Managing the risk of owning and operating dams requires the best available knowledge on both the potential failure modes and associated consequences. Furthermore, CEATI believes that most dam failures are not the result of extreme loading conditions. Dams typically fail due to unforeseen or unrecognised conditions that result in piping failures, overtopping, and errors in operation or foundation failures.
In order to minimise the risk to those who live downstream of dams or who otherwise benefit from them, it is vital that the combined knowledge of the dam safety community is shared. Case studies of dam failures and incidents must be exchanged. The lessons of the past must not be forgotten.
Learning from incidents and failures
According to CEATI, sharing of incidents is one of the most effective tools in driving down the risks in dam safety. And this appears to be the motivating force behind the Dam Safety Interest Group’s workshop which was held in Los Angeles, California, US in March 2009. Called Learning from International Dam Incidents and Failures, the event was planned in co-ordination with the US Federal Energy Regulatory Commission and the Division of Dam Safety and Inspections.
The goal of the workshop was to:
• Educate and provide knowledge necessary to manage the risks associated with dams.
• Identify the areas where additional research may improve knowledge of potential failure modes and identify ways to prevent or mitigate their development.
• Identify ways to ensure that the knowledge of the past is available for the future.
• Develop a plan to ensure the continuation of global communication, co-ordination and collaboration.
To further its goals of sharing important dam safety case studies, CEATI has kindly allowed IWP&DC to give an insight into the information exchange which took place at the workshop in 2009. A selection of case studies below illustrates the most common causes of dam failures – ranging from liquefaction, piping and seismic deformation, plus operation controls and human interaction.
Arapuni dam, New Zealand
Arapuni dam is a 64m high curved concrete gravity structure on the Waikato river in the North Island of New Zealand. It was completed in 1927 to form a reservoir for the 186MW hydroelectric power station.
A series of foundation leakage events have occurred since water was first impounded in 1927. Several remedial techniques were tried over the years, including a grout curtain in 1930 and bitumen grouting from 1935 to 1942. Overall there was no long-term effect on seepage flows.
In 1995 eight observation well holes were drilled into the foundation to provide the opportunity for uplift measurement beneath the dam. Two of these eight holes intersected high water pressures and each flowed at several hundred litres per minute after drilling. The high-pressure flow was encountered at a depth that coincided with one of the sub-vertical fractures in the dam foundation recorded during dam construction. Dirty water observed in one of the holes indicated that some foundation material may have eroded.
From 1995 to 1999 pressures and flows increased slightly. One of the two high pressure holes was used as a relief drain throughout this period, while the other monitored pressure in the fracture. During 1999-2000 there was a steep increase in pressure and flow. In September 2000, solid material (clay and bitumen particles up to 20mm in diameter) was observed to be exiting the relief drain. If erosion migrated along the line of the fracture and downstream of the two drill holes, it could have had major consequences for the integrity of the dam.
While an immediate action involving grouting the two high pressure holes could have been carried out in September 2000, this was not considered the best solution. Clearly an eroded path through the dam foundation had developed, but little was known about the seepage path, including the size of the void, nature of eroding material, the source of seepage water or the path of the leak.
Grouting at this stage would have given an unknown result as to the effectiveness of the grouting. Excessive grouting pressures were also considered to have the potential to initiate blowout of the remaining fracture infill between the holes and the downstream face.
Therefore it was considered more prudent to investigate the leak and treat it when better understood. The investigations would be undertaken with the full reservoir in place but with the option to grout the two drill holes as a contingency should the dam’s condition become unacceptable and action be necessary to stem the leakage flow.
In October 2001 it was decided that sufficient investigation information had been gathered to identify the leakage path and successfully fill the void. This well engineered operation was programmed for early December 2001.
It is now evident that all the previous incidents dating back to 1927 were related to piping within, and erosion of, the weak clay infilling the defects within the volcanic ignimbrite rock foundation. The various grouting works only filled voids where the vertical drillholes connected to open voids in vertical joints, leaving other leakage paths open. The most likely cause of seepage reduction in the intervening time is considered to be migration of fracture infill material gradually sealing seepage exit points.
