Teton dam was designed and built to the modern standards of the 1970s; however, the 123m high zoned earthfill dam failed during its first filling on 5 June 1976. The dam was constructed on the Teton river (a tributary of the Snake river) as part of a major irrigation-power-flood control scheme in the high plateau of southeastern Idaho, US (Figure 1). The failure of the dam resulted in 14 fatalities and a very large economic loss. Its failure was one of the most publicised events at that time involving a large earthfill dam built using current standards. Therefore, it received considerable attention from engineering experts around the world. However, the failure assessment and prognosis by experts including those by the Independent Panel (IP, 1976) and the Interior Review Group (IRG, 1980) failed to arrive at a conclusive explanation. Failure mechanisms suggested included hydraulic fractures, internal erosion, the wet-seam theory, and defects in the abutment rock. However, there remained an unanswered question as to why the dam breached when the reservoir reached el. 5301.7ft (1616m) and initiated only in the vicinity of Sta.14+00 on the right abutment.

The impervious core/water barrier (Zone-1) of Teton was constructed of uniform clayey silt of low plasticity and low liquidity index. Highly compacted soils of low plasticity tend to crack in an environment of low liquidity index, low confining stresses and high shear stresses. None of the previous investigations focused on the possibility of the presence of cracks in the upper portions of the dam. This critical feature is investigated here using the fundamental concepts of the modern framework of ‘state based soil mechanics’ (Pillai and Muhunthan, 2001, 2002). The investigation consisted of laboratory tests on Zone-1 material to determine the physical and mechanical parameters and finite element analysis conducted using ABAQUS to simulate field stress conditions.

Background

The design cross section of the Teton dam at the river valley and the right abutment are as shown in Figure 2 and Figure 3, respectively. The construction of the dam began in June 1972 and was completed in November 1975. The dam was conservatively designed to have a wide impervious core with a head to width ratio of about 1.5 (Figures 2 and 3). The bedrock consisted of open-jointed rhyolite and basalt but was well treated with blanket and curtain grouting. The abutment rock was trenched to provide a large core-rock contact and a long flow path to have a low seepage gradient (Figure 3).

The impervious core (Zone-1) of the dam consisted of clayey silts of aeolion origin with low plasticity (PI ~ 4) and USCS classification of CL- ML. As per the design and specifications Zone-1 material was placed at average water contents of 1% dry of optimum and compacted to a maximum dry density of 98-102% of the Standard Proctor test (Figure 4). Similarly, the support zone (Zone-2) (chimney filter/drain) was compacted to a high relative density of the order of 65-70% (IRG, 1980).

The first filling of the reservoir began on 3 October 1975. The rate of filling of the reservoir was about a foot per day (0.3m) in the early stages; however, it was increased to about 3ft (0.9m) per day for the most part of May and June 1976. When the dam breached on 5 June 1976 the reservoir had reached only el. 5301.7ft (1616m), which was about 22ft (6.7m) less than the design full pool elevation.

Surface manifestations

On or before 3 June 1976 (reservoir level was at or below el. 5297ft [1614.5m]), no unusual signs of distress or springs or other water seepage were noticed downstream of the dam. On 4 June, minor evidence of clear seepages appeared downstream, a good 1300-1500ft (396-457m) distance from the toe, which was consistent with the raising of the ground water regime due to rising reservoir water level. Late in the evening of 4 June (reservoir el. 5300ft [1615m]), some dampness was noticed in the downstream dam slope at the right abutment at el. 5200ft (1585m). The following morning on 5 June, shortly after 7:00AM (reservoir el. 5301.3ft [1615.8m]), some muddy water was first observed to be flowing from the junction of the embankment and the abutment at el. 5200ft [1585m]. At 10:30AM, a large leak of about 15cfs (0.4m3/sec) appeared with a ‘burst’ on the downstream at el 5200ft (1585m). The leak appeared to emerge from a tunnel of about 6ft (1.8m) in diameter from inside the embankment and roughly perpendicular to the dam axis at Sta. 15+25. At about 11:00AM, a vortex appeared in the reservoir near Sta. 14+00 above the upstream slope of the embankment. At 11:30AM, a sinkhole on the downstream slope (el. 5315ft [1620m]) developed near the crest and above the leaky tunnel. At 11:55AM, the crest of the dam began to collapse between the vortex and the sinkhole, leading to a full breach at 11:59AM (IP 1976).

