Assessing internal erosion10 January 2008
Hans Ronnqvist assesses the potential for internal erosion in embankment dams in Sweden, and investigates methods for identifying structures with a high potential for erosion
The majority of Swedish embankment dams are comprised of broadly graded glacial moraines (tills) in the core, protected by filters of sand and gravel, which are, in many cases, widely graded. It was recognized by Sherard  that dams of this type exhibited signs of internal erosion to a much larger extent than dams consisting of other types of materials. With this comes the complexity that broadly graded core materials bring on the design of protective filters (Sherard & Dunnigan , Lafleur et al. ). The typical Swedish embankment dam is a zoned embankment with a vertical and centrally placed moraine (till) core, with filters of glacifluvial, sand and gravel, and shoulders (support fills) of earth or rockfill (Figure 1). There have been several incidents of reported sinkholes in Swedish embankment dams. Sinkholes can often be attributed to internal erosion in the dam body or its foundation and can serve as an indicator of internal erosion if occurring on the dam crest or the upstream face, together with dirty water (turbidity) and unexpected increases in seepage downstream. A survey carried out by Nilsson et al , to investigate the ageing and deterioration processes of 84 rock and earthfill embankment dams in Sweden, revealed that 27 dams had developed sinkholes on the crest. Of these, 20% were estimated to have experienced some kind of internal erosion either in the dam body or its foundation (Norstedt & Nilsson ). It is necessary to point out that, to date, there have been no dam failures due to internal erosion of large hydro power embankment dams in Sweden, but in an international perspective, icold  statistics show that internal erosion in the dam body or its foundation is a major cause of dam failures.
The process of internal erosion is usually described by the initiation, continuation and progression phases (Foster & Fell ). Initiated internal erosion, resulting in the loss of impervious core function, can in many cases be traced back to a 'root cause' for internal erosion, and the subsequent continuation of the internal erosion is mainly dependent on the protective filter function (Nilsson & Rönnqvist ). These causes include internal instability of the core, coarsely graded filters which are unable to protect the core, or different structural or construction related issues that unintentionally create concentrated leaks and pathways (eg arching effects, low effective stresses or unintentional seepage paths).
Using the example of 45 existing Swedish embankment dams with moraine cores and filters of sand and gravel, this paper illustrates that the majority (approximately 70%) do not satisfy current filter criteria. However, only approximately 20% of these dams have actually developed confirmed internal erosion. This suggests that current filter criteria have built-in factors of safety making them more suitable for design use, while being less effective from an analysis point of view. This paper investigates if a more discriminatory tool can be derived for assessing potential for surfacing internal erosion in embankment dams with widely graded glacial cores. Surfacing internal erosion (Rönnqvist ) refers to the point when the process has exceeded the initiation phase, and passed into the continuation and progression phase of internal erosion and manifested itself in a sinkhole, or other signs.
Questions can be raised about whether it is possible to analytically distinguish glacial moraine core embankment dams with high potential for surfacing internal erosion from dams with low potential, apparently internal erosion free. Introduced in Rönnqvist [2005, 2006], an evaluation of existing Swedish embankment dams is initiated and further developed on inventoried dam records of core and filter gradation curves, mainly from the time of construction, but also in some cases on soil samples taken after the dam has been built.
The notation and terminology used in this paper is: D15 = particle size for which 15% by weight is finer; DF15 = particle size in filter for which 15% by weight is finer; DB95 = particle size in core base soil for which 95% by weight is finer.; H/F = the ratio introduced by Kenney & Lau [1985, 1986] of percentage passing weight between d and 4d on the sieve curve and passing weight at d.
Proposed dam categories
The dams included here are part of an ongoing inventory of Swedish embankment dams being carried out by the author. To date there are 45 dams included in this study and out of these there are 10 cases of glacial moraine core embankment dams with confirmed internal erosion. The distribution of the dams is shown in Figure 2. The composition and structure of the dams in the study varies slightly, but common features are the core of glacial moraine and a sand-gravel filter of some sort. All the dams impervious moraine cores classifies as Sherard & Dunnigan  soil group 2 base soils (40-85% finer than No. 200 (0.075mm) sieve determined from base soils re-graded on the 4.75mm sieve). In this study it has been necessary to categorize the dams with regards to the occurrence of internal erosion:
• 1 dams – embankment dams with probable occurrence of internal erosion. The dam has had incidents with visible sinkholes and settlements on the surface of the dam and leakage with dirty waters (eroded material in suspension) or other observed signs of internal erosion. The presence of internal erosion is documented.
• 2 dams – embankment dams where observations may be signs of internal erosion. Sinkholes and settlements have occurred on the dam, but no leakage with eroded material has been noted.
• 3 dams – embankment dams with no observations to indicate internal erosion. No sinkholes or settlements have occurred on the dam and no leakage with eroded material has been sighted.
