Documente Academic
Documente Profesional
Documente Cultură
TO
INTRODUCTION
Decision analysis for the management of an inventory of
aging embankment dams should include a consideration of the
overall risk or probability of a failure within the inventory.
Engineering risk can be defined as the probability of failure
multiplied by the cost of failure and hence is the expected cost
of failure. The price of a failure can include but is not limited
to: replacement cost; loss of service cost; loss of life; damage
to the environment; or other societal costs.
The price of a classical risk analysis and the time required
for its implementation may not be justified for dams on which
there has been no modern engineering analysis and for which
there is little or no information available concerning as-built
conditions and performance history and for which there is little
or no instrumentation. For such dams, current budgets available to state regulators may only enable a rapid walk-through
inspection once every several years. Information necessary to
perform a risk analysis may not be obtainable under current
budgeting constraints and in order to help prioritize maintenance, repair, and evaluation tasks on such dams, a simplified
procedure that can be conducted in real time during a walkthrough inspection at nominal additional cost may be justified.
The indexing system proposed herein is intended for this purpose.
In an effort to develop decision analysis tools that can be
more readily applied to large inventories of civil engineering
1
Sr. Engr., Burns Cooley Dennis Inc., Ridgeland, MS 39157; formerly,
Asst. Prof. of Civ. Engrg., Michigan State Univ., East Lansing, MI 48823.
2
Assoc. Prof. of Civ. Engrg., McGill Univ., Montreal, Quebec, Canada
H3A 2K6.
3
Prin. and Sr. Vice Pres., GZA GeoEnvironmental Technologies, Inc.,
Norwood, MA 02062.
4
Asst. Proj. Mgr., GZA GeoEnvironmental Technologies, Inc., Norwood, MA 02062.
Note. Discussion open until September 1, 2001. Separate discussions
should be submitted for the individual papers in this symposium. To
extend the closing date one month, a written request must be filed with
the ASCE Manager of Journals. The manuscript for this paper was submitted for review and possible publication on September 27, 1999; revised December 8, 2000. This paper is part of the Journal of Geotechnical and Geoenvironmental Engineering, Vol. 127, No. 4, April, 2001.
ASCE, ISSN 1090-0241/01/0004-03250334/$8.00 $.50 per page.
Paper No. 21939.
guidelines presented by FEMA (1998). The relative importance of various observable physical conditions that could lead
to a failure is then determined through a Bayesian updating
procedure and a simplified failure criticality analysis based
upon conditional probabilities of failure determined by expert
elicitation. An onsite inspection is then performed to determine
the current physical state. Inspection forms that describe various indicators of physical state are provided herein. The relative importance of the physical conditions is combined with
the importance of the dam in the inventory to obtain the overall importance with respect to the entire inventory. The results
of the importance determination are then combined with the
current physical condition (in a simple multiplication) to compute a risk index for each observed deficiency. For a particular
dam, these risk indices are summed to estimate an overall risk
index. Prioritization can be accomplished based on particular
physical deficiencies or upon the overall risk index.
FIG. 1.
where all terms have been previously defined. Note that the
maximum vulnerability score is 1,000 in accordance with (1).
Thus, a three-order-of-magnitude range is available for classifying dams within the same inventory according to vulnerability. This three-order-of-magnitude range might be similar
to that from a probability of failure analysis to dams in similar
inventories. Tables 18 present each of the vulnerability factors, the suggested criteria, and the suggested scores that are
used in the calculation. Note that the relationship between the
criteria and the scores proposed is approximately linear.
These criteria are in general conformance with similar criteria
Intrinsic Characteristics (I1 = Height of Dam)
Height of dam
[m (ft)]
(1)
Score
(2)
<2.7 (<9)
2.712.2 (940)
12.230.5 (40100)
>30.5 (>100)
1
3
6
10
TABLE 3.
Type of foundation
(1)
Score
(2)
Rock
Moraine
Alluvium
1
5
10
TABLE 4.