Fracture, rupture and ductile behaviour of soils

Aggregates of grains that form natural and man-made soil deposits exhibit three distinct classes of behaviour (Figure 5); at large depths, high pressures cause ductile yielding of the aggregates and the layer of sediments to fold; above these depths and at lower pressures aggregates rupture and a layer of sediment faults with the presence of gouge material along the slip planes; near the surface where the pressure is even lower, a layer of sediment fractures or cracks (Muhunthan and Schofield, 2000).

Critical state soil mechanics (Schofield and Wroth, 1968) captures these simple depositional and structural phenomena of folds, faults, and fractures in soil and sedimentary as well as man-made deposits in a scientific manner. It explicitly recognises that soil is an aggregate of interlocking frictional particles and the regimes of soil behaviour depend in a major way on its density and effective pressure.

In the critical state framework, the state of soils is defined in 3D: mean effective normal stress (p), shear stress (q) and void ratio or specific volume (v) space. Limits to stable states of yielding are defined by the state boundary surface in the 3-D, p-q-e space. The 2D representations of the normalised state boundary surface in the q/pcrit – p/pcrit and e – ln p spaces are as shown in Figure 6.

Critical state soil mechanics divides the soil behaviour at limiting states into three distinct classes of failure; the limiting lines OA and OG (Figure 6a) indicate states of soils undergoing fractures or cracks; AB and GE indicate that Hvorslev’s Coulomb faults on rupture planes; BD and ED indicate Cam-clay yield and fold of a sediment layer.

Soil states on the crack surface result in the development of unstable fissures and crack openings. Heavily overconsolidated clays and overcompacted sands at low confining stresses could reach this limiting state. Collapse similar to fracture on the dilative side can also exist on the contractive domain but outside the normal consolidation line (Figure 6b). Such states outside the stable yielding exist in wind deposited loose sands, air pluviated or moist-tamped sands and result in abrupt collapse upon shearing of these materials (Pillai and Muhunthan, 2001, 2002). For sands and clayey silts of low plasticity, stable yield (rupture and ductile) behaviour occurs only within a narrow band on both the denser and looser side of the critical state line (Figure 6b).

The ‘no tension’ or ‘limiting tensile strain’ criteria are the most widely used among the alternative theories to quantify tensile fracture (Schofield, 1980). For the triaxial specimen the no tension criterion with s3 = 0 results in p = s1/3 or q/p = 3 and leads to vertical split cracks which is the case of line OA. For horizontally spalling cracks, s1= 0 results in p = 2/3 s3, q = -s3, or q/p = 1.5 which is the case of line OG. For clays or silty clays, Schofield (1980) had suggested that the change from rupture to tensile crack occurs at a pressure p = 0.1 pc, where pc is the effective confining stress at critical state. This is equivalent to an over consolidation ratio of approximately 20 (Figure 6a). When the effective stress path crosses the crack surface OA, the soil element in that location begins to disintegrate into a clastic body and unstressed grains become free to slide apart. In that case the average specific volume of the clastic mass can increase (large voids/cracks) and consequently its permeability can increase significantly and instantly. A significant internal/external shear stress at low confining stresses can cause the crossover of the crack-surface OA and a large increase in specific volume. When such condition occurs, the opening within the soil body may be an extensive crack or a local pipe or channel. If such opening (crack/channel) daylights into the water body it could lead to a free flow of water into the downstream slope.

