What Caused the Accident at the Sayano–

Shushenskaya Hydroelectric Power Plant

(SSHPP): A Seismologist’s Point of View
by Victor S. Seleznev, Alexey V. Liseikin, Alexey A. Bryksin, and Pavel
V. Gromyko

INTRODUCTION Different versions of events were discussed by the mass

media (Klyukach, 2011; Lobanovsky, 2011; Rassokhin, 2011,
The Sayano–Shushenskaya hydroelectric power plant (SSHPP) 2012), but we present another one here.
is the greatest power plant in Russian history due to its highly The investigation committee concluded that the accident
rated capacity. This ranks it at number 7 in a list of hydroelec- was caused by “…destruction of generator’s (HG2) cover studs,
tric power plants worldwide (Fig. 1). caused by additional dynamic loads with variable features… the
It is an upstage of Enisey’s series of hydroelectric stations. fatigue damage of attachment points made possible the cover
The station’s unique arch-gravity dam is the highest in Russia breakage and flooding of the turbine hall” (Rostekhnadzor,
and one of the highest in the world. It has a maximum height 2009). Topics about resonant phenomena at large hydropower
of 242 m, its upstream face is delineated with an arc of 600 m constructions and the influence of the equipment on stability
in radius, with a width of 105.7 m in its basement and 25 m in of hydropower plants have been widely discussed (De Jaeger
its crown. The length of the dam’s crown with waterside inserts et al., 1994; Strah et al., 2005; Khodabakhshian and Hoosh-
is 1074.4 m. Ten axial galleries are located in the body of the mand, 2010). The objective of this paper is to try and explain
dam, which contain equipment to monitorthe dam’s condition to explain the underlying cause of the accident using data re-
and allow access for repair work. corded by seismological instruments. The Cheremushki station
Ten hydroelectric generators with centrifugal turbines are (CERR) is 4 km from the plant. Data were recorded before the
located in the SSHPP building, each has a power of 640 MWand accident and after the plant was restored; these data are exam-
works under a water head from 175 up to 220 m. Construction ined to identify the reason for the tragedy and the origin of
of the plant was started in 1963 and was completed in 2000. additional dynamic loads.
There were a number of problems concerning the destruction To begin with, we should understand the theoretical fea-
of spillways and forming of cracks in the body of the dam, which sibility of using seismological data to recover the process of
occurred during the beginning period of construction and op- tragedy. Forensic seismology now acts as one of the instru-
eration, but these were successfully resolved. ments that could help in the investigation of different acci-
The greatest anthropogenic accident happened within the dents. The case with the ship Costa Concordia gave us a good
plant at 8:13 a.m. local time (0:13 UTC) on 17 August 2008. example (Mucciarelli, 2012). Station MAON was situated
Hydroelectric generator number 2 (HG2) suddenly destroyed 18 km from the site of impact, and recordings from this station
itself during operation and was thrown from its position by helped establish a timeline of that particular accident. In our
water pressure. Water began to come into the turbine hall at case, an instrument recorded oscillations of different plant
a vast rate. At the moment of destruction, the total power mechanisms at the distance of about 100 wavelengths.
within the plant was about 4100 MW; there were nine hydro-
electric generators in operation. Unfortunately, most of the au-
tomatic self-protection systems were not actuated. The power METHODS AND DATA
supply to auxiliaries was lost, which necessitated the manual
release of an emergency repair diversion gate in order to stop In this study, we examine an oscillation recording produced
water flow. Seventy-five people died as a result of the catastro- by the working equipment. The main task in this case is to
phe. All hydroelectric generators in the plant were badly dam- separate and recognize signals produced by this equipment.
aged, the turbine hall building was partially destroyed, and By analyzing the changes in spectral composition of the
electrical and additional equipment was significantly broken. seismic recording when switching the equipment operating
The following day, the Russian government created an in- mode, we selected the oscillation frequencies that corresponded
vestigative committee. V. S. Seleznev has had the honor of par- to relevant equipment. Constructing plots of signal swings at
ticipating in this committee. selected frequencies provides information about changes in the

doi: 10.1785/0220130163 Seismological Research Letters Volume 85, Number 4 July/August 2014 817
▴ Figure 1. Location of the Sayano–Shushenskaya hydroelectric power plant (SSHPP).