Storage capacity
[ha m (acre-ft)]
(1)
Score
(2)
<6.17 (<50)
6.17123 (50999)
1236,170 (1,00050,000)
>6,170 (>50,000)
1
3
6
10
TABLE 5.
4
2
2
TABLE 1.
Age of dam
(years)
(1)
Score
(2)
09
1029
3059
6099
>100
10
8
5
2
1
TABLE 6.
Score
(2)
V or lower
VI
VII
VIII
IX
1
2
6
8
10
TABLE 7.
Note: Ranges and scores can be modified for specific owner requirements.
Conditions
(1)
Score
(2)
Score
(3)
4
10
10
5
1
5
2
Note: Only select known conditions if a hydrologic and hydraulic analysis has been performed.
TABLE 8.
Score
(2)
10
1
7
2
(2)
and assign these probabilities directly using subjective judgment. In this case, it is suggested that a precision no greater
than 10% be used and that any failure mode with a conditional
probability of less than 10% be given zero conditional probability. If actual probabilities of failure have been determined
for each of the failure modes in a risk or reliability analysis,
these can be used in place of the direct elicitation. Note that
the consequence of assigning a zero conditional probability to
a failure mode at this stage is to cause all physical deficiencies
associated with it to have a zero risk index in the final analysis.
General guidance can be provided for the selection of these
conditional probabilities from the reported dam incidents summarized by the United States Committee on Large Dams Subcommittee of Dam Incidents and Accidents (USCOLD 1988).
For those incidents involving a failure of the dam (failure type
1 or failure type 2 as outlined in the report), 81 were associated
Overtopping
Piping
Mass movement
Surface erosion
0.49
0.32
0.09
0.10
FIG. 3.
P [Mi Xi F] =
P[Xi Mi F] P[Mi F ]
(3)
P[Xi Mi F] P[Mi F ]
i=1
where Xi = an observation of the presence of an attribute relative to failure mode i, and P[Mi Xi F ] is the updated
conditional probability of failure mode i in the presence of
attribute Xi and failure. In the above equation, the denominator
is a summation over the four possible failure modes. The equation is applied recursively for all the attributes Xi that are present at the dam using the latest posterior probabilities as the
new prior probabilities. Eq. (3) assumes that each attribute has
a first-order effect and that there is no synergy between attributes (in order to simplify the analysis). The conditional probabilities relating the attributes to the failure modes, P[Xi Mi
F ] in (3), must be determined and placed in Fig. 3. The writers
propose the use of expert elicitation for this purpose relying
upon engineering judgment and where appropriate upon an
analysis of existing data bases on dam performance. Note that
the probabilities of the complementary events can be obtained
by subtraction.
A conditional probability of 0.5 indicates that an attribute
has no effect on the updated conditional probability of the
failure mode while a conditional probability larger/smaller
than 0.5 indicate that the presence of the attribute increases/
decreases the likelihood of the failure mode, respectively. In
other words, if the likelihood of an attribute being present
given a failure mode and failure is 0.5, then its converse (likelihood that the attribute is absent) is (10.5) or 0.5 and the
attribute is a neutral indicator for that particular failure mode
given failure. These updating conditional probabilities are defined for four cases: (1) when the attributes are known to be
present (A in Fig. 3); (2) when they are suspected to be present
(B in Fig. 3); (3) when they are suspected to be absent (C in
Fig. 3); and (4) when they are known to be absent (D in Fig.
3). In other words, for each attribute that is applicable to a
particular dam, the evaluator answers whether or not the at-
Downloaded 20 May 2009 to 140.194.193.5. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
TABLE 10.