While the 2001 grouting had successfully filled the void identified in the fracture infill, there were significant quantities of erodable clay infill remaining in the foundation. Furthermore, open voids were identified by subsequent investigations in other fractures in the dam foundation. Based on previous history, future incidents with similar potential consequences were considered likely.
A targeted and cost effective cut-off wall project to reduce the risk of further piping incidents was implemented by the owner, from 2005-7. Operation of the reservoir was not affected and electricity generation also continued during the cut-off works.
Misinterpretation
Gradually declining seepage from the 1950s was misinterpreted at Arapuni dam up until 2000. It was not until the incident occurred that detailed research of construction and early incident records was carried out. The seepage history was not taken into consideration when the 1995 drill holes were planned. No filtering of the drains was applied and considerable joint infill material would have eroded over the following five years. The frequency of monitoring readings prior to 2001 had been insufficient to identify sudden seepage changes that indicated the erosion process of fines migration with blocking and unblocking along the seepage paths
Lower San Fernanado dam, US
Lower San Fernando dam was part of the Lower Van Norman complex and Los Angeles aqueduct system in the US. The dam was an earth embankment approximately 43m high at its maximum section. It was constructed by the hydraulic fill method in 1912.
The foundation in the channel section and lower portions of the abutment rests on alluvium consisting of stiff clay with lenses of sand and gravel. The alluvium has a maximum thickness of about 11m. Underlying this and forming the upper parts of the abutments are shales, siltstones and sandstones.
On 9 February 1971 an earthquake of magnitude 6.6 on the Richter scale caused a major slide on the upstream slope of the dam. The dam did not fail or release the contents of the reservoir, which was immediately drawn down. The downstream population of 80,000 people was evacuated to a safe elevation over a period of four days.
The possible occurrence of the slide 25 seconds after shaking had stopped led to multiple interpretations of the possible cause of failure. Field investigations found a highly disturbed zone of hydraulic fill in the lower 6m of the dam that liquefied and sheared. The dam above this zone remained largely intact moving in large undisturbed blocks.
One explanation has suggested the increase in pore pressure in the hydraulic fill corresponded to a loss of strength in those soils. As a result, stronger zones accumulated the stresses the weaker zones were carrying, eventually exceeding their strengths at which point failure occurs.
The timing has been suggested to relate to the flow of water that occurs due to the excess pore pressure and the associated hydraulic gradient. It was hypothesised that the flow of water caused a progressive loosening and loss of strength in the initially stronger but dilative starter dikes leading to the failure.
Other explanations of the delayed failure mechanism have suggested the possibility of a progressive failure where a lower block fails leading to an increase in driving stresses on the remaining in-tact slope and a subsequent failure of the next up-slope block.
The experience at Lower San Fernando is a critical case history on dams and liquefaction, and forms the basis of understanding residual strengths of soils subjected to liquefaction. Loose material within foundations or embankments will not perform well during earthquakes. Dams with potentially liquefiable material and subject to earthquake shaking need careful evaluation.
The dam and reservoir were abandoned. Other facilities were constructed in the complex to replace the storage facility.
Belci dam, Romania
Limited hydrological data prior to construction had a detrimental impact on the Belci dam on the Tazlaur river, near Slobozia in Romania. This 18.5m high, 432m long earthen structure was built in 1962. However hydrological measurements for the estimation and pre-calculation of design floods had only been collected from a gauging station 10km upstream from the dam site for ten years prior to construction.
During its 29 years of operation, floods occurred on Tazlau river with peak values much more than those estimated at Belci dam. On 7 July 1970 a peak inflow of 980m3/sec caused overtopping of the dam and part of the left wing was eroded. After further floods on 29 May 1971 and in August 1979 with peak values of 890m3/sec and 855m3/sec respectively, a new calculation for the design floods was commissioned. However the spillway capacity was never changed because at the same time the risk classification of the dam was reduced.