vibratory status of the oscillation’s source—the power plant The study of observed data gave us an opportunity to fix
equipment. links between the monochromatic-signal amplitude changes
As source data, we used seismic recordings from the seis- and the generator working parameters.
mic station CERR and from temporary seismic stations located
in the body of the dam and in the turbine hall (Fig. 2). All
RESULTS AND DISCUSSION
seismic equipment used in our analysis was produced in Russia.
At the moment of the accident, the seismic instruments at
Many analysts thought a hydraulic shock was the reason the
CERR were equipped with the three-component SM-3 seismic studs broke. We examined this hypothesis using seismological
detector, which has a natural frequency of 0.5 Hz, and with a instrument recordings. During the accident, one of the hydro-
20-bit Baykal-11 digital recorder. Recording was performed electric generators increased to 12–14 m above its zero point,
consistently with a sample rate of 100 Hz within the 0.5– which was determined from marks on a pillar in the turbine
45 Hz frequency band. hall of the plant. It is possible to roughly estimate the energy of
Seismic stations located in the body of the dam and the this action using recordings from the Cheremushki station and
turbine hall of SSHPP included three-component piezoceramic from the seismological methods that are used to rate the energy
A1638 accelerometers with a low-frequency bandlimit of of a seismic event.
0.5 Hz and 24-bit Baykal-AS digital recorders. Stations were As was mentioned previously, Cheremushki is located
able to perform a persistent recording over —three to five days, about 4 km from SSHPP. P-wave arrival from the earthquake
with a sample rate of 500 Hz in the 0.5–200 Hz fre- that happened near the Japanese islands (M w 6.5, origin
quency band. time 00:05:51.4, after GFZ data, http://www.gfz‑potsdam.de/
Both new and old hydroelectric generators were turned on portal/gfz/home; last accessed May 2013) was registered at
when the plant was restored. Data received from the interim 00:13:06 GMT on 17 August 2009 by this instrument (Fig. 4).
seismological instruments that were placed near the generators The next part of the recording, beginning at 00:13:28
(Fig. 2) and from Cheremushki demonstrated that a working GMT on 17 August 2009 (the exact time of accident), shows
generator radiates a set of monochromatic frequencies that are more high-frequency oscillations than from the earthquake,
multiples of a rotation frequency and their amplitudes essen- but the amplitudes are not bigger than the amplitudes associ-
tially depend on the working conditions of a generator (Fig. 3). ated with the earthquake.
Some harmonics of oscillations are well recognized on the re- The maximum amplitude of oscillations (full swing) that
cordings from the station. were registered before the accident was about 1:83 μm=s

818 Seismological Research Letters Volume 85, Number 4 July/August 2014

▴ Figure 2. (a) Seismic observation scheme at seismic station Cheremushki (CERR; a background photo from Google Earth). Seismic
observation schemes (b) in the dam and (c) in the turbine hall. In the legend, 1 indicates seismic equipment placement and 2 indicates axis
directions of seismic receivers.

(vertical channel). Converting an amplitude to displacement Thus, if there was any dynamic action upon SSHPP, then
and considering the dominant period of the oscillations as an amplitude caused by it was unable to exceed the expected
1 s, we have 1:83=
2π ≈ 0:3 μm. value of 1:83 μm=s at 4 km distance. Otherwise, it should have

Seismological Research Letters Volume 85, Number 4 July/August 2014 819

▴ Figure 3. Example of relationships between SSHPP’s working regimes and registered seismic data. In the legend, 1 and 2 indicate
oscillation amplitude changes at a frequency of 38.1 Hz (vertical channel, dam body [μm=s2 ]) and CERR (1500 × μm=s2 ), respectively; 3
indicates the change in power produced by HG6 (MW).

that energy of dynamic action that could reach SSHPP at the

moment of the accident did not exceed the total energy pro-
duced by all generators in one minute. These facts indicate the
generator, with a weight of about 150 tons, was rising quite
slowly after its cover was torn.
Studying seismic recordings from instruments placed in
the body of the dam near HG2, we tried to define whether
powerful low-frequency oscillations were registered at the mo-
ment of accident.
The frequency band of these instruments spans the range
of 1–50 Hz, but powerful low-frequency oscillations should
be detectable in any case. We conducted an experiment with
seismometers (with an active horizontal component) mounted
on a moving platform that was moved over a distance of 0.2 m
and then back to its original location within 10–15 s in one
direction. Observed seismic recordings were filtered at a target
frequency of 0.05 Hz (Fig. 5a). Such slow but high-amplitude
▴ Figure 4. CERR vertical-channel data recording at the moment movement is detectable by seismometers with low working
of accident. frequencies of 1 and 10 Hz. Then the similar filtration pro-
cedure was done for instrument recordings from SSHPP
dam (Fig. 5b). Using this data, we concluded that HG2 was
been noticeable on the seismogram. Using seismological meth- rising for more than 10 s, which is the reason we cannot see
ods for rating the energy of earthquakes, we can say that seismic large amplitudes on recording from Cheremushki.
energy of such dynamic action cannot exceed the fourth energy Oscillations with frequencies of 2.381 and 38.095 Hz (the
class, or 10 kJ. For example, a seismic event with such energy working and paddle frequencies of SSHPP generators, respec-
can be initiated by an explosion of 0.5 kg of trinitrotoluene tively), were clearly detected in the wavefield registered by
equivalent. Cheremushki (Fig. 6). The electric motor of the pump house
When the energy level of such dynamic action was com- located near the power plant is the source of oscillations with
pared with dynamic actions caused by all the generators’ radi- 24.85 Hz frequency. Changes of oscillation amplitudes at the
ation at the working frequency of 2.38 Hz, it was determined working frequency are correlated with changes in generator