Liquidity index, confining stress and soil behaviour

Critical state soil mechanics (Schofield and Wroth, 1968) has shown that it is possible to generalise the density or specific volume axis by converting to a liquidity basis. It was further shown that the critical pressure is about 5kPa at the liquid limit and 500kPa at the plastic limit. In his Rankine lecture, Schofield (1980) mapped the remolded soil behaviour on a liquidity against pressure diagram (as shown in Figure 7) utilising the hundred fold increase in pressure from the liquid limit critical state to the plastic limit critical state which is two log cycles, so the rupture band has half the width of PI and will intersect the line p = 5 kPa at LI = 0.5. This intersection is a consequence of putting the lower limit of Coulomb rupture at p/pcrit= 0.1 (Schofield, 1980). In the LI-p space, clear boundaries exists that separate the regions of fracture, rupture, and ductile behaviour. This provides for an independent and convenient approach to separate the states of fracture/rupture/ductile yield behaviour of soil using its physical properties.

Consider a body of soil initially at LI = 0.5 subjected to an elastic compression; the map suggests at shallow depths where p < 5kPa there may be cracks, but for depths where 5 kPa < p < 50 kPa the soil will remain watertight while deforming. In contrast, a body of soil initially at LI = 0 will undergo fracture at depths for which p < 50 kPa or about 3m of the overburden depth. In other words, the overburden depth should be larger than 3m to ensure that deformation caused rupture planes (watertight) rather than open cracks. If LI= -0.25, the depth could be about 100kPa or 6m of depth. In this view the vertical face of the breach in Teton dam can be seen as an open fracture in very strong soil, standing to a near vertical height of 6m+.

In order to identify the band of behaviour in which various states of soil lie in the LI-p space, Schofield (1980) defined their equivalent liquidities by projecting these states in the direction parallel to the critical state line towards the ordinate through p = 5 kN/m2. The equivalent liquidity LI5 can be shown to be LI5 = LI+1/2 log (p/5) (Schofield, 1980). Thus, equivalent liquidity equals liquidity as found in the ground plus a correction for mean stress. A value of LI5 of less than 0.5 would generally indicate the fracture zone. Values of 0.5 to 1.0 represent the rupture zone. Values larger than 1.0 represent ductile zone.

The inset of Figure 7 shows the section of the behaviour map at constant p: stress ratios q/p will increase as equivalent liquidity falls. In the high equivalent liquidity range, stress ratio increases linearly as liquidity of cam-clay falls. The Hvorslev surface gives the rupture limits which allow higher stress ratios as lower values of p/pcrit are approached, but at the no tension limits, q/p = 3 in compression, and -1.5 in extension. There is a general increase of limiting stress ratio as equivalent liquidity falls, but this is not a continuous change because there is a change of limiting behaviour from contours yield, to discrete rupture, to fracture of stiff fissured soil at equivalent liquidity below 0.5 (Schofield, 1980).

The above concepts provide two independent approaches to analysing the cracking of soils particularly in an earth dam. The first approach makes use of mechanical properties determined from triaxial tests and oedometer tests to separate the three regions of soil behaviour, the fractures, the faults, and the ductile yield while the second approach relies on physical properties, plasticity index, and liquidity index to identify such regions.

Parameters of the impervious core

A large database of field and laboratory tests carried out during the post-failure investigations by the IRG and the IP exists in their reports. The laboratory testing herein was focused on the verification of some of the index and mechanical properties. About 1000 lbs of the zone-1 material was obtained from the remnants of the failed Teton dam. The material was tested for physical and mechanical properties in the laboratory. Tests for physical properties included grain size, Atterberg limits, and proctor compaction curves. Mechanical tests included CU triaxial tests on remolded soils, UU triaxial compression tests, and 1-D compression curves on compacted samples at wopt-1, wopt, and wopt+1 to obtain constrained modulus at various confining stress levels.