On 28 July 1991 heavy rainfall occurred. Failure of the main telephone lines meant that no prediction of flood warnings could be sent from the upstream catchment areas to the dam site. Power failure at the site also meant that the bottom outlet could only be opened 40cm. Emergency power was also unavailable and the gates could not be opened manually as timber and debris were blocking the bottom outlet.
At 02.15am on 29 July the dam began to overtop and the reservoir was emptied by 07.15am. Twenty five people were killed by the flood wave caused by the dam break and 119 houses were destroyed.
The peak inflow of 1200m3/sec was lower than the later estimated 1-in-100-year flood of 1515m3/sec. The initial dam break occurred at the same point where the dam had been affected by erosion in 1970. Post-failure measurements of the intact dam crest near this initial break showed that the repair works had produced the effect of a natural overflow section. A cable trench that had been dug along the dam crest also had a negative and accelerating influence with regards to the process and dynamics of the breach formation.
The dam has not been reconstructed.
Zipingpu dam, China
On 12 May 2008 the epicentre of an earthquake, measuring 8.0 on the Richter scale, was located only 17km away from the Zipingpu dam. Located on the Minjiang river, in China’s Sichuan Province, the 156m high CFRD survived the strong earthquake but suffered considerable damage, especially on the upper part of the structure.
During the earthquake acceleration at the dam crest reached 2.0g. As the dam design is based on the assumed peak ground acceleration (PGA) of 0.26g, the actual PGA is far beyond the intensity accepted by the design.
During the earthquake significant deformation occurred at the dam. Immediately afterwards there was obvious settlement of the dam crest in the riverbed section. The maximum settlement value was 683.9mm, located at the river centre section. After the main earthquake, the observed settlement of dam crest still developed but the rate dropped rapidly and stabilised on 22 May.
The strong earthquake had damaged the concrete face slab, which included cracks and ruptures. Spalling of the concrete was also noticed. As the maximum dynamic response of the dam occurred at the dam crest, the earthquake caused many cracks and ruptures of the parapet wall.
Immediately after the earthquake, a dam safety expert group was dispatched by the Ministry of Water Resources to check the dam status. The damaged spillway tunnels were repaired and the reservoir level controlled and lowered. The repairs to the ruptured concrete face slab and separation of the face slab and rockfill began on 30 May 2008.
Stable and safe
Although the Sichuan earthquake had produced severe damage to the dam, it is still structurally stable and safe. The performance of Zipingpu has demonstrated the safety features of CFRD during such an event. As there is no serious slide of the downstream slope, it shows that the two-stage slope design (1:1.5 for the upper part slope and 1:1.4 for the lower part slope) for the downstream slope is appropriate.
Although the waterstop system was not seriously damaged and there was no significant increase in dam leakage at Zipingpu, it should be noticed that this is one of the vulnerable parts of a CFRD upon extreme loading.
Steinaker dam, US
Located in Utah, US the 49m high Steinaker dam was completed in 1961. In November 1962, a sinkhole area, approximately 3m in diameter and 2m deep, appeared on the downstream face of the dam, 47m downstream from the crest centreline.
The engineers who examined the sinkhole concluded that saturation collapse had occurred. The supposedly benign sinkhole was filled. Additional depressions were reported in this area in June and October of 1963. The sinkhole was again backfilled. No further subsidence was reported at this location.
In August 1965 a second sinkhole was reported, approximately 6m in diameter and 1.5m deep, 45m downstream from the centreline. An engineer examining the site judged that the dam was in no danger of failing, but recommended a remedial grouting programme which was carried out on the left abutment between December 1965 and April 1966.
In 1992, US Bureau of Reclamation decided to modify Steinaker dam to rectify a seismic stability issue. It was also recommended that investigations were carried out to assure that the core’s integrity had not been compromised by the internal erosion. Three lines of three boreholes were drilled in the vicinity of the sinkholes, and cross-hole tomography was performed from the boreholes in an attempt to locate and determine the size of any voids that might remain. The tomography results were inconclusive, but borehole completions prior to testing did provide some indication that voids might still be present.
