Show simple item record

dc.contributor.authorChen, Pengen_US
dc.date.accessioned2015-01-07T20:57:36Z
dc.date.available2019-08-19T06:01:11Z
dc.date.issued2014-08-18en_US
dc.identifier.otherbibid: 8793429
dc.identifier.urihttps://hdl.handle.net/1813/38898
dc.description.abstractUncertainty propagation (UP) in physical systems governed by PDEs is a challenging problem. This thesis addresses the development of a number of innovative techniques that emphasize the need for high-dimensionality modeling, resolving discontinuities in the stochastic space and considering the computational expense of forward solvers. Both Bayesian and non-Bayesian approaches are considered. Applications demonstrating the developed techniques are investigated in the context of flow in porous media and reservoir engineering applications. An adaptive locally weighted projection method (ALWPR) is firstly developed. It adaptively selects the needed runs of the forward solver (data collection) to maximize the predictive capability of the method. The methodology effectively learns the local features and accurately quantifies the uncertainty in the prediction of the statistics. It could provide predictions and confidence intervals at any query input and can deal with multi-output responses. A probabilistic graphical model framework for uncertainty quantification is next introduced. The high dimensionality issue of the input is addressed by a local model reduction framework. Then the conditional distribution of the multi-output responses on the low dimensional representation of the input field is factorized into a product of local potential functions that are represented non-parametrically. A nonparametric loopy belief propagation algorithm is developed for studying uncertainty quantification directly on the graph. The nonparametric nature of the model is able to efficiently capture non-Gaussian features of the response. Finally an infinite mixture of Multi-output Gaussian Process (MGP) models is presented to effectively deal with many of the difficulties of current UQ methods. This model involves an infinite mixture of MGP's using Dirichlet process priors and is trained using Variational Bayesian Inference. The Bayesian nature of the model allows for the quantification of the uncertainties due to the limited number of simulations. The automatic detection of the mixture components by the Variational Inference algorithm is able to capture discontinuities and localized features without adhering to ad hoc constructions. Finally, correlations between the components of multi-variate responses are captured by the underlying MGP model in a natural way. A summary of suggestions for future research in the area of uncertainty quantification field are given at the end of the thesis.en_US
dc.language.isoen_USen_US
dc.subjectUncertainty quantificationen_US
dc.subjectstochastic partial differential equationen_US
dc.titleNovel Uncertainty Quantification Techniques For Problems Described By Stochastic Partial Differential Equationsen_US
dc.typedissertation or thesisen_US
thesis.degree.disciplineMechanical Engineering
thesis.degree.grantorCornell Universityen_US
thesis.degree.levelDoctor of Philosophy
thesis.degree.namePh. D., Mechanical Engineering
dc.contributor.chairZabaras, Nicholas Johnen_US
dc.contributor.committeeMemberSamorodnitsky, Gennadyen_US
dc.contributor.committeeMemberWarner, Derek H.en_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record

Statistics