Placing existing dams into categories of internal erosion is difficult because in most cases the process of internal erosion is concealed and often progresses very slowly. Ten dams qualify as category 1 dams (Figure 2), and on two of these dams internal erosion – in the shape of sinkholes and turbidity – surfaced almost immediately after the first raising of the reservoir (Nilsson et al ). However, as shown in Figure 2, the majority of the category 1 dams did not show signs of internal erosion until 5-15 years after commissioning.
Core/filter internal stability and filter geometry – results and evaluation
Out of the 45 dams in the study, 10 are documented cases of dams with internal erosion. In Rönnqvist , a selection of methods (i.e. Kenney & Lau [1985, 1986], Sherard & Dunnigan , USACE  and Burenkova ) were applied to the inventoried gradation curves of the core and the protective filter of the embankment dams. It revealed that attempts to single out dams with internal erosion from dams without internal erosion were unsuccessful when using the selected methods mentioned above.
However, as shown in Figure 3, the Kenney & Lau [1985, 1986] method of assessing internal stability in materials proved to be successful in separating the categories, as long as both the base (core) and filter is assessed and the results compared. Of the category 1 dams tested, 80% were found to include an unstable core and unstable filter. The extremes of the categories are effectively separated, as only approximately 12% of the category 3 dams have an unstable core and filter. A material is considered internally unstable if the Kenney & Lau [1985, 1986] H/F ratio is smaller than 1, meaning that the soil has a smaller fines volume in comparison with the volume of voids between coarser load-bearing particles.
Generally due to less strict requirements at the time of construction, and uncertainties in the guidelines, the majority of Swedish dams feature protective filters that are too coarse to meet today's standards. For example, out of the 45 dams included in this study, approximately 70% have filters with D15 coarser than today's guideline of 0.7mm (filter criteria according to Sherard & Dunnigan , soil group 2 base soil). In another inventory of 44 Swedish embankment dams (Norstedt & Nilsson ), the share amounts to more than 64%. The number of dams that actually developed surfacing internal erosion is not related to the number of dams that fail to meet today's filter requirements. This is probably due to the fact that current filter criteria are design-oriented with built-in factors of safety, and are thereby less effective from an analysis standpoint. Hence, no clear pattern can be seen in this case as to why a dam is prone to develop surfacing internal erosion or not, solely from the basis of current filter criteria. The distribution of the dams in relation to filter coarseness is shown in Figure 4.
As can be seen in this figure, if the distribution of the max filter D15 is plotted in terms of category 1 to 3 dams, none of the category 1 dams have a protective filter up to today's standard of 0.7mm (according to Sherard & Dunnigan , soil group 2 base soil). The figure shows that that a more discriminating boundary can be suggested for analysis purposes, namely D15 = 1.4mm. This limit is adjusted for the dams in the study, and provides a more discernible boundary between category 1 and 3 dams compared to current filter criteria – since 90% of the category 1 dams have a filter D15 coarser or equal to 1.4mm, as opposed to about 30% of the dams without internal erosion. The CE-boundary in Figure 4 (D15 = 93d95), is the continuing erosion boundary from Foster & Fell  and is a proposed boundary beyond which a filter is considered too coarse to seize the migration of fines from a base soil.
When combining filter internal stability (according to the method developed by Kenney & Lau [1985, 1986]) with filter D15, a distribution comes up as plotted in Figure 5. When the filter internal stability is plotted against the filter coarseness (here in the shape of max D15) a possible correlation appears indicating that the coarser D15 of the filter, the more likely the filter is unstable. Furthermore, from Figure 5, it appears that as the filter gets increasingly coarser and passes the boundary into internal instability, the more likely it is that the dam is a category 1 dam. This indicates that terms like internal instability and filter coarseness can be combined in order to efficiently identify category 1 dams from category 3 dams. A material is considered unstable if the Kenney & Lau [1985, 1986] H/F ratio is smaller than 1. The '(H/F)min' on the y-axis in Figure 5 is the minimum value obtained of the ratio along the sieve curve within the evaluation range (0-20% passing), stipulated by Kenney & Lau [1985, 1986].
Filter performance - results and evaluation
The Foster & Fell  filter testing method for dams makes it possible to assess the filter performance of existing embankment dams with regards to the no, excessive and continuing erosion boundaries in case of a concentrated leak. Foster & Fell  introduced a link between the filter D15 and the percentage of fine-medium sand in the core material to define the large erosion boundary, suggesting that for Group 2A base soils, the lower the degree of fine-medium sand in the core and the higher filter D15, the more likely is large erosion. Three categories for filter performance in existing dams are suggested in Foster & Fell , namely seal with no erosion, seal with some erosion, and partial or no seal with large erosion.