The sinkholes are believed to have formed by the following mechanism:
• Seepage travelled through left abutment bedrock fractures and flowed through pervious alluvial foundation materials, through Zone 3 shell material, and through the gravel envelope surrounding the toe drain.
• Voids in the gravel envelope and between cobbles in the Zone 3 were large enough to provide an unfiltered exit for the fine-grained foundation alluvial material, and were extensive enough to store some of the eroded material.
• A large tear in the toe drain also provided an unfiltered exit point and a means to remove foundation alluvial material.
• The constant seepage flow carried the alluvial material with it, eventually forming a large void beneath the Zone 3 material.
The sinkhole mechanism by itself was unlikely to have resulted in dam failure. However, if the abutment had not been grouted, it is possible some of the joints or fractures could have directed seepage water along the embankment/bedrock contact, causing erosion at contact points, possibly bringing full reservoir head closer to the unfiltered exit points, increasing gradients and initiating a pipe completely along the embankment/foundation contact. The pipe would then transmit higher-velocity seepage, rapidly increase erosion, and eventually initiate a dam breach.
Engineers on the project warn that there should be a greater awareness of potential stiffness changes beneath a planned toe drain, so that differential settlement from embankment loading does not tear openings in the drain pipe. At the project the original toe drain was replaced with a perforated PVC pipe, embedded in a sand and gravel mixture having a gradation designed to filter the fine-grained foundation materials. The filter material was also placed as a chimney drain on the upstream face of the excavation slope.
Saluda dam, US
Seismic remediation using a back-up dam was a milestone project carried out at the Saluda dam in the US. It was performed to prevent reservoir releases due to seismic loading, and subsequent potential liquefaction of the original dam.
Originally constructed from 1927-30, Saluda Dam is owned and operated by South Carolina Electric & Gas Company. It is a 64m high, 2388m long semi-hydraulic earthfill structure made from locally obtained silty clay and sandy silt. This construction method resulted in a relatively low density embankment with extensive zones of internal saturation.
In the 1980s, a series of geotechnical investigations were undertaken to assess the safety of the original dam, particularly under seismic loading. Analysis indicated that the dam could be susceptible to liquefaction failure during large earthquakes. A major portion of the embankment could fail if an earthquake similar to the 1886 Charleston event were to occur again.
It was decided that the dam required remediation of some kind to resist such an event. The primary objectives of this were to prevent catastrophic flooding downstream of the dam, and to ensure safe shutdown of the facilities, including lowering the water level of Lake Murray in a controlled manner.
After considerable evaluation of alternatives, the remediation concept was developed as a backup dam immediately downstream of the existing earth embankment. The original dam remains in place and functions as the primary water impounding structure for Lake Murray. The backup dam will become a water retention structure if the original dam ever fails. It is also the first time a dam of this size has been so extensively modified under a virtually full reservoir head. This minimised economic and environmental impacts to the local area.
Approximately five million cubic yards of rock fill and aggregate were required to construct the backup dam, which consists of a combination RCC gravity section and two rockfill embankments. These required an on-site rock quarry, with a large crushing plant, stockpiles, and concrete mixing facilities. In order to ensure the safety of the original dam during construction, an extensive dewatering programme was required.
The backup dam itself had two seismic instruments installed, one in the foundation and another on the crest of the RCC section. New vibrating wire piezometers were installed in the foundation of the RCC section to monitor uplift pressures should the dam ever be called upon to retain Lake Murray. A network of reference benchmarks were installed on site, and survey monuments were installed on the crest of the backup dam to allow precise deformation and settlement measurements. The back-up dam was completed in 2005.
Matahina dam, New Zealand
Severe internal erosion damage with crest subsidence on the left abutment followed an earthquake at New Zealand’s Matahina dam on 2 March 1987. The epicentre of the earthquake was only 23km from the dam, and emergency drawdown of the reservoir prevented failure.
