Print Go Back Next Page

Engineering Design Methods for Service Life

Prediction
K Moser1 &C Edvardsen2
1
EMPA Research & Testing Switzerland
2
COWI Consulting Engineers and Planners Denmark

Summary: Service Life prediction is required when designing large infrastructure projects and is
one of the objectives to be covered according to the European Construction Products Directive. It is
of major concern for the owners of any kind of structure or asset.
Large investment bodies (transport ministries, property funds, universities, consulting companies,
etc.) collect data on service life for specific components. The need arises however; to set up
generally applicable methods using basic data and adapting these to e.g. different application,
exposure and user conditions.
Today, two main methods can be distinguished:
1. For large infrastructure projects, teams of specialists are set up, investigating service life under
the exact conditions envisaged (Great Belt, Western Scheldt Tunnel, etc.) and developing tailor
made solutions, quite often using probabilistic methods.
2. In ISO 15686 – 1:2000 the so called "factor method" is proposed where seven factors are applied
to the basic value of service life catering for the individual quality, exposure and in use condition of
the building component considered.
These two methods exhibit heavy drawbacks. The probabilistic methods and most other tailor
made techniques applied are mostly based on probabilistics and are too elaborate to be used on
standard applications such as office buildings or ordinary road bridges. The factor method on the
other hand is fairly simple, but identifies the main parameters influencing service life. The result
however, is only a single figure for service life and does not take into account at all the variability of
the processes involved.
Thus within the CIB/RILEM Working Commission 175-SLM: "Service Life Methodologies", three
subtask groups were set up under the heading "Performance based methods of service life
prediction", one of them with the goal of setting up generally applicable "Engineering Design
Methods" fitting in between the two methods described above. This was thought to be done either
by simplifying scientific models or by expanding the factor method towards the more sophisticated
models.
The subtask groups have produced state of the art reports to be published in 2002 as a basis
document for further developments. As engineering design method, a principle solution is proposed
in this paper which can be applied to the factorial method for standard cases as well as to other set-
ups employing mathematical relations for service life. This is achieved by not using plain factors
but only, probability density functions instead. These are established using reliable and
understandable engineering techniques. Three Examples are shown to illustrate the proposed
procedure.

Keywords. Service life, engineering design method, factor method, probabilistic method, ISO 15686

9DBMC-2002 Paper 222 Page 1

 Go Back Next Page

1 INTRODUCTION
This paper reports on work done with regard to the service life prediction under the wings of the CIB/RILEM Working
Commission 175: "Service Life Methodologies", in close co-operation with ISO TC 59 / SC 14 “ Building Construction -
Design Life”. Three subtask-groups were established in 1999 under the heading "Performance based methods of service life
prediction". For one subtask-group the goal was set to developing generally applicable "Engineering Design Methods (EDM)"
for service life design, methodically fitting in between the probabilistic and the factorial methods. This was thought to be done
either by simplifying scientific models or by expanding the factor method towards the more sophisticated models. The three
subtask-groups compiled state of the art reports as a basis for future development (CIB/RILEM 2002).
2 INCENTIVES FOR SERVICE LIFE PREDICTION METHODS
Service life is of major concern for the owner of any kind of structure or asset. Its prediction is therefore required by investors,
be it a state organisation or a private body, when designing large infrastructure projects, and for private investors holding large
property assets. Service life is as main parameter involved in all economic considerations as return of the invested capital,
investment planning for maintenance and refurbishment, etc.
Service life is also the essential requirement to be covered according to the European Construction Products Directive: Within
the period of service life all main qualities of a constructed asset or product have to be met, viz. the work has to be
mechanically safe, protected against failure as a consequence of fire, it has to be hygienic, safe to use, with an acceptable noise
level, and has to maintain sufficient comfort to the user at a low energy consumption (ECC 1988, Appendix 1).
3 STATE OF THE ART
3.1 Relevant data available
Large investment bodies (transport ministries, property funds, universities, consulting companies, etc.) collect data on service
life for specific components. For large populations of similar structures, service life planning in this case can be a relatively
simple matter of applying the respective collected data. In this case no elaborate models are required, as the data fit more ore
less closely to the problem under consideration.

