Recently, one of the most challenging tasks in the CALPHAD community is the development of the third generation CALPHAD databases and their application to re-assessment of binary, ternary and high-order systems from 0K [1-5, 13]. During the development of the third generation of CALPHAD databases, not only newly available DFT [4] and experimental data [5] should be considered to build a new pure elements database from 0 K, but the existing physical laws [6] and newly discovered relationships [7-9] between relevant thermodynamic properties should be established and integrated. Considering that the classical re-assessment procedure of high order thermodynamic systems is very time consuming, a combination of machine-learning (ML) methods with the well-established CALPHAD-type assessment will be presented as one of possible solution [10-12]. In this work, several successful partial applications of machine learning to accelerate and support the development the 3d generation CALPHAD database from 0K will be demonstrated.

References
[1] I. Roslyakova, et al., CALPHAD 55, 165-180, (2016).
[2] Y. Jiang, et al., CALPHAD 62, 109-118, (2018)
[3] Y. Jiang, et al., Journal of Materials Research, 110, 797-807, (2019)
[4] S. Bigdeli, et al., CALPHAD 65, 79–85, (2019)
[5] A. Khvan, et al. CALPHAD 60, 144-155, (2018)
[6] G. Grimvall, Thermodynamic properties of materials (1986).
[7] C. A. Becker, Phys. Status Solidi B 251, 1, 33–52 (2014)
[8] A. Obaied, et al., CALPHAD 69 (2020) 101762
[9] D. Sergeev, et al., J. Chem. Thermodynamics, 134, 187-194, (2019)
[10] B. Bocklund, et al., MRS Communications: Artificial Intelligence Research Letter, 9, 618-627, (2019)
[11] S. Zomorodpoosh, et al., CALPHAD 71, 101994, (2020)
[12] N. H. Paulson, et al., CALPHAD 68, 1-9, (2020)
[13] E. Zhang, et al., International Journal of Refractory Metals and Hard Materials 103 (2022) 105780