Investigation of backwards erosion piping by data driven modeling

Abstract

This thesis investigates backwards erosion piping (BEP) as a failure mechanism; it focuses on the ability to use pore water pressure (PWP) measurements to (1) monitor, (2) investigate progression of, and (3) better predict BEP. This thesis starts with understanding PWP trends as BEP progresses, analyzed from PWP collected during a full scale experiment. The result is the understanding that the PWP trends are spatially and temporally complex, indicating that BEP does not progress in a spatially linear pattern, such as a longitudinally progressing BEP channel. The comparison of PWP trends from BEP with PWP trends calculated from finite element models corroborate previous findings that BEP does not progress in a simple manner. BEP does not progress smoothly in time nor does it progress symmetrically or only longitudinally (directly from downstream to upstream) in space. This research further provides an understanding that finite element models generate similar trends in PWP as observed during the full scale experiment, suggesting that the use of inversion analysis to predict spatial progression of BEP from PWP measurements is appropriate. The inversion analysis is validated by building synthetic models that simulate BEP in a finite element software, and investigating how the inversion analysis is able to re-create these synthetic models. Once the inversion analysis is validated, PWP measurements from the IJkdijk 2009 Test 2 full scale experiment are used to estimate the spatial progression of BEP at consecutive time steps. The results indicate a highly non-uniform progression of BEP. Multiple BEP channels progress transversely and longitudinally to bulk fluid flow (from upstream to downstream) and appear to progress, halt, and progress again later in time. Observations taken during the full scale experiment corroborate these findings. To complete the thesis, PWP measurements in space and time are used to investigate the ability to use Bayesian updating to decrease understanding of parameters that are spatially heterogeneous and update the prediction of BEP leading to failure.