FEM analysis of teton

Finite element analyses were carried out for the longitudinal section of the dam. This section was chosen because it captures all of the variation along the bottom profile (berms, slopes, etc.). Plane strain condition is assumed to prevail along the section.

The analyses used an elasto-plastic model with modified cam-clay yield curve (Roscoe and Burland 1968). The CSL line with a slope M divides the yield curve into two regions, dry and wet sides. Porous elastic option is used to describe elastic behaviour inside the yield curve. It is assumed valid for small strains (<5%) and is a nonlinear isotropic model in which the pressure varies as an exponential function of volumetric strain. The model parameters used in the analysis are shown in Table 1 above.

The model had five layers to simulate the construction of the dam. In the first step, the top four layers were removed and the remaining layer was analysed. This was to allow the geostatic stress field to reach equilibrium with initial conditions, applied load, and boundary conditions. Subsequently, each layer was activated strain-free to simulate the construction steps. The strain-free activation scheme was adopted to avoid the creation of strain by the deformation of the previous layer. From the analysis, the shear stress (q) and the mean stress (p) were obtained along the longitudinal section and contours of q/p ratio were drawn as shown in Figure 8.

Teton soil behaviour in LI5-p space

As described earlier, the transition of soil behaviour from the crack surface region to stable Hvorslev fault region occurs at an equivalent liquidity index of 0.5 corresponding to a confining stress of 0.8psi or 5kPa, or zero liquidity index at confining stress of 8psi (50kPa). Similarly, the Hvorslev-Coulomb rupture regime changes to ductile Cam-clay regime at 80psi (500kPa) (Figure 7). This was further confirmed by a series of consolidometer tests with Zone 1 samples compacted at varying initial liquidity indices. For various confining stresses, the corresponding equivalent liquidity indices LI5 were determined and their position in the LI5-p space was identified. This was transferred to the cross-valley section for the respective confining stresses. A mapping of the contours of equivalent liquidity index for the valley crosses section of the Teton dam was made as shown in Figure 9.

New mechanism of teton failure

The state based soil mechanics theory presented earlier suggests that zones with stress ratio q/p larger than 3 would indicate the presence of a vertical split or crack (Figure 5). It can be seen that the majority of the soil elements of the Teton dam have q/p stress ratios significantly less than 3, indicating that they were intact (Figure 8). However, there were two zones that have q/p ratio larger than 3 (Figure 8). They are from Sta. 13+00 to Sta.15+00 in the right side and from Sta. 26+00 to Sta. 28+00 in the left side.

The results show that at the end of construction the state of stress in the dam core had significantly reached into the crack surface (q/p = 3) region which is an indication of the existence of internal cracks at two locations, Sta.14 + 50 in the right abutment and Sta.26 + 50 in the left abutment. The cracks at Sta. 14+50 were 32ft (9.75m) deep from top of the crest while they were only 10ft (3m) deep at Sta.26+50 (Figure 8). The state based theory further suggests that contours of the q/p ratio less than 3 would indicate the stable nature of the compacted soil, which is the case for soil elements at depth and particularly below 32ft (9.75m) (Figure 8). Therefore, it is concluded that the failure of the Teton dam was initiated as a result of water flowing through the deepest open vertical crack on the right abutment near Sta. 14+50 during the first filling when the water level reached the bottom of the crack, which slowly eroded the crack into a large tunnel leading to the major breach hours later.

The zone-1 core was capped by a 3ft (0.9m) layer of sand and gravel roadbed, which was subjected to continual vibration and compaction by vehicular traffic inhibiting cracks in the layer. Further, the material parameters of the granular bed, their packing, and the characteristics were different from zone-1 material to exhibit cracking. As a result, it was likely that the cracks below in the core zone had apparently not daylighted onto the roadbed and were not visible during first filling. However, numerous transverse cracks daylighted onto the roadbed in the left abutment soon after the dam breach, mostly near Sta. 26+50, where the q/p ratio was close to or larger than 3 for shallow depths.

