There are many old dams in Sweden and, since few dams are constructed today, the main task in risk assessment on the existing dams is often to reduce the risk of failure. RIDAS (Swedish Guidelines on Dam Safety) is used when designing new dams and assess the existing ones. The guidelines include stability requirements for different failure modes, e.g. overturning and sliding, which imply that the load effect from e.g. uplift and ice load need to be less than the resisting loads or moments. The ice load in RIDAS is given as a deter­ministic value depending on where in Sweden the dam is located. For many years, ice and ice load have been researched, but there is still a lack of knowledge regarding the magni­tude of the ice load and how it affects the probability of failure for dams. More knowledge about the actual ice load would result in a lower calculated probability of failure for the dam which could be used to design slender dams or avoid unnecessary reinforcement of existing dams. Dam safety evaluation is often performed with deterministic methods based on safety factors. In recent years, the use of probabilistic methods in dam design has increased. The method has an advantage compared to deterministic methods in safety evaluations of existing dams, since probabilistic methods provide an answer to which parameters that have the greatest impact on the stability of the dam and take into account the variations in each parameter. I this master thesis, a statistical distribution for the variation of the ice load’s annual maxi­mum value was calculated. This was used in the analysis of the probability of failure for solid gravity concrete dams and buttress concrete dams. The probability of failure was cal­culated for dams of different sizes for overturning and sliding failure modes, and also for three different load cases; without ice load, with a truncated ice load distribution and with an ice load distribution that has not been truncated. The probabilistic stability analysis was conducted in Comrel with ice load as one of the stochastic variables. It was found which sizes of the dams that have the largest impact from the ice load; also what effect extreme values on the ice load has on the failure probability of the dam. The results indicated that the probability of failure for dams lower than 15 m is more affected by the ice load, for both failure modes analyzed. The probability of failure is reduced for all dam types when eliminating extreme values of the ice load, particularly for dams lower than 15 m. In several cases, truncation of the ice load distribution is the differ­ence between an accepted and a non-accepted level of the probability of failure. It is also shown that reduced coefficient of variance for the ice load results in a decreased probability of failure. The conclusion is that solid gravity dams and buttress dams lower than 15 m, with a high consequence class, should be risk assessed with the ice load as a stochastic vari­able. The statistical distribution of the ice load is still uncertain and the distribution used in this report should not be used globally, rather in areas with a climate similar to northern Sweden. The reason for this is that the measurements that were used to derive the global distribution were mainly performed in areas with conditions similar to those in northern Sweden. A recommendation for further research is to focus on determining statistical distributions for the ice load for southern, central and northern Sweden. An alternative is to use differ­ent ice load distributions for the different areas. Another alternative could be to use the same statistical distribution for southern, central and northern Sweden but with different values for where the ice load distribution is truncated, depending on the maximal ice thick­ness in each area. The recommendation is also to develop a reliable method for measuring the ice load. In addition, attempts should be made to determine whether extreme values on the ice load really exist or if they are effects of measurement errors. Key words: concrete dams, ice load, probabilistic stability analysis, probability of failure

Vid en ombyggnation av Henriksdals reningsverk har ett vertikalschakt utformats och anlagts. Under berguttaget tillämpades extensometermätningar för att följa upp deformationer i en bergpelare mot en befintlig tunnel. Mätningar utfördes automatiserat och regelbundet. För uppföljning av deformationer tillämpades tre larmnivåer avgränsade av två larmgränser vid 1.0 mm samt 3.0 mm deformation och var förbundna till olika motåtgärder. Målet med detta arbete harvarit att studera hur tillämpade larmgränser har presterat relativt dokumenterade och prognostiserade deformationer ur ett sannolikhetsbaserat perspektiv. Samt vid behov formulera och föreslå nya larmgränser. En litteraturstudie om modellering av bergmassor samt mekaniska egenskaper hos förekommande bergmassa av gnejs genomfördes. Därefter modellerades objektet för vidare numerisk analys genom FEM i Plaxis 3D. Med grund i uppställd modell kunde in-situ spänningsfältet skattas och modellen optimeras. En känslighetsanalys för bergmassans hållfasthetsparametrar utfördes och de fyra mest känsliga parametrarna: γ, ν, σci, Qbas, applicerades statistiskt i efterföljande studier. Föreslagna och tillämpade larmgränser studerades genom utförande av Monte Carlo-simuleringar. Använda deformationsbaserade larmgränser återfanns inte vara tillfredsställande och nya individuella töjningsbaserade gränsvärden framställdes. De ursprungliga deformationsbaserade larmgränserna resulterade för längre mätankare i utlösta larm och visades vara ouppnåeliga för kortare mätankare. En resulterande sannolikhet för överskridande av föreslagna varningsgränser vid dragtöjning skattades till storleksordningen 10−2 till 10−1 samt för stoppgränserna vid dragtöjning till 10−4. Sannolikheterna för överskridande av föreslagna larmgränser vid trycktöjning var inte möjliga att uppskatta för det studerade fallet. / During a rebuild of Henriksdal wastewater treatment plant, a vertical rock shaft was constructed. During the excavation, extensometers were used for monitoring deformations in a rock pillar against an existing tunnel. Measurements were automatically and regularly performed. Three alarm levels for follow up of the deformations were used and associated with different counter measures. The alarm levels were divided by two alarm limits at 1.0 mm and 3.0 mm deformation. The goal with this work was to study how the applied alarm limits from a probabilistic point of view have performed relative to obtained and forecasted deformations. And, if necessary, propose new alarm limits. A literature study about modeling of rock masses and mechanical properties of the occurring gneiss at site was performed. The object was thereafter modeled for further numerical analyzes using FEM in Plaxis 3D. Based on the performed model, the in-situ stress field could be estimated,and the model was further optimized. A sensitivity analysis of the rock mass strength parameters was conducted and the four most sensitive parameters: γ, ν, σci, Qbas, were applied as probabilistic in the following studies. Proposed and applied alarm limits were studied through Monte Carlo simulation. Applied deformation-based alarm limits were found not to be sufficient and two new strain-based were prescribed. The deformation-based limits were for the longer extensometer anchors resulting in triggered alarms and were only considered to be applicable for anchor lengths between 3.0 to 5.0 m. The probability of exceeding the proposed warning limits for tensile strain was estimated to be of a magnitude 10−2 to 10−1 and for the stop limits to be of a magnitude 10−4. The probabilities of exceeding the proposed alarm limits during compressive strain were not possible to estimate for the studied case.