By applying the Foster & Fell  method on the 45 existing Swedish glacial moraine core embankment dams comprising the study and comparing the result to the dam’s history of internal erosion, the distribution shown in Figure 6 is developed. Dams with probable occurrence of internal erosion are dams where the Foster & Fell  method says erosion is possible (some or large) under a concentrated leak. Furthermore, none of the category 1 dams are likely to seal with no erosion in the case of a leak, which corresponds well to the dams in the study. Approximately 90% of the dams with probable occurrence of internal erosion are assessed as being likely to highly likely to suffer large erosion in the case of a leak, which relates well with the operative history of internal erosion (Figure 6).
On the other hand over 40% of the dams with no observations of internal erosion (category 3 dams) are also placed in the large erosion category, suggesting that the method is on the conservative side. Only one dam out of four in category 3 falls under the 'no erosion' boundary, which suggests that there could be an over-prediction of likelihood for internal erosion. As with the other methods applied to the dams in this study (Kenney & Lau [1985, 1986], Sherard & Dunnigan , USACE  and Burenkova , in detail described in Rönnqvist ), the Foster & Fell  method by itself is unable to clearly distinguish between dams with internal erosion (category 1 dams) from dams without (category 2 and 3 dams). Comparing the overlapping results from the Foster & Fell  and Kenney & Lau [1985, 1986] method reveals that 80% of the category 1 dams have unstable cores and unstable filters, running together with likely large erosion in case of a concentrated leak. The category 3 dams form a sharp contrast to that with 11%.
Proposed method to assess potential for internal erosion in moraine core dams
By putting the filter internal stability (in the shape of the H/Fmin, the minimum value of ratio H/F along the stipulated evaluation range 0-20% passing weight according to Kenney&Lau [1985, 1986]) against the filter coarseness (log of max filter D15) a spread of dams is obtained (Figure 7). The filters at the interface to a core assessed as unstable are indicated with a circle on the plot, and the boundaries proposed from methods by others are also shown. That is the boundary between internally stable and unstable material (Kenney & Lau [1985, 1986] shape curve ratio H/F = 1) and the Sherard & Dunnigan  filter criteria, and later on adopted no erosion boundary in Foster & Fell  D15 = 0.7mm (soil group 2 base soils, 40-85 % finer than No. 200 0.075 mm). Furthermore the proposed filter boundary of D15 = 1.4mm is also included.
As can be interpreted from the plot; most dams without internal erosion (category 2 and 3 dams) have an internally stable filter and a filter D15 finer than 1.4mm. Dams with internal erosion on the other hand generally have an unstable filter and a filter D15 coarser or equal to 1.4mm. The core is assessed as being unstable in most cases, but most distinct in the case of dams with internal erosion. From this it is reasonable to suggest that the potential for surfacing internal erosion is high for dams with a filter and core placing in the shaded area in the lower right corner (where the filter is coarser or equal to 1.4mm and the filter and core are assessed as unstable) or at the far right where the filter's D15 exceeds the CE-boundary, D15>93d95 (Foster & Fell [2001). Furthermore, it is reasonable to suggest that the potential is low for dams in the upper left corner box in Figure 7 (where the filter is finer grained than the ‘no erosion boundary’ of 0.7mm and the filters and cores are evaluated as being internally stable).
This suggests that surfacing internal erosion in embankment dams composed of moraine (till) cores is dependent not only on filter coarseness at this stage, but also on internal stability in the core and filter. This is relevant for dams with filters with somewhat too coarse filters. However, it has been recognized that at some 'tipping point' of the filter coarseness, high potential for surfacing internal erosion is quite likely reached regardless of the internal stability. Logically, at some critical coarseness of the filter, a sufficiently coarse filter would, independent of the internal stability or instability of the core or filter, wash out the core enough to create sinkholes. In Foster & Fell [1999, 2001] a continuing erosion boundary is suggested, i.e. DF15>9DB95, that suggests, when exceeded, a filter is too coarse to allow the eroded core particles to seal the filter. Foster & Fell  confirmed that the filter D15 is an appropriate indicator for the opening size of a filter (based on D15/9 as first concluded by Sherard et al. [1984a]). With regards to the dams in this study, this would result in a CE-boundary varying approximately between 6<DF15<30mm depending on the dam (the CE-boundary is plotted as indicated in Figure 7). The majority of dams in the study, at least according to the gradation curves from the dam records, do not reach the CE-boundary – the maximum filter D15 for most of the dams in the study vary around 1.0-1.5mm with only a few dams peaking at 3.5-3.7mm (Rönnqvist ).