3.2 Lack of relevant data

In most cases, directly applicable data are not available. This is namely the case, when the works are
• not reasonably comparable, or
• exposure conditions are different, e.g. due to sites in other climatic environments, or
• they are built in areas where no sufficient service life data are available, or
• new products have to be applied in new climatic conditions based on tests under other conditions.
Therefore, the need arises, to set up generally applicable methods using basic data and adapting these to e.g. different
application, exposure conditions and user conditions, etc.
Today, two main methods of service life prediction can be distinguished: the probabilistic methods and the factorial method.

3.3 Probabilistic methods

For large infrastructure projects, teams of specialists are set up, investigating service life under the exact conditions envisaged
(Great Belt, Western Scheldt Tunnel, etc.). They develop tailor made solutions, most often using quite elaborate probabilistic
methods (see literature, viz. DuraCrete Design Manual, Edvardsen & Mohr 2000, Breitenbüchner et al, Fagerlund, Leira,
Lounis et al, Siemes, all in 1999, etc.).
These techniques are out of the reach of the ordinary planner, at least as long as he is not in the position to acquire the inherent
knowledge easily and quickly, and as long as there are no easy-to-use design aids available, properly introduced to the
engineering community.

3.4 Factorial method

ISO TC 59/SC 14 “ Building Construction - Design Life” has set up six working groups so far successfully producing several
parts of ISO 15686 "Buildings and constructed assets: Service Life Planning". In the first part, the so called "Factor method" is
proposed applying seven factors to the basic value of reference service life, catering for the individual quality, exposure and in
use condition of the building component considered (ISO 15686-1, 2000, for working examples see annexes E & F).
This method, applied plainly as set up in the code, yields one single value for the average service life. A customer however, is
not only interested in the average value; he has to know as from when substantial renovations or replacements have to be
expected. This usually is the case long before the average service life is reached.

9DBMC-2002 Paper 222 Page 2

 Go Back Next Page

3.5 Appraisal of the state of the art

These two methods shortly described above exhibit to heavy drawbacks:
• The probabilistic methods and most other tailor made techniques which are mostly based on the theory of
probabilistics are too specialised, too cumbersome or too complicated to be used by ordinary planners on standard
applications such as office buildings or ordinary road bridges.
• The factor method requires the estimation or evaluation of up to seven individual factors and nevertheless gives a
single figure for service life only. It does not at all take into account at all the variability of the processes involved in
the ageing of the structure or component involved.

3.6 Performance based methods of service life prediction

Within the CIB/RILEM Working Commission 175: "Service Life Methodologies", three subtask groups were set up under the
heading "Performance based methods of service life prediction". The three groups are titled as follows:
• Scientific (probability, stochastic) methods,
• Engineering design methods and
• Factorial method.
The second group with the goal of setting up generally applicable "Engineering Design Methods (EDM)" fits in between the
two other methods already described above. EDMs were thought to be gained by either by expanding the factor method
towards the more sophisticated models or by simplifying scientific models for general use.
Firstly, the three subtask-groups have produced state of the art reports being put together as a basis document for further
developments (CIB/RILEM 2002).
4 ENGINEERING DESIGN METHODS
As a first engineering design method, a principle solution is proposed which can be applied to the factorial method for standard
cases as well as to other set-ups employing mathematical relations for service life. This is achieved by not using plain factors
but probability density functions instead. These are established using reliable and understandable engineering techniques.

4.1 Data acquisition: the recursive Delphi method

In many cases, data are not available for ready use in the form of values for the main factors of influence defined in the
equation for service life, let alone in the form of distributions. A very valid method is the so-called (recursive) Delphi method.
(The author has been engaged in this type of data acquisition as far back as the 1980’s in the field of (industrial) risk
engineering.)
1. In a first step, a panel of experts is called together and asked on their professional opinion on the distributions of
the different factors, their type of distribution (normal, lognormal, Gumbel, etc), their mean values, standard
deviations.
Usually, it is easier to define fractiles based on experience and professional judgement, say 5 or 10%, mean
values and 90 or 95%. Experts can quite well define these values, when asked precisely.
2. The second step involves the service life calculation using the expert panel’s input. The distributions are used
instead of plain factors in the mathematical formulations for service life.
3. The third step is a thorough discussion of the results and of the dominating parameters. Sensitivity analysis has
proved to be an important tool at this stage. Very often, as a consequence of this appraisal, the data or the
models have to be adjusted to yield results judged reasonable in those areas of the problem where the experts
have sufficient practical experience. After this fine-tuning of the model and the density distributions of the
factors, the general problem can be tackled successfully.