820 Seismological Research Letters Volume 85, Number 4 July/August 2014

▴ Figure 5. Data analyzed from instruments located in the dam of SSHPP. (a) Recordings of modeled slow movement. (b) Real data from
different locations within the SSHPP dam body.

loading: stepped increase or decrease of amplitude is observed Amplitudes of leaps associated with HG2 are 1.5 times
due to turning a generator on or off. Knowing the working greater in comparison with other generators. At the same time,
history of generators, it is possible to analyze these amplitudes there is no evidence of growth of these amplitudes directly be-
in a timeline and make conclusions about the vibration state of fore the accident, which indicates the stability of HG2.
a generator. An increase in oscillation amplitudes at low frequencies
The analysis of recordings from Cheremushki was con- (< 5 Hz, with the exception of the frequency associated with
ducted in three time intervals: 17 July 2009–17 August HG2) was observed five minutes before the accident (Fig. 6).
2009 (date of the accident), 1 January 2010–14 January However, the detailed studies of recordings preceding the acci-
2010 (just before the start of HG2 repairing works), and dent have shown the repeatable character of this feature.
14 March 2010–to 14 April 2010 (immediately after the We determined that these oscillations appear at specific
end of repair works). During all three time intervals, there were frequencies: seven correspond to the first seven forms of natural
leaps that could be associated with turning on and off HG2 oscillations of the SSHPP dam, and the eighth has a frequency of
and other generators (HG1, HG3, HG4, HG9, and HG10). 1.85 Hz, which does not correspond to natural oscillations.

Seismological Research Letters Volume 85, Number 4 July/August 2014 821

▴ Figure 6. Spectrogram of seismic recording from CERR prior to the accident. The color scale shows amplitude.

▴ Figure 7. Changes of oscillations amplitudes with frequencies of 1.55 and 1.85 Hz, registered at CERR.

Oscillation amplitudes observed during the five minutes ∼475–525 MW a half-month after the HG2 repair works
preceding the accident are larger than mean values but are were finished. There is not a direct link with HG2, because
not the greatest that were observed (Fig. 7). They could origi- low-frequency oscillations were also observed when HG2 was
nate under the specific working conditions of the generators stopped.
and could depend on fall level, such as the load of The analysis of turbine-cover oscillations has shown some
∼560–580 MW a month before the accident or the load of interesting details. A comparison of oscillation spectra of

822 Seismological Research Letters Volume 85, Number 4 July/August 2014

▴ Figure 8. Oscillations spectra registered on turbine covers under a generators’ load of 640 MW. HG9 is a new design, and HG3 is the
old design that was also found in HG2. X, Y, and Z are the axis directions of seismic receivers.

turbine covers, calculated for two different types of generators, new design with the same power as older ones (such as HG2)
is shown in Figure 8. Here, HG3 has the same constructional but with significant improvements in stability and safety. Both
features as HG2 (which caused the disaster), and HG9 has a generators have a load of 640 MW. The oscillation amplitudes