The 85m high dam is located on the Rangitaiki river in the North Island. This earthquake-induced internal erosion damage needs to be understood in context of an earlier incident at the dam.
During the final stages of lake filling at the construction site in January 1967, a large temporary increase in flow was observed from the drainage blanket. Approximately two weeks later an erosion cavity appeared on the crest downstream of the core in the right abutment area. It was found that the core had cracked adjacent to a 1.8m wide bench in the abutment at a depth of 12m below the crest. It was concluded that the erosion had occurred where a large boulder had been placed against the abutment creating an unprotected exit. The area was repaired and performed satisfactorily until 1987.
After the earthquake the initial damage appeared minor. There was transverse cracking at each abutment, minor spreading of the crest, moderate deformation of the rockfill and a small initial increase in flow from the earth dam drainage blanket.
Although observations indicated the damage was superficial, in light of the 1967 incident, a comprehensive monitoring and investigation programme was initiated. It was concluded that core cracking and internal erosion had occurred but that it was not known if this was caused by the earthquake or predated it. On 17 December 1987, while investigations were in progress, an erosion cavity appeared at the surface above the area of damaged inner transition. This was confirmation that internal erosion had occurred and was continuing.
Over the next few days, the lake was lowered in a number of stages as fluctuating piezometer measurements indicated continuing erosion. It was concluded that cracking of the core and inner transition had occurred prior to the earthquake and probably during lake filling in 1967. (A few hours after the 1967 leakage flows had peaked, a second but smaller peak occurred. A possible explanation for the second peak is that it was due to cracking and leakage on the left abutment).
The dam was repaired in 1988 but did not address the prevention of internal erosion in the dam away from the dam abutments. This was rectified during major re-building of the dam in 1996-98.
Lessons learned from the experiences at Matahina dam are summarised here:
• The availability of design, construction and performance information, together with personnel familiar with it was essential in the earthquake emergency and the investigations that followed.
• The six-yearly dam safety examinations provided the means for a new generation of engineers to become familiar with the dam. The dam safety report completed before the earthquake correctly identified the main features of the dam’s subsequent post-earthquake behaviour.
• As the dam was operated below its normal generation level after the earthquake the monitoring information from the 1967 lake filling was the only precedent information available and demonstrated the importance of these records.
• Internal erosion can be a rapid process or a slow process over many years.
• At intensities above MM7 many people are affected by shock, including dam operational staff. Shock may prevent staff from carrying out planned post earthquake procedures. Relief staff are required on site to continue dam inspections and operations.
• Duplicate power sources for the spillway gates are essential and have become commonplace. At Matahina external power was unavailable and the station generators had tripped off. It was a few hours before river flows could be passed again.
Swinging Bridge dam, US
The near failure of Swinging Bridge dam in New York state, US was the result of a condition that had been recognised and understood. However, the criticality of the condition was not recognised until a sinkhole had formed at the dam. The damage that finally manifested itself was the result of 75 years of accumulated distress.
There are number of other ageing dams in the US, many with a history of problems. But when monitoring problematic dams, when is enough enough? How does the team of owner, regulator, engineer, and contractor determine when major and costly rehabilitations are needed before failure occurs? And how do these decisions get made without undertaking work that may not be needed?
A failure of Swinging Bridge dam would have major consequences. A flood wave would incrementally increase the river level by 7.6m before reaching the community of Port Jervis, 30km downstream. The flood wave would travel over 241km downstream before its effect was attenuated, resulting in cascading dam failures with major consequences in the Delaware Valley.
Constructed from 1927-30 on the Mongaup river, Swinging Bridge dam is an earth embankment, 304m long with a maximum height of 41m. There are two powerhouses with separate water conveyance structures. An approximately 213m long, low-level conduit was constructed through the base of the embankment as part of the original construction and is founded on glaciated alluvium.
A sinkhole was observed on 5 May 2005 along the upstream edge of the crest of the dam, directly over the low-level conduit servicing unit 1. Although reservoir levels had been within normal ranges in the weeks preceding the sinkhole formation, the project had experienced back-to-back flooding events in late March and early April, with the record flood event at the site occurring on 3 April 2005.