The contours of LI5 (Figure 9) independently confirm that only shallow depths to about 30ft (9m) between Sta. 14+00 and Sta.+ 16+00 are prone for cracking. Because of the low plasticity (PI ~ 4), the liquidity index was very sensitive to placement water content and its influence on the performance of the soil core, under rapidly changing confining and shear stress conditions, particularly at the abutments. At the steep abutments, depth of the soil column decreases; consequently the soil elements were subjected to decreased confining stress. In effect, the stress states of the soil in the abutments were in the Hvorslev regime and were stiffer while those in the valley section of the dam were at or near the ductile regime, which were more deformable. Again the changes in the deformability were further disrupted by the benches, which apparently caused significant differential deformations and increased shear stresses at some locations. These aspects were well reflected in the stress analysis.

Stress path of teton soil and critique of past investigations

The concepts presented also help explain some of the misgivings of previous investigations. Consider the states of soil element shown in Figure 10 (A1, A2, A3, and A4). At the placement condition, the state of soil would have been in the Hvorslev region at point A1 in Figure 11. As the dam was built up, the confining stress would increase and the state of soil would move along the path A1A2A3A4 (Figure 11). Thus, the soil, which was in the key trench, would move to the stable yielding region when it was wetted.

It is, therefore, concluded that the hydraulic fracture in the key trench (Seed et al, 1976, Sherard, 1987) and its relevance to the failure of the dam is fundamentally flawed (See also Muhunthan and Schofield, 2000). Except for the shallow depths of 30-35ft (9-10.7m) in some locations, the q/p stress ratio is significantly lower than 3 (fracture level), which indicates that fracturing of the soil would be difficult with increasing depth (Figure 8). For hydraulic fracture to occur, the soil element must be subjected to seepage water, which can cause (a) physical wetting of the soil first and then (b) a corresponding hydraulic pressure in the soil. The physical wetting and saturation of the soil increases the liquidity index of the in-situ soil and consequently the soil element becomes more ductile and the material tighter and less permeable (Figure 5) (also the q/p ratio drops off quickly, Figure 11 (a)). That is the stress-path moves significantly to the right to a more ductile and stable yield (Cam-clay) regime.

Some researchers (Leonards and Davidson, 1984) characterised the phenomenon of yield as ‘collapse on wetting’, which is a misnomer considering that the stress path simply migrated from the stable Hvorslev regime to the stable ductile Cam-clay regime. On the second point, (b), the hydraulic pressure due to the water seepage would have a limited opposite effect of reducing the effective stress of the soil element. Any such reduction in effective stress due to the seepage pressure will be more than offset by changes in the mechanical properties (ductility) of the soil. The net effect is the movement of the stress-path of the soil element is to the right and towards the Cam-clay regime (Figure 11). Therefore, the notion of ‘hydraulic fracture’ by water pressures equal to or less than the reservoir head, which could initiate a failure of the dam, has no scientific basis. In fact, to cause hydraulic fracture in the soil at the base of the dam (Cam-clay state), one needs to apply a hydraulic head of about 800ft (243.8m) of water!

It is also concluded that the ‘wet seam’ theory postulated during post-failure investigations (Leonards, 1987; Hilf, 1987) is flawed. The majority of the core material on Zone-1 was placed at a negative liquidity index (0.25 – 0.50) or in the Hvorslev regime in the stress-space (Figures 5 and 11). When seasonal rains and snow condition interrupted the material placement during construction, some layers might have been placed at wetter than the average or near liquidity index of unity. When subjected to large stresses, such pockets of material would fall into the Cam-clay ductile regime and deform like potter’s clay, ‘wet-seams’ or wet-pockets producing positive pore water pressure. This was the case for a few random pockets/layers of fill that were affected by the rain/snow when full stripping and replacement of such layers were not possible during the construction. Although such layers were of low strength and stiffness, they provide more impermeable mass relative to the surrounding material and would have had no adverse effect on the performance of the dam.

