In the realm of energy research, a team from Chalmers University of Technology in Sweden has made significant strides in the discovery and design of new materials crucial for green technologies. The researchers, Joakim Brorsson, Henrik Klein Moberg, Joel Hildingsson, Jonatan Gastaldi, Tobias Mattisson, and Anders Hellman, have leveraged machine learning to tackle the complex task of navigating vast compositional spaces in high entropy oxides. Their work, published in the journal npj Computational Materials, focuses on identifying high-performance oxygen carriers for chemical looping processes, a promising technology for clean energy production.
The team employed active learning-based strategies combined with first-principles calculations to efficiently explore and identify promising materials. Active learning is a machine learning approach that iteratively selects the most informative data points to learn from, significantly reducing the number of calculations needed. This method was validated using an established computational framework for identifying high entropy perovskites, which are materials with a specific crystal structure that can be used in chemical looping air separation and dry reforming. The researchers found that greedy or Thompson-based sampling, which are strategies for selecting the most promising candidates based on uncertainty estimates from Gaussian processes, were the most effective approaches.
Building on these insights, the team applied their method to a more complex problem: the discovery of high entropy oxygen carriers for chemical looping oxygen uncoupling. This process is crucial for various energy applications, including clean combustion and hydrogen production. The results yielded both qualitative and quantitative outcomes, including lists of specific materials with high oxygen transfer capacities and configurational entropies. Notably, the best candidates were based on the known oxygen carrier CaMnO3 but also included a variety of additional species, some expected (such as Ti, Co, and Cu) and others not (such as Y and Sm). This suggests that active learning approaches can uncover unexpected but highly effective materials.
The practical applications of this research are significant for the energy sector. High entropy oxides with optimized oxygen transfer capacities can enhance the efficiency and reduce the costs of chemical looping processes. These processes are integral to developing clean energy technologies, such as carbon capture and storage, hydrogen production, and low-emission power generation. By adopting active learning approaches, researchers can accelerate the discovery of new materials, making these methods critical for advancing green technologies. The team’s work highlights the transformative potential of machine learning in materials science and underscores its growing importance in energy research.
This article is based on research available at arXiv.