162
11 Dam Stability Failures
11.2.3 Does It Really Work?
Here is an example of “simplified” transformation of a deterministic slope stability
analysis into a “probabilistic one” using the simplified approach described in Silva
et al. (2008). The dam under consideration is well maintained and not damaged.
However, there was little control during construction and only partial monitoring
and observations. So, SLM method would propose a Class II–III structure, which is
definitely not a high quality dam.
With a FoS between 1.09 and 1.36 obtained by the engineers through their classic
slope stability analyses, we would have a p f = 10 −1 to 5 × 10 −2 or 5 × 10 −2 to 10 −3 .
It is now easy to include progressive release of attention, analysis of anomalies such as
deformations … and evaluate the resulting increased p f . It would then also be easy to
simulate progressive increase of attention, maintenance, monitoring, or hardening of
the structure through repairs and better monitoring, maintenance and see the effects.
We have applied this type of “symptom-based” approach, similar to the SLM
approach, to a variety of cases around the world, ranging from the evaluation of
unexploded ordnance risks in countries such as Lao PDR and Cambodia to the risk
assessment of mountainous roads networks. The success of the approach has been
measured by third parties through extensive reality checks.
For dams, we include thirty diagnostic points specific to tailings dams (see
Sect. 11.2.1). The range of probabilities stems out of a “optimistic” and a “pessimistic” evaluation of the thirty diagnostic points (a re-calibration of Table 11.4).
As many forensic studies have shown, the failure of a dam can rarely be attributed
to a single cause, but rather to a set of conditions (hence the thirty diagnostic points)
which, together, can bring the category evaluation to a higher value and thus significantly affect the probability of failure even if the FoS seems “reasonable”.
In this type of application where the knowledge about symptoms and dysfunctions
is encoded to finally provide a probability of an event, the “ignorance”—i.e., “not
knowing enough about a symptom”—is an extremely important piece of information
and is carefully used in the analyses. This applies, for example to boreholes and
instrumentation that may be concentrated (for practical/access reasons) in an area
and thus generate a lack of information that is geographically well distributed.
Of course, as new information becomes available, the category of the structures
will change (hopefully towards a higher category) and Bayesian updates can be performed. The benchmarking against the world portfolio and recent failures is preserved
to help understating emerging trends.
Before closing this review of semi-empirical methodologies three more points
must be discussed:
(1) Is there a correspondence between the results of a semi-empirical method and
a direct slope stability approach?
(2) Can a semi-empirical approach be used for pseudo-static analyses as well?
(3) What about liquefaction?
11 Dam Stability Failures
11.2.3 Does It Really Work?
Here is an example of “simplified” transformation of a deterministic slope stability
analysis into a “probabilistic one” using the simplified approach described in Silva
et al. (2008). The dam under consideration is well maintained and not damaged.
However, there was little control during construction and only partial monitoring
and observations. So, SLM method would propose a Class II–III structure, which is
definitely not a high quality dam.
With a FoS between 1.09 and 1.36 obtained by the engineers through their classic
slope stability analyses, we would have a p f = 10 −1 to 5 × 10 −2 or 5 × 10 −2 to 10 −3 .
It is now easy to include progressive release of attention, analysis of anomalies such as
deformations … and evaluate the resulting increased p f . It would then also be easy to
simulate progressive increase of attention, maintenance, monitoring, or hardening of
the structure through repairs and better monitoring, maintenance and see the effects.
We have applied this type of “symptom-based” approach, similar to the SLM
approach, to a variety of cases around the world, ranging from the evaluation of
unexploded ordnance risks in countries such as Lao PDR and Cambodia to the risk
assessment of mountainous roads networks. The success of the approach has been
measured by third parties through extensive reality checks.
For dams, we include thirty diagnostic points specific to tailings dams (see
Sect. 11.2.1). The range of probabilities stems out of a “optimistic” and a “pessimistic” evaluation of the thirty diagnostic points (a re-calibration of Table 11.4).
As many forensic studies have shown, the failure of a dam can rarely be attributed
to a single cause, but rather to a set of conditions (hence the thirty diagnostic points)
which, together, can bring the category evaluation to a higher value and thus significantly affect the probability of failure even if the FoS seems “reasonable”.
In this type of application where the knowledge about symptoms and dysfunctions
is encoded to finally provide a probability of an event, the “ignorance”—i.e., “not
knowing enough about a symptom”—is an extremely important piece of information
and is carefully used in the analyses. This applies, for example to boreholes and
instrumentation that may be concentrated (for practical/access reasons) in an area
and thus generate a lack of information that is geographically well distributed.
Of course, as new information becomes available, the category of the structures
will change (hopefully towards a higher category) and Bayesian updates can be performed. The benchmarking against the world portfolio and recent failures is preserved
to help understating emerging trends.
Before closing this review of semi-empirical methodologies three more points
must be discussed:
(1) Is there a correspondence between the results of a semi-empirical method and
a direct slope stability approach?
(2) Can a semi-empirical approach be used for pseudo-static analyses as well?
(3) What about liquefaction?