The original design specifications of Teton dam stipulated placement water content of optimum minus 1% to optimum for the core, which had only a small plastic index (PI<4). Based on the analysis here, it is believed that this was the fundamental error in the design concept in leading to the demise of the dam. The placement water content represented an initial liquidity index of zero or negative, which allowed considerable depth of the core to be prone to fracture (Figure 11). Without compromising the compacted density, for this material an additional 1-2% water content would have provided adequate equivalent liquidity index of at least 0.5 or more for most of the placed fill. This would have kept the entire fill intact in the Hvorslev regime where the material would have been stiffer, stronger and more watertight except for the top 5-10ft (1.5-3m) (freeboard regime). Therefore, it is evident that the lack of knowledge at that time of the combined effect of liquidity and confining stress in controlling the mechanical behaviour of Zone 1 contributed in a major way to the Teton dam failure. For the design of earth-structures, the theory based on the ‘state based soil mechanics’ provides a better understanding of the physical and mechanical behaviour of a broad spectrum of soils including that of Teton dam, which are subjected to different loading conditions.

Conclusion

The ‘sunny day’ failure of Teton dam was well captured in many films available on numerous websites. A new theory based on the concepts of fundamental soil mechanics explains conclusively the manifestations and the mode of failure of Teton dam as it occurred on 5 June 1976. The theory also identified the problems with using materials with low plasticity for the impervious core and the role of liquidity index. Unless these aspects are recognised in the design and liquidity index is normalised to fit the geometry of the core, potential for cracks in earth dams exist. Based on the investigation and discussion documented here, the following can be concluded.

An internal transverse crack(s) or large opening(s) had developed in the core (Zone-1) to a maximum depth of 32ft (9.75m) below the crest (top of the core) at the right abutment near Sta. 14+00. The analysis further indicates that much shallower cracks existed in the core in both abutments under the steep rock slopes. When the reservoir level rose to the level of the deepest crack, water flowed freely barrelling downstream into the chimney drain (Zone 2).

The internal cracks might not have day lighted through the 3ft (0.9m) thick granular roadbed, which was subjected to constant vehicular traffic and compaction. Also, the parameters that affected the core were different from those of the overlying roadbed granular fill. The uniform clayey silt (CL-ML) that was used for the core of Teton dam fitted well into the CSSM model that was developed for other soils with different plasticity. Although the clayey silt had relatively high values for the liquid limit (LL~23) and plastic limit (PL~19), the plastic index was relatively small (PI~4 or less). Consequently the liquidity index was very sensitive to the initial placement water content and its subsequent variability in mechanical properties due to varying confining stress. The initial liquidity index and its variation played a key role in the cracking of the dam. Therefore, for clay-silt cores, it is more prudent to have the construction specification refer to the liquidity index or the ‘placement water content’ with respect to the plastic limit (PL), than of the optimum water content.

A combination of material parameters such as the low plasticity of the core, the sensitivity of the liquidity index of the material to water content, its variation under the subsequent confining stress condition, and their influence on the constrained modulus played a key role in the cracking of the core. It appears that these aspects of fundamental soil mechanics and the phenomenon of cracking were not recognised in the original design of the dam.


Author Info:

N Sasiharan, Graduate Student, and B Muhunthan, Professor, Department of Civil and Environmental Engineering, Washington State University, Pullman, WA, US. V S Pillai, Geotechnical Consultant, 2839 W.Kennewick Ave., #204, Kennewick, WA, US 99336

This study was sponsored by the Geomechanics and Geotechnical systems of the National Science Foundation under the grant CMS-0234103. The authors also wish to thank Chris Ketchum of USBR for supplying samples and allowing them to collect additional material from the remnants of the failed dam.

They also thank Kathy Cox of the Department of Civil Engineering at Washington State University for her assistance with the preparation of the paper.

The opinions expressed in this paper are those of the authors and do not necessarily represent those of any other individual or organisation.

Tables

Table 1