9DBMC-2002 Paper 222 Page 3

 Go Back Next Page

Model Parameters & Model Input Results & Sensitivity Analysis

Panel of Experts

Figure 1: Recursive Delphi method

4.2 Application of Engineering Design Methods

The general engineering design method (EDM), at this stage, is defined as “any simple mathematical relation (as simple as
possible, but not too simple) worked on, using distributions of any kind for the individual factors in the relation”. This
procedure yields as result distributions for the expected service lives, information, which can easily be understood and
interpreted by the decision-makers.
The following examples show three variations of the engineering design method EDM:
The first example uses all seven factors of ISO 15686-1, 2000 under the assumption, that the information for defining the
respective distributions is readily available.
The second example is working on limited information only. The equation has to be modified, and the respective distributions
are set up indirectly, partially from information on the resultant differences of service life.
The third example bases on a completely different equation, which is established using common engineering sense. It is
normalised using an average result calculated on the basis of an equation based on an error-function.
5 EXAMPLE 1: EXPANSION OF THE FACTORIAL METHOD
This example demonstrates the basic procedure with the following full equation (1). All factors are applied as indicated, using
four different distributions in order to demonstrate the ease of application of this expanded factorial method. The example is
based on the worked example for softwood windows given in ISO/CD 15686-1 (1997) but the factors are indexed as in the
edition 2000 of the code:

PSLDC = RSLC ⋅ f A ⋅ f B ⋅ f C ⋅ f D ⋅ f E ⋅ f F ⋅ f G (1)

PSLDC is the predicted service life distribution of the component based on the reference service life RSLC. The factors indices
are: A for the quality of the component, B for the design level, C for the work execution level, D for the indoor environment, E
for outdoor environment, F for in-use condition and G for maintenance level (see Tab. 1).

5.1 Estimated service lives for the windows in all four faces
The basis of the numerical example is a squared building of a length of 50 m, a width of 25 m and a height of 30 m. The long
sides are facing due south and north respectively. The windows of the four façades of the building are in the example treated
separately. Table 1 shows the assumed relevant conditions for all factors and faces. Therefrom the factors for the three fractiles
5%, 50% and 95% are defined in the sense of the Delphi method, in this case based on the factors and their description given in
ISO/CD 15686-1:2000.

