Technological breakthroughs

Artificial intelligence (AI) is emerging as a powerful tool for uncovering the complex electrochemical mechanisms that govern battery performance. According to a new review by researchers at Tsinghua University, AI is enabling scientists to analyze vast, multidimensional datasets, accelerating the development of safer, longer-lasting, and more energy-dense batteries.

Batteries are at the heart of modern technologies, from electric vehicles to grid-scale energy storage. Yet improving their energy density, lifespan, and safety ultimately depends on understanding the electrochemical processes occurring inside battery materials.

In a review published in National Science Review, a team led by Zhiyuan Han and Jiaqi Zhou from Tsinghua University argues that AI is becoming a game-changing tool for addressing this challenge.

Three Major Challenges in Battery Research

According to the researchers, battery scientists face three fundamental challenges.

The first is tracking the dynamic evolution of electrode materials and interfacial structures throughout charging and discharging cycles.

The second involves analyzing complex, multi-dimensional and multi-scale datasets that encompass atomic structures, chemical compositions, and electrical, thermal, and mechanical properties—from the atomic level to the performance of an entire battery cell.

The third challenge is disentangling multiple electrochemical processes that occur simultaneously, including thermodynamics, reaction kinetics, and mechanical stress, all of which influence battery performance in interconnected ways.

Modern characterization techniques such as transmission electron microscopy (TEM), Raman spectroscopy, and X-ray fluorescence generate enormous volumes of high-resolution data. However, these datasets are often noisy and difficult to analyze manually.

This is where AI offers a significant advantage. Machine learning algorithms can process millions of data points simultaneously, identify nonlinear relationships, and detect hidden patterns that conventional statistical approaches may overlook.

AI Reveals What the Human Eye Cannot See

One of AI's most successful applications is the automated analysis of data collected through advanced in situ characterization techniques.

Machine learning models have been used to remove noise from spectroscopic data tracking lithium-ion diffusion, enabling researchers to identify anisotropic lithium transport within electrode materials.

In another study, convolutional neural networks (CNNs) achieved an accuracy of 98.8% in automatically identifying lithium structures, including dendrites—needle-like formations that can trigger short circuits and compromise battery safety.

AI has also been applied to TEM image analysis, allowing researchers to identify crystal defects and reconstruct the three-dimensional structures of thousands of electrode particles. These analyses revealed that smaller particles or those subjected to fast charging are more likely to lose electrical contact with the conductive network, contributing to battery degradation over time.

Discovering New Electrochemical Insights

Beyond image analysis, AI is helping scientists uncover the factors that determine battery performance.

In electrolyte research, an AI model trained on data from 150 lithium–copper half-cells identified the oxygen content of solvents as the dominant factor influencing Coulombic efficiency. This insight guided the design of fluorine-free electrolytes capable of achieving efficiencies close to 99.7%.

For electrode materials, transfer learning models revealed that covalent atomic radius is the most influential descriptor governing electrode potential, while internal resistance plays a critical role in determining areal energy density.

AI has also shown considerable promise in catalyst discovery. By analyzing more than 2,900 published studies on hydrodesulfurization reactions, researchers predicted the catalytic activity of over 800 candidates and identified dozens of promising catalysts from a pool of approximately 375,000 potential materials.

In carbon dioxide reduction research, AI has further enabled scientists to determine the adsorption sites of carbon monoxide on catalyst surfaces—an inverse structural problem that has long challenged conventional computational methods.

Understanding Why Batteries Degrade

Battery aging remains one of the most complex issues in electrochemistry.

According to the review, performance degradation rarely results from a single mechanism. Instead, it arises from the interaction of multiple processes, including chemical reactions, mechanical stress, and phase transformations.

Deep learning models combined with expert knowledge have successfully distinguished different degradation mechanisms in solid-state cathodes, separating surface instability from stress-induced phase changes.

Meanwhile, AI-powered generative algorithms integrated with high-throughput robotic laboratories are dramatically accelerating the discovery of new battery materials while reducing the number of experiments required.

AI Cannot Replace Scientific Reasoning

Despite its impressive achievements, the authors caution that AI still has important limitations.

Most current models prioritize predictive accuracy without demonstrating whether the relationships they identify truly reflect the underlying electrochemical mechanisms. High prediction accuracy does not necessarily imply causal understanding; many AI-discovered patterns may simply represent statistical correlations within limited datasets.

The researchers argue that AI models should be integrated with physical principles, electrochemical knowledge, and causal reasoning to become genuine scientific discovery tools rather than sophisticated prediction engines.

They also note that large language models (LLMs) and autonomous AI agents currently serve primarily as tools for literature mining and data extraction. Their ability to explain fundamental electrochemical mechanisms remains limited because battery research data differ substantially from the text-based information on which LLMs are trained.

Looking ahead, the authors believe the future of battery research will depend on close collaboration between human scientists and AI. While AI excels at processing data and identifying patterns, researchers remain essential for interpreting results, formulating hypotheses, and validating new scientific discoveries. Only through this partnership, they argue, can AI evolve from a powerful analytical tool into a true collaborator in scientific exploration.

Source: Han, Z. et al., "Uncovering battery electrochemical mechanisms by artificial intelligence," National Science Review, 2025. https://doi.org/10.1093/nsr/nwaf442