Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Surrogate Modeling and Global Sensitivity Analysis towards Efficient Simulation of Nuclear Reactor Stochastic Dynamics

Banyay, Gregory (2019) Surrogate Modeling and Global Sensitivity Analysis towards Efficient Simulation of Nuclear Reactor Stochastic Dynamics. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

This is the latest version of this item.

[img] Other (XML Bibliography)

Download (42kB)
Download (3MB) | Preview


Surrogate modeling is used to support global sensitivity analyses (GSA) for the modeling and simulation of nuclear reactor assembly structural dynamics to demonstrate the pertinence of such methods to this application as well as the significant physical insights provided by GSA. In addition to the knowledge gained related to the system sensitivity, insight gained from the accuracy of the GSA results may be used to compare with goodness-of-fit metrics that are traditionally used for verification of the surrogate model. The coupled use of surrogate modeling and GSA reduces the number of full-order simulations required, substantially reducing total computational cost. This work focuses on the use of Gaussian Process surrogates in particular, and examines the robustness of these techniques to evaluate sensitivity by considering a variety of design of experiment strategies used to create the surrogate models.
Numerical experiments based upon two finite element models representing stochastic dynamics for a pressurized water reactor, are used to evaluate the relationship between sensitivities computed from a full-order model versus those computed from a surrogate model, highlighting the effectiveness of utilizing GSA and surrogate modeling. For the examples presented herein the historical significance of both forcing function characterization and model parameter definition is substantiated, in terms of the GSA providing insight as to dominant contributors to structural dynamic behavior. For large sample sizes, negligible variation in the resultant sensitivities is shown with respect to the particular method by which a computational design of experiment is constructed to train the surrogates, that demonstrates stability and veracity of the results. For small sample sizes, the use of Latinized Partially Stratified Sampling (LPSS) provided surrogates and associated sensitivities with lower error as compared to Latin Hypercube Sampling (LHS) and sampling via the Fourier Amplitude Sensitivity Test (FAST). Differences in GSA results imparted by examining time-domain versus spectral acceleration results, as well as increasing model parameter variation further illustrated the effectiveness of advanced sampling methods. Furthermore, the use of adaptive sampling and aggregate surrogate modeling techniques are introduced, with which incremental improvements were realized regarding the number of samples required to achieve accurate surrogate models.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Banyay, Gregorygab58@pitt.edugab58
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairBrigham,
Committee MemberKhanna,
Committee MemberLin,
Committee MemberShields,
Date: 21 February 2019
Defense Date: 8 February 2019
Approval Date: 18 June 2019
Submission Date: 23 February 2019
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Number of Pages: 175
Institution: University of Pittsburgh
Schools and Programs: Swanson School of Engineering > Civil and Environmental Engineering
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: flow-induced vibration, surrogate modeling, global sensitivity analysis, nuclear reactor, stochastic dynamics, computational mechanics
Additional Information:, e-mail after graduation
Date Deposited: 18 Jun 2019 20:07
Last Modified: 18 Jun 2019 20:07

Available Versions of this Item


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item