9DBMC-2002 Paper 222 Page 4

 Go Back Next Page

Table 1: Fractile values of factors

Factors for the
Factor Face Relevant conditions fractiles
5% / 50% / 95%
fA Quality of all general variations of components 1.2 / 1.5 / 1.8
component
fB Design level all good, identical value 1.2
fC Work execution level all general variation, 1.0 / 1.2 / 1.5
but insufficient quality repaired
fD Indoor environment S occasional risk of condensation 0.9 / 1.0 / 1.2
W medium risk of condensation 0.8 / 0.9 / 1.1
N high risk of condensation 0.7 / 0.8 / 0.95
E medium risk of condensation 0.8 / 0.9 / 1.1
fE Outdoor environment S occasional cycling dry / damp 0.8 / 1.0 / 1.3
W regular cycling dry / damp 0.6 / 0.8 / 1.0
N sheltered from rain 1.0 / 1.2 / 1.5
E occasional cycling dry / damp 0.8 / 1.0 / 1.3
1
fF In use conditions S occasional access by children ) 0.8 / 1.0 / 1.2
1
W regular access by children ) 0.6 / 0.8 / 1.0
1
N occ. / reg. access by children ) 0.7 / 0.9 / 1.1
1
E occasional access by children ) 0.8 / 1.0 / 1.2
fG Maintenance level all painted on judgement of caretaker 0.9 / 1.0 / 1.1
1
Note: ) according to example in ISO/CD 15868-1 (1997), other descriptions for wear and tear may appear more realistic.
The values for the fractiles given in table 1 are approximated by the density functions given in Table 2 for the ease of
processing. The functions chosen represent those generally used: deterministic, normal, lognormal and Gumbel (extreme-
value) distributions. The program used for the data processing, VaP 1.6 (1996), supports 11 types of distributions and user
defined functions. It only requires 2.2 MB of disc-space in the expanded state.
The factors for C reflect the fact, that a quality of workmanship usually is such, that all parts not being sufficient are upgraded
to meet the requirements, whereas those exceeding the requirements are naturally left at their higher level. This procedure
leads typically to asymmetric distributions (in this example approximated using an extreme value “Gumbel” distribution) with
very few components below the satisfactory level of 1.0 and a fairly wide spread upper area of the distribution.
The variables m and s are the first and second moments of the respective distributions (see VaP 1.6 1996).
Table 2: Resultant predicted service life distribution of the components (PSLDC)
Face
Factor Type of South West North East
Distribution m/s m/s m/s m/s
RSLC Deterministic 25 years 25 years 25 years 25 years
fA Normal 1.5 / 0.185 1.5 / 0.185 1.5 / 0.185 1.5 / 0.185
fB Deterministic 1.20 1.20 1.20 1.20
fC Gumbel 1.25 / 0.10 1.25 / 0.10 1.25 / 0.10 1.25 / 0.10
fD Lognormal 1.05 / 0.10 0.95 / 0.10 0.80 / 0.10 0.95 / 0.10
fE Lognormal 1.05 / 0.20 0.80 / 0.20 1.25 / 0.20 1.05 / 0.20
fF Normal 1.0 / 0.12 0.80 / 0.12 0.90 / 0.12 1.0 / 0.12
fG Normal 1.0 / 0.06 1.0 / 0.06 1.0 / 0.06 1.0 / 0.06
1
PSLDC (years) Lognormal ) 62.0 / 20.4 34.2 / 11.8 50.6 / 14.8 56.1 / 18.6
1
Note: ) close fit
The results were calculated by direct methods using VaP 1.6 (1996) and are shown on the last line of Table 2. The average
results of two runs of a Monte Carlo simulation match the mathematically calculated results by a maximum difference on the
average value of 0.1 years. These simulations yielded detail results as graphically shown for each facade as in Fig. 2. Dividing

9DBMC-2002 Paper 222 Page 5

 Go Back Next Page

the relative densities shown on the vertical axis through the number of runs, in this case by 100’000, derives at the absolute
densities.
5.2 Comparison of the four façades
The results for the estimated service lives of the four façades in Fig. 2 are different with respect to several aspects. First one
notices the different widths of the distributions, in accordance to the values (second moments) in Fig.2). The spread is largest
for the south face and narrowest for the west face. Some of this effect is relative: Due to the higher average value, the same
relative spread is larger in years.
The west face shows the shortest service life, as expected, mainly due to the unfavourable outdoor climate and the in-use
conditions. The effect of the higher risk of condensation, assumed for the north face indoor climate, is offset by the more
favourable outdoor climate. The main difference originates from the in-use conditions. Both effects combined yield some 15%
less estimated service life.

Fig. 5: Density distributions for the windows in the four faces

Note: Densities are the result of 105 runs of a Monte Carlo simulation
Figure 2: Distributions of predicted service lives PSLDC for all four facades

5.3 Financial demand

For the planning of the maintenance funds, the functions for the service lives of the similar building parts can be superimposed.
In general, this has to be done for all parts of a building considered. For the superposition, costs have to be allocated to the
different groups of building parts.
In this example, the superposition of all window areas to be replaced is executed only, in order to be able to show typical
results. (It is assumed for this purpose, that the windows cover 40% of the area of the respective façades.)

