Sodium-containing transition-metal layered oxides are emerging as promising electrode materials for sodium-ion batteries, which could be a viable substitute for lithium-ion batteries. However, due to the large variety of possible elemental compositions, finding the best ones presents significant challenges. A recent study utilized extensive experimental data and machine learning techniques to identify the optimal compositions for sodium-ion batteries, potentially streamlining the research process and facilitating the transition to renewable energy.
Energy storage plays a crucial role in many rapidly advancing sustainable technologies, such as electric vehicles and renewable energy systems. Currently, lithium-ion batteries (LIBs) dominate the market, but lithium is limited in availability and can be costly, resulting in economic and supply chain issues. Consequently, researchers globally are investigating new battery technologies that utilize more abundant materials.
Sodium-ion (Na-ion) batteries, which utilize sodium ions for energy transport, represent a promising alternative to LIBs due to sodium’s abundance, enhanced safety features, and potentially lower costs. Sodium-containing transition-metal layered oxides (NaMeO2) are particularly effective materials for the positive electrode in Na-ion batteries, exhibiting impressive energy density and capacity. However, with multiple transition metals used in these layered oxides, the vast number of possible combinations complicates the process of discovering the best composition. Even slight modifications in the type and ratio of transition metals can lead to significant variations in crystal structure and overall battery performance.
In a recent study led by Professor Shinichi Komaba, alongside Ms. Saaya Sekine and Dr. Tomooki Hosaka from Tokyo University of Science (TUS), Japan, as well as collaborators from Chalmers University of Technology and Professor Masanobu Nakayama from the Nagoya Institute of Technology, machine learning was utilized to enhance the search for optimal compositions. The findings of this research, shared on September 05, 2024, were published online in the Journal of Materials Chemistry A on November 06, 2024, following final edits. This study received support from funding agencies including JST-CREST, DX-GEM, and JST-GteX.
The research team aimed to automate the evaluation process of various elemental compositions in O3-type NaMeO2 materials. They began by compiling a comprehensive database of 100 samples from O3-type sodium half-cells, featuring 68 distinct compositions collected over a span of 11 years by Komaba’s team. “The database detailed the compositions of NaMeO2 samples, where Me represents transition metals such as Mn, Ti, Zn, Ni, Fe, and Sn. It also included the voltage limits for charge-discharge tests, initial discharge capacities, average discharge voltages, and capacity retention after 20 cycles,” Komaba explains.
The team then utilized this extensive database to train a model that employed several machine learning algorithms along with Bayesian optimization for conducting an efficient exploration. The model was designed to understand how factors like operating voltage, lifespan capacity retention, and energy density are interconnected with NaMeO2 layered oxide composition, enabling it to predict the optimal element ratios needed for the best balance of these characteristics.
After examining the model’s predictions, the researchers identified Na[Mn0.36Ni0.44Ti0.15Fe0.05]O2 as the composition that would deliver maximum energy density, a crucial quality in electrode materials. To validate these predictions, they synthesized samples of this composition and created standard coin cells for charge-discharge testing.
The experimental measurements largely corroborated the model’s predictions, demonstrating both its accuracy and the model’s potential for discovering new battery materials. “The methodology established in our study presents an effective approach for isolating promising compositions from a vast array of candidates,” notes Komaba. “Additionally, this method can be adapted for more complicated material systems, such as quinary transition metal oxides.”
Using machine learning to identify promising directions for research is becoming more prevalent in materials science, as it significantly reduces the number of experiments and the time needed for new material evaluations. The strategy introduced in this study may expedite the advancement of next-generation batteries, which could transform energy storage technologies widely. This includes applications in renewable energy systems and electric or hybrid vehicles, as well as consumer electronics like laptops and smartphones. Furthermore, successful machine learning applications in battery research could provide a framework for material development in other sectors, potentially accelerating progress across the materials science field.
“By utilizing machine learning, we can minimize the number of experiments required, bringing us closer to faster and more cost-effective materials development. With ongoing improvements in the performance of electrode materials for Na-ion batteries, we anticipate the emergence of high-capacity, long-life batteries at lower costs in the future,” concludes Komaba.
