Description: Propagation of parameter uncertainties through large computer models can be very resource intensive. Frameworks and tools for uncertainty quantification are generally geared to individual codes, are research codes, or are single-purpose tools such as LHS matrix generators. The Reduced-Order-Clustering-Uncertainty-Quantification (ROCUQ) methodology discussed in this proposal is specifically designed to circumvent many of the issues associated with uncertainty quantification of large simulation codes. The ROCUQ methodology has been applied in several different physical disciplines with good results. The computational methodology is a combination of reduced-order modeling, stratified sampling (Latin Hypercube Sampling LHS), statistical clustering of results (K-means clustering) and a few (five to ten) full-physics runs of the high-fidelity model under investigation. The method should be applicable to hundreds of uncertain variables when required. ROCUQ enables estimates of system response quantities (SRQ) uncertainty distributions for situations where it is not feasible to use purely sampling, collocation, or other techniques where many runs would be required. For some organizations, uncertainty analysis has never been possible due to resource limitations, and thus is not part of the organizational culture. Many analysts know that uncertainties can be important, but have no way to expend sufficient resources (money, CPU cycles, time) to do the work needed to quantify uncertainties. A methodology such as ROCUQ promises to open doors in organizations that know that they have the need, and may now be able to actually perform the analyses. Successful completion of the Phase II project will produce not only new software that will be able to be used by researchers and industry, but will assemble insights on the use of reduced order models in a variety of disciplines, and provide guidance and rules for the use of ROCUQ for the estimation of SRQ uncertainty distributions.

Benefits: This program will provide pathways to two commercial products: software and engineering services. Engineering services: Consulting services will be available based on the extensibility of the proposed system. IllinoisRocstar has the broad-based experience with a wide variety of supercomputing platforms to allow support of the proposed system on platforms located at NASA, DoD components, DOE, and private companies. Assisting companies and government agencies with customization of the reduced-order models for their specific applications will provide a market, as well as a source of reduced-order library models for these services.

This program will provide pathways to two commercial products: software and engineering services. Software: The completed module will be architected in such as manner as to allow introduction of specialized reduced-order model modules without requiring changes to the base software. IllinoisRocstar has significant expertise in building modular, extensible software. As a module operable within the open-source Dakota framework, it will be useable by a wide variety of entities and organizations.