Insights into dendritic growth mechanisms in batteries: A combined machine learning and computational study

TL;DR

Dendritic growth in aqueous metal-ion batteries poses safety and performance risks that are not well captured by traditional models. The authors introduce a UNet-based CNN (CNN-2) augmented with seven physical parameters and validate it against literature data and VASP simulations, achieving training and validation correlations of approximately $R \approx 0.93$ and $R \approx 0.80$, respectively. Compared with a physics-agnostic CNN-1, CNN-2 shows markedly improved accuracy and robustness, accompanied by sensitivity analyses and visualizations that reflect real dendrite dynamics. The approach offers practical routes to optimize electrolyte formulations, guide electrode material design, and enable virtual testing to accelerate battery development. The integration of data-driven learning with physics-informed insights represents a significant advance for predicting and mitigating dendritic growth in diverse battery systems.

Abstract

In recent years, researchers have increasingly sought batteries as an efficient and cost-effective solution for energy storage and supply, owing to their high energy density, low cost, and environmental resilience. However, the issue of dendrite growth has emerged as a significant obstacle in battery development. Excessive dendrite growth during charging and discharging processes can lead to battery short-circuiting, degradation of electrochemical performance, reduced cycle life, and abnormal exothermic events. Consequently, understanding the dendrite growth process has become a key challenge for researchers. In this study, we investigated dendrite growth mechanisms in batteries using a combined machine learning approach, specifically a two-dimensional artificial convolutional neural network (CNN) model, along with computational methods. We developed two distinct computer models to predict dendrite growth in batteries. The CNN-1 model employs standard convolutional neural network techniques for dendritic growth prediction, while CNN-2 integrates additional physical parameters to enhance model robustness. Our results demonstrate that CNN-2 significantly enhances prediction accuracy, offering deeper insights into the impact of physical factors on dendritic growth. This improved model effectively captures the dynamic nature of dendrite formation, exhibiting high accuracy and sensitivity. These findings contribute to the advancement of safer and more reliable energy storage systems.

Figures (7)

• Figure 1: Structure of a typical aqueous metal-ion battery.
• Figure 2: Dendritic deposition/growth on the anode surface, utilizing aqueous metal-ion batteries as a case study.
• Figure 3: 2D convolutional neural network constructed for predicting dendritic growth processes based on simulated data collected.
• Figure 4: Performance evaluation of the dendritic growth modes utilizing our trained artificial convolutional neural network. (A) and (B) illustrate the dendritic growth rates of network 1 (CNN mode 1) (omitting various physical variables), alongside their respective prediction accuracies in both training (A) and testing (B) datasets. (C) Specific performance metrics for the predicted data points. (D) and (E) The dendritic growth rates of network 2 (CNN model 2) (considering specific physical variables as input), along with their corresponding prediction accuracies in the training (D) and testing (E) datasets, respectively. (F) Additional demonstration of the performance on the predicted data points. The CNN model-2 architecture integrates physical quantities. This neural network model incorporates seven selected physical parameters as inputs, facilitating a more thorough analysis of their influence on dendritic growth rates. The incorporation of these parameters is intended to bolster the predictive capability of the CNN model by integrating pertinent physical insights.
• Figure 5: Sensitivity assessment of the two modes is highlighted. (A) The overall assessment. (B) The sensitivity assessment across trickle charging (stage 1), constant current charging (stage 2), and constant voltage charging (stage 3). The left panels in (A) and (B) represent the assessments using VASP mode without data-driven computation, whereas the right panels correspond to CNN mode 2 with data-driven computation. The evaluation incorporates seven distinct physical parameters (t$_{\textrm{i}-1}$: VASP mode; t$_\textrm{i}$: CNN mode 2). The specific parameters are as follows, 1: temperature, 2: concentration gradient, 3: electric field strength, 4: ion concentration, 5: surface energy of dendrites, 6: fluidity of the solution, and 7: number of lattice defects.