9DBMC-2002 Paper 222 Page 6

 Go Back Next Page

60

50

Total area of windows

(=integral of function)

window area [m ]
40

2
2
A = 1800 m

30

20

10

0
0 20 40 60 80 100 120
service life [years]

Figure 3: Financial demand

The superposition yields an asymmetric function having a steep increase up to a peak demand of replacements of 48 m2/year
after 37 years (see Fig. 3). Then the demand decreases at a gentle slope down to 10 m2/year after 70 years.
In a next step, the same service life functions can again be applied to the replaced windows, and the results of the multiple
replacements are summed up, leading to a fairly constant replacement function. These steps are omitted here for clarity.
In general, the financial demand for similar parts tends to merge into a one-peak function. The superposition of the functions of
all different parts of a building is more likely to result in several peaks or even a relatively steady demand over the lifetime of
the building considered, starting at a certain age of the building.
6 EXAMPLE 2: MODIFICATION OF THE FACTORIAL METHOD
This example deals with service life of fibre cement slates used as wall cladding. The input data is fairly scarce, far form being
complete and not directly suited for application in service life calculation. The basis of this calculation is the factorial method
as set up in ISO 15686-1:2000, modified to suit this specific case. The example shows that interpretation of available limited
data can nevertheless lead to a coherent and satisfactory service life prediction.

6.1 Available data

A manufacturer supplied data from his experience as follows:
• The quality of production of the slates can be derived from the bending strength, assumed to be characteristic value
for the mechanical strength. The mean value m lies 20% above the strength required and the standard deviation s is
very small, normalised: v = m/s = 0.015.
• The design level is such, that out of all designs some 10% to 15% are considered to be inadequate.
• The quality of work execution is at a fairly high level and some 5% are judged to be inadequate.
• Outdoor environment under normal conditions results in the following modifications of service life:

Table 3: Service lives and equivalent factors for different expositions

exposition East North West South
difference in service life ± 0 years - 2 to 3 years -7 to 10 years - 5 to 7 years
equivalent factor 1.0 0.95 0.85 0.90
The equivalent factors have been estimated knowing that the expected service life lies somewhere between 50 and 60 years.
• In-use conditions do not have to be considered, as only direct mechanical destruction can be a result of use. These
cases are however not of statistical significance.
• Maintenance level does not have to be considered, as basically no maintenance is required. Flat sheets are installed
and do not receive any, or minimal maintenance only, throughout their service life.

6.2 Input data

For the calculation of service life, the following equation is set up according to the factorial method:

PSLDC = RSLC ⋅ f A ⋅ f B ⋅ f C ⋅ f E (2)

9DBMC-2002 Paper 222 Page 7

 Go Back Next Page

From above inputs, the following mean factors and standard deviations, or second moments respectively, are derived at:
• The density distribution of the factor for the quality of the component is, on the basis of the mechanical strength, set
to a mean value of fA = 1.2. The standard deviation is, on the basis of the normalised standard deviation from
production, set to sA = 0.02.
• The density distribution of the factor for the design level is set to a mean of fB = 1.1 with a standard deviation of sB =
0.12, resulting in some 13% of the cases being below 1.0, i.e. exhibiting insufficient quality.
• The density distribution of the factor for the work execution level is from experience asymmetric and a lognormal
distribution is defined by a mean value (first moment) of fC = 1.1 and a second moment of sC = 0.06, resulting in some
5% of the cases being below 1.0, i.e. insufficient.
• The density distribution of the factors for the outdoor environment are set to the mean values fE in Tab. 3 above. The
standard deviation is, for the sake of simplification, set to an estimated sE = 0.1 for all four expositions.

6.3 Calculation of service life

The calculation using VaP (1996) yields the following results for the predicted service life PSLC, basing on a RSLC of 50
years (deterministic value, for the purpose of this example):

Table 4: Service lives and equivalent factors for different expositions

Exposition East North West South

PSLC, mean value [years] 72. 6 69.0 61.7 65.3
m, standard deviation [years] 11.5 11.2 10.5 10.9
for 16% damaged: SLC ≈ PSLC - m 61 58 51 54
The distributions for the service lives of two facades are shown in Fig. 4.

