A Data-Mining Framework for Uncertainty Analysis in Pipeline Erosion Modeling
Type of DegreePhD Dissertation
MetadataShow full item record
The transport of solids in multiphase flows is common practice in energy industries due to the unavoidable extraction of solids from oil and gas bearing reservoirs either from onshore or offshore sites. The safe and efficient operation requires reliable estimates of erosion happening in the pipelines. The phenomenon that leads to erosion, especially in multiphase flow systems, is very complex and depends on many factors including the fluid and solid characteristics, the pipeline material properties, and the geometry of the flow lines. Semi-mechanistic models have been developed to quantify the erosion rate within pipelines that transport solids in both single-phase and multiphase flows. However, due to the process complexity, most of the developed models tend to have large model uncertainties, even in interpolated regions, and fail to provide an accurate estimates of confidence levels of their predictions based on sound uncertainty analysis. The overall goal of this dissertation is to develop uncertainty quantification and propagation approaches for multiphase flow model predictions focusing on cases where solids are present in the system. Three sources of uncertainty identified and implemented are model form uncertainty, input uncertainty, and experimental data uncertainty. The first two uncertainties are resolved based on the Gaussian Process Modeling (GPM) framework, where data clustering, data transformation and dimensional analysis are applied to enhance the prediction reliability. For data uncertainty, a unique approach to estimate and combine variable-dependent data uncertainty is proposed. The proposed framework is applied to one commonly used erosion model. Uncertainty of the model, which spans six orders of magnitude has been accurately captured by the framework. By using a novel clustering approach, prediction accuracy has been further enhanced according to grouping data from a wide spread of operating conditions. The physical explanation for regions that lead to high uncertainty has also been investigated using the dimensionless numbers obtained from dimensional analysis. The methodology developed as a result of this dissertation can be applied to uncertainty analysis for other erosion models. For given operating conditions, uncertainty of erosion predictions can be obtained together with confidence intervals. The uncertainty analysis can be used as a guideline for field production and future research on erosion process.