Hypothetical Calculation for Relative Importance of Physical Conditions (from Direct Elicitation)
Conditional probabilities
for failure modes
(1)
Overtopping (60%)
Overtopping (60%)
Overtopping (60%)
External Erosion (0%)
Piping (30%)
Piping (30%)
Mass movement (10%)
Mass movement (10%)
a
Relative importance
of observable physical
conditions to dam
(3)
30% 60% =
10% 60% =
60% 60% =
a
70% 30% =
30% 30% =
50% 10% =
50% 10% =
18%
6%
36%
21%
9%
5%
5%
Relative importance
of observable physical
conditions to inventory
IDam = 1,000
(4)
0.18 1,000 =
0.06 1,000 =
0.36 1,000 =
a
0.21 1,000 =
0.09 1,000 =
0.05 1,000 =
0.05 1,000 =
180
60
360
210
90
50
50
No further calculations are necessary because the initial conditional probability was assessed to be less than 10%.
(4)
where all terms have been previously defined. Under this definition, the relative importance of the jth physical condition
on a dam is the conditional probability that it would lead to a
failure of the dam multiplied by the importance of the dam in
the inventory. Table 10 presents a hypothetical relative importance determination for the physical conditions on a fictitious
embankment dam. In this example the importance of the embankment dam in the inventory is assumed to be 1,000. According to the hypothetical calculation presented in Table 10,
the most significant changes in physical conditions at this particular dam would be the loss of function in the outflow works,
piping in the embankment, and the loss of the spillway capacity.
Note that this analysis for relative importance of physical
conditions is generally conducted after an onsite inspection or
after an analysis of existing dam safety inspection reports,
when the engineer would have more information about the
current physical condition of the dam and the most likely
modes of failure. In other words, the current physical state can
and should have an impact on the perceived importance of the
physical conditions.
Condition
function
(C)
(1)
10
98
76
54
32
1
0
TABLE 12.
pacity
Indicator
(1)
Part of the spillway cross section if obstructed
010% obstructed
1025% obstructed
>25% obstructed
710
47
04
Indicator
(1)
Deviations from original crest elevation in terms of
loss of designed or estimated freeboard
010% loss
1025% loss
>25% loss
Trees on or near crest
Suggested
condition
range
(2)
Indicator
(1)
Obstructions in cross section of outlet pipes
010% obstructed
1025% obstructed
>25% obstructed
Suspected but unverified obstruction
Valves and gates for outlet works
Functioning properly or recently repaired
Not operated recently
Owner not willing to exercise valves and gates
Valves and gates cannot be opened
710
47
04
710
610
05
01
0
Indicator
(1)
Observed erosion/deterioration of spillway channel
None to minor
Some to moderate
Serious to extensive
Critical with sill lost
710
47
14
0
410
14
0
710
57
05
710
47
04
05
TABLE 14. Condition Inspection Checklist for Loss of Function of Low-Level Outlet Works
Suggested
condition
range
(2)
Indicator
(1)
TABLE 17.
bankment
TABLE 15.
sion
TABLE 16. Condition Inspection Checklist for Loss of Embankment Surface Protection Material
Indicator
(1)
Turbid flows
Evidence of prior occurrence that has gone uncorrected
Actively occurring
Sinkholes or depressions on the surface of dam
Build-up of pore water pressure in embankment as inferred by uncontrolled seepage areas
Changes in surface vegetation
Soft/wet areas on the surface
Constant surface flow
Increasing surface flow
Stumps and root systems left in place on embankment
or small animal burrows present
27
02
05
510
48
27
04
05
TABLE 18.
dation
Indicator
(1)
Turbid flows
Evidence of prior occurrence
Actively occurring
Sinkholes or depressions on dam, toe, or abutments
Buildup of pore water pressure in foundation as inferred by uncontrolled seepage areas in toe and
abutment areas
Changes in surface vegetation
Soft/wet areas on surface
Constant surface flow
Increasing surface flow
27
02
05
510
48
27
04
(10 CFj)
10
(5)
TABLE 19.
