The magnetocaloric effect (MCE), often characterized by the magnetic entropy change (SM) for a givenapplied field change (H), is one of the most promising alternative paths for the development ofgreenhouse-gas free refrigeration devices (Franco et al., 2018). The magnitude of SM tends to peak at amaterial`s magnetic ordering temperature (Tmag), such as Curie (TC) or Néel (TN) temperature, and itsmaximum value for a H strongly depends on the material. Since the discovery of gigantic MCE inmaterials such as Gd5Si2Ge2 (Pecharsky and Gschneidner, 1997) and La(Fe,Si)13 (Shen et al., 2009) anexplosive increase in the search for materials which could exhibit such effect lead to the accumulation ofmagnetocaloric properties of a vast number of magnetic materials. However, it remains a challenge todesign materials that can exhibit such a remarkable effect. To tackle this challenge, we constructed amachine learning model for predicting SMMAX purely based on chemical composition descriptors. Forthis, we gathered the accumulated data of magnetocaloric materials, and trained a model for theprediction of SMMAX for a given material composition and an applied H. Then, by exploring a textmineddatabase called MagneticMaterials (Court and Cole, 2018) the obtained model was used inconjunction with domain expertise to filter possible candidates for experimental verification. Throughthis approach, we found HoB2 (TC = 15 K), exhibits that highest volumetric SM of 0.35 J cm-3 K-1 for H= 5 T (Castro et al., 2020), to the best of our knowledge, of all known characterized second-ordermagnetic phase transition materials in the temperature range between 4.2 and 77 K, as shown in thefigure below.In this talk, we will discuss the process of model building, the choice of compositional based features,the current challenges and will briefly introduce and compare the experimental results with othermaterials in the same temperature range.
展开▼
