Active Learning Accelerates the Discovery of Multicomponent Oxide Oxygen Carriers for Chemical Looping
Prof. Anders Hellman
Chalmers University of Technology, Sweden
Chemical looping is a promising technology for carbon utilization, but to implement it on a large scale, we need improved oxygen carrier materials. We present an active-learning framework that helps the discovery of multicomponent oxide oxygen carriers for various chemical looping reactions. The approach integrates atomistic calculations with a Wasserstein Autoencoder to iteratively explore perovskite-based compositions using explorative, greedy, and Thompson sampling strategies. Validation against previous studies shows that the active-learning cycle efficiently finds promising candidates while using much less data than traditional high-throughput methods. Using the Thompson strategy helps identify a variety of promising oxygen carriers, thereby expanding the number of viable materials for chemical looping. This work shows an effective way to speed up the discovery of materials for sustainable energy applications.
