Statistical Methods for Model Building Coupled with the MOOSE Framework

Angelo Cassiadoro*, David S. Mebane and Alejandro Mejia, Benjamin M. Statler College of Engineering and Mineral Resources, West Virginia University, Morgantown, WV 26506

Field (Broad Category): Material Science (Physical Sciences & Engineering) 

Student’s Major: Mechanical and Aerospace Engineering 

The Energy Systems and Materials Simulation group is focused on analysis and model building of chemical and electrochemical systems. The group utilizes a Bayesian approach to model-building, thus requiring acquisition of experimental data. The model-building processes are a series of Bayesian calibrations, in which model parameters are estimated as probability distributions. The calibrations incorporate Markov chain Monte Carlo routines to sample from the “posterior”: the calibration process yields a probability distribution for the parameter space called a posterior distribution. The solver for the current project model is being developed in an external software known as MOOSE (Multiphysics Object Oriented Simulation Environment). MOOSE was chosen due to its extensive capabilities with finite elements and model solving. My objective is to create an interface routine that will allow the MOOSE solver results to be utilized within the C++ calibrations in an effective manner. This involves research into the operation of the calibration routines as well as ways to communicate data effectively with C++. The end goal is to incorporate the general routine into the sequential routine which will allow for parallel computing with MPI. A successful base calibration will allow for more complex model versions to be tested. From each different calibration, one can then utilize different statistical tools to determine which model has the best data coverage and least required complexity. Overall, this relatively new data science technique for model building is efficient, powerful, and may bring greater possibilities when coupled with MOOSE. 

Funding: National Science Foundation 

Program/mechanism supporting research/creative efforts: WVU's Research Apprenticeship Program (RAP) & accompanying HONR 297-level course