CALPHAD in the Cloud

In this Phase II SBIR program, QuesTek Innovations LLC, a leader in the field of integrated computational materials engineering (ICME), proposes to develop a methodology and software suite for applying Bayesian statistics to fit CALculation of PHAse Diagrams (CALPHAD)-based thermodynamic databases to quantify, store, and propagate the uncertainty in the chemical thermodynamic models. QuesTek will build off of the success of the Phase I program at demonstrating that such a methodology can be used to quantify and propagate uncertainty through thermodynamic calculations to apply Bayesian CALPHAD to databases of quaternary and higher order materials systems. As a partner of the Center for Hierarchical Materials Design (CHiMaD), QuesTek will connect this software suite with web-aware tools for sharing materials data developed in CHiMaD such as the Materials Data Curation System (MDCS). QuesTek will work with Thermo-Calc Inc., world-leader in CALPHAD thermodynamic software, to ensure the developed methodology will be robust and well-suited for commercial CALPHAD software. In Phase I of this program, QuesTek has applied its expertise in ICME and CALPHAD to develop web-based tools for performing chemical thermodynamics calculations and for working with CALPHAD databases. QuesTek has worked with Citrine Informatics to develop CALPHAD database and experimental data file search functionality for Citrine’s web-based platform. QuesTek developed a python-based web application for performing CALPHAD thermodynamic calculations using the Thermo-Calc software platform. QuesTek has also developed a Bayesian framework for incorporating uncertainty quantification into CALPHAD model databases and propagating this uncertainty through chemical thermodynamic calculations to provide confidence intervals on predicted thermo- chemical quantities. QuesTek has demonstrated this framework on a thermoelectric alloy system PbS—PbTe that has potential applications for waste-heat recovery and power generation. In Phase II of this program, QuesTek will build out this uncertainty quantification/propagation framework to apply to large CALPHAD thermodynamic databases containing up to hundreds or thousands of model parameters. These types of databases are used in practice to design and develop new materials. A key component of incorporating uncertainty quantification into CALPHAD thermodynamics is a tight coupling between the underlying thermodynamic models, the numerical minimizer used to calculate equilibrium, and the Monte Carlo simulations used to sample model parameter space. To this end, QuesTek will develop the methodology of Bayesian-based CALPHAD thermodynamics will be developed on pycalphad, open-source python-based thermodynamics calculation software that can be tightly integrated with pymc3, open-source python-based Bayesian modeling software. Another key component of incorporating uncertainty quantification into CALPHAD thermodynamics is the storage and dissemination of the uncertainty associated with a fitted thermodynamic model. QuesTek will leverage its partnership with the Center for Hierarchical Materials Design (CHiMaD), a NIST Center of Excellence, to disseminate the uncertainty-quantified data generated in this program. CHiMaD has several web-based materials tools that enable sharing of materials data including the Materials Data Curation System (MDCS) and Materials Data Facility (MDF). QuesTek will develop schema for the Bayesian CALPHAD databases developed in this program that will allow this type of data structure to be shared through these web-aware materials platforms.