Experimentally validated screening strategy for alloys as anode in Mg-air battery with multi-target machine learning predictions

The anode in Mg-air cell plays the important role in determining the overall discharge performance. To refine the Mg-air cell performance, numerous Mg alloys have been developed in past decades by extensive trial-and-error efforts. However, it is a tough task to determine the suitable elements and their compositions to fabricate the Mg alloys with the expected feature. Machine learning (ML), as a hot artificial intelligence technology, has shown significant potential in screening and predicting the new materials. Herein, a machine learning (ML) model is trained via Extreme Gradient Boosting Regressor (XGB) algorithm to assist the exploration of binary Mg alloys that is suitable to be efficient anode in Mg-air cell. Importantly, four metrics related with magnesium-air batteries: cell voltage, utilization efficiency, specific capacity, and specific energy, are all considered to screen out the promising Mg anodes instead of depending on only one feature. After that, five potential Mg alloys, Mg-Ca, Mg-Zn, Mg-Sn, Mg-Al, and Mg-Li, are fabricated according to the machine learning results and their discharge performance is determined. The different discharge performance of above five anodes are elucidated by the emission scanning electron microscope (SEM), potentiodynamic polarization curves (PDP), and real-time hydrogen evolution measurement. The Mg-Ca anode exhibits the highest specific energy of 1797.1 Wh kg- 1 at 10 mA cm-2, while the Mg-Li anode has the worst discharge performance. It is consistent with the predicted results by machine learning. Overall, this work breaks through the limitations of traditional material design, offering new insights for the development of magnesium-air batteries, which could be extended to other materials.