bankment
Indicator
(1)
Buildup of pore water pressure in embankment as inferred by uncontrolled seepage areas
Changes in surface vegetation
Soft/wet areas on the surface
Constant surface flow
Increasing surface flow
Surface evidence of impending mass movement such
as cracking, shallow slides, and differential
movement in the embankment or between the
embankment and foundation
Minor and localized
Major and extensive
Suggested
condition
range
(2)
510
48
27
04
28
02
TABLE 20. Condition Inspection Checklist for Slide in Embankment and Foundation
Indicator
(1)
Buildup of pore water pressure in embankment and
foundation as inferred by uncontrolled seepage
areas
Changes in surface vegetation
Soft/wet areas on the surface
Constant surface flow
Increasing surface flow
Surface evidence of impending mass movement such
as cracking, shallow slides, and bulging
Minor and localized
Major and extensive
Suggested
condition
range
(2)
510
48
27
04
28
02
tance. The overall risk index of the dam is obtained by summing risk index over all potential failure modes.
DISCUSSION
Relationship of Current Approach to Classical
Risk Analysis
The proposed methodology to calculate risk indices is compatible with a classical risk analysis in that: (1) the importance
of the dam (IDam) is estimated through a consideration of surrogate factors that are related to the likelihood of failure and
consequences of failure; and (2) the relative importance of
potential changes in physical state are determined in a failure
criticality analysis based upon their relative likelihood of causing a failure. It is possible to directly incorporate the results
of a classical risk or reliability analysis into this methodology
by using estimated probabilities of failure to determine the
conditional probabilities of failure for each failure mode. Additionally, the dam importance determination can be taken directly from a risk analysis by using the estimated risk instead
of the vulnerability and hazard considerations proposed herein.
The dam with the highest risk would be deemed the most
important. Thus, information determined during a detailed risk
or reliability analysis could supplant the information developed herein.
The proposed methodology provides no measure of the actual probability of failure. This is by design. In the writers
opinion, the determination of probability of failure for a small
embankment dam carries a considerable amount of uncertainty
and is currently the subject of much controversy. There is disagreement within the dam safety profession as to whether or
not it is possible to reliably determine the probability of failure
of a dam and whether or not the probability of failure is of
value for the prioritization of maintenance, repair, and evaluation tasks in an inventory of dams. At the onset of this development, the writers decided to table this controversial issue
until the profession could agree on the ability and usefulness
of determining absolute probabilities of failure. They decided
instead to base the analysis on the use of conditional probabilities. The expert panel members during the development of
the initial condition assessment methodology (senior dam
safety engineers from the Corp of Engineers and Hydro-Quebec) were much more comfortable answering questions dealing with conditional probabilities than with absolute probabilities.
The prioritizations accomplished by this methodology rely
solely on observational factors coupled with engineering judgment. This can be potentially misleading when dealing with
actual dam failures. However, in the absence of engineering
analysis, with the lack of knowledge of a performance history,
with the lack of knowledge of as-built conditions, and with
severely restricted budgets, the writers feel that expert elicitation based on observational factors (as developed herein) is
an appropriate manner in which to develop a tool that can be
used to help prioritize improvements to such inventories of
embankment dams.
A fully developed methodology for condition indexing of
embankment dams (from which this approach has been extracted) has been field tested in one of the management regions
of Hydro-Quebec using approximately 30 dams. The results
were so encouraging that their upper management gave their
authorization to implement the condition indexing methodology across their entire inventory of 300 plus embankment
dams and to use the results for the prioritization of maintenance and repair tasks. Additionally, Hydro-Quebec, the Corps
of Engineers, and the Bureau of Reclamation are currently
working together in a joint research project to develop a condition indexing methodology for spillways and flood water discharge facilities based upon the same principles used in the
methodology for embankment dams. Hence, the indexing
approach is being accepted by owners of large inventories of
dams as a viable alternative to risk analysis for the management of maintenance and repair tasks. The use of this tool
does not rule out the use of good judgment.