Cracking of the crest access road extended approximately 27m east and west of the sinkhole, indicating that the effect of the disturbance was distributed over a wide area.
Operations staff took immediate and appropriate action, including:
• Implementation of the emergency action plan, advising emergency management agencies of a potentially hazardous situation.
• Opening the spillway gates and operating unit 2 at full capacity to begin immediate drawdown of the reservoir.
• Shutting down unit 1 while leaving the penstock shutoff valve open to prevent piping of soil into the penstock should a significant crack in the penstock exist.
Following these actions, it was determined that the dam, while significantly distressed, was stable and the reservoir drawdown was continued as quickly as possible. Remedial measures occurred in mid-2007. The controlled refilling of the reservoir was concluded in late 2007.
Upon review it became apparent that the Swinging Bridge project had a troubled past. No provisions for settlement of the conduit were included in the original design. Settlement and cracking of the conduit was observed during initial filling of the reservoir, along with silt accumulation, although there was little reference to the volume of silt that accumulated. Historical records are sparse between 1933 and the early 1980s, though piping into the conduit was reported periodically in the years between 1933 and 2005. Although it had slowed, settlement of the conduit was still occurring.
The record flood event occurred on 3 April 3 2005 approximately one month before the sinkhole formation. The peak flood stage was 0.8m above normal full and should have resulted in failure of the spillway flashboards. High tailwater in the flat spillway discharge channel prevented the flashboard failure. Although no definite relation between the flood and the subsequent sinkhole has been established, it seems an odd coincidence that the two events were unrelated.
The high reservoir level may have reached a seepage path that previously had not been exposed. It may have saturated crest soils causing the soil bridging voids to collapse; or the sustained high tailwater level resulting from continuous turbine operation may have caused incremental settlement sufficient to open the penstock joints. As piping into the tunnel started, the relationship between penstock leakage, foundation piping, and settlement may have resulted in a spiralling, accelerating failure mechanism.
Three piezometers had been identified as reading higher than normal prior to the sinkhole formation. These piezometers, which had a history of unreliability, were to be monitored in more detail. However attention was diverted away as the flooding had caused significant damage that required action elsewhere.
A number of important lessons can be learned from the Swinging Bridge project with respect to design, construction, monitoring and rehabilitation:
• Avoid having conduits pass through embankment dams. If this cannot be avoided, found the conduit on or within rock so that settlement can be effectively eliminated.
• Install instrumentation that is appropriate for the potential failure mode being monitored. Instrumentation must be trusted and behaviour that is out of character must be investigated immediately.
• Operations and maintenance personnel are the primary line of defence with respect to dam safety. Functional exercises would be very valuable in training personnel how to identify problems and how to react.
• Potential failure mode analyses are an excellent tool to identify problematic behaviour, and it is an appropriate way to identify and prioritise remediation activities, while considering risk and consequences.
This article was produced by IWP&DC in collaboration with Chris Hayes, Director, Generation and Utilisation at CEATI. The case studies used were reproduced with the kind permission of CEATI International. The papers were presented at the workshop Case Studies: Learning from International Dam Incidents and Failures held in Los Angeles, California, US on 24-25 March 2009.
Readers can obtain the full proceedings CD from http://www.ceati.com/pricequote?publicationid=6057 or publications@ceati.com.
| CEATI International |
|
Based in North America, CEATI International is a user driven multi-national development and exchange programme for utilities as well as select government agencies. Its Dam Safety Interest Group is part of a larger CEATI hydro programme which brings together over 60 of the world”™s leading hydroelectric utilities. Goals and objectives include technical networking, best practices exchange as well as research and development project cost sharing opportunities on a practical scale such that deliverables have an immediate impact for participants. Hydro generation capacities of participating companies range from under 50MW to over 35,000MW. |
| CEATI Workshop 2011 |
|
Case Studies: Reliability of Discharge Facilities – Learning from International Incidents, Failures and Best Practices. |
