Understanding the Correlation between Lithium Dendrite Growth and Local Material Properties by Machine Learning

Abstract

Lithium metal batteries are attractive for next-generation energy storage because of their high energy density. A major obstacle to their commercialization is the uncontrollable growth of lithium dendrites, which arises from complicated but poorly understood interactions at the electrolyte/electrode interface. In this work, we use a machine learning-based artificial neural network (ANN) model to explore how the lithium growth rate is affected by local material properties, such as surface curvature, ion concentration in the electrolyte, and the lithium growth rates at previous moments. The ion concentration in the electrolyte was acquired by Stimulated Raman Scattering Microscopy, which is often missing in past experimental data-based modeling. The ANN network reached a high correlation coefficient of 0.8 between predicted and experimental values. Further sensitivity analysis based on the ANN model demonstrated that the salt concentration and concentration gradient, as well as the prior lithium growth rate, have the highest impacts on the lithium dendrite growth rate at the next moment. This work shows the potential capability of the ANN model to forecast lithium growth rate, and unveil the inner dependency of the lithium dendrite growth rate on various factors.

Reshaping the material research paradigm of electrochemical energy storage and conversion by machine learning

“…For more complex EESC systems, such as fuel cells, lithium metal batteries (LMBs), lithium-sulfur (Li-S) batteries, and sodium-ion batteries (SIBs), ML affords the similar key effects. [131][132][133][134][135] Proton exchange membrane fuel cell (PEMFC) is susceptible to the impurities in H 2 and operating conditions, which generally results in deteriorated performance over time. Thus, the degradation prediction is crucial in evaluating the reliability of the PEMFC system.…”

Graphene Quantum Dots with Improved Fluorescence Activity via Machine Learning: Implications for Fluorescence Monitoring

“…At the same time, by the greedy random walk method and the corresponding evaluation standard performed by optimizing 100 times, we confirmed that B,N-GQDs synthesized in the condition of "184-10-2. 23 Synthesis of B,N-Codoped Graphene Quantum Dots (B,N-GQDs). APBA (0.1 g) was dissolved in acetone (60 mL), followed by ultrasonic stirring for 30 min.…”

“…(2) Macroscopicity and objectivity. The fitting results of machine learning are based on the obtained data, but their analysis is not completely limited by a specific result; thus, this method can be used to objectively reveal the process parameters and their effects in the material synthesis process. , (3) Machine learning technology can be utilized to optimize the synthesis process by selecting the reasonable evaluation standards. (4) Machine learning can be used to reflect the relevance in the synthesis process in a complex system, beyond the traditional material/process analysis methods.…”