Stochastic Data Assimilation with Application to Multi-Phase Flow and....

Stochastic Data Assimilation with Application to Multi-Phase
Flow and Health Monitoring ProblemsPh.D. Defense by:
George Saad, CE Ph.D. CandidateDepartment of Civil and Environmental Engineering
University of Southern California
Los Angeles, California 90089
gsaad@usc.eduModel-based predictions are critically dependent on assumptions and hypotheses that are not based on first principles and that cannot necessarily be justified based on known prevalent physics. Constitutive models, for instance, fall under this category. While these predictive tools are typically calibrated using observational data, little is usually done with the scatter in the thus-calibrated model parameters. In this study, this scatter is used to characterize the parameters as stochastic processes and a procedure is developed to carry out model validation for ascertaining the confidence in the predictions from the model. Most parameters in model-based predictive tools are heterogeneous in nature and have a large range of variability. Thus the study aims at improving these predictive tools by using the Polynomial Chaos methodology to capture this heterogeneity and provide a more realistic description of the system's behavior. Consequently, a data assimilation technique based on forecasting the error statistics using the Polynomial Chaos methodology is developed. The proposed method allows the propagation of a stochastic representation of the unknown variables using Polynomial Chaos instead of propagating an ensemble of model states forward in time as is suggested within the framework of the Ensemble Kalman Filter (EnKF). This overcomes some of the drawbacks of the EnKF. Using the proposed method, the update preserves all the statistics of the posterior unlike the EnKF which maintains the first two moments only. At any instant in time, the probability density function of the model state or parameters can be easily obtained by simulating the Polynomial Chaos basis. Furthermore it allows representation of non-Gaussian measurement and parameter uncertainties in a simpler, less taxing way without the necessity of managing a large ensemble. The proposed method is used for realistic nonlinear models, and its efficiency is first demonstrated for reservoir characterization using automatic history matching and then for tracking the fluid front dynamics to maximize the waterflooding sweeping efficiency by controlling the injection rates. The developed methodology is also used for system identification of civil structures with strong nonlinear behavior.

Location: Kaprielian Hall (KAP) - rielian Hall 209

Audiences: Everyone Is Invited

Contact: Evangeline Reyes