Clearly one dam deviates quite significantly. This is a glacial moraine core embankment dam having a filter of tunnel spoil with D15 of 32mm. At this particular dam, leakage with dirty water occurred almost immediately after first filling and subsequent investigations revealed that this was due to internal erosion of the core. The dam's filter exceeds the Foster & Fell [1999, 2001] CE-boundary, confirming the relevance of this method. In order to present an upper boundary to which the core and filter's internal stability is still relevant, and a lower boundary beyond which only the filter coarseness is important (regardless of whether the filter and core are stable or not), the Foster & Fell  CE-boundary should be considered.
The study shows that in order to be able to differentiate between dams with internal erosion and dams without it is necessary to include the properties of both the filter as well as the core into the assessment, assuming that the internal erosion process initiates from basic suffusion (internal instability) and backward erosion from an unfiltered interface. In regards to the conceptual event tree shown in Figure 8 three basic questions can be raised while assessing potential for surfacing internal erosion: (1) Is the core internally stable or is it likely that fines from the core can migrate?; (2) Is the filter appropriately graded and sufficiently fine-grained to stop migration of the core fines at proximity of the interface?
Seeing that it is not uncommon with a considerable 'fine tail' even in the filters of some Swedish embankment dams (some of the dams in the study had fines content (material finer than 0.075mm) in the filter up to 20-30% (Rönnqvist ) it is reasonable to suggest that the filter itself can experience internal erosion if the seepage velocity is high enough. Foster & Fell  indicate that particle sizes up to fine sand fraction (<0.2mm) can be readily transported in concentrated leaks, and if the concentrated leak increases in size, particles up to about 5mm would also be transported. With this in mind, assuming that the seepage velocity increases due to internal erosion (causing a property change in the core) the filter's internal stability can come into play. The third question can therefore be raised regarding continuation: (3) Is the filter internally unstable permitting fine particles in the filter to move in case the seepage velocity is sufficiently high?
Based on the research results, a method is proposed in Table 1 in which internal erosion is approached from the core/filter interaction. The method addresses the potential for surfacing internal erosion in moraine core embankment dams with levels from high, increased or neutral if the filter D15 = 1.4mm. High potential is located as indicated in Figure 7 as the shaded area. With finer grained filter the potential ranges from Neutral, Reduced to Low. The basic internal erosion scenario that can be suggested from the result of the assessment is schematically outlined in Figure 8, which shows a possible chain of events leading up to high potential for surfacing internal erosion in glacial moraine core embankment dams.
The method indicates that an unstable filter appears necessary for high potential for surfacing internal erosion, unless the filter is excessively coarse (exceeding the Foster & Fell [1999, 2001] continuing erosion boundary). The underlying logic to the relevancy of filter instability for potential of surfacing internal erosion is that if the seepage velocity gets sufficiently high at the core/filter interface, the filter, if unstable, starts losing fines progressively perhaps to the point that it is closing in on some critical coarseness, i.e. approaching the CE-boundary of Foster & Fell [1999, 2001].
When dealing with surfacing internal erosion in moraine core embankment dams, not only should the filter coarseness be addressed, but also the internal stability of the filter and core material. Analyzing these properties enhances the accuracy in analytically finding embankment dams prone to develop surfacing internal erosion; dams that previously were, from an internal erosion perspective, more or less undetectable.
The overall objective of the proposed method in Table 1 is to offer a means for assessing glacial moraine core embankment dams and their potential for developing surfacing internal erosion. Experience of Swedish embankment dams shows that from a dam safety point of view there is often a need to improve the resistance of existing dams against internal erosion (Nilsson & Rönnqvist ). The proposed method can provide a possible screening tool that can assist in tracing dams prone to developing internal erosion. A screening tool can prove particularly useful when prioritizing remedial works for dams. The method in Table 1 is still in development, but once complete it will help provide a qualitative indication of the potential for sinkhole development.
It is necessary to point out that when it comes to the design of protective filters for moraine (till) cores this paper does not recommend a relaxation of the generally accepted design criteria (such as Sherard & Dunnigan ), but rather to promote these in order to create reliable filters that operate effectively in the event of an initiated internal erosion process.
Hans Rönnqvist, Vattenfall Power Consultant AB, Hydropower Division, Dam Safety department
Postgraduate student, The Royal Institute of Technology, Dep. of Land and Water Resources Engineering, Hydraulic Engineering
The author wishes to thank Åke Nilsson, Vattenfall Power Consultant, for valuable discussions. Financial support has been received from the Swedish Hydropower Centre (www.SVC.nu) and Vattenfall AB Vattenkraft
Burenkova, V.V. (1993) Assessment of suffusion in non-cohesive and graded soils, Proceedings, the First International Conference Ã¢€Å“Geo-Filters", Karlsruhe, Germany, 20-22 Oct 1992, Filters in Geotechnical Engineering, Brauns, Heibum & Schuler (eds), 1993 Balkema Rotterdam, pp. 357-360.