Broad Context for Decision-Making on Improvement
to Dams
In a broader context, having the ability to accurately assess
the current condition of an embankment dam and to prioritize
the observed deficiencies in terms of safety-related concerns
if of little value unless such information can be used to generate scope-of-work statements for proposed further investigations and improvements and to generate cost estimates. Improved decision-making must also consider the following
issues that are not directly addressed by the methodology:
1. Does the current physical state represent an emergency
situation that must be immediately remedied?
2. Is it necessary to completely remedy the observed deficiencies to achieve optimal performance or is there some
level of performance that is acceptable but less expensive?
3. What is the cost variance between various improvement
options, and how confident can we be in the cost estimates?
4. Is the physical condition changing at such a rate that if
the repairs are not effected today, there may be unac-
5.
6.
7.
8.
9.
Answering questions such as these requires that the risk indexing methodology described herein be linked to a structured
methodology for scope-of-work definition and cost estimation.
These tools can be used at the network level to prioritize limited funds. A well-conceived management system for improvements to embankment dams must be capable of evaluating
alternatives ranging from an initial do nothing and defer
scenario through all available repair options. The ultimate
choice between improvement alternatives must be a balanced
evaluation of impacts to both embankment dam performance
and cost. A quantitatively derived risk indexing tool (such as
the one presented herein) can be used to more accurately define performance levels (benefits) as well as scope of
work(cost) thus enabling an improved benefit/cost analysis
over the entire service life of an inventory of embankment
dams.
Precision of Risk Indices
In order to interpret the results of this risk indexing procedure, it is necessary to consider the precision with which these
risk indices can be determined. The precision of any derived
quantity can be approximated by using a propagation of errors
technique known as first-order second moment method, Benjamin and Cornell (1970). The risk index defined by (4) is the
product of four terms: (1) the importance of the dam; (2) the
conditional probability of failure for a failure mode; (3) the
conditional probability that each of the nine physical conditions leading to a particular failure mode; and (4) a condition
function. Assume that: (1) the importance of the dam is deterministic; (2) the conditional probabilities of failure for the
four failure modes from expert elicitation will be normally
distributed with a standard deviation (cp) of 10%; (3) the conditional probabilities of the nine physical conditions will be
normally distributed with a standard deviation (cp) of 10%;
and (4) the condition functions would be normally distributed
with a standard deviation (cf) of 10%. The standard deviation
for each risk index (RI) will depend upon the importance of
the dam (IDam), upon the conditional probabilities for the physical conditions (P[Mi F ], P[Cj Mi]), and upon the condition
function (CFj). Under the assumption that the conditional
probabilities for the failure modes and physical conditions and
the condition functions are independent, the standard deviation
of the risk index (RI) for any condition j can be expressed by
the following equation:
RI = IDam
P[Mi F ]
(10 CFj)
10
2cp
Condition: Current physical state of embankment dam as determined from observational data during walk-through inspection.
Condition function: Numerical score ranging from 0 (poor) to
10 (excellent) that rates current physical state of portion of
embankment dam in terms of its function in preventing one
or more failure modes, CF for jth condition.
Condition indexing: Process of quantifying current physical
state of facility in terms of set of predetermined rules.
Conditional probability: Probability of particular event under
premise that another event has occurred.
Failure criticality analysis: Systematic assessment of events
that can lead to failure and determination of their corresponding importance.
Hazard potential: Measure of specific types of damage such
as property loss or loss of life that may result in event of
dam breach.
P[Cj Mi]
where all terms have been previously defined. For dams indexed according to this methodology, the writers have found
that the standard deviation of a risk index [from (6)] has a
coefficient of variation that is approximately 25% under the
uncertainty assumptions given previously. Thus, the interpretation of the risk index for prioritization of improvements
GLOSSARY
(10 CFj)
10
2cp
2cf
(6)
APPENDIX II.
REFERENCES
APPENDIX III.
NOTATION