Data-driven Thermodynamic Modelling and Uncertainty Quantification of the Binary Iron-Carbon System

Johannes Bechter

Research output: ThesisMaster's Thesis

6 Downloads (Pure)


The mostly accepted CALPHAD assessment of the binary iron-carbon system is from Gustafson (Scand. J. Metall. 14.5 (1985): 259-267). As is the case for most CALPHAD assessments, the proposed parameterization reports the chosen values of parameters without providing details about the procedure used to identify the optimal values and without information about their uncertainty or reliability. Therefore, the parametrization is not fully reproducible. In this thesis, a database was created which contains the original thermodynamic data on phase boundaries, activities and formation enthalpies from experiments and ab-initio calculations along with the relevant meta-data specifying e.g. the original reference or experimental details. This python-based database allows adaption to user-specific requirements and easy reassessment at a later time, for example when new data is added. The parameter optimization for the CALPHAD assessment was performed with ESPEI. The open-source software optimizes the parameters with Markov Chain Monte Carlo in the Bayesian framework and provides the associated probability distribution. This allows exploring propagation of parameter uncertainties and investigating the effect of choosing different sets of input data or model structure. As a result, a new parametrization for the Fe-C system is presented which is fully reproducible and explains the thermodynamic data points with a higher probability compared to the parameter set proposed by Gustafson within the chosen error definition. The optimized phase diagram is presented along with the relating parameters in form of a tdb-file which considers the underlying data and estimates the uncertainties of the calculated phase boundaries. It is found that various data types help select reliable datasets and increase the accuracy of an assessment. Based on the calculated uncertainties it is pointed out that providing new datapoints in the eutectic and the high carbon-region can further improve the reliability of the assessment.
Translated title of the contributionDatengestützte thermodynamische Modellierung und Quantifizierung der Unsicherheitsvorhersage für das binäre Eisen-Kohlenstoff-System
Original languageEnglish
Awarding Institution
  • Montanuniversität
  • Romaner, Lorenz, Supervisor (internal)
Award date21 Oct 2022
Publication statusPublished - 2022

Bibliographical note

no embargo


  • thermodynamic modelling
  • thermodynamic database
  • iron-carbon
  • phase diagram
  • uncertainty quantification

Cite this