression models to study the correlations of the structural, elemental feature of 168 distinct doped nickel–cobalt–manganese (NCM) systems.…”

Crystallography of Active Particles Defining Battery Electrochemistry

“…These tools also allow for the effective integration of different advanced characterization techniques to allow more advanced and succinct analyses of electrode materials to be made. [ 208 ] In imaging, AI based techniques have been shown as crucial tools to systematically recognize pattern formation and structural intricacies within large data sets using TEM, which traditional data analytic techniques struggle to interpret. [ 209 ] Artificial‐Intelligence‐based Structure Evaluation (ARISE) to predict crystal structures based on Bayesian deep learning has also recently been employed to predict single and polycrystalline crystal and amorphous structures from experimental and synthetic information.…”

Design of a High-Performance Titanium Nitride Metastructure-Based Solar Absorber Using Quantum Computing-Assisted Optimization

“…In the recent decade, optimization approaches leveraging machine learning have emerged as efficient tools for the inverse design of complex materials such as metastructures. − The optimization could successfully design various photonic structures to achieve high performance. , However, designing discretized structures, such as those seen in metastructures, can be difficult for conventional gradient-based schemes to find the global optimum due to the lack of gradient. In discrete optimization spaces, the only approach that guarantees the finding of the global optimum is an exhaustive search, but when the degree of freedom is large, it becomes insurmountable for classical computers to exhaustively enumerate all possible candidates.…”