Research

Energy storage devices consist of multiple components, including the cathode, anode, and electrolyte, each of which is fabricated through complex processes involving various materials. These materials must meet specific property requirements to ensure optimal performance.

Atomic simulation methods, such as density functional theory (DFT), classical molecular dynamics (CMD), and machine-learning-based potential (MLP) molecular dynamics, serve as powerful tools for predicting key material properties with high accuracy and low cost. Since the influence of atomic structure on material properties is often difficult to measure experimentally, atomic simulations provide valuable complementary insights.

Our research focuses on designing novel materials to enhance the performance of energy storage devices, such as secondary batteries and fuel cells, through advanced atomic simulation techniques.

Recent advancements in materials research are increasingly driven by artificial intelligence (AI), which relies on large volumes of high-quality materials data. However, obtaining such data is challenging, as most research findings are scattered across individual institutions and are not readily accessible in structured formats.

Despite this, scientific literature contains a wealth of valuable information on battery materials and their performance. These datasets are carefully curated, peer-reviewed, and thus of high quality. However, since they are stored in unstructured formats as human language, manually extracting and compiling them is a time-consuming and labor-intensive task.

Our research focuses on developing an automated data retrieval system using natural language processing (NLP), a subfield of AI specialized in decoding human language. By leveraging NLP, we aim to construct a comprehensive, high-quality, and large-scale battery materials database, facilitating data-driven advancements in energy storage research.

One of the most impactful applications of artificial intelligence (AI) in the battery field is the prediction of battery performance. In theory, if all relevant information about materials synthesis, cell assembly, and testing conditions is known, the performance of a battery cell can be precisely determined. By leveraging this information, AI models can predict key performance metrics, such as energy density and cycle life, under specific conditions.

However, developing accurate AI models for battery performance prediction is challenging due to the limited availability of public data and inconsistencies in experimental conditions, which are often poorly documented. Additionally, the battery manufacturing process is highly complex, with performance being sensitive to even minor variations in production details.

Our research aims to develop advanced AI models capable of predicting missing data within existing datasets, enabling the construction of AI-ready databases. Ultimately, this will allow us to accurately predict the performance of any given battery cell, facilitating accelerated development and optimization of energy storage devices