Seismological Research Letters Volume 85, Number 4 July/August 2014 823

of the turbine cover on HG3 are significantly higher than on REFERENCES
HG9. Moreover, spectra are essentially different: there are a
large number of monochromatic high-frequency signals (each De Jaeger, E., N. Janssens, B. Malfliet, and F. Van De Meulebroeke (1994).
Hydro turbine model for system dynamic studies, IEEE Trans.
of them is a multiple of the rotational frequency of the gen- Power Syst. 9, no. 4, 1709–1715.
erator) on the recordings from the older (HG3) generator, Khodabakhshian, A., and N. Golbon (2005). Robust load frequency con-
whereas we do not find them on recordings from the new troller design for hydro power systems, IEEE Conference on Control
generator(HG9). Thus, the turbine cover and its studs are con- Applications, Toronto, Canada, 28–31 August 2005.
tinuously influenced by high-frequency vibrations. This con- Khodabakhshian, A., and R. Hooshmand (2010). A new PID controller
design for automatic generation control of hydropower systems,
structional feature shared with HG2 (greater oscillation Electr. Power Energ. Syst. 32, 375–382.
amplitudes) is what most likely caused HG2 to be destroyed. Kishor, N., S. P. Singh, and A. S. Raghuvanshi (2006). Dynamic simu-
Inspection of the cover studs of other generators has also lations of hydro turbine and its state estimation based LQ control,
shown that some of them also were in danger of catastrophic Energ. Convers. Manag. 47, 3119–3137.
Klyukach, A. A. (2011). About the vibration at hydraulic generator no. 2
failure. SSHPP prior to the accident, Discussion, http://tayga.info/details/
Turbine-cover oscillations were not too large but contin- 2011/02/02/~102283 (last accessed May 2014) (in Russian).
ued for many years. Strong influences upon turbine-cover studs Lobanovsky, U. (2011). More polemic about accident with SSHPP, or
occurred when the screw mechanism was loosened, and the Myth about Sayan’ disaster-2, http://tayga.info/details/2011/01/
04/~101924 (last accessed May 2014) (in Russian).
smallest gaps caused the growth of dynamic action upon studs
Mucciarelli, M. (2012). The seismic wake of Costa Concordia, Seismol.
such that they began to break. Res. Lett. 83, no. 4, 636–638.
Thus we believe the true reason for the accident is based Rassokhin, G. (2011). Accident with SSHPP: Reality and myths, http://
on the fact that engineering calculations did not take into ac- www.plotina.net/sshges‑rassokhin‑3/ (last accessed May 2014) (in
count the interruptible work of the turbine cover under the Russian).
Rassokhin, G. (2012). Hydrodynamic factors of the accident with
influence of continuous high-frequency vibrations. All new SSHPP, http://www.plotina.net/sshges‑rassokhin‑5/ (last accessed
generators will be equipped with more turbine-cover studs May 2014) (in Russian).
to decrease amplitudes of monochromatic oscillations. Rostekhnadzor (Federal Environmental, Industrial and Nuclear Supervi-
The value of experimental and modeling works conducted over sion Service of Russia) (2009). Act of engineering investigation of
this problem should never be underestimated (Khodabakh- the accident that happened on 08/17/2009 in “RusHydro” branch
—“P.S. Neporozhny Sayano–Shushenskaya Hydroelectric Power
shian and Golbon, 2005; Kishor et al., 2006; Wang Plant”, available at http://www.rushydro.ru/file/main/global/press/
et al., 2009). news/8526.html/Akt_tehrassledovaniya__prichin_avarii_na_SShGES.pdf
Studies have shown that existing a hydroelectric genera- (last accessed May 2014) (in Russian).
tor’s control system doesnot take into account the complex Seleznev, V. S., A. V. Liseikin, and A. A. Bryksin (2012). A method of
continuous monitoring of physical state of buildings and/or con-
interaction of generators and surrounding buildings and con- structions and an appliance for its implementation, Patent for
structions. Vibrations of the working equipment itself generally Invention Number RU 2461847 C2, Bull. Number 26, http://
are in the focus of attention, but different frequency harmon- www.gts.energy‑journals.ru/content/2012/10‑2012/ (last accessed
ics, which are radiated by generators, could coincide with natu- June 2014) (in Russian).
Seleznev, V. S., A. V. Liseikin, and P. V. Gromyko (2012). Were there
ral frequencies of other parts of the equipment as well
raised oscillations of hydroelectric generator No 2 prior to the ac-
(Seleznev, Liseikin, and Gromyko, 2012). Hence continuous cident with SSHPP on 08/17/2009? Hydrotechnical Construct. 10,
vibrations at a stable resonance excitation could lead to emer- 48–50.
gency situations. Strah, B., O. Kuljaca, and Z. Vukic (2005). Speed and active power con-
trol of hydro turbine unit, IEEE Trans. Energ. Convers. 20, no. 2,
424–433.
Wang, F., X. Li, J. Ma, and M. Yang (2009). Experimental investigation of
CONCLUSIONS characteristic frequency in unsteady hydraulic behavior of a large
hydraulic turbine, J. Hydrodyn. 21, no. 1, 12–19.
The analysis of seismic recordings from Cheremushki and
from instruments placed in the body of the dam led to the Victor S. Seleznev
discovery that continuous vibrations at a stable resonance ex- Alexey V. Liseikin
citation (work frequency of the HG2 cover versus natural Alexey A. Bryksin
frequencies of buildings and constructions) most likely caused Pavel V. Gromyko
the accident with SSHPP. Geophysical Survey of Siberian Branch of Russian Academy of
It is necessary to watch over not only the working equip- Sciences
ment, but also the processes of its communications with build- Novosibirsk, Russian Federation
ings, constructions, and distinct parts of the equipment itself 3 Koptuyga Ave
when we deal with large-scale hydraulic structures (Seleznev, Novosibirsk, Russia, 630090
Liseikin, and Bryksin, 2012). [email protected]

824 Seismological Research Letters Volume 85, Number 4 July/August 2014