Comparison of statistically-based methods for automated weighting of experimental data in CALPHAD-type assessment

Journal Article · 09 January 2020 · Calphad

DOI: https://doi.org/10.1016/j.calphad.2019.101728 · OSTI ID:1592101

Paulson, Noah Harris ^[1]; Zomorodpoosh, Setareh ^[2]; Roslyakova, Irina ^[2]; Stan, Marius ^[1]

---

1. Argonne National Lab. (ANL), Lemont, IL (United States)
2. Ruhr-Univ. Bochum (Germany)

The selection and weighting of experimental and simulated datasets is a necessary step in the development of thermodynamic property models in the calculation of phase diagrams (CALPHAD) approach. Currently, this requires significant effort on the part of the researcher and involves subjective evaluation of the reliability of datasets. In this work, we present two novel and independently developed statistical approaches to address outliers and perform automated dataset weighting. The first method, presented here for the first time, applies classical statistical techniques and commonly available optimization algorithms. The second method employs Bayesian statistics via numerical sampling techniques. In this work, we present both approaches and compare their strengths and weaknesses through an assessment of the specific heat of aluminum and hafnium metal versus temperature for several experimental datasets. Here, we finally compare the weightings of each dataset versus a number of metrics employed by experts to evaluate the reliability of datasets.

• Research Organization: Argonne National Laboratory (ANL), Argonne, IL (United States)
• Sponsoring Organization: Argonne National Laboratory, Laboratory Directed Research and Development (LDRD); German Research Foundation (DFG); National Institute of Standards and Technology (NIST), Center for Hierarchical Materials Design (CHiMaD); USDOE; USDOE Office of Science (SC)
• Grant/Contract Number: AC02-06CH11357
• OSTI ID: 1592101
• Journal Information: Calphad, Journal Name: Calphad Journal Issue: C Vol. 68; ISSN 0364-5916
• Publisher: ElsevierCopyright Statement
• Country of Publication: United States
• Language: English