Data science and deep learning for the development of new hydrogen storage materials
Апстракт
Prediction of metal hydride formation enthalpy is one of the key requirements for a rapid design of new hydrogen storage materials. In the last decades, DFT (density functional theory) approach showed good predictive potential for the ground state properties and accurate energies of hydride formation. However, calculating ZPE contribution and temperature effects in addition to formation energy at 0K is computationally and time- consuming and therefore often avoided, resulting in discrepancy to experiment. The development of machine learning and, in particular, deep learning, opens a new perspective for predictive modeling of materials properties. Data collected through DFT calculations can be combined with experimental results in a predictive model, aiming to exploit unexplored compositional space. In this work, we consider the application of MatErials Graph Network (MEGNet) [1] to the prediction of hydrogen formation behavior, and screening of potential dopants in reversible metal hyd...ride materials. Various approaches, relying on transfer learning and both experimental data and computational repositories (MP [2], NOMAD [3]) are proposed as a route to accurate prediction of a structure-property relation for hydrogen storage materials. Domains of applicability of these models are addressed.
Извор:
4th International Meeting on Materials Science for Energy Related Applications : Book of abstracts, 2021, 18-18Издавач:
- Belgrade : Faculty of Physical Chemistry, University of Belgrade
Напомена:
- 4th International Meeting MATERIALS SCIENCE FOR ENERGY RELATED APPLICATIONS; September 22-23, 2021; Belgrade, Serbia
Колекције
Институција/група
VinčaTY - CONF AU - Batalović, Katarina AU - Radaković, Jana AU - Paskaš Mamula, Bojana PY - 2021 UR - https://vinar.vin.bg.ac.rs/handle/123456789/11250 AB - Prediction of metal hydride formation enthalpy is one of the key requirements for a rapid design of new hydrogen storage materials. In the last decades, DFT (density functional theory) approach showed good predictive potential for the ground state properties and accurate energies of hydride formation. However, calculating ZPE contribution and temperature effects in addition to formation energy at 0K is computationally and time- consuming and therefore often avoided, resulting in discrepancy to experiment. The development of machine learning and, in particular, deep learning, opens a new perspective for predictive modeling of materials properties. Data collected through DFT calculations can be combined with experimental results in a predictive model, aiming to exploit unexplored compositional space. In this work, we consider the application of MatErials Graph Network (MEGNet) [1] to the prediction of hydrogen formation behavior, and screening of potential dopants in reversible metal hydride materials. Various approaches, relying on transfer learning and both experimental data and computational repositories (MP [2], NOMAD [3]) are proposed as a route to accurate prediction of a structure-property relation for hydrogen storage materials. Domains of applicability of these models are addressed. PB - Belgrade : Faculty of Physical Chemistry, University of Belgrade C3 - 4th International Meeting on Materials Science for Energy Related Applications : Book of abstracts T1 - Data science and deep learning for the development of new hydrogen storage materials SP - 18 EP - 18 UR - https://hdl.handle.net/21.15107/rcub_vinar_11250 ER -
@conference{ author = "Batalović, Katarina and Radaković, Jana and Paskaš Mamula, Bojana", year = "2021", abstract = "Prediction of metal hydride formation enthalpy is one of the key requirements for a rapid design of new hydrogen storage materials. In the last decades, DFT (density functional theory) approach showed good predictive potential for the ground state properties and accurate energies of hydride formation. However, calculating ZPE contribution and temperature effects in addition to formation energy at 0K is computationally and time- consuming and therefore often avoided, resulting in discrepancy to experiment. The development of machine learning and, in particular, deep learning, opens a new perspective for predictive modeling of materials properties. Data collected through DFT calculations can be combined with experimental results in a predictive model, aiming to exploit unexplored compositional space. In this work, we consider the application of MatErials Graph Network (MEGNet) [1] to the prediction of hydrogen formation behavior, and screening of potential dopants in reversible metal hydride materials. Various approaches, relying on transfer learning and both experimental data and computational repositories (MP [2], NOMAD [3]) are proposed as a route to accurate prediction of a structure-property relation for hydrogen storage materials. Domains of applicability of these models are addressed.", publisher = "Belgrade : Faculty of Physical Chemistry, University of Belgrade", journal = "4th International Meeting on Materials Science for Energy Related Applications : Book of abstracts", title = "Data science and deep learning for the development of new hydrogen storage materials", pages = "18-18", url = "https://hdl.handle.net/21.15107/rcub_vinar_11250" }
Batalović, K., Radaković, J.,& Paskaš Mamula, B.. (2021). Data science and deep learning for the development of new hydrogen storage materials. in 4th International Meeting on Materials Science for Energy Related Applications : Book of abstracts Belgrade : Faculty of Physical Chemistry, University of Belgrade., 18-18. https://hdl.handle.net/21.15107/rcub_vinar_11250
Batalović K, Radaković J, Paskaš Mamula B. Data science and deep learning for the development of new hydrogen storage materials. in 4th International Meeting on Materials Science for Energy Related Applications : Book of abstracts. 2021;:18-18. https://hdl.handle.net/21.15107/rcub_vinar_11250 .
Batalović, Katarina, Radaković, Jana, Paskaš Mamula, Bojana, "Data science and deep learning for the development of new hydrogen storage materials" in 4th International Meeting on Materials Science for Energy Related Applications : Book of abstracts (2021):18-18, https://hdl.handle.net/21.15107/rcub_vinar_11250 .