Backward Erosion Piping

Backward Erosion Piping is an internal erosion mechanism which occurs beneath embankment dams and levees founded on soil. When the foundation contains uniform sand, backward erosion becomes the major cause of incidents and failures. Yet, despite extensive research over the past century, a robust and accurate method for predicting backward erosion has been elusive. This study provides new insights into many aspects of backward erosion and provides improvements to existing prediction methods, with the use of a comprehensive suite of 92 large-scale laboratory flume experiments. For the first time, the four exit geometries of slope, plane, slot and circle were tested in otherwise identical flumes. Experimental results and complementary numerical modelling demonstrated that exit geometries with more confined outflow areas required both lower initiation and critical gradients because these exits caused higher seepage velocities at both the exit and channel tip, thus needing less gradient to generate necessary erosive forces. Few studies had investigated internally stable soils with uniformity coefficients above 3, therefore, seven of these soils were tested. Critical gradients increased exponentially with increase in uniformity and with decrease in permeability at a constant tip width. Additionally, the tip width increased linearly with d50. The latter two findings were combined to form a new empirical model with a coefficient of correlation of 0.95. To investigate the industry's concern that critical gradient decreases with subsequent floods, novel tests were loaded in cycles. The critical gradient did decrease in experiments by 2-13% but not due to cyclic loading. In fact, gradients needed under cyclic loading were higher than under constant loading. It was due to an increase in permeability as the channel lengthened. An unprecedented investigation into the rate of backward erosion revealed an average channel advance rate of 3mm/minute at critical gradient and a 3-fold increase in this rate with each 10% increment in gradient above critical (in 0.3mm uniform sand). Using experimental results from both this present study and those undertaken by others, an assessment was made of the two most widely used prediction methods- Schmertmann (2000) and Sellmeijer (2011). In doing so, modifications are recommended to improve model performance in forms suitable for industry use