East facade [years] West facade [years]

5
Note: Densities are the result of 10 runs of a Monte Carlo simulation

Figure 4: Distributions of predicted service lives (PSLDC) for East and West facades
Under the assumption, that damage to about one out of every six of the slates requires replacement of the entire respective
cladding, the service life of the four facades (for a fractile of about 16% of damaged slates) is also shown in the above table,
varying from 61 to 51 years.

6.4 Discussion of results

The differences in service life given from experience can be found in the results of the prediction. For the purpose of
investment planning, the 16% fractile (or any other fractile deemed to be reasonable) seems to be a good indication of the point
in time of replacement.
This example shows that even on the basis of relatively scarce input, quite sensible service life prediction are possible.
7 EXAMPLE 3: SIMPLIFICATION OF THE PROBABILISTIC METHOD
This example deals with chloride ingress into concrete. It is based on a paper using the probabilistic approach on the basis of
error-functions (Edvardsen & Mohr 2000). In the paper, the authors compare the results of deterministic and of probabilistic
calculations of service life of reinforced concrete structures in two climates (10°C and 30°C). The necessary concrete cover for
a service life of 50 years has to be determined. A chloride content of 0.1 % chloride by mass of concrete defines the service life
at the reinforcement.

9DBMC-2002 Paper 222 Page 8

 Go Back Next Page

The example demonstrates, that service life prediction using a relatively plain formula, different to the one in ISO 15686, 2000,
can be done in the same way, again by introducing densities for the factors involved.
7.1 Procedure for simplified equation
In terms of an engineering design method, the following procedure is used:
1. The mean value of the chloride ingress depth is calculated as x = 34 mm, using the equation for diffusion for a time of 50
years:
  x 
c( x, t ) = c s − (c s − ⋅c0 )1 − erf    (3)
  
  2 D t  
where: c: concentration on chlorides, cs: concentration at the outer face and c0: initial concentration in the concrete. The
front of the ingress is defined by the critical value c = ccrit = 0.1 % of mass of concrete.
2. A simplified diffusion equation is set up for the depth x of chloride ingress. By this, all constants in the equation of
diffusion are rounded up into one single constant K:
x ≈ K (cs − ccrit − c0 ) D , (4)

3. The constant K is calculated by solving the equation for the mean value x. Using the mean value of 34 mm this results in
K = 38⋅103 [s0.5 / wt.-%].
Table 5: Values used for the diffusion calculations
variable distribution mean value standard deviation
Surface chloride concentration cs Lognormal 1.0 [wt-%] 0.3 [wt-%]
Critical chloride content ccrit Normal 0.1 [wt-%] 0.025 [wt-%]
Initial chloride content Normal 0.01 [wt-%] 0.002 [wt-%]
c0 -12 2
Eff. chloride diffusion coefficient (10°C) Normal 1.0·10 [m /s] 0.1·10-12 [m2/s]
D1 -12 2
Eff. chloride diffusion coefficient (30°C) Normal 4.0·10 [m /s] 0.4·10-12 [m2/s]
D2
7.2 Solving equation using distributions
The equation reads now as

x = 38 ⋅103 (c s − ccrit − c0 ) D [ s/% mass] (4)

Solving this equation using the same density distributions as used in the detailed probabilistic solution (see table of Fig. 5),
yields results as shown in Fig. 6. The slight skewness of the resultant density distribution is neglected for indicating standard
deviations.
7.3 Fractile value required
The probability of exceeding the critical content ccrit is set to 10% in Edvardsen & Mohr 2000 Assuming a normal distribution,
the value of the fractile of 90% is derived at by adding λ = 1.28 standard deviations to the mean: x90 = x + λs . These results
are also shown in Fig. 6 and are compared to the exact values of the original paper. The mean values are identical as well as the
standard deviation for Diffusion constant D1. For Diffusion constant D2, the fractile value of this prediction exceeds the exact
value by some 5%.
This accuracy is deemed to satisfy the needs of the customer, bearing in mind, that all input values, although being set up as
distributions, are still never perfectly exact.

9DBMC-2002 Paper 222 Page 9

 Go Back

diffusion mean standard Fracile of 10% exceedance

constant value deviation
eng. design Edvardsen &
method Mohr 2000
D1 34 mm 12 mm 49 mm 49 mm
D2 68 mm 23 mm 97 mm 91 mm

x [mm] for D1 x [mm] for D2

5
Note: Densities shown are the results of 10 runs of a Monte Carlo simulation
Figure 5: Engineering design method: density distributions for the depth of chloride ingress
8 CONCLUSION
In order to calculate service life in the sense of the engineering design method EDM an equation containing the relevant factors
at their relevant levels has first to be set up. Density distributions instead of plain factors in the equation for service life greatly
improve the information content and the relevance of the results at a significantly reduced intellectual, mathematical and
timewise input, compared to the quite often elaborate original equations.
By making the trail to the results clearly understandable, fewer errors will occur and fewer traps stepped into. Thus the method
can be applied by the plain engineer yielding nearly as good results as finely tuned sophisticated probabilistic models.
9 REFERENCES
1. Breitenbüchner R., Gehlen C., Schiessl P., Van den Hoonard J., Siemes T. 1999, Service life design of the Western
Scheldt tunnel, proc. 8DBMC, eds. M.A. Lacasse & D.J. Vanier, NRC Research Press, Ottawa, pp. 3-15.
2. CIB/RILEM: Service life Prediction: State of the Art, 2002, eds. Jostein P.H, Moser K., Siemes T., to be published.
3. DuraCrete Manual, The European Union, Brite EuRam III, contract BRPR-CT95-0132, Project BE95-1347, Report no.
BE95-1347/R17, May 2000
4. ECC 1988, Construction Products Directive, European Community Council 89/106/EWG updated 93/68/EWG,
Appendix 1.
5. Edvardsen C., Mohr L. 2000, Designing and Rehabilitating Concrete Structures: Probabilistic Approach (DuraCrete),
proc. CANMET/ACI 2000, Intl. Conference Durability of Concrete, pp. 1193-1208.
6. Fagerlund G. 1999, Service life with regard to frost attack – a probabilistic approach, proc. 8DBMC, eds. M.A.
Lacasse & D.J. Vanier, NRC Research Press, Ottawa, pp. 1268-1279.
7. ISO 15686-1:2000, Buildings and constructed assets- Service life planning- Part 1: General principles, International
Standard Organisation, Geneva.
8. ISO 15686-2:2001, Buildings and constructed assets: Service life planning, Part 2 – Service life prediction procedures,
International Standard Organisation, Geneva.
9. Lounis Z., Lacasse M.A., Siemes A.J.M., Moser K. 1998, Further steps towards a quantitative approach to durability
design, proc. Materials and Technologies for Sustainable Construction, Construction & Environment, CIB World
Building Congress, Gävle, Sweden, pp. 315-324.
10. Moser K. 1999, Towards the practical evaluation of service life – Illustrative application of the probabilistic approach,
proc. 8DBMC, eds. M.A. Lacasse & D.J. Vanier, NRC Research Press, Ottawa, pp. 1319-1329.
11. Naus D. 2000: Life prediction and aging management of concrete structures, ed. D. Naus, International RILEM
Workshop, Cannes, France.
12. Siemes T., Edvardsen C. 1999, Duracrete: Service life design for concrete structures, proc. 8DBMC, eds. M.A. Lacasse
& D.J. Vanier, NRC Research Press, Ottawa, pp. 1343-1356.
13. Teply B., Novak D., Kersner Z., Lawanwisut W. 1999, Deterioration of reinforced concrete: probabilistic and
sensitivity analyses, Acta Polytecnica and proc. 8DBMC, eds. M.A. Lacasse & D.J. Vanier, NRC Research Press,
Ottawa, pp. 1357-1366.
14. VaP 1.6 for Windows™ (1996) Short course in Variables Processing, ETH (Swiss Federal School of Engineering)
Zürich, Manual 13 p. in English; Papers 33 p. and Examples 22 p. in German.

9DBMC-2002 Paper 222 Page 10